Make the PROC_I() and PDE() macros internal to procfs. This means making
PDE_DATA() out of line. This could be made more optimal by storing
PDE()->data into inode->i_private.
Also provide a __PDE_DATA() that is inline and internal to procfs.
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Supply an accessor function for getting the private data from the parent
proc_dir_entry struct of the proc_dir_entry struct associated with an inode.
ReiserFS, for instance, stores the super_block pointer in the proc directory
it makes for that super_block, and a pointer to the respective seq_file show
function in each of the proc files in that directory.
This allows a reduction in the number of file_operations structs, open
functions and seq_operations structs required. The problem otherwise is that
each show function requires two pieces of data but only has storage for one
per PDE (and this has no release function).
Signed-off-by: David Howells <dhowells@redhat.com>
Acked-by: Mauro Carvalho Chehab <mchehab@redhat.com>
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
cc: Jerry Chuang <jerry-chuang@realtek.com>
cc: Maxim Mikityanskiy <maxtram95@gmail.com>
cc: YAMANE Toshiaki <yamanetoshi@gmail.com>
cc: linux-wireless@vger.kernel.org
cc: linux-scsi@vger.kernel.org
cc: devel@driverdev.osuosl.org
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Add proc_mkdir_data() to allow procfs directories to be created that are
annotated at the time of creation with private data rather than doing this
post-creation. This means no access is then required to the proc_dir_entry
struct to set this.
Signed-off-by: David Howells <dhowells@redhat.com>
Acked-by: Mauro Carvalho Chehab <mchehab@redhat.com>
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
cc: Neela Syam Kolli <megaraidlinux@lsi.com>
cc: Jerry Chuang <jerry-chuang@realtek.com>
cc: linux-scsi@vger.kernel.org
cc: devel@driverdev.osuosl.org
cc: linux-wireless@vger.kernel.org
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Delete create_proc_read_entry() as it no longer has any users.
Also delete read_proc_t, write_proc_t, the read_proc member of the
proc_dir_entry struct and the support functions that use them. This saves a
pointer for every PDE allocated.
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Switch huge if-statement in __proc_file_read() around. This then puts the
single line loop break immediately after the if-statement and allows us to
de-indent the huge comment and make it take fewer lines. The code following
the if-statement then follows naturally from the call to dp->read_proc().
Signed-off-by: David Howells <dhowells@redhat.com>
Kill create_proc_entry() in favour of create_proc_read_entry(), proc_create()
and proc_create_data().
Signed-off-by: David Howells <dhowells@redhat.com>
The only part of proc_dir_entry the code outside of fs/proc
really cares about is PDE(inode)->data. Provide a helper
for that; static inline for now, eventually will be moved
to fs/proc, along with the knowledge of struct proc_dir_entry
layout.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
just what it sounds like; do that only to procfs subtrees you've
created - doing that to something shared with another driver is
not only antisocial, but might cause interesting races with
proc_create() and its ilk.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
- use pr_foo() throughout
- remove a couple of duplicated KERN_WARNINGs, via WARN(KERN_WARNING "...")
- nuke a few warnings which I've never seen happen, ever.
Cc: Joe Perches <joe@perches.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Make it drop the pde in *all* cases when no new reference to it is
put into an inode - both when an inode had already been set up
(as we were already doing) and when inode allocation has failed.
Makes for simpler logics in callers...
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Merge the rest of Andrew's patches for -rc1:
"A bunch of fixes and misc missed-out-on things.
That'll do for -rc1. I still have a batch of IPC patches which still
have a possible bug report which I'm chasing down."
* emailed patches from Andrew Morton <akpm@linux-foundation.org>: (25 commits)
keys: use keyring_alloc() to create module signing keyring
keys: fix unreachable code
sendfile: allows bypassing of notifier events
SGI-XP: handle non-fatal traps
fat: fix incorrect function comment
Documentation: ABI: remove testing/sysfs-devices-node
proc: fix inconsistent lock state
linux/kernel.h: fix DIV_ROUND_CLOSEST with unsigned divisors
memcg: don't register hotcpu notifier from ->css_alloc()
checkpatch: warn on uapi #includes that #include <uapi/...
revert "rtc: recycle id when unloading a rtc driver"
mm: clean up transparent hugepage sysfs error messages
hfsplus: add error message for the case of failure of sync fs in delayed_sync_fs() method
hfsplus: rework processing of hfs_btree_write() returned error
hfsplus: rework processing errors in hfsplus_free_extents()
hfsplus: avoid crash on failed block map free
kcmp: include linux/ptrace.h
drivers/rtc/rtc-imxdi.c: must include <linux/spinlock.h>
mm: cma: WARN if freed memory is still in use
exec: do not leave bprm->interp on stack
...
Lockdep found an inconsistent lock state when rcu is processing delayed
work in softirq. Currently, kernel is using spin_lock/spin_unlock to
protect proc_inum_ida, but proc_free_inum is called by rcu in softirq
context.
Use spin_lock_bh/spin_unlock_bh fix following lockdep warning.
=================================
[ INFO: inconsistent lock state ]
3.7.0 #36 Not tainted
---------------------------------
inconsistent {SOFTIRQ-ON-W} -> {IN-SOFTIRQ-W} usage.
swapper/1/0 [HC0[0]:SC1[1]:HE1:SE0] takes:
(proc_inum_lock){+.?...}, at: proc_free_inum+0x1c/0x50
{SOFTIRQ-ON-W} state was registered at:
__lock_acquire+0x8ae/0xca0
lock_acquire+0x199/0x200
_raw_spin_lock+0x41/0x50
proc_alloc_inum+0x4c/0xd0
alloc_mnt_ns+0x49/0xc0
create_mnt_ns+0x25/0x70
mnt_init+0x161/0x1c7
vfs_caches_init+0x107/0x11a
start_kernel+0x348/0x38c
x86_64_start_reservations+0x131/0x136
x86_64_start_kernel+0x103/0x112
irq event stamp: 2993422
hardirqs last enabled at (2993422): _raw_spin_unlock_irqrestore+0x55/0x80
hardirqs last disabled at (2993421): _raw_spin_lock_irqsave+0x29/0x70
softirqs last enabled at (2993394): _local_bh_enable+0x13/0x20
softirqs last disabled at (2993395): call_softirq+0x1c/0x30
other info that might help us debug this:
Possible unsafe locking scenario:
CPU0
----
lock(proc_inum_lock);
<Interrupt>
lock(proc_inum_lock);
*** DEADLOCK ***
no locks held by swapper/1/0.
stack backtrace:
Pid: 0, comm: swapper/1 Not tainted 3.7.0 #36
Call Trace:
<IRQ> [<ffffffff810a40f1>] ? vprintk_emit+0x471/0x510
print_usage_bug+0x2a5/0x2c0
mark_lock+0x33b/0x5e0
__lock_acquire+0x813/0xca0
lock_acquire+0x199/0x200
_raw_spin_lock+0x41/0x50
proc_free_inum+0x1c/0x50
free_pid_ns+0x1c/0x50
put_pid_ns+0x2e/0x50
put_pid+0x4a/0x60
delayed_put_pid+0x12/0x20
rcu_process_callbacks+0x462/0x790
__do_softirq+0x1b4/0x3b0
call_softirq+0x1c/0x30
do_softirq+0x59/0xd0
irq_exit+0x54/0xd0
smp_apic_timer_interrupt+0x95/0xa3
apic_timer_interrupt+0x72/0x80
cpuidle_enter_tk+0x10/0x20
cpuidle_enter_state+0x17/0x50
cpuidle_idle_call+0x287/0x520
cpu_idle+0xba/0x130
start_secondary+0x2b3/0x2bc
Signed-off-by: Xiaotian Feng <dannyfeng@tencent.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Generalize the proc inode allocation so that it can be
used without having to having to create a proc_dir_entry.
This will allow namespace file descriptors to remain light
weight entitities but still have the same inode number
when the backing namespace is the same.
Acked-by: Serge E. Hallyn <serge.hallyn@ubuntu.com>
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Part of the memory will be written twice after this change, but that
should be negligible.
[akpm@linux-foundation.org: fix __proc_create() coding-style issues, remove unneeded zero-initialisations]
Signed-off-by: yan <clouds.yan@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
If proc_get_inode() returns NULL then presumably it encountered memory
exhaustion. proc_lookup_de() should return -ENOMEM in this case, not
-EINVAL.
Signed-off-by: yan <clouds.yan@gmail.com>
Cc: Ryan Mallon <rmallon@gmail.com>
Cc: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Just the flags; only NFS cares even about that, but there are
legitimate uses for such argument. And getting rid of that
completely would require splitting ->lookup() into a couple
of methods (at least), so let's leave that alone for now...
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Replace remaining direct i_nlink updates with a new set_nlink()
updater function.
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Tested-by: Toshiyuki Okajima <toshi.okajima@jp.fujitsu.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Since __proc_create() appends the name it is given to the end of the PDE
structure that it allocates, there isn't a need to store a name pointer.
Instead we can just replace the name pointer with a terminal char array of
_unspecified_ length. The compiler will simply append the string to statically
defined variables of PDE type overlapping any hole at the end of the structure
and, unlike specifying an explicitly _zero_ length array, won't give a warning
if you try to statically initialise it with a string of more than zero length.
Also, whilst we're at it:
(1) Move namelen to end just prior to name and reduce it to a single byte
(name shouldn't be longer than NAME_MAX).
(2) Move pde_unload_lock two places further on so that if it's four bytes in
size on a 64-bit machine, it won't cause an unused hole in the PDE struct.
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
* use proc_mkdir_mode() instead of create_proc_entry(S_IFDIR|...),
export proc_mkdir_mode() for that, oh well.
* don't supply S_IFREG to proc_create_data(), it's unnecessary
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
1. namelen is declared "unsigned short" which hints for "maybe space savings".
Indeed in 2.4 struct proc_dir_entry looked like:
struct proc_dir_entry {
unsigned short low_ino;
unsigned short namelen;
Now, low_ino is "unsigned int", all savings were gone for a long time.
"struct proc_dir_entry" is not that countless to worry about it's size,
anyway.
2. converting from unsigned short to int/unsigned int can only create
problems, we better play it safe.
Space is not really conserved, because of natural alignment for the next
field. sizeof(struct proc_dir_entry) remains the same.
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
For the common case where a proc entry is being removed and nobody is in
the process of using it, save a LOCK/UNLOCK pair.
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
- ->low_ino is write-once field -- reading it under locks is unnecessary.
- /proc/$PID stuff never reaches pde_put()/free_proc_entry() --
PROC_DYNAMIC_FIRST check never triggers.
- in proc_get_inode(), inode number always matches proc dir entry, so
save one parameter.
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Reduce some branches and memory accesses in dcache lookup by adding dentry
flags to indicate common d_ops are set, rather than having to check them.
This saves a pointer memory access (dentry->d_op) in common path lookup
situations, and saves another pointer load and branch in cases where we
have d_op but not the particular operation.
Patched with:
git grep -E '[.>]([[:space:]])*d_op([[:space:]])*=' | xargs sed -e 's/\([^\t ]*\)->d_op = \(.*\);/d_set_d_op(\1, \2);/' -e 's/\([^\t ]*\)\.d_op = \(.*\);/d_set_d_op(\&\1, \2);/' -i
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Change d_delete from a dentry deletion notification to a dentry caching
advise, more like ->drop_inode. Require it to be constant and idempotent,
and not take d_lock. This is how all existing filesystems use the callback
anyway.
This makes fine grained dentry locking of dput and dentry lru scanning
much simpler.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Replace inode_setattr with opencoded variants of it in all callers. This
moves the remaining call to vmtruncate into the filesystem methods where it
can be replaced with the proper truncate sequence.
In a few cases it was obvious that we would never end up calling vmtruncate
so it was left out in the opencoded variant:
spufs: explicitly checks for ATTR_SIZE earlier
btrfs,hugetlbfs,logfs,dlmfs: explicitly clears ATTR_SIZE earlier
ufs: contains an opencoded simple_seattr + truncate that sets the filesize just above
In addition to that ncpfs called inode_setattr with handcrafted iattrs,
which allowed to trim down the opencoded variant.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
A quick test shows these comments are obsolete, so just remove them.
Signed-off-by: WANG Cong <amwang@redhat.com>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
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>
* warn if creation goes on to non-existent directory
* warn if removal goes on from non-existing directory
* warn if non-existing proc entry is removed
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
remove_proc_entry() does
lock
lookup parent
unlock
lock
unlink proc entry from lists
unlock
which can be made bit more correct by doing parent translation + unlink
without dropping lock.
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
EXPORT_SYMBOL(proc_symlink);
EXPORT_SYMBOL(proc_mkdir);
EXPORT_SYMBOL(create_proc_entry);
EXPORT_SYMBOL(proc_create_data);
EXPORT_SYMBOL(remove_proc_entry);
Those EXPORT_SYMBOL shouldn't be in fs/proc/root.c,
should be in fs/proc/generic.c.
Signed-off-by: Helight.Xu <helight.xu@gmail.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
* de_get() is trivial -- make inline, save a few bits of code, drop
"refcount is 0" check -- it should be done in some generic refcount
code, don't recall it's was helpful
* rename GET and PUT functions to pde_get(), pde_put() for cool prefix!
* remove obvious and incorrent comments
* in remove_proc_entry() use pde_put(), when I fixed PDE refcounting to
be normal one, remove_proc_entry() was supposed to do "-1" and code now
reflects that.
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
struct proc_dir_entry::owner is going to be removed. Now it's only necessary
to protect PDEs which are using ->read_proc, ->write_proc hooks.
However, ->owner assignments are racy and make it very easy for someone to switch
->owner on live PDE (as some subsystems do) without fixing refcounts and so on.
http://bugzilla.kernel.org/show_bug.cgi?id=12454
So, ->owner is on death row.
Proxy file operations exist already (proc_file_operations), just bump usecount
when necessary.
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
so that people will realize that it exists and can update it as needed.
Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
There are four BKL users in proc: de_put(), proc_lookup_de(),
proc_readdir_de(), proc_root_readdir(),
1) de_put()
-----------
de_put() is classic atomic_dec_and_test() refcount wrapper -- no BKL
needed. BKL doesn't matter to possible refcount leak as well.
2) proc_lookup_de()
-------------------
Walking PDE list is protected by proc_subdir_lock(), proc_get_inode() is
potentially blocking, all callers of proc_lookup_de() eventually end up
from ->lookup hooks which is protected by directory's ->i_mutex -- BKL
doesn't protect anything.
3) proc_readdir_de()
--------------------
"." and ".." part doesn't need BKL, walking PDE list is under
proc_subdir_lock, calling filldir callback is potentially blocking
because it writes to luserspace. All proc_readdir_de() callers
eventually come from ->readdir hook which is under directory's
->i_mutex -- BKL doesn't protect anything.
4) proc_root_readdir_de()
-------------------------
proc_root_readdir_de is ->readdir hook, see (3).
Since readdir hooks doesn't use BKL anymore, switch to
generic_file_llseek, since it also takes directory's i_mutex.
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Use WARN() rather than a printk() + backtrace();
this gives a more standard format message as well as complete
information (including line numbers etc) that will be collected
by kerneloops.org
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Print parent directory name as well.
The aim is to catch non-creation of parent directory when proc_mkdir will
return NULL and all subsequent registrations go directly in /proc instead
of intended directory.
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
[ Fixed insane printk string while at it. - Linus ]
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Ouch, if number taken from IDA is too big, the intent was to signal an
error, not check for overflow and still do overflowing addition.
One still needs 2^28 proc entries to notice this.
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
proc doesn't use "associate pointer with id" feature of IDR, so switch
to IDA.
NOTE, NOTE, NOTE:
Do not apply if release_inode_number() still mantions MAX_ID_MASK!
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Id which proc gets from IDR for inode number and id which proc removes
from IDR do not match. E.g. 0x11a transforms into 0x8000011a.
Which stayed unnoticed for a long time because, surprise, idr_remove()
masks out that high bit before doing anything.
All of this due to "| ~MAX_ID_MASK" in release_inode_number().
I still don't understand how it's supposed to work, because "| ~MASK"
is not an inversion for "& MAX" operation.
So, use just one nice, working addition. Make start offset unsigned int,
while I'm at it. It's longness is not used anywhere.
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
