pandora-kernel.git
11 years agobtrfs: remove unnecessary DEFINE_WAIT() declarations
Eric Sandeen [Thu, 31 Jan 2013 00:55:00 +0000 (00:55 +0000)]
btrfs: remove unnecessary DEFINE_WAIT() declarations

No point in DEFINE_WAIT(wait) if it's not used!

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agobtrfs: remove unused "item" in btrfs_insert_delayed_item()
Eric Sandeen [Thu, 31 Jan 2013 00:54:59 +0000 (00:54 +0000)]
btrfs: remove unused "item" in btrfs_insert_delayed_item()

"item" was set but never used in this function.

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agobtrfs: fix varargs in __btrfs_std_error
Eric Sandeen [Thu, 31 Jan 2013 00:54:58 +0000 (00:54 +0000)]
btrfs: fix varargs in __btrfs_std_error

__btrfs_std_error didn't always properly call va_end,
and might call va_start even if fmt was NULL.

Move all the varargs handling into the block where we
have fmt.

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agobtrfs: add missing break in btrfs_print_leaf()
Eric Sandeen [Thu, 31 Jan 2013 00:54:57 +0000 (00:54 +0000)]
btrfs: add missing break in btrfs_print_leaf()

I don't think that BTRFS_DEV_EXTENT_KEY is supposed
to fall through to BTRFS_DEV_STATS_KEY ...

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agobtrfs: annotate intentional switch case fallthroughs
Eric Sandeen [Thu, 31 Jan 2013 00:54:56 +0000 (00:54 +0000)]
btrfs: annotate intentional switch case fallthroughs

This keeps static checkers happy.

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agobtrfs: handle null fs_info in btrfs_panic()
Eric Sandeen [Thu, 31 Jan 2013 00:54:55 +0000 (00:54 +0000)]
btrfs: handle null fs_info in btrfs_panic()

At least backref_tree_panic() can apparently pass
in a null fs_info, so handle that in __btrfs_panic
to get the message out on the console.

The btrfs_panic macro also uses fs_info, but that's
largely pointless; it's testing to see if
BTRFS_MOUNT_PANIC_ON_FATAL_ERROR is not set.
But if it *were* set, __btrfs_panic() would have,
well, paniced and we wouldn't be here, testing it!
So just BUG() at this point.

And since we only use fs_info once now, just use it
directly.

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agobtrfs: remove unused fs_info from btrfs_decode_error()
Eric Sandeen [Thu, 31 Jan 2013 00:54:54 +0000 (00:54 +0000)]
btrfs: remove unused fs_info from btrfs_decode_error()

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agobtrfs: list_entry can't return NULL
Eric Sandeen [Thu, 31 Jan 2013 00:54:53 +0000 (00:54 +0000)]
btrfs: list_entry can't return NULL

No need to test the result, we can't get a
null pointer from list_entry()

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agobtrfs: remove unused fd in btrfs_ioctl_send()
Eric Sandeen [Thu, 31 Jan 2013 00:54:52 +0000 (00:54 +0000)]
btrfs: remove unused fd in btrfs_ioctl_send()

All we do is set it to NULL and test it :)

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: do not overcommit if we don't have enough space for global rsv
Josef Bacik [Wed, 30 Jan 2013 22:02:51 +0000 (17:02 -0500)]
Btrfs: do not overcommit if we don't have enough space for global rsv

Because of how little we allocate chunks now we can get really tight on
metadata space before we will allocate a new chunk.  This resulted in being
unable to add device extents when allocating a new metadata chunk as we did
not have enough space.  This is because we were allowed to overcommit too
much metadata without actually making sure we had enough space to make
allocations.  The idea behind overcommit is that we are allowed to say "sure
you can have that reservation" when most of the free space is occupied by
reservations, not actual allocations.  But in this case where a majority of
the total space is in use by actual allocations we can screw ourselves by
not being able to make real allocations when it matters.  So make sure we
have enough real space for our global reserve, and if not then don't allow
overcommitting.  Thanks,

Reported-and-tested-by: Jim Schutt <jaschut@sandia.gov>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: remove extent mapping if we fail to add chunk
Josef Bacik [Thu, 31 Jan 2013 15:23:04 +0000 (10:23 -0500)]
Btrfs: remove extent mapping if we fail to add chunk

I got a double free error when unmounting a file system that failed to add a
chunk during its operation.  This is because we will kfree the mapping that
we created but leave the extent_map in the em_tree for chunks.  So to fix
this just remove the extent_map when we error out so we don't run into this
problem.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: fix chunk allocation error handling
Josef Bacik [Tue, 29 Jan 2013 20:03:37 +0000 (15:03 -0500)]
Btrfs: fix chunk allocation error handling

If we error out allocating a dev extent we will have already created the
block group and such which will cause problems since the allocator may have
tried to allocate out of the block group that no longer exists.  This will
cause BUG_ON()'s in the bio submission path.  This also makes a failure to
allocate a dev extent a non-abort error, we will just clean up the dev
extents we did allocate and exit.  Now if we fail to delete the dev extents
we will abort since we can't have half of the dev extents hanging around,
but this will make us much less likely to abort.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: use bit operation for ->fs_state
Miao Xie [Tue, 29 Jan 2013 10:14:48 +0000 (10:14 +0000)]
Btrfs: use bit operation for ->fs_state

There is no lock to protect fs_info->fs_state, it will introduce
some problems, such as the value may be covered by the other task
when several tasks modify it. For example:
Task0 - CPU0 Task1 - CPU1
mov %fs_state rax
or $0x1 rax
mov %fs_state rax
or $0x2 rax
mov rax %fs_state
mov rax %fs_state
The expected value is 3, but in fact, it is 2.

Though this problem doesn't happen now (because there is only one
flag currently), the code is error prone, if we add other flags,
the above problem will happen to a certainty.

Now we use bit operation for it to fix the above problem.
In this way, we can make the code more robust and be easy to
add new flags.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: use seqlock to protect fs_info->avail_{data, metadata, system}_alloc_bits
Miao Xie [Tue, 29 Jan 2013 10:13:12 +0000 (10:13 +0000)]
Btrfs: use seqlock to protect fs_info->avail_{data, metadata, system}_alloc_bits

There is no lock to protect
  fs_info->avail_{data, metadata, system}_alloc_bits,
it may introduce some problem, such as the wrong profile
information, so we add a seqlock to protect them.

Signed-off-by: Zhao Lei <zhaolei@cn.fujitsu.com>
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: use the inode own lock to protect its delalloc_bytes
Miao Xie [Tue, 29 Jan 2013 10:11:59 +0000 (10:11 +0000)]
Btrfs: use the inode own lock to protect its delalloc_bytes

We need not use a global lock to protect the delalloc_bytes of the
inode, just use its own lock. In this way, we can reduce the lock
contention and ->delalloc_lock will just protect delalloc inode
list.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: use percpu counter for fs_info->delalloc_bytes
Miao Xie [Tue, 29 Jan 2013 10:10:51 +0000 (10:10 +0000)]
Btrfs: use percpu counter for fs_info->delalloc_bytes

fs_info->delalloc_bytes is accessed very frequently, so use percpu
counter instead of the u64 variant for it to reduce the lock
contention.

This patch also fixed the problem that we access the variant
without the lock protection.At worst, we would not flush the
delalloc inodes, and just return ENOSPC error when we still have
some free space in the fs.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: use percpu counter for dirty metadata count
Miao Xie [Tue, 29 Jan 2013 10:09:20 +0000 (10:09 +0000)]
Btrfs: use percpu counter for dirty metadata count

->dirty_metadata_bytes is accessed very frequently, so use percpu
counter instead of the u64 variant to reduce the contention of
the lock.

This patch also fixed the problem that we access it without
lock protection in __btrfs_btree_balance_dirty(), which may
cause we skip the dirty pages flush.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: protect fs_info->alloc_start
Miao Xie [Tue, 29 Jan 2013 10:07:33 +0000 (10:07 +0000)]
Btrfs: protect fs_info->alloc_start

fs_info->alloc_start is a 64bits variant, can be accessed by
multi-task, but it is not protected strictly, it can be changed
while we are accessing it. On 32bit machine, we will get wrong
value because we access it by two instructions.(In fact, it is
also possible that the same problem happens on the 64bit machine,
because the compiler may split the 64bit operation into two 32bit
operation.)

For example:
Assuming -> alloc_start is 0x0000 0000 0001 0000 at the beginning,
then we remount and set ->alloc_start to 0x0000 0100 0000 0000.
Task0  Task1
load high 32 bits
set high 32 bits
set low 32 bits
load low 32 bits

Task1 will get 0.

This patch fixes this problem by using two locks to protect it
fs_info->chunk_mutex
sb->s_umount
On the read side, we just need get one of these two locks, and on
the write side, we must lock all of them.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: add a comment for fs_info->max_inline
Miao Xie [Tue, 29 Jan 2013 10:05:05 +0000 (10:05 +0000)]
Btrfs: add a comment for fs_info->max_inline

Though ->max_inline is a 64bit variant, and may be accessed by
multi-task, but it is just suggestive number, so we needn't add
anything to protect fs_info->max_inline, just add a comment to
explain wny we don't use a lock to protect it.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: move fs/btrfs/ioctl.h to include/uapi/linux/btrfs.h
Filipe Brandenburger [Tue, 29 Jan 2013 06:04:50 +0000 (06:04 +0000)]
Btrfs: move fs/btrfs/ioctl.h to include/uapi/linux/btrfs.h

The header file will then be installed under /usr/include/linux so that
userspace applications can refer to Btrfs ioctls by name and use the same
structs used internally in the kernel.

Signed-off-by: Filipe Brandenburger <filbranden@google.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: Check CAP_DAC_READ_SEARCH for BTRFS_IOC_INO_PATHS
Kusanagi Kouichi [Mon, 28 Jan 2013 11:33:31 +0000 (11:33 +0000)]
Btrfs: Check CAP_DAC_READ_SEARCH for BTRFS_IOC_INO_PATHS

CAP_DAC_READ_SEARCH overrides read and search permission check on
file and directory. It seems fit for BTRFS_IOC_INO_PATHS.

Signed-off-by: Kusanagi Kouichi <slash@ac.auone-net.jp>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoRevert "Btrfs: fix permissions of empty files not affected by umask"
Josef Bacik [Thu, 24 Jan 2013 18:49:14 +0000 (13:49 -0500)]
Revert "Btrfs: fix permissions of empty files not affected by umask"

This reverts commit 2794ed013b3551cbae887ea1b93c52aaacb7370d.

Wasn't supposed to get used in btrfs_mknod, it was supposed to be in
btrfs_create, which was done in commit
9185aa587b7425f8f4520da2e66792f5f3c2b815.

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: don't traverse the ordered operation list repeatedly
Miao Xie [Tue, 22 Jan 2013 10:52:04 +0000 (10:52 +0000)]
Btrfs: don't traverse the ordered operation list repeatedly

btrfs_run_ordered_operations() needn't traverse the ordered operation list
repeatedly, it is because the transaction commiter will invoke it again when
there is no other writer in this transaction, it can ensure that no one can
add new objects into the ordered operation list.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: traverse and flush the delalloc inodes once
Miao Xie [Tue, 22 Jan 2013 10:50:35 +0000 (10:50 +0000)]
Btrfs: traverse and flush the delalloc inodes once

btrfs_start_delalloc_inodes() needn't traverse and flush the delalloc inodes
repeatedly. It is because we can regard the data that the users write after
we start delalloc inodes flush as the one which is after the delalloc inodes
flush is done, and we can flush it next time.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: check the return value of btrfs_run_ordered_operations()
Miao Xie [Tue, 22 Jan 2013 10:50:01 +0000 (10:50 +0000)]
Btrfs: check the return value of btrfs_run_ordered_operations()

We forget to check the return value of btrfs_run_ordered_operations() when
flushing all the pending stuffs, fix it.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: check the return value of btrfs_start_delalloc_inodes()
Miao Xie [Tue, 22 Jan 2013 10:49:33 +0000 (10:49 +0000)]
Btrfs: check the return value of btrfs_start_delalloc_inodes()

We forget to check the return value of btrfs_start_delalloc_inodes(), fix it.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: make raid attr array more readable
Miao Xie [Thu, 17 Jan 2013 05:38:51 +0000 (05:38 +0000)]
Btrfs: make raid attr array more readable

The current code of raid attr arry is hard to understand and it is easy to
introduce some problem if we modify the array. So I changed it and made it
more readable.

Cc: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: record first logical byte in memory
Liu Bo [Thu, 27 Dec 2012 09:01:23 +0000 (09:01 +0000)]
Btrfs: record first logical byte in memory

This'd save us a rbtree search which may become expensive in large filesystem.

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: save us a read_lock
Liu Bo [Thu, 27 Dec 2012 09:01:22 +0000 (09:01 +0000)]
Btrfs: save us a read_lock

This does not change the logic of code, but can save us a read_lock.

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: use token to avoid times mapping extent buffer
Liu Bo [Thu, 27 Dec 2012 09:01:21 +0000 (09:01 +0000)]
Btrfs: use token to avoid times mapping extent buffer

The API in tree log code has done sort of changes, and it proves that
we can benifit from using token, so do the same thing here.

function_graph tracer's timer shows that it costs nearly half time
of before(39.788us -> 22.391us).

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: kill unused argument of btrfs_pin_extent_for_log_replay
Liu Bo [Thu, 27 Dec 2012 09:01:20 +0000 (09:01 +0000)]
Btrfs: kill unused argument of btrfs_pin_extent_for_log_replay

Argument 'trans' is not used any more.

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: kill unused argument of update_block_group
Liu Bo [Thu, 27 Dec 2012 09:01:19 +0000 (09:01 +0000)]
Btrfs: kill unused argument of update_block_group

Argument 'trans' is not used any more.

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: kill unused arguments of cache_block_group
Liu Bo [Thu, 27 Dec 2012 09:01:18 +0000 (09:01 +0000)]
Btrfs: kill unused arguments of cache_block_group

Argument 'trans' and 'root' are not used any more.

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: remove deprecated comments
Liu Bo [Thu, 27 Dec 2012 09:01:17 +0000 (09:01 +0000)]
Btrfs: remove deprecated comments

commit d53ba47484ed6245e640ee4bfe9d21e9bfc15765
(Btrfs: use commit root when loading free space cache) has remove
the deadlock check, and the related comments can be removed as well.

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: don't re-enter when allocating a chunk
Josef Bacik [Tue, 18 Dec 2012 14:16:16 +0000 (09:16 -0500)]
Btrfs: don't re-enter when allocating a chunk

If we start running low on metadata space we will try to allocate a chunk,
which could then try to allocate a chunk to add the device entry.  The thing
is we allocate a chunk before we try really hard to make the allocation, so
we should be able to find space for the device entry.  Add a flag to the
trans handle so we know we're currently allocating a chunk so we can just
bail out if we try to allocate another chunk.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: wait on ordered extents at the last possible moment
Josef Bacik [Fri, 12 Oct 2012 19:27:49 +0000 (15:27 -0400)]
Btrfs: wait on ordered extents at the last possible moment

Since we don't actually copy the extent information from the source tree in
the fast case we don't need to wait for ordered io to be completed in order
to fsync, we just need to wait for the io to be completed.  So when we're
logging our file just attach all of the ordered extents to the log, and then
when the log syncs just wait for IO_DONE on the ordered extents and then
write the super.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: fix trivial error in btrfs_ioctl_resize()
Miao Xie [Fri, 21 Dec 2012 09:21:30 +0000 (09:21 +0000)]
Btrfs: fix trivial error in btrfs_ioctl_resize()

This patch fixes the following problem:
- improper return value
- unnecessary read-only check

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: use wrapper page_offset
Miao Xie [Fri, 21 Dec 2012 09:17:45 +0000 (09:17 +0000)]
Btrfs: use wrapper page_offset

Use wrapper page_offset to get byte-offset into filesystem object for page.

Signed-off-by: Liu Bo <liubo2009@cn.fujitsu.com>
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: flush all dirty inodes if writeback can not start
Miao Xie [Thu, 20 Dec 2012 11:19:09 +0000 (11:19 +0000)]
Btrfs: flush all dirty inodes if writeback can not start

We may try to flush some dirty pages when there is no enough space to reserve.
But it is possible that this operation fails, in order to get enough space to
reserve successfully, we will sync all the delalloc file. This operation is
safe, we needn't worry about the case that the filesystem goes from r/w to r/o.
because the filesystem should guarantee all the dirty pages have been written
into the disk after it becomes readonly, so the sync operation will do nothing
if the filesystem is already readonly. Though it may waste lots of time,
as a corner case, we needn't care.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: make delayed ref lock logic more readable
Miao Xie [Wed, 19 Dec 2012 08:10:10 +0000 (08:10 +0000)]
Btrfs: make delayed ref lock logic more readable

Locking and unlocking delayed ref mutex are in the different functions,
and the name of lock functions is not uniform, so the readability is not
so good, this patch optimizes the lock logic and makes it more readable.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: fix lots of orphan inodes when the space is not enough
Miao Xie [Wed, 19 Dec 2012 06:59:51 +0000 (06:59 +0000)]
Btrfs: fix lots of orphan inodes when the space is not enough

We're running into having 50-100 orphans left over with xfstests 83
because of ENOSPC when trying to start the transaction for the inode update.
But in fact, it makes no sense in updating the inode for the new size while
we're deleting the stupid thing. This patch fixes this problem.

Reported-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: cleanup similar code in delayed inode
Miao Xie [Wed, 19 Dec 2012 06:59:03 +0000 (06:59 +0000)]
Btrfs: cleanup similar code in delayed inode

The delayed item commit code in several functions is similar, so
cleanup it.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: use common work instead of delayed work
Miao Xie [Thu, 15 Nov 2012 08:14:47 +0000 (08:14 +0000)]
Btrfs: use common work instead of delayed work

Since we do not want to delay the async transaction commit, we should
use common work, not delayed work.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
11 years agoBtrfs: cleanup unnecessary clear when freeing a transaction or a trans handle
Miao Xie [Thu, 15 Nov 2012 08:14:11 +0000 (08:14 +0000)]
Btrfs: cleanup unnecessary clear when freeing a transaction or a trans handle

We clear the transaction object and the trans handle when they are about to be
freed, it is unnecessary, cleanup it.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
11 years agoBtrfs: use slabs for delayed reference allocation
Miao Xie [Wed, 21 Nov 2012 02:21:28 +0000 (02:21 +0000)]
Btrfs: use slabs for delayed reference allocation

The delayed reference allocation is in the fast path of the IO, so use slabs
to improve the speed of the allocation.

And besides that, it can do check for leaked objects when the module is removed.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
11 years agobtrfs: access superblock via pagecache in scan_one_device
David Sterba [Fri, 15 Feb 2013 18:31:02 +0000 (11:31 -0700)]
btrfs: access superblock via pagecache in scan_one_device

btrfs_scan_one_device is calling set_blocksize() which can race
with a concurrent process making dirty page cache pages.  It can end up
dropping dirty page cache pages on the floor, which isn't very nice when
someone is just running btrfs dev scan to find filesystems on the
box.

Now that udev is registering btrfs devices as it discovers them, we can
actually end up racing with our own mkfs program too.  When this
happens, we drop some of the important blocks written by mkfs.

This commit changes scan_one_device to read the super out of the page
cache instead of trying to use bread.  This way we don't have to care
about the blocksize of the device.

This also drops the invalidate_bdev() call.  It wasn't very polite to
invalidate during the scan either.  mkfs is putting the super into the
page cache, there's no reason to invalidate at this point.

Signed-off-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: fix crash in log replay with qgroups enabled
Arne Jansen [Wed, 13 Feb 2013 11:20:01 +0000 (04:20 -0700)]
Btrfs: fix crash in log replay with qgroups enabled

When replaying a log tree with qgroups enabled, tree_mod_log_rewind does a
sanity-check of the number of items against the maximum possible number.
It calculates that number with the nodesize of fs_root. Unfortunately
fs_root is not yet set at this stage. So instead use the nodesize from
tree_root, which is already initialized.

Signed-off-by: Arne Jansen <sensille@gmx.net>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: move d_instantiate outside the transaction during mksubvol
Chris Mason [Wed, 6 Feb 2013 17:06:02 +0000 (12:06 -0500)]
Btrfs: move d_instantiate outside the transaction during mksubvol

Dave Sterba triggered a lockdep complaint about lock ordering
between the sb_internal lock and the cleaner semaphore.

btrfs_lookup_dentry() checks for orphans if we're looking up
the inode for a subvolume, and subvolume creation is triggering
the lookup with a transaction running.

This commit moves the d_instantiate after the transaction closes.

Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: fix EDQUOT handling in btrfs_delalloc_reserve_metadata
Jan Schmidt [Mon, 28 Jan 2013 06:26:00 +0000 (23:26 -0700)]
Btrfs: fix EDQUOT handling in btrfs_delalloc_reserve_metadata

When btrfs_qgroup_reserve returned a failure, we were missing a counter
operation for BTRFS_I(inode)->outstanding_extents++, leading to warning
messages about outstanding extents and space_info->bytes_may_use != 0.
Additionally, the error handling code didn't take into account that we
dropped the inode lock which might require more cleanup.

Luckily, all the cleanup code we need is already there and can be shared
with reserve_metadata_bytes, which is exactly what this patch does.

Reported-by: Lev Vainblat <lev@zadarastorage.com>
Signed-off-by: Jan Schmidt <list.btrfs@jan-o-sch.net>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/josef/btrfs-next.git for-chris...
Chris Mason [Wed, 6 Feb 2013 00:24:44 +0000 (19:24 -0500)]
Merge git://git./linux/kernel/git/josef/btrfs-next.git for-chris into for-linus

11 years agoBtrfs: fix possible stale data exposure
Josef Bacik [Wed, 30 Jan 2013 19:31:31 +0000 (14:31 -0500)]
Btrfs: fix possible stale data exposure

We specifically do not update the disk i_size if there are ordered extents
outstanding for any area between the current disk_i_size and our ordered
extent so that we do not expose stale data.  The problem is the check we
have only checks if the ordered extent starts at or after the current
disk_i_size, which doesn't take into account an ordered extent that starts
before the current disk_i_size and ends past the disk_i_size.  Fix this by
checking if the extent ends past the disk_i_size.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: fix missing i_size update
Josef Bacik [Wed, 30 Jan 2013 19:17:31 +0000 (14:17 -0500)]
Btrfs: fix missing i_size update

If we have an ordered extent before the ordered extent we are currently
completing that is after the current disk_i_size we will put our i_size
update into that ordered extent so that we do not expose stale data.  The
problem is that if our disk i_size is updated past the previous ordered
extent we won't update the i_size with the pending i_size update.  So check
the pending i_size update and if its above the current disk i_size we need
to go ahead and try to update.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: fix race between snapshot deletion and getting inode
Liu Bo [Tue, 29 Jan 2013 03:22:10 +0000 (03:22 +0000)]
Btrfs: fix race between snapshot deletion and getting inode

While running snapshot testscript created by Mitch and David,
the race between autodefrag and snapshot deletion can lead to
corruption of dead_root list so that we can get crash on
btrfs_clean_old_snapshots().

And besides autodefrag, scrub also does the same thing, ie. read
root first and get inode.

Here is the story(take autodefrag as an example):
(1) when we delete a snapshot or subvolume, it will set its root's
refs to zero and do a iput() on its own inode, and if this inode happens
to be the only active in-meory one in root's inode rbtree, it will add
itself to the global dead_roots list for later cleanup.

(2) after (1), the autodefrag thread may read another inode for defrag
and the inode is just in the deleted snapshot/subvolume, but all of these
are without checking if the root is still valid(refs > 0).  So the end up
result is adding the deleted snapshot/subvolume's root to the global
dead_roots list AGAIN.

Fortunately, we already have a srcu lock to avoid the race, ie. subvol_srcu.

So all we need to do is to take the lock to protect 'read root and get inode',
since we synchronize to wait for the rcu grace period before adding something
to the global dead_roots list.

Reported-by: Mitch Harder <mitch.harder@sabayonlinux.org>
Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: fix missing release of the space/qgroup reservation in start_transaction()
Miao Xie [Mon, 28 Jan 2013 12:36:22 +0000 (12:36 +0000)]
Btrfs: fix missing release of the space/qgroup reservation in start_transaction()

When we fail to start a transaction, we need to release the reserved free space
and qgroup space, fix it.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Reviewed-by: Jan Schmidt <list.btrfs@jan-o-sch.net>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: fix wrong sync_writers decrement in btrfs_file_aio_write()
Miao Xie [Mon, 28 Jan 2013 12:34:55 +0000 (12:34 +0000)]
Btrfs: fix wrong sync_writers decrement in btrfs_file_aio_write()

If the checks at the beginning of btrfs_file_aio_write() fail, we needn't
decrease ->sync_writers, because we have not increased it. Fix it.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: do not merge logged extents if we've removed them from the tree
Josef Bacik [Mon, 28 Jan 2013 14:45:20 +0000 (09:45 -0500)]
Btrfs: do not merge logged extents if we've removed them from the tree

You can run into this problem where if somebody is fsyncing and writing out
the existing extents you will have removed the extent map from the em tree,
but it's still valid for the current fsync so we go ahead and write it.  The
problem is we unconditionally try to merge it back into the em tree, but if
we've removed it from the em tree that will cause use after free problems.
Fix this to only merge if we are still a part of the tree.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agobtrfs: don't try to notify udev about missing devices
Eric Sandeen [Thu, 31 Jan 2013 00:55:02 +0000 (00:55 +0000)]
btrfs: don't try to notify udev about missing devices

If we remove a missing device, bdev is null, and if we
send that off to btrfs_kobject_uevent we'll panic.

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: fix repeated delalloc work allocation
Miao Xie [Tue, 22 Jan 2013 10:49:00 +0000 (10:49 +0000)]
Btrfs: fix repeated delalloc work allocation

btrfs_start_delalloc_inodes() locks the delalloc_inodes list, fetches the
first inode, unlocks the list, triggers btrfs_alloc_delalloc_work/
btrfs_queue_worker for this inode, and then it locks the list, checks the
head of the list again. But because we don't delete the first inode that it
deals with before, it will fetch the same inode. As a result, this function
allocates a huge amount of btrfs_delalloc_work structures, and OOM happens.

Fix this problem by splice this delalloc list.

Reported-by: Alex Lyakas <alex.btrfs@zadarastorage.com>
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: fix wrong max device number for single profile
Miao Xie [Wed, 16 Jan 2013 11:27:17 +0000 (11:27 +0000)]
Btrfs: fix wrong max device number for single profile

The max device number of single profile is 1, not 0 (0 means 'as many as
possible'). Fix it.

Cc: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Reviewed-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: fix missed transaction->aborted check
Miao Xie [Tue, 15 Jan 2013 06:29:12 +0000 (06:29 +0000)]
Btrfs: fix missed transaction->aborted check

First, though the current transaction->aborted check can stop the commit early
and avoid unnecessary operations, it is too early, and some transaction handles
don't end, those handles may set transaction->aborted after the check.

Second, when we commit the transaction, we will wake up some worker threads to
flush the space cache and inode cache. Those threads also allocate some transaction
handles and may set transaction->aborted if some serious error happens.

So we need more check for ->aborted when committing the transaction. Fix it.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: Add ACCESS_ONCE() to transaction->abort accesses
Miao Xie [Tue, 15 Jan 2013 06:27:25 +0000 (06:27 +0000)]
Btrfs: Add ACCESS_ONCE() to transaction->abort accesses

We may access and update transaction->aborted on the different CPUs without
lock, so we need ACCESS_ONCE() wrapper to prevent the compiler from creating
unsolicited accesses and make sure we can get the right value.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: put csums on the right ordered extent
Josef Bacik [Tue, 22 Jan 2013 20:43:09 +0000 (15:43 -0500)]
Btrfs: put csums on the right ordered extent

I noticed a WARN_ON going off when adding csums because we were going over
the amount of csum bytes that should have been allowed for an ordered
extent.  This is a leftover from when we used to hold the csums privately
for direct io, but now we use the normal ordered sum stuff so we need to
make sure and check if we've moved on to another extent so that the csums
are added to the right extent.  Without this we could end up with csums for
bytenrs that don't have extents to cover them yet.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: use right range to find checksum for compressed extents
Liu Bo [Sun, 6 Jan 2013 03:38:22 +0000 (03:38 +0000)]
Btrfs: use right range to find checksum for compressed extents

For compressed extents, the range of checksum is covered by disk length,
and the disk length is different with ram length, so we need to use disk
length instead to get us the right checksum.

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: fix panic when recovering tree log
Josef Bacik [Tue, 18 Dec 2012 16:39:19 +0000 (11:39 -0500)]
Btrfs: fix panic when recovering tree log

A user reported a BUG_ON(ret) that occured during tree log replay.  Ret was
-EAGAIN, so what I think happened is that we removed an extent that covered
a bitmap entry and an extent entry.  We remove the part from the bitmap and
return -EAGAIN and then search for the next piece we want to remove, which
happens to be an entire extent entry, so we just free the sucker and return.
The problem is ret is still set to -EAGAIN so we trip the BUG_ON().  The
user used btrfs-zero-log so I'm not 100% sure this is what happened so I've
added a WARN_ON() to catch the other possibility.  Thanks,

Reported-by: Jan Steffens <jan.steffens@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: do not allow logged extents to be merged or removed
Josef Bacik [Thu, 24 Jan 2013 17:02:07 +0000 (12:02 -0500)]
Btrfs: do not allow logged extents to be merged or removed

We drop the extent map tree lock while we're logging extents, so somebody
could come in and merge another extent into this one and screw up our
logging, or they could even remove us from the list which would keep us from
logging the extent or freeing our ref on it, so we need to make sure to not
clear LOGGING until after the extent is logged, and then we can merge it to
adjacent extents.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: fix a regression in balance usage filter
Ilya Dryomov [Mon, 21 Jan 2013 13:15:56 +0000 (15:15 +0200)]
Btrfs: fix a regression in balance usage filter

Commit 3fed40cc ("Btrfs: cleanup duplicated division functions"), which
was merged into 3.8-rc1, has introduced a regression by removing logic
that was guarding us against bad user input.  Bring it back.

Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoMerge branch 'mutex-ops@next-for-chris' of git://github.com/idryomov/btrfs-unstable...
Chris Mason [Tue, 22 Jan 2013 01:39:06 +0000 (20:39 -0500)]
Merge branch 'mutex-ops@next-for-chris' of git://github.com/idryomov/btrfs-unstable into linus

11 years agoMerge branch 'for-chris' of git://git.kernel.org/pub/scm/linux/kernel/git/josef/btrfs...
Chris Mason [Tue, 22 Jan 2013 01:26:55 +0000 (20:26 -0500)]
Merge branch 'for-chris' of git://git./linux/kernel/git/josef/btrfs-next into linus

11 years agoBtrfs: prevent qgroup destroy when there are still relations
Arne Jansen [Thu, 17 Jan 2013 08:22:09 +0000 (01:22 -0700)]
Btrfs: prevent qgroup destroy when there are still relations

Currently you can just destroy a qgroup even though it is in use by other qgroups
or has qgroups assigned to it. This patch prevents destruction of qgroups unless
they are completely unused. Otherwise destroy will return EBUSY.

Reported-by: Eric Hopper <hopper@omnifarious.org>
Signed-off-by: Arne Jansen <sensille@gmx.net>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: ignore orphan qgroup relations
Arne Jansen [Thu, 17 Jan 2013 08:22:08 +0000 (01:22 -0700)]
Btrfs: ignore orphan qgroup relations

If a qgroup that has still assignments is deleted by the user, the corresponding
relations are left in the tree. This leads to an unmountable filesystem.
With this patch, those relations are simple ignored.

Reported-by: Eric Hopper <hopper@omnifarious.org>
Signed-off-by: Arne Jansen <sensille@gmx.net>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: reorder locks and sanity checks in btrfs_ioctl_defrag
Ilya Dryomov [Sun, 20 Jan 2013 13:57:57 +0000 (15:57 +0200)]
Btrfs: reorder locks and sanity checks in btrfs_ioctl_defrag

Operation-specific check (whether subvol is readonly or not) should go
after the mutual exclusiveness check.

Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
11 years agoBtrfs: fix unlock order in btrfs_ioctl_rm_dev
Ilya Dryomov [Sun, 20 Jan 2013 13:57:57 +0000 (15:57 +0200)]
Btrfs: fix unlock order in btrfs_ioctl_rm_dev

Fix unlock order in btrfs_ioctl_rm_dev().

Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
11 years agoBtrfs: fix unlock order in btrfs_ioctl_resize
Ilya Dryomov [Sun, 20 Jan 2013 13:57:57 +0000 (15:57 +0200)]
Btrfs: fix unlock order in btrfs_ioctl_resize

Fix unlock order in btrfs_ioctl_resize().

Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
11 years agoBtrfs: fix "mutually exclusive op is running" error code
Ilya Dryomov [Sun, 20 Jan 2013 13:57:57 +0000 (15:57 +0200)]
Btrfs: fix "mutually exclusive op is running" error code

The error code that is returned in response to starting a mutually
exclusive operation when there is one already running got silently
changed from EINVAL to EINPROGRESS by 5ac00add.  Returning EINPROGRESS
to, say, add_dev, when rm_dev is running is misleading.  Furthermore,
the operation itself may want to use EINPROGRESS for other purposes.

Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
11 years agoBtrfs: bring back balance pause/resume logic
Ilya Dryomov [Sun, 20 Jan 2013 13:57:57 +0000 (15:57 +0200)]
Btrfs: bring back balance pause/resume logic

Balance pause/resume logic got broken by 5ac00add (went in into 3.8-rc1
as part of dev-replace merge).  Offending commit took a stab at making
mutually exclusive volume operations (add_dev, rm_dev, resize, balance,
replace_dev) not block behind volume_mutex if another such operation is
in progress and instead return an error right away.  Balancing front-end
relied on the blocking behaviour, so the fix is ugly, but short of a
complete rework, it's the best we can do.

Reported-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
11 years agobtrfs: update timestamps on truncate()
Eric Sandeen [Sat, 12 Jan 2013 02:57:22 +0000 (02:57 +0000)]
btrfs: update timestamps on truncate()

truncate() vs. ftruncate() differ in the VFS; truncate()
doesn't set (ATTR_CTIME | ATTR_MTIME), and it's up to the
fs to do the timestamp updates if the size changes.

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
11 years agobtrfs: fix btrfs_cont_expand() freeing IS_ERR em
Zach Brown [Tue, 8 Jan 2013 19:37:58 +0000 (19:37 +0000)]
btrfs: fix btrfs_cont_expand() freeing IS_ERR em

btrfs_cont_expand() tries to free an IS_ERR em as it gets an error from
btrfs_get_extent() and breaks out of its loop.

An instance of -EEXIST was reported in the wild:

  https://bugzilla.redhat.com/show_bug.cgi?id=874407

I have no idea if that -EEXIST is surprising, or not.  Regardless, this
error handling should be cleaned up to handle other reasonable errors
(ENOMEM, EIO; whatever).

This seemed to be the only buggy freeing of the relatively rare IS_ERR
em so I opted to fix the caller rather than teach free_extent_map() to
use IS_ERR_OR_NULL().

Signed-off-by: Zach Brown <zab@redhat.com>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
11 years agoBtrfs: fix a bug when llseek for delalloc bytes behind prealloc extents
Liu Bo [Mon, 7 Jan 2013 10:10:12 +0000 (10:10 +0000)]
Btrfs: fix a bug when llseek for delalloc bytes behind prealloc extents

xfstests case 285 complains.

It it because btrfs did not try to find unwritten delalloc
bytes(only dirty pages, not yet writeback) behind prealloc
extents, it ends up finding nothing while we're with SEEK_DATA.

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: fix off-by-one in lseek
Liu Bo [Mon, 7 Jan 2013 03:53:08 +0000 (03:53 +0000)]
Btrfs: fix off-by-one in lseek

Lock end is inclusive.

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: reset path lock state to zero
Liu Bo [Fri, 28 Dec 2012 09:33:19 +0000 (09:33 +0000)]
Btrfs: reset path lock state to zero

We forgot to reset the path lock state to zero after we unlock the path block,
and this can lead to the ASSERT checker in tree unlock API.

Reported-by: Slava Barinov <rayslava@gmail.com>
Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: let allocation start from the right raid type
Liu Bo [Thu, 27 Dec 2012 09:01:24 +0000 (09:01 +0000)]
Btrfs: let allocation start from the right raid type

This'd avoid us empty looping.

Say we have only one disk and the metadata raid type will be defaultly DUP,
and we do not need to start from index=0(RAID10) and get over two empty
loops to index=2(DUP).

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: add orphan before truncating pagecache
Josef Bacik [Mon, 7 Jan 2013 22:03:21 +0000 (17:03 -0500)]
Btrfs: add orphan before truncating pagecache

Running xfstests 83 in a loop would sometimes fail the fsck.  This happens
because if we invalidate a page that already has an ordered extent setup for
it we will complete the ordered extent ourselves, assuming that the truncate
will clean everything up.  The problem with this is there is plenty of time
for the truncate to fail after we've done this work.  So to fix this we need
to add the orphan item first to make sure the cleanup gets done properly,
and then we can truncate the pagecache and all that stuff and be safe.  This
fixes the btrfsck failures I was seeing while running 83 in a loop.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: set flushing if we're limited flushing
Josef Bacik [Tue, 18 Dec 2012 20:16:34 +0000 (15:16 -0500)]
Btrfs: set flushing if we're limited flushing

We still need to say we're flushing if we're limit flushing to keep somebody
from coming in and stealing our reservation.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: fix missing write access release in btrfs_ioctl_resize()
Miao Xie [Fri, 21 Dec 2012 10:38:50 +0000 (10:38 +0000)]
Btrfs: fix missing write access release in btrfs_ioctl_resize()

We forget to give up the write access after we find some device operation
is going on. Fix it.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: fix resize a readonly device
Miao Xie [Fri, 21 Dec 2012 09:19:51 +0000 (09:19 +0000)]
Btrfs: fix resize a readonly device

We should not resize a readonly device, fix it.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
11 years agoBtrfs: do not delete a subvolume which is in a R/O subvolume
Miao Xie [Mon, 22 Oct 2012 11:39:53 +0000 (11:39 +0000)]
Btrfs: do not delete a subvolume which is in a R/O subvolume

Step to reproduce:
 # mkfs.btrfs <disk>
 # mount <disk> <mnt>
 # btrfs sub create <mnt>/subv0
 # btrfs sub snap <mnt> <mnt>/subv0/snap0
 # change <mnt>/subv0 from R/W to R/O
 # btrfs sub del <mnt>/subv0/snap0

We deleted the snapshot successfully. I think we should not be able to delete
the snapshot since the parent subvolume is R/O.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
11 years agoBtrfs: disable qgroup id 0
Miao Xie [Thu, 15 Nov 2012 11:35:41 +0000 (11:35 +0000)]
Btrfs: disable qgroup id 0

Qgroup id 0 is a special number, we should set the id of a qgroup to 0.
Fix it.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
11 years agobtrfs: get the device in write mode when deleting it
Lukas Czerner [Fri, 7 Dec 2012 10:09:19 +0000 (10:09 +0000)]
btrfs: get the device in write mode when deleting it

When we're deleting the device we should get it in write mode since
we're going to re-write the super block magic on that device. And it
should fail if the device is read-only.

Signed-off-by: Lukas Czerner <lczerner@redhat.com>
11 years agoBtrfs: fix memory leak in name_cache_insert()
Tsutomu Itoh [Mon, 17 Dec 2012 06:38:51 +0000 (06:38 +0000)]
Btrfs: fix memory leak in name_cache_insert()

We should free name_cache_entry before returning from the
error handling code.

Signed-off-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com>
11 years agoRevert "Btrfs: reorder tree mod log operations in deleting a pointer"
Chris Mason [Wed, 19 Dec 2012 00:35:32 +0000 (19:35 -0500)]
Revert "Btrfs: reorder tree mod log operations in deleting a pointer"

This reverts commit 6a7a665d78c5dd8bc76a010648c4e7d84517ab5a.

This was bug was fixed differently in 3.6, so this commit
isn't needed.

Conflicts:
fs/btrfs/ctree.c

Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoRevert "Btrfs: MOD_LOG_KEY_REMOVE_WHILE_MOVING never change node's nritems"
Chris Mason [Tue, 18 Dec 2012 20:43:18 +0000 (15:43 -0500)]
Revert "Btrfs: MOD_LOG_KEY_REMOVE_WHILE_MOVING never change node's nritems"

This reverts commit 95c80bb1f6b24b57058d971ed252b2c1c5121b51.

The bug addressed by this commit was fixed differently back in 3.6

Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: fix a bug of per-file nocow
Liu Bo [Tue, 11 Sep 2012 14:33:50 +0000 (08:33 -0600)]
Btrfs: fix a bug of per-file nocow

Users report a bug, the reproducer is:
$ mkfs.btrfs /dev/loop0
$ mount /dev/loop0 /mnt/btrfs/
$ mkdir /mnt/btrfs/dir
$ chattr +C /mnt/btrfs/dir/
$ dd if=/dev/zero of=/mnt/btrfs/dir/foo bs=4K count=10;
$ lsattr /mnt/btrfs/dir/foo
---------------C- /mnt/btrfs/dir/foo
$ filefrag /mnt/btrfs/dir/foo
/mnt/btrfs/dir/foo: 1 extent found    ---> an extent
$ dd if=/dev/zero of=/mnt/btrfs/dir/foo bs=4K count=1 seek=5 conv=notrunc,nocreat; sync
$ filefrag /mnt/btrfs/dir/foo
/mnt/btrfs/dir/foo: 3 extents found   ---> with nocow, btrfs breaks the extent into three parts

The new created file should not only inherit the NODATACOW flag, but also
honor NODATASUM flag, because we must do COW on a file extent with checksum.

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: fix hash overflow handling
Chris Mason [Mon, 17 Dec 2012 19:26:57 +0000 (14:26 -0500)]
Btrfs: fix hash overflow handling

The handling for directory crc hash overflows was fairly obscure,
split_leaf returns EOVERFLOW when we try to extend the item and that is
supposed to bubble up to userland.  For a while it did so, but along the
way we added better handling of errors and forced the FS readonly if we
hit IO errors during the directory insertion.

Along the way, we started testing only for EEXIST and the EOVERFLOW case
was dropped.  The end result is that we may force the FS readonly if we
catch a directory hash bucket overflow.

This fixes a few problem spots.  First I add tests for EOVERFLOW in the
places where we can safely just return the error up the chain.

btrfs_rename is harder though, because it tries to insert the new
directory item only after it has already unlinked anything the rename
was going to overwrite.  Rather than adding very complex logic, I added
a helper to test for the hash overflow case early while it is still safe
to bail out.

Snapshot and subvolume creation had a similar problem, so they are using
the new helper now too.

Signed-off-by: Chris Mason <chris.mason@fusionio.com>
Reported-by: Pascal Junod <pascal@junod.info>
11 years agoBtrfs: don't take inode delalloc mutex if we're a free space inode
Josef Bacik [Fri, 14 Dec 2012 18:48:14 +0000 (13:48 -0500)]
Btrfs: don't take inode delalloc mutex if we're a free space inode

This confuses and angers lockdep even though it's ok.  We don't really need
the lock for free space inodes since only the transaction committer will be
reserving space.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: fix autodefrag and umount lockup
Josef Bacik [Fri, 14 Dec 2012 18:46:43 +0000 (13:46 -0500)]
Btrfs: fix autodefrag and umount lockup

This happens because writeback_inodes_sb_nr_if_idle does down_read.  This
doesn't work for us and it has not been fixed upstream yet, so do it
ourselves and use that instead so we can stop having this stupid long
standing lockup.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: fix permissions of empty files not affected by umask
Filipe Brandenburger [Fri, 30 Nov 2012 03:40:08 +0000 (03:40 +0000)]
Btrfs: fix permissions of empty files not affected by umask

When a new file is created with btrfs_create(), the inode will initially be
created with permissions 0666 and later on in btrfs_init_acl() it will be
adapted to mask out the umask bits. The problem is that this change won't make
it into the btrfs_inode unless there's another change to the inode (e.g. writing
content changing the size or touching the file changing the mtime.)

This fix adds a call to btrfs_update_inode() to btrfs_create() to make sure that
the change will not get lost if the in-memory inode is flushed before other
changes are made to the file.

Signed-off-by: Filipe Brandenburger <filbranden@google.com>
Reviewed-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: put raid properties into global table
Liu Bo [Wed, 21 Nov 2012 14:18:10 +0000 (14:18 +0000)]
Btrfs: put raid properties into global table

Raid properties can be shared among raid calculation code, we can put
them into a global table to keep it simple.

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: fix BUG() in scrub when first superblock reading gives EIO
Stefan Behrens [Wed, 14 Nov 2012 18:57:29 +0000 (18:57 +0000)]
Btrfs: fix BUG() in scrub when first superblock reading gives EIO

This fixes a very special case that can be reproduced by just
disconnecting a disk at runtime, and without unmounting the
filesystem first, start scrub on the filesystem with the
disconnected disk. All read and write EIOs are handled
correctly, only the first superblock is an exception and gives
a BUG() in a subfunction. The BUG() is correct, it would crash
later otherwise. The subfunction must not be called for
superblocks and this is what the fix changes.

Reported-by: Joeri Vanthienen <mail@joerivanthienen.be>
Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: do not call file_update_time in aio_write
Josef Bacik [Fri, 9 Nov 2012 15:53:21 +0000 (10:53 -0500)]
Btrfs: do not call file_update_time in aio_write

This starts a transaction and dirties the inode everytime we call it, which
is super expensive if you have a write heavy workload.  We will be updating
the inode when the IO completes and we reserve the space for the inode
update when we reserve space for the write, so there is no chance of loss of
information or enospc issues.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: only unlock and relock if we have to
Josef Bacik [Wed, 7 Nov 2012 18:44:13 +0000 (13:44 -0500)]
Btrfs: only unlock and relock if we have to

I noticed while doing fsync tests that we were always dropping the path and
re-searching when we first cow the log root even though we've already gotten
the write lock on the root.  That's because we don't take into account that
there might not be a parent node, so fix the check to make sure there is
actually a parent node before we undo all of this work for nothing.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>