linux/fs/btrfs/inode.c

7515 lines
198 KiB
C
Raw Normal View History

/*
* Copyright (C) 2007 Oracle. All rights reserved.
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public
* License v2 as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* General Public License for more details.
*
* You should have received a copy of the GNU General Public
* License along with this program; if not, write to the
* Free Software Foundation, Inc., 59 Temple Place - Suite 330,
* Boston, MA 021110-1307, USA.
*/
#include <linux/kernel.h>
#include <linux/bio.h>
#include <linux/buffer_head.h>
#include <linux/file.h>
#include <linux/fs.h>
#include <linux/pagemap.h>
#include <linux/highmem.h>
#include <linux/time.h>
#include <linux/init.h>
#include <linux/string.h>
#include <linux/backing-dev.h>
#include <linux/mpage.h>
#include <linux/swap.h>
#include <linux/writeback.h>
#include <linux/statfs.h>
#include <linux/compat.h>
#include <linux/bit_spinlock.h>
#include <linux/xattr.h>
#include <linux/posix_acl.h>
#include <linux/falloc.h>
include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit slab.h inclusion from percpu.h percpu.h is included by sched.h and module.h and thus ends up being included when building most .c files. percpu.h includes slab.h which in turn includes gfp.h making everything defined by the two files universally available and complicating inclusion dependencies. percpu.h -> slab.h dependency is about to be removed. Prepare for this change by updating users of gfp and slab facilities include those headers directly instead of assuming availability. As this conversion needs to touch large number of source files, the following script is used as the basis of conversion. http://userweb.kernel.org/~tj/misc/slabh-sweep.py The script does the followings. * Scan files for gfp and slab usages and update includes such that only the necessary includes are there. ie. if only gfp is used, gfp.h, if slab is used, slab.h. * When the script inserts a new include, it looks at the include blocks and try to put the new include such that its order conforms to its surrounding. It's put in the include block which contains core kernel includes, in the same order that the rest are ordered - alphabetical, Christmas tree, rev-Xmas-tree or at the end if there doesn't seem to be any matching order. * If the script can't find a place to put a new include (mostly because the file doesn't have fitting include block), it prints out an error message indicating which .h file needs to be added to the file. The conversion was done in the following steps. 1. The initial automatic conversion of all .c files updated slightly over 4000 files, deleting around 700 includes and adding ~480 gfp.h and ~3000 slab.h inclusions. The script emitted errors for ~400 files. 2. Each error was manually checked. Some didn't need the inclusion, some needed manual addition while adding it to implementation .h or embedding .c file was more appropriate for others. This step added inclusions to around 150 files. 3. The script was run again and the output was compared to the edits from #2 to make sure no file was left behind. 4. Several build tests were done and a couple of problems were fixed. e.g. lib/decompress_*.c used malloc/free() wrappers around slab APIs requiring slab.h to be added manually. 5. The script was run on all .h files but without automatically editing them as sprinkling gfp.h and slab.h inclusions around .h files could easily lead to inclusion dependency hell. Most gfp.h inclusion directives were ignored as stuff from gfp.h was usually wildly available and often used in preprocessor macros. Each slab.h inclusion directive was examined and added manually as necessary. 6. percpu.h was updated not to include slab.h. 7. Build test were done on the following configurations and failures were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my distributed build env didn't work with gcov compiles) and a few more options had to be turned off depending on archs to make things build (like ipr on powerpc/64 which failed due to missing writeq). * x86 and x86_64 UP and SMP allmodconfig and a custom test config. * powerpc and powerpc64 SMP allmodconfig * sparc and sparc64 SMP allmodconfig * ia64 SMP allmodconfig * s390 SMP allmodconfig * alpha SMP allmodconfig * um on x86_64 SMP allmodconfig 8. percpu.h modifications were reverted so that it could be applied as a separate patch and serve as bisection point. Given the fact that I had only a couple of failures from tests on step 6, I'm fairly confident about the coverage of this conversion patch. If there is a breakage, it's likely to be something in one of the arch headers which should be easily discoverable easily on most builds of the specific arch. Signed-off-by: Tejun Heo <tj@kernel.org> Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
2010-03-24 08:04:11 +00:00
#include <linux/slab.h>
#include <linux/ratelimit.h>
#include <linux/mount.h>
#include "compat.h"
#include "ctree.h"
#include "disk-io.h"
#include "transaction.h"
#include "btrfs_inode.h"
#include "ioctl.h"
#include "print-tree.h"
#include "ordered-data.h"
#include "xattr.h"
#include "tree-log.h"
#include "volumes.h"
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
#include "compression.h"
Btrfs: Change btree locking to use explicit blocking points Most of the btrfs metadata operations can be protected by a spinlock, but some operations still need to schedule. So far, btrfs has been using a mutex along with a trylock loop, most of the time it is able to avoid going for the full mutex, so the trylock loop is a big performance gain. This commit is step one for getting rid of the blocking locks entirely. btrfs_tree_lock takes a spinlock, and the code explicitly switches to a blocking lock when it starts an operation that can schedule. We'll be able get rid of the blocking locks in smaller pieces over time. Tracing allows us to find the most common cause of blocking, so we can start with the hot spots first. The basic idea is: btrfs_tree_lock() returns with the spin lock held btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in the extent buffer flags, and then drops the spin lock. The buffer is still considered locked by all of the btrfs code. If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops the spin lock and waits on a wait queue for the blocking bit to go away. Much of the code that needs to set the blocking bit finishes without actually blocking a good percentage of the time. So, an adaptive spin is still used against the blocking bit to avoid very high context switch rates. btrfs_clear_lock_blocking() clears the blocking bit and returns with the spinlock held again. btrfs_tree_unlock() can be called on either blocking or spinning locks, it does the right thing based on the blocking bit. ctree.c has a helper function to set/clear all the locked buffers in a path as blocking. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 14:25:08 +00:00
#include "locking.h"
#include "free-space-cache.h"
#include "inode-map.h"
struct btrfs_iget_args {
u64 ino;
struct btrfs_root *root;
};
static const struct inode_operations btrfs_dir_inode_operations;
static const struct inode_operations btrfs_symlink_inode_operations;
static const struct inode_operations btrfs_dir_ro_inode_operations;
static const struct inode_operations btrfs_special_inode_operations;
static const struct inode_operations btrfs_file_inode_operations;
static const struct address_space_operations btrfs_aops;
static const struct address_space_operations btrfs_symlink_aops;
static const struct file_operations btrfs_dir_file_operations;
static struct extent_io_ops btrfs_extent_io_ops;
static struct kmem_cache *btrfs_inode_cachep;
struct kmem_cache *btrfs_trans_handle_cachep;
struct kmem_cache *btrfs_transaction_cachep;
struct kmem_cache *btrfs_path_cachep;
struct kmem_cache *btrfs_free_space_cachep;
#define S_SHIFT 12
static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
[S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
[S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
[S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
[S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
[S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
[S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
[S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
};
static int btrfs_setsize(struct inode *inode, loff_t newsize);
static int btrfs_truncate(struct inode *inode);
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
static noinline int cow_file_range(struct inode *inode,
struct page *locked_page,
u64 start, u64 end, int *page_started,
unsigned long *nr_written, int unlock);
static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
struct btrfs_root *root, struct inode *inode);
static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
struct inode *inode, struct inode *dir,
const struct qstr *qstr)
{
int err;
err = btrfs_init_acl(trans, inode, dir);
if (!err)
err = btrfs_xattr_security_init(trans, inode, dir, qstr);
return err;
}
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
/*
* this does all the hard work for inserting an inline extent into
* the btree. The caller should have done a btrfs_drop_extents so that
* no overlapping inline items exist in the btree
*/
static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
struct btrfs_root *root, struct inode *inode,
u64 start, size_t size, size_t compressed_size,
int compress_type,
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
struct page **compressed_pages)
{
struct btrfs_key key;
struct btrfs_path *path;
struct extent_buffer *leaf;
struct page *page = NULL;
char *kaddr;
unsigned long ptr;
struct btrfs_file_extent_item *ei;
int err = 0;
int ret;
size_t cur_size = size;
size_t datasize;
unsigned long offset;
if (compressed_size && compressed_pages)
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
cur_size = compressed_size;
path = btrfs_alloc_path();
if (!path)
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
return -ENOMEM;
path->leave_spinning = 1;
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
key.objectid = btrfs_ino(inode);
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
key.offset = start;
btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
datasize = btrfs_file_extent_calc_inline_size(cur_size);
inode_add_bytes(inode, size);
ret = btrfs_insert_empty_item(trans, root, path, &key,
datasize);
BUG_ON(ret);
if (ret) {
err = ret;
goto fail;
}
leaf = path->nodes[0];
ei = btrfs_item_ptr(leaf, path->slots[0],
struct btrfs_file_extent_item);
btrfs_set_file_extent_generation(leaf, ei, trans->transid);
btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
btrfs_set_file_extent_encryption(leaf, ei, 0);
btrfs_set_file_extent_other_encoding(leaf, ei, 0);
btrfs_set_file_extent_ram_bytes(leaf, ei, size);
ptr = btrfs_file_extent_inline_start(ei);
if (compress_type != BTRFS_COMPRESS_NONE) {
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
struct page *cpage;
int i = 0;
while (compressed_size > 0) {
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
cpage = compressed_pages[i];
cur_size = min_t(unsigned long, compressed_size,
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
PAGE_CACHE_SIZE);
kaddr = kmap_atomic(cpage, KM_USER0);
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
write_extent_buffer(leaf, kaddr, ptr, cur_size);
kunmap_atomic(kaddr, KM_USER0);
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
i++;
ptr += cur_size;
compressed_size -= cur_size;
}
btrfs_set_file_extent_compression(leaf, ei,
compress_type);
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
} else {
page = find_get_page(inode->i_mapping,
start >> PAGE_CACHE_SHIFT);
btrfs_set_file_extent_compression(leaf, ei, 0);
kaddr = kmap_atomic(page, KM_USER0);
offset = start & (PAGE_CACHE_SIZE - 1);
write_extent_buffer(leaf, kaddr + offset, ptr, size);
kunmap_atomic(kaddr, KM_USER0);
page_cache_release(page);
}
btrfs_mark_buffer_dirty(leaf);
btrfs_free_path(path);
/*
* we're an inline extent, so nobody can
* extend the file past i_size without locking
* a page we already have locked.
*
* We must do any isize and inode updates
* before we unlock the pages. Otherwise we
* could end up racing with unlink.
*/
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
BTRFS_I(inode)->disk_i_size = inode->i_size;
btrfs_update_inode(trans, root, inode);
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
return 0;
fail:
btrfs_free_path(path);
return err;
}
/*
* conditionally insert an inline extent into the file. This
* does the checks required to make sure the data is small enough
* to fit as an inline extent.
*/
static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
struct btrfs_root *root,
struct inode *inode, u64 start, u64 end,
size_t compressed_size, int compress_type,
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
struct page **compressed_pages)
{
u64 isize = i_size_read(inode);
u64 actual_end = min(end + 1, isize);
u64 inline_len = actual_end - start;
u64 aligned_end = (end + root->sectorsize - 1) &
~((u64)root->sectorsize - 1);
u64 hint_byte;
u64 data_len = inline_len;
int ret;
if (compressed_size)
data_len = compressed_size;
if (start > 0 ||
actual_end >= PAGE_CACHE_SIZE ||
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
(!compressed_size &&
(actual_end & (root->sectorsize - 1)) == 0) ||
end + 1 < isize ||
data_len > root->fs_info->max_inline) {
return 1;
}
ret = btrfs_drop_extents(trans, inode, start, aligned_end,
&hint_byte, 1);
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
BUG_ON(ret);
if (isize > actual_end)
inline_len = min_t(u64, isize, actual_end);
ret = insert_inline_extent(trans, root, inode, start,
inline_len, compressed_size,
compress_type, compressed_pages);
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
BUG_ON(ret);
btrfs_delalloc_release_metadata(inode, end + 1 - start);
btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
return 0;
}
struct async_extent {
u64 start;
u64 ram_size;
u64 compressed_size;
struct page **pages;
unsigned long nr_pages;
int compress_type;
struct list_head list;
};
struct async_cow {
struct inode *inode;
struct btrfs_root *root;
struct page *locked_page;
u64 start;
u64 end;
struct list_head extents;
struct btrfs_work work;
};
static noinline int add_async_extent(struct async_cow *cow,
u64 start, u64 ram_size,
u64 compressed_size,
struct page **pages,
unsigned long nr_pages,
int compress_type)
{
struct async_extent *async_extent;
async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
BUG_ON(!async_extent);
async_extent->start = start;
async_extent->ram_size = ram_size;
async_extent->compressed_size = compressed_size;
async_extent->pages = pages;
async_extent->nr_pages = nr_pages;
async_extent->compress_type = compress_type;
list_add_tail(&async_extent->list, &cow->extents);
return 0;
}
/*
* we create compressed extents in two phases. The first
* phase compresses a range of pages that have already been
* locked (both pages and state bits are locked).
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
*
* This is done inside an ordered work queue, and the compression
* is spread across many cpus. The actual IO submission is step
* two, and the ordered work queue takes care of making sure that
* happens in the same order things were put onto the queue by
* writepages and friends.
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
*
* If this code finds it can't get good compression, it puts an
* entry onto the work queue to write the uncompressed bytes. This
* makes sure that both compressed inodes and uncompressed inodes
* are written in the same order that pdflush sent them down.
*/
static noinline int compress_file_range(struct inode *inode,
struct page *locked_page,
u64 start, u64 end,
struct async_cow *async_cow,
int *num_added)
{
struct btrfs_root *root = BTRFS_I(inode)->root;
struct btrfs_trans_handle *trans;
u64 num_bytes;
u64 blocksize = root->sectorsize;
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
u64 actual_end;
u64 isize = i_size_read(inode);
int ret = 0;
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
struct page **pages = NULL;
unsigned long nr_pages;
unsigned long nr_pages_ret = 0;
unsigned long total_compressed = 0;
unsigned long total_in = 0;
unsigned long max_compressed = 128 * 1024;
unsigned long max_uncompressed = 128 * 1024;
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
int i;
int will_compress;
int compress_type = root->fs_info->compress_type;
/* if this is a small write inside eof, kick off a defragbot */
if (end <= BTRFS_I(inode)->disk_i_size && (end - start + 1) < 16 * 1024)
btrfs_add_inode_defrag(NULL, inode);
actual_end = min_t(u64, isize, end + 1);
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
again:
will_compress = 0;
nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
/*
* we don't want to send crud past the end of i_size through
* compression, that's just a waste of CPU time. So, if the
* end of the file is before the start of our current
* requested range of bytes, we bail out to the uncompressed
* cleanup code that can deal with all of this.
*
* It isn't really the fastest way to fix things, but this is a
* very uncommon corner.
*/
if (actual_end <= start)
goto cleanup_and_bail_uncompressed;
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
total_compressed = actual_end - start;
/* we want to make sure that amount of ram required to uncompress
* an extent is reasonable, so we limit the total size in ram
* of a compressed extent to 128k. This is a crucial number
* because it also controls how easily we can spread reads across
* cpus for decompression.
*
* We also want to make sure the amount of IO required to do
* a random read is reasonably small, so we limit the size of
* a compressed extent to 128k.
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
*/
total_compressed = min(total_compressed, max_uncompressed);
num_bytes = (end - start + blocksize) & ~(blocksize - 1);
num_bytes = max(blocksize, num_bytes);
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
total_in = 0;
ret = 0;
/*
* we do compression for mount -o compress and when the
* inode has not been flagged as nocompress. This flag can
* change at any time if we discover bad compression ratios.
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
*/
if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
(btrfs_test_opt(root, COMPRESS) ||
(BTRFS_I(inode)->force_compress) ||
(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
WARN_ON(pages);
pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
if (!pages) {
/* just bail out to the uncompressed code */
goto cont;
}
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
if (BTRFS_I(inode)->force_compress)
compress_type = BTRFS_I(inode)->force_compress;
ret = btrfs_compress_pages(compress_type,
inode->i_mapping, start,
total_compressed, pages,
nr_pages, &nr_pages_ret,
&total_in,
&total_compressed,
max_compressed);
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
if (!ret) {
unsigned long offset = total_compressed &
(PAGE_CACHE_SIZE - 1);
struct page *page = pages[nr_pages_ret - 1];
char *kaddr;
/* zero the tail end of the last page, we might be
* sending it down to disk
*/
if (offset) {
kaddr = kmap_atomic(page, KM_USER0);
memset(kaddr + offset, 0,
PAGE_CACHE_SIZE - offset);
kunmap_atomic(kaddr, KM_USER0);
}
will_compress = 1;
}
}
cont:
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
if (start == 0) {
trans = btrfs_join_transaction(root);
BUG_ON(IS_ERR(trans));
trans->block_rsv = &root->fs_info->delalloc_block_rsv;
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
/* lets try to make an inline extent */
if (ret || total_in < (actual_end - start)) {
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
/* we didn't compress the entire range, try
* to make an uncompressed inline extent.
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
*/
ret = cow_file_range_inline(trans, root, inode,
start, end, 0, 0, NULL);
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
} else {
/* try making a compressed inline extent */
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
ret = cow_file_range_inline(trans, root, inode,
start, end,
total_compressed,
compress_type, pages);
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
}
if (ret == 0) {
/*
* inline extent creation worked, we don't need
* to create any more async work items. Unlock
* and free up our temp pages.
*/
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
extent_clear_unlock_delalloc(inode,
&BTRFS_I(inode)->io_tree,
start, end, NULL,
EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
EXTENT_CLEAR_DELALLOC |
EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
btrfs_end_transaction(trans, root);
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
goto free_pages_out;
}
btrfs_end_transaction(trans, root);
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
}
if (will_compress) {
/*
* we aren't doing an inline extent round the compressed size
* up to a block size boundary so the allocator does sane
* things
*/
total_compressed = (total_compressed + blocksize - 1) &
~(blocksize - 1);
/*
* one last check to make sure the compression is really a
* win, compare the page count read with the blocks on disk
*/
total_in = (total_in + PAGE_CACHE_SIZE - 1) &
~(PAGE_CACHE_SIZE - 1);
if (total_compressed >= total_in) {
will_compress = 0;
} else {
num_bytes = total_in;
}
}
if (!will_compress && pages) {
/*
* the compression code ran but failed to make things smaller,
* free any pages it allocated and our page pointer array
*/
for (i = 0; i < nr_pages_ret; i++) {
WARN_ON(pages[i]->mapping);
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
page_cache_release(pages[i]);
}
kfree(pages);
pages = NULL;
total_compressed = 0;
nr_pages_ret = 0;
/* flag the file so we don't compress in the future */
if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
!(BTRFS_I(inode)->force_compress)) {
BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
}
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
}
if (will_compress) {
*num_added += 1;
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
/* the async work queues will take care of doing actual
* allocation on disk for these compressed pages,
* and will submit them to the elevator.
*/
add_async_extent(async_cow, start, num_bytes,
total_compressed, pages, nr_pages_ret,
compress_type);
if (start + num_bytes < end) {
start += num_bytes;
pages = NULL;
cond_resched();
goto again;
}
} else {
cleanup_and_bail_uncompressed:
/*
* No compression, but we still need to write the pages in
* the file we've been given so far. redirty the locked
* page if it corresponds to our extent and set things up
* for the async work queue to run cow_file_range to do
* the normal delalloc dance
*/
if (page_offset(locked_page) >= start &&
page_offset(locked_page) <= end) {
__set_page_dirty_nobuffers(locked_page);
/* unlocked later on in the async handlers */
}
add_async_extent(async_cow, start, end - start + 1,
0, NULL, 0, BTRFS_COMPRESS_NONE);
*num_added += 1;
}
out:
return 0;
free_pages_out:
for (i = 0; i < nr_pages_ret; i++) {
WARN_ON(pages[i]->mapping);
page_cache_release(pages[i]);
}
kfree(pages);
goto out;
}
/*
* phase two of compressed writeback. This is the ordered portion
* of the code, which only gets called in the order the work was
* queued. We walk all the async extents created by compress_file_range
* and send them down to the disk.
*/
static noinline int submit_compressed_extents(struct inode *inode,
struct async_cow *async_cow)
{
struct async_extent *async_extent;
u64 alloc_hint = 0;
struct btrfs_trans_handle *trans;
struct btrfs_key ins;
struct extent_map *em;
struct btrfs_root *root = BTRFS_I(inode)->root;
struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
struct extent_io_tree *io_tree;
int ret = 0;
if (list_empty(&async_cow->extents))
return 0;
while (!list_empty(&async_cow->extents)) {
async_extent = list_entry(async_cow->extents.next,
struct async_extent, list);
list_del(&async_extent->list);
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
io_tree = &BTRFS_I(inode)->io_tree;
retry:
/* did the compression code fall back to uncompressed IO? */
if (!async_extent->pages) {
int page_started = 0;
unsigned long nr_written = 0;
lock_extent(io_tree, async_extent->start,
async_extent->start +
async_extent->ram_size - 1);
/* allocate blocks */
ret = cow_file_range(inode, async_cow->locked_page,
async_extent->start,
async_extent->start +
async_extent->ram_size - 1,
&page_started, &nr_written, 0);
/*
* if page_started, cow_file_range inserted an
* inline extent and took care of all the unlocking
* and IO for us. Otherwise, we need to submit
* all those pages down to the drive.
*/
if (!page_started && !ret)
extent_write_locked_range(io_tree,
inode, async_extent->start,
async_extent->start +
async_extent->ram_size - 1,
btrfs_get_extent,
WB_SYNC_ALL);
kfree(async_extent);
cond_resched();
continue;
}
lock_extent(io_tree, async_extent->start,
async_extent->start + async_extent->ram_size - 1);
trans = btrfs_join_transaction(root);
BUG_ON(IS_ERR(trans));
trans->block_rsv = &root->fs_info->delalloc_block_rsv;
ret = btrfs_reserve_extent(trans, root,
async_extent->compressed_size,
async_extent->compressed_size,
0, alloc_hint,
(u64)-1, &ins, 1);
btrfs_end_transaction(trans, root);
if (ret) {
int i;
for (i = 0; i < async_extent->nr_pages; i++) {
WARN_ON(async_extent->pages[i]->mapping);
page_cache_release(async_extent->pages[i]);
}
kfree(async_extent->pages);
async_extent->nr_pages = 0;
async_extent->pages = NULL;
unlock_extent(io_tree, async_extent->start,
async_extent->start +
async_extent->ram_size - 1);
goto retry;
}
/*
* here we're doing allocation and writeback of the
* compressed pages
*/
btrfs_drop_extent_cache(inode, async_extent->start,
async_extent->start +
async_extent->ram_size - 1, 0);
em = alloc_extent_map();
BUG_ON(!em);
em->start = async_extent->start;
em->len = async_extent->ram_size;
em->orig_start = em->start;
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
em->block_start = ins.objectid;
em->block_len = ins.offset;
em->bdev = root->fs_info->fs_devices->latest_bdev;
em->compress_type = async_extent->compress_type;
set_bit(EXTENT_FLAG_PINNED, &em->flags);
set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
while (1) {
write_lock(&em_tree->lock);
ret = add_extent_mapping(em_tree, em);
write_unlock(&em_tree->lock);
if (ret != -EEXIST) {
free_extent_map(em);
break;
}
btrfs_drop_extent_cache(inode, async_extent->start,
async_extent->start +
async_extent->ram_size - 1, 0);
}
ret = btrfs_add_ordered_extent_compress(inode,
async_extent->start,
ins.objectid,
async_extent->ram_size,
ins.offset,
BTRFS_ORDERED_COMPRESSED,
async_extent->compress_type);
BUG_ON(ret);
/*
* clear dirty, set writeback and unlock the pages.
*/
extent_clear_unlock_delalloc(inode,
&BTRFS_I(inode)->io_tree,
async_extent->start,
async_extent->start +
async_extent->ram_size - 1,
NULL, EXTENT_CLEAR_UNLOCK_PAGE |
EXTENT_CLEAR_UNLOCK |
EXTENT_CLEAR_DELALLOC |
EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
ret = btrfs_submit_compressed_write(inode,
async_extent->start,
async_extent->ram_size,
ins.objectid,
ins.offset, async_extent->pages,
async_extent->nr_pages);
BUG_ON(ret);
alloc_hint = ins.objectid + ins.offset;
kfree(async_extent);
cond_resched();
}
return 0;
}
static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
u64 num_bytes)
{
struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
struct extent_map *em;
u64 alloc_hint = 0;
read_lock(&em_tree->lock);
em = search_extent_mapping(em_tree, start, num_bytes);
if (em) {
/*
* if block start isn't an actual block number then find the
* first block in this inode and use that as a hint. If that
* block is also bogus then just don't worry about it.
*/
if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
free_extent_map(em);
em = search_extent_mapping(em_tree, 0, 0);
if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
alloc_hint = em->block_start;
if (em)
free_extent_map(em);
} else {
alloc_hint = em->block_start;
free_extent_map(em);
}
}
read_unlock(&em_tree->lock);
return alloc_hint;
}
/*
* when extent_io.c finds a delayed allocation range in the file,
* the call backs end up in this code. The basic idea is to
* allocate extents on disk for the range, and create ordered data structs
* in ram to track those extents.
*
* locked_page is the page that writepage had locked already. We use
* it to make sure we don't do extra locks or unlocks.
*
* *page_started is set to one if we unlock locked_page and do everything
* required to start IO on it. It may be clean and already done with
* IO when we return.
*/
static noinline int cow_file_range(struct inode *inode,
struct page *locked_page,
u64 start, u64 end, int *page_started,
unsigned long *nr_written,
int unlock)
{
struct btrfs_root *root = BTRFS_I(inode)->root;
struct btrfs_trans_handle *trans;
u64 alloc_hint = 0;
u64 num_bytes;
unsigned long ram_size;
u64 disk_num_bytes;
u64 cur_alloc_size;
u64 blocksize = root->sectorsize;
struct btrfs_key ins;
struct extent_map *em;
struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
int ret = 0;
BUG_ON(btrfs_is_free_space_inode(root, inode));
trans = btrfs_join_transaction(root);
BUG_ON(IS_ERR(trans));
trans->block_rsv = &root->fs_info->delalloc_block_rsv;
num_bytes = (end - start + blocksize) & ~(blocksize - 1);
num_bytes = max(blocksize, num_bytes);
disk_num_bytes = num_bytes;
ret = 0;
/* if this is a small write inside eof, kick off defrag */
if (end <= BTRFS_I(inode)->disk_i_size && num_bytes < 64 * 1024)
btrfs_add_inode_defrag(trans, inode);
if (start == 0) {
/* lets try to make an inline extent */
ret = cow_file_range_inline(trans, root, inode,
start, end, 0, 0, NULL);
if (ret == 0) {
extent_clear_unlock_delalloc(inode,
&BTRFS_I(inode)->io_tree,
start, end, NULL,
EXTENT_CLEAR_UNLOCK_PAGE |
EXTENT_CLEAR_UNLOCK |
EXTENT_CLEAR_DELALLOC |
EXTENT_CLEAR_DIRTY |
EXTENT_SET_WRITEBACK |
EXTENT_END_WRITEBACK);
*nr_written = *nr_written +
(end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
*page_started = 1;
ret = 0;
goto out;
}
}
BUG_ON(disk_num_bytes >
btrfs_super_total_bytes(root->fs_info->super_copy));
alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
while (disk_num_bytes > 0) {
unsigned long op;
cur_alloc_size = disk_num_bytes;
ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
root->sectorsize, 0, alloc_hint,
(u64)-1, &ins, 1);
BUG_ON(ret);
em = alloc_extent_map();
BUG_ON(!em);
em->start = start;
em->orig_start = em->start;
ram_size = ins.offset;
em->len = ins.offset;
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
em->block_start = ins.objectid;
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
em->block_len = ins.offset;
em->bdev = root->fs_info->fs_devices->latest_bdev;
set_bit(EXTENT_FLAG_PINNED, &em->flags);
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
while (1) {
write_lock(&em_tree->lock);
ret = add_extent_mapping(em_tree, em);
write_unlock(&em_tree->lock);
if (ret != -EEXIST) {
free_extent_map(em);
break;
}
btrfs_drop_extent_cache(inode, start,
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
start + ram_size - 1, 0);
}
cur_alloc_size = ins.offset;
ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
ram_size, cur_alloc_size, 0);
BUG_ON(ret);
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
if (root->root_key.objectid ==
BTRFS_DATA_RELOC_TREE_OBJECTID) {
ret = btrfs_reloc_clone_csums(inode, start,
cur_alloc_size);
BUG_ON(ret);
}
if (disk_num_bytes < cur_alloc_size)
break;
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
/* we're not doing compressed IO, don't unlock the first
* page (which the caller expects to stay locked), don't
* clear any dirty bits and don't set any writeback bits
*
* Do set the Private2 bit so we know this page was properly
* setup for writepage
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
*/
op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
EXTENT_SET_PRIVATE2;
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
start, start + ram_size - 1,
locked_page, op);
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
disk_num_bytes -= cur_alloc_size;
num_bytes -= cur_alloc_size;
alloc_hint = ins.objectid + ins.offset;
start += cur_alloc_size;
}
out:
ret = 0;
btrfs_end_transaction(trans, root);
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
return ret;
}
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
/*
* work queue call back to started compression on a file and pages
*/
static noinline void async_cow_start(struct btrfs_work *work)
{
struct async_cow *async_cow;
int num_added = 0;
async_cow = container_of(work, struct async_cow, work);
compress_file_range(async_cow->inode, async_cow->locked_page,
async_cow->start, async_cow->end, async_cow,
&num_added);
if (num_added == 0)
async_cow->inode = NULL;
}
/*
* work queue call back to submit previously compressed pages
*/
static noinline void async_cow_submit(struct btrfs_work *work)
{
struct async_cow *async_cow;
struct btrfs_root *root;
unsigned long nr_pages;
async_cow = container_of(work, struct async_cow, work);
root = async_cow->root;
nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
PAGE_CACHE_SHIFT;
atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
if (atomic_read(&root->fs_info->async_delalloc_pages) <
5 * 1042 * 1024 &&
waitqueue_active(&root->fs_info->async_submit_wait))
wake_up(&root->fs_info->async_submit_wait);
if (async_cow->inode)
submit_compressed_extents(async_cow->inode, async_cow);
}
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
static noinline void async_cow_free(struct btrfs_work *work)
{
struct async_cow *async_cow;
async_cow = container_of(work, struct async_cow, work);
kfree(async_cow);
}
static int cow_file_range_async(struct inode *inode, struct page *locked_page,
u64 start, u64 end, int *page_started,
unsigned long *nr_written)
{
struct async_cow *async_cow;
struct btrfs_root *root = BTRFS_I(inode)->root;
unsigned long nr_pages;
u64 cur_end;
int limit = 10 * 1024 * 1042;
clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1, 0, NULL, GFP_NOFS);
while (start < end) {
async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
BUG_ON(!async_cow);
async_cow->inode = inode;
async_cow->root = root;
async_cow->locked_page = locked_page;
async_cow->start = start;
if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
cur_end = end;
else
cur_end = min(end, start + 512 * 1024 - 1);
async_cow->end = cur_end;
INIT_LIST_HEAD(&async_cow->extents);
async_cow->work.func = async_cow_start;
async_cow->work.ordered_func = async_cow_submit;
async_cow->work.ordered_free = async_cow_free;
async_cow->work.flags = 0;
nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
PAGE_CACHE_SHIFT;
atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
btrfs_queue_worker(&root->fs_info->delalloc_workers,
&async_cow->work);
if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
wait_event(root->fs_info->async_submit_wait,
(atomic_read(&root->fs_info->async_delalloc_pages) <
limit));
}
while (atomic_read(&root->fs_info->async_submit_draining) &&
atomic_read(&root->fs_info->async_delalloc_pages)) {
wait_event(root->fs_info->async_submit_wait,
(atomic_read(&root->fs_info->async_delalloc_pages) ==
0));
}
*nr_written += nr_pages;
start = cur_end + 1;
}
*page_started = 1;
return 0;
}
static noinline int csum_exist_in_range(struct btrfs_root *root,
u64 bytenr, u64 num_bytes)
{
int ret;
struct btrfs_ordered_sum *sums;
LIST_HEAD(list);
ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
bytenr + num_bytes - 1, &list, 0);
if (ret == 0 && list_empty(&list))
return 0;
while (!list_empty(&list)) {
sums = list_entry(list.next, struct btrfs_ordered_sum, list);
list_del(&sums->list);
kfree(sums);
}
return 1;
}
/*
* when nowcow writeback call back. This checks for snapshots or COW copies
* of the extents that exist in the file, and COWs the file as required.
*
* If no cow copies or snapshots exist, we write directly to the existing
* blocks on disk
*/
static noinline int run_delalloc_nocow(struct inode *inode,
struct page *locked_page,
u64 start, u64 end, int *page_started, int force,
unsigned long *nr_written)
{
struct btrfs_root *root = BTRFS_I(inode)->root;
struct btrfs_trans_handle *trans;
struct extent_buffer *leaf;
struct btrfs_path *path;
struct btrfs_file_extent_item *fi;
struct btrfs_key found_key;
u64 cow_start;
u64 cur_offset;
u64 extent_end;
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 14:45:14 +00:00
u64 extent_offset;
u64 disk_bytenr;
u64 num_bytes;
int extent_type;
int ret;
int type;
int nocow;
int check_prev = 1;
bool nolock;
u64 ino = btrfs_ino(inode);
path = btrfs_alloc_path();
if (!path)
return -ENOMEM;
nolock = btrfs_is_free_space_inode(root, inode);
if (nolock)
trans = btrfs_join_transaction_nolock(root);
else
trans = btrfs_join_transaction(root);
BUG_ON(IS_ERR(trans));
trans->block_rsv = &root->fs_info->delalloc_block_rsv;
cow_start = (u64)-1;
cur_offset = start;
while (1) {
ret = btrfs_lookup_file_extent(trans, root, path, ino,
cur_offset, 0);
BUG_ON(ret < 0);
if (ret > 0 && path->slots[0] > 0 && check_prev) {
leaf = path->nodes[0];
btrfs_item_key_to_cpu(leaf, &found_key,
path->slots[0] - 1);
if (found_key.objectid == ino &&
found_key.type == BTRFS_EXTENT_DATA_KEY)
path->slots[0]--;
}
check_prev = 0;
next_slot:
leaf = path->nodes[0];
if (path->slots[0] >= btrfs_header_nritems(leaf)) {
ret = btrfs_next_leaf(root, path);
if (ret < 0)
BUG_ON(1);
if (ret > 0)
break;
leaf = path->nodes[0];
}
nocow = 0;
disk_bytenr = 0;
num_bytes = 0;
btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
if (found_key.objectid > ino ||
found_key.type > BTRFS_EXTENT_DATA_KEY ||
found_key.offset > end)
break;
if (found_key.offset > cur_offset) {
extent_end = found_key.offset;
extent_type = 0;
goto out_check;
}
fi = btrfs_item_ptr(leaf, path->slots[0],
struct btrfs_file_extent_item);
extent_type = btrfs_file_extent_type(leaf, fi);
if (extent_type == BTRFS_FILE_EXTENT_REG ||
extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 14:45:14 +00:00
extent_offset = btrfs_file_extent_offset(leaf, fi);
extent_end = found_key.offset +
btrfs_file_extent_num_bytes(leaf, fi);
if (extent_end <= start) {
path->slots[0]++;
goto next_slot;
}
if (disk_bytenr == 0)
goto out_check;
if (btrfs_file_extent_compression(leaf, fi) ||
btrfs_file_extent_encryption(leaf, fi) ||
btrfs_file_extent_other_encoding(leaf, fi))
goto out_check;
if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
goto out_check;
if (btrfs_extent_readonly(root, disk_bytenr))
goto out_check;
if (btrfs_cross_ref_exist(trans, root, ino,
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 14:45:14 +00:00
found_key.offset -
extent_offset, disk_bytenr))
goto out_check;
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 14:45:14 +00:00
disk_bytenr += extent_offset;
disk_bytenr += cur_offset - found_key.offset;
num_bytes = min(end + 1, extent_end) - cur_offset;
/*
* force cow if csum exists in the range.
* this ensure that csum for a given extent are
* either valid or do not exist.
*/
if (csum_exist_in_range(root, disk_bytenr, num_bytes))
goto out_check;
nocow = 1;
} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
extent_end = found_key.offset +
btrfs_file_extent_inline_len(leaf, fi);
extent_end = ALIGN(extent_end, root->sectorsize);
} else {
BUG_ON(1);
}
out_check:
if (extent_end <= start) {
path->slots[0]++;
goto next_slot;
}
if (!nocow) {
if (cow_start == (u64)-1)
cow_start = cur_offset;
cur_offset = extent_end;
if (cur_offset > end)
break;
path->slots[0]++;
goto next_slot;
}
btrfs_release_path(path);
if (cow_start != (u64)-1) {
ret = cow_file_range(inode, locked_page, cow_start,
found_key.offset - 1, page_started,
nr_written, 1);
BUG_ON(ret);
cow_start = (u64)-1;
}
if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
struct extent_map *em;
struct extent_map_tree *em_tree;
em_tree = &BTRFS_I(inode)->extent_tree;
em = alloc_extent_map();
BUG_ON(!em);
em->start = cur_offset;
em->orig_start = em->start;
em->len = num_bytes;
em->block_len = num_bytes;
em->block_start = disk_bytenr;
em->bdev = root->fs_info->fs_devices->latest_bdev;
set_bit(EXTENT_FLAG_PINNED, &em->flags);
while (1) {
write_lock(&em_tree->lock);
ret = add_extent_mapping(em_tree, em);
write_unlock(&em_tree->lock);
if (ret != -EEXIST) {
free_extent_map(em);
break;
}
btrfs_drop_extent_cache(inode, em->start,
em->start + em->len - 1, 0);
}
type = BTRFS_ORDERED_PREALLOC;
} else {
type = BTRFS_ORDERED_NOCOW;
}
ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
num_bytes, num_bytes, type);
BUG_ON(ret);
if (root->root_key.objectid ==
BTRFS_DATA_RELOC_TREE_OBJECTID) {
ret = btrfs_reloc_clone_csums(inode, cur_offset,
num_bytes);
BUG_ON(ret);
}
extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
cur_offset, cur_offset + num_bytes - 1,
locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
EXTENT_SET_PRIVATE2);
cur_offset = extent_end;
if (cur_offset > end)
break;
}
btrfs_release_path(path);
if (cur_offset <= end && cow_start == (u64)-1)
cow_start = cur_offset;
if (cow_start != (u64)-1) {
ret = cow_file_range(inode, locked_page, cow_start, end,
page_started, nr_written, 1);
BUG_ON(ret);
}
if (nolock) {
ret = btrfs_end_transaction_nolock(trans, root);
BUG_ON(ret);
} else {
ret = btrfs_end_transaction(trans, root);
BUG_ON(ret);
}
btrfs_free_path(path);
return 0;
}
/*
* extent_io.c call back to do delayed allocation processing
*/
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
static int run_delalloc_range(struct inode *inode, struct page *locked_page,
u64 start, u64 end, int *page_started,
unsigned long *nr_written)
{
int ret;
struct btrfs_root *root = BTRFS_I(inode)->root;
if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
ret = run_delalloc_nocow(inode, locked_page, start, end,
page_started, 1, nr_written);
else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
ret = run_delalloc_nocow(inode, locked_page, start, end,
page_started, 0, nr_written);
else if (!btrfs_test_opt(root, COMPRESS) &&
!(BTRFS_I(inode)->force_compress) &&
!(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
ret = cow_file_range(inode, locked_page, start, end,
page_started, nr_written, 1);
else
ret = cow_file_range_async(inode, locked_page, start, end,
page_started, nr_written);
return ret;
}
static void btrfs_split_extent_hook(struct inode *inode,
struct extent_state *orig, u64 split)
Btrfs: proper -ENOSPC handling At the start of a transaction we do a btrfs_reserve_metadata_space() and specify how many items we plan on modifying. Then once we've done our modifications and such, just call btrfs_unreserve_metadata_space() for the same number of items we reserved. For keeping track of metadata needed for data I've had to add an extent_io op for when we merge extents. This lets us track space properly when we are doing sequential writes, so we don't end up reserving way more metadata space than what we need. The only place where the metadata space accounting is not done is in the relocation code. This is because Yan is going to be reworking that code in the near future, so running btrfs-vol -b could still possibly result in a ENOSPC related panic. This patch also turns off the metadata_ratio stuff in order to allow users to more efficiently use their disk space. This patch makes it so we track how much metadata we need for an inode's delayed allocation extents by tracking how many extents are currently waiting for allocation. It introduces two new callbacks for the extent_io tree's, merge_extent_hook and split_extent_hook. These help us keep track of when we merge delalloc extents together and split them up. Reservations are handled prior to any actually dirty'ing occurs, and then we unreserve after we dirty. btrfs_unreserve_metadata_for_delalloc() will make the appropriate unreservations as needed based on the number of reservations we currently have and the number of extents we currently have. Doing the reservation outside of doing any of the actual dirty'ing lets us do things like filemap_flush() the inode to try and force delalloc to happen, or as a last resort actually start allocation on all delalloc inodes in the fs. This has survived dbench, fs_mark and an fsx torture test. Signed-off-by: Josef Bacik <jbacik@redhat.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-09-11 20:12:44 +00:00
{
/* not delalloc, ignore it */
Btrfs: proper -ENOSPC handling At the start of a transaction we do a btrfs_reserve_metadata_space() and specify how many items we plan on modifying. Then once we've done our modifications and such, just call btrfs_unreserve_metadata_space() for the same number of items we reserved. For keeping track of metadata needed for data I've had to add an extent_io op for when we merge extents. This lets us track space properly when we are doing sequential writes, so we don't end up reserving way more metadata space than what we need. The only place where the metadata space accounting is not done is in the relocation code. This is because Yan is going to be reworking that code in the near future, so running btrfs-vol -b could still possibly result in a ENOSPC related panic. This patch also turns off the metadata_ratio stuff in order to allow users to more efficiently use their disk space. This patch makes it so we track how much metadata we need for an inode's delayed allocation extents by tracking how many extents are currently waiting for allocation. It introduces two new callbacks for the extent_io tree's, merge_extent_hook and split_extent_hook. These help us keep track of when we merge delalloc extents together and split them up. Reservations are handled prior to any actually dirty'ing occurs, and then we unreserve after we dirty. btrfs_unreserve_metadata_for_delalloc() will make the appropriate unreservations as needed based on the number of reservations we currently have and the number of extents we currently have. Doing the reservation outside of doing any of the actual dirty'ing lets us do things like filemap_flush() the inode to try and force delalloc to happen, or as a last resort actually start allocation on all delalloc inodes in the fs. This has survived dbench, fs_mark and an fsx torture test. Signed-off-by: Josef Bacik <jbacik@redhat.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-09-11 20:12:44 +00:00
if (!(orig->state & EXTENT_DELALLOC))
return;
Btrfs: proper -ENOSPC handling At the start of a transaction we do a btrfs_reserve_metadata_space() and specify how many items we plan on modifying. Then once we've done our modifications and such, just call btrfs_unreserve_metadata_space() for the same number of items we reserved. For keeping track of metadata needed for data I've had to add an extent_io op for when we merge extents. This lets us track space properly when we are doing sequential writes, so we don't end up reserving way more metadata space than what we need. The only place where the metadata space accounting is not done is in the relocation code. This is because Yan is going to be reworking that code in the near future, so running btrfs-vol -b could still possibly result in a ENOSPC related panic. This patch also turns off the metadata_ratio stuff in order to allow users to more efficiently use their disk space. This patch makes it so we track how much metadata we need for an inode's delayed allocation extents by tracking how many extents are currently waiting for allocation. It introduces two new callbacks for the extent_io tree's, merge_extent_hook and split_extent_hook. These help us keep track of when we merge delalloc extents together and split them up. Reservations are handled prior to any actually dirty'ing occurs, and then we unreserve after we dirty. btrfs_unreserve_metadata_for_delalloc() will make the appropriate unreservations as needed based on the number of reservations we currently have and the number of extents we currently have. Doing the reservation outside of doing any of the actual dirty'ing lets us do things like filemap_flush() the inode to try and force delalloc to happen, or as a last resort actually start allocation on all delalloc inodes in the fs. This has survived dbench, fs_mark and an fsx torture test. Signed-off-by: Josef Bacik <jbacik@redhat.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-09-11 20:12:44 +00:00
spin_lock(&BTRFS_I(inode)->lock);
BTRFS_I(inode)->outstanding_extents++;
spin_unlock(&BTRFS_I(inode)->lock);
Btrfs: proper -ENOSPC handling At the start of a transaction we do a btrfs_reserve_metadata_space() and specify how many items we plan on modifying. Then once we've done our modifications and such, just call btrfs_unreserve_metadata_space() for the same number of items we reserved. For keeping track of metadata needed for data I've had to add an extent_io op for when we merge extents. This lets us track space properly when we are doing sequential writes, so we don't end up reserving way more metadata space than what we need. The only place where the metadata space accounting is not done is in the relocation code. This is because Yan is going to be reworking that code in the near future, so running btrfs-vol -b could still possibly result in a ENOSPC related panic. This patch also turns off the metadata_ratio stuff in order to allow users to more efficiently use their disk space. This patch makes it so we track how much metadata we need for an inode's delayed allocation extents by tracking how many extents are currently waiting for allocation. It introduces two new callbacks for the extent_io tree's, merge_extent_hook and split_extent_hook. These help us keep track of when we merge delalloc extents together and split them up. Reservations are handled prior to any actually dirty'ing occurs, and then we unreserve after we dirty. btrfs_unreserve_metadata_for_delalloc() will make the appropriate unreservations as needed based on the number of reservations we currently have and the number of extents we currently have. Doing the reservation outside of doing any of the actual dirty'ing lets us do things like filemap_flush() the inode to try and force delalloc to happen, or as a last resort actually start allocation on all delalloc inodes in the fs. This has survived dbench, fs_mark and an fsx torture test. Signed-off-by: Josef Bacik <jbacik@redhat.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-09-11 20:12:44 +00:00
}
/*
* extent_io.c merge_extent_hook, used to track merged delayed allocation
* extents so we can keep track of new extents that are just merged onto old
* extents, such as when we are doing sequential writes, so we can properly
* account for the metadata space we'll need.
*/
static void btrfs_merge_extent_hook(struct inode *inode,
struct extent_state *new,
struct extent_state *other)
Btrfs: proper -ENOSPC handling At the start of a transaction we do a btrfs_reserve_metadata_space() and specify how many items we plan on modifying. Then once we've done our modifications and such, just call btrfs_unreserve_metadata_space() for the same number of items we reserved. For keeping track of metadata needed for data I've had to add an extent_io op for when we merge extents. This lets us track space properly when we are doing sequential writes, so we don't end up reserving way more metadata space than what we need. The only place where the metadata space accounting is not done is in the relocation code. This is because Yan is going to be reworking that code in the near future, so running btrfs-vol -b could still possibly result in a ENOSPC related panic. This patch also turns off the metadata_ratio stuff in order to allow users to more efficiently use their disk space. This patch makes it so we track how much metadata we need for an inode's delayed allocation extents by tracking how many extents are currently waiting for allocation. It introduces two new callbacks for the extent_io tree's, merge_extent_hook and split_extent_hook. These help us keep track of when we merge delalloc extents together and split them up. Reservations are handled prior to any actually dirty'ing occurs, and then we unreserve after we dirty. btrfs_unreserve_metadata_for_delalloc() will make the appropriate unreservations as needed based on the number of reservations we currently have and the number of extents we currently have. Doing the reservation outside of doing any of the actual dirty'ing lets us do things like filemap_flush() the inode to try and force delalloc to happen, or as a last resort actually start allocation on all delalloc inodes in the fs. This has survived dbench, fs_mark and an fsx torture test. Signed-off-by: Josef Bacik <jbacik@redhat.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-09-11 20:12:44 +00:00
{
/* not delalloc, ignore it */
if (!(other->state & EXTENT_DELALLOC))
return;
Btrfs: proper -ENOSPC handling At the start of a transaction we do a btrfs_reserve_metadata_space() and specify how many items we plan on modifying. Then once we've done our modifications and such, just call btrfs_unreserve_metadata_space() for the same number of items we reserved. For keeping track of metadata needed for data I've had to add an extent_io op for when we merge extents. This lets us track space properly when we are doing sequential writes, so we don't end up reserving way more metadata space than what we need. The only place where the metadata space accounting is not done is in the relocation code. This is because Yan is going to be reworking that code in the near future, so running btrfs-vol -b could still possibly result in a ENOSPC related panic. This patch also turns off the metadata_ratio stuff in order to allow users to more efficiently use their disk space. This patch makes it so we track how much metadata we need for an inode's delayed allocation extents by tracking how many extents are currently waiting for allocation. It introduces two new callbacks for the extent_io tree's, merge_extent_hook and split_extent_hook. These help us keep track of when we merge delalloc extents together and split them up. Reservations are handled prior to any actually dirty'ing occurs, and then we unreserve after we dirty. btrfs_unreserve_metadata_for_delalloc() will make the appropriate unreservations as needed based on the number of reservations we currently have and the number of extents we currently have. Doing the reservation outside of doing any of the actual dirty'ing lets us do things like filemap_flush() the inode to try and force delalloc to happen, or as a last resort actually start allocation on all delalloc inodes in the fs. This has survived dbench, fs_mark and an fsx torture test. Signed-off-by: Josef Bacik <jbacik@redhat.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-09-11 20:12:44 +00:00
spin_lock(&BTRFS_I(inode)->lock);
BTRFS_I(inode)->outstanding_extents--;
spin_unlock(&BTRFS_I(inode)->lock);
Btrfs: proper -ENOSPC handling At the start of a transaction we do a btrfs_reserve_metadata_space() and specify how many items we plan on modifying. Then once we've done our modifications and such, just call btrfs_unreserve_metadata_space() for the same number of items we reserved. For keeping track of metadata needed for data I've had to add an extent_io op for when we merge extents. This lets us track space properly when we are doing sequential writes, so we don't end up reserving way more metadata space than what we need. The only place where the metadata space accounting is not done is in the relocation code. This is because Yan is going to be reworking that code in the near future, so running btrfs-vol -b could still possibly result in a ENOSPC related panic. This patch also turns off the metadata_ratio stuff in order to allow users to more efficiently use their disk space. This patch makes it so we track how much metadata we need for an inode's delayed allocation extents by tracking how many extents are currently waiting for allocation. It introduces two new callbacks for the extent_io tree's, merge_extent_hook and split_extent_hook. These help us keep track of when we merge delalloc extents together and split them up. Reservations are handled prior to any actually dirty'ing occurs, and then we unreserve after we dirty. btrfs_unreserve_metadata_for_delalloc() will make the appropriate unreservations as needed based on the number of reservations we currently have and the number of extents we currently have. Doing the reservation outside of doing any of the actual dirty'ing lets us do things like filemap_flush() the inode to try and force delalloc to happen, or as a last resort actually start allocation on all delalloc inodes in the fs. This has survived dbench, fs_mark and an fsx torture test. Signed-off-by: Josef Bacik <jbacik@redhat.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-09-11 20:12:44 +00:00
}
/*
* extent_io.c set_bit_hook, used to track delayed allocation
* bytes in this file, and to maintain the list of inodes that
* have pending delalloc work to be done.
*/
static void btrfs_set_bit_hook(struct inode *inode,
struct extent_state *state, int *bits)
{
Btrfs: proper -ENOSPC handling At the start of a transaction we do a btrfs_reserve_metadata_space() and specify how many items we plan on modifying. Then once we've done our modifications and such, just call btrfs_unreserve_metadata_space() for the same number of items we reserved. For keeping track of metadata needed for data I've had to add an extent_io op for when we merge extents. This lets us track space properly when we are doing sequential writes, so we don't end up reserving way more metadata space than what we need. The only place where the metadata space accounting is not done is in the relocation code. This is because Yan is going to be reworking that code in the near future, so running btrfs-vol -b could still possibly result in a ENOSPC related panic. This patch also turns off the metadata_ratio stuff in order to allow users to more efficiently use their disk space. This patch makes it so we track how much metadata we need for an inode's delayed allocation extents by tracking how many extents are currently waiting for allocation. It introduces two new callbacks for the extent_io tree's, merge_extent_hook and split_extent_hook. These help us keep track of when we merge delalloc extents together and split them up. Reservations are handled prior to any actually dirty'ing occurs, and then we unreserve after we dirty. btrfs_unreserve_metadata_for_delalloc() will make the appropriate unreservations as needed based on the number of reservations we currently have and the number of extents we currently have. Doing the reservation outside of doing any of the actual dirty'ing lets us do things like filemap_flush() the inode to try and force delalloc to happen, or as a last resort actually start allocation on all delalloc inodes in the fs. This has survived dbench, fs_mark and an fsx torture test. Signed-off-by: Josef Bacik <jbacik@redhat.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-09-11 20:12:44 +00:00
/*
* set_bit and clear bit hooks normally require _irqsave/restore
* but in this case, we are only testing for the DELALLOC
* bit, which is only set or cleared with irqs on
*/
if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
struct btrfs_root *root = BTRFS_I(inode)->root;
u64 len = state->end + 1 - state->start;
bool do_list = !btrfs_is_free_space_inode(root, inode);
Btrfs: proper -ENOSPC handling At the start of a transaction we do a btrfs_reserve_metadata_space() and specify how many items we plan on modifying. Then once we've done our modifications and such, just call btrfs_unreserve_metadata_space() for the same number of items we reserved. For keeping track of metadata needed for data I've had to add an extent_io op for when we merge extents. This lets us track space properly when we are doing sequential writes, so we don't end up reserving way more metadata space than what we need. The only place where the metadata space accounting is not done is in the relocation code. This is because Yan is going to be reworking that code in the near future, so running btrfs-vol -b could still possibly result in a ENOSPC related panic. This patch also turns off the metadata_ratio stuff in order to allow users to more efficiently use their disk space. This patch makes it so we track how much metadata we need for an inode's delayed allocation extents by tracking how many extents are currently waiting for allocation. It introduces two new callbacks for the extent_io tree's, merge_extent_hook and split_extent_hook. These help us keep track of when we merge delalloc extents together and split them up. Reservations are handled prior to any actually dirty'ing occurs, and then we unreserve after we dirty. btrfs_unreserve_metadata_for_delalloc() will make the appropriate unreservations as needed based on the number of reservations we currently have and the number of extents we currently have. Doing the reservation outside of doing any of the actual dirty'ing lets us do things like filemap_flush() the inode to try and force delalloc to happen, or as a last resort actually start allocation on all delalloc inodes in the fs. This has survived dbench, fs_mark and an fsx torture test. Signed-off-by: Josef Bacik <jbacik@redhat.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-09-11 20:12:44 +00:00
if (*bits & EXTENT_FIRST_DELALLOC) {
*bits &= ~EXTENT_FIRST_DELALLOC;
} else {
spin_lock(&BTRFS_I(inode)->lock);
BTRFS_I(inode)->outstanding_extents++;
spin_unlock(&BTRFS_I(inode)->lock);
}
spin_lock(&root->fs_info->delalloc_lock);
BTRFS_I(inode)->delalloc_bytes += len;
root->fs_info->delalloc_bytes += len;
if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
&root->fs_info->delalloc_inodes);
}
spin_unlock(&root->fs_info->delalloc_lock);
}
}
/*
* extent_io.c clear_bit_hook, see set_bit_hook for why
*/
static void btrfs_clear_bit_hook(struct inode *inode,
struct extent_state *state, int *bits)
{
/*
* set_bit and clear bit hooks normally require _irqsave/restore
* but in this case, we are only testing for the DELALLOC
* bit, which is only set or cleared with irqs on
*/
if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
struct btrfs_root *root = BTRFS_I(inode)->root;
u64 len = state->end + 1 - state->start;
bool do_list = !btrfs_is_free_space_inode(root, inode);
if (*bits & EXTENT_FIRST_DELALLOC) {
*bits &= ~EXTENT_FIRST_DELALLOC;
} else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
spin_lock(&BTRFS_I(inode)->lock);
BTRFS_I(inode)->outstanding_extents--;
spin_unlock(&BTRFS_I(inode)->lock);
}
if (*bits & EXTENT_DO_ACCOUNTING)
btrfs_delalloc_release_metadata(inode, len);
if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
&& do_list)
btrfs_free_reserved_data_space(inode, len);
Btrfs: proper -ENOSPC handling At the start of a transaction we do a btrfs_reserve_metadata_space() and specify how many items we plan on modifying. Then once we've done our modifications and such, just call btrfs_unreserve_metadata_space() for the same number of items we reserved. For keeping track of metadata needed for data I've had to add an extent_io op for when we merge extents. This lets us track space properly when we are doing sequential writes, so we don't end up reserving way more metadata space than what we need. The only place where the metadata space accounting is not done is in the relocation code. This is because Yan is going to be reworking that code in the near future, so running btrfs-vol -b could still possibly result in a ENOSPC related panic. This patch also turns off the metadata_ratio stuff in order to allow users to more efficiently use their disk space. This patch makes it so we track how much metadata we need for an inode's delayed allocation extents by tracking how many extents are currently waiting for allocation. It introduces two new callbacks for the extent_io tree's, merge_extent_hook and split_extent_hook. These help us keep track of when we merge delalloc extents together and split them up. Reservations are handled prior to any actually dirty'ing occurs, and then we unreserve after we dirty. btrfs_unreserve_metadata_for_delalloc() will make the appropriate unreservations as needed based on the number of reservations we currently have and the number of extents we currently have. Doing the reservation outside of doing any of the actual dirty'ing lets us do things like filemap_flush() the inode to try and force delalloc to happen, or as a last resort actually start allocation on all delalloc inodes in the fs. This has survived dbench, fs_mark and an fsx torture test. Signed-off-by: Josef Bacik <jbacik@redhat.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-09-11 20:12:44 +00:00
spin_lock(&root->fs_info->delalloc_lock);
root->fs_info->delalloc_bytes -= len;
BTRFS_I(inode)->delalloc_bytes -= len;
if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
list_del_init(&BTRFS_I(inode)->delalloc_inodes);
}
spin_unlock(&root->fs_info->delalloc_lock);
}
}
/*
* extent_io.c merge_bio_hook, this must check the chunk tree to make sure
* we don't create bios that span stripes or chunks
*/
int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
size_t size, struct bio *bio,
unsigned long bio_flags)
{
struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
struct btrfs_mapping_tree *map_tree;
u64 logical = (u64)bio->bi_sector << 9;
u64 length = 0;
u64 map_length;
int ret;
if (bio_flags & EXTENT_BIO_COMPRESSED)
return 0;
length = bio->bi_size;
map_tree = &root->fs_info->mapping_tree;
map_length = length;
ret = btrfs_map_block(map_tree, READ, logical,
&map_length, NULL, 0);
/* Will always return 0 or 1 with map_multi == NULL */
BUG_ON(ret < 0);
if (map_length < length + size)
return 1;
return 0;
}
/*
* in order to insert checksums into the metadata in large chunks,
* we wait until bio submission time. All the pages in the bio are
* checksummed and sums are attached onto the ordered extent record.
*
* At IO completion time the cums attached on the ordered extent record
* are inserted into the btree
*/
static int __btrfs_submit_bio_start(struct inode *inode, int rw,
struct bio *bio, int mirror_num,
unsigned long bio_flags,
u64 bio_offset)
{
struct btrfs_root *root = BTRFS_I(inode)->root;
int ret = 0;
Btrfs: move data checksumming into a dedicated tree Btrfs stores checksums for each data block. Until now, they have been stored in the subvolume trees, indexed by the inode that is referencing the data block. This means that when we read the inode, we've probably read in at least some checksums as well. But, this has a few problems: * The checksums are indexed by logical offset in the file. When compression is on, this means we have to do the expensive checksumming on the uncompressed data. It would be faster if we could checksum the compressed data instead. * If we implement encryption, we'll be checksumming the plain text and storing that on disk. This is significantly less secure. * For either compression or encryption, we have to get the plain text back before we can verify the checksum as correct. This makes the raid layer balancing and extent moving much more expensive. * It makes the front end caching code more complex, as we have touch the subvolume and inodes as we cache extents. * There is potentitally one copy of the checksum in each subvolume referencing an extent. The solution used here is to store the extent checksums in a dedicated tree. This allows us to index the checksums by phyiscal extent start and length. It means: * The checksum is against the data stored on disk, after any compression or encryption is done. * The checksum is stored in a central location, and can be verified without following back references, or reading inodes. This makes compression significantly faster by reducing the amount of data that needs to be checksummed. It will also allow much faster raid management code in general. The checksums are indexed by a key with a fixed objectid (a magic value in ctree.h) and offset set to the starting byte of the extent. This allows us to copy the checksum items into the fsync log tree directly (or any other tree), without having to invent a second format for them. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-08 21:58:54 +00:00
ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
BUG_ON(ret);
Btrfs: Add ordered async work queues Btrfs uses kernel threads to create async work queues for cpu intensive operations such as checksumming and decompression. These work well, but they make it difficult to keep IO order intact. A single writepages call from pdflush or fsync will turn into a number of bios, and each bio is checksummed in parallel. Once the checksum is computed, the bio is sent down to the disk, and since we don't control the order in which the parallel operations happen, they might go down to the disk in almost any order. The code deals with this somewhat by having deep work queues for a single kernel thread, making it very likely that a single thread will process all the bios for a single inode. This patch introduces an explicitly ordered work queue. As work structs are placed into the queue they are put onto the tail of a list. They have three callbacks: ->func (cpu intensive processing here) ->ordered_func (order sensitive processing here) ->ordered_free (free the work struct, all processing is done) The work struct has three callbacks. The func callback does the cpu intensive work, and when it completes the work struct is marked as done. Every time a work struct completes, the list is checked to see if the head is marked as done. If so the ordered_func callback is used to do the order sensitive processing and the ordered_free callback is used to do any cleanup. Then we loop back and check the head of the list again. This patch also changes the checksumming code to use the ordered workqueues. One a 4 drive array, it increases streaming writes from 280MB/s to 350MB/s. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-11-07 03:03:00 +00:00
return 0;
}
Btrfs: Add ordered async work queues Btrfs uses kernel threads to create async work queues for cpu intensive operations such as checksumming and decompression. These work well, but they make it difficult to keep IO order intact. A single writepages call from pdflush or fsync will turn into a number of bios, and each bio is checksummed in parallel. Once the checksum is computed, the bio is sent down to the disk, and since we don't control the order in which the parallel operations happen, they might go down to the disk in almost any order. The code deals with this somewhat by having deep work queues for a single kernel thread, making it very likely that a single thread will process all the bios for a single inode. This patch introduces an explicitly ordered work queue. As work structs are placed into the queue they are put onto the tail of a list. They have three callbacks: ->func (cpu intensive processing here) ->ordered_func (order sensitive processing here) ->ordered_free (free the work struct, all processing is done) The work struct has three callbacks. The func callback does the cpu intensive work, and when it completes the work struct is marked as done. Every time a work struct completes, the list is checked to see if the head is marked as done. If so the ordered_func callback is used to do the order sensitive processing and the ordered_free callback is used to do any cleanup. Then we loop back and check the head of the list again. This patch also changes the checksumming code to use the ordered workqueues. One a 4 drive array, it increases streaming writes from 280MB/s to 350MB/s. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-11-07 03:03:00 +00:00
/*
* in order to insert checksums into the metadata in large chunks,
* we wait until bio submission time. All the pages in the bio are
* checksummed and sums are attached onto the ordered extent record.
*
* At IO completion time the cums attached on the ordered extent record
* are inserted into the btree
*/
static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
int mirror_num, unsigned long bio_flags,
u64 bio_offset)
Btrfs: Add ordered async work queues Btrfs uses kernel threads to create async work queues for cpu intensive operations such as checksumming and decompression. These work well, but they make it difficult to keep IO order intact. A single writepages call from pdflush or fsync will turn into a number of bios, and each bio is checksummed in parallel. Once the checksum is computed, the bio is sent down to the disk, and since we don't control the order in which the parallel operations happen, they might go down to the disk in almost any order. The code deals with this somewhat by having deep work queues for a single kernel thread, making it very likely that a single thread will process all the bios for a single inode. This patch introduces an explicitly ordered work queue. As work structs are placed into the queue they are put onto the tail of a list. They have three callbacks: ->func (cpu intensive processing here) ->ordered_func (order sensitive processing here) ->ordered_free (free the work struct, all processing is done) The work struct has three callbacks. The func callback does the cpu intensive work, and when it completes the work struct is marked as done. Every time a work struct completes, the list is checked to see if the head is marked as done. If so the ordered_func callback is used to do the order sensitive processing and the ordered_free callback is used to do any cleanup. Then we loop back and check the head of the list again. This patch also changes the checksumming code to use the ordered workqueues. One a 4 drive array, it increases streaming writes from 280MB/s to 350MB/s. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-11-07 03:03:00 +00:00
{
struct btrfs_root *root = BTRFS_I(inode)->root;
return btrfs_map_bio(root, rw, bio, mirror_num, 1);
}
/*
* extent_io.c submission hook. This does the right thing for csum calculation
* on write, or reading the csums from the tree before a read
*/
static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
int mirror_num, unsigned long bio_flags,
u64 bio_offset)
{
struct btrfs_root *root = BTRFS_I(inode)->root;
int ret = 0;
int skip_sum;
int metadata = 0;
skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
if (btrfs_is_free_space_inode(root, inode))
metadata = 2;
ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
if (ret)
return ret;
if (!(rw & REQ_WRITE)) {
Btrfs: move data checksumming into a dedicated tree Btrfs stores checksums for each data block. Until now, they have been stored in the subvolume trees, indexed by the inode that is referencing the data block. This means that when we read the inode, we've probably read in at least some checksums as well. But, this has a few problems: * The checksums are indexed by logical offset in the file. When compression is on, this means we have to do the expensive checksumming on the uncompressed data. It would be faster if we could checksum the compressed data instead. * If we implement encryption, we'll be checksumming the plain text and storing that on disk. This is significantly less secure. * For either compression or encryption, we have to get the plain text back before we can verify the checksum as correct. This makes the raid layer balancing and extent moving much more expensive. * It makes the front end caching code more complex, as we have touch the subvolume and inodes as we cache extents. * There is potentitally one copy of the checksum in each subvolume referencing an extent. The solution used here is to store the extent checksums in a dedicated tree. This allows us to index the checksums by phyiscal extent start and length. It means: * The checksum is against the data stored on disk, after any compression or encryption is done. * The checksum is stored in a central location, and can be verified without following back references, or reading inodes. This makes compression significantly faster by reducing the amount of data that needs to be checksummed. It will also allow much faster raid management code in general. The checksums are indexed by a key with a fixed objectid (a magic value in ctree.h) and offset set to the starting byte of the extent. This allows us to copy the checksum items into the fsync log tree directly (or any other tree), without having to invent a second format for them. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-08 21:58:54 +00:00
if (bio_flags & EXTENT_BIO_COMPRESSED) {
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
return btrfs_submit_compressed_read(inode, bio,
mirror_num, bio_flags);
} else if (!skip_sum) {
ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
if (ret)
return ret;
}
goto mapit;
} else if (!skip_sum) {
/* csum items have already been cloned */
if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
goto mapit;
/* we're doing a write, do the async checksumming */
return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
inode, rw, bio, mirror_num,
bio_flags, bio_offset,
__btrfs_submit_bio_start,
Btrfs: Add ordered async work queues Btrfs uses kernel threads to create async work queues for cpu intensive operations such as checksumming and decompression. These work well, but they make it difficult to keep IO order intact. A single writepages call from pdflush or fsync will turn into a number of bios, and each bio is checksummed in parallel. Once the checksum is computed, the bio is sent down to the disk, and since we don't control the order in which the parallel operations happen, they might go down to the disk in almost any order. The code deals with this somewhat by having deep work queues for a single kernel thread, making it very likely that a single thread will process all the bios for a single inode. This patch introduces an explicitly ordered work queue. As work structs are placed into the queue they are put onto the tail of a list. They have three callbacks: ->func (cpu intensive processing here) ->ordered_func (order sensitive processing here) ->ordered_free (free the work struct, all processing is done) The work struct has three callbacks. The func callback does the cpu intensive work, and when it completes the work struct is marked as done. Every time a work struct completes, the list is checked to see if the head is marked as done. If so the ordered_func callback is used to do the order sensitive processing and the ordered_free callback is used to do any cleanup. Then we loop back and check the head of the list again. This patch also changes the checksumming code to use the ordered workqueues. One a 4 drive array, it increases streaming writes from 280MB/s to 350MB/s. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-11-07 03:03:00 +00:00
__btrfs_submit_bio_done);
}
mapit:
return btrfs_map_bio(root, rw, bio, mirror_num, 0);
}
/*
* given a list of ordered sums record them in the inode. This happens
* at IO completion time based on sums calculated at bio submission time.
*/
static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
struct inode *inode, u64 file_offset,
struct list_head *list)
{
struct btrfs_ordered_sum *sum;
list_for_each_entry(sum, list, list) {
Btrfs: move data checksumming into a dedicated tree Btrfs stores checksums for each data block. Until now, they have been stored in the subvolume trees, indexed by the inode that is referencing the data block. This means that when we read the inode, we've probably read in at least some checksums as well. But, this has a few problems: * The checksums are indexed by logical offset in the file. When compression is on, this means we have to do the expensive checksumming on the uncompressed data. It would be faster if we could checksum the compressed data instead. * If we implement encryption, we'll be checksumming the plain text and storing that on disk. This is significantly less secure. * For either compression or encryption, we have to get the plain text back before we can verify the checksum as correct. This makes the raid layer balancing and extent moving much more expensive. * It makes the front end caching code more complex, as we have touch the subvolume and inodes as we cache extents. * There is potentitally one copy of the checksum in each subvolume referencing an extent. The solution used here is to store the extent checksums in a dedicated tree. This allows us to index the checksums by phyiscal extent start and length. It means: * The checksum is against the data stored on disk, after any compression or encryption is done. * The checksum is stored in a central location, and can be verified without following back references, or reading inodes. This makes compression significantly faster by reducing the amount of data that needs to be checksummed. It will also allow much faster raid management code in general. The checksums are indexed by a key with a fixed objectid (a magic value in ctree.h) and offset set to the starting byte of the extent. This allows us to copy the checksum items into the fsync log tree directly (or any other tree), without having to invent a second format for them. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-08 21:58:54 +00:00
btrfs_csum_file_blocks(trans,
BTRFS_I(inode)->root->fs_info->csum_root, sum);
}
return 0;
}
int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
struct extent_state **cached_state)
{
if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
WARN_ON(1);
return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
cached_state, GFP_NOFS);
}
/* see btrfs_writepage_start_hook for details on why this is required */
struct btrfs_writepage_fixup {
struct page *page;
struct btrfs_work work;
};
static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
{
struct btrfs_writepage_fixup *fixup;
struct btrfs_ordered_extent *ordered;
struct extent_state *cached_state = NULL;
struct page *page;
struct inode *inode;
u64 page_start;
u64 page_end;
int ret;
fixup = container_of(work, struct btrfs_writepage_fixup, work);
page = fixup->page;
again:
lock_page(page);
if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
ClearPageChecked(page);
goto out_page;
}
inode = page->mapping->host;
page_start = page_offset(page);
page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
&cached_state);
/* already ordered? We're done */
if (PagePrivate2(page))
goto out;
ordered = btrfs_lookup_ordered_extent(inode, page_start);
if (ordered) {
unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
page_end, &cached_state, GFP_NOFS);
unlock_page(page);
btrfs_start_ordered_extent(inode, ordered, 1);
btrfs_put_ordered_extent(ordered);
goto again;
}
ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
if (ret) {
mapping_set_error(page->mapping, ret);
end_extent_writepage(page, ret, page_start, page_end);
ClearPageChecked(page);
goto out;
}
btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
ClearPageChecked(page);
set_page_dirty(page);
out:
unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
&cached_state, GFP_NOFS);
out_page:
unlock_page(page);
page_cache_release(page);
kfree(fixup);
}
/*
* There are a few paths in the higher layers of the kernel that directly
* set the page dirty bit without asking the filesystem if it is a
* good idea. This causes problems because we want to make sure COW
* properly happens and the data=ordered rules are followed.
*
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
* In our case any range that doesn't have the ORDERED bit set
* hasn't been properly setup for IO. We kick off an async process
* to fix it up. The async helper will wait for ordered extents, set
* the delalloc bit and make it safe to write the page.
*/
static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
{
struct inode *inode = page->mapping->host;
struct btrfs_writepage_fixup *fixup;
struct btrfs_root *root = BTRFS_I(inode)->root;
/* this page is properly in the ordered list */
if (TestClearPagePrivate2(page))
return 0;
if (PageChecked(page))
return -EAGAIN;
fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
if (!fixup)
return -EAGAIN;
SetPageChecked(page);
page_cache_get(page);
fixup->work.func = btrfs_writepage_fixup_worker;
fixup->page = page;
btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
return -EBUSY;
}
static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
struct inode *inode, u64 file_pos,
u64 disk_bytenr, u64 disk_num_bytes,
u64 num_bytes, u64 ram_bytes,
u8 compression, u8 encryption,
u16 other_encoding, int extent_type)
{
struct btrfs_root *root = BTRFS_I(inode)->root;
struct btrfs_file_extent_item *fi;
struct btrfs_path *path;
struct extent_buffer *leaf;
struct btrfs_key ins;
u64 hint;
int ret;
path = btrfs_alloc_path();
if (!path)
return -ENOMEM;
path->leave_spinning = 1;
/*
* we may be replacing one extent in the tree with another.
* The new extent is pinned in the extent map, and we don't want
* to drop it from the cache until it is completely in the btree.
*
* So, tell btrfs_drop_extents to leave this extent in the cache.
* the caller is expected to unpin it and allow it to be merged
* with the others.
*/
ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
&hint, 0);
BUG_ON(ret);
ins.objectid = btrfs_ino(inode);
ins.offset = file_pos;
ins.type = BTRFS_EXTENT_DATA_KEY;
ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
BUG_ON(ret);
leaf = path->nodes[0];
fi = btrfs_item_ptr(leaf, path->slots[0],
struct btrfs_file_extent_item);
btrfs_set_file_extent_generation(leaf, fi, trans->transid);
btrfs_set_file_extent_type(leaf, fi, extent_type);
btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
btrfs_set_file_extent_offset(leaf, fi, 0);
btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
btrfs_set_file_extent_compression(leaf, fi, compression);
btrfs_set_file_extent_encryption(leaf, fi, encryption);
btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
btrfs_unlock_up_safe(path, 1);
btrfs_set_lock_blocking(leaf);
btrfs_mark_buffer_dirty(leaf);
inode_add_bytes(inode, num_bytes);
ins.objectid = disk_bytenr;
ins.offset = disk_num_bytes;
ins.type = BTRFS_EXTENT_ITEM_KEY;
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 14:45:14 +00:00
ret = btrfs_alloc_reserved_file_extent(trans, root,
root->root_key.objectid,
btrfs_ino(inode), file_pos, &ins);
BUG_ON(ret);
btrfs_free_path(path);
return 0;
}
/*
* helper function for btrfs_finish_ordered_io, this
* just reads in some of the csum leaves to prime them into ram
* before we start the transaction. It limits the amount of btree
* reads required while inside the transaction.
*/
/* as ordered data IO finishes, this gets called so we can finish
* an ordered extent if the range of bytes in the file it covers are
* fully written.
*/
static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
{
struct btrfs_root *root = BTRFS_I(inode)->root;
struct btrfs_trans_handle *trans = NULL;
struct btrfs_ordered_extent *ordered_extent = NULL;
struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
struct extent_state *cached_state = NULL;
int compress_type = 0;
int ret;
bool nolock;
ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
end - start + 1);
if (!ret)
return 0;
BUG_ON(!ordered_extent);
nolock = btrfs_is_free_space_inode(root, inode);
if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
BUG_ON(!list_empty(&ordered_extent->list));
ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
if (!ret) {
if (nolock)
trans = btrfs_join_transaction_nolock(root);
else
trans = btrfs_join_transaction(root);
BUG_ON(IS_ERR(trans));
trans->block_rsv = &root->fs_info->delalloc_block_rsv;
ret = btrfs_update_inode_fallback(trans, root, inode);
BUG_ON(ret);
}
goto out;
}
lock_extent_bits(io_tree, ordered_extent->file_offset,
ordered_extent->file_offset + ordered_extent->len - 1,
0, &cached_state);
if (nolock)
trans = btrfs_join_transaction_nolock(root);
else
trans = btrfs_join_transaction(root);
BUG_ON(IS_ERR(trans));
trans->block_rsv = &root->fs_info->delalloc_block_rsv;
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
compress_type = ordered_extent->compress_type;
if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
BUG_ON(compress_type);
ret = btrfs_mark_extent_written(trans, inode,
ordered_extent->file_offset,
ordered_extent->file_offset +
ordered_extent->len);
BUG_ON(ret);
} else {
BUG_ON(root == root->fs_info->tree_root);
ret = insert_reserved_file_extent(trans, inode,
ordered_extent->file_offset,
ordered_extent->start,
ordered_extent->disk_len,
ordered_extent->len,
ordered_extent->len,
compress_type, 0, 0,
BTRFS_FILE_EXTENT_REG);
unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
ordered_extent->file_offset,
ordered_extent->len);
BUG_ON(ret);
}
unlock_extent_cached(io_tree, ordered_extent->file_offset,
ordered_extent->file_offset +
ordered_extent->len - 1, &cached_state, GFP_NOFS);
add_pending_csums(trans, inode, ordered_extent->file_offset,
&ordered_extent->list);
ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
ret = btrfs_update_inode_fallback(trans, root, inode);
BUG_ON(ret);
}
ret = 0;
out:
Btrfs: inline checksums into the disk free space cache Yeah yeah I know this is how we used to do it and then I changed it, but damnit I'm changing it back. The fact is that writing out checksums will modify metadata, which could cause us to dirty a block group we've already written out, so we have to truncate it and all of it's checksums and re-write it which will write new checksums which could dirty a blockg roup that has already been written and you see where I'm going with this? This can cause unmount or really anything that depends on a transaction to commit to take it's sweet damned time to happen. So go back to the way it was, only this time we're specifically setting NODATACOW because we can't go through the COW pathway anyway and we're doing our own built-in cow'ing by truncating the free space cache. The other new thing is once we truncate the old cache and preallocate the new space, we don't need to do that song and dance at all for the rest of the transaction, we can just overwrite the existing space with the new cache if the block group changes for whatever reason, and the NODATACOW will let us do this fine. So keep track of which transaction we last cleared our cache in and if we cleared it in this transaction just say we're all setup and carry on. This survives xfstests and stress.sh. The inode cache will continue to use the normal csum infrastructure since it only gets written once and there will be no more modifications to the fs tree in a transaction commit. Signed-off-by: Josef Bacik <josef@redhat.com>
2011-10-06 12:58:24 +00:00
if (root != root->fs_info->tree_root)
btrfs_delalloc_release_metadata(inode, ordered_extent->len);
Btrfs: inline checksums into the disk free space cache Yeah yeah I know this is how we used to do it and then I changed it, but damnit I'm changing it back. The fact is that writing out checksums will modify metadata, which could cause us to dirty a block group we've already written out, so we have to truncate it and all of it's checksums and re-write it which will write new checksums which could dirty a blockg roup that has already been written and you see where I'm going with this? This can cause unmount or really anything that depends on a transaction to commit to take it's sweet damned time to happen. So go back to the way it was, only this time we're specifically setting NODATACOW because we can't go through the COW pathway anyway and we're doing our own built-in cow'ing by truncating the free space cache. The other new thing is once we truncate the old cache and preallocate the new space, we don't need to do that song and dance at all for the rest of the transaction, we can just overwrite the existing space with the new cache if the block group changes for whatever reason, and the NODATACOW will let us do this fine. So keep track of which transaction we last cleared our cache in and if we cleared it in this transaction just say we're all setup and carry on. This survives xfstests and stress.sh. The inode cache will continue to use the normal csum infrastructure since it only gets written once and there will be no more modifications to the fs tree in a transaction commit. Signed-off-by: Josef Bacik <josef@redhat.com>
2011-10-06 12:58:24 +00:00
if (trans) {
if (nolock)
btrfs_end_transaction_nolock(trans, root);
Btrfs: inline checksums into the disk free space cache Yeah yeah I know this is how we used to do it and then I changed it, but damnit I'm changing it back. The fact is that writing out checksums will modify metadata, which could cause us to dirty a block group we've already written out, so we have to truncate it and all of it's checksums and re-write it which will write new checksums which could dirty a blockg roup that has already been written and you see where I'm going with this? This can cause unmount or really anything that depends on a transaction to commit to take it's sweet damned time to happen. So go back to the way it was, only this time we're specifically setting NODATACOW because we can't go through the COW pathway anyway and we're doing our own built-in cow'ing by truncating the free space cache. The other new thing is once we truncate the old cache and preallocate the new space, we don't need to do that song and dance at all for the rest of the transaction, we can just overwrite the existing space with the new cache if the block group changes for whatever reason, and the NODATACOW will let us do this fine. So keep track of which transaction we last cleared our cache in and if we cleared it in this transaction just say we're all setup and carry on. This survives xfstests and stress.sh. The inode cache will continue to use the normal csum infrastructure since it only gets written once and there will be no more modifications to the fs tree in a transaction commit. Signed-off-by: Josef Bacik <josef@redhat.com>
2011-10-06 12:58:24 +00:00
else
btrfs_end_transaction(trans, root);
}
/* once for us */
btrfs_put_ordered_extent(ordered_extent);
/* once for the tree */
btrfs_put_ordered_extent(ordered_extent);
return 0;
}
static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
struct extent_state *state, int uptodate)
{
Btrfs: add initial tracepoint support for btrfs Tracepoints can provide insight into why btrfs hits bugs and be greatly helpful for debugging, e.g dd-7822 [000] 2121.641088: btrfs_inode_request: root = 5(FS_TREE), gen = 4, ino = 256, blocks = 8, disk_i_size = 0, last_trans = 8, logged_trans = 0 dd-7822 [000] 2121.641100: btrfs_inode_new: root = 5(FS_TREE), gen = 8, ino = 257, blocks = 0, disk_i_size = 0, last_trans = 0, logged_trans = 0 btrfs-transacti-7804 [001] 2146.935420: btrfs_cow_block: root = 2(EXTENT_TREE), refs = 2, orig_buf = 29368320 (orig_level = 0), cow_buf = 29388800 (cow_level = 0) btrfs-transacti-7804 [001] 2146.935473: btrfs_cow_block: root = 1(ROOT_TREE), refs = 2, orig_buf = 29364224 (orig_level = 0), cow_buf = 29392896 (cow_level = 0) btrfs-transacti-7804 [001] 2146.972221: btrfs_transaction_commit: root = 1(ROOT_TREE), gen = 8 flush-btrfs-2-7821 [001] 2155.824210: btrfs_chunk_alloc: root = 3(CHUNK_TREE), offset = 1103101952, size = 1073741824, num_stripes = 1, sub_stripes = 0, type = DATA flush-btrfs-2-7821 [001] 2155.824241: btrfs_cow_block: root = 2(EXTENT_TREE), refs = 2, orig_buf = 29388800 (orig_level = 0), cow_buf = 29396992 (cow_level = 0) flush-btrfs-2-7821 [001] 2155.824255: btrfs_cow_block: root = 4(DEV_TREE), refs = 2, orig_buf = 29372416 (orig_level = 0), cow_buf = 29401088 (cow_level = 0) flush-btrfs-2-7821 [000] 2155.824329: btrfs_cow_block: root = 3(CHUNK_TREE), refs = 2, orig_buf = 20971520 (orig_level = 0), cow_buf = 20975616 (cow_level = 0) btrfs-endio-wri-7800 [001] 2155.898019: btrfs_cow_block: root = 5(FS_TREE), refs = 2, orig_buf = 29384704 (orig_level = 0), cow_buf = 29405184 (cow_level = 0) btrfs-endio-wri-7800 [001] 2155.898043: btrfs_cow_block: root = 7(CSUM_TREE), refs = 2, orig_buf = 29376512 (orig_level = 0), cow_buf = 29409280 (cow_level = 0) Here is what I have added: 1) ordere_extent: btrfs_ordered_extent_add btrfs_ordered_extent_remove btrfs_ordered_extent_start btrfs_ordered_extent_put These provide critical information to understand how ordered_extents are updated. 2) extent_map: btrfs_get_extent extent_map is used in both read and write cases, and it is useful for tracking how btrfs specific IO is running. 3) writepage: __extent_writepage btrfs_writepage_end_io_hook Pages are cirtical resourses and produce a lot of corner cases during writeback, so it is valuable to know how page is written to disk. 4) inode: btrfs_inode_new btrfs_inode_request btrfs_inode_evict These can show where and when a inode is created, when a inode is evicted. 5) sync: btrfs_sync_file btrfs_sync_fs These show sync arguments. 6) transaction: btrfs_transaction_commit In transaction based filesystem, it will be useful to know the generation and who does commit. 7) back reference and cow: btrfs_delayed_tree_ref btrfs_delayed_data_ref btrfs_delayed_ref_head btrfs_cow_block Btrfs natively supports back references, these tracepoints are helpful on understanding btrfs's COW mechanism. 8) chunk: btrfs_chunk_alloc btrfs_chunk_free Chunk is a link between physical offset and logical offset, and stands for space infomation in btrfs, and these are helpful on tracing space things. 9) reserved_extent: btrfs_reserved_extent_alloc btrfs_reserved_extent_free These can show how btrfs uses its space. Signed-off-by: Liu Bo <liubo2009@cn.fujitsu.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-03-24 11:18:59 +00:00
trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
ClearPagePrivate2(page);
return btrfs_finish_ordered_io(page->mapping->host, start, end);
}
/*
* when reads are done, we need to check csums to verify the data is correct
* if there's a match, we allow the bio to finish. If not, the code in
* extent_io.c will try to find good copies for us.
*/
static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
struct extent_state *state)
{
size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
struct inode *inode = page->mapping->host;
struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
char *kaddr;
u64 private = ~(u32)0;
int ret;
struct btrfs_root *root = BTRFS_I(inode)->root;
u32 csum = ~(u32)0;
Btrfs: move data checksumming into a dedicated tree Btrfs stores checksums for each data block. Until now, they have been stored in the subvolume trees, indexed by the inode that is referencing the data block. This means that when we read the inode, we've probably read in at least some checksums as well. But, this has a few problems: * The checksums are indexed by logical offset in the file. When compression is on, this means we have to do the expensive checksumming on the uncompressed data. It would be faster if we could checksum the compressed data instead. * If we implement encryption, we'll be checksumming the plain text and storing that on disk. This is significantly less secure. * For either compression or encryption, we have to get the plain text back before we can verify the checksum as correct. This makes the raid layer balancing and extent moving much more expensive. * It makes the front end caching code more complex, as we have touch the subvolume and inodes as we cache extents. * There is potentitally one copy of the checksum in each subvolume referencing an extent. The solution used here is to store the extent checksums in a dedicated tree. This allows us to index the checksums by phyiscal extent start and length. It means: * The checksum is against the data stored on disk, after any compression or encryption is done. * The checksum is stored in a central location, and can be verified without following back references, or reading inodes. This makes compression significantly faster by reducing the amount of data that needs to be checksummed. It will also allow much faster raid management code in general. The checksums are indexed by a key with a fixed objectid (a magic value in ctree.h) and offset set to the starting byte of the extent. This allows us to copy the checksum items into the fsync log tree directly (or any other tree), without having to invent a second format for them. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-08 21:58:54 +00:00
if (PageChecked(page)) {
ClearPageChecked(page);
goto good;
}
if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
goto good;
if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
GFP_NOFS);
return 0;
}
Btrfs: move data checksumming into a dedicated tree Btrfs stores checksums for each data block. Until now, they have been stored in the subvolume trees, indexed by the inode that is referencing the data block. This means that when we read the inode, we've probably read in at least some checksums as well. But, this has a few problems: * The checksums are indexed by logical offset in the file. When compression is on, this means we have to do the expensive checksumming on the uncompressed data. It would be faster if we could checksum the compressed data instead. * If we implement encryption, we'll be checksumming the plain text and storing that on disk. This is significantly less secure. * For either compression or encryption, we have to get the plain text back before we can verify the checksum as correct. This makes the raid layer balancing and extent moving much more expensive. * It makes the front end caching code more complex, as we have touch the subvolume and inodes as we cache extents. * There is potentitally one copy of the checksum in each subvolume referencing an extent. The solution used here is to store the extent checksums in a dedicated tree. This allows us to index the checksums by phyiscal extent start and length. It means: * The checksum is against the data stored on disk, after any compression or encryption is done. * The checksum is stored in a central location, and can be verified without following back references, or reading inodes. This makes compression significantly faster by reducing the amount of data that needs to be checksummed. It will also allow much faster raid management code in general. The checksums are indexed by a key with a fixed objectid (a magic value in ctree.h) and offset set to the starting byte of the extent. This allows us to copy the checksum items into the fsync log tree directly (or any other tree), without having to invent a second format for them. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-08 21:58:54 +00:00
if (state && state->start == start) {
private = state->private;
ret = 0;
} else {
ret = get_state_private(io_tree, start, &private);
}
kaddr = kmap_atomic(page, KM_USER0);
if (ret)
goto zeroit;
csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
btrfs_csum_final(csum, (char *)&csum);
if (csum != private)
goto zeroit;
kunmap_atomic(kaddr, KM_USER0);
Btrfs: move data checksumming into a dedicated tree Btrfs stores checksums for each data block. Until now, they have been stored in the subvolume trees, indexed by the inode that is referencing the data block. This means that when we read the inode, we've probably read in at least some checksums as well. But, this has a few problems: * The checksums are indexed by logical offset in the file. When compression is on, this means we have to do the expensive checksumming on the uncompressed data. It would be faster if we could checksum the compressed data instead. * If we implement encryption, we'll be checksumming the plain text and storing that on disk. This is significantly less secure. * For either compression or encryption, we have to get the plain text back before we can verify the checksum as correct. This makes the raid layer balancing and extent moving much more expensive. * It makes the front end caching code more complex, as we have touch the subvolume and inodes as we cache extents. * There is potentitally one copy of the checksum in each subvolume referencing an extent. The solution used here is to store the extent checksums in a dedicated tree. This allows us to index the checksums by phyiscal extent start and length. It means: * The checksum is against the data stored on disk, after any compression or encryption is done. * The checksum is stored in a central location, and can be verified without following back references, or reading inodes. This makes compression significantly faster by reducing the amount of data that needs to be checksummed. It will also allow much faster raid management code in general. The checksums are indexed by a key with a fixed objectid (a magic value in ctree.h) and offset set to the starting byte of the extent. This allows us to copy the checksum items into the fsync log tree directly (or any other tree), without having to invent a second format for them. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-12-08 21:58:54 +00:00
good:
return 0;
zeroit:
printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
"private %llu\n",
(unsigned long long)btrfs_ino(page->mapping->host),
(unsigned long long)start, csum,
(unsigned long long)private);
memset(kaddr + offset, 1, end - start + 1);
flush_dcache_page(page);
kunmap_atomic(kaddr, KM_USER0);
if (private == 0)
return 0;
return -EIO;
}
struct delayed_iput {
struct list_head list;
struct inode *inode;
};
void btrfs_add_delayed_iput(struct inode *inode)
{
struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
struct delayed_iput *delayed;
if (atomic_add_unless(&inode->i_count, -1, 1))
return;
delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
delayed->inode = inode;
spin_lock(&fs_info->delayed_iput_lock);
list_add_tail(&delayed->list, &fs_info->delayed_iputs);
spin_unlock(&fs_info->delayed_iput_lock);
}
void btrfs_run_delayed_iputs(struct btrfs_root *root)
{
LIST_HEAD(list);
struct btrfs_fs_info *fs_info = root->fs_info;
struct delayed_iput *delayed;
int empty;
spin_lock(&fs_info->delayed_iput_lock);
empty = list_empty(&fs_info->delayed_iputs);
spin_unlock(&fs_info->delayed_iput_lock);
if (empty)
return;
down_read(&root->fs_info->cleanup_work_sem);
spin_lock(&fs_info->delayed_iput_lock);
list_splice_init(&fs_info->delayed_iputs, &list);
spin_unlock(&fs_info->delayed_iput_lock);
while (!list_empty(&list)) {
delayed = list_entry(list.next, struct delayed_iput, list);
list_del(&delayed->list);
iput(delayed->inode);
kfree(delayed);
}
up_read(&root->fs_info->cleanup_work_sem);
}
enum btrfs_orphan_cleanup_state {
ORPHAN_CLEANUP_STARTED = 1,
ORPHAN_CLEANUP_DONE = 2,
};
/*
* This is called in transaction commit time. If there are no orphan
* files in the subvolume, it removes orphan item and frees block_rsv
* structure.
*/
void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
struct btrfs_root *root)
{
struct btrfs_block_rsv *block_rsv;
int ret;
if (!list_empty(&root->orphan_list) ||
root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
return;
spin_lock(&root->orphan_lock);
if (!list_empty(&root->orphan_list)) {
spin_unlock(&root->orphan_lock);
return;
}
if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
spin_unlock(&root->orphan_lock);
return;
}
block_rsv = root->orphan_block_rsv;
root->orphan_block_rsv = NULL;
spin_unlock(&root->orphan_lock);
if (root->orphan_item_inserted &&
btrfs_root_refs(&root->root_item) > 0) {
ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
root->root_key.objectid);
BUG_ON(ret);
root->orphan_item_inserted = 0;
}
if (block_rsv) {
WARN_ON(block_rsv->size > 0);
btrfs_free_block_rsv(root, block_rsv);
}
}
/*
* This creates an orphan entry for the given inode in case something goes
* wrong in the middle of an unlink/truncate.
*
* NOTE: caller of this function should reserve 5 units of metadata for
* this function.
*/
int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
{
struct btrfs_root *root = BTRFS_I(inode)->root;
struct btrfs_block_rsv *block_rsv = NULL;
int reserve = 0;
int insert = 0;
int ret;
if (!root->orphan_block_rsv) {
block_rsv = btrfs_alloc_block_rsv(root);
if (!block_rsv)
return -ENOMEM;
}
spin_lock(&root->orphan_lock);
if (!root->orphan_block_rsv) {
root->orphan_block_rsv = block_rsv;
} else if (block_rsv) {
btrfs_free_block_rsv(root, block_rsv);
block_rsv = NULL;
}
if (list_empty(&BTRFS_I(inode)->i_orphan)) {
list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
#if 0
/*
* For proper ENOSPC handling, we should do orphan
* cleanup when mounting. But this introduces backward
* compatibility issue.
*/
if (!xchg(&root->orphan_item_inserted, 1))
insert = 2;
else
insert = 1;
#endif
insert = 1;
}
if (!BTRFS_I(inode)->orphan_meta_reserved) {
BTRFS_I(inode)->orphan_meta_reserved = 1;
reserve = 1;
}
spin_unlock(&root->orphan_lock);
/* grab metadata reservation from transaction handle */
if (reserve) {
ret = btrfs_orphan_reserve_metadata(trans, inode);
BUG_ON(ret);
}
/* insert an orphan item to track this unlinked/truncated file */
if (insert >= 1) {
ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
BUG_ON(ret && ret != -EEXIST);
}
/* insert an orphan item to track subvolume contains orphan files */
if (insert >= 2) {
ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
root->root_key.objectid);
BUG_ON(ret);
}
return 0;
}
/*
* We have done the truncate/delete so we can go ahead and remove the orphan
* item for this particular inode.
*/
int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
{
struct btrfs_root *root = BTRFS_I(inode)->root;
int delete_item = 0;
int release_rsv = 0;
int ret = 0;
spin_lock(&root->orphan_lock);
if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
list_del_init(&BTRFS_I(inode)->i_orphan);
delete_item = 1;
}
if (BTRFS_I(inode)->orphan_meta_reserved) {
BTRFS_I(inode)->orphan_meta_reserved = 0;
release_rsv = 1;
}
spin_unlock(&root->orphan_lock);
if (trans && delete_item) {
ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
BUG_ON(ret);
}
if (release_rsv)
btrfs_orphan_release_metadata(inode);
return 0;
}
/*
* this cleans up any orphans that may be left on the list from the last use
* of this root.
*/
int btrfs_orphan_cleanup(struct btrfs_root *root)
{
struct btrfs_path *path;
struct extent_buffer *leaf;
struct btrfs_key key, found_key;
struct btrfs_trans_handle *trans;
struct inode *inode;
u64 last_objectid = 0;
int ret = 0, nr_unlink = 0, nr_truncate = 0;
if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
return 0;
path = btrfs_alloc_path();
if (!path) {
ret = -ENOMEM;
goto out;
}
path->reada = -1;
key.objectid = BTRFS_ORPHAN_OBJECTID;
btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
key.offset = (u64)-1;
while (1) {
ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
if (ret < 0)
goto out;
/*
* if ret == 0 means we found what we were searching for, which
* is weird, but possible, so only screw with path if we didn't
* find the key and see if we have stuff that matches
*/
if (ret > 0) {
ret = 0;
if (path->slots[0] == 0)
break;
path->slots[0]--;
}
/* pull out the item */
leaf = path->nodes[0];
btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
/* make sure the item matches what we want */
if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
break;
if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
break;
/* release the path since we're done with it */
btrfs_release_path(path);
/*
* this is where we are basically btrfs_lookup, without the
* crossing root thing. we store the inode number in the
* offset of the orphan item.
*/
if (found_key.offset == last_objectid) {
printk(KERN_ERR "btrfs: Error removing orphan entry, "
"stopping orphan cleanup\n");
ret = -EINVAL;
goto out;
}
last_objectid = found_key.offset;
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 14:45:14 +00:00
found_key.objectid = found_key.offset;
found_key.type = BTRFS_INODE_ITEM_KEY;
found_key.offset = 0;
Btrfs: change how we mount subvolumes This work is in preperation for being able to set a different root as the default mounting root. There is currently a problem with how we mount subvolumes. We cannot currently mount a subvolume of a subvolume, you can only mount subvolumes/snapshots of the default subvolume. So say you take a snapshot of the default subvolume and call it snap1, and then take a snapshot of snap1 and call it snap2, so now you have / /snap1 /snap1/snap2 as your available volumes. Currently you can only mount / and /snap1, you cannot mount /snap1/snap2. To fix this problem instead of passing subvolid=<name> you must pass in subvolid=<treeid>, where <treeid> is the tree id that gets spit out via the subvolume listing you get from the subvolume listing patches (btrfs filesystem list). This allows us to mount /, /snap1 and /snap1/snap2 as the root volume. In addition to the above, we also now read the default dir item in the tree root to get the root key that it points to. For now this just points at what has always been the default subvolme, but later on I plan to change it to point at whatever root you want to be the new default root, so you can just set the default mount and not have to mount with -o subvolid=<treeid>. I tested this out with the above scenario and it worked perfectly. Thanks, mount -o subvol operates inside the selected subvolid. For example: mount -o subvol=snap1,subvolid=256 /dev/xxx /mnt /mnt will have the snap1 directory for the subvolume with id 256. mount -o subvol=snap /dev/xxx /mnt /mnt will be the snap directory of whatever the default subvolume is. Signed-off-by: Josef Bacik <josef@redhat.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-12-04 17:38:27 +00:00
inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
ret = PTR_RET(inode);
if (ret && ret != -ESTALE)
goto out;
if (ret == -ESTALE && root == root->fs_info->tree_root) {
struct btrfs_root *dead_root;
struct btrfs_fs_info *fs_info = root->fs_info;
int is_dead_root = 0;
/*
* this is an orphan in the tree root. Currently these
* could come from 2 sources:
* a) a snapshot deletion in progress
* b) a free space cache inode
* We need to distinguish those two, as the snapshot
* orphan must not get deleted.
* find_dead_roots already ran before us, so if this
* is a snapshot deletion, we should find the root
* in the dead_roots list
*/
spin_lock(&fs_info->trans_lock);
list_for_each_entry(dead_root, &fs_info->dead_roots,
root_list) {
if (dead_root->root_key.objectid ==
found_key.objectid) {
is_dead_root = 1;
break;
}
}
spin_unlock(&fs_info->trans_lock);
if (is_dead_root) {
/* prevent this orphan from being found again */
key.offset = found_key.objectid - 1;
continue;
}
}
/*
* Inode is already gone but the orphan item is still there,
* kill the orphan item.
*/
if (ret == -ESTALE) {
trans = btrfs_start_transaction(root, 1);
if (IS_ERR(trans)) {
ret = PTR_ERR(trans);
goto out;
}
ret = btrfs_del_orphan_item(trans, root,
found_key.objectid);
BUG_ON(ret);
btrfs_end_transaction(trans, root);
continue;
}
/*
* add this inode to the orphan list so btrfs_orphan_del does
* the proper thing when we hit it
*/
spin_lock(&root->orphan_lock);
list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
spin_unlock(&root->orphan_lock);
/* if we have links, this was a truncate, lets do that */
if (inode->i_nlink) {
if (!S_ISREG(inode->i_mode)) {
WARN_ON(1);
iput(inode);
continue;
}
nr_truncate++;
ret = btrfs_truncate(inode);
} else {
nr_unlink++;
}
/* this will do delete_inode and everything for us */
iput(inode);
if (ret)
goto out;
}
/* release the path since we're done with it */
btrfs_release_path(path);
root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
if (root->orphan_block_rsv)
btrfs_block_rsv_release(root, root->orphan_block_rsv,
(u64)-1);
if (root->orphan_block_rsv || root->orphan_item_inserted) {
trans = btrfs_join_transaction(root);
if (!IS_ERR(trans))
btrfs_end_transaction(trans, root);
}
if (nr_unlink)
printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
if (nr_truncate)
printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
out:
if (ret)
printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
btrfs_free_path(path);
return ret;
}
/*
* very simple check to peek ahead in the leaf looking for xattrs. If we
* don't find any xattrs, we know there can't be any acls.
*
* slot is the slot the inode is in, objectid is the objectid of the inode
*/
static noinline int acls_after_inode_item(struct extent_buffer *leaf,
int slot, u64 objectid)
{
u32 nritems = btrfs_header_nritems(leaf);
struct btrfs_key found_key;
int scanned = 0;
slot++;
while (slot < nritems) {
btrfs_item_key_to_cpu(leaf, &found_key, slot);
/* we found a different objectid, there must not be acls */
if (found_key.objectid != objectid)
return 0;
/* we found an xattr, assume we've got an acl */
if (found_key.type == BTRFS_XATTR_ITEM_KEY)
return 1;
/*
* we found a key greater than an xattr key, there can't
* be any acls later on
*/
if (found_key.type > BTRFS_XATTR_ITEM_KEY)
return 0;
slot++;
scanned++;
/*
* it goes inode, inode backrefs, xattrs, extents,
* so if there are a ton of hard links to an inode there can
* be a lot of backrefs. Don't waste time searching too hard,
* this is just an optimization
*/
if (scanned >= 8)
break;
}
/* we hit the end of the leaf before we found an xattr or
* something larger than an xattr. We have to assume the inode
* has acls
*/
return 1;
}
/*
* read an inode from the btree into the in-memory inode
*/
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 14:45:14 +00:00
static void btrfs_read_locked_inode(struct inode *inode)
{
struct btrfs_path *path;
struct extent_buffer *leaf;
struct btrfs_inode_item *inode_item;
struct btrfs_timespec *tspec;
struct btrfs_root *root = BTRFS_I(inode)->root;
struct btrfs_key location;
int maybe_acls;
u32 rdev;
int ret;
btrfs: fix inconsonant inode information When iputting the inode, We may leave the delayed nodes if they have some delayed items that have not been dealt with. So when the inode is read again, we must look up the relative delayed node, and use the information in it to initialize the inode. Or we will get inconsonant inode information, it may cause that the same directory index number is allocated again, and hit the following oops: [ 5447.554187] err add delayed dir index item(name: pglog_0.965_0) into the insertion tree of the delayed node(root id: 262, inode id: 258, errno: -17) [ 5447.569766] ------------[ cut here ]------------ [ 5447.575361] kernel BUG at fs/btrfs/delayed-inode.c:1301! [SNIP] [ 5447.790721] Call Trace: [ 5447.793191] [<ffffffffa0641c4e>] btrfs_insert_dir_item+0x189/0x1bb [btrfs] [ 5447.800156] [<ffffffffa0651a45>] btrfs_add_link+0x12b/0x191 [btrfs] [ 5447.806517] [<ffffffffa0651adc>] btrfs_add_nondir+0x31/0x58 [btrfs] [ 5447.812876] [<ffffffffa0651d6a>] btrfs_create+0xf9/0x197 [btrfs] [ 5447.818961] [<ffffffff8111f840>] vfs_create+0x72/0x92 [ 5447.824090] [<ffffffff8111fa8c>] do_last+0x22c/0x40b [ 5447.829133] [<ffffffff8112076a>] path_openat+0xc0/0x2ef [ 5447.834438] [<ffffffff810c58e2>] ? __perf_event_task_sched_out+0x24/0x44 [ 5447.841216] [<ffffffff8103ecdd>] ? perf_event_task_sched_out+0x59/0x67 [ 5447.847846] [<ffffffff81121a79>] do_filp_open+0x3d/0x87 [ 5447.853156] [<ffffffff811e126c>] ? strncpy_from_user+0x43/0x4d [ 5447.859072] [<ffffffff8111f1f5>] ? getname_flags+0x2e/0x80 [ 5447.864636] [<ffffffff8111f179>] ? do_getname+0x14b/0x173 [ 5447.870112] [<ffffffff8111f1b7>] ? audit_getname+0x16/0x26 [ 5447.875682] [<ffffffff8112b1ab>] ? spin_lock+0xe/0x10 [ 5447.880882] [<ffffffff81112d39>] do_sys_open+0x69/0xae [ 5447.886153] [<ffffffff81112db1>] sys_open+0x20/0x22 [ 5447.891114] [<ffffffff813b9aab>] system_call_fastpath+0x16/0x1b Fix it by reusing the old delayed node. Reported-by: Jim Schutt <jaschut@sandia.gov> Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Tested-by: Jim Schutt <jaschut@sandia.gov> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-06-23 07:27:13 +00:00
bool filled = false;
ret = btrfs_fill_inode(inode, &rdev);
if (!ret)
filled = true;
path = btrfs_alloc_path();
if (!path)
goto make_bad;
path->leave_spinning = 1;
memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
if (ret)
goto make_bad;
leaf = path->nodes[0];
btrfs: fix inconsonant inode information When iputting the inode, We may leave the delayed nodes if they have some delayed items that have not been dealt with. So when the inode is read again, we must look up the relative delayed node, and use the information in it to initialize the inode. Or we will get inconsonant inode information, it may cause that the same directory index number is allocated again, and hit the following oops: [ 5447.554187] err add delayed dir index item(name: pglog_0.965_0) into the insertion tree of the delayed node(root id: 262, inode id: 258, errno: -17) [ 5447.569766] ------------[ cut here ]------------ [ 5447.575361] kernel BUG at fs/btrfs/delayed-inode.c:1301! [SNIP] [ 5447.790721] Call Trace: [ 5447.793191] [<ffffffffa0641c4e>] btrfs_insert_dir_item+0x189/0x1bb [btrfs] [ 5447.800156] [<ffffffffa0651a45>] btrfs_add_link+0x12b/0x191 [btrfs] [ 5447.806517] [<ffffffffa0651adc>] btrfs_add_nondir+0x31/0x58 [btrfs] [ 5447.812876] [<ffffffffa0651d6a>] btrfs_create+0xf9/0x197 [btrfs] [ 5447.818961] [<ffffffff8111f840>] vfs_create+0x72/0x92 [ 5447.824090] [<ffffffff8111fa8c>] do_last+0x22c/0x40b [ 5447.829133] [<ffffffff8112076a>] path_openat+0xc0/0x2ef [ 5447.834438] [<ffffffff810c58e2>] ? __perf_event_task_sched_out+0x24/0x44 [ 5447.841216] [<ffffffff8103ecdd>] ? perf_event_task_sched_out+0x59/0x67 [ 5447.847846] [<ffffffff81121a79>] do_filp_open+0x3d/0x87 [ 5447.853156] [<ffffffff811e126c>] ? strncpy_from_user+0x43/0x4d [ 5447.859072] [<ffffffff8111f1f5>] ? getname_flags+0x2e/0x80 [ 5447.864636] [<ffffffff8111f179>] ? do_getname+0x14b/0x173 [ 5447.870112] [<ffffffff8111f1b7>] ? audit_getname+0x16/0x26 [ 5447.875682] [<ffffffff8112b1ab>] ? spin_lock+0xe/0x10 [ 5447.880882] [<ffffffff81112d39>] do_sys_open+0x69/0xae [ 5447.886153] [<ffffffff81112db1>] sys_open+0x20/0x22 [ 5447.891114] [<ffffffff813b9aab>] system_call_fastpath+0x16/0x1b Fix it by reusing the old delayed node. Reported-by: Jim Schutt <jaschut@sandia.gov> Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Tested-by: Jim Schutt <jaschut@sandia.gov> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-06-23 07:27:13 +00:00
if (filled)
goto cache_acl;
inode_item = btrfs_item_ptr(leaf, path->slots[0],
struct btrfs_inode_item);
inode->i_mode = btrfs_inode_mode(leaf, inode_item);
set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
inode->i_uid = btrfs_inode_uid(leaf, inode_item);
inode->i_gid = btrfs_inode_gid(leaf, inode_item);
btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
tspec = btrfs_inode_atime(inode_item);
inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
tspec = btrfs_inode_mtime(inode_item);
inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
tspec = btrfs_inode_ctime(inode_item);
inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
inode->i_generation = BTRFS_I(inode)->generation;
inode->i_rdev = 0;
rdev = btrfs_inode_rdev(leaf, inode_item);
BTRFS_I(inode)->index_cnt = (u64)-1;
BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
btrfs: fix inconsonant inode information When iputting the inode, We may leave the delayed nodes if they have some delayed items that have not been dealt with. So when the inode is read again, we must look up the relative delayed node, and use the information in it to initialize the inode. Or we will get inconsonant inode information, it may cause that the same directory index number is allocated again, and hit the following oops: [ 5447.554187] err add delayed dir index item(name: pglog_0.965_0) into the insertion tree of the delayed node(root id: 262, inode id: 258, errno: -17) [ 5447.569766] ------------[ cut here ]------------ [ 5447.575361] kernel BUG at fs/btrfs/delayed-inode.c:1301! [SNIP] [ 5447.790721] Call Trace: [ 5447.793191] [<ffffffffa0641c4e>] btrfs_insert_dir_item+0x189/0x1bb [btrfs] [ 5447.800156] [<ffffffffa0651a45>] btrfs_add_link+0x12b/0x191 [btrfs] [ 5447.806517] [<ffffffffa0651adc>] btrfs_add_nondir+0x31/0x58 [btrfs] [ 5447.812876] [<ffffffffa0651d6a>] btrfs_create+0xf9/0x197 [btrfs] [ 5447.818961] [<ffffffff8111f840>] vfs_create+0x72/0x92 [ 5447.824090] [<ffffffff8111fa8c>] do_last+0x22c/0x40b [ 5447.829133] [<ffffffff8112076a>] path_openat+0xc0/0x2ef [ 5447.834438] [<ffffffff810c58e2>] ? __perf_event_task_sched_out+0x24/0x44 [ 5447.841216] [<ffffffff8103ecdd>] ? perf_event_task_sched_out+0x59/0x67 [ 5447.847846] [<ffffffff81121a79>] do_filp_open+0x3d/0x87 [ 5447.853156] [<ffffffff811e126c>] ? strncpy_from_user+0x43/0x4d [ 5447.859072] [<ffffffff8111f1f5>] ? getname_flags+0x2e/0x80 [ 5447.864636] [<ffffffff8111f179>] ? do_getname+0x14b/0x173 [ 5447.870112] [<ffffffff8111f1b7>] ? audit_getname+0x16/0x26 [ 5447.875682] [<ffffffff8112b1ab>] ? spin_lock+0xe/0x10 [ 5447.880882] [<ffffffff81112d39>] do_sys_open+0x69/0xae [ 5447.886153] [<ffffffff81112db1>] sys_open+0x20/0x22 [ 5447.891114] [<ffffffff813b9aab>] system_call_fastpath+0x16/0x1b Fix it by reusing the old delayed node. Reported-by: Jim Schutt <jaschut@sandia.gov> Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Tested-by: Jim Schutt <jaschut@sandia.gov> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-06-23 07:27:13 +00:00
cache_acl:
/*
* try to precache a NULL acl entry for files that don't have
* any xattrs or acls
*/
maybe_acls = acls_after_inode_item(leaf, path->slots[0],
btrfs_ino(inode));
if (!maybe_acls)
cache_no_acl(inode);
btrfs_free_path(path);
switch (inode->i_mode & S_IFMT) {
case S_IFREG:
inode->i_mapping->a_ops = &btrfs_aops;
inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
inode->i_fop = &btrfs_file_operations;
inode->i_op = &btrfs_file_inode_operations;
break;
case S_IFDIR:
inode->i_fop = &btrfs_dir_file_operations;
if (root == root->fs_info->tree_root)
inode->i_op = &btrfs_dir_ro_inode_operations;
else
inode->i_op = &btrfs_dir_inode_operations;
break;
case S_IFLNK:
inode->i_op = &btrfs_symlink_inode_operations;
inode->i_mapping->a_ops = &btrfs_symlink_aops;
inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
break;
default:
inode->i_op = &btrfs_special_inode_operations;
init_special_inode(inode, inode->i_mode, rdev);
break;
}
btrfs_update_iflags(inode);
return;
make_bad:
btrfs_free_path(path);
make_bad_inode(inode);
}
/*
* given a leaf and an inode, copy the inode fields into the leaf
*/
static void fill_inode_item(struct btrfs_trans_handle *trans,
struct extent_buffer *leaf,
struct btrfs_inode_item *item,
struct inode *inode)
{
btrfs_set_inode_uid(leaf, item, inode->i_uid);
btrfs_set_inode_gid(leaf, item, inode->i_gid);
btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
btrfs_set_inode_mode(leaf, item, inode->i_mode);
btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
inode->i_atime.tv_sec);
btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
inode->i_atime.tv_nsec);
btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
inode->i_mtime.tv_sec);
btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
inode->i_mtime.tv_nsec);
btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
inode->i_ctime.tv_sec);
btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
inode->i_ctime.tv_nsec);
btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
btrfs_set_inode_transid(leaf, item, trans->transid);
btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
btrfs_set_inode_block_group(leaf, item, 0);
}
/*
* copy everything in the in-memory inode into the btree.
*/
static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
struct btrfs_root *root, struct inode *inode)
{
struct btrfs_inode_item *inode_item;
struct btrfs_path *path;
struct extent_buffer *leaf;
int ret;
path = btrfs_alloc_path();
btrfs: implement delayed inode items operation Changelog V5 -> V6: - Fix oom when the memory load is high, by storing the delayed nodes into the root's radix tree, and letting btrfs inodes go. Changelog V4 -> V5: - Fix the race on adding the delayed node to the inode, which is spotted by Chris Mason. - Merge Chris Mason's incremental patch into this patch. - Fix deadlock between readdir() and memory fault, which is reported by Itaru Kitayama. Changelog V3 -> V4: - Fix nested lock, which is reported by Itaru Kitayama, by updating space cache inode in time. Changelog V2 -> V3: - Fix the race between the delayed worker and the task which does delayed items balance, which is reported by Tsutomu Itoh. - Modify the patch address David Sterba's comment. - Fix the bug of the cpu recursion spinlock, reported by Chris Mason Changelog V1 -> V2: - break up the global rb-tree, use a list to manage the delayed nodes, which is created for every directory and file, and used to manage the delayed directory name index items and the delayed inode item. - introduce a worker to deal with the delayed nodes. Compare with Ext3/4, the performance of file creation and deletion on btrfs is very poor. the reason is that btrfs must do a lot of b+ tree insertions, such as inode item, directory name item, directory name index and so on. If we can do some delayed b+ tree insertion or deletion, we can improve the performance, so we made this patch which implemented delayed directory name index insertion/deletion and delayed inode update. Implementation: - introduce a delayed root object into the filesystem, that use two lists to manage the delayed nodes which are created for every file/directory. One is used to manage all the delayed nodes that have delayed items. And the other is used to manage the delayed nodes which is waiting to be dealt with by the work thread. - Every delayed node has two rb-tree, one is used to manage the directory name index which is going to be inserted into b+ tree, and the other is used to manage the directory name index which is going to be deleted from b+ tree. - introduce a worker to deal with the delayed operation. This worker is used to deal with the works of the delayed directory name index items insertion and deletion and the delayed inode update. When the delayed items is beyond the lower limit, we create works for some delayed nodes and insert them into the work queue of the worker, and then go back. When the delayed items is beyond the upper bound, we create works for all the delayed nodes that haven't been dealt with, and insert them into the work queue of the worker, and then wait for that the untreated items is below some threshold value. - When we want to insert a directory name index into b+ tree, we just add the information into the delayed inserting rb-tree. And then we check the number of the delayed items and do delayed items balance. (The balance policy is above.) - When we want to delete a directory name index from the b+ tree, we search it in the inserting rb-tree at first. If we look it up, just drop it. If not, add the key of it into the delayed deleting rb-tree. Similar to the delayed inserting rb-tree, we also check the number of the delayed items and do delayed items balance. (The same to inserting manipulation) - When we want to update the metadata of some inode, we cached the data of the inode into the delayed node. the worker will flush it into the b+ tree after dealing with the delayed insertion and deletion. - We will move the delayed node to the tail of the list after we access the delayed node, By this way, we can cache more delayed items and merge more inode updates. - If we want to commit transaction, we will deal with all the delayed node. - the delayed node will be freed when we free the btrfs inode. - Before we log the inode items, we commit all the directory name index items and the delayed inode update. I did a quick test by the benchmark tool[1] and found we can improve the performance of file creation by ~15%, and file deletion by ~20%. Before applying this patch: Create files: Total files: 50000 Total time: 1.096108 Average time: 0.000022 Delete files: Total files: 50000 Total time: 1.510403 Average time: 0.000030 After applying this patch: Create files: Total files: 50000 Total time: 0.932899 Average time: 0.000019 Delete files: Total files: 50000 Total time: 1.215732 Average time: 0.000024 [1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3 Many thanks for Kitayama-san's help! Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Reviewed-by: David Sterba <dave@jikos.cz> Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com> Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-04-22 10:12:22 +00:00
if (!path)
return -ENOMEM;
path->leave_spinning = 1;
btrfs: implement delayed inode items operation Changelog V5 -> V6: - Fix oom when the memory load is high, by storing the delayed nodes into the root's radix tree, and letting btrfs inodes go. Changelog V4 -> V5: - Fix the race on adding the delayed node to the inode, which is spotted by Chris Mason. - Merge Chris Mason's incremental patch into this patch. - Fix deadlock between readdir() and memory fault, which is reported by Itaru Kitayama. Changelog V3 -> V4: - Fix nested lock, which is reported by Itaru Kitayama, by updating space cache inode in time. Changelog V2 -> V3: - Fix the race between the delayed worker and the task which does delayed items balance, which is reported by Tsutomu Itoh. - Modify the patch address David Sterba's comment. - Fix the bug of the cpu recursion spinlock, reported by Chris Mason Changelog V1 -> V2: - break up the global rb-tree, use a list to manage the delayed nodes, which is created for every directory and file, and used to manage the delayed directory name index items and the delayed inode item. - introduce a worker to deal with the delayed nodes. Compare with Ext3/4, the performance of file creation and deletion on btrfs is very poor. the reason is that btrfs must do a lot of b+ tree insertions, such as inode item, directory name item, directory name index and so on. If we can do some delayed b+ tree insertion or deletion, we can improve the performance, so we made this patch which implemented delayed directory name index insertion/deletion and delayed inode update. Implementation: - introduce a delayed root object into the filesystem, that use two lists to manage the delayed nodes which are created for every file/directory. One is used to manage all the delayed nodes that have delayed items. And the other is used to manage the delayed nodes which is waiting to be dealt with by the work thread. - Every delayed node has two rb-tree, one is used to manage the directory name index which is going to be inserted into b+ tree, and the other is used to manage the directory name index which is going to be deleted from b+ tree. - introduce a worker to deal with the delayed operation. This worker is used to deal with the works of the delayed directory name index items insertion and deletion and the delayed inode update. When the delayed items is beyond the lower limit, we create works for some delayed nodes and insert them into the work queue of the worker, and then go back. When the delayed items is beyond the upper bound, we create works for all the delayed nodes that haven't been dealt with, and insert them into the work queue of the worker, and then wait for that the untreated items is below some threshold value. - When we want to insert a directory name index into b+ tree, we just add the information into the delayed inserting rb-tree. And then we check the number of the delayed items and do delayed items balance. (The balance policy is above.) - When we want to delete a directory name index from the b+ tree, we search it in the inserting rb-tree at first. If we look it up, just drop it. If not, add the key of it into the delayed deleting rb-tree. Similar to the delayed inserting rb-tree, we also check the number of the delayed items and do delayed items balance. (The same to inserting manipulation) - When we want to update the metadata of some inode, we cached the data of the inode into the delayed node. the worker will flush it into the b+ tree after dealing with the delayed insertion and deletion. - We will move the delayed node to the tail of the list after we access the delayed node, By this way, we can cache more delayed items and merge more inode updates. - If we want to commit transaction, we will deal with all the delayed node. - the delayed node will be freed when we free the btrfs inode. - Before we log the inode items, we commit all the directory name index items and the delayed inode update. I did a quick test by the benchmark tool[1] and found we can improve the performance of file creation by ~15%, and file deletion by ~20%. Before applying this patch: Create files: Total files: 50000 Total time: 1.096108 Average time: 0.000022 Delete files: Total files: 50000 Total time: 1.510403 Average time: 0.000030 After applying this patch: Create files: Total files: 50000 Total time: 0.932899 Average time: 0.000019 Delete files: Total files: 50000 Total time: 1.215732 Average time: 0.000024 [1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3 Many thanks for Kitayama-san's help! Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Reviewed-by: David Sterba <dave@jikos.cz> Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com> Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-04-22 10:12:22 +00:00
ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
1);
if (ret) {
if (ret > 0)
ret = -ENOENT;
goto failed;
}
Btrfs: Change btree locking to use explicit blocking points Most of the btrfs metadata operations can be protected by a spinlock, but some operations still need to schedule. So far, btrfs has been using a mutex along with a trylock loop, most of the time it is able to avoid going for the full mutex, so the trylock loop is a big performance gain. This commit is step one for getting rid of the blocking locks entirely. btrfs_tree_lock takes a spinlock, and the code explicitly switches to a blocking lock when it starts an operation that can schedule. We'll be able get rid of the blocking locks in smaller pieces over time. Tracing allows us to find the most common cause of blocking, so we can start with the hot spots first. The basic idea is: btrfs_tree_lock() returns with the spin lock held btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in the extent buffer flags, and then drops the spin lock. The buffer is still considered locked by all of the btrfs code. If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops the spin lock and waits on a wait queue for the blocking bit to go away. Much of the code that needs to set the blocking bit finishes without actually blocking a good percentage of the time. So, an adaptive spin is still used against the blocking bit to avoid very high context switch rates. btrfs_clear_lock_blocking() clears the blocking bit and returns with the spinlock held again. btrfs_tree_unlock() can be called on either blocking or spinning locks, it does the right thing based on the blocking bit. ctree.c has a helper function to set/clear all the locked buffers in a path as blocking. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 14:25:08 +00:00
btrfs_unlock_up_safe(path, 1);
leaf = path->nodes[0];
inode_item = btrfs_item_ptr(leaf, path->slots[0],
btrfs: implement delayed inode items operation Changelog V5 -> V6: - Fix oom when the memory load is high, by storing the delayed nodes into the root's radix tree, and letting btrfs inodes go. Changelog V4 -> V5: - Fix the race on adding the delayed node to the inode, which is spotted by Chris Mason. - Merge Chris Mason's incremental patch into this patch. - Fix deadlock between readdir() and memory fault, which is reported by Itaru Kitayama. Changelog V3 -> V4: - Fix nested lock, which is reported by Itaru Kitayama, by updating space cache inode in time. Changelog V2 -> V3: - Fix the race between the delayed worker and the task which does delayed items balance, which is reported by Tsutomu Itoh. - Modify the patch address David Sterba's comment. - Fix the bug of the cpu recursion spinlock, reported by Chris Mason Changelog V1 -> V2: - break up the global rb-tree, use a list to manage the delayed nodes, which is created for every directory and file, and used to manage the delayed directory name index items and the delayed inode item. - introduce a worker to deal with the delayed nodes. Compare with Ext3/4, the performance of file creation and deletion on btrfs is very poor. the reason is that btrfs must do a lot of b+ tree insertions, such as inode item, directory name item, directory name index and so on. If we can do some delayed b+ tree insertion or deletion, we can improve the performance, so we made this patch which implemented delayed directory name index insertion/deletion and delayed inode update. Implementation: - introduce a delayed root object into the filesystem, that use two lists to manage the delayed nodes which are created for every file/directory. One is used to manage all the delayed nodes that have delayed items. And the other is used to manage the delayed nodes which is waiting to be dealt with by the work thread. - Every delayed node has two rb-tree, one is used to manage the directory name index which is going to be inserted into b+ tree, and the other is used to manage the directory name index which is going to be deleted from b+ tree. - introduce a worker to deal with the delayed operation. This worker is used to deal with the works of the delayed directory name index items insertion and deletion and the delayed inode update. When the delayed items is beyond the lower limit, we create works for some delayed nodes and insert them into the work queue of the worker, and then go back. When the delayed items is beyond the upper bound, we create works for all the delayed nodes that haven't been dealt with, and insert them into the work queue of the worker, and then wait for that the untreated items is below some threshold value. - When we want to insert a directory name index into b+ tree, we just add the information into the delayed inserting rb-tree. And then we check the number of the delayed items and do delayed items balance. (The balance policy is above.) - When we want to delete a directory name index from the b+ tree, we search it in the inserting rb-tree at first. If we look it up, just drop it. If not, add the key of it into the delayed deleting rb-tree. Similar to the delayed inserting rb-tree, we also check the number of the delayed items and do delayed items balance. (The same to inserting manipulation) - When we want to update the metadata of some inode, we cached the data of the inode into the delayed node. the worker will flush it into the b+ tree after dealing with the delayed insertion and deletion. - We will move the delayed node to the tail of the list after we access the delayed node, By this way, we can cache more delayed items and merge more inode updates. - If we want to commit transaction, we will deal with all the delayed node. - the delayed node will be freed when we free the btrfs inode. - Before we log the inode items, we commit all the directory name index items and the delayed inode update. I did a quick test by the benchmark tool[1] and found we can improve the performance of file creation by ~15%, and file deletion by ~20%. Before applying this patch: Create files: Total files: 50000 Total time: 1.096108 Average time: 0.000022 Delete files: Total files: 50000 Total time: 1.510403 Average time: 0.000030 After applying this patch: Create files: Total files: 50000 Total time: 0.932899 Average time: 0.000019 Delete files: Total files: 50000 Total time: 1.215732 Average time: 0.000024 [1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3 Many thanks for Kitayama-san's help! Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Reviewed-by: David Sterba <dave@jikos.cz> Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com> Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-04-22 10:12:22 +00:00
struct btrfs_inode_item);
fill_inode_item(trans, leaf, inode_item, inode);
btrfs_mark_buffer_dirty(leaf);
btrfs_set_inode_last_trans(trans, inode);
ret = 0;
failed:
btrfs_free_path(path);
return ret;
}
/*
* copy everything in the in-memory inode into the btree.
*/
noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
struct btrfs_root *root, struct inode *inode)
{
int ret;
/*
* If the inode is a free space inode, we can deadlock during commit
* if we put it into the delayed code.
*
* The data relocation inode should also be directly updated
* without delay
*/
if (!btrfs_is_free_space_inode(root, inode)
&& root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
ret = btrfs_delayed_update_inode(trans, root, inode);
if (!ret)
btrfs_set_inode_last_trans(trans, inode);
return ret;
}
return btrfs_update_inode_item(trans, root, inode);
}
static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
struct btrfs_root *root, struct inode *inode)
{
int ret;
ret = btrfs_update_inode(trans, root, inode);
if (ret == -ENOSPC)
return btrfs_update_inode_item(trans, root, inode);
return ret;
}
/*
* unlink helper that gets used here in inode.c and in the tree logging
* recovery code. It remove a link in a directory with a given name, and
* also drops the back refs in the inode to the directory
*/
static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
struct btrfs_root *root,
struct inode *dir, struct inode *inode,
const char *name, int name_len)
{
struct btrfs_path *path;
int ret = 0;
struct extent_buffer *leaf;
struct btrfs_dir_item *di;
struct btrfs_key key;
u64 index;
u64 ino = btrfs_ino(inode);
u64 dir_ino = btrfs_ino(dir);
path = btrfs_alloc_path();
if (!path) {
ret = -ENOMEM;
goto out;
}
path->leave_spinning = 1;
di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
name, name_len, -1);
if (IS_ERR(di)) {
ret = PTR_ERR(di);
goto err;
}
if (!di) {
ret = -ENOENT;
goto err;
}
leaf = path->nodes[0];
btrfs_dir_item_key_to_cpu(leaf, di, &key);
ret = btrfs_delete_one_dir_name(trans, root, path, di);
if (ret)
goto err;
btrfs_release_path(path);
ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
dir_ino, &index);
if (ret) {
printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
"inode %llu parent %llu\n", name_len, name,
(unsigned long long)ino, (unsigned long long)dir_ino);
goto err;
}
btrfs: implement delayed inode items operation Changelog V5 -> V6: - Fix oom when the memory load is high, by storing the delayed nodes into the root's radix tree, and letting btrfs inodes go. Changelog V4 -> V5: - Fix the race on adding the delayed node to the inode, which is spotted by Chris Mason. - Merge Chris Mason's incremental patch into this patch. - Fix deadlock between readdir() and memory fault, which is reported by Itaru Kitayama. Changelog V3 -> V4: - Fix nested lock, which is reported by Itaru Kitayama, by updating space cache inode in time. Changelog V2 -> V3: - Fix the race between the delayed worker and the task which does delayed items balance, which is reported by Tsutomu Itoh. - Modify the patch address David Sterba's comment. - Fix the bug of the cpu recursion spinlock, reported by Chris Mason Changelog V1 -> V2: - break up the global rb-tree, use a list to manage the delayed nodes, which is created for every directory and file, and used to manage the delayed directory name index items and the delayed inode item. - introduce a worker to deal with the delayed nodes. Compare with Ext3/4, the performance of file creation and deletion on btrfs is very poor. the reason is that btrfs must do a lot of b+ tree insertions, such as inode item, directory name item, directory name index and so on. If we can do some delayed b+ tree insertion or deletion, we can improve the performance, so we made this patch which implemented delayed directory name index insertion/deletion and delayed inode update. Implementation: - introduce a delayed root object into the filesystem, that use two lists to manage the delayed nodes which are created for every file/directory. One is used to manage all the delayed nodes that have delayed items. And the other is used to manage the delayed nodes which is waiting to be dealt with by the work thread. - Every delayed node has two rb-tree, one is used to manage the directory name index which is going to be inserted into b+ tree, and the other is used to manage the directory name index which is going to be deleted from b+ tree. - introduce a worker to deal with the delayed operation. This worker is used to deal with the works of the delayed directory name index items insertion and deletion and the delayed inode update. When the delayed items is beyond the lower limit, we create works for some delayed nodes and insert them into the work queue of the worker, and then go back. When the delayed items is beyond the upper bound, we create works for all the delayed nodes that haven't been dealt with, and insert them into the work queue of the worker, and then wait for that the untreated items is below some threshold value. - When we want to insert a directory name index into b+ tree, we just add the information into the delayed inserting rb-tree. And then we check the number of the delayed items and do delayed items balance. (The balance policy is above.) - When we want to delete a directory name index from the b+ tree, we search it in the inserting rb-tree at first. If we look it up, just drop it. If not, add the key of it into the delayed deleting rb-tree. Similar to the delayed inserting rb-tree, we also check the number of the delayed items and do delayed items balance. (The same to inserting manipulation) - When we want to update the metadata of some inode, we cached the data of the inode into the delayed node. the worker will flush it into the b+ tree after dealing with the delayed insertion and deletion. - We will move the delayed node to the tail of the list after we access the delayed node, By this way, we can cache more delayed items and merge more inode updates. - If we want to commit transaction, we will deal with all the delayed node. - the delayed node will be freed when we free the btrfs inode. - Before we log the inode items, we commit all the directory name index items and the delayed inode update. I did a quick test by the benchmark tool[1] and found we can improve the performance of file creation by ~15%, and file deletion by ~20%. Before applying this patch: Create files: Total files: 50000 Total time: 1.096108 Average time: 0.000022 Delete files: Total files: 50000 Total time: 1.510403 Average time: 0.000030 After applying this patch: Create files: Total files: 50000 Total time: 0.932899 Average time: 0.000019 Delete files: Total files: 50000 Total time: 1.215732 Average time: 0.000024 [1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3 Many thanks for Kitayama-san's help! Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Reviewed-by: David Sterba <dave@jikos.cz> Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com> Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-04-22 10:12:22 +00:00
ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
if (ret)
goto err;
ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
inode, dir_ino);
BUG_ON(ret != 0 && ret != -ENOENT);
ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
dir, index);
if (ret == -ENOENT)
ret = 0;
err:
btrfs_free_path(path);
if (ret)
goto out;
btrfs_i_size_write(dir, dir->i_size - name_len * 2);
inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
btrfs_update_inode(trans, root, dir);
out:
return ret;
}
int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
struct btrfs_root *root,
struct inode *dir, struct inode *inode,
const char *name, int name_len)
{
int ret;
ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
if (!ret) {
btrfs_drop_nlink(inode);
ret = btrfs_update_inode(trans, root, inode);
}
return ret;
}
/* helper to check if there is any shared block in the path */
static int check_path_shared(struct btrfs_root *root,
struct btrfs_path *path)
{
struct extent_buffer *eb;
int level;
u64 refs = 1;
for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
int ret;
if (!path->nodes[level])
break;
eb = path->nodes[level];
if (!btrfs_block_can_be_shared(root, eb))
continue;
ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
&refs, NULL);
if (refs > 1)
return 1;
}
return 0;
}
/*
* helper to start transaction for unlink and rmdir.
*
* unlink and rmdir are special in btrfs, they do not always free space.
* so in enospc case, we should make sure they will free space before
* allowing them to use the global metadata reservation.
*/
static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
struct dentry *dentry)
{
struct btrfs_trans_handle *trans;
struct btrfs_root *root = BTRFS_I(dir)->root;
struct btrfs_path *path;
struct btrfs_inode_ref *ref;
struct btrfs_dir_item *di;
struct inode *inode = dentry->d_inode;
u64 index;
int check_link = 1;
int err = -ENOSPC;
int ret;
u64 ino = btrfs_ino(inode);
u64 dir_ino = btrfs_ino(dir);
/*
* 1 for the possible orphan item
* 1 for the dir item
* 1 for the dir index
* 1 for the inode ref
* 1 for the inode ref in the tree log
* 2 for the dir entries in the log
* 1 for the inode
*/
trans = btrfs_start_transaction(root, 8);
if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
return trans;
if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
return ERR_PTR(-ENOSPC);
/* check if there is someone else holds reference */
if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
return ERR_PTR(-ENOSPC);
if (atomic_read(&inode->i_count) > 2)
return ERR_PTR(-ENOSPC);
if (xchg(&root->fs_info->enospc_unlink, 1))
return ERR_PTR(-ENOSPC);
path = btrfs_alloc_path();
if (!path) {
root->fs_info->enospc_unlink = 0;
return ERR_PTR(-ENOMEM);
}
/* 1 for the orphan item */
trans = btrfs_start_transaction(root, 1);
if (IS_ERR(trans)) {
btrfs_free_path(path);
root->fs_info->enospc_unlink = 0;
return trans;
}
path->skip_locking = 1;
path->search_commit_root = 1;
ret = btrfs_lookup_inode(trans, root, path,
&BTRFS_I(dir)->location, 0);
if (ret < 0) {
err = ret;
goto out;
}
if (ret == 0) {
if (check_path_shared(root, path))
goto out;
} else {
check_link = 0;
}
btrfs_release_path(path);
ret = btrfs_lookup_inode(trans, root, path,
&BTRFS_I(inode)->location, 0);
if (ret < 0) {
err = ret;
goto out;
}
if (ret == 0) {
if (check_path_shared(root, path))
goto out;
} else {
check_link = 0;
}
btrfs_release_path(path);
if (ret == 0 && S_ISREG(inode->i_mode)) {
ret = btrfs_lookup_file_extent(trans, root, path,
ino, (u64)-1, 0);
if (ret < 0) {
err = ret;
goto out;
}
BUG_ON(ret == 0);
if (check_path_shared(root, path))
goto out;
btrfs_release_path(path);
}
if (!check_link) {
err = 0;
goto out;
}
di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
dentry->d_name.name, dentry->d_name.len, 0);
if (IS_ERR(di)) {
err = PTR_ERR(di);
goto out;
}
if (di) {
if (check_path_shared(root, path))
goto out;
} else {
err = 0;
goto out;
}
btrfs_release_path(path);
ref = btrfs_lookup_inode_ref(trans, root, path,
dentry->d_name.name, dentry->d_name.len,
ino, dir_ino, 0);
if (IS_ERR(ref)) {
err = PTR_ERR(ref);
goto out;
}
BUG_ON(!ref);
if (check_path_shared(root, path))
goto out;
index = btrfs_inode_ref_index(path->nodes[0], ref);
btrfs_release_path(path);
btrfs: implement delayed inode items operation Changelog V5 -> V6: - Fix oom when the memory load is high, by storing the delayed nodes into the root's radix tree, and letting btrfs inodes go. Changelog V4 -> V5: - Fix the race on adding the delayed node to the inode, which is spotted by Chris Mason. - Merge Chris Mason's incremental patch into this patch. - Fix deadlock between readdir() and memory fault, which is reported by Itaru Kitayama. Changelog V3 -> V4: - Fix nested lock, which is reported by Itaru Kitayama, by updating space cache inode in time. Changelog V2 -> V3: - Fix the race between the delayed worker and the task which does delayed items balance, which is reported by Tsutomu Itoh. - Modify the patch address David Sterba's comment. - Fix the bug of the cpu recursion spinlock, reported by Chris Mason Changelog V1 -> V2: - break up the global rb-tree, use a list to manage the delayed nodes, which is created for every directory and file, and used to manage the delayed directory name index items and the delayed inode item. - introduce a worker to deal with the delayed nodes. Compare with Ext3/4, the performance of file creation and deletion on btrfs is very poor. the reason is that btrfs must do a lot of b+ tree insertions, such as inode item, directory name item, directory name index and so on. If we can do some delayed b+ tree insertion or deletion, we can improve the performance, so we made this patch which implemented delayed directory name index insertion/deletion and delayed inode update. Implementation: - introduce a delayed root object into the filesystem, that use two lists to manage the delayed nodes which are created for every file/directory. One is used to manage all the delayed nodes that have delayed items. And the other is used to manage the delayed nodes which is waiting to be dealt with by the work thread. - Every delayed node has two rb-tree, one is used to manage the directory name index which is going to be inserted into b+ tree, and the other is used to manage the directory name index which is going to be deleted from b+ tree. - introduce a worker to deal with the delayed operation. This worker is used to deal with the works of the delayed directory name index items insertion and deletion and the delayed inode update. When the delayed items is beyond the lower limit, we create works for some delayed nodes and insert them into the work queue of the worker, and then go back. When the delayed items is beyond the upper bound, we create works for all the delayed nodes that haven't been dealt with, and insert them into the work queue of the worker, and then wait for that the untreated items is below some threshold value. - When we want to insert a directory name index into b+ tree, we just add the information into the delayed inserting rb-tree. And then we check the number of the delayed items and do delayed items balance. (The balance policy is above.) - When we want to delete a directory name index from the b+ tree, we search it in the inserting rb-tree at first. If we look it up, just drop it. If not, add the key of it into the delayed deleting rb-tree. Similar to the delayed inserting rb-tree, we also check the number of the delayed items and do delayed items balance. (The same to inserting manipulation) - When we want to update the metadata of some inode, we cached the data of the inode into the delayed node. the worker will flush it into the b+ tree after dealing with the delayed insertion and deletion. - We will move the delayed node to the tail of the list after we access the delayed node, By this way, we can cache more delayed items and merge more inode updates. - If we want to commit transaction, we will deal with all the delayed node. - the delayed node will be freed when we free the btrfs inode. - Before we log the inode items, we commit all the directory name index items and the delayed inode update. I did a quick test by the benchmark tool[1] and found we can improve the performance of file creation by ~15%, and file deletion by ~20%. Before applying this patch: Create files: Total files: 50000 Total time: 1.096108 Average time: 0.000022 Delete files: Total files: 50000 Total time: 1.510403 Average time: 0.000030 After applying this patch: Create files: Total files: 50000 Total time: 0.932899 Average time: 0.000019 Delete files: Total files: 50000 Total time: 1.215732 Average time: 0.000024 [1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3 Many thanks for Kitayama-san's help! Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Reviewed-by: David Sterba <dave@jikos.cz> Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com> Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-04-22 10:12:22 +00:00
/*
* This is a commit root search, if we can lookup inode item and other
* relative items in the commit root, it means the transaction of
* dir/file creation has been committed, and the dir index item that we
* delay to insert has also been inserted into the commit root. So
* we needn't worry about the delayed insertion of the dir index item
* here.
*/
di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
dentry->d_name.name, dentry->d_name.len, 0);
if (IS_ERR(di)) {
err = PTR_ERR(di);
goto out;
}
BUG_ON(ret == -ENOENT);
if (check_path_shared(root, path))
goto out;
err = 0;
out:
btrfs_free_path(path);
/* Migrate the orphan reservation over */
if (!err)
err = btrfs_block_rsv_migrate(trans->block_rsv,
&root->fs_info->global_block_rsv,
trans->bytes_reserved);
if (err) {
btrfs_end_transaction(trans, root);
root->fs_info->enospc_unlink = 0;
return ERR_PTR(err);
}
trans->block_rsv = &root->fs_info->global_block_rsv;
return trans;
}
static void __unlink_end_trans(struct btrfs_trans_handle *trans,
struct btrfs_root *root)
{
if (trans->block_rsv == &root->fs_info->global_block_rsv) {
btrfs_block_rsv_release(root, trans->block_rsv,
trans->bytes_reserved);
trans->block_rsv = &root->fs_info->trans_block_rsv;
BUG_ON(!root->fs_info->enospc_unlink);
root->fs_info->enospc_unlink = 0;
}
btrfs_end_transaction(trans, root);
}
static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
{
struct btrfs_root *root = BTRFS_I(dir)->root;
struct btrfs_trans_handle *trans;
struct inode *inode = dentry->d_inode;
int ret;
unsigned long nr = 0;
trans = __unlink_start_trans(dir, dentry);
if (IS_ERR(trans))
return PTR_ERR(trans);
Btrfs: tree logging unlink/rename fixes The tree logging code allows individual files or directories to be logged without including operations on other files and directories in the FS. It tries to commit the minimal set of changes to disk in order to fsync the single file or directory that was sent to fsync or O_SYNC. The tree logging code was allowing files and directories to be unlinked if they were part of a rename operation where only one directory in the rename was in the fsync log. This patch adds a few new rules to the tree logging. 1) on rename or unlink, if the inode being unlinked isn't in the fsync log, we must force a full commit before doing an fsync of the directory where the unlink was done. The commit isn't done during the unlink, but it is forced the next time we try to log the parent directory. Solution: record transid of last unlink/rename per directory when the directory wasn't already logged. For renames this is only done when renaming to a different directory. mkdir foo/some_dir normal commit rename foo/some_dir foo2/some_dir mkdir foo/some_dir fsync foo/some_dir/some_file The fsync above will unlink the original some_dir without recording it in its new location (foo2). After a crash, some_dir will be gone unless the fsync of some_file forces a full commit 2) we must log any new names for any file or dir that is in the fsync log. This way we make sure not to lose files that are unlinked during the same transaction. 2a) we must log any new names for any file or dir during rename when the directory they are being removed from was logged. 2a is actually the more important variant. Without the extra logging a crash might unlink the old name without recreating the new one 3) after a crash, we must go through any directories with a link count of zero and redo the rm -rf mkdir f1/foo normal commit rm -rf f1/foo fsync(f1) The directory f1 was fully removed from the FS, but fsync was never called on f1, only its parent dir. After a crash the rm -rf must be replayed. This must be able to recurse down the entire directory tree. The inode link count fixup code takes care of the ugly details. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-03-24 14:24:20 +00:00
btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
dentry->d_name.name, dentry->d_name.len);
if (ret)
goto out;
if (inode->i_nlink == 0) {
ret = btrfs_orphan_add(trans, inode);
if (ret)
goto out;
}
out:
nr = trans->blocks_used;
__unlink_end_trans(trans, root);
btrfs_btree_balance_dirty(root, nr);
return ret;
}
int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
struct btrfs_root *root,
struct inode *dir, u64 objectid,
const char *name, int name_len)
{
struct btrfs_path *path;
struct extent_buffer *leaf;
struct btrfs_dir_item *di;
struct btrfs_key key;
u64 index;
int ret;
u64 dir_ino = btrfs_ino(dir);
path = btrfs_alloc_path();
if (!path)
return -ENOMEM;
di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
name, name_len, -1);
BUG_ON(IS_ERR_OR_NULL(di));
leaf = path->nodes[0];
btrfs_dir_item_key_to_cpu(leaf, di, &key);
WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
ret = btrfs_delete_one_dir_name(trans, root, path, di);
BUG_ON(ret);
btrfs_release_path(path);
ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
objectid, root->root_key.objectid,
dir_ino, &index, name, name_len);
if (ret < 0) {
BUG_ON(ret != -ENOENT);
di = btrfs_search_dir_index_item(root, path, dir_ino,
name, name_len);
BUG_ON(IS_ERR_OR_NULL(di));
leaf = path->nodes[0];
btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
btrfs_release_path(path);
index = key.offset;
}
btrfs_release_path(path);
btrfs: implement delayed inode items operation Changelog V5 -> V6: - Fix oom when the memory load is high, by storing the delayed nodes into the root's radix tree, and letting btrfs inodes go. Changelog V4 -> V5: - Fix the race on adding the delayed node to the inode, which is spotted by Chris Mason. - Merge Chris Mason's incremental patch into this patch. - Fix deadlock between readdir() and memory fault, which is reported by Itaru Kitayama. Changelog V3 -> V4: - Fix nested lock, which is reported by Itaru Kitayama, by updating space cache inode in time. Changelog V2 -> V3: - Fix the race between the delayed worker and the task which does delayed items balance, which is reported by Tsutomu Itoh. - Modify the patch address David Sterba's comment. - Fix the bug of the cpu recursion spinlock, reported by Chris Mason Changelog V1 -> V2: - break up the global rb-tree, use a list to manage the delayed nodes, which is created for every directory and file, and used to manage the delayed directory name index items and the delayed inode item. - introduce a worker to deal with the delayed nodes. Compare with Ext3/4, the performance of file creation and deletion on btrfs is very poor. the reason is that btrfs must do a lot of b+ tree insertions, such as inode item, directory name item, directory name index and so on. If we can do some delayed b+ tree insertion or deletion, we can improve the performance, so we made this patch which implemented delayed directory name index insertion/deletion and delayed inode update. Implementation: - introduce a delayed root object into the filesystem, that use two lists to manage the delayed nodes which are created for every file/directory. One is used to manage all the delayed nodes that have delayed items. And the other is used to manage the delayed nodes which is waiting to be dealt with by the work thread. - Every delayed node has two rb-tree, one is used to manage the directory name index which is going to be inserted into b+ tree, and the other is used to manage the directory name index which is going to be deleted from b+ tree. - introduce a worker to deal with the delayed operation. This worker is used to deal with the works of the delayed directory name index items insertion and deletion and the delayed inode update. When the delayed items is beyond the lower limit, we create works for some delayed nodes and insert them into the work queue of the worker, and then go back. When the delayed items is beyond the upper bound, we create works for all the delayed nodes that haven't been dealt with, and insert them into the work queue of the worker, and then wait for that the untreated items is below some threshold value. - When we want to insert a directory name index into b+ tree, we just add the information into the delayed inserting rb-tree. And then we check the number of the delayed items and do delayed items balance. (The balance policy is above.) - When we want to delete a directory name index from the b+ tree, we search it in the inserting rb-tree at first. If we look it up, just drop it. If not, add the key of it into the delayed deleting rb-tree. Similar to the delayed inserting rb-tree, we also check the number of the delayed items and do delayed items balance. (The same to inserting manipulation) - When we want to update the metadata of some inode, we cached the data of the inode into the delayed node. the worker will flush it into the b+ tree after dealing with the delayed insertion and deletion. - We will move the delayed node to the tail of the list after we access the delayed node, By this way, we can cache more delayed items and merge more inode updates. - If we want to commit transaction, we will deal with all the delayed node. - the delayed node will be freed when we free the btrfs inode. - Before we log the inode items, we commit all the directory name index items and the delayed inode update. I did a quick test by the benchmark tool[1] and found we can improve the performance of file creation by ~15%, and file deletion by ~20%. Before applying this patch: Create files: Total files: 50000 Total time: 1.096108 Average time: 0.000022 Delete files: Total files: 50000 Total time: 1.510403 Average time: 0.000030 After applying this patch: Create files: Total files: 50000 Total time: 0.932899 Average time: 0.000019 Delete files: Total files: 50000 Total time: 1.215732 Average time: 0.000024 [1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3 Many thanks for Kitayama-san's help! Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Reviewed-by: David Sterba <dave@jikos.cz> Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com> Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-04-22 10:12:22 +00:00
ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
BUG_ON(ret);
btrfs_i_size_write(dir, dir->i_size - name_len * 2);
dir->i_mtime = dir->i_ctime = CURRENT_TIME;
ret = btrfs_update_inode(trans, root, dir);
BUG_ON(ret);
btrfs_free_path(path);
return 0;
}
static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
{
struct inode *inode = dentry->d_inode;
int err = 0;
struct btrfs_root *root = BTRFS_I(dir)->root;
struct btrfs_trans_handle *trans;
unsigned long nr = 0;
if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
return -ENOTEMPTY;
trans = __unlink_start_trans(dir, dentry);
if (IS_ERR(trans))
return PTR_ERR(trans);
if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
err = btrfs_unlink_subvol(trans, root, dir,
BTRFS_I(inode)->location.objectid,
dentry->d_name.name,
dentry->d_name.len);
goto out;
}
err = btrfs_orphan_add(trans, inode);
if (err)
goto out;
/* now the directory is empty */
err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
dentry->d_name.name, dentry->d_name.len);
if (!err)
btrfs_i_size_write(inode, 0);
out:
nr = trans->blocks_used;
__unlink_end_trans(trans, root);
btrfs_btree_balance_dirty(root, nr);
return err;
}
/*
* this can truncate away extent items, csum items and directory items.
* It starts at a high offset and removes keys until it can't find
* any higher than new_size
*
* csum items that cross the new i_size are truncated to the new size
* as well.
*
* min_type is the minimum key type to truncate down to. If set to 0, this
* will kill all the items on this inode, including the INODE_ITEM_KEY.
*/
int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
struct btrfs_root *root,
struct inode *inode,
u64 new_size, u32 min_type)
{
struct btrfs_path *path;
struct extent_buffer *leaf;
struct btrfs_file_extent_item *fi;
struct btrfs_key key;
struct btrfs_key found_key;
u64 extent_start = 0;
u64 extent_num_bytes = 0;
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 14:45:14 +00:00
u64 extent_offset = 0;
u64 item_end = 0;
u64 mask = root->sectorsize - 1;
u32 found_type = (u8)-1;
int found_extent;
int del_item;
int pending_del_nr = 0;
int pending_del_slot = 0;
int extent_type = -1;
int ret;
int err = 0;
u64 ino = btrfs_ino(inode);
BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
path = btrfs_alloc_path();
if (!path)
return -ENOMEM;
path->reada = -1;
if (root->ref_cows || root == root->fs_info->tree_root)
btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
btrfs: implement delayed inode items operation Changelog V5 -> V6: - Fix oom when the memory load is high, by storing the delayed nodes into the root's radix tree, and letting btrfs inodes go. Changelog V4 -> V5: - Fix the race on adding the delayed node to the inode, which is spotted by Chris Mason. - Merge Chris Mason's incremental patch into this patch. - Fix deadlock between readdir() and memory fault, which is reported by Itaru Kitayama. Changelog V3 -> V4: - Fix nested lock, which is reported by Itaru Kitayama, by updating space cache inode in time. Changelog V2 -> V3: - Fix the race between the delayed worker and the task which does delayed items balance, which is reported by Tsutomu Itoh. - Modify the patch address David Sterba's comment. - Fix the bug of the cpu recursion spinlock, reported by Chris Mason Changelog V1 -> V2: - break up the global rb-tree, use a list to manage the delayed nodes, which is created for every directory and file, and used to manage the delayed directory name index items and the delayed inode item. - introduce a worker to deal with the delayed nodes. Compare with Ext3/4, the performance of file creation and deletion on btrfs is very poor. the reason is that btrfs must do a lot of b+ tree insertions, such as inode item, directory name item, directory name index and so on. If we can do some delayed b+ tree insertion or deletion, we can improve the performance, so we made this patch which implemented delayed directory name index insertion/deletion and delayed inode update. Implementation: - introduce a delayed root object into the filesystem, that use two lists to manage the delayed nodes which are created for every file/directory. One is used to manage all the delayed nodes that have delayed items. And the other is used to manage the delayed nodes which is waiting to be dealt with by the work thread. - Every delayed node has two rb-tree, one is used to manage the directory name index which is going to be inserted into b+ tree, and the other is used to manage the directory name index which is going to be deleted from b+ tree. - introduce a worker to deal with the delayed operation. This worker is used to deal with the works of the delayed directory name index items insertion and deletion and the delayed inode update. When the delayed items is beyond the lower limit, we create works for some delayed nodes and insert them into the work queue of the worker, and then go back. When the delayed items is beyond the upper bound, we create works for all the delayed nodes that haven't been dealt with, and insert them into the work queue of the worker, and then wait for that the untreated items is below some threshold value. - When we want to insert a directory name index into b+ tree, we just add the information into the delayed inserting rb-tree. And then we check the number of the delayed items and do delayed items balance. (The balance policy is above.) - When we want to delete a directory name index from the b+ tree, we search it in the inserting rb-tree at first. If we look it up, just drop it. If not, add the key of it into the delayed deleting rb-tree. Similar to the delayed inserting rb-tree, we also check the number of the delayed items and do delayed items balance. (The same to inserting manipulation) - When we want to update the metadata of some inode, we cached the data of the inode into the delayed node. the worker will flush it into the b+ tree after dealing with the delayed insertion and deletion. - We will move the delayed node to the tail of the list after we access the delayed node, By this way, we can cache more delayed items and merge more inode updates. - If we want to commit transaction, we will deal with all the delayed node. - the delayed node will be freed when we free the btrfs inode. - Before we log the inode items, we commit all the directory name index items and the delayed inode update. I did a quick test by the benchmark tool[1] and found we can improve the performance of file creation by ~15%, and file deletion by ~20%. Before applying this patch: Create files: Total files: 50000 Total time: 1.096108 Average time: 0.000022 Delete files: Total files: 50000 Total time: 1.510403 Average time: 0.000030 After applying this patch: Create files: Total files: 50000 Total time: 0.932899 Average time: 0.000019 Delete files: Total files: 50000 Total time: 1.215732 Average time: 0.000024 [1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3 Many thanks for Kitayama-san's help! Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Reviewed-by: David Sterba <dave@jikos.cz> Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com> Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-04-22 10:12:22 +00:00
/*
* This function is also used to drop the items in the log tree before
* we relog the inode, so if root != BTRFS_I(inode)->root, it means
* it is used to drop the loged items. So we shouldn't kill the delayed
* items.
*/
if (min_type == 0 && root == BTRFS_I(inode)->root)
btrfs_kill_delayed_inode_items(inode);
key.objectid = ino;
key.offset = (u64)-1;
key.type = (u8)-1;
search_again:
path->leave_spinning = 1;
ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
if (ret < 0) {
err = ret;
goto out;
}
if (ret > 0) {
/* there are no items in the tree for us to truncate, we're
* done
*/
if (path->slots[0] == 0)
goto out;
path->slots[0]--;
}
while (1) {
fi = NULL;
leaf = path->nodes[0];
btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
found_type = btrfs_key_type(&found_key);
if (found_key.objectid != ino)
break;
if (found_type < min_type)
break;
item_end = found_key.offset;
if (found_type == BTRFS_EXTENT_DATA_KEY) {
fi = btrfs_item_ptr(leaf, path->slots[0],
struct btrfs_file_extent_item);
extent_type = btrfs_file_extent_type(leaf, fi);
if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
item_end +=
btrfs_file_extent_num_bytes(leaf, fi);
} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
item_end += btrfs_file_extent_inline_len(leaf,
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
fi);
}
item_end--;
}
if (found_type > min_type) {
del_item = 1;
} else {
if (item_end < new_size)
break;
if (found_key.offset >= new_size)
del_item = 1;
else
del_item = 0;
}
found_extent = 0;
/* FIXME, shrink the extent if the ref count is only 1 */
if (found_type != BTRFS_EXTENT_DATA_KEY)
goto delete;
if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
u64 num_dec;
extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
if (!del_item) {
u64 orig_num_bytes =
btrfs_file_extent_num_bytes(leaf, fi);
extent_num_bytes = new_size -
found_key.offset + root->sectorsize - 1;
extent_num_bytes = extent_num_bytes &
~((u64)root->sectorsize - 1);
btrfs_set_file_extent_num_bytes(leaf, fi,
extent_num_bytes);
num_dec = (orig_num_bytes -
extent_num_bytes);
if (root->ref_cows && extent_start != 0)
inode_sub_bytes(inode, num_dec);
btrfs_mark_buffer_dirty(leaf);
} else {
extent_num_bytes =
btrfs_file_extent_disk_num_bytes(leaf,
fi);
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 14:45:14 +00:00
extent_offset = found_key.offset -
btrfs_file_extent_offset(leaf, fi);
/* FIXME blocksize != 4096 */
num_dec = btrfs_file_extent_num_bytes(leaf, fi);
if (extent_start != 0) {
found_extent = 1;
if (root->ref_cows)
inode_sub_bytes(inode, num_dec);
}
}
} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
/*
* we can't truncate inline items that have had
* special encodings
*/
if (!del_item &&
btrfs_file_extent_compression(leaf, fi) == 0 &&
btrfs_file_extent_encryption(leaf, fi) == 0 &&
btrfs_file_extent_other_encoding(leaf, fi) == 0) {
u32 size = new_size - found_key.offset;
if (root->ref_cows) {
inode_sub_bytes(inode, item_end + 1 -
new_size);
}
size =
btrfs_file_extent_calc_inline_size(size);
btrfs_truncate_item(trans, root, path,
size, 1);
} else if (root->ref_cows) {
inode_sub_bytes(inode, item_end + 1 -
found_key.offset);
}
}
delete:
if (del_item) {
if (!pending_del_nr) {
/* no pending yet, add ourselves */
pending_del_slot = path->slots[0];
pending_del_nr = 1;
} else if (pending_del_nr &&
path->slots[0] + 1 == pending_del_slot) {
/* hop on the pending chunk */
pending_del_nr++;
pending_del_slot = path->slots[0];
} else {
BUG();
}
} else {
break;
}
if (found_extent && (root->ref_cows ||
root == root->fs_info->tree_root)) {
btrfs_set_path_blocking(path);
ret = btrfs_free_extent(trans, root, extent_start,
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 14:45:14 +00:00
extent_num_bytes, 0,
btrfs_header_owner(leaf),
ino, extent_offset, 0);
BUG_ON(ret);
}
if (found_type == BTRFS_INODE_ITEM_KEY)
break;
if (path->slots[0] == 0 ||
path->slots[0] != pending_del_slot) {
if (root->ref_cows &&
BTRFS_I(inode)->location.objectid !=
BTRFS_FREE_INO_OBJECTID) {
err = -EAGAIN;
goto out;
}
if (pending_del_nr) {
ret = btrfs_del_items(trans, root, path,
pending_del_slot,
pending_del_nr);
BUG_ON(ret);
pending_del_nr = 0;
}
btrfs_release_path(path);
goto search_again;
} else {
path->slots[0]--;
}
}
out:
if (pending_del_nr) {
ret = btrfs_del_items(trans, root, path, pending_del_slot,
pending_del_nr);
BUG_ON(ret);
}
btrfs_free_path(path);
return err;
}
/*
* taken from block_truncate_page, but does cow as it zeros out
* any bytes left in the last page in the file.
*/
static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
{
struct inode *inode = mapping->host;
struct btrfs_root *root = BTRFS_I(inode)->root;
struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
struct btrfs_ordered_extent *ordered;
struct extent_state *cached_state = NULL;
char *kaddr;
u32 blocksize = root->sectorsize;
pgoff_t index = from >> PAGE_CACHE_SHIFT;
unsigned offset = from & (PAGE_CACHE_SIZE-1);
struct page *page;
gfp_t mask = btrfs_alloc_write_mask(mapping);
int ret = 0;
u64 page_start;
u64 page_end;
if ((offset & (blocksize - 1)) == 0)
goto out;
ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
if (ret)
goto out;
ret = -ENOMEM;
again:
page = find_or_create_page(mapping, index, mask);
if (!page) {
btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
goto out;
}
page_start = page_offset(page);
page_end = page_start + PAGE_CACHE_SIZE - 1;
if (!PageUptodate(page)) {
ret = btrfs_readpage(NULL, page);
lock_page(page);
if (page->mapping != mapping) {
unlock_page(page);
page_cache_release(page);
goto again;
}
if (!PageUptodate(page)) {
ret = -EIO;
goto out_unlock;
}
}
wait_on_page_writeback(page);
lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
set_page_extent_mapped(page);
ordered = btrfs_lookup_ordered_extent(inode, page_start);
if (ordered) {
unlock_extent_cached(io_tree, page_start, page_end,
&cached_state, GFP_NOFS);
unlock_page(page);
page_cache_release(page);
btrfs_start_ordered_extent(inode, ordered, 1);
btrfs_put_ordered_extent(ordered);
goto again;
}
clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
0, 0, &cached_state, GFP_NOFS);
ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
&cached_state);
Btrfs: proper -ENOSPC handling At the start of a transaction we do a btrfs_reserve_metadata_space() and specify how many items we plan on modifying. Then once we've done our modifications and such, just call btrfs_unreserve_metadata_space() for the same number of items we reserved. For keeping track of metadata needed for data I've had to add an extent_io op for when we merge extents. This lets us track space properly when we are doing sequential writes, so we don't end up reserving way more metadata space than what we need. The only place where the metadata space accounting is not done is in the relocation code. This is because Yan is going to be reworking that code in the near future, so running btrfs-vol -b could still possibly result in a ENOSPC related panic. This patch also turns off the metadata_ratio stuff in order to allow users to more efficiently use their disk space. This patch makes it so we track how much metadata we need for an inode's delayed allocation extents by tracking how many extents are currently waiting for allocation. It introduces two new callbacks for the extent_io tree's, merge_extent_hook and split_extent_hook. These help us keep track of when we merge delalloc extents together and split them up. Reservations are handled prior to any actually dirty'ing occurs, and then we unreserve after we dirty. btrfs_unreserve_metadata_for_delalloc() will make the appropriate unreservations as needed based on the number of reservations we currently have and the number of extents we currently have. Doing the reservation outside of doing any of the actual dirty'ing lets us do things like filemap_flush() the inode to try and force delalloc to happen, or as a last resort actually start allocation on all delalloc inodes in the fs. This has survived dbench, fs_mark and an fsx torture test. Signed-off-by: Josef Bacik <jbacik@redhat.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-09-11 20:12:44 +00:00
if (ret) {
unlock_extent_cached(io_tree, page_start, page_end,
&cached_state, GFP_NOFS);
Btrfs: proper -ENOSPC handling At the start of a transaction we do a btrfs_reserve_metadata_space() and specify how many items we plan on modifying. Then once we've done our modifications and such, just call btrfs_unreserve_metadata_space() for the same number of items we reserved. For keeping track of metadata needed for data I've had to add an extent_io op for when we merge extents. This lets us track space properly when we are doing sequential writes, so we don't end up reserving way more metadata space than what we need. The only place where the metadata space accounting is not done is in the relocation code. This is because Yan is going to be reworking that code in the near future, so running btrfs-vol -b could still possibly result in a ENOSPC related panic. This patch also turns off the metadata_ratio stuff in order to allow users to more efficiently use their disk space. This patch makes it so we track how much metadata we need for an inode's delayed allocation extents by tracking how many extents are currently waiting for allocation. It introduces two new callbacks for the extent_io tree's, merge_extent_hook and split_extent_hook. These help us keep track of when we merge delalloc extents together and split them up. Reservations are handled prior to any actually dirty'ing occurs, and then we unreserve after we dirty. btrfs_unreserve_metadata_for_delalloc() will make the appropriate unreservations as needed based on the number of reservations we currently have and the number of extents we currently have. Doing the reservation outside of doing any of the actual dirty'ing lets us do things like filemap_flush() the inode to try and force delalloc to happen, or as a last resort actually start allocation on all delalloc inodes in the fs. This has survived dbench, fs_mark and an fsx torture test. Signed-off-by: Josef Bacik <jbacik@redhat.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-09-11 20:12:44 +00:00
goto out_unlock;
}
ret = 0;
if (offset != PAGE_CACHE_SIZE) {
kaddr = kmap(page);
memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
flush_dcache_page(page);
kunmap(page);
}
ClearPageChecked(page);
set_page_dirty(page);
unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
GFP_NOFS);
out_unlock:
if (ret)
btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
unlock_page(page);
page_cache_release(page);
out:
return ret;
}
/*
* This function puts in dummy file extents for the area we're creating a hole
* for. So if we are truncating this file to a larger size we need to insert
* these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
* the range between oldsize and size
*/
int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
{
struct btrfs_trans_handle *trans;
struct btrfs_root *root = BTRFS_I(inode)->root;
struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
struct extent_map *em = NULL;
struct extent_state *cached_state = NULL;
u64 mask = root->sectorsize - 1;
u64 hole_start = (oldsize + mask) & ~mask;
u64 block_end = (size + mask) & ~mask;
u64 last_byte;
u64 cur_offset;
u64 hole_size;
Btrfs: proper -ENOSPC handling At the start of a transaction we do a btrfs_reserve_metadata_space() and specify how many items we plan on modifying. Then once we've done our modifications and such, just call btrfs_unreserve_metadata_space() for the same number of items we reserved. For keeping track of metadata needed for data I've had to add an extent_io op for when we merge extents. This lets us track space properly when we are doing sequential writes, so we don't end up reserving way more metadata space than what we need. The only place where the metadata space accounting is not done is in the relocation code. This is because Yan is going to be reworking that code in the near future, so running btrfs-vol -b could still possibly result in a ENOSPC related panic. This patch also turns off the metadata_ratio stuff in order to allow users to more efficiently use their disk space. This patch makes it so we track how much metadata we need for an inode's delayed allocation extents by tracking how many extents are currently waiting for allocation. It introduces two new callbacks for the extent_io tree's, merge_extent_hook and split_extent_hook. These help us keep track of when we merge delalloc extents together and split them up. Reservations are handled prior to any actually dirty'ing occurs, and then we unreserve after we dirty. btrfs_unreserve_metadata_for_delalloc() will make the appropriate unreservations as needed based on the number of reservations we currently have and the number of extents we currently have. Doing the reservation outside of doing any of the actual dirty'ing lets us do things like filemap_flush() the inode to try and force delalloc to happen, or as a last resort actually start allocation on all delalloc inodes in the fs. This has survived dbench, fs_mark and an fsx torture test. Signed-off-by: Josef Bacik <jbacik@redhat.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-09-11 20:12:44 +00:00
int err = 0;
if (size <= hole_start)
return 0;
while (1) {
struct btrfs_ordered_extent *ordered;
btrfs_wait_ordered_range(inode, hole_start,
block_end - hole_start);
lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
&cached_state);
ordered = btrfs_lookup_ordered_extent(inode, hole_start);
if (!ordered)
break;
unlock_extent_cached(io_tree, hole_start, block_end - 1,
&cached_state, GFP_NOFS);
btrfs_put_ordered_extent(ordered);
}
cur_offset = hole_start;
while (1) {
em = btrfs_get_extent(inode, NULL, 0, cur_offset,
block_end - cur_offset, 0);
BUG_ON(IS_ERR_OR_NULL(em));
last_byte = min(extent_map_end(em), block_end);
last_byte = (last_byte + mask) & ~mask;
if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
u64 hint_byte = 0;
hole_size = last_byte - cur_offset;
Btrfs: proper -ENOSPC handling At the start of a transaction we do a btrfs_reserve_metadata_space() and specify how many items we plan on modifying. Then once we've done our modifications and such, just call btrfs_unreserve_metadata_space() for the same number of items we reserved. For keeping track of metadata needed for data I've had to add an extent_io op for when we merge extents. This lets us track space properly when we are doing sequential writes, so we don't end up reserving way more metadata space than what we need. The only place where the metadata space accounting is not done is in the relocation code. This is because Yan is going to be reworking that code in the near future, so running btrfs-vol -b could still possibly result in a ENOSPC related panic. This patch also turns off the metadata_ratio stuff in order to allow users to more efficiently use their disk space. This patch makes it so we track how much metadata we need for an inode's delayed allocation extents by tracking how many extents are currently waiting for allocation. It introduces two new callbacks for the extent_io tree's, merge_extent_hook and split_extent_hook. These help us keep track of when we merge delalloc extents together and split them up. Reservations are handled prior to any actually dirty'ing occurs, and then we unreserve after we dirty. btrfs_unreserve_metadata_for_delalloc() will make the appropriate unreservations as needed based on the number of reservations we currently have and the number of extents we currently have. Doing the reservation outside of doing any of the actual dirty'ing lets us do things like filemap_flush() the inode to try and force delalloc to happen, or as a last resort actually start allocation on all delalloc inodes in the fs. This has survived dbench, fs_mark and an fsx torture test. Signed-off-by: Josef Bacik <jbacik@redhat.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-09-11 20:12:44 +00:00
trans = btrfs_start_transaction(root, 3);
if (IS_ERR(trans)) {
err = PTR_ERR(trans);
Btrfs: proper -ENOSPC handling At the start of a transaction we do a btrfs_reserve_metadata_space() and specify how many items we plan on modifying. Then once we've done our modifications and such, just call btrfs_unreserve_metadata_space() for the same number of items we reserved. For keeping track of metadata needed for data I've had to add an extent_io op for when we merge extents. This lets us track space properly when we are doing sequential writes, so we don't end up reserving way more metadata space than what we need. The only place where the metadata space accounting is not done is in the relocation code. This is because Yan is going to be reworking that code in the near future, so running btrfs-vol -b could still possibly result in a ENOSPC related panic. This patch also turns off the metadata_ratio stuff in order to allow users to more efficiently use their disk space. This patch makes it so we track how much metadata we need for an inode's delayed allocation extents by tracking how many extents are currently waiting for allocation. It introduces two new callbacks for the extent_io tree's, merge_extent_hook and split_extent_hook. These help us keep track of when we merge delalloc extents together and split them up. Reservations are handled prior to any actually dirty'ing occurs, and then we unreserve after we dirty. btrfs_unreserve_metadata_for_delalloc() will make the appropriate unreservations as needed based on the number of reservations we currently have and the number of extents we currently have. Doing the reservation outside of doing any of the actual dirty'ing lets us do things like filemap_flush() the inode to try and force delalloc to happen, or as a last resort actually start allocation on all delalloc inodes in the fs. This has survived dbench, fs_mark and an fsx torture test. Signed-off-by: Josef Bacik <jbacik@redhat.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-09-11 20:12:44 +00:00
break;
}
err = btrfs_drop_extents(trans, inode, cur_offset,
cur_offset + hole_size,
&hint_byte, 1);
if (err) {
btrfs_update_inode(trans, root, inode);
btrfs_end_transaction(trans, root);
break;
}
err = btrfs_insert_file_extent(trans, root,
btrfs_ino(inode), cur_offset, 0,
0, hole_size, 0, hole_size,
0, 0, 0);
if (err) {
btrfs_update_inode(trans, root, inode);
btrfs_end_transaction(trans, root);
break;
}
btrfs_drop_extent_cache(inode, hole_start,
last_byte - 1, 0);
btrfs_update_inode(trans, root, inode);
btrfs_end_transaction(trans, root);
}
free_extent_map(em);
em = NULL;
cur_offset = last_byte;
if (cur_offset >= block_end)
break;
}
free_extent_map(em);
unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
GFP_NOFS);
return err;
}
static int btrfs_setsize(struct inode *inode, loff_t newsize)
{
struct btrfs_root *root = BTRFS_I(inode)->root;
struct btrfs_trans_handle *trans;
loff_t oldsize = i_size_read(inode);
int ret;
if (newsize == oldsize)
return 0;
if (newsize > oldsize) {
truncate_pagecache(inode, oldsize, newsize);
ret = btrfs_cont_expand(inode, oldsize, newsize);
if (ret)
return ret;
trans = btrfs_start_transaction(root, 1);
if (IS_ERR(trans))
return PTR_ERR(trans);
i_size_write(inode, newsize);
btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
ret = btrfs_update_inode(trans, root, inode);
btrfs_end_transaction(trans, root);
} else {
/*
* We're truncating a file that used to have good data down to
* zero. Make sure it gets into the ordered flush list so that
* any new writes get down to disk quickly.
*/
if (newsize == 0)
BTRFS_I(inode)->ordered_data_close = 1;
/* we don't support swapfiles, so vmtruncate shouldn't fail */
truncate_setsize(inode, newsize);
ret = btrfs_truncate(inode);
}
return ret;
}
static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
{
struct inode *inode = dentry->d_inode;
struct btrfs_root *root = BTRFS_I(inode)->root;
int err;
if (btrfs_root_readonly(root))
return -EROFS;
err = inode_change_ok(inode, attr);
if (err)
return err;
Btrfs: add extra flushing for renames and truncates Renames and truncates are both common ways to replace old data with new data. The filesystem can make an effort to make sure the new data is on disk before actually replacing the old data. This is especially important for rename, which many application use as though it were atomic for both the data and the metadata involved. The current btrfs code will happily replace a file that is fully on disk with one that was just created and still has pending IO. If we crash after transaction commit but before the IO is done, we'll end up replacing a good file with a zero length file. The solution used here is to create a list of inodes that need special ordering and force them to disk before the commit is done. This is similar to the ext3 style data=ordering, except it is only done on selected files. Btrfs is able to get away with this because it does not wait on commits very often, even for fsync (which use a sub-commit). For renames, we order the file when it wasn't already on disk and when it is replacing an existing file. Larger files are sent to filemap_flush right away (before the transaction handle is opened). For truncates, we order if the file goes from non-zero size down to zero size. This is a little different, because at the time of the truncate the file has no dirty bytes to order. But, we flag the inode so that it is added to the ordered list on close (via release method). We also immediately add it to the ordered list of the current transaction so that we can try to flush down any writes the application sneaks in before commit. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-03-31 17:27:11 +00:00
if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
err = btrfs_setsize(inode, attr->ia_size);
if (err)
return err;
}
if (attr->ia_valid) {
setattr_copy(inode, attr);
err = btrfs_dirty_inode(inode);
if (!err && attr->ia_valid & ATTR_MODE)
err = btrfs_acl_chmod(inode);
}
return err;
}
void btrfs_evict_inode(struct inode *inode)
{
struct btrfs_trans_handle *trans;
struct btrfs_root *root = BTRFS_I(inode)->root;
struct btrfs_block_rsv *rsv, *global_rsv;
u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
unsigned long nr;
int ret;
Btrfs: add initial tracepoint support for btrfs Tracepoints can provide insight into why btrfs hits bugs and be greatly helpful for debugging, e.g dd-7822 [000] 2121.641088: btrfs_inode_request: root = 5(FS_TREE), gen = 4, ino = 256, blocks = 8, disk_i_size = 0, last_trans = 8, logged_trans = 0 dd-7822 [000] 2121.641100: btrfs_inode_new: root = 5(FS_TREE), gen = 8, ino = 257, blocks = 0, disk_i_size = 0, last_trans = 0, logged_trans = 0 btrfs-transacti-7804 [001] 2146.935420: btrfs_cow_block: root = 2(EXTENT_TREE), refs = 2, orig_buf = 29368320 (orig_level = 0), cow_buf = 29388800 (cow_level = 0) btrfs-transacti-7804 [001] 2146.935473: btrfs_cow_block: root = 1(ROOT_TREE), refs = 2, orig_buf = 29364224 (orig_level = 0), cow_buf = 29392896 (cow_level = 0) btrfs-transacti-7804 [001] 2146.972221: btrfs_transaction_commit: root = 1(ROOT_TREE), gen = 8 flush-btrfs-2-7821 [001] 2155.824210: btrfs_chunk_alloc: root = 3(CHUNK_TREE), offset = 1103101952, size = 1073741824, num_stripes = 1, sub_stripes = 0, type = DATA flush-btrfs-2-7821 [001] 2155.824241: btrfs_cow_block: root = 2(EXTENT_TREE), refs = 2, orig_buf = 29388800 (orig_level = 0), cow_buf = 29396992 (cow_level = 0) flush-btrfs-2-7821 [001] 2155.824255: btrfs_cow_block: root = 4(DEV_TREE), refs = 2, orig_buf = 29372416 (orig_level = 0), cow_buf = 29401088 (cow_level = 0) flush-btrfs-2-7821 [000] 2155.824329: btrfs_cow_block: root = 3(CHUNK_TREE), refs = 2, orig_buf = 20971520 (orig_level = 0), cow_buf = 20975616 (cow_level = 0) btrfs-endio-wri-7800 [001] 2155.898019: btrfs_cow_block: root = 5(FS_TREE), refs = 2, orig_buf = 29384704 (orig_level = 0), cow_buf = 29405184 (cow_level = 0) btrfs-endio-wri-7800 [001] 2155.898043: btrfs_cow_block: root = 7(CSUM_TREE), refs = 2, orig_buf = 29376512 (orig_level = 0), cow_buf = 29409280 (cow_level = 0) Here is what I have added: 1) ordere_extent: btrfs_ordered_extent_add btrfs_ordered_extent_remove btrfs_ordered_extent_start btrfs_ordered_extent_put These provide critical information to understand how ordered_extents are updated. 2) extent_map: btrfs_get_extent extent_map is used in both read and write cases, and it is useful for tracking how btrfs specific IO is running. 3) writepage: __extent_writepage btrfs_writepage_end_io_hook Pages are cirtical resourses and produce a lot of corner cases during writeback, so it is valuable to know how page is written to disk. 4) inode: btrfs_inode_new btrfs_inode_request btrfs_inode_evict These can show where and when a inode is created, when a inode is evicted. 5) sync: btrfs_sync_file btrfs_sync_fs These show sync arguments. 6) transaction: btrfs_transaction_commit In transaction based filesystem, it will be useful to know the generation and who does commit. 7) back reference and cow: btrfs_delayed_tree_ref btrfs_delayed_data_ref btrfs_delayed_ref_head btrfs_cow_block Btrfs natively supports back references, these tracepoints are helpful on understanding btrfs's COW mechanism. 8) chunk: btrfs_chunk_alloc btrfs_chunk_free Chunk is a link between physical offset and logical offset, and stands for space infomation in btrfs, and these are helpful on tracing space things. 9) reserved_extent: btrfs_reserved_extent_alloc btrfs_reserved_extent_free These can show how btrfs uses its space. Signed-off-by: Liu Bo <liubo2009@cn.fujitsu.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-03-24 11:18:59 +00:00
trace_btrfs_inode_evict(inode);
truncate_inode_pages(&inode->i_data, 0);
if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
btrfs_is_free_space_inode(root, inode)))
goto no_delete;
if (is_bad_inode(inode)) {
btrfs_orphan_del(NULL, inode);
goto no_delete;
}
/* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
btrfs_wait_ordered_range(inode, 0, (u64)-1);
if (root->fs_info->log_root_recovering) {
BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
goto no_delete;
}
if (inode->i_nlink > 0) {
BUG_ON(btrfs_root_refs(&root->root_item) != 0);
goto no_delete;
}
rsv = btrfs_alloc_block_rsv(root);
if (!rsv) {
btrfs_orphan_del(NULL, inode);
goto no_delete;
}
rsv->size = min_size;
global_rsv = &root->fs_info->global_block_rsv;
btrfs_i_size_write(inode, 0);
/*
* This is a bit simpler than btrfs_truncate since
*
* 1) We've already reserved our space for our orphan item in the
* unlink.
* 2) We're going to delete the inode item, so we don't need to update
* it at all.
*
* So we just need to reserve some slack space in case we add bytes when
* doing the truncate.
*/
while (1) {
ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size);
/*
* Try and steal from the global reserve since we will
* likely not use this space anyway, we want to try as
* hard as possible to get this to work.
*/
if (ret)
ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
if (ret) {
printk(KERN_WARNING "Could not get space for a "
"delete, will truncate on mount %d\n", ret);
btrfs_orphan_del(NULL, inode);
btrfs_free_block_rsv(root, rsv);
goto no_delete;
}
trans = btrfs_start_transaction(root, 0);
if (IS_ERR(trans)) {
btrfs_orphan_del(NULL, inode);
btrfs_free_block_rsv(root, rsv);
goto no_delete;
}
trans->block_rsv = rsv;
ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
if (ret != -EAGAIN)
break;
nr = trans->blocks_used;
btrfs_end_transaction(trans, root);
trans = NULL;
btrfs_btree_balance_dirty(root, nr);
}
btrfs_free_block_rsv(root, rsv);
if (ret == 0) {
trans->block_rsv = root->orphan_block_rsv;
ret = btrfs_orphan_del(trans, inode);
BUG_ON(ret);
}
trans->block_rsv = &root->fs_info->trans_block_rsv;
if (!(root == root->fs_info->tree_root ||
root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
btrfs_return_ino(root, btrfs_ino(inode));
nr = trans->blocks_used;
btrfs_end_transaction(trans, root);
btrfs_btree_balance_dirty(root, nr);
no_delete:
end_writeback(inode);
return;
}
/*
* this returns the key found in the dir entry in the location pointer.
* If no dir entries were found, location->objectid is 0.
*/
static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
struct btrfs_key *location)
{
const char *name = dentry->d_name.name;
int namelen = dentry->d_name.len;
struct btrfs_dir_item *di;
struct btrfs_path *path;
struct btrfs_root *root = BTRFS_I(dir)->root;
int ret = 0;
path = btrfs_alloc_path();
if (!path)
return -ENOMEM;
di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
namelen, 0);
if (IS_ERR(di))
ret = PTR_ERR(di);
if (IS_ERR_OR_NULL(di))
goto out_err;
btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
out:
btrfs_free_path(path);
return ret;
out_err:
location->objectid = 0;
goto out;
}
/*
* when we hit a tree root in a directory, the btrfs part of the inode
* needs to be changed to reflect the root directory of the tree root. This
* is kind of like crossing a mount point.
*/
static int fixup_tree_root_location(struct btrfs_root *root,
struct inode *dir,
struct dentry *dentry,
struct btrfs_key *location,
struct btrfs_root **sub_root)
{
struct btrfs_path *path;
struct btrfs_root *new_root;
struct btrfs_root_ref *ref;
struct extent_buffer *leaf;
int ret;
int err = 0;
path = btrfs_alloc_path();
if (!path) {
err = -ENOMEM;
goto out;
}
err = -ENOENT;
ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
BTRFS_I(dir)->root->root_key.objectid,
location->objectid);
if (ret) {
if (ret < 0)
err = ret;
goto out;
}
leaf = path->nodes[0];
ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
goto out;
ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
(unsigned long)(ref + 1),
dentry->d_name.len);
if (ret)
goto out;
btrfs_release_path(path);
new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
if (IS_ERR(new_root)) {
err = PTR_ERR(new_root);
goto out;
}
if (btrfs_root_refs(&new_root->root_item) == 0) {
err = -ENOENT;
goto out;
}
*sub_root = new_root;
location->objectid = btrfs_root_dirid(&new_root->root_item);
location->type = BTRFS_INODE_ITEM_KEY;
location->offset = 0;
err = 0;
out:
btrfs_free_path(path);
return err;
}
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 14:45:14 +00:00
static void inode_tree_add(struct inode *inode)
{
struct btrfs_root *root = BTRFS_I(inode)->root;
struct btrfs_inode *entry;
struct rb_node **p;
struct rb_node *parent;
u64 ino = btrfs_ino(inode);
again:
p = &root->inode_tree.rb_node;
parent = NULL;
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 14:45:14 +00:00
if (inode_unhashed(inode))
return;
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 14:45:14 +00:00
spin_lock(&root->inode_lock);
while (*p) {
parent = *p;
entry = rb_entry(parent, struct btrfs_inode, rb_node);
if (ino < btrfs_ino(&entry->vfs_inode))
p = &parent->rb_left;
else if (ino > btrfs_ino(&entry->vfs_inode))
p = &parent->rb_right;
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 14:45:14 +00:00
else {
WARN_ON(!(entry->vfs_inode.i_state &
(I_WILL_FREE | I_FREEING)));
rb_erase(parent, &root->inode_tree);
RB_CLEAR_NODE(parent);
spin_unlock(&root->inode_lock);
goto again;
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 14:45:14 +00:00
}
}
rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
spin_unlock(&root->inode_lock);
}
static void inode_tree_del(struct inode *inode)
{
struct btrfs_root *root = BTRFS_I(inode)->root;
int empty = 0;
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 14:45:14 +00:00
spin_lock(&root->inode_lock);
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 14:45:14 +00:00
if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
empty = RB_EMPTY_ROOT(&root->inode_tree);
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 14:45:14 +00:00
}
spin_unlock(&root->inode_lock);
/*
* Free space cache has inodes in the tree root, but the tree root has a
* root_refs of 0, so this could end up dropping the tree root as a
* snapshot, so we need the extra !root->fs_info->tree_root check to
* make sure we don't drop it.
*/
if (empty && btrfs_root_refs(&root->root_item) == 0 &&
root != root->fs_info->tree_root) {
synchronize_srcu(&root->fs_info->subvol_srcu);
spin_lock(&root->inode_lock);
empty = RB_EMPTY_ROOT(&root->inode_tree);
spin_unlock(&root->inode_lock);
if (empty)
btrfs_add_dead_root(root);
}
}
void btrfs_invalidate_inodes(struct btrfs_root *root)
{
struct rb_node *node;
struct rb_node *prev;
struct btrfs_inode *entry;
struct inode *inode;
u64 objectid = 0;
WARN_ON(btrfs_root_refs(&root->root_item) != 0);
spin_lock(&root->inode_lock);
again:
node = root->inode_tree.rb_node;
prev = NULL;
while (node) {
prev = node;
entry = rb_entry(node, struct btrfs_inode, rb_node);
if (objectid < btrfs_ino(&entry->vfs_inode))
node = node->rb_left;
else if (objectid > btrfs_ino(&entry->vfs_inode))
node = node->rb_right;
else
break;
}
if (!node) {
while (prev) {
entry = rb_entry(prev, struct btrfs_inode, rb_node);
if (objectid <= btrfs_ino(&entry->vfs_inode)) {
node = prev;
break;
}
prev = rb_next(prev);
}
}
while (node) {
entry = rb_entry(node, struct btrfs_inode, rb_node);
objectid = btrfs_ino(&entry->vfs_inode) + 1;
inode = igrab(&entry->vfs_inode);
if (inode) {
spin_unlock(&root->inode_lock);
if (atomic_read(&inode->i_count) > 1)
d_prune_aliases(inode);
/*
* btrfs_drop_inode will have it removed from
* the inode cache when its usage count
* hits zero.
*/
iput(inode);
cond_resched();
spin_lock(&root->inode_lock);
goto again;
}
if (cond_resched_lock(&root->inode_lock))
goto again;
node = rb_next(node);
}
spin_unlock(&root->inode_lock);
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 14:45:14 +00:00
}
static int btrfs_init_locked_inode(struct inode *inode, void *p)
{
struct btrfs_iget_args *args = p;
inode->i_ino = args->ino;
BTRFS_I(inode)->root = args->root;
btrfs_set_inode_space_info(args->root, inode);
return 0;
}
static int btrfs_find_actor(struct inode *inode, void *opaque)
{
struct btrfs_iget_args *args = opaque;
return args->ino == btrfs_ino(inode) &&
args->root == BTRFS_I(inode)->root;
}
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 14:45:14 +00:00
static struct inode *btrfs_iget_locked(struct super_block *s,
u64 objectid,
struct btrfs_root *root)
{
struct inode *inode;
struct btrfs_iget_args args;
args.ino = objectid;
args.root = root;
inode = iget5_locked(s, objectid, btrfs_find_actor,
btrfs_init_locked_inode,
(void *)&args);
return inode;
}
/* Get an inode object given its location and corresponding root.
* Returns in *is_new if the inode was read from disk
*/
struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
Btrfs: change how we mount subvolumes This work is in preperation for being able to set a different root as the default mounting root. There is currently a problem with how we mount subvolumes. We cannot currently mount a subvolume of a subvolume, you can only mount subvolumes/snapshots of the default subvolume. So say you take a snapshot of the default subvolume and call it snap1, and then take a snapshot of snap1 and call it snap2, so now you have / /snap1 /snap1/snap2 as your available volumes. Currently you can only mount / and /snap1, you cannot mount /snap1/snap2. To fix this problem instead of passing subvolid=<name> you must pass in subvolid=<treeid>, where <treeid> is the tree id that gets spit out via the subvolume listing you get from the subvolume listing patches (btrfs filesystem list). This allows us to mount /, /snap1 and /snap1/snap2 as the root volume. In addition to the above, we also now read the default dir item in the tree root to get the root key that it points to. For now this just points at what has always been the default subvolme, but later on I plan to change it to point at whatever root you want to be the new default root, so you can just set the default mount and not have to mount with -o subvolid=<treeid>. I tested this out with the above scenario and it worked perfectly. Thanks, mount -o subvol operates inside the selected subvolid. For example: mount -o subvol=snap1,subvolid=256 /dev/xxx /mnt /mnt will have the snap1 directory for the subvolume with id 256. mount -o subvol=snap /dev/xxx /mnt /mnt will be the snap directory of whatever the default subvolume is. Signed-off-by: Josef Bacik <josef@redhat.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-12-04 17:38:27 +00:00
struct btrfs_root *root, int *new)
{
struct inode *inode;
inode = btrfs_iget_locked(s, location->objectid, root);
if (!inode)
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 14:45:14 +00:00
return ERR_PTR(-ENOMEM);
if (inode->i_state & I_NEW) {
BTRFS_I(inode)->root = root;
memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
btrfs_read_locked_inode(inode);
if (!is_bad_inode(inode)) {
inode_tree_add(inode);
unlock_new_inode(inode);
if (new)
*new = 1;
} else {
unlock_new_inode(inode);
iput(inode);
inode = ERR_PTR(-ESTALE);
}
}
return inode;
}
static struct inode *new_simple_dir(struct super_block *s,
struct btrfs_key *key,
struct btrfs_root *root)
{
struct inode *inode = new_inode(s);
if (!inode)
return ERR_PTR(-ENOMEM);
BTRFS_I(inode)->root = root;
memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
BTRFS_I(inode)->dummy_inode = 1;
inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
inode->i_op = &simple_dir_inode_operations;
inode->i_fop = &simple_dir_operations;
inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
return inode;
}
struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
{
struct inode *inode;
struct btrfs_root *root = BTRFS_I(dir)->root;
struct btrfs_root *sub_root = root;
struct btrfs_key location;
int index;
int ret = 0;
if (dentry->d_name.len > BTRFS_NAME_LEN)
return ERR_PTR(-ENAMETOOLONG);
if (unlikely(d_need_lookup(dentry))) {
memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
kfree(dentry->d_fsdata);
dentry->d_fsdata = NULL;
/* This thing is hashed, drop it for now */
d_drop(dentry);
} else {
ret = btrfs_inode_by_name(dir, dentry, &location);
}
if (ret < 0)
return ERR_PTR(ret);
if (location.objectid == 0)
return NULL;
if (location.type == BTRFS_INODE_ITEM_KEY) {
Btrfs: change how we mount subvolumes This work is in preperation for being able to set a different root as the default mounting root. There is currently a problem with how we mount subvolumes. We cannot currently mount a subvolume of a subvolume, you can only mount subvolumes/snapshots of the default subvolume. So say you take a snapshot of the default subvolume and call it snap1, and then take a snapshot of snap1 and call it snap2, so now you have / /snap1 /snap1/snap2 as your available volumes. Currently you can only mount / and /snap1, you cannot mount /snap1/snap2. To fix this problem instead of passing subvolid=<name> you must pass in subvolid=<treeid>, where <treeid> is the tree id that gets spit out via the subvolume listing you get from the subvolume listing patches (btrfs filesystem list). This allows us to mount /, /snap1 and /snap1/snap2 as the root volume. In addition to the above, we also now read the default dir item in the tree root to get the root key that it points to. For now this just points at what has always been the default subvolme, but later on I plan to change it to point at whatever root you want to be the new default root, so you can just set the default mount and not have to mount with -o subvolid=<treeid>. I tested this out with the above scenario and it worked perfectly. Thanks, mount -o subvol operates inside the selected subvolid. For example: mount -o subvol=snap1,subvolid=256 /dev/xxx /mnt /mnt will have the snap1 directory for the subvolume with id 256. mount -o subvol=snap /dev/xxx /mnt /mnt will be the snap directory of whatever the default subvolume is. Signed-off-by: Josef Bacik <josef@redhat.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-12-04 17:38:27 +00:00
inode = btrfs_iget(dir->i_sb, &location, root, NULL);
return inode;
}
BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
index = srcu_read_lock(&root->fs_info->subvol_srcu);
ret = fixup_tree_root_location(root, dir, dentry,
&location, &sub_root);
if (ret < 0) {
if (ret != -ENOENT)
inode = ERR_PTR(ret);
else
inode = new_simple_dir(dir->i_sb, &location, sub_root);
} else {
Btrfs: change how we mount subvolumes This work is in preperation for being able to set a different root as the default mounting root. There is currently a problem with how we mount subvolumes. We cannot currently mount a subvolume of a subvolume, you can only mount subvolumes/snapshots of the default subvolume. So say you take a snapshot of the default subvolume and call it snap1, and then take a snapshot of snap1 and call it snap2, so now you have / /snap1 /snap1/snap2 as your available volumes. Currently you can only mount / and /snap1, you cannot mount /snap1/snap2. To fix this problem instead of passing subvolid=<name> you must pass in subvolid=<treeid>, where <treeid> is the tree id that gets spit out via the subvolume listing you get from the subvolume listing patches (btrfs filesystem list). This allows us to mount /, /snap1 and /snap1/snap2 as the root volume. In addition to the above, we also now read the default dir item in the tree root to get the root key that it points to. For now this just points at what has always been the default subvolme, but later on I plan to change it to point at whatever root you want to be the new default root, so you can just set the default mount and not have to mount with -o subvolid=<treeid>. I tested this out with the above scenario and it worked perfectly. Thanks, mount -o subvol operates inside the selected subvolid. For example: mount -o subvol=snap1,subvolid=256 /dev/xxx /mnt /mnt will have the snap1 directory for the subvolume with id 256. mount -o subvol=snap /dev/xxx /mnt /mnt will be the snap directory of whatever the default subvolume is. Signed-off-by: Josef Bacik <josef@redhat.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-12-04 17:38:27 +00:00
inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
}
srcu_read_unlock(&root->fs_info->subvol_srcu, index);
if (!IS_ERR(inode) && root != sub_root) {
down_read(&root->fs_info->cleanup_work_sem);
if (!(inode->i_sb->s_flags & MS_RDONLY))
ret = btrfs_orphan_cleanup(sub_root);
up_read(&root->fs_info->cleanup_work_sem);
if (ret)
inode = ERR_PTR(ret);
}
return inode;
}
static int btrfs_dentry_delete(const struct dentry *dentry)
{
struct btrfs_root *root;
if (!dentry->d_inode && !IS_ROOT(dentry))
dentry = dentry->d_parent;
if (dentry->d_inode) {
root = BTRFS_I(dentry->d_inode)->root;
if (btrfs_root_refs(&root->root_item) == 0)
return 1;
}
return 0;
}
static void btrfs_dentry_release(struct dentry *dentry)
{
if (dentry->d_fsdata)
kfree(dentry->d_fsdata);
}
static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
struct nameidata *nd)
{
struct dentry *ret;
ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
if (unlikely(d_need_lookup(dentry))) {
spin_lock(&dentry->d_lock);
dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
spin_unlock(&dentry->d_lock);
}
return ret;
}
btrfs: implement delayed inode items operation Changelog V5 -> V6: - Fix oom when the memory load is high, by storing the delayed nodes into the root's radix tree, and letting btrfs inodes go. Changelog V4 -> V5: - Fix the race on adding the delayed node to the inode, which is spotted by Chris Mason. - Merge Chris Mason's incremental patch into this patch. - Fix deadlock between readdir() and memory fault, which is reported by Itaru Kitayama. Changelog V3 -> V4: - Fix nested lock, which is reported by Itaru Kitayama, by updating space cache inode in time. Changelog V2 -> V3: - Fix the race between the delayed worker and the task which does delayed items balance, which is reported by Tsutomu Itoh. - Modify the patch address David Sterba's comment. - Fix the bug of the cpu recursion spinlock, reported by Chris Mason Changelog V1 -> V2: - break up the global rb-tree, use a list to manage the delayed nodes, which is created for every directory and file, and used to manage the delayed directory name index items and the delayed inode item. - introduce a worker to deal with the delayed nodes. Compare with Ext3/4, the performance of file creation and deletion on btrfs is very poor. the reason is that btrfs must do a lot of b+ tree insertions, such as inode item, directory name item, directory name index and so on. If we can do some delayed b+ tree insertion or deletion, we can improve the performance, so we made this patch which implemented delayed directory name index insertion/deletion and delayed inode update. Implementation: - introduce a delayed root object into the filesystem, that use two lists to manage the delayed nodes which are created for every file/directory. One is used to manage all the delayed nodes that have delayed items. And the other is used to manage the delayed nodes which is waiting to be dealt with by the work thread. - Every delayed node has two rb-tree, one is used to manage the directory name index which is going to be inserted into b+ tree, and the other is used to manage the directory name index which is going to be deleted from b+ tree. - introduce a worker to deal with the delayed operation. This worker is used to deal with the works of the delayed directory name index items insertion and deletion and the delayed inode update. When the delayed items is beyond the lower limit, we create works for some delayed nodes and insert them into the work queue of the worker, and then go back. When the delayed items is beyond the upper bound, we create works for all the delayed nodes that haven't been dealt with, and insert them into the work queue of the worker, and then wait for that the untreated items is below some threshold value. - When we want to insert a directory name index into b+ tree, we just add the information into the delayed inserting rb-tree. And then we check the number of the delayed items and do delayed items balance. (The balance policy is above.) - When we want to delete a directory name index from the b+ tree, we search it in the inserting rb-tree at first. If we look it up, just drop it. If not, add the key of it into the delayed deleting rb-tree. Similar to the delayed inserting rb-tree, we also check the number of the delayed items and do delayed items balance. (The same to inserting manipulation) - When we want to update the metadata of some inode, we cached the data of the inode into the delayed node. the worker will flush it into the b+ tree after dealing with the delayed insertion and deletion. - We will move the delayed node to the tail of the list after we access the delayed node, By this way, we can cache more delayed items and merge more inode updates. - If we want to commit transaction, we will deal with all the delayed node. - the delayed node will be freed when we free the btrfs inode. - Before we log the inode items, we commit all the directory name index items and the delayed inode update. I did a quick test by the benchmark tool[1] and found we can improve the performance of file creation by ~15%, and file deletion by ~20%. Before applying this patch: Create files: Total files: 50000 Total time: 1.096108 Average time: 0.000022 Delete files: Total files: 50000 Total time: 1.510403 Average time: 0.000030 After applying this patch: Create files: Total files: 50000 Total time: 0.932899 Average time: 0.000019 Delete files: Total files: 50000 Total time: 1.215732 Average time: 0.000024 [1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3 Many thanks for Kitayama-san's help! Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Reviewed-by: David Sterba <dave@jikos.cz> Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com> Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-04-22 10:12:22 +00:00
unsigned char btrfs_filetype_table[] = {
DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
};
static int btrfs_real_readdir(struct file *filp, void *dirent,
filldir_t filldir)
{
struct inode *inode = filp->f_dentry->d_inode;
struct btrfs_root *root = BTRFS_I(inode)->root;
struct btrfs_item *item;
struct btrfs_dir_item *di;
struct btrfs_key key;
struct btrfs_key found_key;
struct btrfs_path *path;
btrfs: implement delayed inode items operation Changelog V5 -> V6: - Fix oom when the memory load is high, by storing the delayed nodes into the root's radix tree, and letting btrfs inodes go. Changelog V4 -> V5: - Fix the race on adding the delayed node to the inode, which is spotted by Chris Mason. - Merge Chris Mason's incremental patch into this patch. - Fix deadlock between readdir() and memory fault, which is reported by Itaru Kitayama. Changelog V3 -> V4: - Fix nested lock, which is reported by Itaru Kitayama, by updating space cache inode in time. Changelog V2 -> V3: - Fix the race between the delayed worker and the task which does delayed items balance, which is reported by Tsutomu Itoh. - Modify the patch address David Sterba's comment. - Fix the bug of the cpu recursion spinlock, reported by Chris Mason Changelog V1 -> V2: - break up the global rb-tree, use a list to manage the delayed nodes, which is created for every directory and file, and used to manage the delayed directory name index items and the delayed inode item. - introduce a worker to deal with the delayed nodes. Compare with Ext3/4, the performance of file creation and deletion on btrfs is very poor. the reason is that btrfs must do a lot of b+ tree insertions, such as inode item, directory name item, directory name index and so on. If we can do some delayed b+ tree insertion or deletion, we can improve the performance, so we made this patch which implemented delayed directory name index insertion/deletion and delayed inode update. Implementation: - introduce a delayed root object into the filesystem, that use two lists to manage the delayed nodes which are created for every file/directory. One is used to manage all the delayed nodes that have delayed items. And the other is used to manage the delayed nodes which is waiting to be dealt with by the work thread. - Every delayed node has two rb-tree, one is used to manage the directory name index which is going to be inserted into b+ tree, and the other is used to manage the directory name index which is going to be deleted from b+ tree. - introduce a worker to deal with the delayed operation. This worker is used to deal with the works of the delayed directory name index items insertion and deletion and the delayed inode update. When the delayed items is beyond the lower limit, we create works for some delayed nodes and insert them into the work queue of the worker, and then go back. When the delayed items is beyond the upper bound, we create works for all the delayed nodes that haven't been dealt with, and insert them into the work queue of the worker, and then wait for that the untreated items is below some threshold value. - When we want to insert a directory name index into b+ tree, we just add the information into the delayed inserting rb-tree. And then we check the number of the delayed items and do delayed items balance. (The balance policy is above.) - When we want to delete a directory name index from the b+ tree, we search it in the inserting rb-tree at first. If we look it up, just drop it. If not, add the key of it into the delayed deleting rb-tree. Similar to the delayed inserting rb-tree, we also check the number of the delayed items and do delayed items balance. (The same to inserting manipulation) - When we want to update the metadata of some inode, we cached the data of the inode into the delayed node. the worker will flush it into the b+ tree after dealing with the delayed insertion and deletion. - We will move the delayed node to the tail of the list after we access the delayed node, By this way, we can cache more delayed items and merge more inode updates. - If we want to commit transaction, we will deal with all the delayed node. - the delayed node will be freed when we free the btrfs inode. - Before we log the inode items, we commit all the directory name index items and the delayed inode update. I did a quick test by the benchmark tool[1] and found we can improve the performance of file creation by ~15%, and file deletion by ~20%. Before applying this patch: Create files: Total files: 50000 Total time: 1.096108 Average time: 0.000022 Delete files: Total files: 50000 Total time: 1.510403 Average time: 0.000030 After applying this patch: Create files: Total files: 50000 Total time: 0.932899 Average time: 0.000019 Delete files: Total files: 50000 Total time: 1.215732 Average time: 0.000024 [1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3 Many thanks for Kitayama-san's help! Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Reviewed-by: David Sterba <dave@jikos.cz> Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com> Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-04-22 10:12:22 +00:00
struct list_head ins_list;
struct list_head del_list;
struct qstr q;
int ret;
struct extent_buffer *leaf;
int slot;
unsigned char d_type;
int over = 0;
u32 di_cur;
u32 di_total;
u32 di_len;
int key_type = BTRFS_DIR_INDEX_KEY;
char tmp_name[32];
char *name_ptr;
int name_len;
btrfs: implement delayed inode items operation Changelog V5 -> V6: - Fix oom when the memory load is high, by storing the delayed nodes into the root's radix tree, and letting btrfs inodes go. Changelog V4 -> V5: - Fix the race on adding the delayed node to the inode, which is spotted by Chris Mason. - Merge Chris Mason's incremental patch into this patch. - Fix deadlock between readdir() and memory fault, which is reported by Itaru Kitayama. Changelog V3 -> V4: - Fix nested lock, which is reported by Itaru Kitayama, by updating space cache inode in time. Changelog V2 -> V3: - Fix the race between the delayed worker and the task which does delayed items balance, which is reported by Tsutomu Itoh. - Modify the patch address David Sterba's comment. - Fix the bug of the cpu recursion spinlock, reported by Chris Mason Changelog V1 -> V2: - break up the global rb-tree, use a list to manage the delayed nodes, which is created for every directory and file, and used to manage the delayed directory name index items and the delayed inode item. - introduce a worker to deal with the delayed nodes. Compare with Ext3/4, the performance of file creation and deletion on btrfs is very poor. the reason is that btrfs must do a lot of b+ tree insertions, such as inode item, directory name item, directory name index and so on. If we can do some delayed b+ tree insertion or deletion, we can improve the performance, so we made this patch which implemented delayed directory name index insertion/deletion and delayed inode update. Implementation: - introduce a delayed root object into the filesystem, that use two lists to manage the delayed nodes which are created for every file/directory. One is used to manage all the delayed nodes that have delayed items. And the other is used to manage the delayed nodes which is waiting to be dealt with by the work thread. - Every delayed node has two rb-tree, one is used to manage the directory name index which is going to be inserted into b+ tree, and the other is used to manage the directory name index which is going to be deleted from b+ tree. - introduce a worker to deal with the delayed operation. This worker is used to deal with the works of the delayed directory name index items insertion and deletion and the delayed inode update. When the delayed items is beyond the lower limit, we create works for some delayed nodes and insert them into the work queue of the worker, and then go back. When the delayed items is beyond the upper bound, we create works for all the delayed nodes that haven't been dealt with, and insert them into the work queue of the worker, and then wait for that the untreated items is below some threshold value. - When we want to insert a directory name index into b+ tree, we just add the information into the delayed inserting rb-tree. And then we check the number of the delayed items and do delayed items balance. (The balance policy is above.) - When we want to delete a directory name index from the b+ tree, we search it in the inserting rb-tree at first. If we look it up, just drop it. If not, add the key of it into the delayed deleting rb-tree. Similar to the delayed inserting rb-tree, we also check the number of the delayed items and do delayed items balance. (The same to inserting manipulation) - When we want to update the metadata of some inode, we cached the data of the inode into the delayed node. the worker will flush it into the b+ tree after dealing with the delayed insertion and deletion. - We will move the delayed node to the tail of the list after we access the delayed node, By this way, we can cache more delayed items and merge more inode updates. - If we want to commit transaction, we will deal with all the delayed node. - the delayed node will be freed when we free the btrfs inode. - Before we log the inode items, we commit all the directory name index items and the delayed inode update. I did a quick test by the benchmark tool[1] and found we can improve the performance of file creation by ~15%, and file deletion by ~20%. Before applying this patch: Create files: Total files: 50000 Total time: 1.096108 Average time: 0.000022 Delete files: Total files: 50000 Total time: 1.510403 Average time: 0.000030 After applying this patch: Create files: Total files: 50000 Total time: 0.932899 Average time: 0.000019 Delete files: Total files: 50000 Total time: 1.215732 Average time: 0.000024 [1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3 Many thanks for Kitayama-san's help! Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Reviewed-by: David Sterba <dave@jikos.cz> Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com> Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-04-22 10:12:22 +00:00
int is_curr = 0; /* filp->f_pos points to the current index? */
/* FIXME, use a real flag for deciding about the key type */
if (root->fs_info->tree_root == root)
key_type = BTRFS_DIR_ITEM_KEY;
/* special case for "." */
if (filp->f_pos == 0) {
over = filldir(dirent, ".", 1,
filp->f_pos, btrfs_ino(inode), DT_DIR);
if (over)
return 0;
filp->f_pos = 1;
}
/* special case for .., just use the back ref */
if (filp->f_pos == 1) {
u64 pino = parent_ino(filp->f_path.dentry);
over = filldir(dirent, "..", 2,
filp->f_pos, pino, DT_DIR);
if (over)
return 0;
filp->f_pos = 2;
}
path = btrfs_alloc_path();
btrfs: implement delayed inode items operation Changelog V5 -> V6: - Fix oom when the memory load is high, by storing the delayed nodes into the root's radix tree, and letting btrfs inodes go. Changelog V4 -> V5: - Fix the race on adding the delayed node to the inode, which is spotted by Chris Mason. - Merge Chris Mason's incremental patch into this patch. - Fix deadlock between readdir() and memory fault, which is reported by Itaru Kitayama. Changelog V3 -> V4: - Fix nested lock, which is reported by Itaru Kitayama, by updating space cache inode in time. Changelog V2 -> V3: - Fix the race between the delayed worker and the task which does delayed items balance, which is reported by Tsutomu Itoh. - Modify the patch address David Sterba's comment. - Fix the bug of the cpu recursion spinlock, reported by Chris Mason Changelog V1 -> V2: - break up the global rb-tree, use a list to manage the delayed nodes, which is created for every directory and file, and used to manage the delayed directory name index items and the delayed inode item. - introduce a worker to deal with the delayed nodes. Compare with Ext3/4, the performance of file creation and deletion on btrfs is very poor. the reason is that btrfs must do a lot of b+ tree insertions, such as inode item, directory name item, directory name index and so on. If we can do some delayed b+ tree insertion or deletion, we can improve the performance, so we made this patch which implemented delayed directory name index insertion/deletion and delayed inode update. Implementation: - introduce a delayed root object into the filesystem, that use two lists to manage the delayed nodes which are created for every file/directory. One is used to manage all the delayed nodes that have delayed items. And the other is used to manage the delayed nodes which is waiting to be dealt with by the work thread. - Every delayed node has two rb-tree, one is used to manage the directory name index which is going to be inserted into b+ tree, and the other is used to manage the directory name index which is going to be deleted from b+ tree. - introduce a worker to deal with the delayed operation. This worker is used to deal with the works of the delayed directory name index items insertion and deletion and the delayed inode update. When the delayed items is beyond the lower limit, we create works for some delayed nodes and insert them into the work queue of the worker, and then go back. When the delayed items is beyond the upper bound, we create works for all the delayed nodes that haven't been dealt with, and insert them into the work queue of the worker, and then wait for that the untreated items is below some threshold value. - When we want to insert a directory name index into b+ tree, we just add the information into the delayed inserting rb-tree. And then we check the number of the delayed items and do delayed items balance. (The balance policy is above.) - When we want to delete a directory name index from the b+ tree, we search it in the inserting rb-tree at first. If we look it up, just drop it. If not, add the key of it into the delayed deleting rb-tree. Similar to the delayed inserting rb-tree, we also check the number of the delayed items and do delayed items balance. (The same to inserting manipulation) - When we want to update the metadata of some inode, we cached the data of the inode into the delayed node. the worker will flush it into the b+ tree after dealing with the delayed insertion and deletion. - We will move the delayed node to the tail of the list after we access the delayed node, By this way, we can cache more delayed items and merge more inode updates. - If we want to commit transaction, we will deal with all the delayed node. - the delayed node will be freed when we free the btrfs inode. - Before we log the inode items, we commit all the directory name index items and the delayed inode update. I did a quick test by the benchmark tool[1] and found we can improve the performance of file creation by ~15%, and file deletion by ~20%. Before applying this patch: Create files: Total files: 50000 Total time: 1.096108 Average time: 0.000022 Delete files: Total files: 50000 Total time: 1.510403 Average time: 0.000030 After applying this patch: Create files: Total files: 50000 Total time: 0.932899 Average time: 0.000019 Delete files: Total files: 50000 Total time: 1.215732 Average time: 0.000024 [1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3 Many thanks for Kitayama-san's help! Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Reviewed-by: David Sterba <dave@jikos.cz> Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com> Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-04-22 10:12:22 +00:00
if (!path)
return -ENOMEM;
path->reada = 1;
btrfs: implement delayed inode items operation Changelog V5 -> V6: - Fix oom when the memory load is high, by storing the delayed nodes into the root's radix tree, and letting btrfs inodes go. Changelog V4 -> V5: - Fix the race on adding the delayed node to the inode, which is spotted by Chris Mason. - Merge Chris Mason's incremental patch into this patch. - Fix deadlock between readdir() and memory fault, which is reported by Itaru Kitayama. Changelog V3 -> V4: - Fix nested lock, which is reported by Itaru Kitayama, by updating space cache inode in time. Changelog V2 -> V3: - Fix the race between the delayed worker and the task which does delayed items balance, which is reported by Tsutomu Itoh. - Modify the patch address David Sterba's comment. - Fix the bug of the cpu recursion spinlock, reported by Chris Mason Changelog V1 -> V2: - break up the global rb-tree, use a list to manage the delayed nodes, which is created for every directory and file, and used to manage the delayed directory name index items and the delayed inode item. - introduce a worker to deal with the delayed nodes. Compare with Ext3/4, the performance of file creation and deletion on btrfs is very poor. the reason is that btrfs must do a lot of b+ tree insertions, such as inode item, directory name item, directory name index and so on. If we can do some delayed b+ tree insertion or deletion, we can improve the performance, so we made this patch which implemented delayed directory name index insertion/deletion and delayed inode update. Implementation: - introduce a delayed root object into the filesystem, that use two lists to manage the delayed nodes which are created for every file/directory. One is used to manage all the delayed nodes that have delayed items. And the other is used to manage the delayed nodes which is waiting to be dealt with by the work thread. - Every delayed node has two rb-tree, one is used to manage the directory name index which is going to be inserted into b+ tree, and the other is used to manage the directory name index which is going to be deleted from b+ tree. - introduce a worker to deal with the delayed operation. This worker is used to deal with the works of the delayed directory name index items insertion and deletion and the delayed inode update. When the delayed items is beyond the lower limit, we create works for some delayed nodes and insert them into the work queue of the worker, and then go back. When the delayed items is beyond the upper bound, we create works for all the delayed nodes that haven't been dealt with, and insert them into the work queue of the worker, and then wait for that the untreated items is below some threshold value. - When we want to insert a directory name index into b+ tree, we just add the information into the delayed inserting rb-tree. And then we check the number of the delayed items and do delayed items balance. (The balance policy is above.) - When we want to delete a directory name index from the b+ tree, we search it in the inserting rb-tree at first. If we look it up, just drop it. If not, add the key of it into the delayed deleting rb-tree. Similar to the delayed inserting rb-tree, we also check the number of the delayed items and do delayed items balance. (The same to inserting manipulation) - When we want to update the metadata of some inode, we cached the data of the inode into the delayed node. the worker will flush it into the b+ tree after dealing with the delayed insertion and deletion. - We will move the delayed node to the tail of the list after we access the delayed node, By this way, we can cache more delayed items and merge more inode updates. - If we want to commit transaction, we will deal with all the delayed node. - the delayed node will be freed when we free the btrfs inode. - Before we log the inode items, we commit all the directory name index items and the delayed inode update. I did a quick test by the benchmark tool[1] and found we can improve the performance of file creation by ~15%, and file deletion by ~20%. Before applying this patch: Create files: Total files: 50000 Total time: 1.096108 Average time: 0.000022 Delete files: Total files: 50000 Total time: 1.510403 Average time: 0.000030 After applying this patch: Create files: Total files: 50000 Total time: 0.932899 Average time: 0.000019 Delete files: Total files: 50000 Total time: 1.215732 Average time: 0.000024 [1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3 Many thanks for Kitayama-san's help! Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Reviewed-by: David Sterba <dave@jikos.cz> Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com> Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-04-22 10:12:22 +00:00
if (key_type == BTRFS_DIR_INDEX_KEY) {
INIT_LIST_HEAD(&ins_list);
INIT_LIST_HEAD(&del_list);
btrfs_get_delayed_items(inode, &ins_list, &del_list);
}
btrfs_set_key_type(&key, key_type);
key.offset = filp->f_pos;
key.objectid = btrfs_ino(inode);
ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
if (ret < 0)
goto err;
while (1) {
leaf = path->nodes[0];
slot = path->slots[0];
if (slot >= btrfs_header_nritems(leaf)) {
ret = btrfs_next_leaf(root, path);
if (ret < 0)
goto err;
else if (ret > 0)
break;
continue;
}
item = btrfs_item_nr(leaf, slot);
btrfs_item_key_to_cpu(leaf, &found_key, slot);
if (found_key.objectid != key.objectid)
break;
if (btrfs_key_type(&found_key) != key_type)
break;
if (found_key.offset < filp->f_pos)
goto next;
btrfs: implement delayed inode items operation Changelog V5 -> V6: - Fix oom when the memory load is high, by storing the delayed nodes into the root's radix tree, and letting btrfs inodes go. Changelog V4 -> V5: - Fix the race on adding the delayed node to the inode, which is spotted by Chris Mason. - Merge Chris Mason's incremental patch into this patch. - Fix deadlock between readdir() and memory fault, which is reported by Itaru Kitayama. Changelog V3 -> V4: - Fix nested lock, which is reported by Itaru Kitayama, by updating space cache inode in time. Changelog V2 -> V3: - Fix the race between the delayed worker and the task which does delayed items balance, which is reported by Tsutomu Itoh. - Modify the patch address David Sterba's comment. - Fix the bug of the cpu recursion spinlock, reported by Chris Mason Changelog V1 -> V2: - break up the global rb-tree, use a list to manage the delayed nodes, which is created for every directory and file, and used to manage the delayed directory name index items and the delayed inode item. - introduce a worker to deal with the delayed nodes. Compare with Ext3/4, the performance of file creation and deletion on btrfs is very poor. the reason is that btrfs must do a lot of b+ tree insertions, such as inode item, directory name item, directory name index and so on. If we can do some delayed b+ tree insertion or deletion, we can improve the performance, so we made this patch which implemented delayed directory name index insertion/deletion and delayed inode update. Implementation: - introduce a delayed root object into the filesystem, that use two lists to manage the delayed nodes which are created for every file/directory. One is used to manage all the delayed nodes that have delayed items. And the other is used to manage the delayed nodes which is waiting to be dealt with by the work thread. - Every delayed node has two rb-tree, one is used to manage the directory name index which is going to be inserted into b+ tree, and the other is used to manage the directory name index which is going to be deleted from b+ tree. - introduce a worker to deal with the delayed operation. This worker is used to deal with the works of the delayed directory name index items insertion and deletion and the delayed inode update. When the delayed items is beyond the lower limit, we create works for some delayed nodes and insert them into the work queue of the worker, and then go back. When the delayed items is beyond the upper bound, we create works for all the delayed nodes that haven't been dealt with, and insert them into the work queue of the worker, and then wait for that the untreated items is below some threshold value. - When we want to insert a directory name index into b+ tree, we just add the information into the delayed inserting rb-tree. And then we check the number of the delayed items and do delayed items balance. (The balance policy is above.) - When we want to delete a directory name index from the b+ tree, we search it in the inserting rb-tree at first. If we look it up, just drop it. If not, add the key of it into the delayed deleting rb-tree. Similar to the delayed inserting rb-tree, we also check the number of the delayed items and do delayed items balance. (The same to inserting manipulation) - When we want to update the metadata of some inode, we cached the data of the inode into the delayed node. the worker will flush it into the b+ tree after dealing with the delayed insertion and deletion. - We will move the delayed node to the tail of the list after we access the delayed node, By this way, we can cache more delayed items and merge more inode updates. - If we want to commit transaction, we will deal with all the delayed node. - the delayed node will be freed when we free the btrfs inode. - Before we log the inode items, we commit all the directory name index items and the delayed inode update. I did a quick test by the benchmark tool[1] and found we can improve the performance of file creation by ~15%, and file deletion by ~20%. Before applying this patch: Create files: Total files: 50000 Total time: 1.096108 Average time: 0.000022 Delete files: Total files: 50000 Total time: 1.510403 Average time: 0.000030 After applying this patch: Create files: Total files: 50000 Total time: 0.932899 Average time: 0.000019 Delete files: Total files: 50000 Total time: 1.215732 Average time: 0.000024 [1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3 Many thanks for Kitayama-san's help! Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Reviewed-by: David Sterba <dave@jikos.cz> Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com> Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-04-22 10:12:22 +00:00
if (key_type == BTRFS_DIR_INDEX_KEY &&
btrfs_should_delete_dir_index(&del_list,
found_key.offset))
goto next;
filp->f_pos = found_key.offset;
btrfs: implement delayed inode items operation Changelog V5 -> V6: - Fix oom when the memory load is high, by storing the delayed nodes into the root's radix tree, and letting btrfs inodes go. Changelog V4 -> V5: - Fix the race on adding the delayed node to the inode, which is spotted by Chris Mason. - Merge Chris Mason's incremental patch into this patch. - Fix deadlock between readdir() and memory fault, which is reported by Itaru Kitayama. Changelog V3 -> V4: - Fix nested lock, which is reported by Itaru Kitayama, by updating space cache inode in time. Changelog V2 -> V3: - Fix the race between the delayed worker and the task which does delayed items balance, which is reported by Tsutomu Itoh. - Modify the patch address David Sterba's comment. - Fix the bug of the cpu recursion spinlock, reported by Chris Mason Changelog V1 -> V2: - break up the global rb-tree, use a list to manage the delayed nodes, which is created for every directory and file, and used to manage the delayed directory name index items and the delayed inode item. - introduce a worker to deal with the delayed nodes. Compare with Ext3/4, the performance of file creation and deletion on btrfs is very poor. the reason is that btrfs must do a lot of b+ tree insertions, such as inode item, directory name item, directory name index and so on. If we can do some delayed b+ tree insertion or deletion, we can improve the performance, so we made this patch which implemented delayed directory name index insertion/deletion and delayed inode update. Implementation: - introduce a delayed root object into the filesystem, that use two lists to manage the delayed nodes which are created for every file/directory. One is used to manage all the delayed nodes that have delayed items. And the other is used to manage the delayed nodes which is waiting to be dealt with by the work thread. - Every delayed node has two rb-tree, one is used to manage the directory name index which is going to be inserted into b+ tree, and the other is used to manage the directory name index which is going to be deleted from b+ tree. - introduce a worker to deal with the delayed operation. This worker is used to deal with the works of the delayed directory name index items insertion and deletion and the delayed inode update. When the delayed items is beyond the lower limit, we create works for some delayed nodes and insert them into the work queue of the worker, and then go back. When the delayed items is beyond the upper bound, we create works for all the delayed nodes that haven't been dealt with, and insert them into the work queue of the worker, and then wait for that the untreated items is below some threshold value. - When we want to insert a directory name index into b+ tree, we just add the information into the delayed inserting rb-tree. And then we check the number of the delayed items and do delayed items balance. (The balance policy is above.) - When we want to delete a directory name index from the b+ tree, we search it in the inserting rb-tree at first. If we look it up, just drop it. If not, add the key of it into the delayed deleting rb-tree. Similar to the delayed inserting rb-tree, we also check the number of the delayed items and do delayed items balance. (The same to inserting manipulation) - When we want to update the metadata of some inode, we cached the data of the inode into the delayed node. the worker will flush it into the b+ tree after dealing with the delayed insertion and deletion. - We will move the delayed node to the tail of the list after we access the delayed node, By this way, we can cache more delayed items and merge more inode updates. - If we want to commit transaction, we will deal with all the delayed node. - the delayed node will be freed when we free the btrfs inode. - Before we log the inode items, we commit all the directory name index items and the delayed inode update. I did a quick test by the benchmark tool[1] and found we can improve the performance of file creation by ~15%, and file deletion by ~20%. Before applying this patch: Create files: Total files: 50000 Total time: 1.096108 Average time: 0.000022 Delete files: Total files: 50000 Total time: 1.510403 Average time: 0.000030 After applying this patch: Create files: Total files: 50000 Total time: 0.932899 Average time: 0.000019 Delete files: Total files: 50000 Total time: 1.215732 Average time: 0.000024 [1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3 Many thanks for Kitayama-san's help! Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Reviewed-by: David Sterba <dave@jikos.cz> Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com> Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-04-22 10:12:22 +00:00
is_curr = 1;
di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
di_cur = 0;
di_total = btrfs_item_size(leaf, item);
while (di_cur < di_total) {
struct btrfs_key location;
struct dentry *tmp;
if (verify_dir_item(root, leaf, di))
break;
name_len = btrfs_dir_name_len(leaf, di);
if (name_len <= sizeof(tmp_name)) {
name_ptr = tmp_name;
} else {
name_ptr = kmalloc(name_len, GFP_NOFS);
if (!name_ptr) {
ret = -ENOMEM;
goto err;
}
}
read_extent_buffer(leaf, name_ptr,
(unsigned long)(di + 1), name_len);
d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
btrfs_dir_item_key_to_cpu(leaf, di, &location);
q.name = name_ptr;
q.len = name_len;
q.hash = full_name_hash(q.name, q.len);
tmp = d_lookup(filp->f_dentry, &q);
if (!tmp) {
struct btrfs_key *newkey;
newkey = kzalloc(sizeof(struct btrfs_key),
GFP_NOFS);
if (!newkey)
goto no_dentry;
tmp = d_alloc(filp->f_dentry, &q);
if (!tmp) {
kfree(newkey);
dput(tmp);
goto no_dentry;
}
memcpy(newkey, &location,
sizeof(struct btrfs_key));
tmp->d_fsdata = newkey;
tmp->d_flags |= DCACHE_NEED_LOOKUP;
d_rehash(tmp);
dput(tmp);
} else {
dput(tmp);
}
no_dentry:
/* is this a reference to our own snapshot? If so
* skip it
*/
if (location.type == BTRFS_ROOT_ITEM_KEY &&
location.objectid == root->root_key.objectid) {
over = 0;
goto skip;
}
over = filldir(dirent, name_ptr, name_len,
found_key.offset, location.objectid,
d_type);
skip:
if (name_ptr != tmp_name)
kfree(name_ptr);
if (over)
goto nopos;
di_len = btrfs_dir_name_len(leaf, di) +
btrfs_dir_data_len(leaf, di) + sizeof(*di);
di_cur += di_len;
di = (struct btrfs_dir_item *)((char *)di + di_len);
}
next:
path->slots[0]++;
}
btrfs: implement delayed inode items operation Changelog V5 -> V6: - Fix oom when the memory load is high, by storing the delayed nodes into the root's radix tree, and letting btrfs inodes go. Changelog V4 -> V5: - Fix the race on adding the delayed node to the inode, which is spotted by Chris Mason. - Merge Chris Mason's incremental patch into this patch. - Fix deadlock between readdir() and memory fault, which is reported by Itaru Kitayama. Changelog V3 -> V4: - Fix nested lock, which is reported by Itaru Kitayama, by updating space cache inode in time. Changelog V2 -> V3: - Fix the race between the delayed worker and the task which does delayed items balance, which is reported by Tsutomu Itoh. - Modify the patch address David Sterba's comment. - Fix the bug of the cpu recursion spinlock, reported by Chris Mason Changelog V1 -> V2: - break up the global rb-tree, use a list to manage the delayed nodes, which is created for every directory and file, and used to manage the delayed directory name index items and the delayed inode item. - introduce a worker to deal with the delayed nodes. Compare with Ext3/4, the performance of file creation and deletion on btrfs is very poor. the reason is that btrfs must do a lot of b+ tree insertions, such as inode item, directory name item, directory name index and so on. If we can do some delayed b+ tree insertion or deletion, we can improve the performance, so we made this patch which implemented delayed directory name index insertion/deletion and delayed inode update. Implementation: - introduce a delayed root object into the filesystem, that use two lists to manage the delayed nodes which are created for every file/directory. One is used to manage all the delayed nodes that have delayed items. And the other is used to manage the delayed nodes which is waiting to be dealt with by the work thread. - Every delayed node has two rb-tree, one is used to manage the directory name index which is going to be inserted into b+ tree, and the other is used to manage the directory name index which is going to be deleted from b+ tree. - introduce a worker to deal with the delayed operation. This worker is used to deal with the works of the delayed directory name index items insertion and deletion and the delayed inode update. When the delayed items is beyond the lower limit, we create works for some delayed nodes and insert them into the work queue of the worker, and then go back. When the delayed items is beyond the upper bound, we create works for all the delayed nodes that haven't been dealt with, and insert them into the work queue of the worker, and then wait for that the untreated items is below some threshold value. - When we want to insert a directory name index into b+ tree, we just add the information into the delayed inserting rb-tree. And then we check the number of the delayed items and do delayed items balance. (The balance policy is above.) - When we want to delete a directory name index from the b+ tree, we search it in the inserting rb-tree at first. If we look it up, just drop it. If not, add the key of it into the delayed deleting rb-tree. Similar to the delayed inserting rb-tree, we also check the number of the delayed items and do delayed items balance. (The same to inserting manipulation) - When we want to update the metadata of some inode, we cached the data of the inode into the delayed node. the worker will flush it into the b+ tree after dealing with the delayed insertion and deletion. - We will move the delayed node to the tail of the list after we access the delayed node, By this way, we can cache more delayed items and merge more inode updates. - If we want to commit transaction, we will deal with all the delayed node. - the delayed node will be freed when we free the btrfs inode. - Before we log the inode items, we commit all the directory name index items and the delayed inode update. I did a quick test by the benchmark tool[1] and found we can improve the performance of file creation by ~15%, and file deletion by ~20%. Before applying this patch: Create files: Total files: 50000 Total time: 1.096108 Average time: 0.000022 Delete files: Total files: 50000 Total time: 1.510403 Average time: 0.000030 After applying this patch: Create files: Total files: 50000 Total time: 0.932899 Average time: 0.000019 Delete files: Total files: 50000 Total time: 1.215732 Average time: 0.000024 [1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3 Many thanks for Kitayama-san's help! Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Reviewed-by: David Sterba <dave@jikos.cz> Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com> Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-04-22 10:12:22 +00:00
if (key_type == BTRFS_DIR_INDEX_KEY) {
if (is_curr)
filp->f_pos++;
ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
&ins_list);
if (ret)
goto nopos;
}
/* Reached end of directory/root. Bump pos past the last item. */
if (key_type == BTRFS_DIR_INDEX_KEY)
/*
* 32-bit glibc will use getdents64, but then strtol -
* so the last number we can serve is this.
*/
filp->f_pos = 0x7fffffff;
else
filp->f_pos++;
nopos:
ret = 0;
err:
btrfs: implement delayed inode items operation Changelog V5 -> V6: - Fix oom when the memory load is high, by storing the delayed nodes into the root's radix tree, and letting btrfs inodes go. Changelog V4 -> V5: - Fix the race on adding the delayed node to the inode, which is spotted by Chris Mason. - Merge Chris Mason's incremental patch into this patch. - Fix deadlock between readdir() and memory fault, which is reported by Itaru Kitayama. Changelog V3 -> V4: - Fix nested lock, which is reported by Itaru Kitayama, by updating space cache inode in time. Changelog V2 -> V3: - Fix the race between the delayed worker and the task which does delayed items balance, which is reported by Tsutomu Itoh. - Modify the patch address David Sterba's comment. - Fix the bug of the cpu recursion spinlock, reported by Chris Mason Changelog V1 -> V2: - break up the global rb-tree, use a list to manage the delayed nodes, which is created for every directory and file, and used to manage the delayed directory name index items and the delayed inode item. - introduce a worker to deal with the delayed nodes. Compare with Ext3/4, the performance of file creation and deletion on btrfs is very poor. the reason is that btrfs must do a lot of b+ tree insertions, such as inode item, directory name item, directory name index and so on. If we can do some delayed b+ tree insertion or deletion, we can improve the performance, so we made this patch which implemented delayed directory name index insertion/deletion and delayed inode update. Implementation: - introduce a delayed root object into the filesystem, that use two lists to manage the delayed nodes which are created for every file/directory. One is used to manage all the delayed nodes that have delayed items. And the other is used to manage the delayed nodes which is waiting to be dealt with by the work thread. - Every delayed node has two rb-tree, one is used to manage the directory name index which is going to be inserted into b+ tree, and the other is used to manage the directory name index which is going to be deleted from b+ tree. - introduce a worker to deal with the delayed operation. This worker is used to deal with the works of the delayed directory name index items insertion and deletion and the delayed inode update. When the delayed items is beyond the lower limit, we create works for some delayed nodes and insert them into the work queue of the worker, and then go back. When the delayed items is beyond the upper bound, we create works for all the delayed nodes that haven't been dealt with, and insert them into the work queue of the worker, and then wait for that the untreated items is below some threshold value. - When we want to insert a directory name index into b+ tree, we just add the information into the delayed inserting rb-tree. And then we check the number of the delayed items and do delayed items balance. (The balance policy is above.) - When we want to delete a directory name index from the b+ tree, we search it in the inserting rb-tree at first. If we look it up, just drop it. If not, add the key of it into the delayed deleting rb-tree. Similar to the delayed inserting rb-tree, we also check the number of the delayed items and do delayed items balance. (The same to inserting manipulation) - When we want to update the metadata of some inode, we cached the data of the inode into the delayed node. the worker will flush it into the b+ tree after dealing with the delayed insertion and deletion. - We will move the delayed node to the tail of the list after we access the delayed node, By this way, we can cache more delayed items and merge more inode updates. - If we want to commit transaction, we will deal with all the delayed node. - the delayed node will be freed when we free the btrfs inode. - Before we log the inode items, we commit all the directory name index items and the delayed inode update. I did a quick test by the benchmark tool[1] and found we can improve the performance of file creation by ~15%, and file deletion by ~20%. Before applying this patch: Create files: Total files: 50000 Total time: 1.096108 Average time: 0.000022 Delete files: Total files: 50000 Total time: 1.510403 Average time: 0.000030 After applying this patch: Create files: Total files: 50000 Total time: 0.932899 Average time: 0.000019 Delete files: Total files: 50000 Total time: 1.215732 Average time: 0.000024 [1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3 Many thanks for Kitayama-san's help! Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Reviewed-by: David Sterba <dave@jikos.cz> Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com> Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-04-22 10:12:22 +00:00
if (key_type == BTRFS_DIR_INDEX_KEY)
btrfs_put_delayed_items(&ins_list, &del_list);
btrfs_free_path(path);
return ret;
}
int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
{
struct btrfs_root *root = BTRFS_I(inode)->root;
struct btrfs_trans_handle *trans;
int ret = 0;
bool nolock = false;
if (BTRFS_I(inode)->dummy_inode)
return 0;
if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
nolock = true;
if (wbc->sync_mode == WB_SYNC_ALL) {
if (nolock)
trans = btrfs_join_transaction_nolock(root);
else
trans = btrfs_join_transaction(root);
if (IS_ERR(trans))
return PTR_ERR(trans);
if (nolock)
ret = btrfs_end_transaction_nolock(trans, root);
else
ret = btrfs_commit_transaction(trans, root);
}
return ret;
}
/*
* This is somewhat expensive, updating the tree every time the
* inode changes. But, it is most likely to find the inode in cache.
* FIXME, needs more benchmarking...there are no reasons other than performance
* to keep or drop this code.
*/
int btrfs_dirty_inode(struct inode *inode)
{
struct btrfs_root *root = BTRFS_I(inode)->root;
struct btrfs_trans_handle *trans;
int ret;
if (BTRFS_I(inode)->dummy_inode)
return 0;
trans = btrfs_join_transaction(root);
if (IS_ERR(trans))
return PTR_ERR(trans);
ret = btrfs_update_inode(trans, root, inode);
if (ret && ret == -ENOSPC) {
/* whoops, lets try again with the full transaction */
btrfs_end_transaction(trans, root);
trans = btrfs_start_transaction(root, 1);
if (IS_ERR(trans))
return PTR_ERR(trans);
ret = btrfs_update_inode(trans, root, inode);
}
btrfs_end_transaction(trans, root);
btrfs: implement delayed inode items operation Changelog V5 -> V6: - Fix oom when the memory load is high, by storing the delayed nodes into the root's radix tree, and letting btrfs inodes go. Changelog V4 -> V5: - Fix the race on adding the delayed node to the inode, which is spotted by Chris Mason. - Merge Chris Mason's incremental patch into this patch. - Fix deadlock between readdir() and memory fault, which is reported by Itaru Kitayama. Changelog V3 -> V4: - Fix nested lock, which is reported by Itaru Kitayama, by updating space cache inode in time. Changelog V2 -> V3: - Fix the race between the delayed worker and the task which does delayed items balance, which is reported by Tsutomu Itoh. - Modify the patch address David Sterba's comment. - Fix the bug of the cpu recursion spinlock, reported by Chris Mason Changelog V1 -> V2: - break up the global rb-tree, use a list to manage the delayed nodes, which is created for every directory and file, and used to manage the delayed directory name index items and the delayed inode item. - introduce a worker to deal with the delayed nodes. Compare with Ext3/4, the performance of file creation and deletion on btrfs is very poor. the reason is that btrfs must do a lot of b+ tree insertions, such as inode item, directory name item, directory name index and so on. If we can do some delayed b+ tree insertion or deletion, we can improve the performance, so we made this patch which implemented delayed directory name index insertion/deletion and delayed inode update. Implementation: - introduce a delayed root object into the filesystem, that use two lists to manage the delayed nodes which are created for every file/directory. One is used to manage all the delayed nodes that have delayed items. And the other is used to manage the delayed nodes which is waiting to be dealt with by the work thread. - Every delayed node has two rb-tree, one is used to manage the directory name index which is going to be inserted into b+ tree, and the other is used to manage the directory name index which is going to be deleted from b+ tree. - introduce a worker to deal with the delayed operation. This worker is used to deal with the works of the delayed directory name index items insertion and deletion and the delayed inode update. When the delayed items is beyond the lower limit, we create works for some delayed nodes and insert them into the work queue of the worker, and then go back. When the delayed items is beyond the upper bound, we create works for all the delayed nodes that haven't been dealt with, and insert them into the work queue of the worker, and then wait for that the untreated items is below some threshold value. - When we want to insert a directory name index into b+ tree, we just add the information into the delayed inserting rb-tree. And then we check the number of the delayed items and do delayed items balance. (The balance policy is above.) - When we want to delete a directory name index from the b+ tree, we search it in the inserting rb-tree at first. If we look it up, just drop it. If not, add the key of it into the delayed deleting rb-tree. Similar to the delayed inserting rb-tree, we also check the number of the delayed items and do delayed items balance. (The same to inserting manipulation) - When we want to update the metadata of some inode, we cached the data of the inode into the delayed node. the worker will flush it into the b+ tree after dealing with the delayed insertion and deletion. - We will move the delayed node to the tail of the list after we access the delayed node, By this way, we can cache more delayed items and merge more inode updates. - If we want to commit transaction, we will deal with all the delayed node. - the delayed node will be freed when we free the btrfs inode. - Before we log the inode items, we commit all the directory name index items and the delayed inode update. I did a quick test by the benchmark tool[1] and found we can improve the performance of file creation by ~15%, and file deletion by ~20%. Before applying this patch: Create files: Total files: 50000 Total time: 1.096108 Average time: 0.000022 Delete files: Total files: 50000 Total time: 1.510403 Average time: 0.000030 After applying this patch: Create files: Total files: 50000 Total time: 0.932899 Average time: 0.000019 Delete files: Total files: 50000 Total time: 1.215732 Average time: 0.000024 [1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3 Many thanks for Kitayama-san's help! Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Reviewed-by: David Sterba <dave@jikos.cz> Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com> Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-04-22 10:12:22 +00:00
if (BTRFS_I(inode)->delayed_node)
btrfs_balance_delayed_items(root);
return ret;
}
/*
* This is a copy of file_update_time. We need this so we can return error on
* ENOSPC for updating the inode in the case of file write and mmap writes.
*/
int btrfs_update_time(struct file *file)
{
struct inode *inode = file->f_path.dentry->d_inode;
struct timespec now;
int ret;
enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
/* First try to exhaust all avenues to not sync */
if (IS_NOCMTIME(inode))
return 0;
now = current_fs_time(inode->i_sb);
if (!timespec_equal(&inode->i_mtime, &now))
sync_it = S_MTIME;
if (!timespec_equal(&inode->i_ctime, &now))
sync_it |= S_CTIME;
if (IS_I_VERSION(inode))
sync_it |= S_VERSION;
if (!sync_it)
return 0;
/* Finally allowed to write? Takes lock. */
if (mnt_want_write_file(file))
return 0;
/* Only change inode inside the lock region */
if (sync_it & S_VERSION)
inode_inc_iversion(inode);
if (sync_it & S_CTIME)
inode->i_ctime = now;
if (sync_it & S_MTIME)
inode->i_mtime = now;
ret = btrfs_dirty_inode(inode);
if (!ret)
mark_inode_dirty_sync(inode);
mnt_drop_write(file->f_path.mnt);
return ret;
}
/*
* find the highest existing sequence number in a directory
* and then set the in-memory index_cnt variable to reflect
* free sequence numbers
*/
static int btrfs_set_inode_index_count(struct inode *inode)
{
struct btrfs_root *root = BTRFS_I(inode)->root;
struct btrfs_key key, found_key;
struct btrfs_path *path;
struct extent_buffer *leaf;
int ret;
key.objectid = btrfs_ino(inode);
btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
key.offset = (u64)-1;
path = btrfs_alloc_path();
if (!path)
return -ENOMEM;
ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
if (ret < 0)
goto out;
/* FIXME: we should be able to handle this */
if (ret == 0)
goto out;
ret = 0;
/*
* MAGIC NUMBER EXPLANATION:
* since we search a directory based on f_pos we have to start at 2
* since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
* else has to start at 2
*/
if (path->slots[0] == 0) {
BTRFS_I(inode)->index_cnt = 2;
goto out;
}
path->slots[0]--;
leaf = path->nodes[0];
btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
if (found_key.objectid != btrfs_ino(inode) ||
btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
BTRFS_I(inode)->index_cnt = 2;
goto out;
}
BTRFS_I(inode)->index_cnt = found_key.offset + 1;
out:
btrfs_free_path(path);
return ret;
}
/*
* helper to find a free sequence number in a given directory. This current
* code is very simple, later versions will do smarter things in the btree
*/
int btrfs_set_inode_index(struct inode *dir, u64 *index)
{
int ret = 0;
if (BTRFS_I(dir)->index_cnt == (u64)-1) {
btrfs: implement delayed inode items operation Changelog V5 -> V6: - Fix oom when the memory load is high, by storing the delayed nodes into the root's radix tree, and letting btrfs inodes go. Changelog V4 -> V5: - Fix the race on adding the delayed node to the inode, which is spotted by Chris Mason. - Merge Chris Mason's incremental patch into this patch. - Fix deadlock between readdir() and memory fault, which is reported by Itaru Kitayama. Changelog V3 -> V4: - Fix nested lock, which is reported by Itaru Kitayama, by updating space cache inode in time. Changelog V2 -> V3: - Fix the race between the delayed worker and the task which does delayed items balance, which is reported by Tsutomu Itoh. - Modify the patch address David Sterba's comment. - Fix the bug of the cpu recursion spinlock, reported by Chris Mason Changelog V1 -> V2: - break up the global rb-tree, use a list to manage the delayed nodes, which is created for every directory and file, and used to manage the delayed directory name index items and the delayed inode item. - introduce a worker to deal with the delayed nodes. Compare with Ext3/4, the performance of file creation and deletion on btrfs is very poor. the reason is that btrfs must do a lot of b+ tree insertions, such as inode item, directory name item, directory name index and so on. If we can do some delayed b+ tree insertion or deletion, we can improve the performance, so we made this patch which implemented delayed directory name index insertion/deletion and delayed inode update. Implementation: - introduce a delayed root object into the filesystem, that use two lists to manage the delayed nodes which are created for every file/directory. One is used to manage all the delayed nodes that have delayed items. And the other is used to manage the delayed nodes which is waiting to be dealt with by the work thread. - Every delayed node has two rb-tree, one is used to manage the directory name index which is going to be inserted into b+ tree, and the other is used to manage the directory name index which is going to be deleted from b+ tree. - introduce a worker to deal with the delayed operation. This worker is used to deal with the works of the delayed directory name index items insertion and deletion and the delayed inode update. When the delayed items is beyond the lower limit, we create works for some delayed nodes and insert them into the work queue of the worker, and then go back. When the delayed items is beyond the upper bound, we create works for all the delayed nodes that haven't been dealt with, and insert them into the work queue of the worker, and then wait for that the untreated items is below some threshold value. - When we want to insert a directory name index into b+ tree, we just add the information into the delayed inserting rb-tree. And then we check the number of the delayed items and do delayed items balance. (The balance policy is above.) - When we want to delete a directory name index from the b+ tree, we search it in the inserting rb-tree at first. If we look it up, just drop it. If not, add the key of it into the delayed deleting rb-tree. Similar to the delayed inserting rb-tree, we also check the number of the delayed items and do delayed items balance. (The same to inserting manipulation) - When we want to update the metadata of some inode, we cached the data of the inode into the delayed node. the worker will flush it into the b+ tree after dealing with the delayed insertion and deletion. - We will move the delayed node to the tail of the list after we access the delayed node, By this way, we can cache more delayed items and merge more inode updates. - If we want to commit transaction, we will deal with all the delayed node. - the delayed node will be freed when we free the btrfs inode. - Before we log the inode items, we commit all the directory name index items and the delayed inode update. I did a quick test by the benchmark tool[1] and found we can improve the performance of file creation by ~15%, and file deletion by ~20%. Before applying this patch: Create files: Total files: 50000 Total time: 1.096108 Average time: 0.000022 Delete files: Total files: 50000 Total time: 1.510403 Average time: 0.000030 After applying this patch: Create files: Total files: 50000 Total time: 0.932899 Average time: 0.000019 Delete files: Total files: 50000 Total time: 1.215732 Average time: 0.000024 [1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3 Many thanks for Kitayama-san's help! Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Reviewed-by: David Sterba <dave@jikos.cz> Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com> Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-04-22 10:12:22 +00:00
ret = btrfs_inode_delayed_dir_index_count(dir);
if (ret) {
ret = btrfs_set_inode_index_count(dir);
if (ret)
return ret;
}
}
*index = BTRFS_I(dir)->index_cnt;
BTRFS_I(dir)->index_cnt++;
return ret;
}
static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
struct btrfs_root *root,
struct inode *dir,
const char *name, int name_len,
u64 ref_objectid, u64 objectid,
umode_t mode, u64 *index)
{
struct inode *inode;
struct btrfs_inode_item *inode_item;
struct btrfs_key *location;
struct btrfs_path *path;
struct btrfs_inode_ref *ref;
struct btrfs_key key[2];
u32 sizes[2];
unsigned long ptr;
int ret;
int owner;
path = btrfs_alloc_path();
if (!path)
return ERR_PTR(-ENOMEM);
inode = new_inode(root->fs_info->sb);
if (!inode) {
btrfs_free_path(path);
return ERR_PTR(-ENOMEM);
}
/*
* we have to initialize this early, so we can reclaim the inode
* number if we fail afterwards in this function.
*/
inode->i_ino = objectid;
if (dir) {
Btrfs: add initial tracepoint support for btrfs Tracepoints can provide insight into why btrfs hits bugs and be greatly helpful for debugging, e.g dd-7822 [000] 2121.641088: btrfs_inode_request: root = 5(FS_TREE), gen = 4, ino = 256, blocks = 8, disk_i_size = 0, last_trans = 8, logged_trans = 0 dd-7822 [000] 2121.641100: btrfs_inode_new: root = 5(FS_TREE), gen = 8, ino = 257, blocks = 0, disk_i_size = 0, last_trans = 0, logged_trans = 0 btrfs-transacti-7804 [001] 2146.935420: btrfs_cow_block: root = 2(EXTENT_TREE), refs = 2, orig_buf = 29368320 (orig_level = 0), cow_buf = 29388800 (cow_level = 0) btrfs-transacti-7804 [001] 2146.935473: btrfs_cow_block: root = 1(ROOT_TREE), refs = 2, orig_buf = 29364224 (orig_level = 0), cow_buf = 29392896 (cow_level = 0) btrfs-transacti-7804 [001] 2146.972221: btrfs_transaction_commit: root = 1(ROOT_TREE), gen = 8 flush-btrfs-2-7821 [001] 2155.824210: btrfs_chunk_alloc: root = 3(CHUNK_TREE), offset = 1103101952, size = 1073741824, num_stripes = 1, sub_stripes = 0, type = DATA flush-btrfs-2-7821 [001] 2155.824241: btrfs_cow_block: root = 2(EXTENT_TREE), refs = 2, orig_buf = 29388800 (orig_level = 0), cow_buf = 29396992 (cow_level = 0) flush-btrfs-2-7821 [001] 2155.824255: btrfs_cow_block: root = 4(DEV_TREE), refs = 2, orig_buf = 29372416 (orig_level = 0), cow_buf = 29401088 (cow_level = 0) flush-btrfs-2-7821 [000] 2155.824329: btrfs_cow_block: root = 3(CHUNK_TREE), refs = 2, orig_buf = 20971520 (orig_level = 0), cow_buf = 20975616 (cow_level = 0) btrfs-endio-wri-7800 [001] 2155.898019: btrfs_cow_block: root = 5(FS_TREE), refs = 2, orig_buf = 29384704 (orig_level = 0), cow_buf = 29405184 (cow_level = 0) btrfs-endio-wri-7800 [001] 2155.898043: btrfs_cow_block: root = 7(CSUM_TREE), refs = 2, orig_buf = 29376512 (orig_level = 0), cow_buf = 29409280 (cow_level = 0) Here is what I have added: 1) ordere_extent: btrfs_ordered_extent_add btrfs_ordered_extent_remove btrfs_ordered_extent_start btrfs_ordered_extent_put These provide critical information to understand how ordered_extents are updated. 2) extent_map: btrfs_get_extent extent_map is used in both read and write cases, and it is useful for tracking how btrfs specific IO is running. 3) writepage: __extent_writepage btrfs_writepage_end_io_hook Pages are cirtical resourses and produce a lot of corner cases during writeback, so it is valuable to know how page is written to disk. 4) inode: btrfs_inode_new btrfs_inode_request btrfs_inode_evict These can show where and when a inode is created, when a inode is evicted. 5) sync: btrfs_sync_file btrfs_sync_fs These show sync arguments. 6) transaction: btrfs_transaction_commit In transaction based filesystem, it will be useful to know the generation and who does commit. 7) back reference and cow: btrfs_delayed_tree_ref btrfs_delayed_data_ref btrfs_delayed_ref_head btrfs_cow_block Btrfs natively supports back references, these tracepoints are helpful on understanding btrfs's COW mechanism. 8) chunk: btrfs_chunk_alloc btrfs_chunk_free Chunk is a link between physical offset and logical offset, and stands for space infomation in btrfs, and these are helpful on tracing space things. 9) reserved_extent: btrfs_reserved_extent_alloc btrfs_reserved_extent_free These can show how btrfs uses its space. Signed-off-by: Liu Bo <liubo2009@cn.fujitsu.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-03-24 11:18:59 +00:00
trace_btrfs_inode_request(dir);
ret = btrfs_set_inode_index(dir, index);
if (ret) {
btrfs_free_path(path);
iput(inode);
return ERR_PTR(ret);
}
}
/*
* index_cnt is ignored for everything but a dir,
* btrfs_get_inode_index_count has an explanation for the magic
* number
*/
BTRFS_I(inode)->index_cnt = 2;
BTRFS_I(inode)->root = root;
BTRFS_I(inode)->generation = trans->transid;
inode->i_generation = BTRFS_I(inode)->generation;
btrfs_set_inode_space_info(root, inode);
if (S_ISDIR(mode))
owner = 0;
else
owner = 1;
key[0].objectid = objectid;
btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
key[0].offset = 0;
key[1].objectid = objectid;
btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
key[1].offset = ref_objectid;
sizes[0] = sizeof(struct btrfs_inode_item);
sizes[1] = name_len + sizeof(*ref);
path->leave_spinning = 1;
ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
if (ret != 0)
goto fail;
inode_init_owner(inode, dir, mode);
inode_set_bytes(inode, 0);
inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
struct btrfs_inode_item);
fill_inode_item(trans, path->nodes[0], inode_item, inode);
ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
struct btrfs_inode_ref);
btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
ptr = (unsigned long)(ref + 1);
write_extent_buffer(path->nodes[0], name, ptr, name_len);
btrfs_mark_buffer_dirty(path->nodes[0]);
btrfs_free_path(path);
location = &BTRFS_I(inode)->location;
location->objectid = objectid;
location->offset = 0;
btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
btrfs_inherit_iflags(inode, dir);
if (S_ISREG(mode)) {
if (btrfs_test_opt(root, NODATASUM))
BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
if (btrfs_test_opt(root, NODATACOW) ||
(BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
}
insert_inode_hash(inode);
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 14:45:14 +00:00
inode_tree_add(inode);
Btrfs: add initial tracepoint support for btrfs Tracepoints can provide insight into why btrfs hits bugs and be greatly helpful for debugging, e.g dd-7822 [000] 2121.641088: btrfs_inode_request: root = 5(FS_TREE), gen = 4, ino = 256, blocks = 8, disk_i_size = 0, last_trans = 8, logged_trans = 0 dd-7822 [000] 2121.641100: btrfs_inode_new: root = 5(FS_TREE), gen = 8, ino = 257, blocks = 0, disk_i_size = 0, last_trans = 0, logged_trans = 0 btrfs-transacti-7804 [001] 2146.935420: btrfs_cow_block: root = 2(EXTENT_TREE), refs = 2, orig_buf = 29368320 (orig_level = 0), cow_buf = 29388800 (cow_level = 0) btrfs-transacti-7804 [001] 2146.935473: btrfs_cow_block: root = 1(ROOT_TREE), refs = 2, orig_buf = 29364224 (orig_level = 0), cow_buf = 29392896 (cow_level = 0) btrfs-transacti-7804 [001] 2146.972221: btrfs_transaction_commit: root = 1(ROOT_TREE), gen = 8 flush-btrfs-2-7821 [001] 2155.824210: btrfs_chunk_alloc: root = 3(CHUNK_TREE), offset = 1103101952, size = 1073741824, num_stripes = 1, sub_stripes = 0, type = DATA flush-btrfs-2-7821 [001] 2155.824241: btrfs_cow_block: root = 2(EXTENT_TREE), refs = 2, orig_buf = 29388800 (orig_level = 0), cow_buf = 29396992 (cow_level = 0) flush-btrfs-2-7821 [001] 2155.824255: btrfs_cow_block: root = 4(DEV_TREE), refs = 2, orig_buf = 29372416 (orig_level = 0), cow_buf = 29401088 (cow_level = 0) flush-btrfs-2-7821 [000] 2155.824329: btrfs_cow_block: root = 3(CHUNK_TREE), refs = 2, orig_buf = 20971520 (orig_level = 0), cow_buf = 20975616 (cow_level = 0) btrfs-endio-wri-7800 [001] 2155.898019: btrfs_cow_block: root = 5(FS_TREE), refs = 2, orig_buf = 29384704 (orig_level = 0), cow_buf = 29405184 (cow_level = 0) btrfs-endio-wri-7800 [001] 2155.898043: btrfs_cow_block: root = 7(CSUM_TREE), refs = 2, orig_buf = 29376512 (orig_level = 0), cow_buf = 29409280 (cow_level = 0) Here is what I have added: 1) ordere_extent: btrfs_ordered_extent_add btrfs_ordered_extent_remove btrfs_ordered_extent_start btrfs_ordered_extent_put These provide critical information to understand how ordered_extents are updated. 2) extent_map: btrfs_get_extent extent_map is used in both read and write cases, and it is useful for tracking how btrfs specific IO is running. 3) writepage: __extent_writepage btrfs_writepage_end_io_hook Pages are cirtical resourses and produce a lot of corner cases during writeback, so it is valuable to know how page is written to disk. 4) inode: btrfs_inode_new btrfs_inode_request btrfs_inode_evict These can show where and when a inode is created, when a inode is evicted. 5) sync: btrfs_sync_file btrfs_sync_fs These show sync arguments. 6) transaction: btrfs_transaction_commit In transaction based filesystem, it will be useful to know the generation and who does commit. 7) back reference and cow: btrfs_delayed_tree_ref btrfs_delayed_data_ref btrfs_delayed_ref_head btrfs_cow_block Btrfs natively supports back references, these tracepoints are helpful on understanding btrfs's COW mechanism. 8) chunk: btrfs_chunk_alloc btrfs_chunk_free Chunk is a link between physical offset and logical offset, and stands for space infomation in btrfs, and these are helpful on tracing space things. 9) reserved_extent: btrfs_reserved_extent_alloc btrfs_reserved_extent_free These can show how btrfs uses its space. Signed-off-by: Liu Bo <liubo2009@cn.fujitsu.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-03-24 11:18:59 +00:00
trace_btrfs_inode_new(inode);
btrfs_set_inode_last_trans(trans, inode);
Btrfs: add initial tracepoint support for btrfs Tracepoints can provide insight into why btrfs hits bugs and be greatly helpful for debugging, e.g dd-7822 [000] 2121.641088: btrfs_inode_request: root = 5(FS_TREE), gen = 4, ino = 256, blocks = 8, disk_i_size = 0, last_trans = 8, logged_trans = 0 dd-7822 [000] 2121.641100: btrfs_inode_new: root = 5(FS_TREE), gen = 8, ino = 257, blocks = 0, disk_i_size = 0, last_trans = 0, logged_trans = 0 btrfs-transacti-7804 [001] 2146.935420: btrfs_cow_block: root = 2(EXTENT_TREE), refs = 2, orig_buf = 29368320 (orig_level = 0), cow_buf = 29388800 (cow_level = 0) btrfs-transacti-7804 [001] 2146.935473: btrfs_cow_block: root = 1(ROOT_TREE), refs = 2, orig_buf = 29364224 (orig_level = 0), cow_buf = 29392896 (cow_level = 0) btrfs-transacti-7804 [001] 2146.972221: btrfs_transaction_commit: root = 1(ROOT_TREE), gen = 8 flush-btrfs-2-7821 [001] 2155.824210: btrfs_chunk_alloc: root = 3(CHUNK_TREE), offset = 1103101952, size = 1073741824, num_stripes = 1, sub_stripes = 0, type = DATA flush-btrfs-2-7821 [001] 2155.824241: btrfs_cow_block: root = 2(EXTENT_TREE), refs = 2, orig_buf = 29388800 (orig_level = 0), cow_buf = 29396992 (cow_level = 0) flush-btrfs-2-7821 [001] 2155.824255: btrfs_cow_block: root = 4(DEV_TREE), refs = 2, orig_buf = 29372416 (orig_level = 0), cow_buf = 29401088 (cow_level = 0) flush-btrfs-2-7821 [000] 2155.824329: btrfs_cow_block: root = 3(CHUNK_TREE), refs = 2, orig_buf = 20971520 (orig_level = 0), cow_buf = 20975616 (cow_level = 0) btrfs-endio-wri-7800 [001] 2155.898019: btrfs_cow_block: root = 5(FS_TREE), refs = 2, orig_buf = 29384704 (orig_level = 0), cow_buf = 29405184 (cow_level = 0) btrfs-endio-wri-7800 [001] 2155.898043: btrfs_cow_block: root = 7(CSUM_TREE), refs = 2, orig_buf = 29376512 (orig_level = 0), cow_buf = 29409280 (cow_level = 0) Here is what I have added: 1) ordere_extent: btrfs_ordered_extent_add btrfs_ordered_extent_remove btrfs_ordered_extent_start btrfs_ordered_extent_put These provide critical information to understand how ordered_extents are updated. 2) extent_map: btrfs_get_extent extent_map is used in both read and write cases, and it is useful for tracking how btrfs specific IO is running. 3) writepage: __extent_writepage btrfs_writepage_end_io_hook Pages are cirtical resourses and produce a lot of corner cases during writeback, so it is valuable to know how page is written to disk. 4) inode: btrfs_inode_new btrfs_inode_request btrfs_inode_evict These can show where and when a inode is created, when a inode is evicted. 5) sync: btrfs_sync_file btrfs_sync_fs These show sync arguments. 6) transaction: btrfs_transaction_commit In transaction based filesystem, it will be useful to know the generation and who does commit. 7) back reference and cow: btrfs_delayed_tree_ref btrfs_delayed_data_ref btrfs_delayed_ref_head btrfs_cow_block Btrfs natively supports back references, these tracepoints are helpful on understanding btrfs's COW mechanism. 8) chunk: btrfs_chunk_alloc btrfs_chunk_free Chunk is a link between physical offset and logical offset, and stands for space infomation in btrfs, and these are helpful on tracing space things. 9) reserved_extent: btrfs_reserved_extent_alloc btrfs_reserved_extent_free These can show how btrfs uses its space. Signed-off-by: Liu Bo <liubo2009@cn.fujitsu.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-03-24 11:18:59 +00:00
return inode;
fail:
if (dir)
BTRFS_I(dir)->index_cnt--;
btrfs_free_path(path);
iput(inode);
return ERR_PTR(ret);
}
static inline u8 btrfs_inode_type(struct inode *inode)
{
return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
}
/*
* utility function to add 'inode' into 'parent_inode' with
* a give name and a given sequence number.
* if 'add_backref' is true, also insert a backref from the
* inode to the parent directory.
*/
int btrfs_add_link(struct btrfs_trans_handle *trans,
struct inode *parent_inode, struct inode *inode,
const char *name, int name_len, int add_backref, u64 index)
{
int ret = 0;
struct btrfs_key key;
struct btrfs_root *root = BTRFS_I(parent_inode)->root;
u64 ino = btrfs_ino(inode);
u64 parent_ino = btrfs_ino(parent_inode);
if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
} else {
key.objectid = ino;
btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
key.offset = 0;
}
if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
key.objectid, root->root_key.objectid,
parent_ino, index, name, name_len);
} else if (add_backref) {
ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
parent_ino, index);
}
if (ret == 0) {
ret = btrfs_insert_dir_item(trans, root, name, name_len,
btrfs: implement delayed inode items operation Changelog V5 -> V6: - Fix oom when the memory load is high, by storing the delayed nodes into the root's radix tree, and letting btrfs inodes go. Changelog V4 -> V5: - Fix the race on adding the delayed node to the inode, which is spotted by Chris Mason. - Merge Chris Mason's incremental patch into this patch. - Fix deadlock between readdir() and memory fault, which is reported by Itaru Kitayama. Changelog V3 -> V4: - Fix nested lock, which is reported by Itaru Kitayama, by updating space cache inode in time. Changelog V2 -> V3: - Fix the race between the delayed worker and the task which does delayed items balance, which is reported by Tsutomu Itoh. - Modify the patch address David Sterba's comment. - Fix the bug of the cpu recursion spinlock, reported by Chris Mason Changelog V1 -> V2: - break up the global rb-tree, use a list to manage the delayed nodes, which is created for every directory and file, and used to manage the delayed directory name index items and the delayed inode item. - introduce a worker to deal with the delayed nodes. Compare with Ext3/4, the performance of file creation and deletion on btrfs is very poor. the reason is that btrfs must do a lot of b+ tree insertions, such as inode item, directory name item, directory name index and so on. If we can do some delayed b+ tree insertion or deletion, we can improve the performance, so we made this patch which implemented delayed directory name index insertion/deletion and delayed inode update. Implementation: - introduce a delayed root object into the filesystem, that use two lists to manage the delayed nodes which are created for every file/directory. One is used to manage all the delayed nodes that have delayed items. And the other is used to manage the delayed nodes which is waiting to be dealt with by the work thread. - Every delayed node has two rb-tree, one is used to manage the directory name index which is going to be inserted into b+ tree, and the other is used to manage the directory name index which is going to be deleted from b+ tree. - introduce a worker to deal with the delayed operation. This worker is used to deal with the works of the delayed directory name index items insertion and deletion and the delayed inode update. When the delayed items is beyond the lower limit, we create works for some delayed nodes and insert them into the work queue of the worker, and then go back. When the delayed items is beyond the upper bound, we create works for all the delayed nodes that haven't been dealt with, and insert them into the work queue of the worker, and then wait for that the untreated items is below some threshold value. - When we want to insert a directory name index into b+ tree, we just add the information into the delayed inserting rb-tree. And then we check the number of the delayed items and do delayed items balance. (The balance policy is above.) - When we want to delete a directory name index from the b+ tree, we search it in the inserting rb-tree at first. If we look it up, just drop it. If not, add the key of it into the delayed deleting rb-tree. Similar to the delayed inserting rb-tree, we also check the number of the delayed items and do delayed items balance. (The same to inserting manipulation) - When we want to update the metadata of some inode, we cached the data of the inode into the delayed node. the worker will flush it into the b+ tree after dealing with the delayed insertion and deletion. - We will move the delayed node to the tail of the list after we access the delayed node, By this way, we can cache more delayed items and merge more inode updates. - If we want to commit transaction, we will deal with all the delayed node. - the delayed node will be freed when we free the btrfs inode. - Before we log the inode items, we commit all the directory name index items and the delayed inode update. I did a quick test by the benchmark tool[1] and found we can improve the performance of file creation by ~15%, and file deletion by ~20%. Before applying this patch: Create files: Total files: 50000 Total time: 1.096108 Average time: 0.000022 Delete files: Total files: 50000 Total time: 1.510403 Average time: 0.000030 After applying this patch: Create files: Total files: 50000 Total time: 0.932899 Average time: 0.000019 Delete files: Total files: 50000 Total time: 1.215732 Average time: 0.000024 [1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3 Many thanks for Kitayama-san's help! Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Reviewed-by: David Sterba <dave@jikos.cz> Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com> Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-04-22 10:12:22 +00:00
parent_inode, &key,
btrfs_inode_type(inode), index);
if (ret)
goto fail_dir_item;
btrfs_i_size_write(parent_inode, parent_inode->i_size +
name_len * 2);
parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
ret = btrfs_update_inode(trans, root, parent_inode);
}
return ret;
fail_dir_item:
if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
u64 local_index;
int err;
err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
key.objectid, root->root_key.objectid,
parent_ino, &local_index, name, name_len);
} else if (add_backref) {
u64 local_index;
int err;
err = btrfs_del_inode_ref(trans, root, name, name_len,
ino, parent_ino, &local_index);
}
return ret;
}
static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
struct inode *dir, struct dentry *dentry,
struct inode *inode, int backref, u64 index)
{
int err = btrfs_add_link(trans, dir, inode,
dentry->d_name.name, dentry->d_name.len,
backref, index);
if (err > 0)
err = -EEXIST;
return err;
}
static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
umode_t mode, dev_t rdev)
{
struct btrfs_trans_handle *trans;
struct btrfs_root *root = BTRFS_I(dir)->root;
struct inode *inode = NULL;
int err;
int drop_inode = 0;
u64 objectid;
unsigned long nr = 0;
u64 index = 0;
if (!new_valid_dev(rdev))
return -EINVAL;
Btrfs: proper -ENOSPC handling At the start of a transaction we do a btrfs_reserve_metadata_space() and specify how many items we plan on modifying. Then once we've done our modifications and such, just call btrfs_unreserve_metadata_space() for the same number of items we reserved. For keeping track of metadata needed for data I've had to add an extent_io op for when we merge extents. This lets us track space properly when we are doing sequential writes, so we don't end up reserving way more metadata space than what we need. The only place where the metadata space accounting is not done is in the relocation code. This is because Yan is going to be reworking that code in the near future, so running btrfs-vol -b could still possibly result in a ENOSPC related panic. This patch also turns off the metadata_ratio stuff in order to allow users to more efficiently use their disk space. This patch makes it so we track how much metadata we need for an inode's delayed allocation extents by tracking how many extents are currently waiting for allocation. It introduces two new callbacks for the extent_io tree's, merge_extent_hook and split_extent_hook. These help us keep track of when we merge delalloc extents together and split them up. Reservations are handled prior to any actually dirty'ing occurs, and then we unreserve after we dirty. btrfs_unreserve_metadata_for_delalloc() will make the appropriate unreservations as needed based on the number of reservations we currently have and the number of extents we currently have. Doing the reservation outside of doing any of the actual dirty'ing lets us do things like filemap_flush() the inode to try and force delalloc to happen, or as a last resort actually start allocation on all delalloc inodes in the fs. This has survived dbench, fs_mark and an fsx torture test. Signed-off-by: Josef Bacik <jbacik@redhat.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-09-11 20:12:44 +00:00
/*
* 2 for inode item and ref
* 2 for dir items
* 1 for xattr if selinux is on
*/
trans = btrfs_start_transaction(root, 5);
if (IS_ERR(trans))
return PTR_ERR(trans);
err = btrfs_find_free_ino(root, &objectid);
if (err)
goto out_unlock;
inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
dentry->d_name.len, btrfs_ino(dir), objectid,
mode, &index);
if (IS_ERR(inode)) {
err = PTR_ERR(inode);
goto out_unlock;
}
err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
if (err) {
drop_inode = 1;
goto out_unlock;
}
/*
* If the active LSM wants to access the inode during
* d_instantiate it needs these. Smack checks to see
* if the filesystem supports xattrs by looking at the
* ops vector.
*/
inode->i_op = &btrfs_special_inode_operations;
err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
if (err)
drop_inode = 1;
else {
init_special_inode(inode, inode->i_mode, rdev);
btrfs_update_inode(trans, root, inode);
d_instantiate(dentry, inode);
}
out_unlock:
nr = trans->blocks_used;
btrfs_end_transaction(trans, root);
btrfs_btree_balance_dirty(root, nr);
if (drop_inode) {
inode_dec_link_count(inode);
iput(inode);
}
return err;
}
static int btrfs_create(struct inode *dir, struct dentry *dentry,
umode_t mode, struct nameidata *nd)
{
struct btrfs_trans_handle *trans;
struct btrfs_root *root = BTRFS_I(dir)->root;
struct inode *inode = NULL;
int drop_inode = 0;
int err;
unsigned long nr = 0;
u64 objectid;
u64 index = 0;
Btrfs: proper -ENOSPC handling At the start of a transaction we do a btrfs_reserve_metadata_space() and specify how many items we plan on modifying. Then once we've done our modifications and such, just call btrfs_unreserve_metadata_space() for the same number of items we reserved. For keeping track of metadata needed for data I've had to add an extent_io op for when we merge extents. This lets us track space properly when we are doing sequential writes, so we don't end up reserving way more metadata space than what we need. The only place where the metadata space accounting is not done is in the relocation code. This is because Yan is going to be reworking that code in the near future, so running btrfs-vol -b could still possibly result in a ENOSPC related panic. This patch also turns off the metadata_ratio stuff in order to allow users to more efficiently use their disk space. This patch makes it so we track how much metadata we need for an inode's delayed allocation extents by tracking how many extents are currently waiting for allocation. It introduces two new callbacks for the extent_io tree's, merge_extent_hook and split_extent_hook. These help us keep track of when we merge delalloc extents together and split them up. Reservations are handled prior to any actually dirty'ing occurs, and then we unreserve after we dirty. btrfs_unreserve_metadata_for_delalloc() will make the appropriate unreservations as needed based on the number of reservations we currently have and the number of extents we currently have. Doing the reservation outside of doing any of the actual dirty'ing lets us do things like filemap_flush() the inode to try and force delalloc to happen, or as a last resort actually start allocation on all delalloc inodes in the fs. This has survived dbench, fs_mark and an fsx torture test. Signed-off-by: Josef Bacik <jbacik@redhat.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-09-11 20:12:44 +00:00
/*
* 2 for inode item and ref
* 2 for dir items
* 1 for xattr if selinux is on
*/
trans = btrfs_start_transaction(root, 5);
if (IS_ERR(trans))
return PTR_ERR(trans);
Btrfs: proper -ENOSPC handling At the start of a transaction we do a btrfs_reserve_metadata_space() and specify how many items we plan on modifying. Then once we've done our modifications and such, just call btrfs_unreserve_metadata_space() for the same number of items we reserved. For keeping track of metadata needed for data I've had to add an extent_io op for when we merge extents. This lets us track space properly when we are doing sequential writes, so we don't end up reserving way more metadata space than what we need. The only place where the metadata space accounting is not done is in the relocation code. This is because Yan is going to be reworking that code in the near future, so running btrfs-vol -b could still possibly result in a ENOSPC related panic. This patch also turns off the metadata_ratio stuff in order to allow users to more efficiently use their disk space. This patch makes it so we track how much metadata we need for an inode's delayed allocation extents by tracking how many extents are currently waiting for allocation. It introduces two new callbacks for the extent_io tree's, merge_extent_hook and split_extent_hook. These help us keep track of when we merge delalloc extents together and split them up. Reservations are handled prior to any actually dirty'ing occurs, and then we unreserve after we dirty. btrfs_unreserve_metadata_for_delalloc() will make the appropriate unreservations as needed based on the number of reservations we currently have and the number of extents we currently have. Doing the reservation outside of doing any of the actual dirty'ing lets us do things like filemap_flush() the inode to try and force delalloc to happen, or as a last resort actually start allocation on all delalloc inodes in the fs. This has survived dbench, fs_mark and an fsx torture test. Signed-off-by: Josef Bacik <jbacik@redhat.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-09-11 20:12:44 +00:00
err = btrfs_find_free_ino(root, &objectid);
if (err)
goto out_unlock;
inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
dentry->d_name.len, btrfs_ino(dir), objectid,
mode, &index);
if (IS_ERR(inode)) {
err = PTR_ERR(inode);
goto out_unlock;
}
err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
if (err) {
drop_inode = 1;
goto out_unlock;
}
/*
* If the active LSM wants to access the inode during
* d_instantiate it needs these. Smack checks to see
* if the filesystem supports xattrs by looking at the
* ops vector.
*/
inode->i_fop = &btrfs_file_operations;
inode->i_op = &btrfs_file_inode_operations;
err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
if (err)
drop_inode = 1;
else {
inode->i_mapping->a_ops = &btrfs_aops;
inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
d_instantiate(dentry, inode);
}
out_unlock:
nr = trans->blocks_used;
btrfs_end_transaction(trans, root);
if (drop_inode) {
inode_dec_link_count(inode);
iput(inode);
}
btrfs_btree_balance_dirty(root, nr);
return err;
}
static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
struct dentry *dentry)
{
struct btrfs_trans_handle *trans;
struct btrfs_root *root = BTRFS_I(dir)->root;
struct inode *inode = old_dentry->d_inode;
u64 index;
unsigned long nr = 0;
int err;
int drop_inode = 0;
/* do not allow sys_link's with other subvols of the same device */
if (root->objectid != BTRFS_I(inode)->root->objectid)
return -EXDEV;
if (inode->i_nlink == ~0U)
return -EMLINK;
err = btrfs_set_inode_index(dir, &index);
if (err)
goto fail;
/*
* 2 items for inode and inode ref
* 2 items for dir items
* 1 item for parent inode
*/
trans = btrfs_start_transaction(root, 5);
if (IS_ERR(trans)) {
err = PTR_ERR(trans);
goto fail;
}
btrfs_inc_nlink(inode);
inode->i_ctime = CURRENT_TIME;
ihold(inode);
err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
if (err) {
drop_inode = 1;
} else {
struct dentry *parent = dentry->d_parent;
err = btrfs_update_inode(trans, root, inode);
BUG_ON(err);
d_instantiate(dentry, inode);
btrfs_log_new_name(trans, inode, NULL, parent);
}
nr = trans->blocks_used;
btrfs_end_transaction(trans, root);
fail:
if (drop_inode) {
inode_dec_link_count(inode);
iput(inode);
}
btrfs_btree_balance_dirty(root, nr);
return err;
}
static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
{
struct inode *inode = NULL;
struct btrfs_trans_handle *trans;
struct btrfs_root *root = BTRFS_I(dir)->root;
int err = 0;
int drop_on_err = 0;
u64 objectid = 0;
u64 index = 0;
unsigned long nr = 1;
Btrfs: proper -ENOSPC handling At the start of a transaction we do a btrfs_reserve_metadata_space() and specify how many items we plan on modifying. Then once we've done our modifications and such, just call btrfs_unreserve_metadata_space() for the same number of items we reserved. For keeping track of metadata needed for data I've had to add an extent_io op for when we merge extents. This lets us track space properly when we are doing sequential writes, so we don't end up reserving way more metadata space than what we need. The only place where the metadata space accounting is not done is in the relocation code. This is because Yan is going to be reworking that code in the near future, so running btrfs-vol -b could still possibly result in a ENOSPC related panic. This patch also turns off the metadata_ratio stuff in order to allow users to more efficiently use their disk space. This patch makes it so we track how much metadata we need for an inode's delayed allocation extents by tracking how many extents are currently waiting for allocation. It introduces two new callbacks for the extent_io tree's, merge_extent_hook and split_extent_hook. These help us keep track of when we merge delalloc extents together and split them up. Reservations are handled prior to any actually dirty'ing occurs, and then we unreserve after we dirty. btrfs_unreserve_metadata_for_delalloc() will make the appropriate unreservations as needed based on the number of reservations we currently have and the number of extents we currently have. Doing the reservation outside of doing any of the actual dirty'ing lets us do things like filemap_flush() the inode to try and force delalloc to happen, or as a last resort actually start allocation on all delalloc inodes in the fs. This has survived dbench, fs_mark and an fsx torture test. Signed-off-by: Josef Bacik <jbacik@redhat.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-09-11 20:12:44 +00:00
/*
* 2 items for inode and ref
* 2 items for dir items
* 1 for xattr if selinux is on
*/
trans = btrfs_start_transaction(root, 5);
if (IS_ERR(trans))
return PTR_ERR(trans);
err = btrfs_find_free_ino(root, &objectid);
if (err)
goto out_fail;
inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
dentry->d_name.len, btrfs_ino(dir), objectid,
S_IFDIR | mode, &index);
if (IS_ERR(inode)) {
err = PTR_ERR(inode);
goto out_fail;
}
drop_on_err = 1;
err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
if (err)
goto out_fail;
inode->i_op = &btrfs_dir_inode_operations;
inode->i_fop = &btrfs_dir_file_operations;
btrfs_i_size_write(inode, 0);
err = btrfs_update_inode(trans, root, inode);
if (err)
goto out_fail;
err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
dentry->d_name.len, 0, index);
if (err)
goto out_fail;
d_instantiate(dentry, inode);
drop_on_err = 0;
out_fail:
nr = trans->blocks_used;
btrfs_end_transaction(trans, root);
if (drop_on_err)
iput(inode);
btrfs_btree_balance_dirty(root, nr);
return err;
}
/* helper for btfs_get_extent. Given an existing extent in the tree,
* and an extent that you want to insert, deal with overlap and insert
* the new extent into the tree.
*/
static int merge_extent_mapping(struct extent_map_tree *em_tree,
struct extent_map *existing,
struct extent_map *em,
u64 map_start, u64 map_len)
{
u64 start_diff;
BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
start_diff = map_start - em->start;
em->start = map_start;
em->len = map_len;
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
if (em->block_start < EXTENT_MAP_LAST_BYTE &&
!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
em->block_start += start_diff;
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
em->block_len -= start_diff;
}
return add_extent_mapping(em_tree, em);
}
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
static noinline int uncompress_inline(struct btrfs_path *path,
struct inode *inode, struct page *page,
size_t pg_offset, u64 extent_offset,
struct btrfs_file_extent_item *item)
{
int ret;
struct extent_buffer *leaf = path->nodes[0];
char *tmp;
size_t max_size;
unsigned long inline_size;
unsigned long ptr;
int compress_type;
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
WARN_ON(pg_offset != 0);
compress_type = btrfs_file_extent_compression(leaf, item);
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
max_size = btrfs_file_extent_ram_bytes(leaf, item);
inline_size = btrfs_file_extent_inline_item_len(leaf,
btrfs_item_nr(leaf, path->slots[0]));
tmp = kmalloc(inline_size, GFP_NOFS);
if (!tmp)
return -ENOMEM;
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
ptr = btrfs_file_extent_inline_start(item);
read_extent_buffer(leaf, tmp, ptr, inline_size);
max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
ret = btrfs_decompress(compress_type, tmp, page,
extent_offset, inline_size, max_size);
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
if (ret) {
char *kaddr = kmap_atomic(page, KM_USER0);
unsigned long copy_size = min_t(u64,
PAGE_CACHE_SIZE - pg_offset,
max_size - extent_offset);
memset(kaddr + pg_offset, 0, copy_size);
kunmap_atomic(kaddr, KM_USER0);
}
kfree(tmp);
return 0;
}
/*
* a bit scary, this does extent mapping from logical file offset to the disk.
* the ugly parts come from merging extents from the disk with the in-ram
* representation. This gets more complex because of the data=ordered code,
* where the in-ram extents might be locked pending data=ordered completion.
*
* This also copies inline extents directly into the page.
*/
struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
size_t pg_offset, u64 start, u64 len,
int create)
{
int ret;
int err = 0;
u64 bytenr;
u64 extent_start = 0;
u64 extent_end = 0;
u64 objectid = btrfs_ino(inode);
u32 found_type;
struct btrfs_path *path = NULL;
struct btrfs_root *root = BTRFS_I(inode)->root;
struct btrfs_file_extent_item *item;
struct extent_buffer *leaf;
struct btrfs_key found_key;
struct extent_map *em = NULL;
struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
struct btrfs_trans_handle *trans = NULL;
int compress_type;
again:
read_lock(&em_tree->lock);
em = lookup_extent_mapping(em_tree, start, len);
if (em)
em->bdev = root->fs_info->fs_devices->latest_bdev;
read_unlock(&em_tree->lock);
if (em) {
if (em->start > start || em->start + em->len <= start)
free_extent_map(em);
else if (em->block_start == EXTENT_MAP_INLINE && page)
free_extent_map(em);
else
goto out;
}
em = alloc_extent_map();
if (!em) {
err = -ENOMEM;
goto out;
}
em->bdev = root->fs_info->fs_devices->latest_bdev;
em->start = EXTENT_MAP_HOLE;
em->orig_start = EXTENT_MAP_HOLE;
em->len = (u64)-1;
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
em->block_len = (u64)-1;
if (!path) {
path = btrfs_alloc_path();
if (!path) {
err = -ENOMEM;
goto out;
}
/*
* Chances are we'll be called again, so go ahead and do
* readahead
*/
path->reada = 1;
}
ret = btrfs_lookup_file_extent(trans, root, path,
objectid, start, trans != NULL);
if (ret < 0) {
err = ret;
goto out;
}
if (ret != 0) {
if (path->slots[0] == 0)
goto not_found;
path->slots[0]--;
}
leaf = path->nodes[0];
item = btrfs_item_ptr(leaf, path->slots[0],
struct btrfs_file_extent_item);
/* are we inside the extent that was found? */
btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
found_type = btrfs_key_type(&found_key);
if (found_key.objectid != objectid ||
found_type != BTRFS_EXTENT_DATA_KEY) {
goto not_found;
}
found_type = btrfs_file_extent_type(leaf, item);
extent_start = found_key.offset;
compress_type = btrfs_file_extent_compression(leaf, item);
if (found_type == BTRFS_FILE_EXTENT_REG ||
found_type == BTRFS_FILE_EXTENT_PREALLOC) {
extent_end = extent_start +
btrfs_file_extent_num_bytes(leaf, item);
} else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
size_t size;
size = btrfs_file_extent_inline_len(leaf, item);
extent_end = (extent_start + size + root->sectorsize - 1) &
~((u64)root->sectorsize - 1);
}
if (start >= extent_end) {
path->slots[0]++;
if (path->slots[0] >= btrfs_header_nritems(leaf)) {
ret = btrfs_next_leaf(root, path);
if (ret < 0) {
err = ret;
goto out;
}
if (ret > 0)
goto not_found;
leaf = path->nodes[0];
}
btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
if (found_key.objectid != objectid ||
found_key.type != BTRFS_EXTENT_DATA_KEY)
goto not_found;
if (start + len <= found_key.offset)
goto not_found;
em->start = start;
em->len = found_key.offset - start;
goto not_found_em;
}
if (found_type == BTRFS_FILE_EXTENT_REG ||
found_type == BTRFS_FILE_EXTENT_PREALLOC) {
em->start = extent_start;
em->len = extent_end - extent_start;
em->orig_start = extent_start -
btrfs_file_extent_offset(leaf, item);
bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
if (bytenr == 0) {
em->block_start = EXTENT_MAP_HOLE;
goto insert;
}
if (compress_type != BTRFS_COMPRESS_NONE) {
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
em->compress_type = compress_type;
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
em->block_start = bytenr;
em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
item);
} else {
bytenr += btrfs_file_extent_offset(leaf, item);
em->block_start = bytenr;
em->block_len = em->len;
if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
}
goto insert;
} else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
unsigned long ptr;
char *map;
size_t size;
size_t extent_offset;
size_t copy_size;
em->block_start = EXTENT_MAP_INLINE;
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
if (!page || create) {
em->start = extent_start;
em->len = extent_end - extent_start;
goto out;
}
size = btrfs_file_extent_inline_len(leaf, item);
extent_offset = page_offset(page) + pg_offset - extent_start;
copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
size - extent_offset);
em->start = extent_start + extent_offset;
em->len = (copy_size + root->sectorsize - 1) &
~((u64)root->sectorsize - 1);
em->orig_start = EXTENT_MAP_INLINE;
if (compress_type) {
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
em->compress_type = compress_type;
}
ptr = btrfs_file_extent_inline_start(item) + extent_offset;
if (create == 0 && !PageUptodate(page)) {
if (btrfs_file_extent_compression(leaf, item) !=
BTRFS_COMPRESS_NONE) {
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
ret = uncompress_inline(path, inode, page,
pg_offset,
extent_offset, item);
BUG_ON(ret);
} else {
map = kmap(page);
read_extent_buffer(leaf, map + pg_offset, ptr,
copy_size);
if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
memset(map + pg_offset + copy_size, 0,
PAGE_CACHE_SIZE - pg_offset -
copy_size);
}
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
kunmap(page);
}
flush_dcache_page(page);
} else if (create && PageUptodate(page)) {
BUG();
if (!trans) {
kunmap(page);
free_extent_map(em);
em = NULL;
btrfs_release_path(path);
trans = btrfs_join_transaction(root);
if (IS_ERR(trans))
return ERR_CAST(trans);
goto again;
}
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
map = kmap(page);
write_extent_buffer(leaf, map + pg_offset, ptr,
copy_size);
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
kunmap(page);
btrfs_mark_buffer_dirty(leaf);
}
set_extent_uptodate(io_tree, em->start,
extent_map_end(em) - 1, NULL, GFP_NOFS);
goto insert;
} else {
printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
WARN_ON(1);
}
not_found:
em->start = start;
em->len = len;
not_found_em:
em->block_start = EXTENT_MAP_HOLE;
set_bit(EXTENT_FLAG_VACANCY, &em->flags);
insert:
btrfs_release_path(path);
if (em->start > start || extent_map_end(em) <= start) {
printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
"[%llu %llu]\n", (unsigned long long)em->start,
(unsigned long long)em->len,
(unsigned long long)start,
(unsigned long long)len);
err = -EIO;
goto out;
}
err = 0;
write_lock(&em_tree->lock);
ret = add_extent_mapping(em_tree, em);
/* it is possible that someone inserted the extent into the tree
* while we had the lock dropped. It is also possible that
* an overlapping map exists in the tree
*/
if (ret == -EEXIST) {
struct extent_map *existing;
ret = 0;
existing = lookup_extent_mapping(em_tree, start, len);
if (existing && (existing->start > start ||
existing->start + existing->len <= start)) {
free_extent_map(existing);
existing = NULL;
}
if (!existing) {
existing = lookup_extent_mapping(em_tree, em->start,
em->len);
if (existing) {
err = merge_extent_mapping(em_tree, existing,
em, start,
root->sectorsize);
free_extent_map(existing);
if (err) {
free_extent_map(em);
em = NULL;
}
} else {
err = -EIO;
free_extent_map(em);
em = NULL;
}
} else {
free_extent_map(em);
em = existing;
err = 0;
}
}
write_unlock(&em_tree->lock);
out:
Btrfs: add initial tracepoint support for btrfs Tracepoints can provide insight into why btrfs hits bugs and be greatly helpful for debugging, e.g dd-7822 [000] 2121.641088: btrfs_inode_request: root = 5(FS_TREE), gen = 4, ino = 256, blocks = 8, disk_i_size = 0, last_trans = 8, logged_trans = 0 dd-7822 [000] 2121.641100: btrfs_inode_new: root = 5(FS_TREE), gen = 8, ino = 257, blocks = 0, disk_i_size = 0, last_trans = 0, logged_trans = 0 btrfs-transacti-7804 [001] 2146.935420: btrfs_cow_block: root = 2(EXTENT_TREE), refs = 2, orig_buf = 29368320 (orig_level = 0), cow_buf = 29388800 (cow_level = 0) btrfs-transacti-7804 [001] 2146.935473: btrfs_cow_block: root = 1(ROOT_TREE), refs = 2, orig_buf = 29364224 (orig_level = 0), cow_buf = 29392896 (cow_level = 0) btrfs-transacti-7804 [001] 2146.972221: btrfs_transaction_commit: root = 1(ROOT_TREE), gen = 8 flush-btrfs-2-7821 [001] 2155.824210: btrfs_chunk_alloc: root = 3(CHUNK_TREE), offset = 1103101952, size = 1073741824, num_stripes = 1, sub_stripes = 0, type = DATA flush-btrfs-2-7821 [001] 2155.824241: btrfs_cow_block: root = 2(EXTENT_TREE), refs = 2, orig_buf = 29388800 (orig_level = 0), cow_buf = 29396992 (cow_level = 0) flush-btrfs-2-7821 [001] 2155.824255: btrfs_cow_block: root = 4(DEV_TREE), refs = 2, orig_buf = 29372416 (orig_level = 0), cow_buf = 29401088 (cow_level = 0) flush-btrfs-2-7821 [000] 2155.824329: btrfs_cow_block: root = 3(CHUNK_TREE), refs = 2, orig_buf = 20971520 (orig_level = 0), cow_buf = 20975616 (cow_level = 0) btrfs-endio-wri-7800 [001] 2155.898019: btrfs_cow_block: root = 5(FS_TREE), refs = 2, orig_buf = 29384704 (orig_level = 0), cow_buf = 29405184 (cow_level = 0) btrfs-endio-wri-7800 [001] 2155.898043: btrfs_cow_block: root = 7(CSUM_TREE), refs = 2, orig_buf = 29376512 (orig_level = 0), cow_buf = 29409280 (cow_level = 0) Here is what I have added: 1) ordere_extent: btrfs_ordered_extent_add btrfs_ordered_extent_remove btrfs_ordered_extent_start btrfs_ordered_extent_put These provide critical information to understand how ordered_extents are updated. 2) extent_map: btrfs_get_extent extent_map is used in both read and write cases, and it is useful for tracking how btrfs specific IO is running. 3) writepage: __extent_writepage btrfs_writepage_end_io_hook Pages are cirtical resourses and produce a lot of corner cases during writeback, so it is valuable to know how page is written to disk. 4) inode: btrfs_inode_new btrfs_inode_request btrfs_inode_evict These can show where and when a inode is created, when a inode is evicted. 5) sync: btrfs_sync_file btrfs_sync_fs These show sync arguments. 6) transaction: btrfs_transaction_commit In transaction based filesystem, it will be useful to know the generation and who does commit. 7) back reference and cow: btrfs_delayed_tree_ref btrfs_delayed_data_ref btrfs_delayed_ref_head btrfs_cow_block Btrfs natively supports back references, these tracepoints are helpful on understanding btrfs's COW mechanism. 8) chunk: btrfs_chunk_alloc btrfs_chunk_free Chunk is a link between physical offset and logical offset, and stands for space infomation in btrfs, and these are helpful on tracing space things. 9) reserved_extent: btrfs_reserved_extent_alloc btrfs_reserved_extent_free These can show how btrfs uses its space. Signed-off-by: Liu Bo <liubo2009@cn.fujitsu.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-03-24 11:18:59 +00:00
trace_btrfs_get_extent(root, em);
if (path)
btrfs_free_path(path);
if (trans) {
ret = btrfs_end_transaction(trans, root);
if (!err)
err = ret;
}
if (err) {
free_extent_map(em);
return ERR_PTR(err);
}
return em;
}
struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
size_t pg_offset, u64 start, u64 len,
int create)
{
struct extent_map *em;
struct extent_map *hole_em = NULL;
u64 range_start = start;
u64 end;
u64 found;
u64 found_end;
int err = 0;
em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
if (IS_ERR(em))
return em;
if (em) {
/*
* if our em maps to a hole, there might
* actually be delalloc bytes behind it
*/
if (em->block_start != EXTENT_MAP_HOLE)
return em;
else
hole_em = em;
}
/* check to see if we've wrapped (len == -1 or similar) */
end = start + len;
if (end < start)
end = (u64)-1;
else
end -= 1;
em = NULL;
/* ok, we didn't find anything, lets look for delalloc */
found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
end, len, EXTENT_DELALLOC, 1);
found_end = range_start + found;
if (found_end < range_start)
found_end = (u64)-1;
/*
* we didn't find anything useful, return
* the original results from get_extent()
*/
if (range_start > end || found_end <= start) {
em = hole_em;
hole_em = NULL;
goto out;
}
/* adjust the range_start to make sure it doesn't
* go backwards from the start they passed in
*/
range_start = max(start,range_start);
found = found_end - range_start;
if (found > 0) {
u64 hole_start = start;
u64 hole_len = len;
em = alloc_extent_map();
if (!em) {
err = -ENOMEM;
goto out;
}
/*
* when btrfs_get_extent can't find anything it
* returns one huge hole
*
* make sure what it found really fits our range, and
* adjust to make sure it is based on the start from
* the caller
*/
if (hole_em) {
u64 calc_end = extent_map_end(hole_em);
if (calc_end <= start || (hole_em->start > end)) {
free_extent_map(hole_em);
hole_em = NULL;
} else {
hole_start = max(hole_em->start, start);
hole_len = calc_end - hole_start;
}
}
em->bdev = NULL;
if (hole_em && range_start > hole_start) {
/* our hole starts before our delalloc, so we
* have to return just the parts of the hole
* that go until the delalloc starts
*/
em->len = min(hole_len,
range_start - hole_start);
em->start = hole_start;
em->orig_start = hole_start;
/*
* don't adjust block start at all,
* it is fixed at EXTENT_MAP_HOLE
*/
em->block_start = hole_em->block_start;
em->block_len = hole_len;
} else {
em->start = range_start;
em->len = found;
em->orig_start = range_start;
em->block_start = EXTENT_MAP_DELALLOC;
em->block_len = found;
}
} else if (hole_em) {
return hole_em;
}
out:
free_extent_map(hole_em);
if (err) {
free_extent_map(em);
return ERR_PTR(err);
}
return em;
}
static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
struct extent_map *em,
u64 start, u64 len)
{
struct btrfs_root *root = BTRFS_I(inode)->root;
struct btrfs_trans_handle *trans;
struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
struct btrfs_key ins;
u64 alloc_hint;
int ret;
bool insert = false;
/*
* Ok if the extent map we looked up is a hole and is for the exact
* range we want, there is no reason to allocate a new one, however if
* it is not right then we need to free this one and drop the cache for
* our range.
*/
if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
em->len != len) {
free_extent_map(em);
em = NULL;
insert = true;
btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
}
trans = btrfs_join_transaction(root);
if (IS_ERR(trans))
return ERR_CAST(trans);
if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
btrfs_add_inode_defrag(trans, inode);
trans->block_rsv = &root->fs_info->delalloc_block_rsv;
alloc_hint = get_extent_allocation_hint(inode, start, len);
ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
alloc_hint, (u64)-1, &ins, 1);
if (ret) {
em = ERR_PTR(ret);
goto out;
}
if (!em) {
em = alloc_extent_map();
if (!em) {
em = ERR_PTR(-ENOMEM);
goto out;
}
}
em->start = start;
em->orig_start = em->start;
em->len = ins.offset;
em->block_start = ins.objectid;
em->block_len = ins.offset;
em->bdev = root->fs_info->fs_devices->latest_bdev;
/*
* We need to do this because if we're using the original em we searched
* for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
*/
em->flags = 0;
set_bit(EXTENT_FLAG_PINNED, &em->flags);
while (insert) {
write_lock(&em_tree->lock);
ret = add_extent_mapping(em_tree, em);
write_unlock(&em_tree->lock);
if (ret != -EEXIST)
break;
btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
}
ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
ins.offset, ins.offset, 0);
if (ret) {
btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
em = ERR_PTR(ret);
}
out:
btrfs_end_transaction(trans, root);
return em;
}
/*
* returns 1 when the nocow is safe, < 1 on error, 0 if the
* block must be cow'd
*/
static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
struct inode *inode, u64 offset, u64 len)
{
struct btrfs_path *path;
int ret;
struct extent_buffer *leaf;
struct btrfs_root *root = BTRFS_I(inode)->root;
struct btrfs_file_extent_item *fi;
struct btrfs_key key;
u64 disk_bytenr;
u64 backref_offset;
u64 extent_end;
u64 num_bytes;
int slot;
int found_type;
path = btrfs_alloc_path();
if (!path)
return -ENOMEM;
ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
offset, 0);
if (ret < 0)
goto out;
slot = path->slots[0];
if (ret == 1) {
if (slot == 0) {
/* can't find the item, must cow */
ret = 0;
goto out;
}
slot--;
}
ret = 0;
leaf = path->nodes[0];
btrfs_item_key_to_cpu(leaf, &key, slot);
if (key.objectid != btrfs_ino(inode) ||
key.type != BTRFS_EXTENT_DATA_KEY) {
/* not our file or wrong item type, must cow */
goto out;
}
if (key.offset > offset) {
/* Wrong offset, must cow */
goto out;
}
fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
found_type = btrfs_file_extent_type(leaf, fi);
if (found_type != BTRFS_FILE_EXTENT_REG &&
found_type != BTRFS_FILE_EXTENT_PREALLOC) {
/* not a regular extent, must cow */
goto out;
}
disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
backref_offset = btrfs_file_extent_offset(leaf, fi);
extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
if (extent_end < offset + len) {
/* extent doesn't include our full range, must cow */
goto out;
}
if (btrfs_extent_readonly(root, disk_bytenr))
goto out;
/*
* look for other files referencing this extent, if we
* find any we must cow
*/
if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
key.offset - backref_offset, disk_bytenr))
goto out;
/*
* adjust disk_bytenr and num_bytes to cover just the bytes
* in this extent we are about to write. If there
* are any csums in that range we have to cow in order
* to keep the csums correct
*/
disk_bytenr += backref_offset;
disk_bytenr += offset - key.offset;
num_bytes = min(offset + len, extent_end) - offset;
if (csum_exist_in_range(root, disk_bytenr, num_bytes))
goto out;
/*
* all of the above have passed, it is safe to overwrite this extent
* without cow
*/
ret = 1;
out:
btrfs_free_path(path);
return ret;
}
static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
struct buffer_head *bh_result, int create)
{
struct extent_map *em;
struct btrfs_root *root = BTRFS_I(inode)->root;
u64 start = iblock << inode->i_blkbits;
u64 len = bh_result->b_size;
struct btrfs_trans_handle *trans;
em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
if (IS_ERR(em))
return PTR_ERR(em);
/*
* Ok for INLINE and COMPRESSED extents we need to fallback on buffered
* io. INLINE is special, and we could probably kludge it in here, but
* it's still buffered so for safety lets just fall back to the generic
* buffered path.
*
* For COMPRESSED we _have_ to read the entire extent in so we can
* decompress it, so there will be buffering required no matter what we
* do, so go ahead and fallback to buffered.
*
* We return -ENOTBLK because thats what makes DIO go ahead and go back
* to buffered IO. Don't blame me, this is the price we pay for using
* the generic code.
*/
if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
em->block_start == EXTENT_MAP_INLINE) {
free_extent_map(em);
return -ENOTBLK;
}
/* Just a good old fashioned hole, return */
if (!create && (em->block_start == EXTENT_MAP_HOLE ||
test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
free_extent_map(em);
/* DIO will do one hole at a time, so just unlock a sector */
unlock_extent(&BTRFS_I(inode)->io_tree, start,
start + root->sectorsize - 1);
return 0;
}
/*
* We don't allocate a new extent in the following cases
*
* 1) The inode is marked as NODATACOW. In this case we'll just use the
* existing extent.
* 2) The extent is marked as PREALLOC. We're good to go here and can
* just use the extent.
*
*/
if (!create) {
len = em->len - (start - em->start);
goto map;
}
if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
em->block_start != EXTENT_MAP_HOLE)) {
int type;
int ret;
u64 block_start;
if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
type = BTRFS_ORDERED_PREALLOC;
else
type = BTRFS_ORDERED_NOCOW;
len = min(len, em->len - (start - em->start));
block_start = em->block_start + (start - em->start);
/*
* we're not going to log anything, but we do need
* to make sure the current transaction stays open
* while we look for nocow cross refs
*/
trans = btrfs_join_transaction(root);
if (IS_ERR(trans))
goto must_cow;
if (can_nocow_odirect(trans, inode, start, len) == 1) {
ret = btrfs_add_ordered_extent_dio(inode, start,
block_start, len, len, type);
btrfs_end_transaction(trans, root);
if (ret) {
free_extent_map(em);
return ret;
}
goto unlock;
}
btrfs_end_transaction(trans, root);
}
must_cow:
/*
* this will cow the extent, reset the len in case we changed
* it above
*/
len = bh_result->b_size;
em = btrfs_new_extent_direct(inode, em, start, len);
if (IS_ERR(em))
return PTR_ERR(em);
len = min(len, em->len - (start - em->start));
unlock:
clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
0, NULL, GFP_NOFS);
map:
bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
inode->i_blkbits;
bh_result->b_size = len;
bh_result->b_bdev = em->bdev;
set_buffer_mapped(bh_result);
if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
set_buffer_new(bh_result);
free_extent_map(em);
return 0;
}
struct btrfs_dio_private {
struct inode *inode;
u64 logical_offset;
u64 disk_bytenr;
u64 bytes;
u32 *csums;
void *private;
/* number of bios pending for this dio */
atomic_t pending_bios;
/* IO errors */
int errors;
struct bio *orig_bio;
};
static void btrfs_endio_direct_read(struct bio *bio, int err)
{
struct btrfs_dio_private *dip = bio->bi_private;
struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
struct bio_vec *bvec = bio->bi_io_vec;
struct inode *inode = dip->inode;
struct btrfs_root *root = BTRFS_I(inode)->root;
u64 start;
u32 *private = dip->csums;
start = dip->logical_offset;
do {
if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
struct page *page = bvec->bv_page;
char *kaddr;
u32 csum = ~(u32)0;
unsigned long flags;
local_irq_save(flags);
kaddr = kmap_atomic(page, KM_IRQ0);
csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
csum, bvec->bv_len);
btrfs_csum_final(csum, (char *)&csum);
kunmap_atomic(kaddr, KM_IRQ0);
local_irq_restore(flags);
flush_dcache_page(bvec->bv_page);
if (csum != *private) {
printk(KERN_ERR "btrfs csum failed ino %llu off"
" %llu csum %u private %u\n",
(unsigned long long)btrfs_ino(inode),
(unsigned long long)start,
csum, *private);
err = -EIO;
}
}
start += bvec->bv_len;
private++;
bvec++;
} while (bvec <= bvec_end);
unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
dip->logical_offset + dip->bytes - 1);
bio->bi_private = dip->private;
kfree(dip->csums);
kfree(dip);
/* If we had a csum failure make sure to clear the uptodate flag */
if (err)
clear_bit(BIO_UPTODATE, &bio->bi_flags);
dio_end_io(bio, err);
}
static void btrfs_endio_direct_write(struct bio *bio, int err)
{
struct btrfs_dio_private *dip = bio->bi_private;
struct inode *inode = dip->inode;
struct btrfs_root *root = BTRFS_I(inode)->root;
struct btrfs_trans_handle *trans;
struct btrfs_ordered_extent *ordered = NULL;
struct extent_state *cached_state = NULL;
u64 ordered_offset = dip->logical_offset;
u64 ordered_bytes = dip->bytes;
int ret;
if (err)
goto out_done;
again:
ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
&ordered_offset,
ordered_bytes);
if (!ret)
goto out_test;
BUG_ON(!ordered);
trans = btrfs_join_transaction(root);
if (IS_ERR(trans)) {
err = -ENOMEM;
goto out;
}
trans->block_rsv = &root->fs_info->delalloc_block_rsv;
if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
ret = btrfs_ordered_update_i_size(inode, 0, ordered);
if (!ret)
err = btrfs_update_inode_fallback(trans, root, inode);
goto out;
}
lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
ordered->file_offset + ordered->len - 1, 0,
&cached_state);
if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
ret = btrfs_mark_extent_written(trans, inode,
ordered->file_offset,
ordered->file_offset +
ordered->len);
if (ret) {
err = ret;
goto out_unlock;
}
} else {
ret = insert_reserved_file_extent(trans, inode,
ordered->file_offset,
ordered->start,
ordered->disk_len,
ordered->len,
ordered->len,
0, 0, 0,
BTRFS_FILE_EXTENT_REG);
unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
ordered->file_offset, ordered->len);
if (ret) {
err = ret;
WARN_ON(1);
goto out_unlock;
}
}
add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
ret = btrfs_ordered_update_i_size(inode, 0, ordered);
if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags))
btrfs_update_inode_fallback(trans, root, inode);
ret = 0;
out_unlock:
unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
ordered->file_offset + ordered->len - 1,
&cached_state, GFP_NOFS);
out:
btrfs_delalloc_release_metadata(inode, ordered->len);
btrfs_end_transaction(trans, root);
ordered_offset = ordered->file_offset + ordered->len;
btrfs_put_ordered_extent(ordered);
btrfs_put_ordered_extent(ordered);
out_test:
/*
* our bio might span multiple ordered extents. If we haven't
* completed the accounting for the whole dio, go back and try again
*/
if (ordered_offset < dip->logical_offset + dip->bytes) {
ordered_bytes = dip->logical_offset + dip->bytes -
ordered_offset;
goto again;
}
out_done:
bio->bi_private = dip->private;
kfree(dip->csums);
kfree(dip);
/* If we had an error make sure to clear the uptodate flag */
if (err)
clear_bit(BIO_UPTODATE, &bio->bi_flags);
dio_end_io(bio, err);
}
static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
struct bio *bio, int mirror_num,
unsigned long bio_flags, u64 offset)
{
int ret;
struct btrfs_root *root = BTRFS_I(inode)->root;
ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
BUG_ON(ret);
return 0;
}
static void btrfs_end_dio_bio(struct bio *bio, int err)
{
struct btrfs_dio_private *dip = bio->bi_private;
if (err) {
printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
"sector %#Lx len %u err no %d\n",
(unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
(unsigned long long)bio->bi_sector, bio->bi_size, err);
dip->errors = 1;
/*
* before atomic variable goto zero, we must make sure
* dip->errors is perceived to be set.
*/
smp_mb__before_atomic_dec();
}
/* if there are more bios still pending for this dio, just exit */
if (!atomic_dec_and_test(&dip->pending_bios))
goto out;
if (dip->errors)
bio_io_error(dip->orig_bio);
else {
set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
bio_endio(dip->orig_bio, 0);
}
out:
bio_put(bio);
}
static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
u64 first_sector, gfp_t gfp_flags)
{
int nr_vecs = bio_get_nr_vecs(bdev);
return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
}
static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
int rw, u64 file_offset, int skip_sum,
u32 *csums, int async_submit)
{
int write = rw & REQ_WRITE;
struct btrfs_root *root = BTRFS_I(inode)->root;
int ret;
bio_get(bio);
ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
if (ret)
goto err;
if (skip_sum)
goto map;
if (write && async_submit) {
ret = btrfs_wq_submit_bio(root->fs_info,
inode, rw, bio, 0, 0,
file_offset,
__btrfs_submit_bio_start_direct_io,
__btrfs_submit_bio_done);
goto err;
} else if (write) {
/*
* If we aren't doing async submit, calculate the csum of the
* bio now.
*/
ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
if (ret)
goto err;
} else if (!skip_sum) {
ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
file_offset, csums);
if (ret)
goto err;
}
map:
ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
err:
bio_put(bio);
return ret;
}
static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
int skip_sum)
{
struct inode *inode = dip->inode;
struct btrfs_root *root = BTRFS_I(inode)->root;
struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
struct bio *bio;
struct bio *orig_bio = dip->orig_bio;
struct bio_vec *bvec = orig_bio->bi_io_vec;
u64 start_sector = orig_bio->bi_sector;
u64 file_offset = dip->logical_offset;
u64 submit_len = 0;
u64 map_length;
int nr_pages = 0;
u32 *csums = dip->csums;
int ret = 0;
int async_submit = 0;
int write = rw & REQ_WRITE;
map_length = orig_bio->bi_size;
ret = btrfs_map_block(map_tree, READ, start_sector << 9,
&map_length, NULL, 0);
if (ret) {
bio_put(orig_bio);
return -EIO;
}
if (map_length >= orig_bio->bi_size) {
bio = orig_bio;
goto submit;
}
async_submit = 1;
bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
if (!bio)
return -ENOMEM;
bio->bi_private = dip;
bio->bi_end_io = btrfs_end_dio_bio;
atomic_inc(&dip->pending_bios);
while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
if (unlikely(map_length < submit_len + bvec->bv_len ||
bio_add_page(bio, bvec->bv_page, bvec->bv_len,
bvec->bv_offset) < bvec->bv_len)) {
/*
* inc the count before we submit the bio so
* we know the end IO handler won't happen before
* we inc the count. Otherwise, the dip might get freed
* before we're done setting it up
*/
atomic_inc(&dip->pending_bios);
ret = __btrfs_submit_dio_bio(bio, inode, rw,
file_offset, skip_sum,
csums, async_submit);
if (ret) {
bio_put(bio);
atomic_dec(&dip->pending_bios);
goto out_err;
}
/* Write's use the ordered csums */
if (!write && !skip_sum)
csums = csums + nr_pages;
start_sector += submit_len >> 9;
file_offset += submit_len;
submit_len = 0;
nr_pages = 0;
bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
start_sector, GFP_NOFS);
if (!bio)
goto out_err;
bio->bi_private = dip;
bio->bi_end_io = btrfs_end_dio_bio;
map_length = orig_bio->bi_size;
ret = btrfs_map_block(map_tree, READ, start_sector << 9,
&map_length, NULL, 0);
if (ret) {
bio_put(bio);
goto out_err;
}
} else {
submit_len += bvec->bv_len;
nr_pages ++;
bvec++;
}
}
submit:
ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
csums, async_submit);
if (!ret)
return 0;
bio_put(bio);
out_err:
dip->errors = 1;
/*
* before atomic variable goto zero, we must
* make sure dip->errors is perceived to be set.
*/
smp_mb__before_atomic_dec();
if (atomic_dec_and_test(&dip->pending_bios))
bio_io_error(dip->orig_bio);
/* bio_end_io() will handle error, so we needn't return it */
return 0;
}
static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
loff_t file_offset)
{
struct btrfs_root *root = BTRFS_I(inode)->root;
struct btrfs_dio_private *dip;
struct bio_vec *bvec = bio->bi_io_vec;
int skip_sum;
int write = rw & REQ_WRITE;
int ret = 0;
skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
dip = kmalloc(sizeof(*dip), GFP_NOFS);
if (!dip) {
ret = -ENOMEM;
goto free_ordered;
}
dip->csums = NULL;
/* Write's use the ordered csum stuff, so we don't need dip->csums */
if (!write && !skip_sum) {
dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
if (!dip->csums) {
kfree(dip);
ret = -ENOMEM;
goto free_ordered;
}
}
dip->private = bio->bi_private;
dip->inode = inode;
dip->logical_offset = file_offset;
dip->bytes = 0;
do {
dip->bytes += bvec->bv_len;
bvec++;
} while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
dip->disk_bytenr = (u64)bio->bi_sector << 9;
bio->bi_private = dip;
dip->errors = 0;
dip->orig_bio = bio;
atomic_set(&dip->pending_bios, 0);
if (write)
bio->bi_end_io = btrfs_endio_direct_write;
else
bio->bi_end_io = btrfs_endio_direct_read;
ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
if (!ret)
return;
free_ordered:
/*
* If this is a write, we need to clean up the reserved space and kill
* the ordered extent.
*/
if (write) {
struct btrfs_ordered_extent *ordered;
ordered = btrfs_lookup_ordered_extent(inode, file_offset);
if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
!test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
btrfs_free_reserved_extent(root, ordered->start,
ordered->disk_len);
btrfs_put_ordered_extent(ordered);
btrfs_put_ordered_extent(ordered);
}
bio_endio(bio, ret);
}
static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
const struct iovec *iov, loff_t offset,
unsigned long nr_segs)
{
int seg;
int i;
size_t size;
unsigned long addr;
unsigned blocksize_mask = root->sectorsize - 1;
ssize_t retval = -EINVAL;
loff_t end = offset;
if (offset & blocksize_mask)
goto out;
/* Check the memory alignment. Blocks cannot straddle pages */
for (seg = 0; seg < nr_segs; seg++) {
addr = (unsigned long)iov[seg].iov_base;
size = iov[seg].iov_len;
end += size;
if ((addr & blocksize_mask) || (size & blocksize_mask))
goto out;
/* If this is a write we don't need to check anymore */
if (rw & WRITE)
continue;
/*
* Check to make sure we don't have duplicate iov_base's in this
* iovec, if so return EINVAL, otherwise we'll get csum errors
* when reading back.
*/
for (i = seg + 1; i < nr_segs; i++) {
if (iov[seg].iov_base == iov[i].iov_base)
goto out;
}
}
retval = 0;
out:
return retval;
}
static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
const struct iovec *iov, loff_t offset,
unsigned long nr_segs)
{
struct file *file = iocb->ki_filp;
struct inode *inode = file->f_mapping->host;
struct btrfs_ordered_extent *ordered;
struct extent_state *cached_state = NULL;
u64 lockstart, lockend;
ssize_t ret;
int writing = rw & WRITE;
int write_bits = 0;
size_t count = iov_length(iov, nr_segs);
if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
offset, nr_segs)) {
return 0;
}
lockstart = offset;
lockend = offset + count - 1;
if (writing) {
ret = btrfs_delalloc_reserve_space(inode, count);
if (ret)
goto out;
}
while (1) {
lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
0, &cached_state);
/*
* We're concerned with the entire range that we're going to be
* doing DIO to, so we need to make sure theres no ordered
* extents in this range.
*/
ordered = btrfs_lookup_ordered_range(inode, lockstart,
lockend - lockstart + 1);
if (!ordered)
break;
unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
&cached_state, GFP_NOFS);
btrfs_start_ordered_extent(inode, ordered, 1);
btrfs_put_ordered_extent(ordered);
cond_resched();
}
/*
* we don't use btrfs_set_extent_delalloc because we don't want
* the dirty or uptodate bits
*/
if (writing) {
write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
EXTENT_DELALLOC, NULL, &cached_state,
GFP_NOFS);
if (ret) {
clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
lockend, EXTENT_LOCKED | write_bits,
1, 0, &cached_state, GFP_NOFS);
goto out;
}
}
free_extent_state(cached_state);
cached_state = NULL;
ret = __blockdev_direct_IO(rw, iocb, inode,
BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
btrfs_submit_direct, 0);
if (ret < 0 && ret != -EIOCBQUEUED) {
clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
offset + iov_length(iov, nr_segs) - 1,
EXTENT_LOCKED | write_bits, 1, 0,
&cached_state, GFP_NOFS);
} else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
/*
* We're falling back to buffered, unlock the section we didn't
* do IO on.
*/
clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
offset + iov_length(iov, nr_segs) - 1,
EXTENT_LOCKED | write_bits, 1, 0,
&cached_state, GFP_NOFS);
}
out:
free_extent_state(cached_state);
return ret;
}
static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
__u64 start, __u64 len)
{
return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
}
int btrfs_readpage(struct file *file, struct page *page)
{
struct extent_io_tree *tree;
tree = &BTRFS_I(page->mapping->host)->io_tree;
return extent_read_full_page(tree, page, btrfs_get_extent, 0);
}
static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
{
struct extent_io_tree *tree;
if (current->flags & PF_MEMALLOC) {
redirty_page_for_writepage(wbc, page);
unlock_page(page);
return 0;
}
tree = &BTRFS_I(page->mapping->host)->io_tree;
return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
}
int btrfs_writepages(struct address_space *mapping,
struct writeback_control *wbc)
{
struct extent_io_tree *tree;
tree = &BTRFS_I(mapping->host)->io_tree;
return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
}
static int
btrfs_readpages(struct file *file, struct address_space *mapping,
struct list_head *pages, unsigned nr_pages)
{
struct extent_io_tree *tree;
tree = &BTRFS_I(mapping->host)->io_tree;
return extent_readpages(tree, mapping, pages, nr_pages,
btrfs_get_extent);
}
static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
{
struct extent_io_tree *tree;
struct extent_map_tree *map;
int ret;
tree = &BTRFS_I(page->mapping->host)->io_tree;
map = &BTRFS_I(page->mapping->host)->extent_tree;
ret = try_release_extent_mapping(map, tree, page, gfp_flags);
if (ret == 1) {
ClearPagePrivate(page);
set_page_private(page, 0);
page_cache_release(page);
}
return ret;
}
static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
{
if (PageWriteback(page) || PageDirty(page))
return 0;
return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
}
static void btrfs_invalidatepage(struct page *page, unsigned long offset)
{
struct extent_io_tree *tree;
struct btrfs_ordered_extent *ordered;
struct extent_state *cached_state = NULL;
u64 page_start = page_offset(page);
u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
/*
* we have the page locked, so new writeback can't start,
* and the dirty bit won't be cleared while we are here.
*
* Wait for IO on this page so that we can safely clear
* the PagePrivate2 bit and do ordered accounting
*/
wait_on_page_writeback(page);
tree = &BTRFS_I(page->mapping->host)->io_tree;
if (offset) {
btrfs_releasepage(page, GFP_NOFS);
return;
}
lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
ordered = btrfs_lookup_ordered_extent(page->mapping->host,
page_offset(page));
if (ordered) {
/*
* IO on this page will never be started, so we need
* to account for any ordered extents now
*/
clear_extent_bit(tree, page_start, page_end,
EXTENT_DIRTY | EXTENT_DELALLOC |
EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
&cached_state, GFP_NOFS);
/*
* whoever cleared the private bit is responsible
* for the finish_ordered_io
*/
if (TestClearPagePrivate2(page)) {
btrfs_finish_ordered_io(page->mapping->host,
page_start, page_end);
}
btrfs_put_ordered_extent(ordered);
cached_state = NULL;
lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
}
clear_extent_bit(tree, page_start, page_end,
EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
__btrfs_releasepage(page, GFP_NOFS);
ClearPageChecked(page);
if (PagePrivate(page)) {
ClearPagePrivate(page);
set_page_private(page, 0);
page_cache_release(page);
}
}
/*
* btrfs_page_mkwrite() is not allowed to change the file size as it gets
* called from a page fault handler when a page is first dirtied. Hence we must
* be careful to check for EOF conditions here. We set the page up correctly
* for a written page which means we get ENOSPC checking when writing into
* holes and correct delalloc and unwritten extent mapping on filesystems that
* support these features.
*
* We are not allowed to take the i_mutex here so we have to play games to
* protect against truncate races as the page could now be beyond EOF. Because
* vmtruncate() writes the inode size before removing pages, once we have the
* page lock we can determine safely if the page is beyond EOF. If it is not
* beyond EOF, then the page is guaranteed safe against truncation until we
* unlock the page.
*/
int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
{
struct page *page = vmf->page;
struct inode *inode = fdentry(vma->vm_file)->d_inode;
struct btrfs_root *root = BTRFS_I(inode)->root;
struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
struct btrfs_ordered_extent *ordered;
struct extent_state *cached_state = NULL;
char *kaddr;
unsigned long zero_start;
loff_t size;
int ret;
int reserved = 0;
u64 page_start;
u64 page_end;
ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
if (!ret) {
ret = btrfs_update_time(vma->vm_file);
reserved = 1;
}
if (ret) {
if (ret == -ENOMEM)
ret = VM_FAULT_OOM;
else /* -ENOSPC, -EIO, etc */
ret = VM_FAULT_SIGBUS;
if (reserved)
goto out;
goto out_noreserve;
}
ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
again:
lock_page(page);
size = i_size_read(inode);
page_start = page_offset(page);
page_end = page_start + PAGE_CACHE_SIZE - 1;
if ((page->mapping != inode->i_mapping) ||
(page_start >= size)) {
/* page got truncated out from underneath us */
goto out_unlock;
}
wait_on_page_writeback(page);
lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
set_page_extent_mapped(page);
/*
* we can't set the delalloc bits if there are pending ordered
* extents. Drop our locks and wait for them to finish
*/
ordered = btrfs_lookup_ordered_extent(inode, page_start);
if (ordered) {
unlock_extent_cached(io_tree, page_start, page_end,
&cached_state, GFP_NOFS);
unlock_page(page);
btrfs_start_ordered_extent(inode, ordered, 1);
btrfs_put_ordered_extent(ordered);
goto again;
}
/*
* XXX - page_mkwrite gets called every time the page is dirtied, even
* if it was already dirty, so for space accounting reasons we need to
* clear any delalloc bits for the range we are fixing to save. There
* is probably a better way to do this, but for now keep consistent with
* prepare_pages in the normal write path.
*/
clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
0, 0, &cached_state, GFP_NOFS);
ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
&cached_state);
Btrfs: proper -ENOSPC handling At the start of a transaction we do a btrfs_reserve_metadata_space() and specify how many items we plan on modifying. Then once we've done our modifications and such, just call btrfs_unreserve_metadata_space() for the same number of items we reserved. For keeping track of metadata needed for data I've had to add an extent_io op for when we merge extents. This lets us track space properly when we are doing sequential writes, so we don't end up reserving way more metadata space than what we need. The only place where the metadata space accounting is not done is in the relocation code. This is because Yan is going to be reworking that code in the near future, so running btrfs-vol -b could still possibly result in a ENOSPC related panic. This patch also turns off the metadata_ratio stuff in order to allow users to more efficiently use their disk space. This patch makes it so we track how much metadata we need for an inode's delayed allocation extents by tracking how many extents are currently waiting for allocation. It introduces two new callbacks for the extent_io tree's, merge_extent_hook and split_extent_hook. These help us keep track of when we merge delalloc extents together and split them up. Reservations are handled prior to any actually dirty'ing occurs, and then we unreserve after we dirty. btrfs_unreserve_metadata_for_delalloc() will make the appropriate unreservations as needed based on the number of reservations we currently have and the number of extents we currently have. Doing the reservation outside of doing any of the actual dirty'ing lets us do things like filemap_flush() the inode to try and force delalloc to happen, or as a last resort actually start allocation on all delalloc inodes in the fs. This has survived dbench, fs_mark and an fsx torture test. Signed-off-by: Josef Bacik <jbacik@redhat.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-09-11 20:12:44 +00:00
if (ret) {
unlock_extent_cached(io_tree, page_start, page_end,
&cached_state, GFP_NOFS);
Btrfs: proper -ENOSPC handling At the start of a transaction we do a btrfs_reserve_metadata_space() and specify how many items we plan on modifying. Then once we've done our modifications and such, just call btrfs_unreserve_metadata_space() for the same number of items we reserved. For keeping track of metadata needed for data I've had to add an extent_io op for when we merge extents. This lets us track space properly when we are doing sequential writes, so we don't end up reserving way more metadata space than what we need. The only place where the metadata space accounting is not done is in the relocation code. This is because Yan is going to be reworking that code in the near future, so running btrfs-vol -b could still possibly result in a ENOSPC related panic. This patch also turns off the metadata_ratio stuff in order to allow users to more efficiently use their disk space. This patch makes it so we track how much metadata we need for an inode's delayed allocation extents by tracking how many extents are currently waiting for allocation. It introduces two new callbacks for the extent_io tree's, merge_extent_hook and split_extent_hook. These help us keep track of when we merge delalloc extents together and split them up. Reservations are handled prior to any actually dirty'ing occurs, and then we unreserve after we dirty. btrfs_unreserve_metadata_for_delalloc() will make the appropriate unreservations as needed based on the number of reservations we currently have and the number of extents we currently have. Doing the reservation outside of doing any of the actual dirty'ing lets us do things like filemap_flush() the inode to try and force delalloc to happen, or as a last resort actually start allocation on all delalloc inodes in the fs. This has survived dbench, fs_mark and an fsx torture test. Signed-off-by: Josef Bacik <jbacik@redhat.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-09-11 20:12:44 +00:00
ret = VM_FAULT_SIGBUS;
goto out_unlock;
}
ret = 0;
/* page is wholly or partially inside EOF */
if (page_start + PAGE_CACHE_SIZE > size)
zero_start = size & ~PAGE_CACHE_MASK;
else
zero_start = PAGE_CACHE_SIZE;
if (zero_start != PAGE_CACHE_SIZE) {
kaddr = kmap(page);
memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
flush_dcache_page(page);
kunmap(page);
}
ClearPageChecked(page);
set_page_dirty(page);
SetPageUptodate(page);
Btrfs: add extra flushing for renames and truncates Renames and truncates are both common ways to replace old data with new data. The filesystem can make an effort to make sure the new data is on disk before actually replacing the old data. This is especially important for rename, which many application use as though it were atomic for both the data and the metadata involved. The current btrfs code will happily replace a file that is fully on disk with one that was just created and still has pending IO. If we crash after transaction commit but before the IO is done, we'll end up replacing a good file with a zero length file. The solution used here is to create a list of inodes that need special ordering and force them to disk before the commit is done. This is similar to the ext3 style data=ordering, except it is only done on selected files. Btrfs is able to get away with this because it does not wait on commits very often, even for fsync (which use a sub-commit). For renames, we order the file when it wasn't already on disk and when it is replacing an existing file. Larger files are sent to filemap_flush right away (before the transaction handle is opened). For truncates, we order if the file goes from non-zero size down to zero size. This is a little different, because at the time of the truncate the file has no dirty bytes to order. But, we flag the inode so that it is added to the ordered list on close (via release method). We also immediately add it to the ordered list of the current transaction so that we can try to flush down any writes the application sneaks in before commit. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-03-31 17:27:11 +00:00
BTRFS_I(inode)->last_trans = root->fs_info->generation;
BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
out_unlock:
if (!ret)
return VM_FAULT_LOCKED;
unlock_page(page);
out:
btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
out_noreserve:
return ret;
}
static int btrfs_truncate(struct inode *inode)
{
struct btrfs_root *root = BTRFS_I(inode)->root;
struct btrfs_block_rsv *rsv;
int ret;
int err = 0;
struct btrfs_trans_handle *trans;
unsigned long nr;
u64 mask = root->sectorsize - 1;
u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
if (ret)
return ret;
btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
/*
* Yes ladies and gentelment, this is indeed ugly. The fact is we have
* 3 things going on here
*
* 1) We need to reserve space for our orphan item and the space to
* delete our orphan item. Lord knows we don't want to have a dangling
* orphan item because we didn't reserve space to remove it.
*
* 2) We need to reserve space to update our inode.
*
* 3) We need to have something to cache all the space that is going to
* be free'd up by the truncate operation, but also have some slack
* space reserved in case it uses space during the truncate (thank you
* very much snapshotting).
*
* And we need these to all be seperate. The fact is we can use alot of
* space doing the truncate, and we have no earthly idea how much space
* we will use, so we need the truncate reservation to be seperate so it
* doesn't end up using space reserved for updating the inode or
* removing the orphan item. We also need to be able to stop the
* transaction and start a new one, which means we need to be able to
* update the inode several times, and we have no idea of knowing how
* many times that will be, so we can't just reserve 1 item for the
* entirety of the opration, so that has to be done seperately as well.
* Then there is the orphan item, which does indeed need to be held on
* to for the whole operation, and we need nobody to touch this reserved
* space except the orphan code.
*
* So that leaves us with
*
* 1) root->orphan_block_rsv - for the orphan deletion.
* 2) rsv - for the truncate reservation, which we will steal from the
* transaction reservation.
* 3) fs_info->trans_block_rsv - this will have 1 items worth left for
* updating the inode.
*/
rsv = btrfs_alloc_block_rsv(root);
if (!rsv)
return -ENOMEM;
rsv->size = min_size;
/*
* 1 for the truncate slack space
* 1 for the orphan item we're going to add
* 1 for the orphan item deletion
* 1 for updating the inode.
*/
trans = btrfs_start_transaction(root, 4);
if (IS_ERR(trans)) {
err = PTR_ERR(trans);
goto out;
}
/* Migrate the slack space for the truncate to our reserve */
ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
min_size);
BUG_ON(ret);
ret = btrfs_orphan_add(trans, inode);
if (ret) {
btrfs_end_transaction(trans, root);
goto out;
}
Btrfs: add extra flushing for renames and truncates Renames and truncates are both common ways to replace old data with new data. The filesystem can make an effort to make sure the new data is on disk before actually replacing the old data. This is especially important for rename, which many application use as though it were atomic for both the data and the metadata involved. The current btrfs code will happily replace a file that is fully on disk with one that was just created and still has pending IO. If we crash after transaction commit but before the IO is done, we'll end up replacing a good file with a zero length file. The solution used here is to create a list of inodes that need special ordering and force them to disk before the commit is done. This is similar to the ext3 style data=ordering, except it is only done on selected files. Btrfs is able to get away with this because it does not wait on commits very often, even for fsync (which use a sub-commit). For renames, we order the file when it wasn't already on disk and when it is replacing an existing file. Larger files are sent to filemap_flush right away (before the transaction handle is opened). For truncates, we order if the file goes from non-zero size down to zero size. This is a little different, because at the time of the truncate the file has no dirty bytes to order. But, we flag the inode so that it is added to the ordered list on close (via release method). We also immediately add it to the ordered list of the current transaction so that we can try to flush down any writes the application sneaks in before commit. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-03-31 17:27:11 +00:00
/*
* setattr is responsible for setting the ordered_data_close flag,
* but that is only tested during the last file release. That
* could happen well after the next commit, leaving a great big
* window where new writes may get lost if someone chooses to write
* to this file after truncating to zero
*
* The inode doesn't have any dirty data here, and so if we commit
* this is a noop. If someone immediately starts writing to the inode
* it is very likely we'll catch some of their writes in this
* transaction, and the commit will find this file on the ordered
* data list with good things to send down.
*
* This is a best effort solution, there is still a window where
* using truncate to replace the contents of the file will
* end up with a zero length file after a crash.
*/
if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
btrfs_add_ordered_operation(trans, root, inode);
while (1) {
ret = btrfs_block_rsv_refill(root, rsv, min_size);
if (ret) {
/*
* This can only happen with the original transaction we
* started above, every other time we shouldn't have a
* transaction started yet.
*/
if (ret == -EAGAIN)
goto end_trans;
err = ret;
break;
}
if (!trans) {
/* Just need the 1 for updating the inode */
trans = btrfs_start_transaction(root, 1);
if (IS_ERR(trans)) {
ret = err = PTR_ERR(trans);
trans = NULL;
break;
}
}
trans->block_rsv = rsv;
ret = btrfs_truncate_inode_items(trans, root, inode,
inode->i_size,
BTRFS_EXTENT_DATA_KEY);
if (ret != -EAGAIN) {
err = ret;
break;
}
trans->block_rsv = &root->fs_info->trans_block_rsv;
ret = btrfs_update_inode(trans, root, inode);
if (ret) {
err = ret;
break;
}
end_trans:
nr = trans->blocks_used;
btrfs_end_transaction(trans, root);
trans = NULL;
btrfs_btree_balance_dirty(root, nr);
}
if (ret == 0 && inode->i_nlink > 0) {
trans->block_rsv = root->orphan_block_rsv;
ret = btrfs_orphan_del(trans, inode);
if (ret)
err = ret;
} else if (ret && inode->i_nlink > 0) {
/*
* Failed to do the truncate, remove us from the in memory
* orphan list.
*/
ret = btrfs_orphan_del(NULL, inode);
}
if (trans) {
trans->block_rsv = &root->fs_info->trans_block_rsv;
ret = btrfs_update_inode(trans, root, inode);
if (ret && !err)
err = ret;
nr = trans->blocks_used;
ret = btrfs_end_transaction(trans, root);
btrfs_btree_balance_dirty(root, nr);
}
out:
btrfs_free_block_rsv(root, rsv);
if (ret && !err)
err = ret;
return err;
}
/*
* create a new subvolume directory/inode (helper for the ioctl).
*/
int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
struct btrfs_root *new_root, u64 new_dirid)
{
struct inode *inode;
int err;
u64 index = 0;
inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
new_dirid, new_dirid,
S_IFDIR | (~current_umask() & S_IRWXUGO),
&index);
if (IS_ERR(inode))
return PTR_ERR(inode);
inode->i_op = &btrfs_dir_inode_operations;
inode->i_fop = &btrfs_dir_file_operations;
set_nlink(inode, 1);
btrfs_i_size_write(inode, 0);
err = btrfs_update_inode(trans, new_root, inode);
BUG_ON(err);
iput(inode);
return 0;
}
struct inode *btrfs_alloc_inode(struct super_block *sb)
{
struct btrfs_inode *ei;
struct inode *inode;
ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
if (!ei)
return NULL;
ei->root = NULL;
ei->space_info = NULL;
ei->generation = 0;
ei->sequence = 0;
ei->last_trans = 0;
ei->last_sub_trans = 0;
ei->logged_trans = 0;
ei->delalloc_bytes = 0;
ei->disk_i_size = 0;
ei->flags = 0;
ei->csum_bytes = 0;
ei->index_cnt = (u64)-1;
ei->last_unlink_trans = 0;
spin_lock_init(&ei->lock);
ei->outstanding_extents = 0;
ei->reserved_extents = 0;
ei->ordered_data_close = 0;
ei->orphan_meta_reserved = 0;
ei->dummy_inode = 0;
ei->in_defrag = 0;
Btrfs: fix our reservations for updating an inode when completing io People have been reporting ENOSPC crashes in finish_ordered_io. This is because we try to steal from the delalloc block rsv to satisfy a reservation to update the inode. The problem with this is we don't explicitly save space for updating the inode when doing delalloc. This is kind of a problem and we've gotten away with this because way back when we just stole from the delalloc reserve without any questions, and this worked out fine because generally speaking the leaf had been modified either by the mtime update when we did the original write or because we just updated the leaf when we inserted the file extent item, only on rare occasions had the leaf not actually been modified, and that was still ok because we'd just use a block or two out of the over-reservation that is delalloc. Then came the delayed inode stuff. This is amazing, except it wants a full reservation for updating the inode since it may do it at some point down the road after we've written the blocks and we have to recow everything again. This worked out because the delayed inode stuff just stole from the global reserve, that is until recently when I changed that because it caused other problems. So here we are, we're doing everything right and being screwed for it. So take an extra reservation for the inode at delalloc reservation time and carry it through the life of the delalloc reservation. If we need it we can steal it in the delayed inode stuff. If we have already stolen it try and do a normal metadata reservation. If that fails try to steal from the delalloc reservation. If _that_ fails we'll get a WARN_ON() so I can start thinking of a better way to solve this and in the meantime we'll steal from the global reserve. With this patch I ran xfstests 13 in a loop for a couple of hours and didn't see any problems. Signed-off-by: Josef Bacik <josef@redhat.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-11-08 20:47:34 +00:00
ei->delalloc_meta_reserved = 0;
ei->force_compress = BTRFS_COMPRESS_NONE;
btrfs: implement delayed inode items operation Changelog V5 -> V6: - Fix oom when the memory load is high, by storing the delayed nodes into the root's radix tree, and letting btrfs inodes go. Changelog V4 -> V5: - Fix the race on adding the delayed node to the inode, which is spotted by Chris Mason. - Merge Chris Mason's incremental patch into this patch. - Fix deadlock between readdir() and memory fault, which is reported by Itaru Kitayama. Changelog V3 -> V4: - Fix nested lock, which is reported by Itaru Kitayama, by updating space cache inode in time. Changelog V2 -> V3: - Fix the race between the delayed worker and the task which does delayed items balance, which is reported by Tsutomu Itoh. - Modify the patch address David Sterba's comment. - Fix the bug of the cpu recursion spinlock, reported by Chris Mason Changelog V1 -> V2: - break up the global rb-tree, use a list to manage the delayed nodes, which is created for every directory and file, and used to manage the delayed directory name index items and the delayed inode item. - introduce a worker to deal with the delayed nodes. Compare with Ext3/4, the performance of file creation and deletion on btrfs is very poor. the reason is that btrfs must do a lot of b+ tree insertions, such as inode item, directory name item, directory name index and so on. If we can do some delayed b+ tree insertion or deletion, we can improve the performance, so we made this patch which implemented delayed directory name index insertion/deletion and delayed inode update. Implementation: - introduce a delayed root object into the filesystem, that use two lists to manage the delayed nodes which are created for every file/directory. One is used to manage all the delayed nodes that have delayed items. And the other is used to manage the delayed nodes which is waiting to be dealt with by the work thread. - Every delayed node has two rb-tree, one is used to manage the directory name index which is going to be inserted into b+ tree, and the other is used to manage the directory name index which is going to be deleted from b+ tree. - introduce a worker to deal with the delayed operation. This worker is used to deal with the works of the delayed directory name index items insertion and deletion and the delayed inode update. When the delayed items is beyond the lower limit, we create works for some delayed nodes and insert them into the work queue of the worker, and then go back. When the delayed items is beyond the upper bound, we create works for all the delayed nodes that haven't been dealt with, and insert them into the work queue of the worker, and then wait for that the untreated items is below some threshold value. - When we want to insert a directory name index into b+ tree, we just add the information into the delayed inserting rb-tree. And then we check the number of the delayed items and do delayed items balance. (The balance policy is above.) - When we want to delete a directory name index from the b+ tree, we search it in the inserting rb-tree at first. If we look it up, just drop it. If not, add the key of it into the delayed deleting rb-tree. Similar to the delayed inserting rb-tree, we also check the number of the delayed items and do delayed items balance. (The same to inserting manipulation) - When we want to update the metadata of some inode, we cached the data of the inode into the delayed node. the worker will flush it into the b+ tree after dealing with the delayed insertion and deletion. - We will move the delayed node to the tail of the list after we access the delayed node, By this way, we can cache more delayed items and merge more inode updates. - If we want to commit transaction, we will deal with all the delayed node. - the delayed node will be freed when we free the btrfs inode. - Before we log the inode items, we commit all the directory name index items and the delayed inode update. I did a quick test by the benchmark tool[1] and found we can improve the performance of file creation by ~15%, and file deletion by ~20%. Before applying this patch: Create files: Total files: 50000 Total time: 1.096108 Average time: 0.000022 Delete files: Total files: 50000 Total time: 1.510403 Average time: 0.000030 After applying this patch: Create files: Total files: 50000 Total time: 0.932899 Average time: 0.000019 Delete files: Total files: 50000 Total time: 1.215732 Average time: 0.000024 [1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3 Many thanks for Kitayama-san's help! Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Reviewed-by: David Sterba <dave@jikos.cz> Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com> Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-04-22 10:12:22 +00:00
ei->delayed_node = NULL;
inode = &ei->vfs_inode;
extent_map_tree_init(&ei->extent_tree);
extent_io_tree_init(&ei->io_tree, &inode->i_data);
extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
mutex_init(&ei->log_mutex);
mutex_init(&ei->delalloc_mutex);
btrfs_ordered_inode_tree_init(&ei->ordered_tree);
INIT_LIST_HEAD(&ei->i_orphan);
INIT_LIST_HEAD(&ei->delalloc_inodes);
Btrfs: add extra flushing for renames and truncates Renames and truncates are both common ways to replace old data with new data. The filesystem can make an effort to make sure the new data is on disk before actually replacing the old data. This is especially important for rename, which many application use as though it were atomic for both the data and the metadata involved. The current btrfs code will happily replace a file that is fully on disk with one that was just created and still has pending IO. If we crash after transaction commit but before the IO is done, we'll end up replacing a good file with a zero length file. The solution used here is to create a list of inodes that need special ordering and force them to disk before the commit is done. This is similar to the ext3 style data=ordering, except it is only done on selected files. Btrfs is able to get away with this because it does not wait on commits very often, even for fsync (which use a sub-commit). For renames, we order the file when it wasn't already on disk and when it is replacing an existing file. Larger files are sent to filemap_flush right away (before the transaction handle is opened). For truncates, we order if the file goes from non-zero size down to zero size. This is a little different, because at the time of the truncate the file has no dirty bytes to order. But, we flag the inode so that it is added to the ordered list on close (via release method). We also immediately add it to the ordered list of the current transaction so that we can try to flush down any writes the application sneaks in before commit. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-03-31 17:27:11 +00:00
INIT_LIST_HEAD(&ei->ordered_operations);
RB_CLEAR_NODE(&ei->rb_node);
return inode;
}
2011-01-07 06:49:49 +00:00
static void btrfs_i_callback(struct rcu_head *head)
{
struct inode *inode = container_of(head, struct inode, i_rcu);
kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
}
void btrfs_destroy_inode(struct inode *inode)
{
struct btrfs_ordered_extent *ordered;
Btrfs: add extra flushing for renames and truncates Renames and truncates are both common ways to replace old data with new data. The filesystem can make an effort to make sure the new data is on disk before actually replacing the old data. This is especially important for rename, which many application use as though it were atomic for both the data and the metadata involved. The current btrfs code will happily replace a file that is fully on disk with one that was just created and still has pending IO. If we crash after transaction commit but before the IO is done, we'll end up replacing a good file with a zero length file. The solution used here is to create a list of inodes that need special ordering and force them to disk before the commit is done. This is similar to the ext3 style data=ordering, except it is only done on selected files. Btrfs is able to get away with this because it does not wait on commits very often, even for fsync (which use a sub-commit). For renames, we order the file when it wasn't already on disk and when it is replacing an existing file. Larger files are sent to filemap_flush right away (before the transaction handle is opened). For truncates, we order if the file goes from non-zero size down to zero size. This is a little different, because at the time of the truncate the file has no dirty bytes to order. But, we flag the inode so that it is added to the ordered list on close (via release method). We also immediately add it to the ordered list of the current transaction so that we can try to flush down any writes the application sneaks in before commit. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-03-31 17:27:11 +00:00
struct btrfs_root *root = BTRFS_I(inode)->root;
WARN_ON(!list_empty(&inode->i_dentry));
WARN_ON(inode->i_data.nrpages);
WARN_ON(BTRFS_I(inode)->outstanding_extents);
WARN_ON(BTRFS_I(inode)->reserved_extents);
WARN_ON(BTRFS_I(inode)->delalloc_bytes);
WARN_ON(BTRFS_I(inode)->csum_bytes);
/*
* This can happen where we create an inode, but somebody else also
* created the same inode and we need to destroy the one we already
* created.
*/
if (!root)
goto free;
Btrfs: add extra flushing for renames and truncates Renames and truncates are both common ways to replace old data with new data. The filesystem can make an effort to make sure the new data is on disk before actually replacing the old data. This is especially important for rename, which many application use as though it were atomic for both the data and the metadata involved. The current btrfs code will happily replace a file that is fully on disk with one that was just created and still has pending IO. If we crash after transaction commit but before the IO is done, we'll end up replacing a good file with a zero length file. The solution used here is to create a list of inodes that need special ordering and force them to disk before the commit is done. This is similar to the ext3 style data=ordering, except it is only done on selected files. Btrfs is able to get away with this because it does not wait on commits very often, even for fsync (which use a sub-commit). For renames, we order the file when it wasn't already on disk and when it is replacing an existing file. Larger files are sent to filemap_flush right away (before the transaction handle is opened). For truncates, we order if the file goes from non-zero size down to zero size. This is a little different, because at the time of the truncate the file has no dirty bytes to order. But, we flag the inode so that it is added to the ordered list on close (via release method). We also immediately add it to the ordered list of the current transaction so that we can try to flush down any writes the application sneaks in before commit. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-03-31 17:27:11 +00:00
/*
* Make sure we're properly removed from the ordered operation
* lists.
*/
smp_mb();
if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
spin_lock(&root->fs_info->ordered_extent_lock);
list_del_init(&BTRFS_I(inode)->ordered_operations);
spin_unlock(&root->fs_info->ordered_extent_lock);
}
spin_lock(&root->orphan_lock);
if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
(unsigned long long)btrfs_ino(inode));
list_del_init(&BTRFS_I(inode)->i_orphan);
}
spin_unlock(&root->orphan_lock);
while (1) {
ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
if (!ordered)
break;
else {
printk(KERN_ERR "btrfs found ordered "
"extent %llu %llu on inode cleanup\n",
(unsigned long long)ordered->file_offset,
(unsigned long long)ordered->len);
btrfs_remove_ordered_extent(inode, ordered);
btrfs_put_ordered_extent(ordered);
btrfs_put_ordered_extent(ordered);
}
}
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE) This commit introduces a new kind of back reference for btrfs metadata. Once a filesystem has been mounted with this commit, IT WILL NO LONGER BE MOUNTABLE BY OLDER KERNELS. When a tree block in subvolume tree is cow'd, the reference counts of all extents it points to are increased by one. At transaction commit time, the old root of the subvolume is recorded in a "dead root" data structure, and the btree it points to is later walked, dropping reference counts and freeing any blocks where the reference count goes to 0. The increments done during cow and decrements done after commit cancel out, and the walk is a very expensive way to go about freeing the blocks that are no longer referenced by the new btree root. This commit reduces the transaction overhead by avoiding the need for dead root records. When a non-shared tree block is cow'd, we free the old block at once, and the new block inherits old block's references. When a tree block with reference count > 1 is cow'd, we increase the reference counts of all extents the new block points to by one, and decrease the old block's reference count by one. This dead tree avoidance code removes the need to modify the reference counts of lower level extents when a non-shared tree block is cow'd. But we still need to update back ref for all pointers in the block. This is because the location of the block is recorded in the back ref item. We can solve this by introducing a new type of back ref. The new back ref provides information about pointer's key, level and in which tree the pointer lives. This information allow us to find the pointer by searching the tree. The shortcoming of the new back ref is that it only works for pointers in tree blocks referenced by their owner trees. This is mostly a problem for snapshots, where resolving one of these fuzzy back references would be O(number_of_snapshots) and quite slow. The solution used here is to use the fuzzy back references in the common case where a given tree block is only referenced by one root, and use the full back references when multiple roots have a reference on a given block. This commit adds per subvolume red-black tree to keep trace of cached inodes. The red-black tree helps the balancing code to find cached inodes whose inode numbers within a given range. This commit improves the balancing code by introducing several data structures to keep the state of balancing. The most important one is the back ref cache. It caches how the upper level tree blocks are referenced. This greatly reduce the overhead of checking back ref. The improved balancing code scales significantly better with a large number of snapshots. This is a very large commit and was written in a number of pieces. But, they depend heavily on the disk format change and were squashed together to make sure git bisect didn't end up in a bad state wrt space balancing or the format change. Signed-off-by: Yan Zheng <zheng.yan@oracle.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 14:45:14 +00:00
inode_tree_del(inode);
btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
free:
btrfs: implement delayed inode items operation Changelog V5 -> V6: - Fix oom when the memory load is high, by storing the delayed nodes into the root's radix tree, and letting btrfs inodes go. Changelog V4 -> V5: - Fix the race on adding the delayed node to the inode, which is spotted by Chris Mason. - Merge Chris Mason's incremental patch into this patch. - Fix deadlock between readdir() and memory fault, which is reported by Itaru Kitayama. Changelog V3 -> V4: - Fix nested lock, which is reported by Itaru Kitayama, by updating space cache inode in time. Changelog V2 -> V3: - Fix the race between the delayed worker and the task which does delayed items balance, which is reported by Tsutomu Itoh. - Modify the patch address David Sterba's comment. - Fix the bug of the cpu recursion spinlock, reported by Chris Mason Changelog V1 -> V2: - break up the global rb-tree, use a list to manage the delayed nodes, which is created for every directory and file, and used to manage the delayed directory name index items and the delayed inode item. - introduce a worker to deal with the delayed nodes. Compare with Ext3/4, the performance of file creation and deletion on btrfs is very poor. the reason is that btrfs must do a lot of b+ tree insertions, such as inode item, directory name item, directory name index and so on. If we can do some delayed b+ tree insertion or deletion, we can improve the performance, so we made this patch which implemented delayed directory name index insertion/deletion and delayed inode update. Implementation: - introduce a delayed root object into the filesystem, that use two lists to manage the delayed nodes which are created for every file/directory. One is used to manage all the delayed nodes that have delayed items. And the other is used to manage the delayed nodes which is waiting to be dealt with by the work thread. - Every delayed node has two rb-tree, one is used to manage the directory name index which is going to be inserted into b+ tree, and the other is used to manage the directory name index which is going to be deleted from b+ tree. - introduce a worker to deal with the delayed operation. This worker is used to deal with the works of the delayed directory name index items insertion and deletion and the delayed inode update. When the delayed items is beyond the lower limit, we create works for some delayed nodes and insert them into the work queue of the worker, and then go back. When the delayed items is beyond the upper bound, we create works for all the delayed nodes that haven't been dealt with, and insert them into the work queue of the worker, and then wait for that the untreated items is below some threshold value. - When we want to insert a directory name index into b+ tree, we just add the information into the delayed inserting rb-tree. And then we check the number of the delayed items and do delayed items balance. (The balance policy is above.) - When we want to delete a directory name index from the b+ tree, we search it in the inserting rb-tree at first. If we look it up, just drop it. If not, add the key of it into the delayed deleting rb-tree. Similar to the delayed inserting rb-tree, we also check the number of the delayed items and do delayed items balance. (The same to inserting manipulation) - When we want to update the metadata of some inode, we cached the data of the inode into the delayed node. the worker will flush it into the b+ tree after dealing with the delayed insertion and deletion. - We will move the delayed node to the tail of the list after we access the delayed node, By this way, we can cache more delayed items and merge more inode updates. - If we want to commit transaction, we will deal with all the delayed node. - the delayed node will be freed when we free the btrfs inode. - Before we log the inode items, we commit all the directory name index items and the delayed inode update. I did a quick test by the benchmark tool[1] and found we can improve the performance of file creation by ~15%, and file deletion by ~20%. Before applying this patch: Create files: Total files: 50000 Total time: 1.096108 Average time: 0.000022 Delete files: Total files: 50000 Total time: 1.510403 Average time: 0.000030 After applying this patch: Create files: Total files: 50000 Total time: 0.932899 Average time: 0.000019 Delete files: Total files: 50000 Total time: 1.215732 Average time: 0.000024 [1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3 Many thanks for Kitayama-san's help! Signed-off-by: Miao Xie <miaox@cn.fujitsu.com> Reviewed-by: David Sterba <dave@jikos.cz> Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com> Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2011-04-22 10:12:22 +00:00
btrfs_remove_delayed_node(inode);
2011-01-07 06:49:49 +00:00
call_rcu(&inode->i_rcu, btrfs_i_callback);
}
int btrfs_drop_inode(struct inode *inode)
{
struct btrfs_root *root = BTRFS_I(inode)->root;
if (btrfs_root_refs(&root->root_item) == 0 &&
!btrfs_is_free_space_inode(root, inode))
return 1;
else
return generic_drop_inode(inode);
}
static void init_once(void *foo)
{
struct btrfs_inode *ei = (struct btrfs_inode *) foo;
inode_init_once(&ei->vfs_inode);
}
void btrfs_destroy_cachep(void)
{
if (btrfs_inode_cachep)
kmem_cache_destroy(btrfs_inode_cachep);
if (btrfs_trans_handle_cachep)
kmem_cache_destroy(btrfs_trans_handle_cachep);
if (btrfs_transaction_cachep)
kmem_cache_destroy(btrfs_transaction_cachep);
if (btrfs_path_cachep)
kmem_cache_destroy(btrfs_path_cachep);
if (btrfs_free_space_cachep)
kmem_cache_destroy(btrfs_free_space_cachep);
}
int btrfs_init_cachep(void)
{
btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
sizeof(struct btrfs_inode), 0,
SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
if (!btrfs_inode_cachep)
goto fail;
btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
sizeof(struct btrfs_trans_handle), 0,
SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
if (!btrfs_trans_handle_cachep)
goto fail;
btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
sizeof(struct btrfs_transaction), 0,
SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
if (!btrfs_transaction_cachep)
goto fail;
btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
sizeof(struct btrfs_path), 0,
SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
if (!btrfs_path_cachep)
goto fail;
btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
sizeof(struct btrfs_free_space), 0,
SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
if (!btrfs_free_space_cachep)
goto fail;
return 0;
fail:
btrfs_destroy_cachep();
return -ENOMEM;
}
static int btrfs_getattr(struct vfsmount *mnt,
struct dentry *dentry, struct kstat *stat)
{
struct inode *inode = dentry->d_inode;
u32 blocksize = inode->i_sb->s_blocksize;
generic_fillattr(inode, stat);
stat->dev = BTRFS_I(inode)->root->anon_dev;
stat->blksize = PAGE_CACHE_SIZE;
stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
return 0;
}
/*
* If a file is moved, it will inherit the cow and compression flags of the new
* directory.
*/
static void fixup_inode_flags(struct inode *dir, struct inode *inode)
{
struct btrfs_inode *b_dir = BTRFS_I(dir);
struct btrfs_inode *b_inode = BTRFS_I(inode);
if (b_dir->flags & BTRFS_INODE_NODATACOW)
b_inode->flags |= BTRFS_INODE_NODATACOW;
else
b_inode->flags &= ~BTRFS_INODE_NODATACOW;
if (b_dir->flags & BTRFS_INODE_COMPRESS)
b_inode->flags |= BTRFS_INODE_COMPRESS;
else
b_inode->flags &= ~BTRFS_INODE_COMPRESS;
}
static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
struct inode *new_dir, struct dentry *new_dentry)
{
struct btrfs_trans_handle *trans;
struct btrfs_root *root = BTRFS_I(old_dir)->root;
struct btrfs_root *dest = BTRFS_I(new_dir)->root;
struct inode *new_inode = new_dentry->d_inode;
struct inode *old_inode = old_dentry->d_inode;
struct timespec ctime = CURRENT_TIME;
u64 index = 0;
u64 root_objectid;
int ret;
u64 old_ino = btrfs_ino(old_inode);
if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
return -EPERM;
/* we only allow rename subvolume link between subvolumes */
if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
return -EXDEV;
if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
(new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
return -ENOTEMPTY;
if (S_ISDIR(old_inode->i_mode) && new_inode &&
new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
return -ENOTEMPTY;
Btrfs: add extra flushing for renames and truncates Renames and truncates are both common ways to replace old data with new data. The filesystem can make an effort to make sure the new data is on disk before actually replacing the old data. This is especially important for rename, which many application use as though it were atomic for both the data and the metadata involved. The current btrfs code will happily replace a file that is fully on disk with one that was just created and still has pending IO. If we crash after transaction commit but before the IO is done, we'll end up replacing a good file with a zero length file. The solution used here is to create a list of inodes that need special ordering and force them to disk before the commit is done. This is similar to the ext3 style data=ordering, except it is only done on selected files. Btrfs is able to get away with this because it does not wait on commits very often, even for fsync (which use a sub-commit). For renames, we order the file when it wasn't already on disk and when it is replacing an existing file. Larger files are sent to filemap_flush right away (before the transaction handle is opened). For truncates, we order if the file goes from non-zero size down to zero size. This is a little different, because at the time of the truncate the file has no dirty bytes to order. But, we flag the inode so that it is added to the ordered list on close (via release method). We also immediately add it to the ordered list of the current transaction so that we can try to flush down any writes the application sneaks in before commit. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-03-31 17:27:11 +00:00
/*
* we're using rename to replace one file with another.
* and the replacement file is large. Start IO on it now so
* we don't add too much work to the end of the transaction
*/
if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
Btrfs: add extra flushing for renames and truncates Renames and truncates are both common ways to replace old data with new data. The filesystem can make an effort to make sure the new data is on disk before actually replacing the old data. This is especially important for rename, which many application use as though it were atomic for both the data and the metadata involved. The current btrfs code will happily replace a file that is fully on disk with one that was just created and still has pending IO. If we crash after transaction commit but before the IO is done, we'll end up replacing a good file with a zero length file. The solution used here is to create a list of inodes that need special ordering and force them to disk before the commit is done. This is similar to the ext3 style data=ordering, except it is only done on selected files. Btrfs is able to get away with this because it does not wait on commits very often, even for fsync (which use a sub-commit). For renames, we order the file when it wasn't already on disk and when it is replacing an existing file. Larger files are sent to filemap_flush right away (before the transaction handle is opened). For truncates, we order if the file goes from non-zero size down to zero size. This is a little different, because at the time of the truncate the file has no dirty bytes to order. But, we flag the inode so that it is added to the ordered list on close (via release method). We also immediately add it to the ordered list of the current transaction so that we can try to flush down any writes the application sneaks in before commit. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-03-31 17:27:11 +00:00
old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
filemap_flush(old_inode->i_mapping);
/* close the racy window with snapshot create/destroy ioctl */
if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
down_read(&root->fs_info->subvol_sem);
/*
* We want to reserve the absolute worst case amount of items. So if
* both inodes are subvols and we need to unlink them then that would
* require 4 item modifications, but if they are both normal inodes it
* would require 5 item modifications, so we'll assume their normal
* inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
* should cover the worst case number of items we'll modify.
*/
trans = btrfs_start_transaction(root, 20);
if (IS_ERR(trans)) {
ret = PTR_ERR(trans);
goto out_notrans;
}
if (dest != root)
btrfs_record_root_in_trans(trans, dest);
ret = btrfs_set_inode_index(new_dir, &index);
if (ret)
goto out_fail;
Btrfs: add extra flushing for renames and truncates Renames and truncates are both common ways to replace old data with new data. The filesystem can make an effort to make sure the new data is on disk before actually replacing the old data. This is especially important for rename, which many application use as though it were atomic for both the data and the metadata involved. The current btrfs code will happily replace a file that is fully on disk with one that was just created and still has pending IO. If we crash after transaction commit but before the IO is done, we'll end up replacing a good file with a zero length file. The solution used here is to create a list of inodes that need special ordering and force them to disk before the commit is done. This is similar to the ext3 style data=ordering, except it is only done on selected files. Btrfs is able to get away with this because it does not wait on commits very often, even for fsync (which use a sub-commit). For renames, we order the file when it wasn't already on disk and when it is replacing an existing file. Larger files are sent to filemap_flush right away (before the transaction handle is opened). For truncates, we order if the file goes from non-zero size down to zero size. This is a little different, because at the time of the truncate the file has no dirty bytes to order. But, we flag the inode so that it is added to the ordered list on close (via release method). We also immediately add it to the ordered list of the current transaction so that we can try to flush down any writes the application sneaks in before commit. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-03-31 17:27:11 +00:00
if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
/* force full log commit if subvolume involved. */
root->fs_info->last_trans_log_full_commit = trans->transid;
} else {
ret = btrfs_insert_inode_ref(trans, dest,
new_dentry->d_name.name,
new_dentry->d_name.len,
old_ino,
btrfs_ino(new_dir), index);
if (ret)
goto out_fail;
/*
* this is an ugly little race, but the rename is required
* to make sure that if we crash, the inode is either at the
* old name or the new one. pinning the log transaction lets
* us make sure we don't allow a log commit to come in after
* we unlink the name but before we add the new name back in.
*/
btrfs_pin_log_trans(root);
}
Btrfs: add extra flushing for renames and truncates Renames and truncates are both common ways to replace old data with new data. The filesystem can make an effort to make sure the new data is on disk before actually replacing the old data. This is especially important for rename, which many application use as though it were atomic for both the data and the metadata involved. The current btrfs code will happily replace a file that is fully on disk with one that was just created and still has pending IO. If we crash after transaction commit but before the IO is done, we'll end up replacing a good file with a zero length file. The solution used here is to create a list of inodes that need special ordering and force them to disk before the commit is done. This is similar to the ext3 style data=ordering, except it is only done on selected files. Btrfs is able to get away with this because it does not wait on commits very often, even for fsync (which use a sub-commit). For renames, we order the file when it wasn't already on disk and when it is replacing an existing file. Larger files are sent to filemap_flush right away (before the transaction handle is opened). For truncates, we order if the file goes from non-zero size down to zero size. This is a little different, because at the time of the truncate the file has no dirty bytes to order. But, we flag the inode so that it is added to the ordered list on close (via release method). We also immediately add it to the ordered list of the current transaction so that we can try to flush down any writes the application sneaks in before commit. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-03-31 17:27:11 +00:00
/*
* make sure the inode gets flushed if it is replacing
* something.
*/
if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
Btrfs: add extra flushing for renames and truncates Renames and truncates are both common ways to replace old data with new data. The filesystem can make an effort to make sure the new data is on disk before actually replacing the old data. This is especially important for rename, which many application use as though it were atomic for both the data and the metadata involved. The current btrfs code will happily replace a file that is fully on disk with one that was just created and still has pending IO. If we crash after transaction commit but before the IO is done, we'll end up replacing a good file with a zero length file. The solution used here is to create a list of inodes that need special ordering and force them to disk before the commit is done. This is similar to the ext3 style data=ordering, except it is only done on selected files. Btrfs is able to get away with this because it does not wait on commits very often, even for fsync (which use a sub-commit). For renames, we order the file when it wasn't already on disk and when it is replacing an existing file. Larger files are sent to filemap_flush right away (before the transaction handle is opened). For truncates, we order if the file goes from non-zero size down to zero size. This is a little different, because at the time of the truncate the file has no dirty bytes to order. But, we flag the inode so that it is added to the ordered list on close (via release method). We also immediately add it to the ordered list of the current transaction so that we can try to flush down any writes the application sneaks in before commit. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-03-31 17:27:11 +00:00
btrfs_add_ordered_operation(trans, root, old_inode);
old_dir->i_ctime = old_dir->i_mtime = ctime;
new_dir->i_ctime = new_dir->i_mtime = ctime;
old_inode->i_ctime = ctime;
Btrfs: tree logging unlink/rename fixes The tree logging code allows individual files or directories to be logged without including operations on other files and directories in the FS. It tries to commit the minimal set of changes to disk in order to fsync the single file or directory that was sent to fsync or O_SYNC. The tree logging code was allowing files and directories to be unlinked if they were part of a rename operation where only one directory in the rename was in the fsync log. This patch adds a few new rules to the tree logging. 1) on rename or unlink, if the inode being unlinked isn't in the fsync log, we must force a full commit before doing an fsync of the directory where the unlink was done. The commit isn't done during the unlink, but it is forced the next time we try to log the parent directory. Solution: record transid of last unlink/rename per directory when the directory wasn't already logged. For renames this is only done when renaming to a different directory. mkdir foo/some_dir normal commit rename foo/some_dir foo2/some_dir mkdir foo/some_dir fsync foo/some_dir/some_file The fsync above will unlink the original some_dir without recording it in its new location (foo2). After a crash, some_dir will be gone unless the fsync of some_file forces a full commit 2) we must log any new names for any file or dir that is in the fsync log. This way we make sure not to lose files that are unlinked during the same transaction. 2a) we must log any new names for any file or dir during rename when the directory they are being removed from was logged. 2a is actually the more important variant. Without the extra logging a crash might unlink the old name without recreating the new one 3) after a crash, we must go through any directories with a link count of zero and redo the rm -rf mkdir f1/foo normal commit rm -rf f1/foo fsync(f1) The directory f1 was fully removed from the FS, but fsync was never called on f1, only its parent dir. After a crash the rm -rf must be replayed. This must be able to recurse down the entire directory tree. The inode link count fixup code takes care of the ugly details. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-03-24 14:24:20 +00:00
if (old_dentry->d_parent != new_dentry->d_parent)
btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
old_dentry->d_name.name,
old_dentry->d_name.len);
} else {
ret = __btrfs_unlink_inode(trans, root, old_dir,
old_dentry->d_inode,
old_dentry->d_name.name,
old_dentry->d_name.len);
if (!ret)
ret = btrfs_update_inode(trans, root, old_inode);
}
BUG_ON(ret);
if (new_inode) {
new_inode->i_ctime = CURRENT_TIME;
if (unlikely(btrfs_ino(new_inode) ==
BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
root_objectid = BTRFS_I(new_inode)->location.objectid;
ret = btrfs_unlink_subvol(trans, dest, new_dir,
root_objectid,
new_dentry->d_name.name,
new_dentry->d_name.len);
BUG_ON(new_inode->i_nlink == 0);
} else {
ret = btrfs_unlink_inode(trans, dest, new_dir,
new_dentry->d_inode,
new_dentry->d_name.name,
new_dentry->d_name.len);
}
BUG_ON(ret);
if (new_inode->i_nlink == 0) {
ret = btrfs_orphan_add(trans, new_dentry->d_inode);
BUG_ON(ret);
}
}
fixup_inode_flags(new_dir, old_inode);
ret = btrfs_add_link(trans, new_dir, old_inode,
new_dentry->d_name.name,
new_dentry->d_name.len, 0, index);
BUG_ON(ret);
if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
struct dentry *parent = new_dentry->d_parent;
btrfs_log_new_name(trans, old_inode, old_dir, parent);
btrfs_end_log_trans(root);
}
out_fail:
btrfs_end_transaction(trans, root);
out_notrans:
if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
up_read(&root->fs_info->subvol_sem);
Btrfs: proper -ENOSPC handling At the start of a transaction we do a btrfs_reserve_metadata_space() and specify how many items we plan on modifying. Then once we've done our modifications and such, just call btrfs_unreserve_metadata_space() for the same number of items we reserved. For keeping track of metadata needed for data I've had to add an extent_io op for when we merge extents. This lets us track space properly when we are doing sequential writes, so we don't end up reserving way more metadata space than what we need. The only place where the metadata space accounting is not done is in the relocation code. This is because Yan is going to be reworking that code in the near future, so running btrfs-vol -b could still possibly result in a ENOSPC related panic. This patch also turns off the metadata_ratio stuff in order to allow users to more efficiently use their disk space. This patch makes it so we track how much metadata we need for an inode's delayed allocation extents by tracking how many extents are currently waiting for allocation. It introduces two new callbacks for the extent_io tree's, merge_extent_hook and split_extent_hook. These help us keep track of when we merge delalloc extents together and split them up. Reservations are handled prior to any actually dirty'ing occurs, and then we unreserve after we dirty. btrfs_unreserve_metadata_for_delalloc() will make the appropriate unreservations as needed based on the number of reservations we currently have and the number of extents we currently have. Doing the reservation outside of doing any of the actual dirty'ing lets us do things like filemap_flush() the inode to try and force delalloc to happen, or as a last resort actually start allocation on all delalloc inodes in the fs. This has survived dbench, fs_mark and an fsx torture test. Signed-off-by: Josef Bacik <jbacik@redhat.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-09-11 20:12:44 +00:00
return ret;
}
/*
* some fairly slow code that needs optimization. This walks the list
* of all the inodes with pending delalloc and forces them to disk.
*/
int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
{
struct list_head *head = &root->fs_info->delalloc_inodes;
struct btrfs_inode *binode;
struct inode *inode;
if (root->fs_info->sb->s_flags & MS_RDONLY)
return -EROFS;
spin_lock(&root->fs_info->delalloc_lock);
while (!list_empty(head)) {
binode = list_entry(head->next, struct btrfs_inode,
delalloc_inodes);
inode = igrab(&binode->vfs_inode);
if (!inode)
list_del_init(&binode->delalloc_inodes);
spin_unlock(&root->fs_info->delalloc_lock);
if (inode) {
filemap_flush(inode->i_mapping);
if (delay_iput)
btrfs_add_delayed_iput(inode);
else
iput(inode);
}
cond_resched();
spin_lock(&root->fs_info->delalloc_lock);
}
spin_unlock(&root->fs_info->delalloc_lock);
/* the filemap_flush will queue IO into the worker threads, but
* we have to make sure the IO is actually started and that
* ordered extents get created before we return
*/
atomic_inc(&root->fs_info->async_submit_draining);
while (atomic_read(&root->fs_info->nr_async_submits) ||
atomic_read(&root->fs_info->async_delalloc_pages)) {
wait_event(root->fs_info->async_submit_wait,
(atomic_read(&root->fs_info->nr_async_submits) == 0 &&
atomic_read(&root->fs_info->async_delalloc_pages) == 0));
}
atomic_dec(&root->fs_info->async_submit_draining);
return 0;
}
static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
const char *symname)
{
struct btrfs_trans_handle *trans;
struct btrfs_root *root = BTRFS_I(dir)->root;
struct btrfs_path *path;
struct btrfs_key key;
struct inode *inode = NULL;
int err;
int drop_inode = 0;
u64 objectid;
u64 index = 0 ;
int name_len;
int datasize;
unsigned long ptr;
struct btrfs_file_extent_item *ei;
struct extent_buffer *leaf;
unsigned long nr = 0;
name_len = strlen(symname) + 1;
if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
return -ENAMETOOLONG;
Btrfs: proper -ENOSPC handling At the start of a transaction we do a btrfs_reserve_metadata_space() and specify how many items we plan on modifying. Then once we've done our modifications and such, just call btrfs_unreserve_metadata_space() for the same number of items we reserved. For keeping track of metadata needed for data I've had to add an extent_io op for when we merge extents. This lets us track space properly when we are doing sequential writes, so we don't end up reserving way more metadata space than what we need. The only place where the metadata space accounting is not done is in the relocation code. This is because Yan is going to be reworking that code in the near future, so running btrfs-vol -b could still possibly result in a ENOSPC related panic. This patch also turns off the metadata_ratio stuff in order to allow users to more efficiently use their disk space. This patch makes it so we track how much metadata we need for an inode's delayed allocation extents by tracking how many extents are currently waiting for allocation. It introduces two new callbacks for the extent_io tree's, merge_extent_hook and split_extent_hook. These help us keep track of when we merge delalloc extents together and split them up. Reservations are handled prior to any actually dirty'ing occurs, and then we unreserve after we dirty. btrfs_unreserve_metadata_for_delalloc() will make the appropriate unreservations as needed based on the number of reservations we currently have and the number of extents we currently have. Doing the reservation outside of doing any of the actual dirty'ing lets us do things like filemap_flush() the inode to try and force delalloc to happen, or as a last resort actually start allocation on all delalloc inodes in the fs. This has survived dbench, fs_mark and an fsx torture test. Signed-off-by: Josef Bacik <jbacik@redhat.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-09-11 20:12:44 +00:00
/*
* 2 items for inode item and ref
* 2 items for dir items
* 1 item for xattr if selinux is on
*/
trans = btrfs_start_transaction(root, 5);
if (IS_ERR(trans))
return PTR_ERR(trans);
err = btrfs_find_free_ino(root, &objectid);
if (err)
goto out_unlock;
inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
dentry->d_name.len, btrfs_ino(dir), objectid,
S_IFLNK|S_IRWXUGO, &index);
if (IS_ERR(inode)) {
err = PTR_ERR(inode);
goto out_unlock;
}
err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
if (err) {
drop_inode = 1;
goto out_unlock;
}
/*
* If the active LSM wants to access the inode during
* d_instantiate it needs these. Smack checks to see
* if the filesystem supports xattrs by looking at the
* ops vector.
*/
inode->i_fop = &btrfs_file_operations;
inode->i_op = &btrfs_file_inode_operations;
err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
if (err)
drop_inode = 1;
else {
inode->i_mapping->a_ops = &btrfs_aops;
inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
}
if (drop_inode)
goto out_unlock;
path = btrfs_alloc_path();
if (!path) {
err = -ENOMEM;
drop_inode = 1;
goto out_unlock;
}
key.objectid = btrfs_ino(inode);
key.offset = 0;
btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
datasize = btrfs_file_extent_calc_inline_size(name_len);
err = btrfs_insert_empty_item(trans, root, path, &key,
datasize);
if (err) {
drop_inode = 1;
btrfs_free_path(path);
goto out_unlock;
}
leaf = path->nodes[0];
ei = btrfs_item_ptr(leaf, path->slots[0],
struct btrfs_file_extent_item);
btrfs_set_file_extent_generation(leaf, ei, trans->transid);
btrfs_set_file_extent_type(leaf, ei,
BTRFS_FILE_EXTENT_INLINE);
Btrfs: Add zlib compression support This is a large change for adding compression on reading and writing, both for inline and regular extents. It does some fairly large surgery to the writeback paths. Compression is off by default and enabled by mount -o compress. Even when the -o compress mount option is not used, it is possible to read compressed extents off the disk. If compression for a given set of pages fails to make them smaller, the file is flagged to avoid future compression attempts later. * While finding delalloc extents, the pages are locked before being sent down to the delalloc handler. This allows the delalloc handler to do complex things such as cleaning the pages, marking them writeback and starting IO on their behalf. * Inline extents are inserted at delalloc time now. This allows us to compress the data before inserting the inline extent, and it allows us to insert an inline extent that spans multiple pages. * All of the in-memory extent representations (extent_map.c, ordered-data.c etc) are changed to record both an in-memory size and an on disk size, as well as a flag for compression. From a disk format point of view, the extent pointers in the file are changed to record the on disk size of a given extent and some encoding flags. Space in the disk format is allocated for compression encoding, as well as encryption and a generic 'other' field. Neither the encryption or the 'other' field are currently used. In order to limit the amount of data read for a single random read in the file, the size of a compressed extent is limited to 128k. This is a software only limit, the disk format supports u64 sized compressed extents. In order to limit the ram consumed while processing extents, the uncompressed size of a compressed extent is limited to 256k. This is a software only limit and will be subject to tuning later. Checksumming is still done on compressed extents, and it is done on the uncompressed version of the data. This way additional encodings can be layered on without having to figure out which encoding to checksum. Compression happens at delalloc time, which is basically singled threaded because it is usually done by a single pdflush thread. This makes it tricky to spread the compression load across all the cpus on the box. We'll have to look at parallel pdflush walks of dirty inodes at a later time. Decompression is hooked into readpages and it does spread across CPUs nicely. Signed-off-by: Chris Mason <chris.mason@oracle.com>
2008-10-29 18:49:59 +00:00
btrfs_set_file_extent_encryption(leaf, ei, 0);
btrfs_set_file_extent_compression(leaf, ei, 0);
btrfs_set_file_extent_other_encoding(leaf, ei, 0);
btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
ptr = btrfs_file_extent_inline_start(ei);
write_extent_buffer(leaf, symname, ptr, name_len);
btrfs_mark_buffer_dirty(leaf);
btrfs_free_path(path);
inode->i_op = &btrfs_symlink_inode_operations;
inode->i_mapping->a_ops = &btrfs_symlink_aops;
inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
inode_set_bytes(inode, name_len);
btrfs_i_size_write(inode, name_len - 1);
err = btrfs_update_inode(trans, root, inode);
if (err)
drop_inode = 1;
out_unlock:
if (!err)
d_instantiate(dentry, inode);
nr = trans->blocks_used;
btrfs_end_transaction(trans, root);
if (drop_inode) {
inode_dec_link_count(inode);
iput(inode);
}
btrfs_btree_balance_dirty(root, nr);
return err;
}
static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
u64 start, u64 num_bytes, u64 min_size,
loff_t actual_len, u64 *alloc_hint,
struct btrfs_trans_handle *trans)
{
struct btrfs_root *root = BTRFS_I(inode)->root;
struct btrfs_key ins;
u64 cur_offset = start;
u64 i_size;
int ret = 0;
bool own_trans = true;
if (trans)
own_trans = false;
while (num_bytes > 0) {
if (own_trans) {
trans = btrfs_start_transaction(root, 3);
if (IS_ERR(trans)) {
ret = PTR_ERR(trans);
break;
}
}
ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
0, *alloc_hint, (u64)-1, &ins, 1);
if (ret) {
if (own_trans)
btrfs_end_transaction(trans, root);
break;
}
ret = insert_reserved_file_extent(trans, inode,
cur_offset, ins.objectid,
ins.offset, ins.offset,
ins.offset, 0, 0, 0,
BTRFS_FILE_EXTENT_PREALLOC);
BUG_ON(ret);
btrfs_drop_extent_cache(inode, cur_offset,
cur_offset + ins.offset -1, 0);
num_bytes -= ins.offset;
cur_offset += ins.offset;
*alloc_hint = ins.objectid + ins.offset;
inode->i_ctime = CURRENT_TIME;
BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
if (!(mode & FALLOC_FL_KEEP_SIZE) &&
(actual_len > inode->i_size) &&
(cur_offset > inode->i_size)) {
if (cur_offset > actual_len)
i_size = actual_len;
else
i_size = cur_offset;
i_size_write(inode, i_size);
btrfs_ordered_update_i_size(inode, i_size, NULL);
}
ret = btrfs_update_inode(trans, root, inode);
BUG_ON(ret);
if (own_trans)
btrfs_end_transaction(trans, root);
}
return ret;
}
int btrfs_prealloc_file_range(struct inode *inode, int mode,
u64 start, u64 num_bytes, u64 min_size,
loff_t actual_len, u64 *alloc_hint)
{
return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
min_size, actual_len, alloc_hint,
NULL);
}
int btrfs_prealloc_file_range_trans(struct inode *inode,
struct btrfs_trans_handle *trans, int mode,
u64 start, u64 num_bytes, u64 min_size,
loff_t actual_len, u64 *alloc_hint)
{
return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
min_size, actual_len, alloc_hint, trans);
}
static int btrfs_set_page_dirty(struct page *page)
{
return __set_page_dirty_nobuffers(page);
}
static int btrfs_permission(struct inode *inode, int mask)
{
struct btrfs_root *root = BTRFS_I(inode)->root;
umode_t mode = inode->i_mode;
if (mask & MAY_WRITE &&
(S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
if (btrfs_root_readonly(root))
return -EROFS;
if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
return -EACCES;
}
return generic_permission(inode, mask);
}
static const struct inode_operations btrfs_dir_inode_operations = {
.getattr = btrfs_getattr,
.lookup = btrfs_lookup,
.create = btrfs_create,
.unlink = btrfs_unlink,
.link = btrfs_link,
.mkdir = btrfs_mkdir,
.rmdir = btrfs_rmdir,
.rename = btrfs_rename,
.symlink = btrfs_symlink,
.setattr = btrfs_setattr,
.mknod = btrfs_mknod,
.setxattr = btrfs_setxattr,
.getxattr = btrfs_getxattr,
.listxattr = btrfs_listxattr,
.removexattr = btrfs_removexattr,
.permission = btrfs_permission,
.get_acl = btrfs_get_acl,
};
static const struct inode_operations btrfs_dir_ro_inode_operations = {
.lookup = btrfs_lookup,
.permission = btrfs_permission,
.get_acl = btrfs_get_acl,
};
static const struct file_operations btrfs_dir_file_operations = {
.llseek = generic_file_llseek,
.read = generic_read_dir,
.readdir = btrfs_real_readdir,
.unlocked_ioctl = btrfs_ioctl,
#ifdef CONFIG_COMPAT
.compat_ioctl = btrfs_ioctl,
#endif
.release = btrfs_release_file,
.fsync = btrfs_sync_file,
};
static struct extent_io_ops btrfs_extent_io_ops = {
.fill_delalloc = run_delalloc_range,
.submit_bio_hook = btrfs_submit_bio_hook,
.merge_bio_hook = btrfs_merge_bio_hook,
.readpage_end_io_hook = btrfs_readpage_end_io_hook,
.writepage_end_io_hook = btrfs_writepage_end_io_hook,
.writepage_start_hook = btrfs_writepage_start_hook,
.set_bit_hook = btrfs_set_bit_hook,
.clear_bit_hook = btrfs_clear_bit_hook,
Btrfs: proper -ENOSPC handling At the start of a transaction we do a btrfs_reserve_metadata_space() and specify how many items we plan on modifying. Then once we've done our modifications and such, just call btrfs_unreserve_metadata_space() for the same number of items we reserved. For keeping track of metadata needed for data I've had to add an extent_io op for when we merge extents. This lets us track space properly when we are doing sequential writes, so we don't end up reserving way more metadata space than what we need. The only place where the metadata space accounting is not done is in the relocation code. This is because Yan is going to be reworking that code in the near future, so running btrfs-vol -b could still possibly result in a ENOSPC related panic. This patch also turns off the metadata_ratio stuff in order to allow users to more efficiently use their disk space. This patch makes it so we track how much metadata we need for an inode's delayed allocation extents by tracking how many extents are currently waiting for allocation. It introduces two new callbacks for the extent_io tree's, merge_extent_hook and split_extent_hook. These help us keep track of when we merge delalloc extents together and split them up. Reservations are handled prior to any actually dirty'ing occurs, and then we unreserve after we dirty. btrfs_unreserve_metadata_for_delalloc() will make the appropriate unreservations as needed based on the number of reservations we currently have and the number of extents we currently have. Doing the reservation outside of doing any of the actual dirty'ing lets us do things like filemap_flush() the inode to try and force delalloc to happen, or as a last resort actually start allocation on all delalloc inodes in the fs. This has survived dbench, fs_mark and an fsx torture test. Signed-off-by: Josef Bacik <jbacik@redhat.com> Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-09-11 20:12:44 +00:00
.merge_extent_hook = btrfs_merge_extent_hook,
.split_extent_hook = btrfs_split_extent_hook,
};
/*
* btrfs doesn't support the bmap operation because swapfiles
* use bmap to make a mapping of extents in the file. They assume
* these extents won't change over the life of the file and they
* use the bmap result to do IO directly to the drive.
*
* the btrfs bmap call would return logical addresses that aren't
* suitable for IO and they also will change frequently as COW
* operations happen. So, swapfile + btrfs == corruption.
*
* For now we're avoiding this by dropping bmap.
*/
static const struct address_space_operations btrfs_aops = {
.readpage = btrfs_readpage,
.writepage = btrfs_writepage,
.writepages = btrfs_writepages,
.readpages = btrfs_readpages,
.direct_IO = btrfs_direct_IO,
.invalidatepage = btrfs_invalidatepage,
.releasepage = btrfs_releasepage,
.set_page_dirty = btrfs_set_page_dirty,
.error_remove_page = generic_error_remove_page,
};
static const struct address_space_operations btrfs_symlink_aops = {
.readpage = btrfs_readpage,
.writepage = btrfs_writepage,
.invalidatepage = btrfs_invalidatepage,
.releasepage = btrfs_releasepage,
};
static const struct inode_operations btrfs_file_inode_operations = {
.getattr = btrfs_getattr,
.setattr = btrfs_setattr,
.setxattr = btrfs_setxattr,
.getxattr = btrfs_getxattr,
.listxattr = btrfs_listxattr,
.removexattr = btrfs_removexattr,
.permission = btrfs_permission,
.fiemap = btrfs_fiemap,
.get_acl = btrfs_get_acl,
};
static const struct inode_operations btrfs_special_inode_operations = {
.getattr = btrfs_getattr,
.setattr = btrfs_setattr,
.permission = btrfs_permission,
.setxattr = btrfs_setxattr,
.getxattr = btrfs_getxattr,
.listxattr = btrfs_listxattr,
.removexattr = btrfs_removexattr,
.get_acl = btrfs_get_acl,
};
static const struct inode_operations btrfs_symlink_inode_operations = {
.readlink = generic_readlink,
.follow_link = page_follow_link_light,
.put_link = page_put_link,
.getattr = btrfs_getattr,
.setattr = btrfs_setattr,
.permission = btrfs_permission,
.setxattr = btrfs_setxattr,
.getxattr = btrfs_getxattr,
.listxattr = btrfs_listxattr,
.removexattr = btrfs_removexattr,
.get_acl = btrfs_get_acl,
};
const struct dentry_operations btrfs_dentry_operations = {
.d_delete = btrfs_dentry_delete,
.d_release = btrfs_dentry_release,
};