platform/kernel/linux-rpi.git
11 years agoBtrfs: fix printing of non NULL terminated string
Filipe David Borba Manana [Tue, 20 Aug 2013 16:51:48 +0000 (17:51 +0100)]
Btrfs: fix printing of non NULL terminated string

The name buffer is not terminated by a '\0' character,
therefore it needs to be printed with %.*s and use the
length of the buffer.

Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: Use %z to format size_t
Geert Uytterhoeven [Tue, 20 Aug 2013 11:20:18 +0000 (13:20 +0200)]
Btrfs: Use %z to format size_t

Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: Do not truncate sector_t on 32-bit with CONFIG_LBDAF=y
Geert Uytterhoeven [Tue, 20 Aug 2013 11:20:17 +0000 (13:20 +0200)]
Btrfs: Do not truncate sector_t on 32-bit with CONFIG_LBDAF=y

sector_t may be either "u64" (always 64 bit) or "unsigned long" (32 or 64
bit).  Casting it to "unsigned long" will truncate it on 32-bit platforms
where CONFIG_LBDAF=y.

Cast to "unsigned long long" and format using "ll" instead.

Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: PAGE_CACHE_SIZE is already unsigned long
Geert Uytterhoeven [Tue, 20 Aug 2013 11:20:16 +0000 (13:20 +0200)]
Btrfs: PAGE_CACHE_SIZE is already unsigned long

PAGE_CACHE_SIZE == PAGE_SIZE is "unsigned long" everywhere, so there's no
need to cast it to "unsigned long".

Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: Make btrfs_header_chunk_tree_uuid() return unsigned long
Geert Uytterhoeven [Tue, 20 Aug 2013 11:20:15 +0000 (13:20 +0200)]
Btrfs: Make btrfs_header_chunk_tree_uuid() return unsigned long

Internally, btrfs_header_chunk_tree_uuid() calculates an unsigned long, but
casts it to a pointer, while all callers cast it to unsigned long again.

Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: Make btrfs_header_fsid() return unsigned long
Geert Uytterhoeven [Tue, 20 Aug 2013 11:20:14 +0000 (13:20 +0200)]
Btrfs: Make btrfs_header_fsid() return unsigned long

Internally, btrfs_header_fsid() calculates an unsigned long, but casts
it to a pointer, while all callers cast it to unsigned long again.

Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: Make btrfs_dev_extent_chunk_tree_uuid() return unsigned long
Geert Uytterhoeven [Tue, 20 Aug 2013 11:20:13 +0000 (13:20 +0200)]
Btrfs: Make btrfs_dev_extent_chunk_tree_uuid() return unsigned long

Internally, btrfs_dev_extent_chunk_tree_uuid() calculates an unsigned long,
but casts it to a pointer, while all callers cast it to unsigned long
again.

Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: Make btrfs_device_fsid() return unsigned long
Geert Uytterhoeven [Tue, 20 Aug 2013 11:20:12 +0000 (13:20 +0200)]
Btrfs: Make btrfs_device_fsid() return unsigned long

All callers of btrfs_device_fsid() cast its return type to unsigned long.

Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: Make btrfs_device_uuid() return unsigned long
Geert Uytterhoeven [Tue, 20 Aug 2013 11:20:11 +0000 (13:20 +0200)]
Btrfs: Make btrfs_device_uuid() return unsigned long

All callers of btrfs_device_uuid() cast its return type to unsigned long.

Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: Format mirror_num as int
Geert Uytterhoeven [Tue, 20 Aug 2013 11:20:10 +0000 (13:20 +0200)]
Btrfs: Format mirror_num as int

mirror_num is always "int", hence don't cast it to "unsigned long long" and
format it as a 64-bit number.

Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: Format PAGE_SIZE as unsigned long
Geert Uytterhoeven [Tue, 20 Aug 2013 11:20:09 +0000 (13:20 +0200)]
Btrfs: Format PAGE_SIZE as unsigned long

PAGE_SIZE is "unsigned long" everywhere, so there's no need to cast it to
"unsigned long long" and format it as a 64-bit number.

Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: Make BTRFS_DEV_REPLACE_DEVID an unsigned long long constant
Geert Uytterhoeven [Tue, 20 Aug 2013 11:20:08 +0000 (13:20 +0200)]
Btrfs: Make BTRFS_DEV_REPLACE_DEVID an unsigned long long constant

The internal btrfs device id is a u64, hence make the constant
BTRFS_DEV_REPLACE_DEVID "unsigned long long" as well, so we no longer need
a cast to print it.

Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: Remove superfluous casts from u64 to unsigned long long
Geert Uytterhoeven [Tue, 20 Aug 2013 11:20:07 +0000 (13:20 +0200)]
Btrfs: Remove superfluous casts from u64 to unsigned long long

u64 is "unsigned long long" on all architectures now, so there's no need to
cast it when formatting it using the "ll" length modifier.

Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: fix memory leak of orphan block rsv
Filipe David Borba Manana [Mon, 19 Aug 2013 23:52:51 +0000 (00:52 +0100)]
Btrfs: fix memory leak of orphan block rsv

This issue is simple to reproduce and observe if kmemleak is enabled.
Two simple ways to reproduce it:

** 1

$ mkfs.btrfs -f /dev/loop0
$ mount /dev/loop0 /mnt/btrfs
$ btrfs balance start /mnt/btrfs
$ umount /mnt/btrfs

** 2

$ mkfs.btrfs -f /dev/loop0
$ mount /dev/loop0 /mnt/btrfs
$ touch /mnt/btrfs/foobar
$ rm -f /mnt/btrfs/foobar
$ umount /mnt/btrfs

After a while, kmemleak reports the leak:

$ cat /sys/kernel/debug/kmemleak
unreferenced object 0xffff880402b13e00 (size 128):
  comm "btrfs", pid 19621, jiffies 4341648183 (age 70057.844s)
  hex dump (first 32 bytes):
    00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
    00 fc c6 b1 04 88 ff ff 04 00 04 00 ad 4e ad de  .............N..
  backtrace:
    [<ffffffff817275a6>] kmemleak_alloc+0x26/0x50
    [<ffffffff8117832b>] kmem_cache_alloc_trace+0xeb/0x1d0
    [<ffffffffa04db499>] btrfs_alloc_block_rsv+0x39/0x70 [btrfs]
    [<ffffffffa04f8bad>] btrfs_orphan_add+0x13d/0x1b0 [btrfs]
    [<ffffffffa04e2b13>] btrfs_remove_block_group+0x143/0x500 [btrfs]
    [<ffffffffa0518158>] btrfs_relocate_chunk.isra.63+0x618/0x790 [btrfs]
    [<ffffffffa051bc27>] btrfs_balance+0x8f7/0xe90 [btrfs]
    [<ffffffffa05240a0>] btrfs_ioctl_balance+0x250/0x550 [btrfs]
    [<ffffffffa05269ca>] btrfs_ioctl+0xdfa/0x25f0 [btrfs]
    [<ffffffff8119c936>] do_vfs_ioctl+0x96/0x570
    [<ffffffff8119cea1>] SyS_ioctl+0x91/0xb0
    [<ffffffff81750242>] system_call_fastpath+0x16/0x1b
    [<ffffffffffffffff>] 0xffffffffffffffff

This affects btrfs-next, revision be8e3cd00d7293dd177e3f8a4a1645ce09ca3acb
(Btrfs: separate out tests into their own directory).

Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: rollback btrfs_device fields on umount
Ilya Dryomov [Mon, 12 Aug 2013 11:33:04 +0000 (14:33 +0300)]
Btrfs: rollback btrfs_device fields on umount

It turns out we don't properly rollback in-core btrfs_device state on
umount.  We zero out ->bdev, ->in_fs_metadata and that's about it.  In
particular, we don't zero out ->generation, and this can lead to us
refusing a mount -- a non-NULL fs_devices->latest_bdev is essential, but
btrfs_close_extra_devices will happily assign NULL to ->latest_bdev if
the first device on the dev_list happens to be missing and consequently
has no bdev attached.  This happens because since commit a6b0d5c8
btrfs_close_extra_devices adjusts ->latest_bdev, and in doing that,
relies on the ->generation.  Fix this, and possibly other problems, by
zeroing out everything except for what device_list_add sets, so that a
mount right after insmod and 'btrfs dev scan' is no different from any
later mount in this respect.

Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: add alloc_fs_devices and switch to it
Ilya Dryomov [Mon, 12 Aug 2013 11:33:03 +0000 (14:33 +0300)]
Btrfs: add alloc_fs_devices and switch to it

In the spirit of btrfs_alloc_device, add a helper for allocating and
doing some common initialization of btrfs_fs_devices struct.

Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: add btrfs_alloc_device and switch to it
Ilya Dryomov [Fri, 23 Aug 2013 10:20:17 +0000 (13:20 +0300)]
Btrfs: add btrfs_alloc_device and switch to it

Currently btrfs_device is allocated ad-hoc in a few different places,
and as a result not all fields are initialized properly.  In particular,
readahead state is only initialized in device_list_add (at scan time),
and not in btrfs_init_new_device (when the new device is added with
'btrfs dev add').  Fix this by adding an allocation helper and switch
everybody but __btrfs_close_devices to it.  (__btrfs_close_devices is
dealt with in a later commit.)

Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: find_next_devid: root -> fs_info
Ilya Dryomov [Mon, 12 Aug 2013 11:33:01 +0000 (14:33 +0300)]
Btrfs: find_next_devid: root -> fs_info

find_next_devid() knows which root to search, so it should take an
fs_info instead of an arbitrary root.

Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: don't allow the replace procedure on read only filesystems
Stefan Behrens [Mon, 19 Aug 2013 16:51:13 +0000 (18:51 +0200)]
Btrfs: don't allow the replace procedure on read only filesystems

If you start the replace procedure on a read only filesystem, at
the end the procedure fails to write the updated dev_items to the
chunk tree. The problem is that this error is not indicated except
for a WARN_ON(). If the user now thinks that everything was done
as expected and destroys the source device (with mkfs or with a
hammer). The next mount fails with "failed to read chunk root" and
the filesystem is gone.

This commit adds code to fail the attempt to start the replace
procedure if the filesystem is mounted read-only.

Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de>
Cc: <stable@vger.kernel.org> # 3.10+
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: use __u64 in exported user headers
Mike Frysinger [Fri, 16 Aug 2013 18:02:33 +0000 (14:02 -0400)]
Btrfs: use __u64 in exported user headers

Signed-off-by: Mike Frysinger <vapier@gentoo.org>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: reset force_compress on btrfs_file_defrag failure
Filipe David Borba Manana [Fri, 16 Aug 2013 14:23:33 +0000 (15:23 +0100)]
Btrfs: reset force_compress on btrfs_file_defrag failure

After we set force_compress with a new value (which was not being done
while holding the inode mutex), if an error happens and we jump to
the label out_ra, the force_compress property of the inode is not set
to BTRFS_COMPRESS_NONE (unlike in the case where no errors happen).

Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: add mount option to force UUID tree checking
Stefan Behrens [Thu, 15 Aug 2013 15:11:24 +0000 (17:11 +0200)]
Btrfs: add mount option to force UUID tree checking

This should never be needed, but since all functions are there
to check and rebuild the UUID tree, a mount option is added that
allows to force this check and rebuild procedure.

Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: check UUID tree during mount if required
Stefan Behrens [Thu, 15 Aug 2013 15:11:23 +0000 (17:11 +0200)]
Btrfs: check UUID tree during mount if required

If the filesystem was mounted with an old kernel that was not
aware of the UUID tree, this is detected by looking at the
uuid_tree_generation field of the superblock (similar to how
the free space cache is doing it). If a mismatch is detected
at mount time, a thread is started that does two things:
1. Iterate through the UUID tree, check each entry, delete those
   entries that are not valid anymore (i.e., the subvol does not
   exist anymore or the value changed).
2. Iterate through the root tree, for each found subvolume, add
   the UUID tree entries for the subvolume (if they are not
   already there).

This mechanism is also used to handle and repair errors that
happened during the initial creation and filling of the tree.
The update of the uuid_tree_generation field (which indicates
that the state of the UUID tree is up to date) is blocked until
all create and repair operations are successfully completed.

Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: introduce uuid-tree-gen field
Stefan Behrens [Thu, 15 Aug 2013 15:11:22 +0000 (17:11 +0200)]
Btrfs: introduce uuid-tree-gen field

In order to be able to detect the case that a filesystem is mounted
with an old kernel, add a uuid-tree-gen field like the free space
cache is doing it. It is part of the super block and written with
each commit. Old kernels do not know this field and don't update it.

Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: fill UUID tree initially
Stefan Behrens [Thu, 15 Aug 2013 15:11:21 +0000 (17:11 +0200)]
Btrfs: fill UUID tree initially

When the UUID tree is initially created, a task is spawned that
walks through the root tree. For each found subvolume root_item,
the uuid and received_uuid entries in the UUID tree are added.
This is such a quick operation so that in case somebody wants
to unmount the filesystem while the task is still running, the
unmount is delayed until the UUID tree building task is finished.

Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: maintain subvolume items in the UUID tree
Stefan Behrens [Thu, 15 Aug 2013 15:11:20 +0000 (17:11 +0200)]
Btrfs: maintain subvolume items in the UUID tree

When a new subvolume or snapshot is created, a new UUID item is added
to the UUID tree. Such items are removed when the subvolume is deleted.
The ioctl to set the received subvolume UUID is also touched and will
now also add this received UUID into the UUID tree together with the
ID of the subvolume. The latter is also done when read-only snapshots
are created which inherit all the send/receive information from the
parent subvolume.

User mode programs use the BTRFS_IOC_TREE_SEARCH ioctl to search and
read in the UUID tree.

Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: create UUID tree if required
Stefan Behrens [Thu, 15 Aug 2013 15:11:19 +0000 (17:11 +0200)]
Btrfs: create UUID tree if required

This tree is not created by mkfs.btrfs. Therefore when a filesystem
is mounted writable and the UUID tree does not exist, this tree is
created if required. The tree is also added to the fs_info structure
and initialized, but this commit does not yet read or write UUID tree
elements.

Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: support printing UUID tree elements
Stefan Behrens [Thu, 15 Aug 2013 15:11:18 +0000 (17:11 +0200)]
Btrfs: support printing UUID tree elements

This commit adds support to print UUID tree elements to print-tree.c.

Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: introduce a tree for items that map UUIDs to something
Stefan Behrens [Thu, 15 Aug 2013 15:11:17 +0000 (17:11 +0200)]
Btrfs: introduce a tree for items that map UUIDs to something

Mapping UUIDs to subvolume IDs is an operation with a high effort
today. Today, the algorithm even has quadratic effort (based on the
number of existing subvolumes), which means, that it takes minutes
to send/receive a single subvolume if 10,000 subvolumes exist. But
even linear effort would be too much since it is a waste. And these
data structures to allow mapping UUIDs to subvolume IDs are created
every time a btrfs send/receive instance is started.

It is much more efficient to maintain a searchable persistent data
structure in the filesystem, one that is updated whenever a
subvolume/snapshot is created and deleted, and when the received
subvolume UUID is set by the btrfs-receive tool.

Therefore kernel code is added with this commit that is able to
maintain data structures in the filesystem that allow to quickly
search for a given UUID and to retrieve data that is assigned to
this UUID, like which subvolume ID is related to this UUID.

This commit adds a new tree to hold UUID-to-data mapping items. The
key of the items is the full UUID plus the key type BTRFS_UUID_KEY.
Multiple data blocks can be stored for a given UUID, a type/length/
value scheme is used.

Now follows the lengthy justification, why a new tree was added
instead of using the existing root tree:

The first approach was to not create another tree that holds UUID
items. Instead, the items should just go into the top root tree.
Unfortunately this confused the algorithm to assign the objectid
of subvolumes and snapshots. The reason is that
btrfs_find_free_objectid() calls btrfs_find_highest_objectid() for
the first created subvol or snapshot after mounting a filesystem,
and this function simply searches for the largest used objectid in
the root tree keys to pick the next objectid to assign. Of course,
the UUID keys have always been the ones with the highest offset
value, and the next assigned subvol ID was wastefully huge.

To use any other existing tree did not look proper. To apply a
workaround such as setting the objectid to zero in the UUID item
key and to implement collision handling would either add
limitations (in case of a btrfs_extend_item() approach to handle
the collisions) or a lot of complexity and source code (in case a
key would be looked up that is free of collisions). Adding new code
that introduces limitations is not good, and adding code that is
complex and lengthy for no good reason is also not good. That's the
justification why a completely new tree was introduced.

Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agobtrfs: mark some local function as 'static'
Sergei Trofimovich [Wed, 14 Aug 2013 20:27:46 +0000 (23:27 +0300)]
btrfs: mark some local function as 'static'

Cc: Josef Bacik <jbacik@fusionio.com>
Cc: Chris Mason <chris.mason@fusionio.com>
Signed-off-by: Sergei Trofimovich <slyfox@gentoo.org>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: get rid of sparse warnings
Stefan Behrens [Wed, 14 Aug 2013 16:12:25 +0000 (18:12 +0200)]
Btrfs: get rid of sparse warnings

make C=2 fs/btrfs/ CF=-D__CHECK_ENDIAN__

I tried to filter out the warnings for which patches have already
been sent to the mailing list, pending for inclusion in btrfs-next.

All these changes should be obviously safe.

Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: don't miss inode ref items in BTRFS_IOC_INO_LOOKUP
Filipe David Borba Manana [Wed, 14 Aug 2013 02:00:21 +0000 (03:00 +0100)]
Btrfs: don't miss inode ref items in BTRFS_IOC_INO_LOOKUP

If the inode ref key was not found and the current leaf slot
was 0 (first item in the leaf) the code would always return
-ENOENT. This was not correct because the desired inode ref
item might be the last item in the previous leaf.

Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: add missing error code to BTRFS_IOC_INO_LOOKUP handler
Filipe David Borba Manana [Wed, 14 Aug 2013 02:00:20 +0000 (03:00 +0100)]
Btrfs: add missing error code to BTRFS_IOC_INO_LOOKUP handler

If the path doesn't fit in the input buffer, return ENAMETOOLONG
instead of returning with a success code (0) and a partially
filled and right justified buffer.

Also removed useless buffer pointer check outside the while loop.

Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: remove reduplicate check when disabling quota
Wang Shilong [Wed, 14 Aug 2013 01:13:38 +0000 (09:13 +0800)]
Btrfs: remove reduplicate check when disabling quota

We have checked 'quota_root' with qgroup_ioctl_lock held before,So
here the check is reduplicate, remove it.

Signed-off-by: Wang Shilong <wangsl.fnst@cn.fujitsu.com>
Reviewed-by: Miao Xie <miaox@cn.fujitsu.com>
Reviewed-by: Arne Jansen <sensille@gmx.net>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: move btrfs_free_qgroup_config() out of spin_lock and fix comments
Wang Shilong [Wed, 14 Aug 2013 01:13:37 +0000 (09:13 +0800)]
Btrfs: move btrfs_free_qgroup_config() out of spin_lock and fix comments

btrfs_free_qgroup_config() is not only called by open/close_ctree(),but
also btrfs_disable_quota().And for btrfs_disable_quota(),we have set
'quota_root' to be null before calling btrfs_free_qgroup_config(),so it
is safe to cleanup in-memory structures without lock held.

Signed-off-by: Wang Shilong <wangsl.fnst@cn.fujitsu.com>
Reviewed-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: fix oops when writing dirty qgroups to disk
Wang Shilong [Wed, 14 Aug 2013 01:13:36 +0000 (09:13 +0800)]
Btrfs: fix oops when writing dirty qgroups to disk

When disabling quota, we should clear out list 'dirty_qgroups',otherwise,
we will get oops if enabling quota again. Fix this by abstracting similar
code from del_qgroup_rb().

Signed-off-by: Wang Shilong <wangsl.fnst@cn.fujitsu.com>
Reviewed-by: Miao Xie <miaox@cn.fujitsu.com>
Reviewed-by: Arne Jansen <sensille@gmx.net>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: fix send issues related to inode number reuse
Josef Bacik [Fri, 16 Aug 2013 20:52:55 +0000 (16:52 -0400)]
Btrfs: fix send issues related to inode number reuse

If you are sending a snapshot and specifying a parent snapshot we will walk the
trees and figure out where they differ and send the differences only.  The way
we check for differences are if the leaves aren't the same and if the keys are
not the same within the leaves.  So if neither leaf is the same (ie the leaf has
been cow'ed from the parent snapshot) we walk each item in the send root and
check it against the parent root.  If the items match exactly then we don't do
anything.  This doesn't quite work for inode refs, since they will just have the
name and the parent objectid.  If you move the file from a directory and then
remove that directory and re-create a directory with the same inode number as
the old directory and then move that file back into that directory we will
assume that nothing changed and you will get errors when you try to receive.

In order to fix this we need to do extra checking to see if the inode ref really
is the same or not.  So do this by passing down BTRFS_COMPARE_TREE_SAME if the
items match.  Then if the key type is an inode ref we can do some extra
checking, otherwise we just keep processing.  The extra checking is to look up
the generation of the directory in the parent volume and compare it to the
generation of the send volume.  If they match then they are the same directory
and we are good to go.  If they don't we have to add them to the changed refs
list.

This means we have to track the generation of the ref we're trying to lookup
when we iterate all the refs for a particular inode.  So in the case of looking
for new refs we have to get the generation from the parent volume, and in the
case of looking for deleted refs we have to get the generation from the send
volume to compare with.

There was also the issue of using a ulist to keep track of the directories we
needed to check.  Because we can get a deleted ref and a new ref for the same
inode number the ulist won't work since it indexes based on the value.  So
instead just dup any directory ref we find and add it to a local list, and then
process that list as normal and do away with using a ulist for this altogether.

Before we would fail all of the tests in the far-progs that related to moving
directories (test group 32).  With this patch we now pass these tests, and all
of the tests in the far-progs send testing suite.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: separate out tests into their own directory
Josef Bacik [Wed, 14 Aug 2013 19:05:12 +0000 (15:05 -0400)]
Btrfs: separate out tests into their own directory

The plan is to have a bunch of unit tests that run when btrfs is loaded when you
build with the appropriate config option.  My ultimate goal is to have a test
for every non-static function we have, but at first I'm going to focus on the
things that cause us the most problems.  To start out with this just adds a
tests/ directory and moves the existing free space cache tests into that
directory and sets up all of the infrastructure.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: avoid starting a transaction in the write path
Josef Bacik [Wed, 14 Aug 2013 18:02:47 +0000 (14:02 -0400)]
Btrfs: avoid starting a transaction in the write path

I noticed while looking at a deadlock that we are always starting a transaction
in cow_file_range().  This isn't really needed since we only need a transaction
if we are doing an inline extent, or if the allocator needs to allocate a chunk.
So push down all the transaction start stuff to be closer to where we actually
need a transaction in all of these cases.  This will hopefully reduce our write
latency when we are committing often.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: fix heavy delalloc related deadlock
Josef Bacik [Wed, 14 Aug 2013 15:33:56 +0000 (11:33 -0400)]
Btrfs: fix heavy delalloc related deadlock

I added a patch where we started taking the ordered operations mutex when we
waited on ordered extents.  We need this because we splice the list and process
it, so if a flusher came in during this scenario it would think the list was
empty and we'd usually get an early ENOSPC.  The problem with this is that this
lock is used in transaction committing.  So we end up with something like this

Transaction commit
-> wait on writers

Delalloc flusher
-> run_ordered_operations (holds mutex)
->wait for filemap-flush to do its thing

flush task
-> cow_file_range
->wait on btrfs_join_transaction because we're commiting

some other task
-> commit_transaction because we notice trans->transaction->flush is set
-> run_ordered_operations (hang on mutex)

We need to disentangle the ordered operations flushing from the delalloc
flushing, since they are separate things.  This solves the deadlock issue I was
seeing.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: fix the error handling wrt orphan items
Josef Bacik [Tue, 13 Aug 2013 18:10:08 +0000 (14:10 -0400)]
Btrfs: fix the error handling wrt orphan items

There are several places where we BUG_ON() if we fail to remove the orphan items
and such, which is not ok, so remove those and either abort or just carry on.
This also fixes a problem where if we couldn't start a transaction we wouldn't
actually remove the orphan item reserve for the inode.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: don't allow a subvol to be deleted if it is the default subovl
Josef Bacik [Mon, 12 Aug 2013 19:36:44 +0000 (15:36 -0400)]
Btrfs: don't allow a subvol to be deleted if it is the default subovl

Eric pointed out that btrfs will happily allow you to delete the default subvol.
This is a problem obviously since the next time you go to mount the file system
it will freak out because it can't find the root.  Fix this by adding a check to
see if our default subvol points to the subvol we are trying to delete, and if
it does not allowing it to happen.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: skip subvol entries when checking if we've created a dir already
Josef Bacik [Mon, 12 Aug 2013 14:56:14 +0000 (10:56 -0400)]
Btrfs: skip subvol entries when checking if we've created a dir already

We have logic to see if we've already created a parent directory by check to see
if an inode inside of that directory has a lower inode number than the one we
are currently processing.  The logic is that if there is a lower inode number
then we would have had to made sure the directory was created at that previous
point.  The problem is that subvols inode numbers count from the lowest objectid
in the root tree, which may be less than our current progress.  So just skip if
our dir item key is a root item.  This fixes the original test and the xfstest
version I made that added an extra subvol create.  Thanks,

Reported-by: Emil Karlson <jekarlson@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agobtrfs: offline dedupe
Mark Fasheh [Tue, 6 Aug 2013 18:42:51 +0000 (11:42 -0700)]
btrfs: offline dedupe

This patch adds an ioctl, BTRFS_IOC_FILE_EXTENT_SAME which will try to
de-duplicate a list of extents across a range of files.

Internally, the ioctl re-uses code from the clone ioctl. This avoids
rewriting a large chunk of extent handling code.

Userspace passes in an array of file, offset pairs along with a length
argument. The ioctl will then (for each dedupe) do a byte-by-byte comparison
of the user data before deduping the extent. Status and number of bytes
deduped are returned for each operation.

Signed-off-by: Mark Fasheh <mfasheh@suse.de>
Reviewed-by: Zach Brown <zab@redhat.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agobtrfs: Introduce extent_read_full_page_nolock()
Mark Fasheh [Tue, 6 Aug 2013 18:42:50 +0000 (11:42 -0700)]
btrfs: Introduce extent_read_full_page_nolock()

We want this for btrfs_extent_same. Basically readpage and friends do their
own extent locking but for the purposes of dedupe, we want to have both
files locked down across a set of readpage operations (so that we can
compare data). Introduce this variant and a flag which can be set for
extent_read_full_page() to indicate that we are already locked.

Partial credit for this patch goes to Gabriel de Perthuis <g2p.code@gmail.com>
as I have included a fix from him to the original patch which avoids a
deadlock on compressed extents.

Signed-off-by: Mark Fasheh <mfasheh@suse.de>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agobtrfs_ioctl_clone: Move clone code into it's own function
Mark Fasheh [Tue, 6 Aug 2013 18:42:49 +0000 (11:42 -0700)]
btrfs_ioctl_clone: Move clone code into it's own function

There's some 250+ lines here that are easily encapsulated into their own
function. I don't change how anything works here, just create and document
the new btrfs_clone() function from btrfs_ioctl_clone() code.

Signed-off-by: Mark Fasheh <mfasheh@suse.de>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agobtrfs: abtract out range locking in clone ioctl()
Mark Fasheh [Tue, 6 Aug 2013 18:42:48 +0000 (11:42 -0700)]
btrfs: abtract out range locking in clone ioctl()

The range locking in btrfs_ioctl_clone is trivially broken out into it's own
function. This reduces the complexity of btrfs_ioctl_clone() by a small bit
and makes that locking code available to future functions in
fs/btrfs/ioctl.c

Signed-off-by: Mark Fasheh <mfasheh@suse.de>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: fix possible memory leak in find_parent_nodes()
Wang Shilong [Thu, 8 Aug 2013 05:04:17 +0000 (13:04 +0800)]
Btrfs: fix possible memory leak in find_parent_nodes()

Signed-off-by: Wang Shilong <wangsl.fnst@cn.fujitsu.com>
Reviewed-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: return ENOSPC when target space is full
Filipe David Borba Manana [Mon, 5 Aug 2013 15:25:12 +0000 (16:25 +0100)]
Btrfs: return ENOSPC when target space is full

In extent-tree.c:do_chunk_alloc(), early on we returned 0 (success)
when the target space was full and when chunk allocation is needed.
However, later on in that same function we return ENOSPC if
btrfs_alloc_chunk() fails (and chunk allocation was needed) and
set the space's full flag.

This was inconsistent, as -ENOSPC should be returned if the space
is full and a chunk allocation needs to performed. If the space is
full but no chunk allocation is needed, just return 0 (success).

Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: don't ignore errors from btrfs_run_delayed_items
Filipe David Borba Manana [Mon, 5 Aug 2013 08:25:47 +0000 (09:25 +0100)]
Btrfs: don't ignore errors from btrfs_run_delayed_items

tree-log.c was ignoring the return value from btrfs_run_delayed_items()
in several places.

Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Reviewed-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: fix inode leak on kmalloc failure in tree-log.c
Filipe David Borba Manana [Sun, 4 Aug 2013 18:58:57 +0000 (19:58 +0100)]
Btrfs: fix inode leak on kmalloc failure in tree-log.c

In tree-log.c:replay_one_name(), if memory allocation for
the name fails, ensure we iput the dir inode we got before
before we return.

Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: allow compressed extents to be merged during defragment
Liu Bo [Fri, 2 Aug 2013 08:30:40 +0000 (16:30 +0800)]
Btrfs: allow compressed extents to be merged during defragment

The rule originally comes from nocow writing, but snapshot-aware
defrag is a different case, the extent has been writen and we're
not going to change the extent but add a reference on the data.

So we're able to allow such compressed extents to be merged into
one bigger extent if they're pointing to the same data.

Reviewed-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agobtrfs: add mount option to set commit interval
David Sterba [Thu, 1 Aug 2013 16:14:52 +0000 (18:14 +0200)]
btrfs: add mount option to set commit interval

I'ts hardcoded to 30 seconds which is fine for most users. Higher values
defer data being synced to permanent storage with obvious consequences
when the system crashes. The upper bound is not forced, but a warning is
printed if it's more than 300 seconds (5 minutes).

Signed-off-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: stop using GFP_ATOMIC when allocating rewind ebs
Josef Bacik [Wed, 7 Aug 2013 20:57:23 +0000 (16:57 -0400)]
Btrfs: stop using GFP_ATOMIC when allocating rewind ebs

There is no reason we can't just set the path to blocking and then do normal
GFP_NOFS allocations for these extent buffers.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: deal with enomem in the rewind path
Josef Bacik [Wed, 7 Aug 2013 18:54:37 +0000 (14:54 -0400)]
Btrfs: deal with enomem in the rewind path

We can get ENOMEM trying to allocate dummy bufs for the rewind operation of the
tree mod log.  Instead of BUG_ON()'ing in this case pass up ENOMEM.  I looked
back through the callers and I'm pretty sure I got everybody who did BUG_ON(ret)
in this path.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: check our parent dir when doing a compare send
Josef Bacik [Tue, 6 Aug 2013 20:47:48 +0000 (16:47 -0400)]
Btrfs: check our parent dir when doing a compare send

When doing a send with a parent subvol we will check to see if the file we are
acting on is being overwritten and move it if we think it may be needed further
down the line during the send.  We check this by checking its directory and
making sure it existed in the parent and making sure the file existed in the
parent.  The problem with this check is that if we create a directory and a file
in that directory, and then snapshot, and then remove and re-create that same
directory and file with different inode numbers and then try to snapshot and
send with the original parent we will try and save the original file inside of
that directory.  This is a problem because during the receive we move the
directory out of the way because it is a completely new inode, which makes us
unable to find the old file inside of the directory when we try to move that out
of the way for the overwrite.  We fix this by checking the parent directory of
the inode we think we are overwriting.  If the parent directory generation in
the send root != the parent directory generation in the parent root then we know
it is a completely new directory and we need not bother with moving the file out
of the way because it would have been completely destroyed.  This fixes bz
60673.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: handle errors when doing slow caching
Josef Bacik [Mon, 5 Aug 2013 15:15:21 +0000 (11:15 -0400)]
Btrfs: handle errors when doing slow caching

Alex Lyakas reported a bug where wait_block_group_cache_progress() would wait
forever if a drive failed.  This is because we just bail out if there is an
error while trying to cache a block group, we don't update anybody who may be
waiting.  So this introduces a new enum for the cache state in case of error and
makes everybody bail out if we have an error.  Alex tested and verified this
patch fixed his problem.  This fixes bz 59431.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: add missing error handling to read_tree_block
Filipe David Borba Manana [Tue, 30 Jul 2013 23:39:56 +0000 (00:39 +0100)]
Btrfs: add missing error handling to read_tree_block

Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoFix leak in __btrfs_map_block error path
Dave Jones [Tue, 30 Jul 2013 17:42:17 +0000 (13:42 -0400)]
Fix leak in __btrfs_map_block error path

If we bail out when the stripe alloc fails, we need to undo the
earlier allocation of raid_map.

Signed-off-by: Dave Jones <davej@redhat.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: add missing error check to find_parent_nodes
Filipe David Borba Manana [Tue, 30 Jul 2013 23:26:35 +0000 (00:26 +0100)]
Btrfs: add missing error check to find_parent_nodes

Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Reviewed-by: Jan Schmidt <list.btrfs@jan-o-sch.net>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: optimize function btrfs_read_chunk_tree
Filipe David Borba Manana [Tue, 30 Jul 2013 11:03:04 +0000 (12:03 +0100)]
Btrfs: optimize function btrfs_read_chunk_tree

After reading all device items from the chunk tree, don't
exit the loop and then navigate down the tree again to find
the chunk items. Instead just read all device items and
chunk items with a single tree search. This is possible
because all device items are found before any chunk item in
the chunks tree.

Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Reviewed-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: don't bug_on when we fail when cleaning up transactions
Josef Bacik [Wed, 31 Jul 2013 14:28:05 +0000 (10:28 -0400)]
Btrfs: don't bug_on when we fail when cleaning up transactions

There is no reason for this sort of jackassery.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: change how we queue blocks for backref checking
Josef Bacik [Tue, 30 Jul 2013 20:30:30 +0000 (16:30 -0400)]
Btrfs: change how we queue blocks for backref checking

Previously we only added blocks to the list to have their backrefs checked if
the level of the block is right above the one we are searching for.  This is
because we want to make sure we don't add the entire path up to the root to the
lists to make sure we process things one at a time.  This assumes that if any
blocks in the path to the root are going to be not checked (shared in other
words) then they will be in the level right above the current block on up.  This
isn't quite right though since we can have blocks higher up the list that are
shared because they are attached to a reloc root.  But we won't add this block
to be checked and then later on we will BUG_ON(!upper->checked).  So instead
keep track of wether or not we've queued a block to be checked in this current
search, and if we haven't go ahead and queue it to be checked.  This patch fixed
the panic I was seeing where we BUG_ON(!upper->checked).  Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: check to see if we have an inline item properly
Josef Bacik [Tue, 30 Jul 2013 19:44:09 +0000 (15:44 -0400)]
Btrfs: check to see if we have an inline item properly

If our item isn't big enough to have an actual inline item when we have skinny
metadata enabled just return 1 in find_inline_backref so we can move on to the
next item.  This probably wasn't causing a problem since we check the values of
ptr and end properly, but just in case this will keep us from doing extra work.
Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: fix what bits we clear when erroring out from delalloc
Josef Bacik [Mon, 29 Jul 2013 17:22:24 +0000 (13:22 -0400)]
Btrfs: fix what bits we clear when erroring out from delalloc

First of all we no longer set EXTENT_DIRTY when we dirty an extent so this patch
removes the clearing of EXTENT_DIRTY since it confuses me.  This patch also adds
clearing EXTENT_DEFRAG and also doing EXTENT_DO_ACCOUNTING when we have errors.
This is because if we are clearing delalloc without adding an ordered extent
then we need to make sure the enospc handling stuff is accounted for.  Also if
this range was DEFRAG we need to make sure that bit is cleared so we dont leak
it.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: cleanup arguments to extent_clear_unlock_delalloc
Josef Bacik [Mon, 29 Jul 2013 15:20:47 +0000 (11:20 -0400)]
Btrfs: cleanup arguments to extent_clear_unlock_delalloc

This patch removes the io_tree argument for extent_clear_unlock_delalloc since
we always use &BTRFS_I(inode)->io_tree, and it separates out the extent tree
operations from the page operations.  This way we just pass in the extent bits
we want to clear and then pass in the operations we want done to the pages.
This is because I'm going to fix what extent bits we clear in some cases and
rather than add a bunch of new flags we'll just use the actual extent bits we
want to clear.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agobtrfs: use BTRFS_SUPER_INFO_SIZE macro at btrfs_read_dev_super()
Anand Jain [Thu, 25 Jul 2013 17:29:35 +0000 (01:29 +0800)]
btrfs: use BTRFS_SUPER_INFO_SIZE macro at btrfs_read_dev_super()

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: cache the extent map struct when reading several pages
Miao Xie [Thu, 25 Jul 2013 11:22:37 +0000 (19:22 +0800)]
Btrfs: cache the extent map struct when reading several pages

When we read several pages at once, we needn't get the extent map object
every time we deal with a page, and we can cache the extent map object.
So, we can reduce the search time of the extent map, and besides that, we
also can reduce the lock contention of the extent map tree.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: batch the extent state operation when reading pages
Miao Xie [Thu, 25 Jul 2013 11:22:36 +0000 (19:22 +0800)]
Btrfs: batch the extent state operation when reading pages

In the past, we cached the checksum value in the extent state object, so we
had to split the extent state object by the block size, or we had no space
to keep this checksum value. But it increased the lock contention of the
extent state tree.

Now we removed this limit by caching the checksum into the bio object, so
it is unnecessary to do the extent state operations by the block size, we
can do it in batches, in this way, we can reduce the lock contention of
the extent state tree.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: batch the extent state operation in the end io handle of the read page
Miao Xie [Thu, 25 Jul 2013 11:22:35 +0000 (19:22 +0800)]
Btrfs: batch the extent state operation in the end io handle of the read page

Before applying this patch, we set the uptodate flag and unlock the extent
by the page size, it is unnecessary, we can do it in batches, it can reduce
the lock contention of the extent state tree.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: don't cache the csum value into the extent state tree
Miao Xie [Thu, 25 Jul 2013 11:22:34 +0000 (19:22 +0800)]
Btrfs: don't cache the csum value into the extent state tree

Before applying this patch, we cached the csum value into the extent state
tree when reading some data from the disk, this operation increased the lock
contention of the state tree.

Now, we just store the csum value into the bio structure or other unshared
structure, so we can reduce the lock contention.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: add branch prediction hints in the read page end IO function
Miao Xie [Thu, 25 Jul 2013 11:22:33 +0000 (19:22 +0800)]
Btrfs: add branch prediction hints in the read page end IO function

This patch add some branch prediction hints into the end IO function
of the read page, it reduced the percentage of the branch misses from
5.5% to 4.9%.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: remove unnecessary argument of bio_readpage_error()
Miao Xie [Thu, 25 Jul 2013 11:22:32 +0000 (19:22 +0800)]
Btrfs: remove unnecessary argument of bio_readpage_error()

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: add missing mounting options in btrfs_show_options()
Wang Shilong [Wed, 24 Jul 2013 02:30:02 +0000 (10:30 +0800)]
Btrfs: add missing mounting options in btrfs_show_options()

Some options are missing in btrfs_show_options(), this patch
adds them.

Signed-off-by: Wang Shilong <wangsl.fnst@cn.fujitsu.com>
Reviewed-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: use u64 for subvolid when parsing mount options
Wang Shilong [Wed, 24 Jul 2013 02:29:20 +0000 (10:29 +0800)]
Btrfs: use u64 for subvolid when parsing mount options

Although for most time, int is enough for subvolid, we should
ensure safety in theory.

Signed-off-by: Wang Shilong <wangsl.fnst@cn.fujitsu.com>
Reviewed-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: add sanity checks regarding to parsing mount options
Wang Shilong [Wed, 24 Jul 2013 02:29:05 +0000 (10:29 +0800)]
Btrfs: add sanity checks regarding to parsing mount options

I just notice the following commands succeed:
mount <dev> <mnt> -o thread_pool=-1

This is ridiculous, only positive thread_pool makes sense,this
patch adds sanity checks for them, and also catches the error of
ENOMEM if allocating memory fails.

Signed-off-by: Wang Shilong <wangsl.fnst@cn.fujitsu.com>
Reviewed-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs, raid56: fix memory leak when allocating pages for p/q stripes failed
Miao Xie [Mon, 22 Jul 2013 08:36:57 +0000 (16:36 +0800)]
Btrfs, raid56: fix memory leak when allocating pages for p/q stripes failed

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agobtrfs/raid56: fix and cleanup some error paths
Dan Carpenter [Mon, 22 Jul 2013 06:55:15 +0000 (09:55 +0300)]
btrfs/raid56: fix and cleanup some error paths

The alloc_rbio() frees "raid_map" and "bbio" on error, so there is a
potential double free bug in raid56_parity_write().  The
raid56_parity_write() and raid56_parity_recover() functions should still
free "raid_map" and "bbio" on error if other errors occur though, so I
have added some more calls to kfree().

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: don't bother autodefragging if our root is going away
Josef Bacik [Fri, 26 Jul 2013 13:13:15 +0000 (09:13 -0400)]
Btrfs: don't bother autodefragging if our root is going away

We can end up with inodes on the auto defrag list that exist on roots that are
going to be deleted.  This is extra work we don't need to do, so just bail if
our root has 0 root refs.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: cleanup reloc roots properly on error
Josef Bacik [Tue, 23 Jul 2013 20:57:15 +0000 (16:57 -0400)]
Btrfs: cleanup reloc roots properly on error

I was hitting the BUG_ON() at the end of merge_reloc_roots() because we were
aborting the transaction at some point previously and then getting an error when
we tried to drop the reloc root.  I fixed btrfs_drop_snapshot to re-add us to
the dead roots list if we failed, but this isn't the right thing to do for reloc
roots since it uses root->root_list for it's own stuff in order to know what
needs to be cleaned up.  So fix btrfs_drop_snapshot to only do the re-add if we
aren't dropping for reloc, and handle errors from merge_reloc_root() by dropping
the reloc root we are processing since it won't be on the list of roots to
cleanup.  With this patch my reproducer no longer panics.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: reset ret in record_one_backref
Josef Bacik [Mon, 22 Jul 2013 16:50:37 +0000 (12:50 -0400)]
Btrfs: reset ret in record_one_backref

I was getting warnings when running find ./ -type f -exec btrfs fi defrag -f {}
\; from record_one_backref because ret was set.  Turns out it was because it was
set to 1 because the search slot didn't come out exact and we never reset it.
So reset it to 0 right after the search so we don't leak this and get
uneccessary warnings.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agobtrfs: fix get set label blocking against balance
Anand Jain [Fri, 19 Jul 2013 09:39:32 +0000 (17:39 +0800)]
btrfs: fix get set label blocking against balance

btrfs_ioctl_get_fslabel() and btrfs_ioctl_set_fslabel()
used root->fs_info->volume_mutex mutex which caused operations
like balance to block set/get label operation until its
completion and generally balance operation takes a long
time to complete, so it will be annoying to the user when
cli appears hung

also this patch will add a bit of optimization within
the btrfs_ioctl_get_falabel() function.

v1->v2:
   use fs_info->super_lock instead of uuid_mutex

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: Print key type in decimal everywhere
Stefan Behrens [Tue, 16 Jul 2013 13:38:33 +0000 (15:38 +0200)]
Btrfs: Print key type in decimal everywhere

This is confusing, sometimes the key type is printed in hex (without
a leading "0x" which makes things even more complicated), sometimes
in decimal...
Change it to be in decimal everywhere.

Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs/tracepoint: update delayed ref tracepoints
Liu Bo [Tue, 16 Jul 2013 11:03:36 +0000 (19:03 +0800)]
Btrfs/tracepoint: update delayed ref tracepoints

This shows exactly how btrfs processes the delayed refs onto disks,
which is very helpful on understanding delayed ref mechanism and
debugging related bugs.

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agobtrfs_read_block_groups: Use enums to index
chandan [Tue, 16 Jul 2013 06:58:56 +0000 (12:28 +0530)]
btrfs_read_block_groups: Use enums to index

 btrfs_space_info->block_groups.

The current code uses integer literals to index
btrfs_space_info->block_groups[] array. Instead use corresponding
enums from 'enum btrfs_raid_types'.

Signed-off-by: chandan <chandan@linux.vnet.ibm.com>
Reviewed-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agobtrfs: Cleanup for using BTRFS_SETGET_STACK instead of raw convert
Qu Wenruo [Tue, 16 Jul 2013 03:19:18 +0000 (11:19 +0800)]
btrfs: Cleanup for using BTRFS_SETGET_STACK instead of raw convert

Some codes still use the cpu_to_lexx instead of the
BTRFS_SETGET_STACK_FUNCS declared in ctree.h.

Also added some BTRFS_SETGET_STACK_FUNCS for btrfs_header btrfs_timespec
and other structures.

Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Reviewed-by: Miao Xie <miaoxie@cn.fujitsu.com>
Reviewed-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: set qgroup_ulist to be null after calling ulist_free()
Wang Shilong [Sat, 13 Jul 2013 13:02:54 +0000 (21:02 +0800)]
Btrfs: set qgroup_ulist to be null after calling ulist_free()

We call ulist_free(qgroup_ulist) in btrfs_free_qgroup_config(),
and btrfs_free_qgroup_config() may be called in two cases:

(1)umount filesystem
(2)disabling quota

However, if we firstly disable quota and then umount filesystem,
a double free happens. Fix it.

Signed-off-by: Wang Shilong <wangsl.fnst@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: add missing error checks to add_data_references
Filipe David Borba Manana [Sat, 13 Jul 2013 11:25:15 +0000 (12:25 +0100)]
Btrfs: add missing error checks to add_data_references

The function relocation.c:add_data_references() was not checking
if all calls to __add_tree_block() and find_data_references() were
succeeding or not.

Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Reviewed-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agobtrfs: make errors in btrfs_num_copies less noisy
David Sterba [Thu, 11 Jul 2013 13:41:11 +0000 (15:41 +0200)]
btrfs: make errors in btrfs_num_copies less noisy

The log message level 'critical' is verbose enough, 'emergency' beeps on
all terminals.

Signed-off-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: make free space caching faster with many non-inline extent references
Liu Bo [Thu, 11 Jul 2013 09:51:15 +0000 (17:51 +0800)]
Btrfs: make free space caching faster with many non-inline extent references

So to cache free space, we iterate every extent item to gather free space info.

When we have say 10,000 non-inline extent refs(such as BTRFS_EXTENT_DATA_REF),
it takes quite a long time, and since inline extent refs and non-inline ones have
same objectid in their keys, we can just re-search the tree with the next address
to skip non-inline references.

(This is found by dedup feature because dedup extents can end up with many
non-inline extent refs.)

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agobtrfs: fall back to global reservation when removing subvolumes
Jeff Mahoney [Tue, 9 Jul 2013 20:37:21 +0000 (16:37 -0400)]
btrfs: fall back to global reservation when removing subvolumes

I recently did some ENOSPC testing that involved filling the disk
while create and removing snapshots in a loop. During the test cycle,
I ran into an ENOSPC when trying to remove a snapshot, leaving the fs
stuck in ENOSPC even after a umount/mount cycle.

This patch allow subvolume removal to fall back onto the global
block reservation in order to succeed when it would have failed
otherwise.

Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: optimize btrfs_lookup_extent_info()
Filipe David Borba Manana [Fri, 5 Jul 2013 22:12:06 +0000 (23:12 +0100)]
Btrfs: optimize btrfs_lookup_extent_info()

If we're looking for a metadata item in the tree and the
search fails with return value of 1, and the slot doesn't
point to the first item in the leaf, check if the previous
item in the leaf corresponds to an extent item for the same
object id - if it does, then don't do another tree search
to get it.

This optimization is already done by btrfs-progs.

V2: updated commit message.

Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: Release uuid_mutex for shrink during device delete
Carey Underwood [Mon, 4 Mar 2013 22:37:06 +0000 (16:37 -0600)]
Btrfs: Release uuid_mutex for shrink during device delete

Device scanning waits on the uuid_mutex, which can result in a very long
wait if dev delete is shrinking the device.

Signed-off-by: Carey Underwood <cwillu@cwillu.com>
Reviewed-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: set lockdep class before locking new extent buffer
Josef Bacik [Fri, 5 Jul 2013 21:05:38 +0000 (17:05 -0400)]
Btrfs: set lockdep class before locking new extent buffer

We've been seeing spurious complaints out of lockdep because the lock class name
changes.  This is happening because when we drop a snapshot we will lock a block
before we've read it in, which sets the lockdep class to whatever the default
is.  Then once we read the thing in we reset the lockdep class to what it is
supposed to be, which blows lockdeps' mind.  This patch should fix the problem,
it appears to be the only place where we do this sort of thing.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: return -1 when lzo compression makes data bigger
Stefan Agner [Mon, 1 Jul 2013 18:33:39 +0000 (20:33 +0200)]
Btrfs: return -1 when lzo compression makes data bigger

With this fix the lzo code behaves like the zlib code by returning an
error
code when compression does not help reduce the size of the file.
This is currently not a bug since the compressed size is checked again
in
the calling method compress_file_range.

Signed-off-by: Stefan Agner <stefan@agner.ch>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: stop using GFP_ATOMIC for the tree mod log allocations
Josef Bacik [Mon, 1 Jul 2013 20:18:19 +0000 (16:18 -0400)]
Btrfs: stop using GFP_ATOMIC for the tree mod log allocations

Previously we held the tree mod lock when adding stuff because we use it to
check and see if we truly do want to track tree modifications.  This is
admirable, but GFP_ATOMIC in a critical area that is going to get hit pretty
hard and often is not nice.  So instead do our basic checks to see if we don't
need to track modifications, and if those pass then do our allocation, and then
when we go to insert the new modification check if we still care, and if we
don't just free up our mod and return.  Otherwise we're good to go and we can
carry on.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoLinux 3.11-rc7
Linus Torvalds [Mon, 26 Aug 2013 00:43:22 +0000 (17:43 -0700)]
Linux 3.11-rc7

11 years agoMerge tag 'staging-3.11-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh...
Linus Torvalds [Sun, 25 Aug 2013 19:44:15 +0000 (12:44 -0700)]
Merge tag 'staging-3.11-rc7' of git://git./linux/kernel/git/gregkh/staging

Pull staging fixes from Greg KH:
 "Here are two tiny staging tree fixes (well, one is for an iio driver,
  but those updates come through the staging tree due to dependancies)

  One fixes a problem with an IIO driver, and the other fixes a bug in
  the comedi driver core"

* tag 'staging-3.11-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/staging:
  staging: comedi: bug-fix NULL pointer dereference on failed attach
  iio: adjd_s311: Fix non-scan mode data read

11 years agoMerge tag 'usb-3.11-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb
Linus Torvalds [Sun, 25 Aug 2013 19:43:44 +0000 (12:43 -0700)]
Merge tag 'usb-3.11-rc7' of git://git./linux/kernel/git/gregkh/usb

Pull USB fixes from Greg KH:
 "Here are two USB fixes for 3.11-rc7

  One fixes a reported regression in the OHCI driver, and the other
  fixes a reported build breakage in the USB phy drivers"

* tag 'usb-3.11-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb:
  usb: phy: fix build breakage
  USB: OHCI: add missing PCI PM callbacks to ohci-pci.c

11 years agoMerge branch 'fixes' of git://git.linaro.org/people/rmk/linux-arm
Linus Torvalds [Sun, 25 Aug 2013 19:41:37 +0000 (12:41 -0700)]
Merge branch 'fixes' of git://git.linaro.org/people/rmk/linux-arm

Pull ARM fixes from Russell King:
 "This round of fixes is smaller than previous: a couple more updates
  for the security fixes, and a one-liner kexec fix"

* 'fixes' of git://git.linaro.org/people/rmk/linux-arm:
  ARM: 7816/1: CONFIG_KUSER_HELPERS: fix help text
  ARM: 7815/1: kexec: offline non panic CPUs on Kdump panic
  ARM: 7819/1: fiq: Cast the first argument of flush_icache_range()