linux/fs/nilfs2/segment.h

250 lines
7.8 KiB
C
Raw Normal View History

/* SPDX-License-Identifier: GPL-2.0+ */
/*
* NILFS Segment constructor prototypes and definitions
*
* Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
*
* Written by Ryusuke Konishi.
*
*/
#ifndef _NILFS_SEGMENT_H
#define _NILFS_SEGMENT_H
#include <linux/types.h>
#include <linux/fs.h>
#include <linux/buffer_head.h>
nilfs2: fix deadlock of segment constructor over I_SYNC flag Nilfs2 eventually hangs in a stress test with fsstress program. This issue was caused by the following deadlock over I_SYNC flag between nilfs_segctor_thread() and writeback_sb_inodes(): nilfs_segctor_thread() nilfs_segctor_thread_construct() nilfs_segctor_unlock() nilfs_dispose_list() iput() iput_final() evict() inode_wait_for_writeback() * wait for I_SYNC flag writeback_sb_inodes() * set I_SYNC flag on inode->i_state __writeback_single_inode() do_writepages() nilfs_writepages() nilfs_construct_dsync_segment() nilfs_segctor_sync() * wait for completion of segment constructor inode_sync_complete() * clear I_SYNC flag after __writeback_single_inode() completed writeback_sb_inodes() calls do_writepages() for dirty inodes after setting I_SYNC flag on inode->i_state. do_writepages() in turn calls nilfs_writepages(), which can run segment constructor and wait for its completion. On the other hand, segment constructor calls iput(), which can call evict() and wait for the I_SYNC flag on inode_wait_for_writeback(). Since segment constructor doesn't know when I_SYNC will be set, it cannot know whether iput() will block or not unless inode->i_nlink has a non-zero count. We can prevent evict() from being called in iput() by implementing sop->drop_inode(), but it's not preferable to leave inodes with i_nlink == 0 for long periods because it even defers file truncation and inode deallocation. So, this instead resolves the deadlock by calling iput() asynchronously with a workqueue for inodes with i_nlink == 0. Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp> Cc: Al Viro <viro@zeniv.linux.org.uk> Tested-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-02-05 20:25:20 +00:00
#include <linux/workqueue.h>
#include "nilfs.h"
struct nilfs_root;
/**
* struct nilfs_recovery_info - Recovery information
* @ri_need_recovery: Recovery status
* @ri_super_root: Block number of the last super root
* @ri_ri_cno: Number of the last checkpoint
* @ri_lsegs_start: Region for roll-forwarding (start block number)
* @ri_lsegs_end: Region for roll-forwarding (end block number)
* @ri_lseg_start_seq: Sequence value of the segment at ri_lsegs_start
* @ri_used_segments: List of segments to be mark active
* @ri_pseg_start: Block number of the last partial segment
* @ri_seq: Sequence number on the last partial segment
* @ri_segnum: Segment number on the last partial segment
* @ri_nextnum: Next segment number on the last partial segment
*/
struct nilfs_recovery_info {
int ri_need_recovery;
sector_t ri_super_root;
__u64 ri_cno;
sector_t ri_lsegs_start;
sector_t ri_lsegs_end;
u64 ri_lsegs_start_seq;
struct list_head ri_used_segments;
sector_t ri_pseg_start;
u64 ri_seq;
__u64 ri_segnum;
__u64 ri_nextnum;
};
/* ri_need_recovery */
#define NILFS_RECOVERY_SR_UPDATED 1 /* The super root was updated */
#define NILFS_RECOVERY_ROLLFORWARD_DONE 2 /* Rollforward was carried out */
/**
* struct nilfs_cstage - Context of collection stage
nilfs2: add a tracepoint for tracking stage transition of segment construction This patch adds a tracepoint for tracking stage transition of block collection in segment construction. With the tracepoint, we can analysis the behavior of segment construction in depth. It would be useful for bottleneck detection and debugging, etc. The tracepoint is created with the standard trace API of linux (like ext3, ext4, f2fs and btrfs). So we can analysis with existing tools easily. Of course, more detailed analysis will be possible if we can create nilfs specific analysis tools. Below is an example of event dump with Brendan Gregg's perf-tools (https://github.com/brendangregg/perf-tools). Time consumption between each stage can be obtained. $ sudo bin/tpoint nilfs2:nilfs2_collection_stage_transition Tracing nilfs2:nilfs2_collection_stage_transition. Ctrl-C to end. segctord-14875 [003] ...1 28311.067794: nilfs2_collection_stage_transition: sci = ffff8800ce6de000 stage = ST_INIT segctord-14875 [003] ...1 28311.068139: nilfs2_collection_stage_transition: sci = ffff8800ce6de000 stage = ST_GC segctord-14875 [003] ...1 28311.068139: nilfs2_collection_stage_transition: sci = ffff8800ce6de000 stage = ST_FILE segctord-14875 [003] ...1 28311.068486: nilfs2_collection_stage_transition: sci = ffff8800ce6de000 stage = ST_IFILE segctord-14875 [003] ...1 28311.068540: nilfs2_collection_stage_transition: sci = ffff8800ce6de000 stage = ST_CPFILE segctord-14875 [003] ...1 28311.068561: nilfs2_collection_stage_transition: sci = ffff8800ce6de000 stage = ST_SUFILE segctord-14875 [003] ...1 28311.068565: nilfs2_collection_stage_transition: sci = ffff8800ce6de000 stage = ST_DAT segctord-14875 [003] ...1 28311.068573: nilfs2_collection_stage_transition: sci = ffff8800ce6de000 stage = ST_SR segctord-14875 [003] ...1 28311.068574: nilfs2_collection_stage_transition: sci = ffff8800ce6de000 stage = ST_DONE For capturing transition correctly, this patch adds wrappers for the member scnt of nilfs_cstage. With this change, every transition of the stage can produce trace event in a correct manner. Signed-off-by: Hitoshi Mitake <mitake.hitoshi@lab.ntt.co.jp> Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp> Cc: Steven Rostedt <rostedt@goodmis.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-11-07 00:31:59 +00:00
* @scnt: Stage count, must be accessed via wrappers:
* nilfs_sc_cstage_inc(), nilfs_sc_cstage_set(), nilfs_sc_cstage_get()
* @flags: State flags
* @dirty_file_ptr: Pointer on dirty_files list, or inode of a target file
* @gc_inode_ptr: Pointer on the list of gc-inodes
*/
struct nilfs_cstage {
int scnt;
unsigned int flags;
struct nilfs_inode_info *dirty_file_ptr;
struct nilfs_inode_info *gc_inode_ptr;
};
struct nilfs_segment_buffer;
struct nilfs_segsum_pointer {
struct buffer_head *bh;
unsigned int offset; /* offset in bytes */
};
/**
* struct nilfs_sc_info - Segment constructor information
* @sc_super: Back pointer to super_block struct
* @sc_root: root object of the current filesystem tree
* @sc_nblk_inc: Block count of current generation
* @sc_dirty_files: List of files to be written
* @sc_gc_inodes: List of GC inodes having blocks to be written
nilfs2: fix deadlock of segment constructor over I_SYNC flag Nilfs2 eventually hangs in a stress test with fsstress program. This issue was caused by the following deadlock over I_SYNC flag between nilfs_segctor_thread() and writeback_sb_inodes(): nilfs_segctor_thread() nilfs_segctor_thread_construct() nilfs_segctor_unlock() nilfs_dispose_list() iput() iput_final() evict() inode_wait_for_writeback() * wait for I_SYNC flag writeback_sb_inodes() * set I_SYNC flag on inode->i_state __writeback_single_inode() do_writepages() nilfs_writepages() nilfs_construct_dsync_segment() nilfs_segctor_sync() * wait for completion of segment constructor inode_sync_complete() * clear I_SYNC flag after __writeback_single_inode() completed writeback_sb_inodes() calls do_writepages() for dirty inodes after setting I_SYNC flag on inode->i_state. do_writepages() in turn calls nilfs_writepages(), which can run segment constructor and wait for its completion. On the other hand, segment constructor calls iput(), which can call evict() and wait for the I_SYNC flag on inode_wait_for_writeback(). Since segment constructor doesn't know when I_SYNC will be set, it cannot know whether iput() will block or not unless inode->i_nlink has a non-zero count. We can prevent evict() from being called in iput() by implementing sop->drop_inode(), but it's not preferable to leave inodes with i_nlink == 0 for long periods because it even defers file truncation and inode deallocation. So, this instead resolves the deadlock by calling iput() asynchronously with a workqueue for inodes with i_nlink == 0. Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp> Cc: Al Viro <viro@zeniv.linux.org.uk> Tested-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-02-05 20:25:20 +00:00
* @sc_iput_queue: list of inodes for which iput should be done
* @sc_iput_work: work struct to defer iput call
* @sc_freesegs: array of segment numbers to be freed
* @sc_nfreesegs: number of segments on @sc_freesegs
* @sc_dsync_inode: inode whose data pages are written for a sync operation
* @sc_dsync_start: start byte offset of data pages
* @sc_dsync_end: end byte offset of data pages (inclusive)
* @sc_segbufs: List of segment buffers
* @sc_write_logs: List of segment buffers to hold logs under writing
* @sc_segbuf_nblocks: Number of available blocks in segment buffers.
* @sc_curseg: Current segment buffer
* @sc_stage: Collection stage
* @sc_finfo_ptr: pointer to the current finfo struct in the segment summary
* @sc_binfo_ptr: pointer to the current binfo struct in the segment summary
* @sc_blk_cnt: Block count of a file
* @sc_datablk_cnt: Data block count of a file
* @sc_nblk_this_inc: Number of blocks included in the current logical segment
* @sc_seg_ctime: Creation time
* @sc_cno: checkpoint number of current log
* @sc_flags: Internal flags
* @sc_state_lock: spinlock for sc_state and so on
* @sc_state: Segctord state flags
* @sc_flush_request: inode bitmap of metadata files to be flushed
* @sc_wait_request: Client request queue
* @sc_wait_daemon: Daemon wait queue
* @sc_wait_task: Start/end wait queue to control segctord task
* @sc_seq_request: Request counter
* @sc_seq_accept: Accepted request count
* @sc_seq_done: Completion counter
* @sc_sync: Request of explicit sync operation
* @sc_interval: Timeout value of background construction
* @sc_mjcp_freq: Frequency of creating checkpoints
* @sc_lseg_stime: Start time of the latest logical segment
* @sc_watermark: Watermark for the number of dirty buffers
* @sc_timer: Timer for segctord
* @sc_task: current thread of segctord
*/
struct nilfs_sc_info {
struct super_block *sc_super;
struct nilfs_root *sc_root;
unsigned long sc_nblk_inc;
struct list_head sc_dirty_files;
struct list_head sc_gc_inodes;
nilfs2: fix deadlock of segment constructor over I_SYNC flag Nilfs2 eventually hangs in a stress test with fsstress program. This issue was caused by the following deadlock over I_SYNC flag between nilfs_segctor_thread() and writeback_sb_inodes(): nilfs_segctor_thread() nilfs_segctor_thread_construct() nilfs_segctor_unlock() nilfs_dispose_list() iput() iput_final() evict() inode_wait_for_writeback() * wait for I_SYNC flag writeback_sb_inodes() * set I_SYNC flag on inode->i_state __writeback_single_inode() do_writepages() nilfs_writepages() nilfs_construct_dsync_segment() nilfs_segctor_sync() * wait for completion of segment constructor inode_sync_complete() * clear I_SYNC flag after __writeback_single_inode() completed writeback_sb_inodes() calls do_writepages() for dirty inodes after setting I_SYNC flag on inode->i_state. do_writepages() in turn calls nilfs_writepages(), which can run segment constructor and wait for its completion. On the other hand, segment constructor calls iput(), which can call evict() and wait for the I_SYNC flag on inode_wait_for_writeback(). Since segment constructor doesn't know when I_SYNC will be set, it cannot know whether iput() will block or not unless inode->i_nlink has a non-zero count. We can prevent evict() from being called in iput() by implementing sop->drop_inode(), but it's not preferable to leave inodes with i_nlink == 0 for long periods because it even defers file truncation and inode deallocation. So, this instead resolves the deadlock by calling iput() asynchronously with a workqueue for inodes with i_nlink == 0. Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp> Cc: Al Viro <viro@zeniv.linux.org.uk> Tested-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-02-05 20:25:20 +00:00
struct list_head sc_iput_queue;
struct work_struct sc_iput_work;
__u64 *sc_freesegs;
size_t sc_nfreesegs;
struct nilfs_inode_info *sc_dsync_inode;
loff_t sc_dsync_start;
loff_t sc_dsync_end;
/* Segment buffers */
struct list_head sc_segbufs;
struct list_head sc_write_logs;
unsigned long sc_segbuf_nblocks;
struct nilfs_segment_buffer *sc_curseg;
struct nilfs_cstage sc_stage;
struct nilfs_segsum_pointer sc_finfo_ptr;
struct nilfs_segsum_pointer sc_binfo_ptr;
unsigned long sc_blk_cnt;
unsigned long sc_datablk_cnt;
unsigned long sc_nblk_this_inc;
time64_t sc_seg_ctime;
__u64 sc_cno;
unsigned long sc_flags;
spinlock_t sc_state_lock;
unsigned long sc_state;
unsigned long sc_flush_request;
wait_queue_head_t sc_wait_request;
wait_queue_head_t sc_wait_daemon;
wait_queue_head_t sc_wait_task;
__u32 sc_seq_request;
__u32 sc_seq_accepted;
__u32 sc_seq_done;
int sc_sync;
unsigned long sc_interval;
unsigned long sc_mjcp_freq;
unsigned long sc_lseg_stime; /* in 1/HZ seconds */
unsigned long sc_watermark;
struct timer_list sc_timer;
struct task_struct *sc_timer_task;
struct task_struct *sc_task;
};
/* sc_flags */
enum {
NILFS_SC_DIRTY, /* One or more dirty meta-data blocks exist */
NILFS_SC_UNCLOSED, /* Logical segment is not closed */
NILFS_SC_SUPER_ROOT, /* The latest segment has a super root */
NILFS_SC_PRIOR_FLUSH, /*
* Requesting immediate flush without making a
* checkpoint
*/
NILFS_SC_HAVE_DELTA, /*
* Next checkpoint will have update of files
* other than DAT, cpfile, sufile, or files
* moved by GC.
*/
};
/* sc_state */
#define NILFS_SEGCTOR_QUIT 0x0001 /* segctord is being destroyed */
#define NILFS_SEGCTOR_COMMIT 0x0004 /* committed transaction exists */
/*
* Constant parameters
*/
#define NILFS_SC_CLEANUP_RETRY 3 /*
* Retry count of construction when
* destroying segctord
*/
/*
* Default values of timeout, in seconds.
*/
#define NILFS_SC_DEFAULT_TIMEOUT 5 /*
* Timeout value of dirty blocks.
* It triggers construction of a
* logical segment with a super root.
*/
#define NILFS_SC_DEFAULT_SR_FREQ 30 /*
* Maximum frequency of super root
* creation
*/
/*
* The default threshold amount of data, in block counts.
*/
#define NILFS_SC_DEFAULT_WATERMARK 3600
/* super.c */
extern struct kmem_cache *nilfs_transaction_cachep;
/* segment.c */
extern void nilfs_relax_pressure_in_lock(struct super_block *);
extern int nilfs_construct_segment(struct super_block *);
extern int nilfs_construct_dsync_segment(struct super_block *, struct inode *,
loff_t, loff_t);
extern void nilfs_flush_segment(struct super_block *, ino_t);
nilfs2: fix lock order reversal in nilfs_clean_segments ioctl This is a companion patch to ("nilfs2: fix possible circular locking for get information ioctls"). This corrects lock order reversal between mm->mmap_sem and nilfs->ns_segctor_sem in nilfs_clean_segments() which was detected by lockdep check: ======================================================= [ INFO: possible circular locking dependency detected ] 2.6.30-rc3-nilfs-00003-g360bdc1 #7 ------------------------------------------------------- mmap/5294 is trying to acquire lock: (&nilfs->ns_segctor_sem){++++.+}, at: [<d0d0e846>] nilfs_transaction_begin+0xb6/0x10c [nilfs2] but task is already holding lock: (&mm->mmap_sem){++++++}, at: [<c043700a>] do_page_fault+0x1d8/0x30a which lock already depends on the new lock. the existing dependency chain (in reverse order) is: -> #1 (&mm->mmap_sem){++++++}: [<c01470a5>] __lock_acquire+0x1066/0x13b0 [<c01474a9>] lock_acquire+0xba/0xdd [<c01836bc>] might_fault+0x68/0x88 [<c023c61d>] copy_from_user+0x2a/0x111 [<d0d120d0>] nilfs_ioctl_prepare_clean_segments+0x1d/0xf1 [nilfs2] [<d0d0e2aa>] nilfs_clean_segments+0x6d/0x1b9 [nilfs2] [<d0d11f68>] nilfs_ioctl+0x2ad/0x318 [nilfs2] [<c01a3be7>] vfs_ioctl+0x22/0x69 [<c01a408e>] do_vfs_ioctl+0x460/0x499 [<c01a4107>] sys_ioctl+0x40/0x5a [<c01031a4>] sysenter_do_call+0x12/0x38 [<ffffffff>] 0xffffffff -> #0 (&nilfs->ns_segctor_sem){++++.+}: [<c0146e0b>] __lock_acquire+0xdcc/0x13b0 [<c01474a9>] lock_acquire+0xba/0xdd [<c0433f1d>] down_read+0x2a/0x3e [<d0d0e846>] nilfs_transaction_begin+0xb6/0x10c [nilfs2] [<d0cfe0e5>] nilfs_page_mkwrite+0xe7/0x154 [nilfs2] [<c0183b0b>] __do_fault+0x165/0x376 [<c01855cd>] handle_mm_fault+0x287/0x5d1 [<c043712d>] do_page_fault+0x2fb/0x30a [<c0435462>] error_code+0x72/0x78 [<ffffffff>] 0xffffffff where nilfs_clean_segments() holds: nilfs->ns_segctor_sem -> copy_from_user() --> page fault -> mm->mmap_sem And, page fault path may hold: page fault -> mm->mmap_sem --> nilfs_page_mkwrite() -> nilfs->ns_segctor_sem Even though nilfs_clean_segments() does not perform write access on given user pages, it may cause deadlock because nilfs->ns_segctor_sem is shared per device and mm->mmap_sem can be shared with other tasks. To avoid this problem, this patch moves all calls of copy_from_user() outside the nilfs->ns_segctor_sem lock in the ioctl. Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
2009-05-10 13:41:43 +00:00
extern int nilfs_clean_segments(struct super_block *, struct nilfs_argv *,
void **);
int nilfs_attach_log_writer(struct super_block *sb, struct nilfs_root *root);
void nilfs_detach_log_writer(struct super_block *sb);
/* recovery.c */
extern int nilfs_read_super_root_block(struct the_nilfs *, sector_t,
struct buffer_head **, int);
extern int nilfs_search_super_root(struct the_nilfs *,
struct nilfs_recovery_info *);
int nilfs_salvage_orphan_logs(struct the_nilfs *nilfs, struct super_block *sb,
struct nilfs_recovery_info *ri);
extern void nilfs_dispose_segment_list(struct list_head *);
#endif /* _NILFS_SEGMENT_H */