It actually goes back to 2004 ([PATCH] Concurrent O_SYNC write support)
when sync_page_range() had been introduced; generic_file_write{,v}() correctly
synced
	pos_after_write - written .. pos_after_write - 1
but generic_file_aio_write() synced
	pos_before_write .. pos_before_write + written - 1
instead.  Which is not the same thing with O_APPEND, obviously.
A couple of years later correct variant had been killed off when
everything switched to use of generic_file_aio_write().

All users of generic_file_aio_write() are affected, and the same bug
has been copied into other instances of ->aio_write().

The fix is trivial; the only subtle point is that generic_write_sync()
ought to be inlined to avoid calculations useless for the majority of
calls.
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

When there are processes heavily creating small files while sync(2) is
running, it can easily happen that quite some new files are created
between WB_SYNC_NONE and WB_SYNC_ALL pass of sync(2).  That can happen
especially if there are several busy filesystems (remember that sync
traverses filesystems sequentially and waits in WB_SYNC_ALL phase on one
fs before starting it on another fs).  Because WB_SYNC_ALL pass is slow
(e.g.  causes a transaction commit and cache flush for each inode in
ext3), resulting sync(2) times are rather large.

The following script reproduces the problem:

  function run_writers
  {
    for (( i = 0; i < 10; i++ )); do
      mkdir $1/dir$i
      for (( j = 0; j < 40000; j++ )); do
        dd if=/dev/zero of=$1/dir$i/$j bs=4k count=4 &>/dev/null
      done &
    done
  }

  for dir in "$@"; do
    run_writers $dir
  done

  sleep 40
  time sync

Fix the problem by disregarding inodes dirtied after sync(2) was called
in the WB_SYNC_ALL pass.  To allow for this, sync_inodes_sb() now takes
a time stamp when sync has started which is used for setting up work for
flusher threads.

To give some numbers, when above script is run on two ext4 filesystems
on simple SATA drive, the average sync time from 10 runs is 267.549
seconds with standard deviation 104.799426.  With the patched kernel,
the average sync time from 10 runs is 2.995 seconds with standard
deviation 0.096.
Signed-off-by: NJan Kara <jack@suse.cz>
Reviewed-by: NFengguang Wu <fengguang.wu@intel.com>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

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

... and convert a bunch of SYSCALL_DEFINE ones to SYSCALL_DEFINE<n>,
killing the boilerplate crap around them.
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

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

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

wakeup_flusher_threads(0) will queue work doing complete writeback for each
flusher thread. Thus there is not much point in submitting another work doing
full inode WB_SYNC_NONE writeback by writeback_inodes_sb().

After this change it does not make sense to call nonblocking ->sync_fs and
block device flush before calling sync_inodes_sb() because
wakeup_flusher_threads() is completely asynchronous and thus these functions
would be called in parallel with inode writeback running which will effectively
void any work they do. So we move sync_inodes_sb() call before these two
functions.
Signed-off-by: NJan Kara <jack@suse.cz>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

It is not necessary to write block devices twice. The reason why we first did
flush and then proper sync is that
  for_each_bdev() {
    write_bdev()
    wait_for_completion()
  }
is much slower than
  for_each_bdev()
    write_bdev()
  for_each_bdev()
    wait_for_completion()
when there is bigger amount of data. But as is seen in the above, there's no real
need to scan pages and submit them twice. We just need to separate the submission
and waiting part. This patch does that.
Signed-off-by: NJan Kara <jack@suse.cz>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

In case block device does not have filesystem mounted on it, sys_sync will just
ignore it and doesn't writeout its dirty pages. This is because writeback code
avoids writing inodes from superblock without backing device and
blockdev_superblock is such a superblock.  Since it's unexpected that sync
doesn't writeout dirty data for block devices be nice to users and change the
behavior to do so. So now we iterate over all block devices on blockdev_super
instead of iterating over all superblocks when syncing block devices.
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJan Kara <jack@suse.cz>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Change the order of operations during sync from

for_each_sb {
        writeback_inodes_sb();
        sync_fs(nowait);
        __sync_blockdev(nowait);
}
for_each_sb {
        sync_inodes_sb();
        sync_fs(wait);
        __sync_blockdev(wait);
}

to

for_each_sb
        writeback_inodes_sb();
for_each_sb
        sync_fs(nowait);
for_each_sb
        __sync_blockdev(nowait);
for_each_sb
        sync_inodes_sb();
for_each_sb
        sync_fs(wait);
for_each_sb
        __sync_blockdev(wait);

This is a preparation for the following patches in this series.
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJan Kara <jack@suse.cz>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Since the moment writes to quota files are using block device page cache and
space for quota structures is reserved at the moment they are first accessed we
have no reason to sync quota before inode writeback. In fact this order is now
only harmful since quota information can easily change during inode writeback
(either because conversion of delayed-allocated extents or simply because of
allocation of new blocks for simple filesystems not using page_mkwrite).

So move syncing of quota information after writeback of inodes into ->sync_fs
method. This way we do not have to use ->quota_sync callback which is primarily
intended for use by quotactl syscall anyway and we get rid of calling
->sync_fs() twice unnecessarily. We skip quota syncing for OCFS2 since it does
proper quota journalling in all cases (unlike ext3, ext4, and reiserfs which
also support legacy non-journalled quotas) and thus there are no dirty quota
structures.

CC: "Theodore Ts'o" <tytso@mit.edu>
CC: Joel Becker <jlbec@evilplan.org>
CC: reiserfs-devel@vger.kernel.org
Acked-by: NSteven Whitehouse <swhiteho@redhat.com>
Acked-by: NDave Kleikamp <shaggy@kernel.org>
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJan Kara <jack@suse.cz>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Split off part of dquot_quota_sync() which writes dquots into a quota file
to a separate function. In the next patch we will use the function from
filesystems and we do not want to abuse ->quota_sync quotactl callback more
than necessary.
Acked-by: NSteven Whitehouse <swhiteho@redhat.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJan Kara <jack@suse.cz>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

In principle, a filesystem may want to have ->sync_fs() called during sync(1)
although it does not have a bdi (i.e. s_bdi is set to noop_backing_dev_info).
Only writeback code really needs bdi set to something reasonable. So move the
checks where they are more logical.
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJan Kara <jack@suse.cz>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Signed-off-by: NWanpeng Li <liwp@linux.vnet.ibm.com>
Signed-off-by: NFengguang Wu <fengguang.wu@intel.com>

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

For files only using THIS_MODULE and/or EXPORT_SYMBOL, map
them onto including export.h -- or if the file isn't even
using those, then just delete the include.  Fix up any implicit
include dependencies that were being masked by module.h along
the way.
Signed-off-by: NPaul Gortmaker <paul.gortmaker@windriver.com>

Move invalidate_bdev, block_sync_page into fs/block_dev.c.  Export
kill_bdev as well, so brd doesn't have to open code it.  Reduce
buffer_head.h requirement accordingly.

Removed a rather large comment from invalidate_bdev, as it looked a bit
obsolete to bother moving.  The small comment replacing it says enough.
Signed-off-by: NNick Piggin <npiggin@suse.de>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

This creates a new 'reason' field in a wb_writeback_work
structure, which unambiguously identifies who initiates
writeback activity.  A 'wb_reason' enumeration has been
added to writeback.h, to enumerate the possible reasons.

The 'writeback_work_class' and tracepoint event class and
'writeback_queue_io' tracepoints are updated to include the
symbolic 'reason' in all trace events.

And the 'writeback_inodes_sbXXX' family of routines has had
a wb_stats parameter added to them, so callers can specify
why writeback is being started.
Acked-by: NJan Kara <jack@suse.cz>
Signed-off-by: NCurt Wohlgemuth <curtw@google.com>
Signed-off-by: NWu Fengguang <fengguang.wu@intel.com>

Btrfs needs to be able to control how filemap_write_and_wait_range() is called
in fsync to make it less of a painful operation, so push down taking i_mutex and
the calling of filemap_write_and_wait() down into the ->fsync() handlers.  Some
file systems can drop taking the i_mutex altogether it seems, like ext3 and
ocfs2.  For correctness sake I just pushed everything down in all cases to make
sure that we keep the current behavior the same for everybody, and then each
individual fs maintainer can make up their mind about what to do from there.
Thanks,
Acked-by: NJan Kara <jack@suse.cz>
Signed-off-by: NJosef Bacik <josef@redhat.com>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

It is frequently useful to sync a single file system, instead of all
mounted file systems via sync(2):

 - On machines with many mounts, it is not at all uncommon for some of
   them to hang (e.g. unresponsive NFS server).  sync(2) will get stuck on
   those and may never get to the one you do care about (e.g., /).
 - Some applications write lots of data to the file system and then
   want to make sure it is flushed to disk.  Calling fsync(2) on each
   file introduces unnecessary ordering constraints that result in a large
   amount of sub-optimal writeback/flush/commit behavior by the file
   system.

There are currently two ways (that I know of) to sync a single super_block:

 - BLKFLSBUF ioctl on the block device: That also invalidates the bdev
   mapping, which isn't usually desirable, and doesn't work for non-block
   file systems.
 - 'mount -o remount,rw' will call sync_filesystem as an artifact of the
   current implemention.  Relying on this little-known side effect for
   something like data safety sounds foolish.

Both of these approaches require root privileges, which some applications
do not have (nor should they need?) given that sync(2) is an unprivileged
operation.

This patch introduces a new system call syncfs(2) that takes an fd and
syncs only the file system it references.  Maybe someday we can

 $ sync /some/path

and not get

 sync: ignoring all arguments

The syscall is motivated by comments by Al and Christoph at the last LSF.
syncfs(2) seems like an appropriate name given statfs(2).

A similar ioctl was also proposed a while back, see
	http://marc.info/?l=linux-fsdevel&m=127970513829285&w=2Signed-off-by: NSage Weil <sage@newdream.net>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

We don't have proper reference counting for this yet, so we run into
cases where the device is pulled and we OOPS on flushing the fs data.
This happens even though the dirty inodes have already been
migrated to the default_backing_dev_info.
Reported-by: NTorsten Hilbrich <torsten.hilbrich@secunet.com>
Tested-by: NTorsten Hilbrich <torsten.hilbrich@secunet.com>
Cc: stable@kernel.org
Signed-off-by: NJens Axboe <jaxboe@fusionio.com>

Copy and simplify in the only two users remaining.
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

This reverts commit e913fc82.

We are investigating a hang associated with the WB_SYNC_NONE changes,
so revert them for now.

Conflicts:

	fs/fs-writeback.c
	mm/page-writeback.c
Signed-off-by: NJens Axboe <jaxboe@fusionio.com>

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

Now that the last user passing a NULL file pointer is gone we can remove
the redundant dentry argument and associated hacks inside vfs_fsynmc_range.

The next step will be removig the dentry argument from ->fsync, but given
the luck with the last round of method prototype changes I'd rather
defer this until after the main merge window.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

... and switch the simple "loop over superblocks and do something"
loops to it.
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

At the same time we can kill s_need_restart and local mutex in there.
__put_super() made public for a while; will be gone later.
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

We used to remove from s_list and s_instances at the same
time.  So let's *not* do the former and skip superblocks
that have empty s_instances in the loops over s_list.

The next step, of course, will be to get rid of rescan logics
in those loops.
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

When umount calls sync_filesystem(), we first do a WB_SYNC_NONE
writeback to kick off writeback of pending dirty inodes, then follow
that up with a WB_SYNC_ALL to wait for it. Since umount already holds
the sb s_umount mutex, WB_SYNC_NONE ends up doing nothing and all
writeback happens as WB_SYNC_ALL. This can greatly slow down umount,
since WB_SYNC_ALL writeback is a data integrity operation and thus
a bigger hammer than simple WB_SYNC_NONE. For barrier aware file systems
it's a lot slower.
Signed-off-by: NJens Axboe <jens.axboe@oracle.com>

noop_backing_dev_info is used only as a flag to mark filesystems that
don't have any backing store, like tmpfs, procfs, spufs, etc.
Signed-off-by: NJoern Engel <joern@logfs.org>

Changed the BUG_ON() to a WARN_ON(). Note that adding dirty inodes
to the noop_backing_dev_info is not legal and will not result in
them being flushed, but we already catch this condition in
__mark_inode_dirty() when checking for a registered bdi.
Signed-off-by: NJens Axboe <jens.axboe@oracle.com>

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: NTejun Heo <tj@kernel.org>
Guess-its-ok-by: NChristoph Lameter <cl@linux-foundation.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>

Currenly sync_quota_sb does a lot of sync and truncate action that only
applies to "VFS" style quotas and is actively harmful for the sync
performance in XFS.  Move it into vfs_quota_sync and add a wait parameter
to ->quota_sync to tell if we need it or not.

My audit of the GFS2 code says it's also not needed given the way GFS2
implements quotas, but I'd be happy if this can get a detailed review.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJan Kara <jack@suse.cz>

We recently go rid of all callers of do_sync_file_range as they're better
served with vfs_fsync or the filemap_write_and_wait.  Now that
do_sync_file_range is down to a single caller fold it into it so that people
don't start using it again accidentally.  While at it also switch it from
using __filemap_fdatawrite_range(..., WB_SYNC_ALL) to the more clear
filemap_fdatawrite_range().
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

All callers really want the more logical filemap_fdatawait_range interface,
so convert them to use it and merge wait_on_page_writeback_range into
filemap_fdatawait_range.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJan Kara <jack@suse.cz>

While Linux provided an O_SYNC flag basically since day 1, it took until
Linux 2.4.0-test12pre2 to actually get it implemented for filesystems,
since that day we had generic_osync_around with only minor changes and the
great "For now, when the user asks for O_SYNC, we'll actually give
O_DSYNC" comment.  This patch intends to actually give us real O_SYNC
semantics in addition to the O_DSYNC semantics.  After Jan's O_SYNC
patches which are required before this patch it's actually surprisingly
simple, we just need to figure out when to set the datasync flag to
vfs_fsync_range and when not.

This patch renames the existing O_SYNC flag to O_DSYNC while keeping it's
numerical value to keep binary compatibility, and adds a new real O_SYNC
flag.  To guarantee backwards compatiblity it is defined as expanding to
both the O_DSYNC and the new additional binary flag (__O_SYNC) to make
sure we are backwards-compatible when compiled against the new headers.

This also means that all places that don't care about the differences can
just check O_DSYNC and get the right behaviour for O_SYNC, too - only
places that actuall care need to check __O_SYNC in addition.  Drivers and
network filesystems have been updated in a fail safe way to always do the
full sync magic if O_DSYNC is set.  The few places setting O_SYNC for
lower layers are kept that way for now to stay failsafe.

We enforce that O_DSYNC is set when __O_SYNC is set early in the open path
to make sure we always get these sane options.

Note that parisc really screwed up their headers as they already define a
O_DSYNC that has always been a no-op.  We try to repair it by using it for
the new O_DSYNC and redefinining O_SYNC to send both the traditional
O_SYNC numerical value _and_ the O_DSYNC one.

Cc: Richard Henderson <rth@twiddle.net>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Grant Grundler <grundler@parisc-linux.org>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Andreas Dilger <adilger@sun.com>
Acked-by: NTrond Myklebust <Trond.Myklebust@netapp.com>
Acked-by: NKyle McMartin <kyle@mcmartin.ca>
Acked-by: NUlrich Drepper <drepper@redhat.com>
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NJan Kara <jack@suse.cz>

According to Documentation/CodingStyle the EXPORT* macro should follow
immediately after the closing function brace line.

Also, mark_buffer_async_write_endio() and do_thaw_all() are not used
elsewhere so they should be marked as static.

In addition, file_fsync() is actually in fs/sync.c so move the EXPORT* to
that file.
Signed-off-by: NH Hartley Sweeten <hsweeten@visionengravers.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

We do this automatically in get_sb_bdev() from the set_bdev_super()
callback. Filesystems that have their own private backing_dev_info
must assign that in ->fill_super().

Note that ->s_bdi assignment is required for proper writeback!
Acked-by: NChristoph Hellwig <hch@infradead.org>
Signed-off-by: NJens Axboe <jens.axboe@oracle.com>

Currenly vfs_fsync(_range) first calls filemap_fdatawrite to write out
the data, the calls into ->fsync to write out the metadata and then finally
calls filemap_fdatawait to wait for the data I/O to complete.  What sounds
like a clever micro-optimization actually is nast trap for many filesystems.

For many modern filesystems i_size or other inode information is only
updated on I/O completion and we need to wait for I/O to finish before
we can write out the metadata.  For old fashionen filesystems that
instanciate blocks during the actual write and also update the metadata
at that point it opens up a large window were we could expose uninitialized
blocks after a crash.  While a few filesystems that need it already wait
for the I/O to finish inside their ->fsync methods it is rather suboptimal
as it is done under the i_mutex and also always for the whole file instead
of just a part as we could do for O_SYNC handling.

Here is a small audit of all fsync instances in the tree:

 - spufs_mfc_fsync:
 - ps3flash_fsync:
 - vol_cdev_fsync:
 - printer_fsync:
 - fb_deferred_io_fsync:
 - bad_file_fsync:
 - simple_sync_file:

	don't care - filesystems/drivers do't use the page cache or are
	purely in-memory.

 - simple_fsync:
 - file_fsync:
 - affs_file_fsync:
 - fat_file_fsync:
 - jfs_fsync:
 - ubifs_fsync:
 - reiserfs_dir_fsync:
 - reiserfs_sync_file:

	never touch pagecache themselves.  We need to wait before if we do
	not want to expose stale data after an allocation.

 - afs_fsync:
 - fuse_fsync_common:

	do the waiting writeback itself in awkward ways, would benefit from
	proper semantics

 - block_fsync:

	Does a filemap_write_and_wait on the block device inode.  Because we
	now have f_mapping that is the same inode we call it on in vfs_fsync.
	So just removing it and letting the VFS do the work in one go would
	be an improvement.

 - btrfs_sync_file:
 - cifs_fsync:
 - xfs_file_fsync:

	need the wait first and currently do it themselves. would benefit from
	doing it outside i_mutex.

 - coda_fsync:
 - ecryptfs_fsync:
 - exofs_file_fsync:
 - shm_fsync:

	only passes the fsync through to the lower layer

 - ext3_sync_file:

	doesn't seem to care, comments are confusing.

 - ext4_sync_file:

	would need the wait to work correctly for delalloc mode with late
	i_size updates.  Otherwise the ext3 comment applies.

	currently implemens it's own writeback and wait in an odd way,
	could benefit from doing it properly.

 - gfs2_fsync:

	not needed for journaled data mode, but probably harmless there.
	Currently writes back data asynchronously itself.  Needs some
	major audit.

 - hostfs_fsync:

	just calls fsync/datasync on the host FD.  Without the wait before
	data might not even be inflight yet if we're unlucky.

 - hpfs_file_fsync:
 - ncp_fsync:

	no-ops.  Dangerous before and after.

 - jffs2_fsync:

	just calls jffs2_flush_wbuf_gc, not sure how this relates to data.

 - nfs_fsync_dir:

	just increments stats, claims all directory operations are synchronous

 - nfs_file_fsync:

	only writes out data???  Looks very odd.

 - nilfs_sync_file:

	looks like it expects all data done, but not sure from the code

 - ntfs_dir_fsync:
 - ntfs_file_fsync:

	appear to do their own data writeback.  Very convoluted code.

 - ocfs2_sync_file:

	does it's own data writeback, but no wait.  probably needs the wait.

 - smb_fsync:

	according to a comment expects all pages written already, probably needs
	the wait before.

This patch only changes vfs_fsync_range, removal of the wait in the methods
that have it is left to the filesystem maintainers.  Note that most
filesystems really do need an audit for their fsync methods given the
gems found in this very brief audit.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJan Kara <jack@suse.cz>

Introduce new function for generic inode syncing (vfs_fsync_range) and use
it from fsync() path. Introduce also new helper for syncing after a sync
write (generic_write_sync) using the generic function.

Use these new helpers for syncing from generic VFS functions. This makes
O_SYNC writes to block devices acquire i_mutex for syncing. If we really
care about this, we can make block_fsync() drop the i_mutex and reacquire
it before it returns.

CC: Evgeniy Polyakov <zbr@ioremap.net>
CC: ocfs2-devel@oss.oracle.com
CC: Joel Becker <joel.becker@oracle.com>
CC: Felix Blyakher <felixb@sgi.com>
CC: xfs@oss.sgi.com
CC: Anton Altaparmakov <aia21@cantab.net>
CC: linux-ntfs-dev@lists.sourceforge.net
CC: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
CC: linux-ext4@vger.kernel.org
CC: tytso@mit.edu
Acked-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJan Kara <jack@suse.cz>

This gets rid of pdflush for bdi writeout and kupdated style cleaning.
pdflush writeout suffers from lack of locality and also requires more
threads to handle the same workload, since it has to work in a
non-blocking fashion against each queue. This also introduces lumpy
behaviour and potential request starvation, since pdflush can be starved
for queue access if others are accessing it. A sample ffsb workload that
does random writes to files is about 8% faster here on a simple SATA drive
during the benchmark phase. File layout also seems a LOT more smooth in
vmstat:

 r  b   swpd   free   buff  cache   si   so    bi    bo   in    cs us sy id wa
 0  1      0 608848   2652 375372    0    0     0 71024  604    24  1 10 48 42
 0  1      0 549644   2712 433736    0    0     0 60692  505    27  1  8 48 44
 1  0      0 476928   2784 505192    0    0     4 29540  553    24  0  9 53 37
 0  1      0 457972   2808 524008    0    0     0 54876  331    16  0  4 38 58
 0  1      0 366128   2928 614284    0    0     4 92168  710    58  0 13 53 34
 0  1      0 295092   3000 684140    0    0     0 62924  572    23  0  9 53 37
 0  1      0 236592   3064 741704    0    0     4 58256  523    17  0  8 48 44
 0  1      0 165608   3132 811464    0    0     0 57460  560    21  0  8 54 38
 0  1      0 102952   3200 873164    0    0     4 74748  540    29  1 10 48 41
 0  1      0  48604   3252 926472    0    0     0 53248  469    29  0  7 47 45

where vanilla tends to fluctuate a lot in the creation phase:

 r  b   swpd   free   buff  cache   si   so    bi    bo   in    cs us sy id wa
 1  1      0 678716   5792 303380    0    0     0 74064  565    50  1 11 52 36
 1  0      0 662488   5864 319396    0    0     4   352  302   329  0  2 47 51
 0  1      0 599312   5924 381468    0    0     0 78164  516    55  0  9 51 40
 0  1      0 519952   6008 459516    0    0     4 78156  622    56  1 11 52 37
 1  1      0 436640   6092 541632    0    0     0 82244  622    54  0 11 48 41
 0  1      0 436640   6092 541660    0    0     0     8  152    39  0  0 51 49
 0  1      0 332224   6200 644252    0    0     4 102800  728    46  1 13 49 36
 1  0      0 274492   6260 701056    0    0     4 12328  459    49  0  7 50 43
 0  1      0 211220   6324 763356    0    0     0 106940  515    37  1 10 51 39
 1  0      0 160412   6376 813468    0    0     0  8224  415    43  0  6 49 45
 1  1      0  85980   6452 886556    0    0     4 113516  575    39  1 11 54 34
 0  2      0  85968   6452 886620    0    0     0  1640  158   211  0  0 46 54

A 10 disk test with btrfs performs 26% faster with per-bdi flushing. A
SSD based writeback test on XFS performs over 20% better as well, with
the throughput being very stable around 1GB/sec, where pdflush only
manages 750MB/sec and fluctuates wildly while doing so. Random buffered
writes to many files behave a lot better as well, as does random mmap'ed
writes.

A separate thread is added to sync the super blocks. In the long term,
adding sync_supers_bdi() functionality could get rid of this thread again.
Signed-off-by: NJens Axboe <jens.axboe@oracle.com>