platform/kernel/linux-starfive.git
10 years agoBtrfs: move the missing device to its own fs device list
Miao Xie [Wed, 3 Sep 2014 13:35:46 +0000 (21:35 +0800)]
Btrfs: move the missing device to its own fs device list

For a missing device, we don't know it belong to which fs before we read its
fsid from the chunk tree. So we add them into the current fs device list at first.
When we get its fsid, we should move them to their own fs device list.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: stop mounting the fs if the non-ENOENT errors happen when opening seed fs
Miao Xie [Wed, 3 Sep 2014 13:35:45 +0000 (21:35 +0800)]
Btrfs: stop mounting the fs if the non-ENOENT errors happen when opening seed fs

When we open a seed filesystem, if the degraded mount option is set, we continue to
mount the fs if we don't find some devices in the seed filesystem. But we should stop
mounting if other errors happen. Fix it

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: make the logic of source device removing more clear
Miao Xie [Wed, 3 Sep 2014 13:35:44 +0000 (21:35 +0800)]
Btrfs: make the logic of source device removing more clear

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: fix use-after-free problem of the device during device replace
Miao Xie [Wed, 3 Sep 2014 13:35:43 +0000 (21:35 +0800)]
Btrfs: fix use-after-free problem of the device during device replace

The problem is:
Task0(device scan task) Task1(device replace task)
scan_one_device()
mutex_lock(&uuid_mutex)
device = find_device()
mutex_lock(&device_list_mutex)
lock_chunk()
rm_and_free_source_device
unlock_chunk()
mutex_unlock(&device_list_mutex)
check device

Destroying the target device if device replace fails also has the same problem.

We fix this problem by locking uuid_mutex during destroying source device or
target device, just like the device remove operation.

It is a temporary solution, we can fix this problem and make the code more
clear by atomic counter in the future.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: fix unprotected device list access when cloning fs devices
Miao Xie [Wed, 3 Sep 2014 13:35:42 +0000 (21:35 +0800)]
Btrfs: fix unprotected device list access when cloning fs devices

We can build a new filesystem based a seed filesystem, and we need clone
the fs devices when we open the new filesystem. But someone might clear
the seed flag of the seed filesystem, then mount that filesystem and
remove some device. If we mount the new filesystem, we might access
a device list which was being changed when we clone the fs devices.
Fix it.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: Fix misuse of chunk mutex
Miao Xie [Wed, 3 Sep 2014 13:35:41 +0000 (21:35 +0800)]
Btrfs: Fix misuse of chunk mutex

There were several problems about chunk mutex usage:
- Lock chunk mutex when updating metadata. It would cause the nested
  deadlock because updating metadata might need allocate new chunks
  that need acquire chunk mutex. We remove chunk mutex at this case,
  because b-tree lock and other lock mechanism can help us.
- ABBA deadlock occured between device_list_mutex and chunk_mutex.
  When we update device status, we must acquire device_list_mutex at the
  beginning, and then we might get chunk_mutex during the device status
  update because we need allocate new chunks for metadata COW. But at
  most place, we acquire chunk_mutex at first and then acquire device list
  mutex. We need change the lock order.
- Some place we needn't acquire chunk_mutex. For example we needn't get
  chunk_mutex when we free a empty seed fs_devices structure.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: fix unprotected device list access when getting the fs information
Miao Xie [Wed, 3 Sep 2014 13:35:40 +0000 (21:35 +0800)]
Btrfs: fix unprotected device list access when getting the fs information

When we get the fs information, we forgot to acquire the mutex of device list,
it might cause the problem we might access a device that was removed. Fix
it by acquiring the device list mutex.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: fix unprotected system chunk array insertion
Miao Xie [Wed, 3 Sep 2014 13:35:39 +0000 (21:35 +0800)]
Btrfs: fix unprotected system chunk array insertion

We didn't protect the system chunk array when we added a new
system chunk into it, it would cause the array be corrupted
if someone remove/add some system chunk into array at the same
time. Fix it by chunk lock.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: fix unprotected device's variants on 32bits machine
Miao Xie [Wed, 3 Sep 2014 13:35:38 +0000 (21:35 +0800)]
Btrfs: fix unprotected device's variants on 32bits machine

->total_bytes,->disk_total_bytes,->bytes_used is protected by chunk
lock when we change them, but sometimes we read them without any lock,
and we might get unexpected value. We fix this problem like inode's
i_size.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: update free_chunk_space during allocting a new chunk
Miao Xie [Wed, 3 Sep 2014 13:35:37 +0000 (21:35 +0800)]
Btrfs: update free_chunk_space during allocting a new chunk

We should update free_chunk_space in time when we allocate a new chunk,
not when we deal with the pending device update and block group insertion,
because we need the real free_chunk_space data to calculate the reserved
space, if we don't update it in time, we would consider the disk space which
has be allocated as free space, and would use it to do overcommit reservation.
Fix it.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: fix unprotected device->bytes_used update
Miao Xie [Wed, 3 Sep 2014 13:35:36 +0000 (21:35 +0800)]
Btrfs: fix unprotected device->bytes_used update

We should update device->bytes_used in the lock context of
chunk_mutex, or we would get wrong data.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: Fix wrong free_chunk_space assignment during removing a device
Miao Xie [Wed, 3 Sep 2014 13:35:35 +0000 (21:35 +0800)]
Btrfs: Fix wrong free_chunk_space assignment during removing a device

During removing a device, we have modified free_chunk_space when we
shrink the device, so we needn't assign a new value to it after
the device shrink. Fix it.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: fix wrong device bytes_used in the super block
Miao Xie [Wed, 3 Sep 2014 13:35:34 +0000 (21:35 +0800)]
Btrfs: fix wrong device bytes_used in the super block

device->bytes_used will be changed when allocating a new chunk, and
disk_total_size will be changed if resizing is successful.
Meanwhile, the on-disk super blocks of the previous transaction
might not be updated. Considering the consistency of the metadata
in the previous transaction, We should use the size in the previous
transaction to check if the super block is beyond the boundary
of the device.

Though it is not big problem because we don't use it now, but anyway
it is better that we make it be consistent with the common metadata,
maybe we will use it in the future.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: fix wrong disk size when writing super blocks
Miao Xie [Wed, 3 Sep 2014 13:35:33 +0000 (21:35 +0800)]
Btrfs: fix wrong disk size when writing super blocks

total_size will be changed when resizing a device, and disk_total_size
will be changed if resizing is successful. Meanwhile, the on-disk super
blocks of the previous transaction might not be updated. Considering
the consistency of the metadata in the previous transaction, We should
use the size in the previous transaction to check if the super block is
beyond the boundary of the device. Fix it.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: fix unprotected assignment of the target device
Miao Xie [Wed, 3 Sep 2014 13:35:32 +0000 (21:35 +0800)]
Btrfs: fix unprotected assignment of the target device

We didn't protect the assignment of the target device, it might cause the
problem that the super block update was skipped because we might find wrong
size of the target device during the assignment. Fix it by moving the
assignment sentences into the initialization function of the target device.
And there is another merit that we can check if the target device is suitable
more early.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: cleanup double assignment of device->bytes_used when device replace finishes
Miao Xie [Wed, 3 Sep 2014 13:35:31 +0000 (21:35 +0800)]
Btrfs: cleanup double assignment of device->bytes_used when device replace finishes

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: cleanup unused num_can_discard in fs_devices
Miao Xie [Wed, 3 Sep 2014 13:35:30 +0000 (21:35 +0800)]
Btrfs: cleanup unused num_can_discard in fs_devices

The member variants - num_can_discard - of fs_devices structure
are set, but no one use them to do anything. so remove them.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agobtrfs: remove the wrong comments
Li RongQing [Mon, 8 Sep 2014 12:41:09 +0000 (20:41 +0800)]
btrfs: remove the wrong comments

This comments became wrong after c3c532[bdi: add helper function for
doing init and register of a bdi for a file system], so remove them.

Signed-off-by: Li RongQing <roy.qing.li@gmail.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: fix directory recovery from fsync log
Filipe Manana [Mon, 8 Sep 2014 21:53:18 +0000 (22:53 +0100)]
Btrfs: fix directory recovery from fsync log

When replaying a directory from the fsync log, if a directory entry
exists both in the fs/subvol tree and in the log, the directory's inode
got its i_size updated incorrectly, accounting for the dentry's name
twice.

Reproducer, from a test for xfstests:

    _scratch_mkfs >> $seqres.full 2>&1
    _init_flakey
    _mount_flakey

    touch $SCRATCH_MNT/foo
    sync

    touch $SCRATCH_MNT/bar
    xfs_io -c "fsync" $SCRATCH_MNT
    xfs_io -c "fsync" $SCRATCH_MNT/bar

    _load_flakey_table $FLAKEY_DROP_WRITES
    _unmount_flakey

    _load_flakey_table $FLAKEY_ALLOW_WRITES
    _mount_flakey

    [ -f $SCRATCH_MNT/foo ] || echo "file foo is missing"
    [ -f $SCRATCH_MNT/bar ] || echo "file bar is missing"

    _unmount_flakey
    _check_scratch_fs $FLAKEY_DEV

The filesystem check at the end failed with the message:
"root 5 root dir 256 error".

A test case for xfstests follows.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: fix loop writing of async reclaim
Liu Bo [Wed, 10 Sep 2014 04:58:50 +0000 (12:58 +0800)]
Btrfs: fix loop writing of async reclaim

One of my tests shows that when we really don't have space to reclaim via
flush_space and also run out of space, this async reclaim work loops on adding
itself into the workqueue and keeps writing something to disk according to
iostat's results, and these writes mainly comes from commit_transaction which
writes super_block.  This's unacceptable as it can be bad to disks, especially
memeory storages.

This adds a check to avoid the above situation.

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: make fiemap not blow when you have lots of snapshots
Josef Bacik [Wed, 10 Sep 2014 20:20:45 +0000 (16:20 -0400)]
Btrfs: make fiemap not blow when you have lots of snapshots

We have been iterating all references for each extent we have in a file when we
do fiemap to see if it is shared.  This is fine when you have a few clones or a
few snapshots, but when you have 5k snapshots suddenly fiemap just sits there
and stares at you.  So add btrfs_check_shared which will use the backref walking
code but will short circuit as soon as it finds a root or inode that doesn't
match the one we currently have.  This makes fiemap on my testbox go from
looking at me blankly for a day to spitting out actual output in a reasonable
amount of time.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: add missing compression property remove in btrfs_ioctl_setflags
Filipe Manana [Thu, 11 Sep 2014 10:44:49 +0000 (11:44 +0100)]
Btrfs: add missing compression property remove in btrfs_ioctl_setflags

The behaviour of a 'chattr -c' consists of getting the current flags,
clearing the FS_COMPR_FL bit and then sending the result to the set
flags ioctl - this means the bit FS_NOCOMP_FL isn't set in the flags
passed to the ioctl. This results in the compression property not being
cleared from the inode - it was cleared only if the bit FS_NOCOMP_FL
was set in the received flags.

Reproducer:

    $ mkfs.btrfs -f /dev/sdd
    $ mount /dev/sdd /mnt && cd /mnt
    $ mkdir a
    $ chattr +c a
    $ touch a/file
    $ lsattr a/file
    --------c------- a/file
    $ chattr -c a
    $ touch a/file2
    $ lsattr a/file2
    --------c------- a/file2
    $ lsattr -d a
    ---------------- a

Reported-by: Andreas Schneider <asn@cryptomilk.org>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agobtrfs: Fix a deadlock in btrfs_dev_replace_finishing()
Qu Wenruo [Wed, 20 Aug 2014 08:10:15 +0000 (16:10 +0800)]
btrfs: Fix a deadlock in btrfs_dev_replace_finishing()

btrfs-transacion:5657
[stack snip]
btrfs_bio_map()
    btrfs_bio_counter_inc_blocked()
        percpu_counter_inc(&fs_info->bio_counter)  ###bio_counter > 0(A)
        __btrfs_bio_map()
            btrfs_dev_replace_lock()
                mutex_lock(dev_replace->lock)    ###wait mutex(B)

btrfs:32612
[stack snip]
btrfs_dev_replace_start()
    btrfs_dev_replace_lock()
mutex_lock(dev_replace->lock)    ###hold mutex(B)
    btrfs_dev_replace_finishing()
        btrfs_rm_dev_replace_blocked()
            wait until percpu_counter_sum == 0    ###wait on bio_counter(A)

This bug can be triggered quite easily by the following test script:
http://pastebin.com/MQmb37Cy

This patch will fix the ABBA problem by calling
btrfs_dev_replace_unlock() before btrfs_rm_dev_replace_blocked().

The consistency of btrfs devices list and their superblocks is protected
by device_list_mutex, not btrfs_dev_replace_lock/unlock().
So it is safe the move btrfs_dev_replace_unlock() before
btrfs_rm_dev_replace_blocked().

Reported-by: Zhao Lei <zhaolei@cn.fujitsu.com>
Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Cc: Stefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: cleanup the same name in end_bio_extent_readpage
Liu Bo [Tue, 19 Aug 2014 15:32:22 +0000 (23:32 +0800)]
Btrfs: cleanup the same name in end_bio_extent_readpage

We've defined a 'offset' out of bio_for_each_segment_all.

This is just a clean rename, no function changes.

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agobtrfs: don't go readonly on existing qgroup items
Mark Fasheh [Mon, 18 Aug 2014 21:01:17 +0000 (14:01 -0700)]
btrfs: don't go readonly on existing qgroup items

btrfs_drop_snapshot() leaves subvolume qgroup items on disk after
completion. This can cause problems with snapshot creation. If a new
snapshot tries to claim the deleted subvolumes id, btrfs will get -EEXIST
from add_qgroup_item() and go read-only. The following commands will
reproduce this problem (assume btrfs is on /dev/sda and is mounted at
/btrfs)

mkfs.btrfs -f /dev/sda
mount -t btrfs /dev/sda /btrfs/
btrfs quota enable /btrfs/
btrfs su sna /btrfs/ /btrfs/snap
btrfs su de /btrfs/snap
sleep 45
umount /btrfs/
mount -t btrfs /dev/sda /btrfs/

We can fix this by catching -EEXIST in add_qgroup_item() and
initializing the existing items. We have the problem of orphaned
relation items being on disk from an old snapshot but that is outside
the scope of this patch.

Signed-off-by: Mark Fasheh <mfasheh@suse.de>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: show real function name in btrfs workqueue tracepoint
Liu Bo [Fri, 15 Aug 2014 15:38:06 +0000 (23:38 +0800)]
Btrfs: show real function name in btrfs workqueue tracepoint

Use %pf instead of %p, just same as kernel workqueue tracepoints.

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: shrink further sizeof(struct extent_buffer)
Filipe Manana [Thu, 14 Aug 2014 11:25:14 +0000 (12:25 +0100)]
Btrfs: shrink further sizeof(struct extent_buffer)

The map_start and map_len fields aren't used anywhere, so just remove
them. On a x86_64 system, this reduced sizeof(struct extent_buffer)
from 296 bytes to 280 bytes, and therefore 14 extent_buffer structs can
now fit into a page instead of 13.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: send, lower mem requirements for processing xattrs
Filipe Manana [Wed, 20 Aug 2014 09:45:45 +0000 (10:45 +0100)]
Btrfs: send, lower mem requirements for processing xattrs

Maximum xattr size can be up to nearly the leaf size. For an fs with a
leaf size larger than the page size, using kmalloc requires allocating
multiple pages that are contiguous, which might not be possible if
there's heavy memory fragmentation. Therefore fallback to vmalloc if
we fail to allocate with kmalloc. Also start with a smaller buffer size,
since xattr values typically are smaller than a page.

Reported-by: Chris Murphy <lists@colorremedies.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agobtrfs: remove stale define after removing ordered operations
David Sterba [Wed, 20 Aug 2014 15:34:22 +0000 (17:34 +0200)]
btrfs: remove stale define after removing ordered operations

Last user removed in commit "btrfs: disable strict file flushes for
renames and truncates" (8d875f95da43c6a8f18f77869f2ef26e9594fecc).

Signed-off-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: improve free space cache management and space allocation
Filipe Manana [Fri, 29 Aug 2014 12:35:13 +0000 (13:35 +0100)]
Btrfs: improve free space cache management and space allocation

While under random IO, a block group's free space cache eventually reaches
a state where it has a mix of extent entries and bitmap entries representing
free space regions.

As later free space regions are returned to the cache, some of them are merged
with existing extent entries if they are contiguous with them. But others are
not merged, because despite the existence of adjacent free space regions in
the cache, the merging doesn't happen because the existing free space regions
are represented in bitmap extents. Even when new free space regions are merged
with existing extent entries (enlarging the free space range they represent),
we create chances of having after an enlarged region that is contiguous with
some other region represented in a bitmap entry.

Both clustered and non-clustered space allocation work by iterating over our
extent and bitmap entries and skipping any that represents a region smaller
then the allocation request (and giving preference to extent entries before
bitmap entries). By having a contiguous free space region that is represented
by 2 (or more) entries (mix of extent and bitmap entries), we end up not
satisfying an allocation request with a size larger than the size of any of
the entries but no larger than the sum of their sizes. Making the caller assume
we're under a ENOSPC condition or force it to allocate multiple smaller space
regions (as we do for file data writes), which adds extra overhead and more
chances of causing fragmentation due to the smaller regions being all spread
apart from each other (more likely when under concurrency).

For example, if we have the following in the cache:

* extent entry representing free space range: [128Mb - 256Kb, 128Mb[

* bitmap entry covering the range [128Mb, 256Mb[, but only with the bits
  representing the range [128Mb, 128Mb + 768Kb[ set - that is, only that
  space in this 128Mb area is marked as free

An allocation request for 1Mb, starting at offset not greater than 128Mb - 256Kb,
would fail before, despite the existence of such contiguous free space area in the
cache. The caller could only allocate up to 768Kb of space at once and later another
256Kb (or vice-versa). In between each smaller allocation request, another task
working on a different file/inode might come in and take that space, preventing the
former task of getting a contiguous 1Mb region of free space.

Therefore this change implements the ability to move free space from bitmap
entries into existing and new free space regions represented with extent
entries. This is done when a space region is added to the cache.

A test was added to the sanity tests that explains in detail the issue too.

Some performance test results with compilebench on a 4 cores machine, with
32Gb of ram and using an HDD follow.

Test: compilebench -D /mnt -i 30 -r 1000 --makej

Before this change:

   intial create total runs 30 avg 69.02 MB/s (user 0.28s sys 0.57s)
   compile total runs 30 avg 314.96 MB/s (user 0.12s sys 0.25s)
   read compiled tree total runs 3 avg 27.14 MB/s (user 1.52s sys 0.90s)
   delete compiled tree total runs 30 avg 3.14 seconds (user 0.15s sys 0.66s)

After this change:

   intial create total runs 30 avg 68.37 MB/s (user 0.29s sys 0.55s)
   compile total runs 30 avg 382.83 MB/s (user 0.12s sys 0.24s)
   read compiled tree total runs 3 avg 27.82 MB/s (user 1.45s sys 0.97s)
   delete compiled tree total runs 30 avg 3.18 seconds (user 0.17s sys 0.65s)

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agobtrfs: rename total_bytes to avoid confusion
Anand Jain [Wed, 20 Aug 2014 02:54:17 +0000 (10:54 +0800)]
btrfs: rename total_bytes to avoid confusion

we are assigning number_devices to the total_bytes,
that's very confusing for a moment

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agobtrfs: fix typo in the log message
Anand Jain [Wed, 13 Aug 2014 06:24:25 +0000 (14:24 +0800)]
btrfs: fix typo in the log message

there is no matching open parenthesis for the closing parenthesis

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agobtrfs: rw_devices shouldn't be incremented for seed fs in btrfs_rm_dev_replace_srcdev()
Anand Jain [Wed, 13 Aug 2014 06:24:24 +0000 (14:24 +0800)]
btrfs: rw_devices shouldn't be incremented for seed fs in btrfs_rm_dev_replace_srcdev()

seed fs devices don't participate as rw_device, so don't increment
rw_devices when the device being handled belongs to a seed fs.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agobtrfs: fix memory leak when there is no more seed device
Anand Jain [Wed, 13 Aug 2014 06:24:23 +0000 (14:24 +0800)]
btrfs: fix memory leak when there is no more seed device

When we replace all the seed device in the system there is
no point in just keeping the btrfs_fs_devices with out
any device

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agobtrfs: update sprout seed pointer when seed fs is relinquished
Anand Jain [Wed, 13 Aug 2014 06:24:22 +0000 (14:24 +0800)]
btrfs: update sprout seed pointer when seed fs is relinquished

We are not updating sprout fs seed pointer when all seed device
is replaced. This patch will check if all seed device has been
replaced and then update the sprout pointer accordingly.

Same reproducer as in the previous patch would apply here.
And notice that btrfs_close_device will check if seed fs is
present and spits out the error with out this patch.

int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
{
::
                seed_devices = fs_devices->seed;
::
        while (seed_devices) {
                fs_devices = seed_devices;
                seed_devices = fs_devices->seed;
                __btrfs_close_devices(fs_devices);
                free_fs_devices(fs_devices);
        }

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agobtrfs: fix rw_devices miss match after seed replace
Anand Jain [Wed, 13 Aug 2014 06:24:21 +0000 (14:24 +0800)]
btrfs: fix rw_devices miss match after seed replace

reproducer:
    reproducer:
    mount /dev/sdb /btrfs
    btrfs dev add /dev/sdc /btrfs
    btrfs rep start -B /dev/sdb /dev/sdd /btrfs
    umount /btrfs

WARNING: CPU: 0 PID: 3882 at fs/btrfs/volumes.c:892 __btrfs_close_devices+0x1c8/0x200 [btrfs]()

which is

        WARN_ON(fs_devices->rw_devices);

   The problem here is that we did not add one to the rw_devices when
   we replace the seed device with a writable device.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agobtrfs: replace seed device followed by unmount causes kernel WARNING
Anand Jain [Wed, 20 Aug 2014 02:56:56 +0000 (10:56 +0800)]
btrfs: replace seed device followed by unmount causes kernel WARNING

reproducer:
mount /dev/sdb /btrfs
btrfs dev add /dev/sdc /btrfs
btrfs rep start -B /dev/sdb /dev/sdd /btrfs
umount /btrfs

WARNING: CPU: 0 PID: 12661 at fs/btrfs/volumes.c:891 __btrfs_close_devices+0x1b0/0x200 [btrfs]()
::

__btrfs_close_devices()
::
        WARN_ON(fs_devices->open_devices);

After the seed device has been replaced the new target device
is no more a seed device. So we need to update the device
numbers in the fs_devices as pointed by the fs_info.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agobtrfs: preparatory to make btrfs_rm_dev_replace_srcdev() seed aware
Anand Jain [Wed, 13 Aug 2014 06:24:19 +0000 (14:24 +0800)]
btrfs: preparatory to make btrfs_rm_dev_replace_srcdev() seed aware

There is no logical change in this patch, just a preparatory patch,
so that changes can be easily reasoned.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agobtrfs: Drop stray check of fixup_workers creation
Andrey Utkin [Sat, 9 Aug 2014 11:51:15 +0000 (14:51 +0300)]
btrfs: Drop stray check of fixup_workers creation

The issue was introduced in a79b7d4b3e8118f265dcb4bdf9a572c392f02708,
adding allocation of extent_workers, so this stray check is surely not
meant to be a check of something else.

Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=82021
Reported-by: Maks Naumov <maksqwe1@ukr.net>
Signed-off-by: Andrey Utkin <andrey.krieger.utkin@gmail.com>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: make btrfs_search_forward return with nodes unlocked
Filipe Manana [Mon, 4 Aug 2014 18:37:21 +0000 (19:37 +0100)]
Btrfs: make btrfs_search_forward return with nodes unlocked

None of the uses of btrfs_search_forward() need to have the path
nodes (level >= 1) read locked, only the leaf needs to be locked
while the caller processes it. Therefore make it return a path
with all nodes unlocked, except for the leaf.

This change is motivated by the observation that during a file
fsync we repeatdly call btrfs_search_forward() and process the
returned leaf while upper nodes of the returned path (level >= 1)
are read locked, which unnecessarily blocks other tasks that want
to write to the same fs/subvol btree.
Therefore instead of modifying the fsync code to unlock all nodes
with level >= 1 immediately after calling btrfs_search_forward(),
change btrfs_search_forward() to do it, so that it benefits all
callers.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agobtrfs: sysfs label interface should check for read only FS
Anand Jain [Wed, 30 Jul 2014 12:04:10 +0000 (20:04 +0800)]
btrfs: sysfs label interface should check for read only FS

Not sure how this escaped many eyes so far

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agobtrfs: code optimize: BTRFS_ATTR_RW could set the mode
Anand Jain [Wed, 30 Jul 2014 12:04:09 +0000 (20:04 +0800)]
btrfs: code optimize: BTRFS_ATTR_RW could set the mode

BTRFS_ATTR_RW could set the mode and be inline with BTRFS_ATTR

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agobtrfs: code optimize: BTRFS_ATTR could handle the mode
Anand Jain [Wed, 30 Jul 2014 12:04:08 +0000 (20:04 +0800)]
btrfs: code optimize: BTRFS_ATTR could handle the mode

All that uses BTRFS_ATTR want mode to be set at 0444 so just do
it at the define.  And few spacing alignments.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agobtrfs: use BTRFS_ATTR instead of btrfs_no_store()
Anand Jain [Wed, 30 Jul 2014 12:04:07 +0000 (20:04 +0800)]
btrfs: use BTRFS_ATTR instead of btrfs_no_store()

we have BTRFS_ATTR define to create sysfs RO file, use that.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: avoid unnecessary switch of path locks to blocking mode
Filipe Manana [Mon, 28 Jul 2014 18:37:17 +0000 (19:37 +0100)]
Btrfs: avoid unnecessary switch of path locks to blocking mode

If we need to cow a node, increase the write lock level and retry the
tree search, there's no point of changing the node locks in our path
to blocking mode, as we only waste time and unnecessarily wake up other
tasks waiting on the spinning locks (just to block them again shortly
after) because we release our path before repeating the tree search.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: unlock nodes earlier when inserting items in a btree
Filipe Manana [Mon, 28 Jul 2014 18:34:35 +0000 (19:34 +0100)]
Btrfs: unlock nodes earlier when inserting items in a btree

In ctree.c:setup_items_for_insert(), we can unlock all nodes in our
path before we process the leaf (shift items and data, adjust data
offsets, etc). This allows for better btree concurrency, as we're
often holding a write lock on at least the node at level 1.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agobtrfs: use IS_ALIGNED() for assertion in btrfs_lookup_csums_range() for simplicity
Satoru Takeuchi [Fri, 25 Jul 2014 08:07:48 +0000 (17:07 +0900)]
btrfs: use IS_ALIGNED() for assertion in btrfs_lookup_csums_range() for simplicity

btrfs_lookup_csums_range() uses ALIGN() to check if "start"
and "end + 1" are aligned to "root->sectorsize". It's better to
replace these with IS_ALIGNED() for simplicity.

Signed-off-by: Satoru Takeuchi <takeuchi_satoru@jp.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: cleanup for btrfs workqueue tracepoints
Liu Bo [Tue, 12 Aug 2014 08:33:18 +0000 (16:33 +0800)]
Btrfs: cleanup for btrfs workqueue tracepoints

Tracepoint trace_btrfs_normal_work_done never has an user, just cleanup it.

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: add work_struct information for workqueue tracepoint
Liu Bo [Tue, 12 Aug 2014 08:33:17 +0000 (16:33 +0800)]
Btrfs: add work_struct information for workqueue tracepoint

Kernel workqueue's tracepoints print the address of work_struct, while btrfs
workqueue's tracepoints print the address of btrfs_work.

We need a connection between this two, for example when debuging, we usually
grep an address in the trace output.  So it'd be better to also print
work_struct in btrfs workqueue's tracepoint.

Please note that we can only add this into those tracepoints whose work is still
available in memory because we need to reference the work.

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agobtrfs: add trace for qgroup accounting
Mark Fasheh [Thu, 17 Jul 2014 19:39:00 +0000 (12:39 -0700)]
btrfs: add trace for qgroup accounting

We want this to debug qgroup changes on live systems.

Signed-off-by: Mark Fasheh <mfasheh@suse.de>
Reviewed-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: cleanup unused latest_devid and latest_trans in fs_devices
Miao Xie [Thu, 24 Jul 2014 03:37:15 +0000 (11:37 +0800)]
Btrfs: cleanup unused latest_devid and latest_trans in fs_devices

The member variants - latest_devid and latest_trans - of fs_devices structure
are set, but no one use them to do anything. so remove them.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: update the comment of total_bytes and disk_total_bytes of btrfs_devie
Miao Xie [Thu, 24 Jul 2014 03:37:12 +0000 (11:37 +0800)]
Btrfs: update the comment of total_bytes and disk_total_bytes of btrfs_devie

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: Fix the problem that the dirty flag of dev stats is cleared
Miao Xie [Thu, 24 Jul 2014 03:37:11 +0000 (11:37 +0800)]
Btrfs: Fix the problem that the dirty flag of dev stats is cleared

The io error might happen during writing out the device stats, and the
device stats information and dirty flag would be update at that time,
but the current code didn't consider this case, just clear the dirty
flag, it would cause that we forgot to write out the new device stats
information. Fix it.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: make the device lock and its protected data in the same cacheline
Miao Xie [Thu, 24 Jul 2014 03:37:10 +0000 (11:37 +0800)]
Btrfs: make the device lock and its protected data in the same cacheline

The lock in btrfs_device structure was far away from its protected data, it would
make CPU load the cache line twice when we accessed them, move them together.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: fix wrong generation check of super block on a seed device
Miao Xie [Thu, 24 Jul 2014 03:37:09 +0000 (11:37 +0800)]
Btrfs: fix wrong generation check of super block on a seed device

The super block generation of the seed devices is not the same as the
filesystem which sprouted from them because we don't update the super
block on the seed devices when we change that new filesystem. So we
should not use the generation of that new filesystem to check the super
block generation on the seed devices, Fix it.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Reviewed-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: fix wrong fsid check of scrub
Miao Xie [Thu, 24 Jul 2014 03:37:08 +0000 (11:37 +0800)]
Btrfs: fix wrong fsid check of scrub

All the metadata in the seed devices has the same fsid as the fsid
of the seed filesystem which is on the seed device, so we should check
them by the current filesystem. Fix it.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Reviewed-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agobtrfs: wake up transaction thread from SYNC_FS ioctl
David Sterba [Wed, 23 Jul 2014 12:39:35 +0000 (14:39 +0200)]
btrfs: wake up transaction thread from SYNC_FS ioctl

The transaction thread may want to do more work, namely it pokes the
cleaner ktread that will start processing uncleaned subvols.

This can be triggered by user via the 'btrfs fi sync' command, otherwise
there was a delay up to 30 seconds before the cleaner started to clean
old snapshots.

Signed-off-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: fix wrong max inline data size limit
Wang Shilong [Thu, 17 Jul 2014 03:44:12 +0000 (11:44 +0800)]
Btrfs: fix wrong max inline data size limit

inline data is stored from offset of @disk_bytenr in
struct btrfs_file_extent_item. So substracting total
size of struct btrfs_file_extent_item is wrong, fix it.

Signed-off-by: Wang Shilong <wangsl.fnst@cn.fujitsu.com>
Reviewed-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: fix off-by-one in cow_file_range_inline()
Wang Shilong [Thu, 17 Jul 2014 03:44:11 +0000 (11:44 +0800)]
Btrfs: fix off-by-one in cow_file_range_inline()

Btrfs could still inline file data if its size is same as
page size, so don't skip max value here.

Signed-off-by: Wang Shilong <wangsl.fnst@cn.fujitsu.com>
Reviewed-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: fall into nocompression codes quickly if possible
Wang Shilong [Thu, 17 Jul 2014 03:44:10 +0000 (11:44 +0800)]
Btrfs: fall into nocompression codes quickly if possible

If flag NOCOMPRESS is set which means bad compression ratio,
we could avoid call cow_file_range_async() for this case earlier.

Signed-off-by: Wang Shilong <wangsl.fnst@cn.fujitsu.com>
Reviewed-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: fix wrong skipping compression for an inode
Wang Shilong [Thu, 17 Jul 2014 03:44:09 +0000 (11:44 +0800)]
Btrfs: fix wrong skipping compression for an inode

If a file's compression ratios is bad, we will set NOCOMPRESS
flag for it, and it will skip compression for that inode next time.

However, if we remount fs to COMPRESS_FORCE, it still should try
if we could compress pages for that inode, this patch fix wrong
check for this problem.

Signed-off-by: Wang Shilong <wangsl.fnst@cn.fujitsu.com>
Reviewed-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: fix sparse warning
Fabian Frederick [Tue, 15 Jul 2014 19:17:17 +0000 (21:17 +0200)]
Btrfs: fix sparse warning

Fix the following sparse warning:
fs/btrfs/send.c:518:51: warning: incorrect type in argument 2 (different address spaces)
fs/btrfs/send.c:518:51:    expected char const [noderef] <asn:1>*<noident>
fs/btrfs/send.c:518:51:    got char *

We can safely use (const char __user *) with set_fs(KERNEL_DS)

__force added to avoid sparse-all warning:
fs/btrfs/send.c:518:40: warning: cast adds address space to expression (<asn:1>)

Signed-off-by: Fabian Frederick <fabf@skynet.be>
Reviewed-by: Zach Brown <zab@zabbo.net>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: use BUG_ON
HIMANGI SARAOGI [Tue, 8 Jul 2014 22:21:41 +0000 (03:51 +0530)]
Btrfs: use BUG_ON

Use BUG_ON(x) rather than if(x) BUG();

The semantic patch that fixes this problem is as follows:

// <smpl>
@@ identifier x; @@
-if (x) BUG();
+BUG_ON(x);
// </smpl>

Signed-off-by: Himangi Saraogi <himangi774@gmail.com>
Acked-by: Julia Lawall <julia.lawall@lip6.fr>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agobtrfs compression: merge inflate and deflate z_streams
Sergey Senozhatsky [Mon, 7 Jul 2014 14:38:29 +0000 (23:38 +0900)]
btrfs compression: merge inflate and deflate z_streams

`struct workspace' used for zlib compression contains two zlib
z_stream-s: `def_strm' used in zlib_compress_pages(), and `inf_strm'
used in zlib_decompress/zlib_decompress_biovec(). None of these
functions use `inf_strm' and `def_strm' simultaniously, meaning that
for every compress/decompress operation we need only one z_stream
(out of two available).

`inf_strm' and `def_strm' are different in size of ->workspace. For
inflate stream we vmalloc() zlib_inflate_workspacesize() bytes, for
deflate stream - zlib_deflate_workspacesize() bytes. On my system zlib
returns the following workspace sizes, correspondingly: 42312 and 268104
(+ guard pages).

Keep only one `z_stream' in `struct workspace' and use it for both
compression and decompression. Hence, instead of vmalloc() of two
z_stream->worskpace-s, allocate only one of size:
max(zlib_deflate_workspacesize(), zlib_inflate_workspacesize())

Reviewed-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: set error return value in btrfs_get_blocks_direct
Filipe Manana [Mon, 7 Jul 2014 11:35:21 +0000 (12:35 +0100)]
Btrfs: set error return value in btrfs_get_blocks_direct

We were returning with 0 (success) because we weren't extracting the
error code from em (PTR_ERR(em)). Fix it.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: Satoru Takeuchi <takeuchi_satoru@jp.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: reduce size of struct extent_state
Filipe Manana [Sun, 6 Jul 2014 19:09:59 +0000 (20:09 +0100)]
Btrfs: reduce size of struct extent_state

The tree field of struct extent_state was only used to figure out if
an extent state was connected to an inode's io tree or not. For this
we can just use the rb_node field itself.

On a x86_64 system with this change the sizeof(struct extent_state) is
reduced from 96 bytes down to 88 bytes, meaning that with a page size
of 4096 bytes we can now store 46 extent states per page instead of 42.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agobtrfs: use PTR_ERR_OR_ZERO
Fabian Frederick [Fri, 4 Jul 2014 19:10:27 +0000 (21:10 +0200)]
btrfs: use PTR_ERR_OR_ZERO

replace IS_ERR/PTR_ERR

Cc: Chris Mason <clm@fb.com>
Cc: Josef Bacik <jbacik@fb.com>
Cc: linux-btrfs@vger.kernel.org
Signed-off-by: Fabian Frederick <fabf@skynet.be>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: print btrfs specific info for some fatal error cases
Wang Shilong [Fri, 4 Jul 2014 09:59:06 +0000 (17:59 +0800)]
Btrfs: print btrfs specific info for some fatal error cases

Marc argued that if there are several btrfs filesystems mounted,
while users even don't know which filesystem hit the corrupted
errors something like generation verification failure.

Since @extent_buffer structure has a member @fs_info, let's output
btrfs device info.

Reported-by: Marc MERLIN <marc@merlins.org>
Signed-off-by: Wang Shilong <wangsl.fnst@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: fix writing data into the seed filesystem
Miao Xie [Thu, 3 Jul 2014 10:22:13 +0000 (18:22 +0800)]
Btrfs: fix writing data into the seed filesystem

If we mounted a seed filesystem with degraded option, and then added a new
device into the seed filesystem, then we found adding device failed because
of the IO failure.

Steps to reproduce:
 # mkfs.btrfs -d raid1 -m raid1 <dev0> <dev1>
 # btrfstune -S 1 <dev0>
 # mount <dev0> -o degraded <mnt>
 # btrfs device add -f <dev2> <mnt>

It is because the original didn't set the chunk on the seed device to be
read-only if the degraded flag was set. It was introduced by patch f48b90756,
which fixed the problem the raid1 filesystem became read-only after one device
of it was missing. But this fix method was not right, we should set the read-only
flag according to the number of the missing devices, not the degraded mount
option, if the number of the missing devices is less than the max error number
that the profile of the chunk tolerates, we don't set it to be read-only.

Cc: Josef Bacik <jbacik@fb.com>
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Reviewed-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: make defragment work with nodatacow option
Wang Shilong [Thu, 3 Jul 2014 10:22:07 +0000 (18:22 +0800)]
Btrfs: make defragment work with nodatacow option

Btrfs defragment will utilize COW feature, which means this
did not work for nodatacow option, this problem was detected
by xfstests generic/018 with nodatacow mount option.

Fix this problem by forcing cow for a extent with state
@EXTETN_DEFRAG setting.

Signed-off-by: Wang Shilong <wangsl.fnst@cn.fujitsu.com>
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agobtrfs: label should not contain return char
Satoru Takeuchi [Tue, 1 Jul 2014 08:00:07 +0000 (17:00 +0900)]
btrfs: label should not contain return char

Rediffed remaining parts of original patch from Anand Jain.  This makes
sure to avoid trailing newlines in the btrfs label output

reproducer.sh:

===============================================================================

TEST_DEV=/dev/vdb
TEST_DIR=/home/sat/mnt

umount /home/sat/mnt

mkfs.btrfs -f $TEST_DEV
UUID=$(btrfs fi show $TEST_DEV | head -1 | sed -e 's/.*uuid: \([-0-9a-z]*\)$/\1/')
mount $TEST_DEV $TEST_DIR
LABELFILE=/sys/fs/btrfs/$UUID/label

echo "Test for empty label..." >&2
LINES="$(cat $LABELFILE | wc -l | awk '{print $1}')"
RET=0

if [ $LINES -eq 0 ] ; then
    echo '[PASS] Trailing \n is removed correctly.' >&2
else
    echo '[FAIL] Trailing \n still exists.' >&2
    RET=1
fi

echo "Test for non-empty label..." >&2

echo testlabel >$LABELFILE
LINES="$(cat $LABELFILE | wc -l | awk '{print $1}')"

if [ $LINES -eq 1 ] ; then
    echo '[PASS] Trailing \n is removed correctly.' >&2
else
    echo '[FAIL] Trailing \n still exists.' >&2
    RET=1
fi

exit $RET
===============================================================================

Signed-off-by: Satoru Takeuchi <takeuchi_satoru@jp.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agobtrfs: device delete must be sysloged
Anand Jain [Mon, 30 Jun 2014 16:58:57 +0000 (00:58 +0800)]
btrfs: device delete must be sysloged

as in the disk add patch, disk detached from the volume must be
recorded in the syslog as well for the same reason.

Signed-off-by: Anand Jain <Anand.Jain@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agobtrfs: device add must be sysloged
Anand Jain [Mon, 30 Jun 2014 16:58:56 +0000 (00:58 +0800)]
btrfs: device add must be sysloged

when we add a new disk to the mounted btrfs we don't record it
as of now, disk add is a critical change of btrfs configuration,
it must be recorded in the syslog to help offline investigations
of customer problems when reported.

Signed-off-by: Anand Jain <Anand.Jain@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoBtrfs: clear compress-force when remounting with compress option
Wang Shilong [Mon, 30 Jun 2014 02:51:25 +0000 (10:51 +0800)]
Btrfs: clear compress-force when remounting with compress option

Steps to reproduce:
 # mkfs.btrfs -f /dev/sdb
 # mount /dev/sdb /mnt -o compress-force=lzo
 # mount /dev/sdb /mnt -o remount,compress=zlib
 # cat /proc/mounts

Remounting from compress-force to compress could not clear compress-force
option. The problem is there is no way for users to clear compress-force
option separately.

Fix this problem by clearing @FORCE_COMPRESS flag when remounting to
compress=xxx.

Suggested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com>
Signed-off-by: Wang Shilong <wangsl.fnst@cn.fujitsu.com>
Reviewed-by: David Sterba <dsterba@suse.cz>
Reviewed-by: Satoru Takeuchi <takeuchi_satoru@jp.fujitsu.com>
Tested-by: Satoru Takeuchi <takeuchi_satoru@jp.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agobtrfs: use DIV_ROUND_UP instead of open-coded variants
David Sterba [Wed, 4 Jun 2014 23:59:57 +0000 (01:59 +0200)]
btrfs: use DIV_ROUND_UP instead of open-coded variants

The form

  (value + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT

is equivalent to

  (value + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE

The rest is a simple subsitution, no difference in the generated
assembly code.

Signed-off-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agobtrfs: clean away stripe_align helper
David Sterba [Wed, 4 Jun 2014 23:39:19 +0000 (01:39 +0200)]
btrfs: clean away stripe_align helper

Only wraps the ALIGN macro.

Signed-off-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agobtrfs: use nodesize everywhere, kill leafsize
David Sterba [Wed, 4 Jun 2014 17:22:26 +0000 (19:22 +0200)]
btrfs: use nodesize everywhere, kill leafsize

The nodesize and leafsize were never of different values. Unify the
usage and make nodesize the one. Cleanup the redundant checks and
helpers.

Shaves a few bytes from .text:

  text    data     bss     dec     hex filename
852418   24560   23112  900090   dbbfa btrfs.ko.before
851074   24584   23112  898770   db6d2 btrfs.ko.after

Signed-off-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agobtrfs: kill the key type accessor helpers
David Sterba [Wed, 4 Jun 2014 16:41:45 +0000 (18:41 +0200)]
btrfs: kill the key type accessor helpers

btrfs_set_key_type and btrfs_key_type are used inconsistently along with
open coded variants. Other members of btrfs_key are accessed directly
without any helpers anyway.

Signed-off-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agobtrfs: make close_ctree return void
David Sterba [Wed, 4 Jun 2014 16:10:45 +0000 (18:10 +0200)]
btrfs: make close_ctree return void

There's no user of the return value and we can get rid of the comment in
put_super.

Signed-off-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agobtrfs: cleanup ino cache members of btrfs_root
David Sterba [Wed, 5 Feb 2014 01:37:48 +0000 (02:37 +0100)]
btrfs: cleanup ino cache members of btrfs_root

The naming is confusing, generic yet used for a specific cache. Add a
prefix 'ino_' or rename appropriately.

Signed-off-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agobtrfs: clenaup: don't call btrfs_release_path before free_path
David Sterba [Wed, 5 Feb 2014 01:11:25 +0000 (02:11 +0100)]
btrfs: clenaup: don't call btrfs_release_path before free_path

Signed-off-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agobtrfs: remove obsolete comment in btrfs_clean_one_deleted_snapshot
David Sterba [Wed, 5 Feb 2014 01:03:47 +0000 (02:03 +0100)]
btrfs: remove obsolete comment in btrfs_clean_one_deleted_snapshot

The comment applied when there was a BUG_ON.

Signed-off-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Chris Mason <clm@fb.com>
10 years agoLinux 3.17-rc5
Linus Torvalds [Mon, 15 Sep 2014 00:50:12 +0000 (17:50 -0700)]
Linux 3.17-rc5

10 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
Linus Torvalds [Mon, 15 Sep 2014 00:37:36 +0000 (17:37 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/viro/vfs

Pull vfs fixes from Al Viro:
 "double iput() on failure exit in lustre, racy removal of spliced
  dentries from ->s_anon in __d_materialise_dentry() plus a bunch of
  assorted RCU pathwalk fixes"

The RCU pathwalk fixes end up fixing a couple of cases where we
incorrectly dropped out of RCU walking, due to incorrect initialization
and testing of the sequence locks in some corner cases.  Since dropping
out of RCU walk mode forces the slow locked accesses, those corner cases
slowed down quite dramatically.

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
  be careful with nd->inode in path_init() and follow_dotdot_rcu()
  don't bugger nd->seq on set_root_rcu() from follow_dotdot_rcu()
  fix bogus read_seqretry() checks introduced in b37199e
  move the call of __d_drop(anon) into __d_materialise_unique(dentry, anon)
  [fix] lustre: d_make_root() does iput() on dentry allocation failure

10 years agovfs: avoid non-forwarding large load after small store in path lookup
Linus Torvalds [Mon, 15 Sep 2014 00:28:32 +0000 (17:28 -0700)]
vfs: avoid non-forwarding large load after small store in path lookup

The performance regression that Josef Bacik reported in the pathname
lookup (see commit 99d263d4c5b2 "vfs: fix bad hashing of dentries") made
me look at performance stability of the dcache code, just to verify that
the problem was actually fixed.  That turned up a few other problems in
this area.

There are a few cases where we exit RCU lookup mode and go to the slow
serializing case when we shouldn't, Al has fixed those and they'll come
in with the next VFS pull.

But my performance verification also shows that link_path_walk() turns
out to have a very unfortunate 32-bit store of the length and hash of
the name we look up, followed by a 64-bit read of the combined hash_len
field.  That screws up the processor store to load forwarding, causing
an unnecessary hickup in this critical routine.

It's caused by the ugly calling convention for the "hash_name()"
function, and easily fixed by just making hash_name() fill in the whole
'struct qstr' rather than passing it a pointer to just the hash value.

With that, the profile for this function looks much smoother.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agoMerge branch 'parisc-3.17-1' of git://git.kernel.org/pub/scm/linux/kernel/git/deller...
Linus Torvalds [Sun, 14 Sep 2014 19:28:08 +0000 (12:28 -0700)]
Merge branch 'parisc-3.17-1' of git://git./linux/kernel/git/deller/parisc-linux

Pull parisc updates from Helge Deller:
 "The most important patch is a new Light Weigth Syscall (LWS) for 8,
  16, 32 and 64 bit atomic CAS operations which is required in order to
  be able to implement the atomic gcc builtins on our platform.

  Other than that, we wire up the seccomp, getrandom and memfd_create
  syscalls, fixes a minor off-by-one bug and a wrong printk string"

* 'parisc-3.17-1' of git://git.kernel.org/pub/scm/linux/kernel/git/deller/parisc-linux:
  parisc: Implement new LWS CAS supporting 64 bit operations.
  parisc: Wire up seccomp, getrandom and memfd_create syscalls
  parisc: dino: fix %d confusingly prefixed with 0x in format string
  parisc: sys_hpux: NUL terminator is one past the end

10 years agobe careful with nd->inode in path_init() and follow_dotdot_rcu()
Al Viro [Sun, 14 Sep 2014 01:59:43 +0000 (21:59 -0400)]
be careful with nd->inode in path_init() and follow_dotdot_rcu()

in the former we simply check if dentry is still valid after picking
its ->d_inode; in the latter we fetch ->d_inode in the same places
where we fetch dentry and its ->d_seq, under the same checks.

Cc: stable@vger.kernel.org # 2.6.38+
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
10 years agodon't bugger nd->seq on set_root_rcu() from follow_dotdot_rcu()
Al Viro [Sun, 14 Sep 2014 01:55:46 +0000 (21:55 -0400)]
don't bugger nd->seq on set_root_rcu() from follow_dotdot_rcu()

return the value instead, and have path_init() do the assignment.  Broken by
"vfs: Fix absolute RCU path walk failures due to uninitialized seq number",
which was Cc-stable with 2.6.38+ as destination.  This one should go where
it went.

To avoid dummy value returned in case when root is already set (it would do
no harm, actually, since the only caller that doesn't ignore the return value
is guaranteed to have nd->root *not* set, but it's more obvious that way),
lift the check into callers.  And do the same to set_root(), to keep them
in sync.

Cc: stable@vger.kernel.org # 2.6.38+
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
10 years agoMerge tag 'ntb-3.17' of git://github.com/jonmason/ntb
Linus Torvalds [Sun, 14 Sep 2014 17:54:12 +0000 (10:54 -0700)]
Merge tag 'ntb-3.17' of git://github.com/jonmason/ntb

Pull ntb driver bugfixes from Jon Mason:
 "NTB driver fixes for queue spread and buffer alignment.  Also, update
  to MAINTAINERS to reflect new e-mail address"

* tag 'ntb-3.17' of git://github.com/jonmason/ntb:
  ntb: Add alignment check to meet hardware requirement
  MAINTAINERS: update NTB info
  NTB: correct the spread of queues over mw's

10 years agoMerge branch 'irq-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 14 Sep 2014 17:37:10 +0000 (10:37 -0700)]
Merge branch 'irq-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull ARM irq chip fixes from Thomas Gleixner:
 "Another pile of ARM specific irq chip fixlets:

   - off by one bugs in the crossbar driver
   - missing annotations
   - a bunch of "make it compile" updates

  I pulled the lot today from Jason, but it has been in -next for at
  least a week"

* 'irq-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  irqchip: gic-v3: Declare rdist as __percpu pointer to __iomem pointer
  irqchip: gic: Make gic_default_routable_irq_domain_ops static
  irqchip: exynos-combiner: Fix compilation error on ARM64
  irqchip: crossbar: Off by one bugs in init
  irqchip: gic-v3: Tag all low level accessors __maybe_unused
  irqchip: gic-v3: Only define gic_peek_irq() when building SMP

10 years agoMerge tag 'irqchip-urgent-3.17' of git://git.infradead.org/users/jcooper/linux into...
Thomas Gleixner [Sun, 14 Sep 2014 13:19:19 +0000 (15:19 +0200)]
Merge tag 'irqchip-urgent-3.17' of git://git.infradead.org/users/jcooper/linux into irq/urgent

irqchip fixes for v3.17 from Jason Cooper

 - GIC/GICV3: Various fixlets
 - crossbar: Fix off-by-one bug
 - exynos-combiner: Fix arm64 build error

10 years agontb: Add alignment check to meet hardware requirement
Dave Jiang [Thu, 28 Aug 2014 20:53:02 +0000 (13:53 -0700)]
ntb: Add alignment check to meet hardware requirement

The NTB translate register must have the value to be BAR size aligned.
This alignment check make sure that the DMA memory allocated has the
proper alignment. Another requirement for NTB to function properly with
memory window BAR size greater or equal to 4M is to use the CMA feature
in 3.16 kernel with the appropriate CONFIG_CMA_ALIGNMENT and
CONFIG_CMA_SIZE_MBYTES set.

Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Signed-off-by: Jon Mason <jdmason@kudzu.us>
10 years agoMAINTAINERS: update NTB info
Jon Mason [Thu, 19 Jun 2014 17:44:24 +0000 (10:44 -0700)]
MAINTAINERS: update NTB info

Update my contact info to my personal email address and add Dave Jiang.

Signed-off-by: Jon Mason <jon.mason@intel.com>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
10 years agoNTB: correct the spread of queues over mw's
Jon Mason [Thu, 19 Jun 2014 17:11:13 +0000 (10:11 -0700)]
NTB: correct the spread of queues over mw's

The detection of an uneven number of queues on the given memory windows
was not correct.  The mw_num is zero based and the mod should be
division to spread them evenly over the mw's.

Signed-off-by: Jon Mason <jon.mason@intel.com>
10 years agofix bogus read_seqretry() checks introduced in b37199e
Al Viro [Sun, 14 Sep 2014 01:50:45 +0000 (21:50 -0400)]
fix bogus read_seqretry() checks introduced in b37199e

read_seqretry() returns true on mismatch, not on match...

Cc: stable@vger.kernel.org # 3.15+
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
10 years agomove the call of __d_drop(anon) into __d_materialise_unique(dentry, anon)
Al Viro [Thu, 11 Sep 2014 22:55:50 +0000 (18:55 -0400)]
move the call of __d_drop(anon) into __d_materialise_unique(dentry, anon)

and lock the right list there

Cc: stable@vger.kernel.org
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
10 years ago[fix] lustre: d_make_root() does iput() on dentry allocation failure
Al Viro [Wed, 3 Sep 2014 17:11:09 +0000 (13:11 -0400)]
[fix] lustre: d_make_root() does iput() on dentry allocation failure

double-free is a bad thing

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
10 years agoMerge branches 'locking-urgent-for-linus' and 'timers-urgent-for-linus' of git:/...
Linus Torvalds [Sat, 13 Sep 2014 21:22:12 +0000 (14:22 -0700)]
Merge branches 'locking-urgent-for-linus' and 'timers-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull futex and timer fixes from Thomas Gleixner:
 "A oneliner bugfix for the jinxed futex code:

   - Drop hash bucket lock in the error exit path.  I really could slap
     myself for intruducing that bug while fixing all the other horror
     in that code three month ago ...

  and the timer department is not too proud about the following fixes:

   - Deal with a long standing rounding bug in the timeval to jiffies
     conversion.  It's a real issue and this fix fell through the cracks
     for quite some time.

   - Another round of alarmtimer fixes.  Finally this code gets used
     more widely and the subtle issues hidden for quite some time are
     noticed and fixed.  Nothing really exciting, just the itty bitty
     details which bite the serious users here and there"

* 'locking-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  futex: Unlock hb->lock in futex_wait_requeue_pi() error path

* 'timers-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  alarmtimer: Lock k_itimer during timer callback
  alarmtimer: Do not signal SIGEV_NONE timers
  alarmtimer: Return relative times in timer_gettime
  jiffies: Fix timeval conversion to jiffies

10 years agoparisc: Implement new LWS CAS supporting 64 bit operations.
Guy Martin [Fri, 12 Sep 2014 16:02:34 +0000 (18:02 +0200)]
parisc: Implement new LWS CAS supporting 64 bit operations.

The current LWS cas only works correctly for 32bit. The new LWS allows
for CAS operations of variable size.

Signed-off-by: Guy Martin <gmsoft@tuxicoman.be>
Cc: <stable@vger.kernel.org> # 3.13+
Signed-off-by: Helge Deller <deller@gmx.de>
10 years agovfs: fix bad hashing of dentries
Linus Torvalds [Sat, 13 Sep 2014 18:30:10 +0000 (11:30 -0700)]
vfs: fix bad hashing of dentries

Josef Bacik found a performance regression between 3.2 and 3.10 and
narrowed it down to commit bfcfaa77bdf0 ("vfs: use 'unsigned long'
accesses for dcache name comparison and hashing"). He reports:

 "The test case is essentially

      for (i = 0; i < 1000000; i++)
              mkdir("a$i");

  On xfs on a fio card this goes at about 20k dir/sec with 3.2, and 12k
  dir/sec with 3.10.  This is because we spend waaaaay more time in
  __d_lookup on 3.10 than in 3.2.

  The new hashing function for strings is suboptimal for <
  sizeof(unsigned long) string names (and hell even > sizeof(unsigned
  long) string names that I've tested).  I broke out the old hashing
  function and the new one into a userspace helper to get real numbers
  and this is what I'm getting:

      Old hash table had 1000000 entries, 0 dupes, 0 max dupes
      New hash table had 12628 entries, 987372 dupes, 900 max dupes
      We had 11400 buckets with a p50 of 30 dupes, p90 of 240 dupes, p99 of 567 dupes for the new hash

  My test does the hash, and then does the d_hash into a integer pointer
  array the same size as the dentry hash table on my system, and then
  just increments the value at the address we got to see how many
  entries we overlap with.

  As you can see the old hash function ended up with all 1 million
  entries in their own bucket, whereas the new one they are only
  distributed among ~12.5k buckets, which is why we're using so much
  more CPU in __d_lookup".

The reason for this hash regression is two-fold:

 - On 64-bit architectures the down-mixing of the original 64-bit
   word-at-a-time hash into the final 32-bit hash value is very
   simplistic and suboptimal, and just adds the two 32-bit parts
   together.

   In particular, because there is no bit shuffling and the mixing
   boundary is also a byte boundary, similar character patterns in the
   low and high word easily end up just canceling each other out.

 - the old byte-at-a-time hash mixed each byte into the final hash as it
   hashed the path component name, resulting in the low bits of the hash
   generally being a good source of hash data.  That is not true for the
   word-at-a-time case, and the hash data is distributed among all the
   bits.

The fix is the same in both cases: do a better job of mixing the bits up
and using as much of the hash data as possible.  We already have the
"hash_32|64()" functions to do that.

Reported-by: Josef Bacik <jbacik@fb.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Chris Mason <clm@fb.com>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>