pandora-kernel.git
12 years agobtrfs: fix uninitialized variable warning
David Sterba [Fri, 3 Jun 2011 14:29:08 +0000 (16:29 +0200)]
btrfs: fix uninitialized variable warning

With Linus' tree, today's linux-next build (powercp ppc64_defconfig)
produced this warning:

fs/btrfs/delayed-inode.c: In function 'btrfs_delayed_update_inode':
fs/btrfs/delayed-inode.c:1598:6: warning: 'ret' may be used
uninitialized in this function

Introduced by commit 16cdcec736cd ("btrfs: implement delayed inode items
operation").

This fixes a bug in btrfs_update_inode(): if the returned value from
btrfs_delayed_update_inode is a nonzero garbage, inode stat data are not
updated and several call paths may hit a BUG_ON or fail with strange
code.

Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: David Sterba <dsterba@suse.cz>
12 years agobtrfs: add helper for fs_info->closing
David Sterba [Tue, 31 May 2011 16:07:27 +0000 (18:07 +0200)]
btrfs: add helper for fs_info->closing

wrap checking of filesystem 'closing' flag and fix a few missing memory
barriers.

Signed-off-by: David Sterba <dsterba@suse.cz>
12 years agoBtrfs: add mount -o inode_cache
Chris Mason [Fri, 3 Jun 2011 13:36:29 +0000 (09:36 -0400)]
Btrfs: add mount -o inode_cache

This makes the inode map cache default to off until we
fix the overflow problem when the free space crcs don't fit
inside a single page.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agobtrfs: scrub: add explicit plugging
Arne Jansen [Sat, 28 May 2011 20:58:38 +0000 (20:58 +0000)]
btrfs: scrub: add explicit plugging

With the removal of the implicit plugging scrub ends up doing more and
smaller I/O than necessary. This patch adds explicit plugging per chunk.

Signed-off-by: Arne Jansen <sensille@gmx.net>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agobtrfs: use btrfs_ino to access inode number
David Sterba [Tue, 31 May 2011 17:08:14 +0000 (17:08 +0000)]
btrfs: use btrfs_ino to access inode number

commit 4cb5300bc ("Btrfs: add mount -o auto_defrag") accesses inode
number directly while it should use the helper with the new inode
number allocator.

Signed-off-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agoBtrfs: don't save the inode cache if we are deleting this root
Josef Bacik [Tue, 31 May 2011 19:33:33 +0000 (19:33 +0000)]
Btrfs: don't save the inode cache if we are deleting this root

With xfstest 254 I can panic the box every time with the inode number caching
stuff on.  This is because we clean the inodes out when we delete the subvolume,
but then we write out the inode cache which adds an inode to the subvolume inode
tree, and then when it gets evicted again the root gets added back on the dead
roots list and is deleted again, so we have a double free.  To stop this from
happening just return 0 if refs is 0 (and we're not the tree root since tree
root always has refs of 0).  With this fix 254 no longer panics.  Thanks,

Signed-off-by: Josef Bacik <josef@redhat.com>
Tested-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agobtrfs: false BUG_ON when degraded
Arne Jansen [Mon, 30 May 2011 08:36:16 +0000 (08:36 +0000)]
btrfs: false BUG_ON when degraded

In degraded mode the struct btrfs_device of missing devs don't have
device->name set. A kstrdup of NULL correctly returns NULL. Don't
BUG in this case.

Signed-off-by: Arne Jansen <sensille@gmx.net>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agoBtrfs: don't save the inode cache in non-FS roots
liubo [Wed, 1 Jun 2011 09:42:49 +0000 (09:42 +0000)]
Btrfs: don't save the inode cache in non-FS roots

This adds extra checks to make sure the inode map we are caching really
belongs to a FS root instead of a special relocation tree.  It
prevents crashes during balancing operations.

Signed-off-by: Liu Bo <liubo2009@cn.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agoBtrfs: make sure we don't overflow the free space cache crc page
Chris Mason [Fri, 3 Jun 2011 05:26:53 +0000 (01:26 -0400)]
Btrfs: make sure we don't overflow the free space cache crc page

The free space cache uses only one page for crcs right now,
which means we can't have a cache file bigger than the
crcs we can fit in the first page.  This adds a check to
enforce that restriction.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agoBtrfs: fix uninit variable in the delayed inode code
Chris Mason [Fri, 3 Jun 2011 05:13:45 +0000 (01:13 -0400)]
Btrfs: fix uninit variable in the delayed inode code

The nitems counter needs to start at zero

Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agobtrfs: scrub: don't reuse bios and pages
Arne Jansen [Sat, 28 May 2011 19:57:55 +0000 (21:57 +0200)]
btrfs: scrub: don't reuse bios and pages

The current scrub implementation reuses bios and pages as often as possible,
allocating them only on start and releasing them when finished. This leads
to more problems with the block layer than it's worth. The elevator gets
confused when there are more pages added to the bio than bi_size suggests.
This patch completely rips out the reuse of bios and pages and allocates
them freshly for each submit.

Signed-off-by: Arne Jansen <sensille@gmx.net>
Signed-off-by: Chris Maosn <chris.mason@oracle.com>
12 years agoMerge branch 'for-chris' of
Chris Mason [Sat, 28 May 2011 11:00:39 +0000 (07:00 -0400)]
Merge branch 'for-chris' of
git://git.kernel.org/pub/scm/linux/kernel/git/josef/btrfs-work into for-linus

Conflicts:
fs/btrfs/disk-io.c
fs/btrfs/extent-tree.c
fs/btrfs/free-space-cache.c
fs/btrfs/inode.c
fs/btrfs/transaction.c

Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agoBtrfs: use the device_list_mutex during write_dev_supers
Chris Mason [Fri, 27 May 2011 14:03:58 +0000 (10:03 -0400)]
Btrfs: use the device_list_mutex during write_dev_supers

write_dev_supers was changed to use RCU to protect the list of
devices, but it was then sleeping while it actually wrote the supers.
This fixes it to just use the mutex, since we really don't any
concurrency in write_dev_supers anyway.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agoBtrfs: setup free ino caching in a more asynchronous way
Li Zefan [Thu, 26 May 2011 06:38:30 +0000 (06:38 +0000)]
Btrfs: setup free ino caching in a more asynchronous way

For a filesystem that has lots of files in it, the first time we mount
it with free ino caching support, it can take quite a long time to
setup the caching before we can create new files.

Here we fill the cache with [highest_ino, BTRFS_LAST_FREE_OBJECTID]
before we start the caching thread to search through the extent tree.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agobtrfs scrub: don't coalesce pages that are logically discontiguous
Arne Jansen [Wed, 25 May 2011 12:22:50 +0000 (12:22 +0000)]
btrfs scrub: don't coalesce pages that are logically discontiguous

scrub_page collects several pages into one bio as long as they are physically
contiguous. As we only save one logical address for the whole bio, don't
collect pages that are physically contiguous but logically discontiguous.

Signed-off-by: Arne Jansen <sensille@gmx.net>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agoBtrfs: return -ENOMEM in clear_extent_bit
Chris Mason [Thu, 26 May 2011 21:43:59 +0000 (17:43 -0400)]
Btrfs: return -ENOMEM in clear_extent_bit

The btrfs releasepage function depends on ENOMEM coming
back when it is called atomic.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agoBtrfs: add mount -o auto_defrag
Chris Mason [Tue, 24 May 2011 19:35:30 +0000 (15:35 -0400)]
Btrfs: add mount -o auto_defrag

This will detect small random writes into files and
queue the up for an auto defrag process.  It isn't well suited to
database workloads yet, but works for smaller files such as rpm, sqlite
or bdb databases.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agoMerge branch 'cleanups_and_fixes' into inode_numbers
Chris Mason [Mon, 23 May 2011 18:37:47 +0000 (14:37 -0400)]
Merge branch 'cleanups_and_fixes' into inode_numbers

Conflicts:
fs/btrfs/tree-log.c
fs/btrfs/volumes.c

Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agoBtrfs: using rcu lock in the reader side of devices list
Xiao Guangrong [Wed, 20 Apr 2011 10:09:16 +0000 (10:09 +0000)]
Btrfs: using rcu lock in the reader side of devices list

fs_devices->devices is only updated on remove and add device paths, so we can
use rcu to protect it in the reader side

Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agoBtrfs: drop unnecessary device lock
Xiao Guangrong [Wed, 20 Apr 2011 10:08:47 +0000 (10:08 +0000)]
Btrfs: drop unnecessary device lock

Drop device_list_mutex for the reader side  on clone_fs_devices and
btrfs_rm_device pathes since the fs_info->volume_mutex can ensure the device
list is not updated

btrfs_close_extra_devices is the initialized path, we can not add or remove
device at this time, so we can simply drop the mutex safely, like other
initialized function does(add_missing_dev, __find_device, __btrfs_open_devices
...).

Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agoBtrfs: fix the race between remove dev and alloc chunk
Xiao Guangrong [Wed, 20 Apr 2011 10:08:16 +0000 (10:08 +0000)]
Btrfs: fix the race between remove dev and alloc chunk

On remove device path, it updates device->dev_alloc_list but does not hold
chunk lock

Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agoBtrfs: fix the race between reading and updating devices
Xiao Guangrong [Wed, 20 Apr 2011 10:07:30 +0000 (10:07 +0000)]
Btrfs: fix the race between reading and updating devices

On btrfs_congested_fn and __unplug_io_fn paths, we should hold
device_list_mutex to avoid remove/add device path to
update fs_devices->devices

On __btrfs_close_devices and btrfs_prepare_sprout paths, the devices in
fs_devices->devices or fs_devices->devices is updated, so we should hold
the mutex to avoid the reader side to reach them

Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agoBtrfs: fix bh leak on __btrfs_open_devices path
Xiao Guangrong [Wed, 20 Apr 2011 10:06:40 +0000 (10:06 +0000)]
Btrfs: fix bh leak on __btrfs_open_devices path

'bh' is forgot to release if no error is detected

Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agoBtrfs: fix unsafe usage of merge_state
Xiao Guangrong [Wed, 20 Apr 2011 06:45:49 +0000 (06:45 +0000)]
Btrfs: fix unsafe usage of merge_state

merge_state can free the current state if it can be merged with the next node,
but in set_extent_bit(), after merge_state, we still use the current extent to
get the next node and cache it into cached_state

Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agoBtrfs: allocate extent state and check the result properly
Xiao Guangrong [Wed, 20 Apr 2011 06:44:57 +0000 (06:44 +0000)]
Btrfs: allocate extent state and check the result properly

It doesn't allocate extent_state and check the result properly:
- in set_extent_bit, it doesn't allocate extent_state if the path is not
  allowed wait

- in clear_extent_bit, it doesn't check the result after atomic-ly allocate,
  we trigger BUG_ON() if it's fail

- if allocate fail, we trigger BUG_ON instead of returning -ENOMEM since
  the return value of clear_extent_bit() is ignored by many callers

Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agofs/btrfs: Add missing btrfs_free_path
Julia Lawall [Sat, 14 May 2011 07:10:51 +0000 (07:10 +0000)]
fs/btrfs: Add missing btrfs_free_path

Btrfs_alloc_path should be matched with btrfs_free_path in error-handling code.

A simplified version of the semantic match that finds this problem is as
follows: (http://coccinelle.lip6.fr/)

// <smpl>
@r exists@
local idexpression struct btrfs_path * x;
expression ra,rb;
position p1,p2;
@@

x = btrfs_alloc_path@p1(...)
...  when != btrfs_free_path(x,...)
     when != if (...) { ... btrfs_free_path(x,...) ...}
     when != x = ra
if(...) { ... when != x = rb
     when forall
     when != btrfs_free_path(x,...)
 \(return <+...x...+>; \| return@p2...; \) }

@script:python@
p1 << r.p1;
p2 << r.p2;
@@

cocci.print_main("alloc",p1)
cocci.print_secs("return",p2)
// </smpl>

Signed-off-by: Julia Lawall <julia@diku.dk>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agoBtrfs: check return value of btrfs_inc_extent_ref()
Tsutomu Itoh [Thu, 28 Apr 2011 09:18:21 +0000 (09:18 +0000)]
Btrfs: check return value of btrfs_inc_extent_ref()

If return value of btrfs_inc_extent_ref() is not 0, BUG() is called.

Signed-off-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agoBtrfs: return error to caller if read_one_inode() fails
Tsutomu Itoh [Thu, 28 Apr 2011 09:10:23 +0000 (09:10 +0000)]
Btrfs: return error to caller if read_one_inode() fails

When read_one_inode() fails, error code is returned to caller instead
of BUG_ON().

Signed-off-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agoBtrfs: BUG_ON is deleted from the caller of btrfs_truncate_item & btrfs_extend_item
Tsutomu Itoh [Thu, 19 May 2011 05:19:08 +0000 (05:19 +0000)]
Btrfs: BUG_ON is deleted from the caller of btrfs_truncate_item & btrfs_extend_item

Currently, btrfs_truncate_item and btrfs_extend_item returns only 0.
So, the check by BUG_ON in the caller is unnecessary.

Signed-off-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agoBtrfs: return error code to caller when btrfs_del_item fails
Tsutomu Itoh [Thu, 19 May 2011 04:37:44 +0000 (04:37 +0000)]
Btrfs: return error code to caller when btrfs_del_item fails

The error code is returned instead of calling BUG_ON when
btrfs_del_item returns the error.

Signed-off-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agoBtrfs: return error code to caller when btrfs_previous_item fails
Tsutomu Itoh [Thu, 19 May 2011 07:03:42 +0000 (07:03 +0000)]
Btrfs: return error code to caller when btrfs_previous_item fails

The error code is returned instead of calling BUG_ON when
btrfs_previous_item returns the error.

Signed-off-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agobtrfs: fix typo 'testeing' -> 'testing'
Sergei Trofimovich [Fri, 20 May 2011 20:20:32 +0000 (20:20 +0000)]
btrfs: fix typo 'testeing' -> 'testing'

Signed-off-by: Sergei Trofimovich <slyfox@gentoo.org>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agobtrfs: typo: 'btrfS' -> 'btrfs'
Sergei Trofimovich [Fri, 20 May 2011 20:20:31 +0000 (20:20 +0000)]
btrfs: typo: 'btrfS' -> 'btrfs'

Signed-off-by: Sergei Trofimovich <slyfox@gentoo.org>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agobtrfs: don't spin in shrink_delalloc if there is nothing to free
Sergei Trofimovich [Fri, 20 May 2011 20:20:30 +0000 (20:20 +0000)]
btrfs: don't spin in shrink_delalloc if there is nothing to free

Observed as a large delay when --mixed filesystem is filled up.
Test example:
1. create tiny --mixed FS:
   $ dd if=/dev/zero of=2G.img seek=$((2048 * 1024 * 1024 - 1)) count=1 bs=1
   $ mkfs.btrfs --mixed 2G.img
   $ mount -oloop 2G.img /mnt/ut/
2. Try to fill it up:
   $ dd if=/dev/urandom of=10M.file bs=10240 count=1024
   $ seq 1 256 | while read file_no; do echo $file_no; time cp 10M.file ${file_no}.copy; done

Up to '200.copy' it goes fast, but when disk fills-up each -ENOSPC
message takes 3 seconds to pop-up _every_ ENOSPC (and in usermode linux
it's even more: 30-60 seconds!). (Maybe, time depends on kernel's timer resolution).

No IO, no CPU load, just rescheduling. Some debugging revealed busy spinning
in shrink_delalloc.

Signed-off-by: Sergei Trofimovich <slyfox@gentoo.org>
Reviewed-by: Josef Bacik <josef@redhat.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agobtrfs: Delete unused version.sh script.
Jamey Sharp [Thu, 5 May 2011 19:03:46 +0000 (19:03 +0000)]
btrfs: Delete unused version.sh script.

In 2008, commit b4f6c45dfbf84f47c21f73f6370ad1292b0627fd dropped the use
of fs/btrfs/version.sh, but left the script behind. Kill it.

Commit by Jamey Sharp and Josh Triplett.

Signed-off-by: Jamey Sharp <jamey@minilop.net>
Signed-off-by: Josh Triplett <josh@joshtriplett.org>
Cc: Chris Mason <chris.mason@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agobtrfs: Ensure the tree search ioctl returns the right number of records
Hugo Mills [Sat, 14 May 2011 17:43:41 +0000 (17:43 +0000)]
btrfs: Ensure the tree search ioctl returns the right number of records

Btrfs's tree search ioctl has a field to indicate that no more than a
given number of records should be returned. The ioctl doesn't honour
this, as the tested value is not incremented until the end of the
copy_to_sk function. This patch removes an unnecessary local variable,
and updates the num_found counter as each key is found in the tree.

Signed-off-by: Hugo Mills <hugo@carfax.org.uk>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agoBTRFS: Remove unused node_lock
Andi Kleen [Wed, 18 May 2011 00:11:22 +0000 (00:11 +0000)]
BTRFS: Remove unused node_lock

240f62c8756 replaced the node_lock with rcu_read_lock, but forgot
to remove the actual lock in the data structure. Remove it here.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agoBtrfs: leave spinning on lookup and map the leaf
Josef Bacik [Tue, 17 May 2011 13:50:54 +0000 (09:50 -0400)]
Btrfs: leave spinning on lookup and map the leaf

On lookup we only want to read the inode item, so leave the path spinning.  Also
we're just wholesale reading the leaf off, so map the leaf so we don't do a
bunch of kmap/kunmaps.  Thanks,

Signed-off-by: Josef Bacik <josef@redhat.com>
12 years agoBtrfs: check for duplicate entries in the free space cache
Josef Bacik [Fri, 13 May 2011 18:49:23 +0000 (14:49 -0400)]
Btrfs: check for duplicate entries in the free space cache

If there are duplicate entries in the free space cache, discard the entire cache
and load it the old fashioned way.  Thanks,

Signed-off-by: Josef Bacik <josef@redhat.com>
12 years agoBtrfs: don't try to allocate from a block group that doesn't have enough space
Josef Bacik [Fri, 13 May 2011 15:07:12 +0000 (11:07 -0400)]
Btrfs: don't try to allocate from a block group that doesn't have enough space

If we have a very large filesystem, we can spend a lot of time in
find_free_extent just trying to allocate from empty block groups.  So instead
check to see if the block group even has enough space for the allocation, and if
not go on to the next block group.

Signed-off-by: Josef Bacik <josef@redhat.com>
12 years agoBtrfs: don't always do readahead
Josef Bacik [Fri, 13 May 2011 14:32:11 +0000 (10:32 -0400)]
Btrfs: don't always do readahead

Our readahead is sort of sloppy, and really isn't always needed.  For example if
ls is doing a stating ls (which is the default) it's going to stat in non-disk
order, so if say you have a directory with a stupid amount of files, readahead
is going to do nothing but waste time in the case of doing the stat.  Taking the
unconditional readahead out made my test go from 57 minutes to 36 minutes.  This
means that everywhere we do loop through the tree we want to make sure we do set
path->reada properly, so I went through and found all of the places where we
loop through the path and set reada to 1.  Thanks,

Signed-off-by: Josef Bacik <josef@redhat.com>
12 years agoBtrfs: try not to sleep as much when doing slow caching
Josef Bacik [Wed, 11 May 2011 21:30:53 +0000 (17:30 -0400)]
Btrfs: try not to sleep as much when doing slow caching

When the fs is super full and we unmount the fs, we could get stuck in this
thing where unmount is waiting for the caching kthread to make progress and the
caching kthread keeps scheduling because we're in the middle of a commit.  So
instead just let the caching kthread keep going and only yeild if
need_resched().  This makes my horrible umount case go from taking up to 10
minutes to taking less than 20 seconds.  Thanks,

Signed-off-by: Josef Bacik <josef@redhat.com>
12 years agoBtrfs: kill BTRFS_I(inode)->block_group
Josef Bacik [Wed, 11 May 2011 19:26:06 +0000 (15:26 -0400)]
Btrfs: kill BTRFS_I(inode)->block_group

Originally this was going to be used as a way to give hints to the allocator,
but frankly we can get much better hints elsewhere and it's not even used at all
for anything usefull.  In addition to be completely useless, when we initialize
an inode we try and find a freeish block group to set as the inodes block group,
and with a completely full 40gb fs this takes _forever_, so I imagine with say
1tb fs this is just unbearable.  So just axe the thing altoghether, we don't
need it and it saves us 8 bytes in the inode and saves us 500 microseconds per
inode lookup in my testcase.  Thanks,

Signed-off-by: Josef Bacik <josef@redhat.com>
12 years agoBtrfs: don't look at the extent buffer level 3 times in a row
Josef Bacik [Wed, 11 May 2011 16:25:37 +0000 (12:25 -0400)]
Btrfs: don't look at the extent buffer level 3 times in a row

We have a bit of debugging in btrfs_search_slot to make sure the level of the
cow block is the same as the original block we were cow'ing.  I don't think I've
ever seen this tripped, so kill it.  This saves us 2 kmap's per level in our
search.  Thanks,

Signed-off-by: Josef Bacik <josef@redhat.com>
12 years agoBtrfs: map the node block when looking for readahead targets
Josef Bacik [Wed, 11 May 2011 16:17:34 +0000 (12:17 -0400)]
Btrfs: map the node block when looking for readahead targets

If we have particularly full nodes, we could call btrfs_node_blockptr up to 32
times, which is 32 pairs of kmap/kunmap, which _sucks_.  So go ahead and map the
extent buffer while we look for readahead targets.  Thanks,

Signed-off-by: Josef Bacik <josef@redhat.com>
12 years agoBtrfs: set range_start to the right start in count_range_bits
Josef Bacik [Wed, 4 May 2011 15:11:17 +0000 (11:11 -0400)]
Btrfs: set range_start to the right start in count_range_bits

In count_range_bits we are adjusting total_bytes based on the range we are
searching for, but we don't adjust the range start according to the range we are
searching for, which makes for weird results.  For example, if the range

[0-8192]

is set DELALLOC, but I search for 4096-8192, I will get back 4096 for the number
of bytes found, but the range_start will be 0, which makes it look like the
range is [0-4096].  So instead set range_start = max(cur_start, state->start).
This makes everything come out right.  Thanks,

Signed-off-by: Josef Bacik <josef@redhat.com>
12 years agoBtrfs: fix how we do space reservation for truncate
Josef Bacik [Tue, 3 May 2011 14:40:22 +0000 (10:40 -0400)]
Btrfs: fix how we do space reservation for truncate

The ceph guys keep running into problems where we have space reserved in our
orphan block rsv when freeing it up.  This is because they tend to do snapshots
alot, so their truncates tend to use a bunch of space, so when we go to do
things like update the inode we have to steal reservation space in order to make
the reservation happen.  This happens because truncate can use as much space as
it freaking feels like, but we still have to hold space for removing the orphan
item and updating the inode, which will definitely always happen.  So in order
to fix this we need to split all of the reservation stuf up.  So with this patch
we have

1) The orphan block reserve which only holds the space for deleting our orphan
item when everything is over.

2) The truncate block reserve which gets allocated and used specifically for the
space that the truncate will use on a per truncate basis.

3) The transaction will always have 1 item's worth of data reserved so we can
update the inode normally.

Hopefully this will make the ceph problem go away.  Thanks,

Signed-off-by: Josef Bacik <josef@redhat.com>
12 years agoBtrfs: kill trans_mutex
Josef Bacik [Mon, 11 Apr 2011 21:25:13 +0000 (17:25 -0400)]
Btrfs: kill trans_mutex

We use trans_mutex for lots of things, here's a basic list

1) To serialize trans_handles joining the currently running transaction
2) To make sure that no new trans handles are started while we are committing
3) To protect the dead_roots list and the transaction lists

Really the serializing trans_handles joining is not too hard, and can really get
bogged down in acquiring a reference to the transaction.  So replace the
trans_mutex with a trans_lock spinlock and use it to do the following

1) Protect fs_info->running_transaction.  All trans handles have to do is check
this, and then take a reference of the transaction and keep on going.
2) Protect the fs_info->trans_list.  This doesn't get used too much, basically
it just holds the current transactions, which will usually just be the currently
committing transaction and the currently running transaction at most.
3) Protect the dead roots list.  This is only ever processed by splicing the
list so this is relatively simple.
4) Protect the fs_info->reloc_ctl stuff.  This is very lightweight and was using
the trans_mutex before, so this is a pretty straightforward change.
5) Protect fs_info->no_trans_join.  Because we don't hold the trans_lock over
the entirety of the commit we need to have a way to block new people from
creating a new transaction while we're doing our work.  So we set no_trans_join
and in join_transaction we test to see if that is set, and if it is we do a
wait_on_commit.
6) Make the transaction use count atomic so we don't need to take locks to
modify it when we're dropping references.
7) Add a commit_lock to the transaction to make sure multiple people trying to
commit the same transaction don't race and commit at the same time.
8) Make open_ioctl_trans an atomic so we don't have to take any locks for ioctl
trans.

I have tested this with xfstests, but obviously it is a pretty hairy change so
lots of testing is greatly appreciated.  Thanks,

Signed-off-by: Josef Bacik <josef@redhat.com>
12 years agoBtrfs: if we've already started a trans handle, use that one
Josef Bacik [Wed, 13 Apr 2011 19:15:59 +0000 (15:15 -0400)]
Btrfs: if we've already started a trans handle, use that one

We currently track trans handles in current->journal_info, but we don't actually
use it.  This patch fixes it.  This will cover the case where we have multiple
people starting transactions down the call chain.  This keeps us from having to
allocate a new handle and all of that, we just increase the use count of the
current handle, save the old block_rsv, and return.  I tested this with xfstests
and it worked out fine.  Thanks,

Signed-off-by: Josef Bacik <josef@redhat.com>
12 years agoBtrfs: take away the num_items argument from btrfs_join_transaction
Josef Bacik [Wed, 13 Apr 2011 16:54:33 +0000 (12:54 -0400)]
Btrfs: take away the num_items argument from btrfs_join_transaction

I keep forgetting that btrfs_join_transaction() just ignores the num_items
argument, which leads me to sending pointless patches and looking stupid :).  So
just kill the num_items argument from btrfs_join_transaction and
btrfs_start_ioctl_transaction, since neither of them use it.  Thanks,

Signed-off-by: Josef Bacik <josef@redhat.com>
12 years agoBtrfs: make sure to use the delalloc reserve when filling delalloc
Josef Bacik [Wed, 13 Apr 2011 16:02:53 +0000 (12:02 -0400)]
Btrfs: make sure to use the delalloc reserve when filling delalloc

In the prealloc filling code and compressed code we don't set trans->block_rsv
to the delalloc block reserve properly, which is going to make us use metadata
from the wrong pool, this patch fixes that.  Thanks,

Signed-off-by: Josef Bacik <josef@redhat.com>
12 years agoBtrfs: do not flush csum items of unchanged file data during treelog
liubo [Fri, 6 May 2011 02:36:09 +0000 (10:36 +0800)]
Btrfs: do not flush csum items of unchanged file data during treelog

The current code relogs the entire inode every time during fsync log,
and it is much better suited to small files rather than large ones.

During my performance test, the fsync performace of large files sucks,
and we can ascribe this to the tremendous amount of csum infos of the
large ones, cause we have to flush all of these csum infos into log trees
even when there are only _one_ change in the whole file data.  Apparently,
to optimize fsync, we need to create a filter to skip the unnecessary csum
ones, that is, the corresponding file data remains unchanged before this fsync.

Here I have some test results to show, I use sysbench to do "random write + fsync".

===
sysbench --test=fileio --num-threads=1 --file-num=2 --file-block-size=4K --file-total-size=8G --file-test-mode=rndwr --file-io-mode=sync --file-extra-flags=  [prepare, run]
===

Sysbench args:
  - Number of threads: 1
  - Extra file open flags: 0
  - 2 files, 4Gb each
  - Block size 4Kb
  - Number of random requests for random IO: 10000
  - Read/Write ratio for combined random IO test: 1.50
  - Periodic FSYNC enabled, calling fsync() each 100 requests.
  - Calling fsync() at the end of test, Enabled.
  - Using synchronous I/O mode
  - Doing random write test

Sysbench results:
===
   Operations performed:  0 Read, 10000 Write, 200 Other = 10200 Total
   Read 0b  Written 39.062Mb  Total transferred 39.062Mb
===
a) without patch:  (*SPEED* : 451.01Kb/sec)
   112.75 Requests/sec executed

b) with patch:     (*SPEED* : 4.7533Mb/sec)
   1216.84 Requests/sec executed

PS: I've made a _sub transid_ stuff patch, but it does not perform as effectively as this patch,
and I'm wanderring where the problem is and trying to improve it more.

Signed-off-by: Liu Bo <liubo2009@cn.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agoMerge branch 'for-chris' of git://git.kernel.org/pub/scm/linux/kernel/git/arne/btrfs...
Chris Mason [Mon, 23 May 2011 10:30:52 +0000 (06:30 -0400)]
Merge branch 'for-chris' of git://git./linux/kernel/git/arne/btrfs-unstable-arne into inode_numbers

Conflicts:
fs/btrfs/Makefile
fs/btrfs/ctree.h
fs/btrfs/volumes.h

Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agoMerge branch 'allocator' of git://git.kernel.org/pub/scm/linux/kernel/git/arne/btrfs...
Chris Mason [Sun, 22 May 2011 16:36:34 +0000 (12:36 -0400)]
Merge branch 'allocator' of git://git./linux/kernel/git/arne/btrfs-unstable-arne into inode_numbers

Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agoMerge branch 'cleanups' of git://repo.or.cz/linux-2.6/btrfs-unstable into inode_numbers
Chris Mason [Sun, 22 May 2011 16:33:42 +0000 (12:33 -0400)]
Merge branch 'cleanups' of git://repo.or.cz/linux-2.6/btrfs-unstable into inode_numbers

Conflicts:
fs/btrfs/extent-tree.c
fs/btrfs/free-space-cache.c
fs/btrfs/inode.c
fs/btrfs/tree-log.c

Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agoBtrfs: update the delayed inode code to use the btrfs_ino helper.
Chris Mason [Sun, 22 May 2011 11:11:22 +0000 (07:11 -0400)]
Btrfs: update the delayed inode code to use the btrfs_ino helper.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agoMerge branch 'delayed_inode' into inode_numbers
Chris Mason [Sun, 22 May 2011 11:07:01 +0000 (07:07 -0400)]
Merge branch 'delayed_inode' into inode_numbers

Conflicts:
fs/btrfs/inode.c
fs/btrfs/ioctl.c
fs/btrfs/transaction.c

Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agobtrfs: implement delayed inode items operation
Miao Xie [Fri, 22 Apr 2011 10:12:22 +0000 (18:12 +0800)]
btrfs: implement delayed inode items operation

Changelog V5 -> V6:
- Fix oom when the memory load is high, by storing the delayed nodes into the
  root's radix tree, and letting btrfs inodes go.

Changelog V4 -> V5:
- Fix the race on adding the delayed node to the inode, which is spotted by
  Chris Mason.
- Merge Chris Mason's incremental patch into this patch.
- Fix deadlock between readdir() and memory fault, which is reported by
  Itaru Kitayama.

Changelog V3 -> V4:
- Fix nested lock, which is reported by Itaru Kitayama, by updating space cache
  inode in time.

Changelog V2 -> V3:
- Fix the race between the delayed worker and the task which does delayed items
  balance, which is reported by Tsutomu Itoh.
- Modify the patch address David Sterba's comment.
- Fix the bug of the cpu recursion spinlock, reported by Chris Mason

Changelog V1 -> V2:
- break up the global rb-tree, use a list to manage the delayed nodes,
  which is created for every directory and file, and used to manage the
  delayed directory name index items and the delayed inode item.
- introduce a worker to deal with the delayed nodes.

Compare with Ext3/4, the performance of file creation and deletion on btrfs
is very poor. the reason is that btrfs must do a lot of b+ tree insertions,
such as inode item, directory name item, directory name index and so on.

If we can do some delayed b+ tree insertion or deletion, we can improve the
performance, so we made this patch which implemented delayed directory name
index insertion/deletion and delayed inode update.

Implementation:
- introduce a delayed root object into the filesystem, that use two lists to
  manage the delayed nodes which are created for every file/directory.
  One is used to manage all the delayed nodes that have delayed items. And the
  other is used to manage the delayed nodes which is waiting to be dealt with
  by the work thread.
- Every delayed node has two rb-tree, one is used to manage the directory name
  index which is going to be inserted into b+ tree, and the other is used to
  manage the directory name index which is going to be deleted from b+ tree.
- introduce a worker to deal with the delayed operation. This worker is used
  to deal with the works of the delayed directory name index items insertion
  and deletion and the delayed inode update.
  When the delayed items is beyond the lower limit, we create works for some
  delayed nodes and insert them into the work queue of the worker, and then
  go back.
  When the delayed items is beyond the upper bound, we create works for all
  the delayed nodes that haven't been dealt with, and insert them into the work
  queue of the worker, and then wait for that the untreated items is below some
  threshold value.
- When we want to insert a directory name index into b+ tree, we just add the
  information into the delayed inserting rb-tree.
  And then we check the number of the delayed items and do delayed items
  balance. (The balance policy is above.)
- When we want to delete a directory name index from the b+ tree, we search it
  in the inserting rb-tree at first. If we look it up, just drop it. If not,
  add the key of it into the delayed deleting rb-tree.
  Similar to the delayed inserting rb-tree, we also check the number of the
  delayed items and do delayed items balance.
  (The same to inserting manipulation)
- When we want to update the metadata of some inode, we cached the data of the
  inode into the delayed node. the worker will flush it into the b+ tree after
  dealing with the delayed insertion and deletion.
- We will move the delayed node to the tail of the list after we access the
  delayed node, By this way, we can cache more delayed items and merge more
  inode updates.
- If we want to commit transaction, we will deal with all the delayed node.
- the delayed node will be freed when we free the btrfs inode.
- Before we log the inode items, we commit all the directory name index items
  and the delayed inode update.

I did a quick test by the benchmark tool[1] and found we can improve the
performance of file creation by ~15%, and file deletion by ~20%.

Before applying this patch:
Create files:
        Total files: 50000
        Total time: 1.096108
        Average time: 0.000022
Delete files:
        Total files: 50000
        Total time: 1.510403
        Average time: 0.000030

After applying this patch:
Create files:
        Total files: 50000
        Total time: 0.932899
        Average time: 0.000019
Delete files:
        Total files: 50000
        Total time: 1.215732
        Average time: 0.000024

[1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3

Many thanks for Kitayama-san's help!

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Reviewed-by: David Sterba <dave@jikos.cz>
Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com>
Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agoMerge branch 'ino-alloc' of git://repo.or.cz/linux-btrfs-devel into inode_numbers
Chris Mason [Sat, 21 May 2011 13:27:38 +0000 (09:27 -0400)]
Merge branch 'ino-alloc' of git://repo.or.cz/linux-btrfs-devel into inode_numbers

Conflicts:
fs/btrfs/free-space-cache.c

Signed-off-by: Chris Mason <chris.mason@oracle.com>
12 years agoLinux 2.6.39 v2.6.39
Linus Torvalds [Thu, 19 May 2011 04:06:34 +0000 (21:06 -0700)]
Linux 2.6.39

12 years agoMerge branch 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jlbec/ocfs2
Linus Torvalds [Wed, 18 May 2011 23:50:28 +0000 (16:50 -0700)]
Merge branch 'fixes' of git://git./linux/kernel/git/jlbec/ocfs2

* 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jlbec/ocfs2:
  configfs: Fix race between configfs_readdir() and configfs_d_iput()
  configfs: Don't try to d_delete() negative dentries.
  ocfs2/dlm: Target node death during resource migration leads to thread spin
  ocfs2: Skip mount recovery for hard-ro mounts
  ocfs2/cluster: Heartbeat mismatch message improved
  ocfs2/cluster: Increase the live threshold for global heartbeat
  ocfs2/dlm: Use negotiated o2dlm protocol version
  ocfs2: skip existing hole when removing the last extent_rec in punching-hole codes.
  ocfs2: Initialize data_ac (might be used uninitialized)

12 years agoMerge branch 'devicetree/merge' of git://git.secretlab.ca/git/linux-2.6
Linus Torvalds [Wed, 18 May 2011 20:25:57 +0000 (13:25 -0700)]
Merge branch 'devicetree/merge' of git://git.secretlab.ca/git/linux-2.6

* 'devicetree/merge' of git://git.secretlab.ca/git/linux-2.6:
  drivercore: revert addition of of_match to struct device
  of: fix race when matching drivers

12 years agoMerge branch 'upstream' of git://git.linux-mips.org/pub/scm/upstream-linus
Linus Torvalds [Wed, 18 May 2011 20:21:43 +0000 (13:21 -0700)]
Merge branch 'upstream' of git://git.linux-mips.org/upstream-linus

* 'upstream' of git://git.linux-mips.org/pub/scm/upstream-linus:
  MIPS: Kludge IP27 build for 2.6.39.
  MIPS: AR7: Fix GPIO register size for Titan variant.
  MIPS: Fix duplicate invocation of notify_die.
  MIPS: RB532: Fix iomap resource size miscalculation.

12 years agodrivercore: revert addition of of_match to struct device
Grant Likely [Wed, 18 May 2011 17:19:24 +0000 (11:19 -0600)]
drivercore: revert addition of of_match to struct device

Commit b826291c, "drivercore/dt: add a match table pointer to struct
device" added an of_match pointer to struct device to cache the
of_match_table entry discovered at driver match time.  This was unsafe
because matching is not an atomic operation with probing a driver.  If
two or more drivers are attempted to be matched to a driver at the
same time, then the cached matching entry pointer could get
overwritten.

This patch reverts the of_match cache pointer and reworks all users to
call of_match_device() directly instead.

Signed-off-by: Grant Likely <grant.likely@secretlab.ca>
12 years agoof: fix race when matching drivers
Milton Miller [Wed, 18 May 2011 15:27:39 +0000 (10:27 -0500)]
of: fix race when matching drivers

If two drivers are probing devices at the same time, both will write
their match table result to the dev->of_match cache at the same time.

Only write the result if the device matches.

In a thread titled "SBus devices sometimes detected, sometimes not",
Meelis reported his SBus hme was not detected about 50% of the time.
From the debug suggested by Grant it was obvious another driver matched
some devices between the call to match the hme and the hme discovery
failling.

Reported-by: Meelis Roos <mroos@linux.ee>
Signed-off-by: Milton Miller <miltonm@bga.com>
[grant.likely: modified to only call of_match_device() once]
Signed-off-by: Grant Likely <grant.likely@secretlab.ca>
12 years agoMerge branch 'for-linus' of git://git.kernel.dk/linux-2.6-block
Linus Torvalds [Wed, 18 May 2011 13:49:02 +0000 (06:49 -0700)]
Merge branch 'for-linus' of git://git.kernel.dk/linux-2.6-block

* 'for-linus' of git://git.kernel.dk/linux-2.6-block:
  block: don't delay blk_run_queue_async
  scsi: remove performance regression due to async queue run
  blk-throttle: Use task_subsys_state() to determine a task's blkio_cgroup
  block: rescan partitions on invalidated devices on -ENOMEDIA too
  cdrom: always check_disk_change() on open
  block: unexport DISK_EVENT_MEDIA_CHANGE for legacy/fringe drivers

12 years agoMIPS: Kludge IP27 build for 2.6.39.
Ralf Baechle [Wed, 18 May 2011 12:14:36 +0000 (13:14 +0100)]
MIPS: Kludge IP27 build for 2.6.39.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
12 years agoMIPS: AR7: Fix GPIO register size for Titan variant.
Florian Fainelli [Fri, 13 May 2011 15:41:21 +0000 (17:41 +0200)]
MIPS: AR7: Fix GPIO register size for Titan variant.

The 'size' variable contains the correct register size for both AR7
and Titan, but we never used it to ioremap the correct register size.
This problem only shows up on Titan.

[ralf@linux-mips.org: Fixed the fix.  The original patch as in patchwork
recognizes the problem correctly then fails to fix it ...]

Reported-by: Alexander Clouter <alex@digriz.org.uk>
Signed-off-by: Florian Fainelli <florian@openwrt.org>
Patchwork: https://patchwork.linux-mips.org/patch/2380/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
12 years agoMIPS: Fix duplicate invocation of notify_die.
Ralf Baechle [Fri, 13 May 2011 09:33:28 +0000 (10:33 +0100)]
MIPS: Fix duplicate invocation of notify_die.

Initial patch by Yury Polyanskiy <ypolyans@princeton.edu>.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Patchwork: https://patchwork.linux-mips.org/patch/2373/

12 years agoMIPS: RB532: Fix iomap resource size miscalculation.
Ralf Baechle [Thu, 12 May 2011 12:55:48 +0000 (13:55 +0100)]
MIPS: RB532: Fix iomap resource size miscalculation.

This is the MIPS portion of Joe Perches <joe@perches.com>'s
https://patchwork.linux-mips.org/patch/2172/ which seems to have been
lost in time and space.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
12 years agoconfigfs: Fix race between configfs_readdir() and configfs_d_iput()
Joel Becker [Wed, 18 May 2011 11:08:16 +0000 (04:08 -0700)]
configfs: Fix race between configfs_readdir() and configfs_d_iput()

configfs_readdir() will use the existing inode numbers of inodes in the
dcache, but it makes them up for attribute files that aren't currently
instantiated.  There is a race where a closing attribute file can be
tearing down at the same time as configfs_readdir() is trying to get its
inode number.

We want to get the inode number of open attribute files, because they
should match while instantiated.  We can't lock down the transition
where dentry->d_inode is set to NULL, so we just check for NULL there.
We can, however, ensure that an inode we find isn't iput() in
configfs_d_iput() until after we've accessed it.

Signed-off-by: Joel Becker <jlbec@evilplan.org>
12 years agoconfigfs: Don't try to d_delete() negative dentries.
Joel Becker [Tue, 22 Feb 2011 09:09:49 +0000 (01:09 -0800)]
configfs: Don't try to d_delete() negative dentries.

When configfs is faking mkdir() on its subsystem or default group
objects, it starts by adding a negative dentry.  It then tries to
instantiate the group.  If that should fail, it must clean up after
itself.

I was using d_delete() here, but configfs_attach_group() promises to
return an empty dentry on error.  d_delete() explodes with the entry
dentry.  Let's try d_drop() instead.  The unhashing is what we want for
our dentry.

Signed-off-by: Joel Becker <jlbec@evilplan.org>
12 years agoblock: don't delay blk_run_queue_async
Shaohua Li [Wed, 18 May 2011 09:22:43 +0000 (11:22 +0200)]
block: don't delay blk_run_queue_async

Let's check a scenario:
1. blk_delay_queue(q, SCSI_QUEUE_DELAY);
2. blk_run_queue_async();
the second one will became a noop, because q->delay_work already has
WORK_STRUCT_PENDING_BIT set, so the delayed work will still run after
SCSI_QUEUE_DELAY. But blk_run_queue_async actually hopes the delayed
work runs immediately.

Fix this by doing a cancel on potentially pending delayed work
before queuing an immediate run of the workqueue.

Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
12 years agoMerge branch 'v4l_for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mchehab...
Linus Torvalds [Wed, 18 May 2011 10:16:38 +0000 (03:16 -0700)]
Merge branch 'v4l_for_linus' of git://git./linux/kernel/git/mchehab/linux-2.6

* 'v4l_for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mchehab/linux-2.6:
  [media] V4L: soc-camera: regression fix: calculate .sizeimage in soc_camera.c
  [media] v4l2-subdev: fix broken subdev control enumeration
  [media] Fix cx88 remote control input
  [media] v4l: Release module if subdev registration fails

12 years agoMerge branch 'x86-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Wed, 18 May 2011 10:14:34 +0000 (03:14 -0700)]
Merge branch 'x86-fixes-for-linus' of git://git./linux/kernel/git/tip/linux-2.6-tip

* 'x86-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
  x86, AMD: Fix ARAT feature setting again
  Revert "x86, AMD: Fix APIC timer erratum 400 affecting K8 Rev.A-E processors"
  x86, apic: Fix spurious error interrupts triggering on all non-boot APs
  x86, mce, AMD: Fix leaving freed data in a list
  x86: Fix UV BAU for non-consecutive nasids
  x86, UV: Fix NMI handler for UV platforms

12 years agoMerge branch 'perf-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Wed, 18 May 2011 10:13:46 +0000 (03:13 -0700)]
Merge branch 'perf-fixes-for-linus' of git://git./linux/kernel/git/tip/linux-2.6-tip

* 'perf-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
  perf evlist: Fix per thread mmap setup
  perf tools: Honour the cpu list parameter when also monitoring a thread list
  kprobes, x86: Disable irqs during optimized callback

12 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/sfrench/cifs-2.6
Linus Torvalds [Wed, 18 May 2011 10:13:11 +0000 (03:13 -0700)]
Merge git://git./linux/kernel/git/sfrench/cifs-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/sfrench/cifs-2.6:
  cifs: fix cifsConvertToUCS() for the mapchars case
  cifs: add fallback in is_path_accessible for old servers

12 years agoprocfs: add stub for proc_mkdir_mode()
Randy Dunlap [Tue, 17 May 2011 22:44:12 +0000 (15:44 -0700)]
procfs: add stub for proc_mkdir_mode()

Provide a stub for proc_mkdir_mode() when CONFIG_PROC_FS is not
enabled, just like the stub for proc_mkdir().

Fixes this linux-next build error:

  drivers/net/wireless/airo.c:4504: error: implicit declaration of function 'proc_mkdir_mode'

Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: "John W. Linville" <linville@tuxdriver.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoum: fix abort
Richard Weinberger [Tue, 17 May 2011 22:44:11 +0000 (15:44 -0700)]
um: fix abort

os_dump_core() uses abort() to terminate UML in case of an fatal error.

glibc's abort() calls raise(SIGABRT) which makes use of tgkill().
tgkill() has no effect within UML's kernel threads because they are not
pthreads.  As fallback abort() executes an invalid instruction to
terminate the process.  Therefore UML gets killed by SIGSEGV and leaves a
ugly log entry in the host's kernel ring buffer.

To get rid of this we use our own abort routine.

Signed-off-by: Richard Weinberger <richard@nod.at>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agomemcg: fix zone congestion
KAMEZAWA Hiroyuki [Tue, 17 May 2011 22:44:10 +0000 (15:44 -0700)]
memcg: fix zone congestion

ZONE_CONGESTED should be a state of global memory reclaim.  If not, a busy
memcg sets this and give unnecessary throttoling in wait_iff_congested()
against memory recalim in other contexts.  This makes system performance
bad.

I'll think about "memcg is congested!" flag is required or not, later.
But this fix is required first.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Acked-by: Ying Han <yinghan@google.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Johannes Weiner <jweiner@redhat.com>
Cc: Michal Hocko <mhocko@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agodrivers/leds/leds-lm3530.c: add MODULE_DEVICE_TABLE
Axel Lin [Tue, 17 May 2011 22:44:09 +0000 (15:44 -0700)]
drivers/leds/leds-lm3530.c: add MODULE_DEVICE_TABLE

Adding the necessary MODULE_DEVICE_TABLE() information allows the driver
to be automatically loaded by udev.

Signed-off-by: Axel Lin <axel.lin@gmail.com>
Cc: Shreshtha Kumar SAHU <shreshthakumar.sahu@stericsson.com>
Cc: Richard Purdie <rpurdie@rpsys.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agorapidio: fix default routing initialization
Alexandre Bounine [Tue, 17 May 2011 22:44:08 +0000 (15:44 -0700)]
rapidio: fix default routing initialization

Fix switch initialization to ensure that all switches have default routing
disabled.  This guarantees that no unexpected RapidIO packets arrive to
the default port set by reset and there is no default routing destination
until it is properly configured by software.

This update also unifies handling of unmapped destinations by tsi57x, IDT
Gen1 and IDT Gen2 switches.

Signed-off-by: Alexandre Bounine <alexandre.bounine@idt.com>
Cc: Kumar Gala <galak@kernel.crashing.org>
Cc: Matt Porter <mporter@kernel.crashing.org>
Cc: Li Yang <leoli@freescale.com>
Cc: Thomas Moll <thomas.moll@sysgo.com>
Cc: <stable@kernel.org> [2.6.37+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agocifs: fix cifsConvertToUCS() for the mapchars case
Jeff Layton [Tue, 17 May 2011 19:28:21 +0000 (15:28 -0400)]
cifs: fix cifsConvertToUCS() for the mapchars case

As Metze pointed out, commit 84cdf74e broke mapchars option:

    Commit "cifs: fix unaligned accesses in cifsConvertToUCS"
    (84cdf74e8096a10dd6acbb870dd404b92f07a756) does multiple steps
    in just one commit (moving the function and changing it without
    testing).

    put_unaligned_le16(temp, &target[j]); is never called for any
    codepoint the goes via the 'default' switch statement. As a result
    we put just zero (or maybe uninitialized) bytes into the target
    buffer.

His proposed patch looks correct, but doesn't apply to the current head
of the tree. This patch should also fix it.

Cc: <stable@kernel.org> # .38.x: 581ade4: cifs: clean up various nits in unicode routines (try #2)
Reported-by: Stefan Metzmacher <metze@samba.org>
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Steve French <sfrench@us.ibm.com>
12 years agocifs: add fallback in is_path_accessible for old servers
Jeff Layton [Tue, 17 May 2011 10:40:30 +0000 (06:40 -0400)]
cifs: add fallback in is_path_accessible for old servers

The is_path_accessible check uses a QPathInfo call, which isn't
supported by ancient win9x era servers. Fall back to an older
SMBQueryInfo call if it fails with the magic error codes.

Cc: stable@kernel.org
Reported-and-Tested-by: Sandro Bonazzola <sandro.bonazzola@gmail.com>
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Steve French <sfrench@us.ibm.com>
12 years agoMerge branch 'timers-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Tue, 17 May 2011 15:02:04 +0000 (08:02 -0700)]
Merge branch 'timers-fixes-for-linus' of git://git./linux/kernel/git/tip/linux-2.6-tip

* 'timers-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
  tick: Clear broadcast active bit when switching to oneshot
  rtc: mc13xxx: Don't call rtc_device_register while holding lock
  rtc: rp5c01: Initialize drvdata before registering device
  rtc: pcap: Initialize drvdata before registering device
  rtc: msm6242: Initialize drvdata before registering device
  rtc: max8998: Initialize drvdata before registering device
  rtc: max8925: Initialize drvdata before registering device
  rtc: m41t80: Initialize clientdata before registering device
  rtc: ds1286: Initialize drvdata before registering device
  rtc: ep93xx: Initialize drvdata before registering device
  rtc: davinci: Initialize drvdata before registering device
  rtc: mxc: Initialize drvdata before registering device
  clocksource: Install completely before selecting

12 years agox86, AMD: Fix ARAT feature setting again
Borislav Petkov [Tue, 17 May 2011 12:55:19 +0000 (14:55 +0200)]
x86, AMD: Fix ARAT feature setting again

Trying to enable the local APIC timer on early K8 revisions
uncovers a number of other issues with it, in conjunction with
the C1E enter path on AMD. Fixing those causes much more churn
and troubles than the benefit of using that timer brings so
don't enable it on K8 at all, falling back to the original
functionality the kernel had wrt to that.

Reported-and-bisected-by: Nick Bowler <nbowler@elliptictech.com>
Cc: Boris Ostrovsky <Boris.Ostrovsky@amd.com>
Cc: Andreas Herrmann <andreas.herrmann3@amd.com>
Cc: Greg Kroah-Hartman <greg@kroah.com>
Cc: Hans Rosenfeld <hans.rosenfeld@amd.com>
Cc: Nick Bowler <nbowler@elliptictech.com>
Cc: Joerg-Volker-Peetz <jvpeetz@web.de>
Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
Link: http://lkml.kernel.org/r/1305636919-31165-3-git-send-email-bp@amd64.org
Signed-off-by: Ingo Molnar <mingo@elte.hu>
12 years agoRevert "x86, AMD: Fix APIC timer erratum 400 affecting K8 Rev.A-E processors"
Borislav Petkov [Tue, 17 May 2011 12:55:18 +0000 (14:55 +0200)]
Revert "x86, AMD: Fix APIC timer erratum 400 affecting K8 Rev.A-E processors"

This reverts commit e20a2d205c05cef6b5783df339a7d54adeb50962, as it crashes
certain boxes with specific AMD CPU models.

Moving the lower endpoint of the Erratum 400 check to accomodate
earlier K8 revisions (A-E) opens a can of worms which is simply
not worth to fix properly by tweaking the errata checking
framework:

* missing IntPenging MSR on revisions < CG cause #GP:

http://marc.info/?l=linux-kernel&m=130541471818831

* makes earlier revisions use the LAPIC timer instead of the C1E
idle routine which switches to HPET, thus not waking up in
deeper C-states:

http://lkml.org/lkml/2011/4/24/20

Therefore, leave the original boundary starting with K8-revF.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
12 years agoscsi: remove performance regression due to async queue run
Jens Axboe [Tue, 17 May 2011 09:04:44 +0000 (11:04 +0200)]
scsi: remove performance regression due to async queue run

Commit c21e6beb removed our queue request_fn re-enter
protection, and defaulted to always running the queues from
kblockd to be safe. This was a known potential slow down,
but should be safe.

Unfortunately this is causing big performance regressions for
some, so we need to improve this logic. Looking into the details
of the re-enter, the real issue is on requeue of requests.

Requeue of requests upon seeing a BUSY condition from the device
ends up re-running the queue, causing traces like this:

scsi_request_fn()
        scsi_dispatch_cmd()
                scsi_queue_insert()
                        __scsi_queue_insert()
                                scsi_run_queue()
scsi_request_fn()
...

potentially causing the issue we want to avoid. So special
case the requeue re-run of the queue, but improve it to offload
the entire run of local queue and starved queue from a single
workqueue callback. This is a lot better than potentially
kicking off a workqueue run for each device seen.

This also fixes the issue of the local device going into recursion,
since the above mentioned commit never moved that queue run out
of line.

Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
12 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
Linus Torvalds [Tue, 17 May 2011 01:38:08 +0000 (18:38 -0700)]
Merge git://git./linux/kernel/git/davem/net-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6:
  net: Change netdev_fix_features messages loglevel
  vmxnet3: Fix inconsistent LRO state after initialization
  sfc: Fix oops in register dump after mapping change
  IPVS: fix netns if reading ip_vs_* procfs entries
  bridge: fix forwarding of IPv6

12 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/cjb/mmc
Linus Torvalds [Tue, 17 May 2011 01:36:47 +0000 (18:36 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/cjb/mmc

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/cjb/mmc:
  Revert "mmc: fix a race between card-detect rescan and clock-gate work instances"

12 years agomm: fix kernel-doc warning in page_alloc.c
Randy Dunlap [Mon, 16 May 2011 20:16:54 +0000 (13:16 -0700)]
mm: fix kernel-doc warning in page_alloc.c

Fix new kernel-doc warning in mm/page_alloc.c:

  Warning(mm/page_alloc.c:2370): No description found for parameter 'nid'

Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoPCI: Clear bridge resource flags if requested size is 0
Yinghai Lu [Sat, 14 May 2011 01:06:17 +0000 (18:06 -0700)]
PCI: Clear bridge resource flags if requested size is 0

During pci remove/rescan testing found:

  pci 0000:c0:03.0: PCI bridge to [bus c4-c9]
  pci 0000:c0:03.0:   bridge window [io  0x1000-0x0fff]
  pci 0000:c0:03.0:   bridge window [mem 0xf0000000-0xf00fffff]
  pci 0000:c0:03.0:   bridge window [mem 0xfc180000000-0xfc197ffffff 64bit pref]
  pci 0000:c0:03.0: device not available (can't reserve [io  0x1000-0x0fff])
  pci 0000:c0:03.0: Error enabling bridge (-22), continuing
  pci 0000:c0:03.0: enabling bus mastering
  pci 0000:c0:03.0: setting latency timer to 64
  pcieport 0000:c0:03.0: device not available (can't reserve [io  0x1000-0x0fff])
  pcieport: probe of 0000:c0:03.0 failed with error -22

This bug was caused by commit c8adf9a3e873 ("PCI: pre-allocate
additional resources to devices only after successful allocation of
essential resources.")

After that commit, pci_hotplug_io_size is changed to additional_io_size
from minium size.  So it will not go through resource_size(res) != 0
path, and will not be reset.

The root cause is: pci_bridge_check_ranges will set RESOURCE_IO flag for
pci bridge, and later if children do not need IO resource.  those bridge
resources will not need to be allocated.  but flags is still there.
that will confuse the the pci_enable_bridges later.

related code:

   static void assign_requested_resources_sorted(struct resource_list *head,
                                    struct resource_list_x *fail_head)
   {
           struct resource *res;
           struct resource_list *list;
           int idx;

           for (list = head->next; list; list = list->next) {
                   res = list->res;
                   idx = res - &list->dev->resource[0];
                   if (resource_size(res) && pci_assign_resource(list->dev, idx)) {
   ...
                           reset_resource(res);
                   }
           }
   }

At last, We have to clear the flags in pbus_size_mem/io when requested
size == 0 and !add_head.  becasue this case it will not go through
adjust_resources_sorted().

Just make size1 = size0 when !add_head. it will make flags get cleared.

At the same time when requested size == 0, add_size != 0, will still
have in head and add_list.  because we do not clear the flags for it.

After this, we will get right result:

  pci 0000:c0:03.0: PCI bridge to [bus c4-c9]
  pci 0000:c0:03.0:   bridge window [io  disabled]
  pci 0000:c0:03.0:   bridge window [mem 0xf0000000-0xf00fffff]
  pci 0000:c0:03.0:   bridge window [mem 0xfc180000000-0xfc197ffffff 64bit pref]
  pci 0000:c0:03.0: enabling bus mastering
  pci 0000:c0:03.0: setting latency timer to 64
  pcieport 0000:c0:03.0: setting latency timer to 64
  pcieport 0000:c0:03.0: irq 160 for MSI/MSI-X
  pcieport 0000:c0:03.0: Signaling PME through PCIe PME interrupt
  pci 0000:c4:00.0: Signaling PME through PCIe PME interrupt
  pcie_pme 0000:c0:03.0:pcie01: service driver pcie_pme loaded
  aer 0000:c0:03.0:pcie02: service driver aer loaded
  pciehp 0000:c0:03.0:pcie04: Hotplug Controller:

v3: more simple fix. also fix one typo in pbus_size_mem

Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Reviewed-by: Ram Pai <linuxram@us.ibm.com>
Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
Cc: Bjorn Helgaas <bhelgaas@google.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agotick: Clear broadcast active bit when switching to oneshot
Thomas Gleixner [Mon, 16 May 2011 09:07:48 +0000 (11:07 +0200)]
tick: Clear broadcast active bit when switching to oneshot

The first cpu which switches from periodic to oneshot mode switches
also the broadcast device into oneshot mode. The broadcast device
serves as a backup for per cpu timers which stop in deeper
C-states. To avoid starvation of the cpus which might be in idle and
depend on broadcast mode it marks the other cpus as broadcast active
and sets the brodcast expiry value of those cpus to the next tick.

The oneshot mode broadcast bit for the other cpus is sticky and gets
only cleared when those cpus exit idle. If a cpu was not idle while
the bit got set in consequence the bit prevents that the broadcast
device is armed on behalf of that cpu when it enters idle for the
first time after it switched to oneshot mode.

In most cases that goes unnoticed as one of the other cpus has usually
a timer pending which keeps the broadcast device armed with a short
timeout. Now if the only cpu which has a short timer active has the
bit set then the broadcast device will not be armed on behalf of that
cpu and will fire way after the expected timer expiry. In the case of
Christians bug report it took ~145 seconds which is about half of the
wrap around time of HPET (the limit for that device) due to the fact
that all other cpus had no timers armed which expired before the 145
seconds timeframe.

The solution is simply to clear the broadcast active bit
unconditionally when a cpu switches to oneshot mode after the first
cpu switched the broadcast device over. It's not idle at that point
otherwise it would not be executing that code.

[ I fundamentally hate that broadcast crap. Why the heck thought some
  folks that when going into deep idle it's a brilliant concept to
  switch off the last device which brings the cpu back from that
  state? ]

Thanks to Christian for providing all the valuable debug information!

Reported-and-tested-by: Christian Hoffmann <email@christianhoffmann.info>
Cc: John Stultz <johnstul@us.ibm.com>
Link: http://lkml.kernel.org/r/%3Calpine.LFD.2.02.1105161105170.3078%40ionos%3E
Cc: stable@kernel.org
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
12 years agonet: Change netdev_fix_features messages loglevel
Michał Mirosław [Mon, 16 May 2011 19:14:21 +0000 (15:14 -0400)]
net: Change netdev_fix_features messages loglevel

Those reduced to DEBUG can possibly be triggered by unprivileged processes
and are nothing exceptional. Illegal checksum combinations can only be
caused by driver bug, so promote those messages to WARN.

Since GSO without SG will now only cause DEBUG message from
netdev_fix_features(), remove the workaround from register_netdevice().

Signed-off-by: Michał Mirosław <mirq-linux@rere.qmqm.pl>
Signed-off-by: David S. Miller <davem@davemloft.net>
12 years agovmxnet3: Fix inconsistent LRO state after initialization
Thomas Jarosch [Mon, 16 May 2011 06:28:15 +0000 (06:28 +0000)]
vmxnet3: Fix inconsistent LRO state after initialization

During initialization of vmxnet3, the state of LRO
gets out of sync with netdev->features.

This leads to very poor TCP performance in a IP forwarding
setup and is hitting many VMware users.

Simplified call sequence:
1. vmxnet3_declare_features() initializes "adapter->lro" to true.

2. The kernel automatically disables LRO if IP forwarding is enabled,
so vmxnet3_set_flags() gets called. This also updates netdev->features.

3. Now vmxnet3_setup_driver_shared() is called. "adapter->lro" is still
set to true and LRO gets enabled again, even though
netdev->features shows it's disabled.

Fix it by updating "adapter->lro", too.

The private vmxnet3 adapter flags are scheduled for removal
in net-next, see commit a0d2730c9571aeba793cb5d3009094ee1d8fda35
"net: vmxnet3: convert to hw_features".

Patch applies to 2.6.37 / 2.6.38 and 2.6.39-rc6.

Please CC: comments.

Signed-off-by: Thomas Jarosch <thomas.jarosch@intra2net.com>
Acked-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
12 years agosfc: Fix oops in register dump after mapping change
Ben Hutchings [Mon, 16 May 2011 06:13:49 +0000 (06:13 +0000)]
sfc: Fix oops in register dump after mapping change

Commit 747df2258b1b9a2e25929ef496262c339c380009 ('sfc: Always map MCDI
shared memory as uncacheable') introduced a separate mapping for the
MCDI shared memory (MC_TREG_SMEM).  This means we can no longer easily
include it in the register dump.  Since it is not particularly useful
in debugging, substitute a recognisable dummy value.

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
12 years agoMerge branch 'omap-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Mon, 16 May 2011 15:55:49 +0000 (08:55 -0700)]
Merge branch 'omap-fixes-for-linus' of git://git./linux/kernel/git/tmlind/linux-omap-2.6

* 'omap-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tmlind/linux-omap-2.6:
  OMAP3: set the core dpll clk rate in its set_rate function
  omap: iommu: Return IRQ_HANDLED in fault handler when no fault occured

12 years agoMerge branch 'drm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/airlied...
Linus Torvalds [Mon, 16 May 2011 15:47:31 +0000 (08:47 -0700)]
Merge branch 'drm-fixes' of git://git./linux/kernel/git/airlied/drm-2.6

* 'drm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/airlied/drm-2.6:
  drm: Take lock around probes for drm_fb_helper_hotplug_event
  drm/i915: Revert i915.semaphore=1 default from 47ae63e0
  vga_switcheroo: don't toggle-switch devices
  drm/radeon/kms: add some evergreen/ni safe regs
  drm/radeon/kms: fix extended lvds info parsing
  drm/radeon/kms: fix tiling reg on fusion

12 years agoRevert "mmc: fix a race between card-detect rescan and clock-gate work instances"
Chris Ball [Mon, 16 May 2011 15:32:26 +0000 (11:32 -0400)]
Revert "mmc: fix a race between card-detect rescan and clock-gate work instances"

This reverts commit 26fc8775b51484d8c0a671198639c6d5ae60533e, which has
been reported to cause boot/resume-time crashes for some users:

https://bbs.archlinux.org/viewtopic.php?id=118751.

Signed-off-by: Chris Ball <cjb@laptop.org>
Cc: <stable@kernel.org>
12 years agoblk-throttle: Use task_subsys_state() to determine a task's blkio_cgroup
Vivek Goyal [Mon, 16 May 2011 13:24:08 +0000 (15:24 +0200)]
blk-throttle: Use task_subsys_state() to determine a task's blkio_cgroup

Currentlly we first map the task to cgroup and then cgroup to
blkio_cgroup. There is a more direct way to get to blkio_cgroup
from task using task_subsys_state(). Use that.

The real reason for the fix is that it also avoids a race in generic
cgroup code. During remount/umount rebind_subsystems() is called and
it can do following with and rcu protection.

cgrp->subsys[i] = NULL;

That means if somebody got hold of cgroup under rcu and then it tried
to do cgroup->subsys[] to get to blkio_cgroup, it would get NULL which
is wrong. I was running into this race condition with ltp running on a
upstream derived kernel and that lead to crash.

So ideally we should also fix cgroup generic code to wait for rcu
grace period before setting pointer to NULL. Li Zefan is not very keen
on introducing synchronize_wait() as he thinks it will slow
down moun/remount/umount operations.

So for the time being atleast fix the kernel crash by taking a more
direct route to blkio_cgroup.

One tester had reported a crash while running LTP on a derived kernel
and with this fix crash is no more seen while the test has been
running for over 6 days.

Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Reviewed-by: Li Zefan <lizf@cn.fujitsu.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>