10 years agokmemcheck: enable in the x86 Kconfig
Vegard Nossum [Thu, 3 Apr 2008 22:51:41 +0000]
kmemcheck: enable in the x86 Kconfig

let it rip!

Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

[rebased for mainline inclusion]
Signed-off-by: Vegard Nossum <vegardno@ifi.uio.no>

10 years agokmemcheck: add hooks for the page allocator
Vegard Nossum [Tue, 25 Nov 2008 15:55:53 +0000]
kmemcheck: add hooks for the page allocator

This adds support for tracking the initializedness of memory that
was allocated with the page allocator. Highmem requests are not
tracked.

Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Acked-by: Pekka Enberg <penberg@cs.helsinki.fi>

[build fix for !CONFIG_KMEMCHECK]
Signed-off-by: Ingo Molnar <mingo@elte.hu>

[rebased for mainline inclusion]
Signed-off-by: Vegard Nossum <vegard.nossum@gmail.com>

10 years agokmemcheck: add hooks for page- and sg-dma-mappings
Vegard Nossum [Sat, 21 Feb 2009 12:52:37 +0000]
kmemcheck: add hooks for page- and sg-dma-mappings

This is needed for page allocator support to prevent false positives
when accessing pages which are dma-mapped.

[rebased for mainline inclusion]
Signed-off-by: Vegard Nossum <vegard.nossum@gmail.com>

10 years agokmemcheck: don't track page tables
Vegard Nossum [Sun, 22 Feb 2009 10:28:25 +0000]
kmemcheck: don't track page tables

As these are allocated using the page allocator, we need to pass
__GFP_NOTRACK before we add page allocator support to kmemcheck.

Signed-off-by: Vegard Nossum <vegard.nossum@gmail.com>

10 years agocrypto: don't track xor test pages with kmemcheck
Vegard Nossum [Thu, 26 Feb 2009 08:55:52 +0000]
crypto: don't track xor test pages with kmemcheck

The xor tests are run on uninitialized data, because it is doesn't
really matter what the underlying data is. Annotate this false-
positive warning.

Acked-by: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Vegard Nossum <vegard.nossum@gmail.com>

10 years agoslab: add hooks for kmemcheck
Pekka Enberg [Fri, 9 May 2008 18:35:53 +0000]
slab: add hooks for kmemcheck

We now have SLAB support for kmemcheck! This means that it doesn't matter
whether one chooses SLAB or SLUB, or indeed whether Linus chooses to chuck
SLAB or SLUB.. ;-)

Cc: Ingo Molnar <mingo@elte.hu>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>

[rebased for mainline inclusion]
Signed-off-by: Vegard Nossum <vegard.nossum@gmail.com>

10 years agoslub: add hooks for kmemcheck
Vegard Nossum [Thu, 3 Apr 2008 22:54:48 +0000]
slub: add hooks for kmemcheck

Parts of this patch were contributed by Pekka Enberg but merged for
atomicity.

Cc: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Vegard Nossum <vegardno@ifi.uio.no>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

[rebased for mainline inclusion]
Signed-off-by: Vegard Nossum <vegardno@ifi.uio.no>

10 years agokmemcheck: add DMA hooks
Vegard Nossum [Sun, 20 Jul 2008 08:44:54 +0000]
kmemcheck: add DMA hooks

This patch hooks into the DMA API to prevent the reporting of the
false positives that would otherwise be reported when memory is
accessed that is also used directly by devices.

[rebased for mainline inclusion]
Signed-off-by: Vegard Nossum <vegard.nossum@gmail.com>

10 years agokmemcheck: add mm functions
Vegard Nossum [Sat, 31 May 2008 13:56:17 +0000]
kmemcheck: add mm functions

With kmemcheck enabled, the slab allocator needs to do this:

1. Tell kmemcheck to allocate the shadow memory which stores the status of
   each byte in the allocation proper, e.g. whether it is initialized or
   uninitialized.
2. Tell kmemcheck which parts of memory that should be marked uninitialized.
   There are actually a few more states, such as "not yet allocated" and
   "recently freed".

If a slab cache is set up using the SLAB_NOTRACK flag, it will never return
memory that can take page faults because of kmemcheck.

If a slab cache is NOT set up using the SLAB_NOTRACK flag, callers can still
request memory with the __GFP_NOTRACK flag. This does not prevent the page
faults from occuring, however, but marks the object in question as being
initialized so that no warnings will ever be produced for this object.

In addition to (and in contrast to) __GFP_NOTRACK, the
__GFP_NOTRACK_FALSE_POSITIVE flag indicates that the allocation should
not be tracked _because_ it would produce a false positive. Their values
are identical, but need not be so in the future (for example, we could now
enable/disable false positives with a config option).

Parts of this patch were contributed by Pekka Enberg but merged for
atomicity.

Signed-off-by: Vegard Nossum <vegard.nossum@gmail.com>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

[rebased for mainline inclusion]
Signed-off-by: Vegard Nossum <vegard.nossum@gmail.com>

10 years agox86: add hooks for kmemcheck
Vegard Nossum [Thu, 3 Apr 2008 22:53:23 +0000]
x86: add hooks for kmemcheck

The hooks that we modify are:
- Page fault handler (to handle kmemcheck faults)
- Debug exception handler (to hide pages after single-stepping
  the instruction that caused the page fault)

Also redefine memset() to use the optimized version if kmemcheck is
enabled.

(Thanks to Pekka Enberg for minimizing the impact on the page fault
handler.)

As kmemcheck doesn't handle MMX/SSE instructions (yet), we also disable
the optimized xor code, and rely instead on the generic C implementation
in order to avoid false-positive warnings.

Signed-off-by: Vegard Nossum <vegardno@ifi.uio.no>

[whitespace fixlet]
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

[rebased for mainline inclusion]
Signed-off-by: Vegard Nossum <vegardno@ifi.uio.no>

10 years agokmemcheck: use kmemcheck_pte_lookup() instead of open-coding it
Pekka Enberg [Thu, 26 Feb 2009 09:53:11 +0000]
kmemcheck: use kmemcheck_pte_lookup() instead of open-coding it

Lets use kmemcheck_pte_lookup() in kmemcheck_fault() instead of
open-coding it there.

Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Vegard Nossum <vegard.nossum@gmail.com>

10 years agokmemcheck: move 64-bit ifdef out of kmemcheck_opcode_decode()
Pekka Enberg [Thu, 26 Feb 2009 09:23:52 +0000]
kmemcheck: move 64-bit ifdef out of kmemcheck_opcode_decode()

This patch moves the CONFIG_X86_64 ifdef out of kmemcheck_opcode_decode() by
introducing a version of the function that always returns false for
CONFIG_X86_32.

Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Vegard Nossum <vegard.nossum@gmail.com>

10 years agokmemcheck: remove multiple ifdef'd definitions of the same global variable
Pekka Enberg [Thu, 26 Feb 2009 09:17:31 +0000]
kmemcheck: remove multiple ifdef'd definitions of the same global variable

Multiple ifdef'd definitions of the same global variable is ugly and
error-prone. Fix that up.

Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Vegard Nossum <vegard.nossum@gmail.com>

10 years agokmemcheck: make initialization message less confusing
Pekka Enberg [Thu, 26 Feb 2009 09:11:59 +0000]
kmemcheck: make initialization message less confusing

The "Bugs, beware!" printout during is cute but confuses users that something
bad happened so change the text to the more boring "Initialized" message.

Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Vegard Nossum <vegard.nossum@gmail.com>

10 years agokmemcheck: remove forward declarations from error.c
Pekka Enberg [Thu, 26 Feb 2009 08:47:50 +0000]
kmemcheck: remove forward declarations from error.c

This patch reorders code in error.c so that we can get rid of the forward
declarations.

Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Vegard Nossum <vegard.nossum@gmail.com>

10 years agokmemcheck: include module.h to prevent warnings
Randy Dunlap [Thu, 24 Jul 2008 23:09:32 +0000]
kmemcheck: include module.h to prevent warnings

kmemcheck/shadow.c needs to include <linux/module.h> to prevent
the following warnings:

linux-next-20080724/arch/x86/mm/kmemcheck/shadow.c:64: warning : data definition has no type or storage class
linux-next-20080724/arch/x86/mm/kmemcheck/shadow.c:64: warning : type defaults to 'int' in declaration of 'EXPORT_SYMBOL_GPL'
linux-next-20080724/arch/x86/mm/kmemcheck/shadow.c:64: warning : parameter names (without types) in function declaration

Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Cc: vegardno@ifi.uio.no
Cc: penberg@cs.helsinki.fi
Cc: akpm <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

10 years agokmemcheck: add the kmemcheck core
Vegard Nossum [Thu, 3 Apr 2008 22:51:41 +0000]
kmemcheck: add the kmemcheck core

General description: kmemcheck is a patch to the linux kernel that
detects use of uninitialized memory. It does this by trapping every
read and write to memory that was allocated dynamically (e.g. using
kmalloc()). If a memory address is read that has not previously been
written to, a message is printed to the kernel log.

Thanks to Andi Kleen for the set_memory_4k() solution.

Andrew Morton suggested documenting the shadow member of struct page.

Signed-off-by: Vegard Nossum <vegardno@ifi.uio.no>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>

[export kmemcheck_mark_initialized]
[build fix for setup_max_cpus]
Signed-off-by: Ingo Molnar <mingo@elte.hu>

[rebased for mainline inclusion]
Signed-off-by: Vegard Nossum <vegardno@ifi.uio.no>

10 years agokmemcheck: add the kmemcheck documentation
Vegard Nossum [Sat, 13 Jun 2009 12:15:57 +0000]
kmemcheck: add the kmemcheck documentation

Thanks to Sitsofe Wheeler, Randy Dunlap, and Jonathan Corbet for providing
input and feedback on this!

Signed-off-by: Vegard Nossum <vegard.nossum@gmail.com>

10 years agokmemcheck: add Vegard and Pekka to MAINTAINERS
Pekka Enberg [Mon, 28 Apr 2008 19:47:29 +0000]
kmemcheck: add Vegard and Pekka to MAINTAINERS

Acked-by: Vegard Nossum <vegardno@ifi.uio.no>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

10 years agotasklets: new tasklet scheduling function
Vegard Nossum [Wed, 21 May 2008 20:53:13 +0000]
tasklets: new tasklet scheduling function

Rationale: kmemcheck needs to be able to schedule a tasklet without
touching any dynamically allocated memory _at_ _all_ (since that would
lead to a recursive page fault). This tasklet is used for writing the
error reports to the kernel log.

The new scheduling function avoids touching any other tasklets by
inserting the new tasklist as the head of the "tasklet_hi" list instead
of on the tail.

Also don't wake up the softirq thread lest the scheduler access some
tracked memory and we go down with a recursive page fault.

In this case, we'd better just wait for the maximum time of 1/HZ for the
message to appear.

Signed-off-by: Vegard Nossum <vegard.nossum@gmail.com>

10 years agoslab: move struct kmem_cache to headers
Pekka Enberg [Fri, 9 May 2008 18:32:44 +0000]
slab: move struct kmem_cache to headers

Move the SLAB struct kmem_cache definition to <linux/slab_def.h> like
with SLUB so kmemcheck can access ->ctor and ->flags.

Cc: Ingo Molnar <mingo@elte.hu>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>

[rebased for mainline inclusion]
Signed-off-by: Vegard Nossum <vegard.nossum@gmail.com>

10 years agostacktrace: add forward-declaration struct task_struct
Vegard Nossum [Fri, 13 Jun 2008 13:31:11 +0000]
stacktrace: add forward-declaration struct task_struct

This is needed if the header is to be free-standing.

Signed-off-by: Vegard Nossum <vegard.nossum@gmail.com>

10 years agox86: add save_stack_trace_bp() for tracing from a specific stack frame
Vegard Nossum [Tue, 20 May 2008 09:15:43 +0000]
x86: add save_stack_trace_bp() for tracing from a specific stack frame

This will help kmemcheck (and possibly other debugging tools) since we
can now simply pass regs->bp to the stack tracer instead of specifying
the number of stack frames to skip, which is unreliable if gcc decides
to inline functions, etc.

Note that this makes the API incomplete for other architectures, but I
expect that those can be updated lazily, e.g. when they need it.

Cc: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Vegard Nossum <vegard.nossum@gmail.com>

10 years agoblock: fix kernel-doc in recent block/ changes
Randy Dunlap [Fri, 12 Jun 2009 03:00:41 +0000]
block: fix kernel-doc in recent block/ changes

Fix kernel-doc warnings in recently changed block/ source code.

Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

10 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs-2.6
Linus Torvalds [Fri, 12 Jun 2009 03:05:37 +0000]
Merge branch 'for-linus' of git://git./linux/kernel/git/viro/vfs-2.6

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs-2.6: (87 commits)
  nilfs2: get rid of bd_mount_sem use from nilfs
  nilfs2: correct exclusion control in nilfs_remount function
  nilfs2: simplify remaining sget() use
  nilfs2: get rid of sget use for checking if current mount is present
  nilfs2: get rid of sget use for acquiring nilfs object
  nilfs2: remove meaningless EBUSY case from nilfs_get_sb function
  remove the call to ->write_super in __sync_filesystem
  nilfs2: call nilfs2_write_super from nilfs2_sync_fs
  jffs2: call jffs2_write_super from jffs2_sync_fs
  ufs: add ->sync_fs
  sysv: add ->sync_fs
  hfsplus: add ->sync_fs
  hfs: add ->sync_fs
  fat: add ->sync_fs
  ext2: add ->sync_fs
  exofs: add ->sync_fs
  bfs: add ->sync_fs
  affs: add ->sync_fs
  sanitize ->fsync() for affs
  repair bfs_write_inode(), switch bfs to simple_fsync()
  ...

10 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/gerg/m68knommu
Linus Torvalds [Fri, 12 Jun 2009 03:05:08 +0000]
Merge branch 'for-linus' of git://git./linux/kernel/git/gerg/m68knommu

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/gerg/m68knommu:
  m68knommu: remove unecessary include of thread_info.h in entry.S
  m68knommu: enumerate INIT_THREAD fields properly
  headers_check fix: m68k, swab.h
  arch/m68knommu: Convert #ifdef DEBUG printk(KERN_DEBUG to pr_debug(
  m68knommu: remove obsolete reset code
  m68knommu: move CPU reset code for the 5272 ColdFire into its platform code
  m68knommu: move CPU reset code for the 528x ColdFire into its platform code
  m68knommu: move CPU reset code for the 527x ColdFire into its platform code
  m68knommu: move CPU reset code for the 523x ColdFire into its platform code
  m68knommu: move CPU reset code for the 520x ColdFire into its platform code
  m68knommu: add CPU reset code for the 532x ColdFire
  m68knommu: add CPU reset code for the 5249 ColdFire
  m68knommu: add CPU reset code for the 5206e ColdFire
  m68knommu: add CPU reset code for the 5206 ColdFire
  m68knommu: add CPU reset code for the 5407 ColdFire
  m68knommu: add CPU reset code for the 5307 ColdFire
  m68knommu: merge system reset for code ColdFire 523x family
  m68knommu: fix system reset for ColdFire 527x family

10 years agokvm: remove the duplicated cpumask_clear
Yinghai Lu [Thu, 11 Jun 2009 22:09:00 +0000]
kvm: remove the duplicated cpumask_clear

zalloc_cpumask_var already cleared it.

Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

10 years agox86: use zalloc_cpumask_var in arch_early_irq_init
Yinghai Lu [Thu, 11 Jun 2009 22:07:48 +0000]
x86: use zalloc_cpumask_var in arch_early_irq_init

So we make sure MAXSMP gets a cleared cpumask

Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

10 years agoperfcounters: remove powerpc definitions of perf_counter_do_pending
Stephen Rothwell [Fri, 12 Jun 2009 00:14:22 +0000]
perfcounters: remove powerpc definitions of perf_counter_do_pending

Commit 925d519ab82b6dd7aca9420d809ee83819c08db2 ("perf_counter:
unify and fix delayed counter wakeup") added global definitions.

Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Acked-by: Paul Mackerras <paulus@samba.org>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

10 years agonilfs2: get rid of bd_mount_sem use from nilfs
Ryusuke Konishi [Sun, 7 Jun 2009 16:39:33 +0000]
nilfs2: get rid of bd_mount_sem use from nilfs

This will remove every bd_mount_sem use in nilfs.

The intended exclusion control was replaced by the previous patch
("nilfs2: correct exclusion control in nilfs_remount function") for
nilfs_remount(), and this patch will replace remains with a new mutex
that this inserts in nilfs object.

Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Cc: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agonilfs2: correct exclusion control in nilfs_remount function
Ryusuke Konishi [Sun, 7 Jun 2009 16:39:32 +0000]
nilfs2: correct exclusion control in nilfs_remount function

nilfs_remount() changes mount state of a superblock instance.  Even
though nilfs accesses other superblock instances during mount or
remount, the mount state was not properly protected in
nilfs_remount().

Moreover, nilfs_remount() has a lock order reversal problem;
nilfs_get_sb() holds:

  1. bdev->bd_mount_sem
  2. sb->s_umount  (sget acquires)

and nilfs_remount() holds:

  1. sb->s_umount  (locked by the caller in vfs)
  2. bdev->bd_mount_sem

To avoid these problems, this patch divides a semaphore protecting
super block instances from nilfs->ns_sem, and applies it to the mount
state protection in nilfs_remount().

With this change, bd_mount_sem use is removed from nilfs_remount() and
the lock order reversal will be resolved.  And the new rw-semaphore,
nilfs->ns_super_sem will properly protect the mount state except the
modification from nilfs_error function.

Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agonilfs2: simplify remaining sget() use
Ryusuke Konishi [Sun, 7 Jun 2009 16:39:31 +0000]
nilfs2: simplify remaining sget() use

This simplifies the test function passed on the remaining sget()
callsite in nilfs.

Instead of checking mount type (i.e. ro-mount/rw-mount/snapshot mount)
in the test function passed to sget(), this patch first looks up the
nilfs_sb_info struct which the given mount type matches, and then
acquires the super block instance holding the nilfs_sb_info.

Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agonilfs2: get rid of sget use for checking if current mount is present
Ryusuke Konishi [Sun, 7 Jun 2009 16:39:30 +0000]
nilfs2: get rid of sget use for checking if current mount is present

This stops using sget() for checking if an r/w-mount or an r/o-mount
exists on the device.  This elimination uses a back pointer to the
current mount added to nilfs object.

Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agonilfs2: get rid of sget use for acquiring nilfs object
Ryusuke Konishi [Sun, 7 Jun 2009 16:39:29 +0000]
nilfs2: get rid of sget use for acquiring nilfs object

This will change the way to obtain nilfs object in nilfs_get_sb()
function.

Previously, a preliminary sget() call was performed, and the nilfs
object was acquired from a super block instance found by the sget()
call.

This patch, instead, instroduces a new dedicated function
find_or_create_nilfs(); as the name implies, the function finds an
existent nilfs object from a global list or creates a new one if no
object is found on the device.

Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agonilfs2: remove meaningless EBUSY case from nilfs_get_sb function
Ryusuke Konishi [Sun, 7 Jun 2009 16:39:28 +0000]
nilfs2: remove meaningless EBUSY case from nilfs_get_sb function

The following EBUSY case in nilfs_get_sb() is meaningless.  Indeed,
this error code is never returned to the caller.

    if (!s->s_root) {
          ...
    } else if (!(s->s_flags & MS_RDONLY)) {
        err = -EBUSY;
    }

This simply removes the else case.

Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agoremove the call to ->write_super in __sync_filesystem
Christoph Hellwig [Mon, 8 Jun 2009 08:08:54 +0000]
remove the call to ->write_super in __sync_filesystem

Now that all filesystems provide ->sync_fs methods we can change
__sync_filesystem to only call ->sync_fs.

This gives us a clear separation between periodic writeouts which
are driven by ->write_super and data integrity syncs that go
through ->sync_fs. (modulo file_fsync which is also going away)

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agonilfs2: call nilfs2_write_super from nilfs2_sync_fs
Christoph Hellwig [Mon, 8 Jun 2009 08:08:36 +0000]
nilfs2: call nilfs2_write_super from nilfs2_sync_fs

The call to ->write_super from __sync_filesystem will go away, so make
sure nilfs2 performs the same actions from inside ->sync_fs.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agojffs2: call jffs2_write_super from jffs2_sync_fs
Christoph Hellwig [Mon, 8 Jun 2009 08:08:21 +0000]
jffs2: call jffs2_write_super from jffs2_sync_fs

The call to ->write_super from __sync_filesystem will go away, so make
sure jffs2 performs the same actions from inside ->sync_fs.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agoufs: add ->sync_fs
Christoph Hellwig [Mon, 8 Jun 2009 08:08:05 +0000]
ufs: add ->sync_fs

Add a ->sync_fs method for data integrity syncs, and reimplement
->write_super ontop of it.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agosysv: add ->sync_fs
Christoph Hellwig [Mon, 8 Jun 2009 08:07:45 +0000]
sysv: add ->sync_fs

Add a ->sync_fs method for data integrity syncs, and reimplement
->write_super ontop of it.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agohfsplus: add ->sync_fs
Christoph Hellwig [Mon, 8 Jun 2009 08:05:12 +0000]
hfsplus: add ->sync_fs

Add a ->sync_fs method for data integrity syncs, and reimplement
->write_super ontop of it.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agohfs: add ->sync_fs
Christoph Hellwig [Mon, 8 Jun 2009 08:04:54 +0000]
hfs: add ->sync_fs

Add a ->sync_fs method for data integrity syncs.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agofat: add ->sync_fs
Christoph Hellwig [Mon, 8 Jun 2009 08:04:35 +0000]
fat: add ->sync_fs

Add a ->sync_fs method for data integrity syncs.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agoext2: add ->sync_fs
Christoph Hellwig [Mon, 8 Jun 2009 08:04:17 +0000]
ext2: add ->sync_fs

Add a ->sync_fs method for data integrity syncs, and reimplement
->write_super ontop of it.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agoexofs: add ->sync_fs
Christoph Hellwig [Mon, 8 Jun 2009 08:03:58 +0000]
exofs: add ->sync_fs

Add a ->sync_fs method for data integrity syncs, and reimplement
->write_super ontop of it.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agobfs: add ->sync_fs
Christoph Hellwig [Mon, 8 Jun 2009 08:03:38 +0000]
bfs: add ->sync_fs

Add a ->sync_fs method for data integrity syncs, and reimplement
->write_super ontop of it.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agoaffs: add ->sync_fs
Christoph Hellwig [Mon, 8 Jun 2009 08:03:15 +0000]
affs: add ->sync_fs

Add a ->sync_fs method for data integrity syncs.  Factor out common code
between affs_put_super, affs_write_super and the new affs_sync_fs into
a helper.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agosanitize ->fsync() for affs
Al Viro [Mon, 8 Jun 2009 05:22:00 +0000]
sanitize ->fsync() for affs

unfortunately, for affs (especially for affs directories) we have
no real way to keep track of metadata ownership.  So we have to
do more or less what file_fsync() does, but we do *not* need to
call write_super() there.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agorepair bfs_write_inode(), switch bfs to simple_fsync()
Al Viro [Mon, 8 Jun 2009 05:15:58 +0000]
repair bfs_write_inode(), switch bfs to simple_fsync()

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agoFix adfs GET_FRAG_ID() on big-endian
Al Viro [Mon, 8 Jun 2009 04:46:40 +0000]
Fix adfs GET_FRAG_ID() on big-endian

Missing conversion to host-endian before doing shifts

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agorepair adfs ->write_inode(), switch to simple_fsync()
Al Viro [Mon, 8 Jun 2009 04:44:42 +0000]
repair adfs ->write_inode(), switch to simple_fsync()

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agoswitch omfs to simple_fsync()
Al Viro [Sun, 7 Jun 2009 19:44:50 +0000]
switch omfs to simple_fsync()

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agoswitch udf to simple_fsync()
Al Viro [Sun, 7 Jun 2009 19:40:27 +0000]
switch udf to simple_fsync()

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agoswitch ufs to simple_fsync()
Al Viro [Sun, 7 Jun 2009 19:35:18 +0000]
switch ufs to simple_fsync()

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agorepair sysv_write_inode(), switch sysv to simple_fsync()
Al Viro [Sun, 7 Jun 2009 19:29:45 +0000]
repair sysv_write_inode(), switch sysv to simple_fsync()

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agoswitch minix to simple_fsync()
Al Viro [Sun, 7 Jun 2009 19:21:06 +0000]
switch minix to simple_fsync()

* get minix_write_inode() to honour the second argument
* now we can use simple_fsync() for minixfs

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agoswitch ext2 to simple_fsync()
Al Viro [Sun, 7 Jun 2009 19:14:02 +0000]
switch ext2 to simple_fsync()

kill ext2_sync_file() (along with ext2/fsync.c), get rid of
ext2_update_inode() - it's an alias of ext2_write_inode().

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agoSanitize ->fsync() for FAT
Al Viro [Sun, 7 Jun 2009 17:44:36 +0000]
Sanitize ->fsync() for FAT

* mark directory data blocks as assoc. metadata
* add new inode to deal with FAT, mark FAT blocks as assoc. metadata of that
* now ->fsync() is trivial both for files and directories

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agofs/qnx4: sanitize includes
Al Viro [Sun, 7 Jun 2009 13:47:13 +0000]
fs/qnx4: sanitize includes

fs-internal parts of qnx4_fs.h taken to fs/qnx4/qnx4.h, includes adjusted,
qnx4_fs.h doesn't need unifdef anymore.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agoSanitize qnx4 fsync handling
Al Viro [Sun, 7 Jun 2009 13:30:08 +0000]
Sanitize qnx4 fsync handling

* have directory operations use mark_buffer_dirty_inode(),
  so that sync_mapping_buffers() would get those.
* make qnx4_write_inode() honour its last argument.
* get rid of insane copies of very ancient "walk the indirect blocks"
  in qnx4/fsync - they never matched the actual fs layout and, fortunately,
  never'd been called.  Again, all this junk is not needed; ->fsync()
  should just do sync_mapping_buffers + sync_inode (and if we implement
  block allocation for qnx4, we'll need to use mark_buffer_dirty_inode()
  for extent blocks)

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agoNew helper - simple_fsync()
Al Viro [Sun, 7 Jun 2009 18:56:44 +0000]
New helper - simple_fsync()

writes associated buffers, then does sync_inode() to write
the inode itself (and to make it clean).  Depends on
->write_inode() honouring the second argument.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agoPush BKL down into ->remount_fs()
Alessio Igor Bogani [Tue, 12 May 2009 13:10:54 +0000]
Push BKL down into ->remount_fs()

[xfs, btrfs, capifs, shmem don't need BKL, exempt]

Signed-off-by: Alessio Igor Bogani <abogani@texware.it>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agofs: block_dump missing dentry locking
Nick Piggin [Thu, 28 May 2009 07:01:15 +0000]
fs: block_dump missing dentry locking

I think the block_dump output in __mark_inode_dirty is missing dentry locking.
Surely the i_dentry list can change any time, so we may not even *get* a
dentry there. If we do get one by chance, then it would appear to be able to
go away or get renamed at any time...

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agofs: remove incorrect I_NEW warnings
Nick Piggin [Tue, 2 Jun 2009 10:07:47 +0000]
fs: remove incorrect I_NEW warnings

Some filesystems can call in to sync an inode that is still in the
I_NEW state (eg. ext family, when mounted with -osync). This is OK
because the filesystem has sole access to the new inode, so it can
modify i_state without races (because no other thread should be
modifying it, by definition of I_NEW). Ie. a false positive, so
remove the warnings.

The races are described here 7ef0d7377cb287e08f3ae94cebc919448e1f5dff,
which is also where the warnings were introduced.

Reported-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agolinux/magic.h: move cramfs magic out of cramfs_fs.h
Mike Frysinger [Tue, 26 May 2009 09:45:04 +0000]
linux/magic.h: move cramfs magic out of cramfs_fs.h

Signed-off-by: Mike Frysinger <vapier@gentoo.org>
CC: Alexander Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agoxfs: remove ->write_super and stop maintaining ->s_dirt
Christoph Hellwig [Fri, 5 Jun 2009 10:26:23 +0000]
xfs: remove ->write_super and stop maintaining ->s_dirt

the write_super method is used for

 (1) writing back the superblock periodically from pdflush
 (2) called just before ->sync_fs for data integerity syncs

We don't need (1) because we have our own peridoc writeout through xfssyncd,
and we don't need (2) because xfs_fs_sync_fs performs a proper synchronous
superblock writeout after all other data and metadata has been written out.

Also remove ->s_dirt tracking as it's only used to decide when too call
->write_super.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Eric Sandeen <sandeen@sandeen.net>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agontfs: remove old debug check for dirty data in ntfs_put_super()
Jens Axboe [Mon, 25 May 2009 07:30:45 +0000]
ntfs: remove old debug check for dirty data in ntfs_put_super()

This should not trigger anymore, so kill it.

Acked-by: Anton Altaparmakov <aia21@cam.ac.uk>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agofs: Rearrange inode structure elements to avoid waste due to padding
Theodore Ts'o [Thu, 21 May 2009 20:01:02 +0000]
fs: Rearrange inode structure elements to avoid waste due to padding

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agofs: Remove i_cindex from struct inode
Theodore Ts'o [Thu, 21 May 2009 20:01:00 +0000]
fs: Remove i_cindex from struct inode

The only user of the i_cindex element in the inode structure is used
is by the firewire drivers.  As part of an attempt to slim down the
inode structure to save memory --- since a typical Linux system will
have hundreds of thousands if not millions of inodes cached, a
reduction in the size inode has high leverage.

The firewire driver does not need i_cindex in any fast path, so it's
simple enough to calculate when it is needed, instead of wasting space
in the inode structure.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: krh@redhat.com
Cc: stefanr@s5r6.in-berlin.de
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years ago->write_super lock_super pushdown
Christoph Hellwig [Mon, 11 May 2009 21:35:03 +0000]
->write_super lock_super pushdown

Push down lock_super into ->write_super instances and remove it from the
caller.

Following filesystem don't need ->s_lock in ->write_super and are skipped:

 * bfs, nilfs2 - no other uses of s_lock and have internal locks in
->write_super
 * ext2 - uses BKL in ext2_write_super and has internal calls without s_lock
 * reiserfs - no other uses of s_lock as has reiserfs_write_lock (BKL) in
  ->write_super
 * xfs - no other uses of s_lock and uses internal lock (buffer lock on
superblock buffer) to serialize ->write_super.  Also xfs_fs_write_super
is superflous and will go away in the next merge window

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agojffs2: move jffs2_write_super to super.c
Christoph Hellwig [Mon, 11 May 2009 21:34:27 +0000]
jffs2: move jffs2_write_super to super.c

jffs2_write_super is only called from super.c and doesn't use any
functionality from fs.c.  So move it over to super.c and make it
static there.

[should go in through the vfs tree as it is a requirement for the
 next patch]

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agoPush BKL down into do_remount_sb()
Al Viro [Fri, 8 May 2009 17:36:58 +0000]
Push BKL down into do_remount_sb()

[folded fix from Jiri Slaby]

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agoPush BKL down beyond VFS-only parts of do_mount()
Al Viro [Fri, 8 May 2009 17:34:06 +0000]
Push BKL down beyond VFS-only parts of do_mount()

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agoPush BKL into do_mount()
Al Viro [Fri, 8 May 2009 17:31:17 +0000]
Push BKL into do_mount()

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agoPush lock_super() into the ->remount_fs() of filesystems that care about it
Al Viro [Wed, 6 May 2009 14:43:07 +0000]
Push lock_super() into the ->remount_fs() of filesystems that care about it

Note that since we can't run into contention between remount_fs and write_super
(due to exclusion on s_umount), we have to care only about filesystems that
touch lock_super() on their own.  Out of those ext3, ext4, hpfs, sysv and ufs
do need it; fat doesn't since its ->remount_fs() only accesses assign-once
data (basically, it's "we have no atime on directories and only have atime on
files for vfat; force nodiratime and possibly noatime into *flags").

[folded a build fix from hch]

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agopush BKL down into ->put_super
Christoph Hellwig [Tue, 5 May 2009 13:40:36 +0000]
push BKL down into ->put_super

Move BKL into ->put_super from the only caller.  A couple of
filesystems had trivial enough ->put_super (only kfree and NULLing of
s_fs_info + stuff in there) to not get any locking: coda, cramfs, efs,
hugetlbfs, omfs, qnx4, shmem, all others got the full treatment.  Most
of them probably don't need it, but I'd rather sort that out individually.
Preferably after all the other BKL pushdowns in that area.

[AV: original used to move lock_super() down as well; these changes are
removed since we don't do lock_super() at all in generic_shutdown_super()
now]
[AV: fuse, btrfs and xfs are known to need no damn BKL, exempt]

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agoNo need to do lock_super() for exclusion in generic_shutdown_super()
Al Viro [Wed, 6 May 2009 02:10:44 +0000]
No need to do lock_super() for exclusion in generic_shutdown_super()

We can't run into contention on it.  All other callers of lock_super()
either hold s_umount (and we have it exclusive) or hold an active
reference to superblock in question, which prevents the call of
generic_shutdown_super() while the reference is held.  So we can
replace lock_super(s) with get_fs_excl() in generic_shutdown_super()
(and corresponding change for unlock_super(), of course).

Since ext4 expects s_lock held for its put_super, take lock_super()
into it.  The rest of filesystems do not care at all.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agoTrim a bit of crap from fs.h
Al Viro [Thu, 7 May 2009 07:12:29 +0000]
Trim a bit of crap from fs.h

do_remount_sb() is fs/internal.h fodder, fsync_no_super() is long gone.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agoMake sure that all callers of remount hold s_umount exclusive
Al Viro [Wed, 6 May 2009 03:48:50 +0000]
Make sure that all callers of remount hold s_umount exclusive

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agoenforce ->sync_fs is only called for rw superblock
Christoph Hellwig [Tue, 5 May 2009 13:41:25 +0000]
enforce ->sync_fs is only called for rw superblock

Make sure a superblock really is writeable by checking MS_RDONLY
under s_umount.  sync_filesystems needed some re-arragement for
that, but all but one sync_filesystem caller had the correct locking
already so that we could add that check there.  cachefiles grew
s_umount locking.

I've also added a WARN_ON to sync_filesystem to assert this for
future callers.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agocleanup sync_supers
Christoph Hellwig [Tue, 5 May 2009 14:08:56 +0000]
cleanup sync_supers

Merge the write_super helper into sync_super and move the check for
->write_super earlier so that we can avoid grabbing a reference to
a superblock that doesn't have it.

While we're at it also add a little comment documenting sync_supers.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agodcache: extrace and use d_unlinked()
Alexey Dobriyan [Sun, 3 May 2009 23:32:03 +0000]
dcache: extrace and use d_unlinked()

d_unlinked() will be used in middle-term to ban checkpointing when opened
but unlinked file is detected, and in long term, to detect such situation
and special case on it.

Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agoremove ->write_super call in generic_shutdown_super
Christoph Hellwig [Tue, 28 Apr 2009 16:00:26 +0000]
remove ->write_super call in generic_shutdown_super

We just did a full fs writeout using sync_filesystem before, and if
that's not enough for the filesystem it can perform it's own writeout
in ->put_super, which many filesystems already do.

Move a call to foofs_write_super into every foofs_put_super for now to
guarantee identical behaviour until it's cleaned up by the individual
filesystem maintainers.

Exceptions:

 - affs already has identical copy & pasted code at the beginning of
   affs_put_super so no need to do it twice.
 - xfs does the right thing without it and I have changes pending for
   the xfs tree touching this are so I don't really need conflicts
   here..

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agoqnx4: remove ->write_super
Christoph Hellwig [Mon, 27 Apr 2009 13:46:45 +0000]
qnx4: remove ->write_super

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agoocfs2: remove ->write_super and stop maintaining ->s_dirt
Christoph Hellwig [Mon, 27 Apr 2009 13:46:44 +0000]
ocfs2: remove ->write_super and stop maintaining ->s_dirt

Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agogfs2: remove ->write_super and stop maintaining ->s_dirt
Christoph Hellwig [Mon, 27 Apr 2009 13:46:43 +0000]
gfs2: remove ->write_super and stop maintaining ->s_dirt

Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Steven Whitehouse <swhiteho@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agoext3: remove ->write_super and stop maintaining ->s_dirt
Christoph Hellwig [Mon, 27 Apr 2009 13:46:42 +0000]
ext3: remove ->write_super and stop maintaining ->s_dirt

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agobtrfs: remove ->write_super and stop maintaining ->s_dirt
Christoph Hellwig [Mon, 27 Apr 2009 13:46:41 +0000]
btrfs: remove ->write_super and stop maintaining ->s_dirt

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agoquota: Introduce writeout_quota_sb() (version 4)
Jan Kara [Mon, 27 Apr 2009 14:43:55 +0000]
quota: Introduce writeout_quota_sb() (version 4)

Introduce this function which just writes all the quota structures but
avoids all the syncing and cache pruning work to expose quota structures
to userspace. Use this function from __sync_filesystem when wait == 0.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agoquota: cleanup dquota sync functions (version 4)
Christoph Hellwig [Mon, 27 Apr 2009 14:43:54 +0000]
quota: cleanup dquota sync functions (version 4)

Currently the VFS calls vfs_dq_sync to sync out disk quotas for a given
superblock.  This is a small wrapper around sync_dquots which for the
case of a non-NULL superblock is a small wrapper around quota_sync_sb.

Just make quota_sync_sb global (rename it to sync_quota_sb) and call it
directly.  Also call it directly for those cases in quota.c that have a
superblock and leave sync_dquots purely an iterator over sync_quota_sb and
remove it's superblock argument.

To make this nicer move the check for the lack of a quota_sync method
from the callers into sync_quota_sb.

[folded build fix from Alexander Beregalov <a.beregalov@gmail.com>]

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agovfs: Rename fsync_super() to sync_filesystem() (version 4)
Jan Kara [Mon, 27 Apr 2009 14:43:53 +0000]
vfs: Rename fsync_super() to sync_filesystem() (version 4)

Rename the function so that it better describe what it really does. Also
remove the unnecessary include of buffer_head.h.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agovfs: Move syncing code from super.c to sync.c (version 4)
Jan Kara [Mon, 27 Apr 2009 14:43:52 +0000]
vfs: Move syncing code from super.c to sync.c (version 4)

Move sync_filesystems(), __fsync_super(), fsync_super() from
super.c to sync.c where it fits better.

[build fixes folded]

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agovfs: Make sys_sync() use fsync_super() (version 4)
Jan Kara [Mon, 27 Apr 2009 14:43:51 +0000]
vfs: Make sys_sync() use fsync_super() (version 4)

It is unnecessarily fragile to have two places (fsync_super() and do_sync())
doing data integrity sync of the filesystem. Alter __fsync_super() to
accommodate needs of both callers and use it. So after this patch
__fsync_super() is the only place where we gather all the calls needed to
properly send all data on a filesystem to disk.

Nice bonus is that we get a complete livelock avoidance and write_supers()
is now only used for periodic writeback of superblocks.

sync_blockdevs() introduced a couple of patches ago is gone now.

[build fixes folded]

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agovfs: Make __fsync_super() a static function (version 4)
Jan Kara [Mon, 27 Apr 2009 14:43:50 +0000]
vfs: Make __fsync_super() a static function (version 4)

__fsync_super() does the same thing as fsync_super(). So change the only
caller to use fsync_super() and make __fsync_super() static. This removes
unnecessarily duplicated call to sync_blockdev() and prepares ground
for the changes to __fsync_super() in the following patches.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agovfs: Call ->sync_fs() even if s_dirt is 0 (version 4)
Jan Kara [Mon, 27 Apr 2009 14:43:49 +0000]
vfs: Call ->sync_fs() even if s_dirt is 0 (version 4)

sync_filesystems() has a condition that if wait == 0 and s_dirt == 0, then
->sync_fs() isn't called. This does not really make much sence since s_dirt is
generally used by a filesystem to mean that ->write_super() needs to be called.
But ->sync_fs() does different things. I even suspect that some filesystems
(btrfs?) sets s_dirt just to fool this logic.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agovfs: Fix sys_sync() and fsync_super() reliability (version 4)
Jan Kara [Mon, 27 Apr 2009 14:43:48 +0000]
vfs: Fix sys_sync() and fsync_super() reliability (version 4)

So far, do_sync() called:
  sync_inodes(0);
  sync_supers();
  sync_filesystems(0);
  sync_filesystems(1);
  sync_inodes(1);

This ordering makes it kind of hard for filesystems as sync_inodes(0) need not
submit all the IO (for example it skips inodes with I_SYNC set) so e.g. forcing
transaction to disk in ->sync_fs() is not really enough. Therefore sys_sync has
not been completely reliable on some filesystems (ext3, ext4, reiserfs, ocfs2
and others are hit by this) when racing e.g. with background writeback. A
similar problem hits also other filesystems (e.g. ext2) because of
write_supers() being called before the sync_inodes(1).

Change the ordering of calls in do_sync() - this requires a new function
sync_blockdevs() to preserve the property that block devices are always synced
after write_super() / sync_fs() call.

The same issue is fixed in __fsync_super() function used on umount /
remount read-only.

[AV: build fixes]

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agoremove s_async_list
Christoph Hellwig [Tue, 28 Apr 2009 16:05:55 +0000]
remove s_async_list

Remove the unused s_async_list in the superblock, a leftover of the
broken async inode deletion code that leaked into mainline.  Having this
in the middle of the sync/unmount path is not helpful for the following
cleanups.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agofs: move mark_files_ro into file_table.c
npiggin@suse.de [Sun, 26 Apr 2009 10:25:56 +0000]
fs: move mark_files_ro into file_table.c

This function walks the s_files lock, and operates primarily on the
files in a superblock, so it better belongs here (eg. see also
fs_may_remount_ro).

[AV: ... and it shouldn't be static after that move]

Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agofs: introduce mnt_clone_write
npiggin@suse.de [Sun, 26 Apr 2009 10:25:55 +0000]
fs: introduce mnt_clone_write

This patch speeds up lmbench lat_mmap test by about another 2% after the
first patch.

Before:
 avg = 462.286
 std = 5.46106

After:
 avg = 453.12
 std = 9.58257

(50 runs of each, stddev gives a reasonable confidence)

It does this by introducing mnt_clone_write, which avoids some heavyweight
operations of mnt_want_write if called on a vfsmount which we know already
has a write count; and mnt_want_write_file, which can call mnt_clone_write
if the file is open for write.

After these two patches, mnt_want_write and mnt_drop_write go from 7% on
the profile down to 1.3% (including mnt_clone_write).

[AV: mnt_want_write_file() should take file alone and derive mnt from it;
not only all callers have that form, but that's the only mnt about which
we know that it's already held for write if file is opened for write]

Cc: Dave Hansen <haveblue@us.ibm.com>
Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

10 years agofs: mnt_want_write speedup
npiggin@suse.de [Sun, 26 Apr 2009 10:25:54 +0000]
fs: mnt_want_write speedup

This patch speeds up lmbench lat_mmap test by about 8%. lat_mmap is set up
basically to mmap a 64MB file on tmpfs, fault in its pages, then unmap it.
A microbenchmark yes, but it exercises some important paths in the mm.

Before:
 avg = 501.9
 std = 14.7773

After:
 avg = 462.286
 std = 5.46106

(50 runs of each, stddev gives a reasonable confidence, but there is quite
a bit of variation there still)

It does this by removing the complex per-cpu locking and counter-cache and
replaces it with a percpu counter in struct vfsmount. This makes the code
much simpler, and avoids spinlocks (although the msync is still pretty
costly, unfortunately). It results in about 900 bytes smaller code too. It
does increase the size of a vfsmount, however.

It should also give a speedup on large systems if CPUs are frequently operating
on different mounts (because the existing scheme has to operate on an atomic in
the struct vfsmount when switching between mounts). But I'm most interested in
the single threaded path performance for the moment.

[AV: minor cleanup]

Cc: Dave Hansen <haveblue@us.ibm.com>
Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>