platform/kernel/linux-rpi3.git
6 years agobtrfs: restore uuid_mutex in btrfs_open_devices
David Sterba [Wed, 20 Jun 2018 17:51:28 +0000 (19:51 +0200)]
btrfs: restore uuid_mutex in btrfs_open_devices

Commit 542c5908abfe84f7b4c1 ("btrfs: replace uuid_mutex by
device_list_mutex in btrfs_open_devices") switched to device_list_mutex
as we need that for the device list traversal, but we also need
uuid_mutex to protect access to fs_devices::opened to be consistent with
other users of that.

Fixes: 542c5908abfe84f7b4c1 ("btrfs: replace uuid_mutex by device_list_mutex in btrfs_open_devices")
Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: fix mount failure when qgroup rescan is in progress
Filipe Manana [Tue, 26 Jun 2018 23:43:15 +0000 (00:43 +0100)]
Btrfs: fix mount failure when qgroup rescan is in progress

If a power failure happens while the qgroup rescan kthread is running,
the next mount operation will always fail. This is because of a recent
regression that makes qgroup_rescan_init() incorrectly return -EINVAL
when we are mounting the filesystem (through btrfs_read_qgroup_config()).
This causes the -EINVAL error to be returned regardless of any qgroup
flags being set instead of returning the error only when neither of
the flags BTRFS_QGROUP_STATUS_FLAG_RESCAN nor BTRFS_QGROUP_STATUS_FLAG_ON
are set.

A test case for fstests follows up soon.

Fixes: 9593bf49675e ("btrfs: qgroup: show more meaningful qgroup_rescan_init error message")
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: fix regression in btrfs_page_mkwrite() from vm_fault_t conversion
Chris Mason [Mon, 25 Jun 2018 17:03:41 +0000 (10:03 -0700)]
Btrfs: fix regression in btrfs_page_mkwrite() from vm_fault_t conversion

The vm_fault_t conversion commit introduced a ret2 variable for tracking
the integer return values from internal btrfs functions.  It was
sometimes returning VM_FAULT_LOCKED for pages that were actually invalid
and had been removed from the radix.  Something like this:

    ret2 = btrfs_delalloc_reserve_space() // returns zero on success

    lock_page(page)
    if (page->mapping != inode->i_mapping)
goto out_unlock;

...

out_unlock:
    if (!ret2) {
    ...
    return VM_FAULT_LOCKED;
    }

This ends up triggering this WARNING in btrfs_destroy_inode()
    WARN_ON(BTRFS_I(inode)->block_rsv.size);

xfstests generic/095 was able to reliably reproduce the errors.

Since out_unlock: is only used for errors, this fix moves it below the
if (!ret2) check we use to return VM_FAULT_LOCKED for success.

Fixes: a528a2415087 (btrfs: change return type of btrfs_page_mkwrite to vm_fault_t)
Signed-off-by: Chris Mason <clm@fb.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: quota: Set rescan progress to (u64)-1 if we hit last leaf
Qu Wenruo [Wed, 27 Jun 2018 10:19:55 +0000 (18:19 +0800)]
btrfs: quota: Set rescan progress to (u64)-1 if we hit last leaf

Commit ff3d27a048d9 ("btrfs: qgroup: Finish rescan when hit the last leaf
of extent tree") added a new exit for rescan finish.

However after finishing quota rescan, we set
fs_info->qgroup_rescan_progress to (u64)-1 before we exit through the
original exit path.
While we missed that assignment of (u64)-1 in the new exit path.

The end result is, the quota status item doesn't have the same value.
(-1 vs the last bytenr + 1)
Although it doesn't affect quota accounting, it's still better to keep
the original behavior.

Reported-by: Misono Tomohiro <misono.tomohiro@jp.fujitsu.com>
Fixes: ff3d27a048d9 ("btrfs: qgroup: Finish rescan when hit the last leaf of extent tree")
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Misono Tomohiro <misono.tomohiro@jp.fujitsu.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: fix return value on rename exchange failure
Filipe Manana [Mon, 11 Jun 2018 18:24:16 +0000 (19:24 +0100)]
Btrfs: fix return value on rename exchange failure

If we failed during a rename exchange operation after starting/joining a
transaction, we would end up replacing the return value, stored in the
local 'ret' variable, with the return value from btrfs_end_transaction().
So this could end up returning 0 (success) to user space despite the
operation having failed and aborted the transaction, because if there are
multiple tasks having a reference on the transaction at the time
btrfs_end_transaction() is called by the rename exchange, that function
returns 0 (otherwise it returns -EIO and not the original error value).
So fix this by not overwriting the return value on error after getting
a transaction handle.

Fixes: cdd1fedf8261 ("btrfs: add support for RENAME_EXCHANGE and RENAME_WHITEOUT")
CC: stable@vger.kernel.org # 4.9+
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: fix invalid-free in btrfs_extent_same
Lu Fengqi [Tue, 19 Jun 2018 06:54:38 +0000 (14:54 +0800)]
btrfs: fix invalid-free in btrfs_extent_same

If this condition ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
   (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM))
is hit, we will go to free the uninitialized cmp.src_pages and
cmp.dst_pages.

Fixes: 67b07bd4bec5 ("Btrfs: reuse cmp workspace in EXTENT_SAME ioctl")
Signed-off-by: Lu Fengqi <lufq.fnst@cn.fujitsu.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: fix physical offset reported by fiemap for inline extents
Filipe Manana [Wed, 20 Jun 2018 09:02:30 +0000 (10:02 +0100)]
Btrfs: fix physical offset reported by fiemap for inline extents

Commit 9d311e11fc1f ("Btrfs: fiemap: pass correct bytenr when
fm_extent_count is zero") introduced a regression where we no longer
report 0 as the physical offset for inline extents (and other extents
with a special block_start value). This is because it always sets the
variable used to report the physical offset ("disko") as em->block_start
plus some offset, and em->block_start has the value 18446744073709551614
((u64) -2) for inline extents.

This made the btrfs test 004 (from fstests) often fail, for example, for
a file with an inline extent we have the following items in the subvolume
tree:

    item 101 key (418 INODE_ITEM 0) itemoff 11029 itemsize 160
           generation 25 transid 38 size 1525 nbytes 1525
           block group 0 mode 100666 links 1 uid 0 gid 0 rdev 0
           sequence 0 flags 0x2(none)
           atime 1529342058.461891730 (2018-06-18 18:14:18)
           ctime 1529342058.461891730 (2018-06-18 18:14:18)
           mtime 1529342058.461891730 (2018-06-18 18:14:18)
           otime 1529342055.869892885 (2018-06-18 18:14:15)
    item 102 key (418 INODE_REF 264) itemoff 11016 itemsize 13
           index 25 namelen 3 name: fc7
    item 103 key (418 EXTENT_DATA 0) itemoff 9470 itemsize 1546
           generation 38 type 0 (inline)
           inline extent data size 1525 ram_bytes 1525 compression 0 (none)

Then when test 004 invoked fiemap against the file it got a non-zero
physical offset:

 $ filefrag -v /mnt/p0/d4/d7/fc7
 Filesystem type is: 9123683e
 File size of /mnt/p0/d4/d7/fc7 is 1525 (1 block of 4096 bytes)
  ext:     logical_offset:        physical_offset: length:   expected: flags:
    0:        0..    4095: 18446744073709551614..      4093:   4096:             last,not_aligned,inline,eof
 /mnt/p0/d4/d7/fc7: 1 extent found

This resulted in the test failing like this:

btrfs/004 49s ... [failed, exit status 1]- output mismatch (see /home/fdmanana/git/hub/xfstests/results//btrfs/004.out.bad)
    --- tests/btrfs/004.out 2016-08-23 10:17:35.027012095 +0100
    +++ /home/fdmanana/git/hub/xfstests/results//btrfs/004.out.bad 2018-06-18 18:15:02.385872155 +0100
    @@ -1,3 +1,10 @@
     QA output created by 004
     *** test backref walking
    -*** done
    +./tests/btrfs/004: line 227: [: 7.55578637259143e+22: integer expression expected
    +ERROR: 7.55578637259143e+22 is not a valid numeric value.
    +unexpected output from
    + /home/fdmanana/git/hub/btrfs-progs/btrfs inspect-internal logical-resolve -s 65536 -P 7.55578637259143e+22 /home/fdmanana/btrfs-tests/scratch_1
    ...
    (Run 'diff -u tests/btrfs/004.out /home/fdmanana/git/hub/xfstests/results//btrfs/004.out.bad'  to see the entire diff)
Ran: btrfs/004

The large number in scientific notation reported as an invalid numeric
value is the result from the filter passed to perl which multiplies the
physical offset by the block size reported by fiemap.

So fix this by ensuring the physical offset is always set to 0 when we
are processing an extent with a special block_start value.

Fixes: 9d311e11fc1f ("Btrfs: fiemap: pass correct bytenr when fm_extent_count is zero")
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: scrub: Don't use inode pages for device replace
Qu Wenruo [Tue, 5 Jun 2018 04:36:56 +0000 (12:36 +0800)]
btrfs: scrub: Don't use inode pages for device replace

[BUG]
Btrfs can create compressed extent without checksum (even though it
shouldn't), and if we then try to replace device containing such extent,
the result device will contain all the uncompressed data instead of the
compressed one.

Test case already submitted to fstests:
https://patchwork.kernel.org/patch/10442353/

[CAUSE]
When handling compressed extent without checksum, device replace will
goe into copy_nocow_pages() function.

In that function, btrfs will get all inodes referring to this data
extents and then use find_or_create_page() to get pages direct from that
inode.

The problem here is, pages directly from inode are always uncompressed.
And for compressed data extent, they mismatch with on-disk data.
Thus this leads to corrupted compressed data extent written to replace
device.

[FIX]
In this attempt, we could just remove the "optimization" branch, and let
unified scrub_pages() to handle it.

Although scrub_pages() won't bother reusing page cache, it will be a
little slower, but it does the correct csum checking and won't cause
such data corruption caused by "optimization".

Note about the fix: this is the minimal fix that can be backported to
older stable trees without conflicts. The whole callchain from
copy_nocow_pages() can be deleted, and will be in followup patches.

Fixes: ff023aac3119 ("Btrfs: add code to scrub to copy read data to another disk")
CC: stable@vger.kernel.org # 4.4+
Reported-by: James Harvey <jamespharvey20@gmail.com>
Reviewed-by: James Harvey <jamespharvey20@gmail.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
[ remove code removal, add note why ]
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: change return type of btrfs_page_mkwrite to vm_fault_t
Souptick Joarder [Wed, 6 Jun 2018 14:24:44 +0000 (19:54 +0530)]
btrfs: change return type of btrfs_page_mkwrite to vm_fault_t

Use the new return type vm_fault_t for fault handler. For now, this is
just documenting that the function returns a VM_FAULT value rather than
an errno. Once all instances are converted, vm_fault_t will become a
distinct type.

Reference commit 1c8f422059ae ("mm: change return type to vm_fault_t")

vmf_error() is the newly introduced inline function in 4.17-rc6.

Signed-off-by: Souptick Joarder <jrdr.linux@gmail.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: fiemap: pass correct bytenr when fm_extent_count is zero
Robbie Ko [Mon, 7 May 2018 08:42:04 +0000 (16:42 +0800)]
Btrfs: fiemap: pass correct bytenr when fm_extent_count is zero

[BUG]
fm_mapped_extents is not correct when fm_extent_count is 0
Like:
   # mount /dev/vdb5 /mnt/btrfs
   # dd if=/dev/zero bs=16K count=4 oflag=dsync of=/mnt/btrfs/file
   # xfs_io -c "fiemap -v" /mnt/btrfs/file
   /mnt/btrfs/file:
   EXT: FILE-OFFSET      BLOCK-RANGE      TOTAL FLAGS
     0: [0..127]:        25088..25215       128   0x1

When user space wants to get the number of file extents,
set fm_extent_count to 0 to run fiemap and then read fm_mapped_extents.

In the above example, fiemap will return with fm_mapped_extents set to 4,
but it should be 1 since there's only one entry in the output.

[REASON]
The problem seems to be that disko is only set if
fieinfo->fi_extents_max is set. And this member is initialized, in the
generic ioctl_fiemap function, to the value of used-passed
fm_extent_count. So when the user passes 0 then fi_extent_max is also
set to zero and this causes btrfs to not initialize disko at all.
Eventually this leads emit_fiemap_extent being called with a bogus
'phys' argument preventing proper fiemap entries merging.

[FIX]
Move the disko initialization earlier in extent_fiemap making it
independent of user-passed arguments, allowing emit_fiemap_extent to
properly handle consecutive extent entries.

Signed-off-by: Robbie Ko <robbieko@synology.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Check error of btrfs_iget in btrfs_search_path_in_tree_user
Misono Tomohiro [Mon, 4 Jun 2018 07:41:07 +0000 (16:41 +0900)]
btrfs: Check error of btrfs_iget in btrfs_search_path_in_tree_user

The patch introducing the ioctl was not the latest version at the time
of merging to the mainline and needs a fixup from this patch.

Fixes: ba637a252d30 ("btrfs: Check error of btrfs_iget() in btrfs_search_path_in_tree_user")
Signed-off-by: Misono Tomohiro <misono.tomohiro@jp.fujitsu.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Add unprivileged version of ino_lookup ioctl
Tomohiro Misono [Mon, 21 May 2018 01:09:44 +0000 (10:09 +0900)]
btrfs: Add unprivileged version of ino_lookup ioctl

Add unprivileged version of ino_lookup ioctl BTRFS_IOC_INO_LOOKUP_USER
to allow normal users to call "btrfs subvolume list/show" etc. in
combination with BTRFS_IOC_GET_SUBVOL_INFO/BTRFS_IOC_GET_SUBVOL_ROOTREF.

This can be used like BTRFS_IOC_INO_LOOKUP but the argument is
different. This is  because it always searches the fs/file tree
correspoinding to the fd with which this ioctl is called and also
returns the name of bottom subvolume.

The main differences from original ino_lookup ioctl are:

  1. Read + Exec permission will be checked using inode_permission()
     during path construction. -EACCES will be returned in case
     of failure.
  2. Path construction will be stopped at the inode number which
     corresponds to the fd with which this ioctl is called. If
     constructed path does not exist under fd's inode, -EACCES
     will be returned.
  3. The name of bottom subvolume is also searched and filled.

Note that the maximum length of path is shorter 256 (BTRFS_VOL_NAME_MAX+1)
bytes than ino_lookup ioctl because of space of subvolume's name.

Reviewed-by: Gu Jinxiang <gujx@cn.fujitsu.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Tested-by: Gu Jinxiang <gujx@cn.fujitsu.com>
Signed-off-by: Tomohiro Misono <misono.tomohiro@jp.fujitsu.com>
[ style fixes ]
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Add unprivileged ioctl which returns subvolume's ROOT_REF
Tomohiro Misono [Mon, 21 May 2018 01:09:43 +0000 (10:09 +0900)]
btrfs: Add unprivileged ioctl which returns subvolume's ROOT_REF

Add unprivileged ioctl BTRFS_IOC_GET_SUBVOL_ROOTREF which returns
ROOT_REF information of the subvolume containing this inode except the
subvolume name (this is because to prevent potential name leak). The
subvolume name will be gained by user version of ino_lookup ioctl
(BTRFS_IOC_INO_LOOKUP_USER) which also performs permission check.

The min id of root ref's subvolume to be searched is specified by
@min_id in struct btrfs_ioctl_get_subvol_rootref_args. After the search
ends, @min_id is set to the last searched root ref's subvolid + 1. Also,
if there are more root refs than BTRFS_MAX_ROOTREF_BUFFER_NUM,
-EOVERFLOW is returned. Therefore the caller can just call this ioctl
again without changing the argument to continue search.

Reviewed-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Gu Jinxiang <gujx@cn.fujitsu.com>
Tested-by: Gu Jinxiang <gujx@cn.fujitsu.com>
Signed-off-by: Tomohiro Misono <misono.tomohiro@jp.fujitsu.com>
[ style fixes and struct item renames ]
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Add unprivileged ioctl which returns subvolume information
Tomohiro Misono [Mon, 21 May 2018 01:09:42 +0000 (10:09 +0900)]
btrfs: Add unprivileged ioctl which returns subvolume information

Add new unprivileged ioctl BTRFS_IOC_GET_SUBVOL_INFO which returns
the information of subvolume containing this inode.
(i.e. returns the information in ROOT_ITEM and ROOT_BACKREF.)

Reviewed-by: Gu Jinxiang <gujx@cn.fujitsu.com>
Tested-by: Gu Jinxiang <gujx@cn.fujitsu.com>
Signed-off-by: Tomohiro Misono <misono.tomohiro@jp.fujitsu.com>
[ minor style fixes, update struct comments ]
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: clean up error handling in btrfs_truncate()
Omar Sandoval [Tue, 22 May 2018 16:59:50 +0000 (09:59 -0700)]
Btrfs: clean up error handling in btrfs_truncate()

btrfs_truncate() uses two variables for error handling, ret and err (if
this sounds familiar, it's because btrfs_truncate_inode_items() did
something similar). This is error prone, as was made evident by "Btrfs:
fix error handling in btrfs_truncate()". We only have err because we
don't want to mask an error if we call btrfs_update_inode() and
btrfs_end_transaction(), so let's make that its own scoped return
variable and use ret everywhere else.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Factor out write portion of btrfs_get_blocks_direct
Nikolay Borisov [Wed, 2 May 2018 12:19:33 +0000 (15:19 +0300)]
btrfs: Factor out write portion of btrfs_get_blocks_direct

Now that the read side is extracted into its own function, do the same
to the write side. This leaves btrfs_get_blocks_direct_write with the
sole purpose of handling common locking required. Also flip the
condition in btrfs_get_blocks_direct_write so that the write case
comes first and we check for if (Create) rather than if (!create). This
is purely subjective but I believe makes reading a bit more "linear".
No functional changes.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Factor out read portion of btrfs_get_blocks_direct
Nikolay Borisov [Wed, 2 May 2018 12:19:32 +0000 (15:19 +0300)]
btrfs: Factor out read portion of btrfs_get_blocks_direct

Currently this function handles both the READ and WRITE dio cases. This
is facilitated by a bunch of 'if' statements, a goto short-circuit
statement and a very perverse aliasing of "!created"(READ) case
by setting lockstart = lockend and checking for lockstart < lockend for
detecting the write. Let's simplify this mess by extracting the
READ-only code into a separate __btrfs_get_block_direct_read function.
This is only the first step, the next one will be to factor out the
write side as well. The end goal will be to have the common locking/
unlocking code in btrfs_get_blocks_direct and then it will call either
the read|write subvariants. No functional changes.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: return ENOMEM if path allocation fails in btrfs_cross_ref_exist
Su Yue [Wed, 30 May 2018 06:49:10 +0000 (14:49 +0800)]
btrfs: return ENOMEM if path allocation fails in btrfs_cross_ref_exist

The error code does not match the reason of failure and may confuse the
callers.

Signed-off-by: Su Yue <suy.fnst@cn.fujitsu.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: raid56: Remove VLA usage
Kees Cook [Tue, 29 May 2018 23:44:59 +0000 (16:44 -0700)]
btrfs: raid56: Remove VLA usage

In the quest to remove all stack VLA usage from the kernel[1], this
allocates the working buffers during regular init, instead of using stack
space. This refactors the allocation code a bit to make it easier
to review.

[1] https://lkml.kernel.org/r/CA+55aFzCG-zNmZwX4A2FQpadafLfEzK6CC=qPXydAacU1RqZWA@mail.gmail.com

Signed-off-by: Kees Cook <keescook@chromium.org>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: return error value if create_io_em failed in cow_file_range
Su Yue [Wed, 30 May 2018 08:48:56 +0000 (16:48 +0800)]
btrfs: return error value if create_io_em failed in cow_file_range

In cow_file_range(), create_io_em() may fail, but its return value is
not recorded.  Then return value may be 0 even it failed which is a
wrong behavior.

Let cow_file_range() return PTR_ERR(em) if create_io_em() failed.

Fixes: 6f9994dbabe5 ("Btrfs: create a helper to create em for IO")
CC: stable@vger.kernel.org # 4.11+
Signed-off-by: Su Yue <suy.fnst@cn.fujitsu.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: drop useless member qgroup_reserved of btrfs_pending_snapshot
Gu JinXiang [Wed, 30 May 2018 03:00:39 +0000 (11:00 +0800)]
btrfs: drop useless member qgroup_reserved of btrfs_pending_snapshot

Since there is no more use of qgroup_reserved member in struct
btrfs_pending_snapshot, remove it.

Signed-off-by: Gu JinXiang <gujx@cn.fujitsu.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: drop unused parameter qgroup_reserved
Gu JinXiang [Wed, 30 May 2018 03:00:38 +0000 (11:00 +0800)]
btrfs: drop unused parameter qgroup_reserved

Since commit 7775c8184ec0 ("btrfs: remove unused parameter from
btrfs_subvolume_release_metadata") parameter qgroup_reserved is not used
by caller of function btrfs_subvolume_reserve_metadata.  So remove it.

Signed-off-by: Gu JinXiang <gujx@cn.fujitsu.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: balance dirty metadata pages in btrfs_finish_ordered_io
Ethan Lien [Mon, 28 May 2018 05:48:20 +0000 (13:48 +0800)]
btrfs: balance dirty metadata pages in btrfs_finish_ordered_io

[Problem description and how we fix it]
We should balance dirty metadata pages at the end of
btrfs_finish_ordered_io, since a small, unmergeable random write can
potentially produce dirty metadata which is multiple times larger than
the data itself. For example, a small, unmergeable 4KiB write may
produce:

    16KiB dirty leaf (and possibly 16KiB dirty node) in subvolume tree
    16KiB dirty leaf (and possibly 16KiB dirty node) in checksum tree
    16KiB dirty leaf (and possibly 16KiB dirty node) in extent tree

Although we do call balance dirty pages in write side, but in the
buffered write path, most metadata are dirtied only after we reach the
dirty background limit (which by far only counts dirty data pages) and
wakeup the flusher thread. If there are many small, unmergeable random
writes spread in a large btree, we'll find a burst of dirty pages
exceeds the dirty_bytes limit after we wakeup the flusher thread - which
is not what we expect. In our machine, it caused out-of-memory problem
since a page cannot be dropped if it is marked dirty.

Someone may worry about we may sleep in btrfs_btree_balance_dirty_nodelay,
but since we do btrfs_finish_ordered_io in a separate worker, it will not
stop the flusher consuming dirty pages. Also, we use different worker for
metadata writeback endio, sleep in btrfs_finish_ordered_io help us throttle
the size of dirty metadata pages.

[Reproduce steps]
To reproduce the problem, we need to do 4KiB write randomly spread in a
large btree. In our 2GiB RAM machine:

1) Create 4 subvolumes.
2) Run fio on each subvolume:

   [global]
   direct=0
   rw=randwrite
   ioengine=libaio
   bs=4k
   iodepth=16
   numjobs=1
   group_reporting
   size=128G
   runtime=1800
   norandommap
   time_based
   randrepeat=0

3) Take snapshot on each subvolume and repeat fio on existing files.
4) Repeat step (3) until we get large btrees.
   In our case, by observing btrfs_root_item->bytes_used, we have 2GiB of
   metadata in each subvolume tree and 12GiB of metadata in extent tree.
5) Stop all fio, take snapshot again, and wait until all delayed work is
   completed.
6) Start all fio. Few seconds later we hit OOM when the flusher starts
   to work.

It can be reproduced even when using nocow write.

Signed-off-by: Ethan Lien <ethanlien@synology.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ add comment ]
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: lift some btrfs_cross_ref_exist checks in nocow path
Ethan Lien [Thu, 17 May 2018 06:58:29 +0000 (14:58 +0800)]
btrfs: lift some btrfs_cross_ref_exist checks in nocow path

In nocow path, we check if the extent is snapshotted in
btrfs_cross_ref_exist(). We can do the similar check earlier and avoid
unnecessary search into extent tree.

A fio test on a Intel D-1531, 16GB RAM, SSD RAID-5 machine as follows:

[global]
group_reporting
time_based
thread=1
ioengine=libaio
bs=4k
iodepth=32
size=64G
runtime=180
numjobs=8
rw=randwrite

[file1]
filename=/mnt/nocow/testfile

IOPS result:   unpatched     patched

1 fio round:     46670        46958
snapshot
2 fio round:     51826        54498
3 fio round:     59767        61289

After snapshot, the first fio get about 5% performance gain. As we
continually write to the same file, all writes will resume to nocow mode
and eventually we have no performance gain.

Signed-off-by: Ethan Lien <ethanlien@synology.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ update comments ]
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Remove fs_info argument from btrfs_uuid_tree_rem
Lu Fengqi [Tue, 29 May 2018 07:01:54 +0000 (15:01 +0800)]
btrfs: Remove fs_info argument from btrfs_uuid_tree_rem

This function always takes a transaction handle which contains a
reference to the fs_info. Use that and remove the extra argument.

Signed-off-by: Lu Fengqi <lufq.fnst@cn.fujitsu.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
[ rename the function ]
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Remove fs_info argument from btrfs_uuid_tree_add
Lu Fengqi [Tue, 29 May 2018 07:01:53 +0000 (15:01 +0800)]
btrfs: Remove fs_info argument from btrfs_uuid_tree_add

This function always takes a transaction handle which contains a
reference to the fs_info. Use that and remove the extra argument.

Signed-off-by: Lu Fengqi <lufq.fnst@cn.fujitsu.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: remove unused check of skip_locking
Liu Bo [Tue, 29 May 2018 13:27:06 +0000 (21:27 +0800)]
Btrfs: remove unused check of skip_locking

The check is superfluous since all callers who set search_for_commit
also have skip_locking set.

ASSERT() is put in place to ensure skip_locking is set by new callers.

Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Liu Bo <bo.liu@linux.alibaba.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: remove always true check in unlock_up
Liu Bo [Fri, 18 May 2018 03:00:24 +0000 (11:00 +0800)]
Btrfs: remove always true check in unlock_up

As unlock_up() is written as

for () {
   if (!path->locks[i])
       break;
   ...
   if (... && path->locks[i]) {
   }
}

Apparently, @path->locks[i] is always true at this 'if'.

Signed-off-by: Liu Bo <bo.liu@linux.alibaba.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: grab write lock directly if write_lock_level is the max level
Liu Bo [Fri, 18 May 2018 03:00:23 +0000 (11:00 +0800)]
Btrfs: grab write lock directly if write_lock_level is the max level

Typically, when acquiring root node's lock, btrfs tries its best to get
read lock and trade for write lock if @write_lock_level implies to do so.

In case of (cow && (p->keep_locks || p->lowest_level)), write_lock_level
is set to BTRFS_MAX_LEVEL, which means we need to acquire root node's
write lock directly.

In this particular case, the dance of acquiring read lock and then trading
for write lock can be saved.

Signed-off-by: Liu Bo <bo.liu@linux.alibaba.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: move get root out of btrfs_search_slot to a helper
Liu Bo [Fri, 18 May 2018 03:00:21 +0000 (11:00 +0800)]
Btrfs: move get root out of btrfs_search_slot to a helper

It's good to have a helper instead of having all get-root details
open-coded.  The new helper locks (if necessary) and sets root node of
the path.

Also invert the checks to make the code flow easier to read.  There is
no functional change in this commit.

Signed-off-by: Liu Bo <bo.liu@linux.alibaba.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: use more straightforward extent_buffer_uptodate check
Liu Bo [Fri, 18 May 2018 03:00:20 +0000 (11:00 +0800)]
Btrfs: use more straightforward extent_buffer_uptodate check

If parent_transid "0" is passed to btrfs_buffer_uptodate(),
btrfs_buffer_uptodate() is equivalent to extent_buffer_uptodate(), but
extent_buffer_uptodate() is preferred since we don't have to look into
verify_parent_transid().

Signed-off-by: Liu Bo <bo.liu@linux.alibaba.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: remove superfluous free_extent_buffer in read_block_for_search
Liu Bo [Fri, 18 May 2018 03:00:19 +0000 (11:00 +0800)]
Btrfs: remove superfluous free_extent_buffer in read_block_for_search

read_block_for_search() can be simplified as:

tmp = find_extent_buffer();
if (tmp)
   return;

...

free_extent_buffer();
read_tree_block();

Apparently, @tmp must be NULL at this point, free_extent_buffer() is not
needed.

Signed-off-by: Liu Bo <bo.liu@linux.alibaba.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: drop unused space_info parameter from create_space_info
Lu Fengqi [Mon, 28 May 2018 06:30:27 +0000 (14:30 +0800)]
btrfs: drop unused space_info parameter from create_space_info

Since commit dc2d3005d27d ("btrfs: remove dead create_space_info
calls"), there is only one caller btrfs_init_space_info. However, it
doesn't need create_space_info to return space_info at all.

Signed-off-by: Lu Fengqi <lufq.fnst@cn.fujitsu.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: add parent_transid parameter to veirfy_level_key
Liu Bo [Fri, 18 May 2018 02:59:35 +0000 (10:59 +0800)]
Btrfs: add parent_transid parameter to veirfy_level_key

As verify_level_key() is checked after verify_parent_transid(), i.e.

if (verify_parent_transid())
   ret = -EIO;
else if (verify_level_key())
   ret = -EUCLEAN;

if parent_transid is 0, verify_parent_transid() skips verifying
parent_transid and considers eb as valid, and if verify_level_key()
reports something wrong, we're not going to know if it's caused by
corrupted metadata or non-checkecd eb (e.g. stale eb).

The stale eb can be from an outdated raid1 mirror after a degraded
mount, see eg "btrfs: fix reading stale metadata blocks after degraded
raid1 mounts" (02a3307aa9c20b4f66262) for more details.

@parent_transid is able to tell whether the eb's generation has been
verified by the caller.

Signed-off-by: Liu Bo <bo.liu@linux.alibaba.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: qgroup: show more meaningful qgroup_rescan_init error message
Qu Wenruo [Wed, 2 May 2018 05:28:03 +0000 (13:28 +0800)]
btrfs: qgroup: show more meaningful qgroup_rescan_init error message

Error message from qgroup_rescan_init() mostly looks like:

  BTRFS info (device nvme0n1p1): qgroup_rescan_init failed with -115

Which is far from meaningful, and sometimes confusing as for above
-EINPROGRESS it's mostly (despite the init race) harmless, but sometimes
it can also indicate problem if the return value is -EINVAL.

Change it to some more meaningful messages like:

  BTRFS info (device nvme0n1p1): qgroup rescan is already in progress

And

  BTRFS err(device nvme0n1p1): qgroup rescan init failed, qgroup is not enabled

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
[ update the messages and level ]
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: fix memory and mount leak in btrfs_ioctl_rm_dev_v2()
Omar Sandoval [Tue, 22 May 2018 22:44:01 +0000 (15:44 -0700)]
Btrfs: fix memory and mount leak in btrfs_ioctl_rm_dev_v2()

If we have invalid flags set, when we error out we must drop our writer
counter and free the buffer we allocated for the arguments. This bug is
trivially reproduced with the following program on 4.7+:

#include <fcntl.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <sys/ioctl.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <linux/btrfs.h>
#include <linux/btrfs_tree.h>

int main(int argc, char **argv)
{
struct btrfs_ioctl_vol_args_v2 vol_args = {
.flags = UINT64_MAX,
};
int ret;
int fd;

if (argc != 2) {
fprintf(stderr, "usage: %s PATH\n", argv[0]);
return EXIT_FAILURE;
}

fd = open(argv[1], O_WRONLY);
if (fd == -1) {
perror("open");
return EXIT_FAILURE;
}

ret = ioctl(fd, BTRFS_IOC_RM_DEV_V2, &vol_args);
if (ret == -1)
perror("ioctl");

close(fd);
return EXIT_SUCCESS;
}

When unmounting the filesystem, we'll hit the
WARN_ON(mnt_get_writers(mnt)) in cleanup_mnt() and also may prevent the
filesystem to be remounted read-only as the writer count will stay
lifted.

Fixes: 6b526ed70cf1 ("btrfs: introduce device delete by devid")
CC: stable@vger.kernel.org # 4.9+
Signed-off-by: Omar Sandoval <osandov@fb.com>
Reviewed-by: Su Yue <suy.fnst@cn.fujitsu.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: lzo: Harden inline lzo compressed extent decompression
Qu Wenruo [Thu, 17 May 2018 06:10:29 +0000 (14:10 +0800)]
btrfs: lzo: Harden inline lzo compressed extent decompression

For inlined extent, we only have one segment, thus less things to check.
And further more, inlined extent always has the csum in its leaf header,
it's less probable to have corrupted data.

Anyway, still check header and segment header.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: lzo: Add header length check to avoid potential out-of-bounds access
Qu Wenruo [Tue, 15 May 2018 06:57:51 +0000 (14:57 +0800)]
btrfs: lzo: Add header length check to avoid potential out-of-bounds access

James Harvey reported that some corrupted compressed extent data can
lead to various kernel memory corruption.

Such corrupted extent data belongs to inode with NODATASUM flags, thus
data csum won't help us detecting such bug.

If lucky enough, KASAN could catch it like:

BUG: KASAN: slab-out-of-bounds in lzo_decompress_bio+0x384/0x7a0 [btrfs]
Write of size 4096 at addr ffff8800606cb0f8 by task kworker/u16:0/2338

CPU: 3 PID: 2338 Comm: kworker/u16:0 Tainted: G           O      4.17.0-rc5-custom+ #50
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 0.0.0 02/06/2015
Workqueue: btrfs-endio btrfs_endio_helper [btrfs]
Call Trace:
 dump_stack+0xc2/0x16b
 print_address_description+0x6a/0x270
 kasan_report+0x260/0x380
 memcpy+0x34/0x50
 lzo_decompress_bio+0x384/0x7a0 [btrfs]
 end_compressed_bio_read+0x99f/0x10b0 [btrfs]
 bio_endio+0x32e/0x640
 normal_work_helper+0x15a/0xea0 [btrfs]
 process_one_work+0x7e3/0x1470
 worker_thread+0x1b0/0x1170
 kthread+0x2db/0x390
 ret_from_fork+0x22/0x40
...

The offending compressed data has the following info:

Header: length 32768 (looks completely valid)
Segment 0 Header: length 3472882419 (obviously out of bounds)

Then when handling segment 0, since it's over the current page, we need
the copy the compressed data to temporary buffer in workspace, then such
large size would trigger out-of-bounds memory access, screwing up the
whole kernel.

Fix it by adding extra checks on header and segment headers to ensure we
won't access out-of-bounds, and even checks the decompressed data won't
be out-of-bounds.

Reported-by: James Harvey <jamespharvey20@gmail.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Misono Tomohiro <misono.tomohiro@jp.fujitsu.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ updated comments ]
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: lzo: document the compressed data format
Qu Wenruo [Thu, 17 May 2018 05:10:01 +0000 (13:10 +0800)]
btrfs: lzo: document the compressed data format

Although it's not that complex, but such comment could still save
several minutes for newer reader/reviewer instead of inferring that from
the code.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ minor wording updates ]
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: compression: Add linux/sizes.h for compression.h
Qu Wenruo [Thu, 17 May 2018 05:52:22 +0000 (13:52 +0800)]
btrfs: compression: Add linux/sizes.h for compression.h

Since compression.h is using the SZ_* macros, and if some file includes
only compression.h without linux/sizes.h, it will cause compile error.

One example is lzo.c, if it uses BTRFS_MAX_COMPRESSED.  Fix it by adding
linux/sizes.h in compression.h

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: fix clone vs chattr NODATASUM race
Omar Sandoval [Tue, 22 May 2018 22:02:12 +0000 (15:02 -0700)]
Btrfs: fix clone vs chattr NODATASUM race

In btrfs_clone_files(), we must check the NODATASUM flag while the
inodes are locked. Otherwise, it's possible that btrfs_ioctl_setflags()
will change the flags after we check and we can end up with a party
checksummed file.

The race window is only a few instructions in size, between the if and
the locks which is:

3834         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
3835                 return -EISDIR;

where the setflags must be run and toggle the NODATASUM flag (provided
the file size is 0).  The clone will block on the inode lock, segflags
takes the inode lock, changes flags, releases log and clone continues.

Not impossible but still needs a lot of bad luck to hit unintentionally.

Fixes: 0e7b824c4ef9 ("Btrfs: don't make a file partly checksummed through file clone")
CC: stable@vger.kernel.org # 4.4+
Signed-off-by: Omar Sandoval <osandov@fb.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ update changelog ]
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: propagate failures of __exclude_logged_extent to upper caller
Gu Jinxiang [Tue, 22 May 2018 09:46:51 +0000 (17:46 +0800)]
btrfs: propagate failures of __exclude_logged_extent to upper caller

Function btrfs_exclude_logged_extents may call __exclude_logged_extent
which may fail.
Propagate the failures of __exclude_logged_extent to upper caller.

Signed-off-by: Gu Jinxiang <gujx@cn.fujitsu.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Streamline shared ref check in alloc_reserved_tree_block
Nikolay Borisov [Mon, 21 May 2018 09:27:23 +0000 (12:27 +0300)]
btrfs: Streamline shared ref check in alloc_reserved_tree_block

Instead of setting "parent" to ref->parent only when dealing with
a shared ref and subsequently performing another check to see
if (parent > 0), check the "node->type" directly and act accordingly.
This makes the code more streamline. No functional changes.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Pass btrfs_delayed_extent_op to alloc_reserved_tree_block
Nikolay Borisov [Mon, 21 May 2018 09:27:22 +0000 (12:27 +0300)]
btrfs: Pass btrfs_delayed_extent_op to alloc_reserved_tree_block

Instead of taking only specific member of this structure, which results
in 2 extra arguments, just take the delayed_extent_op struct and
reference the arguments inside the functions. No functional changes.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Simplify alloc_reserved_tree_block interface
Nikolay Borisov [Mon, 21 May 2018 09:27:21 +0000 (12:27 +0300)]
btrfs: Simplify alloc_reserved_tree_block interface

This function currently takes 7 parameters, most of which are proxies
for values from btrfs_delayed_ref_node struct which is not passed. This
patch simplifies the interface of the function by simply passing said
delayed ref node struct to the function. This enables us to:

1. Move locals variables and init code related to them from
   run_delayed_tree_ref which should only be used inside
   alloc_reserved_tree_block, such as skinny_metadata and the btrfs_key,
   representing the extent being inserted. This removes the need for the
   "ins" argument. Instead, it's replaced by a local var with a more
   verbose name - extent_key.

2. Now that we have a reference to the node in alloc_reserved_tree_block
   the delayed_tree_ref struct can be referenced inside the function and
   this enable removing the "ref->level", "parent" and "ref_root"
   arguments.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Remove fs_info argument from alloc_reserved_tree_block
Nikolay Borisov [Mon, 21 May 2018 09:27:20 +0000 (12:27 +0300)]
btrfs: Remove fs_info argument from alloc_reserved_tree_block

This function already takes a transaction handle which contains a
reference to the fs_info. So use this and remove the extra argument.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: tests: drop newline from test_msg strings
David Sterba [Wed, 16 May 2018 22:00:44 +0000 (00:00 +0200)]
btrfs: tests: drop newline from test_msg strings

Now that test_err strings do not need the newline, remove them also from
the test_msg.

Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: tests: add helper for error messages and update them
David Sterba [Wed, 16 May 2018 22:00:42 +0000 (00:00 +0200)]
btrfs: tests: add helper for error messages and update them

The test failures are not clearly visible in the system log as they're
printed at INFO level. Add a new helper that is level ERROR. As this
touches almost all strings, I took the opportunity to unify them:

- decapitalize the first letter as there's a prefix and the text
  continues after ":"
- glue strings split to more lines and un-indent so they fit to 80
  columns
- use %llu instead of %Lu
- drop \n from the modified messages (test_msg is left untouched)

Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: use error code returned by btrfs_read_fs_root_no_name in search ioctl
Misono Tomohiro [Mon, 21 May 2018 04:57:27 +0000 (13:57 +0900)]
btrfs: use error code returned by btrfs_read_fs_root_no_name in search ioctl

btrfs_read_fs_root_no_name() may return ERR_PTR(-ENOENT) or
ERR_PTR(-ENOMEM) and therefore search_ioctl() and
btrfs_search_path_in_tree() should use PTR_ERR() instead of -ENOENT,
which all other callers of btrfs_read_fs_root_no_name() do.

Drop the error message as it would be confusing, the caller of ioctl
will likely interpret the error code and not look into the syslog.

Signed-off-by: Tomohiro Misono <misono.tomohiro@jp.fujitsu.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: allow empty subvol= again
Omar Sandoval [Tue, 22 May 2018 00:07:19 +0000 (17:07 -0700)]
Btrfs: allow empty subvol= again

I got a report that after upgrading to 4.16, someone's filesystems
weren't mounting:

[   23.845852] BTRFS info (device loop0): unrecognized mount option 'subvol='

Before 4.16, this mounted the default subvolume. It turns out that this
empty "subvol=" is actually an application bug, but it was causing the
application to fail, so it's an ABI break if you squint.

The generic parsing code we use for mount options (match_token())
doesn't match an empty string as "%s". Previously, setup_root_args()
removed the "subvol=" string, but the mount path was cleaned up to not
need that. Add a dummy Opt_subvol_empty to fix this.

The simple workaround is to use / or . for the value of 'subvol=' .

Fixes: 312c89fbca06 ("btrfs: cleanup btrfs_mount() using btrfs_mount_root()")
CC: stable@vger.kernel.org # 4.16+
Signed-off-by: Omar Sandoval <osandov@fb.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: fix describe_relocation when printing unknown flags
Anand Jain [Thu, 17 May 2018 13:25:12 +0000 (21:25 +0800)]
btrfs: fix describe_relocation when printing unknown flags

Looks like the original idea was to print the hex of the flags which is
not coded with their flag name. So use the current buf pointer bp
instead of buf.

Reaching the uknown flags should never happen, it's there just in case.

Fixes: ebce0e01b930b ("btrfs: make block group flags in balance printks human-readable")
Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: use kvzalloc for EXTENT_SAME temporary data
David Sterba [Fri, 11 May 2018 15:57:54 +0000 (17:57 +0200)]
btrfs: use kvzalloc for EXTENT_SAME temporary data

The dedupe range is 16 MiB, with 4 KiB pages and 8 byte pointers, the
arrays can be 32KiB large. To avoid allocation failures due to
fragmented memory, use the allocation with fallback to vmalloc.

The arrays are allocated and freed only inside btrfs_extent_same and
reused for all the ranges.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: reuse cmp workspace in EXTENT_SAME ioctl
Timofey Titovets [Wed, 2 May 2018 05:15:38 +0000 (08:15 +0300)]
Btrfs: reuse cmp workspace in EXTENT_SAME ioctl

We support big dedup requests by splitting range to smaller parts, and
call dedupe logic on each of them.

Instead of repeated allocation and deallocation, allocate once at the
beginning and reuse in the iteration.

Signed-off-by: Timofey Titovets <nefelim4ag@gmail.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: dedupe_file_range ioctl: remove 16MiB restriction
Timofey Titovets [Wed, 2 May 2018 05:15:37 +0000 (08:15 +0300)]
Btrfs: dedupe_file_range ioctl: remove 16MiB restriction

Currently btrfs_dedupe_file_range silently restricts the dedupe range to
to 16MiB to limit locking and working memory size and is documented in
manual page as implementation specific.

Let's remove that restriction by iterating over the dedup range in 16MiB
steps.  This is backward compatible and will not change anything for
requests smaller then 16MiB.

Signed-off-by: Timofey Titovets <nefelim4ag@gmail.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: split btrfs_extent_same
Timofey Titovets [Wed, 2 May 2018 05:15:36 +0000 (08:15 +0300)]
Btrfs: split btrfs_extent_same

Split btrfs_extent_same() to two parts where one is the main EXTENT_SAME
entry and a helper that can be repeatedly called on a range.  This will
be used in following patches.

Signed-off-by: Timofey Titovets <nefelim4ag@gmail.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: reserve space for O_TMPFILE orphan item deletion
Omar Sandoval [Fri, 11 May 2018 20:13:40 +0000 (13:13 -0700)]
Btrfs: reserve space for O_TMPFILE orphan item deletion

btrfs_link() calls btrfs_orphan_del() if it's linking an O_TMPFILE but
it doesn't reserve space to do so. Even before the removal of the
orphan_block_rsv it wasn't using it.

Fixes: ef3b9af50bfa ("Btrfs: implement inode_operations callback tmpfile")
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: renumber BTRFS_INODE_ runtime flags and switch to enums
Omar Sandoval [Fri, 11 May 2018 20:13:39 +0000 (13:13 -0700)]
Btrfs: renumber BTRFS_INODE_ runtime flags and switch to enums

We got rid of BTRFS_INODE_HAS_ORPHAN_ITEM and
BTRFS_INODE_ORPHAN_META_RESERVED, so we can renumber the flags to make
them consecutive again.

Signed-off-by: Omar Sandoval <osandov@fb.com>
[ switch them enums so we don't have to do that again ]
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: get rid of unused orphan infrastructure
Omar Sandoval [Fri, 11 May 2018 20:13:38 +0000 (13:13 -0700)]
Btrfs: get rid of unused orphan infrastructure

Now that we don't keep long-standing reservations for orphan items,
root->orphan_block_rsv isn't used. We can git rid of it, along with:

- root->orphan_lock, which was used to protect root->orphan_block_rsv
- root->orphan_inodes, which was used as a refcount for root->orphan_block_rsv
- BTRFS_INODE_ORPHAN_META_RESERVED, which was used to track reservations
  in root->orphan_block_rsv
- btrfs_orphan_commit_root(), which was the last user of any of these
  and does nothing else

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: fix ENOSPC caused by orphan items reservations
Omar Sandoval [Fri, 11 May 2018 20:13:37 +0000 (13:13 -0700)]
Btrfs: fix ENOSPC caused by orphan items reservations

Currently, we keep space reserved for all inode orphan items until the
inode is evicted (i.e., all references to it are dropped). We hit an
issue where an application would keep a bunch of deleted files open (by
design) and thus keep a large amount of space reserved, causing ENOSPC
errors when other operations tried to reserve space. This long-standing
reservation isn't absolutely necessary for a couple of reasons:

- We can almost always make the reservation we need or steal from the
  global reserve for the orphan item
- If we can't, it's not the end of the world if we drop the orphan item
  on the floor and let the next mount clean it up

So, get rid of persistent reservation and just reserve space in
btrfs_evict_inode().

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: refactor btrfs_evict_inode() reserve refill dance
Omar Sandoval [Fri, 11 May 2018 20:13:36 +0000 (13:13 -0700)]
Btrfs: refactor btrfs_evict_inode() reserve refill dance

The truncate loop in btrfs_evict_inode() does two things at once:

- It refills the temporary block reserve, potentially stealing from the
  global reserve or committing
- It calls btrfs_truncate_inode_items()

The tangle of continues hides the fact that these two steps are actually
separate. Split the first step out into a separate function both for
clarity and so that we can reuse it in a later patch.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: don't return ino to ino cache if inode item removal fails
Omar Sandoval [Fri, 11 May 2018 20:13:35 +0000 (13:13 -0700)]
Btrfs: don't return ino to ino cache if inode item removal fails

In btrfs_evict_inode(), if btrfs_truncate_inode_items() fails, the inode
item will still be in the tree but we still return the ino to the ino
cache. That will blow up later when someone tries to allocate that ino,
so don't return it to the cache.

Fixes: 581bb050941b ("Btrfs: Cache free inode numbers in memory")
Reviewed-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: delete dead code in btrfs_orphan_commit_root()
Omar Sandoval [Fri, 11 May 2018 20:13:34 +0000 (13:13 -0700)]
Btrfs: delete dead code in btrfs_orphan_commit_root()

btrfs_orphan_commit_root() tries to delete an orphan item for a
subvolume in the tree root, but we don't actually insert that item in
the first place. See commit 0a0d4415e338 ("Btrfs: delete dead code in
btrfs_orphan_add()"). We can get rid of it.

Reviewed-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: get rid of BTRFS_INODE_HAS_ORPHAN_ITEM
Omar Sandoval [Fri, 11 May 2018 20:13:33 +0000 (13:13 -0700)]
Btrfs: get rid of BTRFS_INODE_HAS_ORPHAN_ITEM

Now that we don't add orphan items for truncate, there can't be races on
adding or deleting an orphan item, so this bit is unnecessary.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: stop creating orphan items for truncate
Omar Sandoval [Fri, 11 May 2018 20:13:32 +0000 (13:13 -0700)]
Btrfs: stop creating orphan items for truncate

Currently, we insert an orphan item during a truncate so that if there's
a crash, we don't leak extents past the on-disk i_size. However, since
commit 7f4f6e0a3f6d ("Btrfs: only update disk_i_size as we remove
extents"), we keep disk_i_size in sync with the extent items as we
truncate, so orphan cleanup will never have any extents to remove. Don't
bother with the superfluous orphan item.

Reviewed-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: don't BUG_ON() in btrfs_truncate_inode_items()
Omar Sandoval [Fri, 11 May 2018 20:13:31 +0000 (13:13 -0700)]
Btrfs: don't BUG_ON() in btrfs_truncate_inode_items()

btrfs_free_extent() can fail because of ENOMEM. There's no reason to
panic here, we can just abort the transaction.

Fixes: f4b9aa8d3b87 ("btrfs_truncate")
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: fix error handling in btrfs_truncate_inode_items()
Omar Sandoval [Fri, 11 May 2018 20:13:30 +0000 (13:13 -0700)]
Btrfs: fix error handling in btrfs_truncate_inode_items()

btrfs_truncate_inode_items() uses two variables for error handling, ret
and err. These are not handled consistently, leading to a couple of
bugs.

- Errors from btrfs_del_items() are handled but not propagated to the
  caller
- If btrfs_run_delayed_refs() fails and aborts the transaction, we
  continue running

Just use ret everywhere and simplify things a bit, fixing both of these
issues.

Fixes: 79787eaab461 ("btrfs: replace many BUG_ONs with proper error handling")
Fixes: 1262133b8d6f ("Btrfs: account for crcs in delayed ref processing")
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: update stale comments referencing vmtruncate()
Omar Sandoval [Fri, 11 May 2018 20:13:29 +0000 (13:13 -0700)]
Btrfs: update stale comments referencing vmtruncate()

Commit a41ad394a03b ("Btrfs: convert to the new truncate sequence")
changed btrfs_setsize() to call truncate_setsize() instead of
vmtruncate() but didn't update the comment above it. truncate_setsize()
never fails (the IS_SWAPFILE() check happens elsewhere), so remove the
comment.

Additionally, the comment above btrfs_page_mkwrite() references
vmtruncate(), but truncate_setsize() does the size write and page
locking now.

Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Remove stale comment about select_delayed_ref
Nikolay Borisov [Thu, 17 May 2018 11:16:29 +0000 (14:16 +0300)]
btrfs: Remove stale comment about select_delayed_ref

select_delayed_ref really just gets the next delayed ref which has to
be processed - either an add ref or drop ref. We never go back for
anything. So the comment is actually bogus, just remove it.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: sysfs: Add entry which shows if rmdir can work on subvolumes
Misono Tomohiro [Thu, 17 May 2018 05:24:51 +0000 (14:24 +0900)]
btrfs: sysfs: Add entry which shows if rmdir can work on subvolumes

Deletion of a subvolume by rmdir(2) has become allowed by the
'commit cd2decf640b1 ("btrfs: Allow rmdir(2) to delete an empty
subvolume")'.

It is a kind of new feature and this commits add a sysfs entry

  /sys/fs/btrfs/features/rmdir_subvol

to indicate the availability of the feature so that a user program
(e.g. fstests) can detect it.

Prior to this commit, all entries in /sys/fs/btrfs/features are feature
which depend on feature bits of superblock (i.e. each feature affects
on-disk format) and managed by attribute_group "btrfs_feature_attr_group".
For each fs, entries in /sys/fs/btrfs/UUID/features indicate which
features are enabled (or can be changed online) for the fs.

However, rmdir_subvol feature only depends on kernel module. Therefore
new attribute_group "btrfs_static_feature_attr_group" is introduced and
sysfs_merge_group() is used to share /sys/fs/btrfs/features directory.
Features in "btrfs_static_feature_attr_group" won't be listed in each
/sys/fs/btrfs/UUID/features.

Signed-off-by: Tomohiro Misono <misono.tomohiro@jp.fujitsu.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: sysfs: Use enum/define value for feature array definitions
Tomohiro Misono [Wed, 16 May 2018 08:09:26 +0000 (17:09 +0900)]
btrfs: sysfs: Use enum/define value for feature array definitions

Use existing named values instead of the raw numbers.

Signed-off-by: Tomohiro Misono <misono.tomohiro@jp.fujitsu.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: add prefix "balance:" for log messages
Anand Jain [Wed, 16 May 2018 02:51:26 +0000 (10:51 +0800)]
btrfs: add prefix "balance:" for log messages

Kernel logs are very important for the forensic investigations of the
issues in general make it easy to use it. This patch adds 'balance:'
prefix so that it can be easily searched.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: unify naming of flags variables for SETFLAGS and XFLAGS
David Sterba [Mon, 23 Apr 2018 13:45:18 +0000 (15:45 +0200)]
btrfs: unify naming of flags variables for SETFLAGS and XFLAGS

* The simple 'flags' refer to the btrfs inode
* ... that's in 'binode
* the FS_*_FL variables are 'fsflags'
* the old copies of the variable are prefixed by 'old_'
* Struct inode flags contain 'i_flags'.

Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: add FS_IOC_FSSETXATTR ioctl
David Sterba [Mon, 26 Mar 2018 17:51:16 +0000 (19:51 +0200)]
btrfs: add FS_IOC_FSSETXATTR ioctl

The new ioctl is an extension to the FS_IOC_SETFLAGS and adds new
flags and is extensible. Don't get fooled by the XATTR in the name, it
does not have anything in common with the extended attributes,
incidentally also abbreviated as XATTRs.

This patch allows to set the xflags portion of the fsxattr structure,
other items have no meaning and non-zero values will result in
EOPNOTSUPP.

Currently supported xflags:

- APPEND
- IMMUTABLE
- NOATIME
- NODUMP
- SYNC

The structure of btrfs_ioctl_fssetxattr copies btrfs_ioctl_setflags but
is simpler on the flag setting side.

The original patch was written by Chandan Jay Sharma but was incomplete
and no further revision has been sent.

Based-on-patches-by: Chandan Jay Sharma <chandansbg@gmail.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: add FS_IOC_FSGETXATTR ioctl
David Sterba [Mon, 26 Mar 2018 17:51:16 +0000 (19:51 +0200)]
btrfs: add FS_IOC_FSGETXATTR ioctl

The new ioctl is an extension to the FS_IOC_GETFLAGS and adds new
flags and is extensible. This patch allows to return the xflags portion
of the fsxattr structure, other items have no meaning for btrfs or can
be added later.

The original patch was written by Chandan Jay Sharma but was incomplete
and no further revision has been sent. Several cleanups were necessary
to avoid confusion with other ioctls, as we have another flavor of
flags.

Based-on-patches-by: Chandan Jay Sharma <chandansbg@gmail.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: add helpers for FS_XFLAG_* conversion
David Sterba [Mon, 26 Mar 2018 17:42:05 +0000 (19:42 +0200)]
btrfs: add helpers for FS_XFLAG_* conversion

Preparatory work for the FS_IOC_FSGETXATTR ioctl, basic conversions and
checking helpers.

Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: rename btrfs_flags_to_ioctl to reflect which flags it touches
David Sterba [Mon, 26 Mar 2018 17:12:25 +0000 (19:12 +0200)]
btrfs: rename btrfs_flags_to_ioctl to reflect which flags it touches

Converts btrfs_inode::flags to the FS_*_FL flags.

Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: rename check_flags to reflect which flags it touches
David Sterba [Mon, 26 Mar 2018 16:52:15 +0000 (18:52 +0200)]
btrfs: rename check_flags to reflect which flags it touches

The FS_*_FL flags cannot be easily identified by a prefix but we still
need to recognize them so the 'fsflags' should be closer to the naming
scheme but again the 'fs' part sounds like it's a filesystem flag. I
don't have a better idea for now.

Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: rename btrfs_mask_flags to reflect which flags it touches
David Sterba [Mon, 26 Mar 2018 16:52:15 +0000 (18:52 +0200)]
btrfs: rename btrfs_mask_flags to reflect which flags it touches

The FS_*_FL flags cannot be easily identified by a variable name prefix
but we still need to recognize them so the 'fsflags' should be closer to
the naming scheme but again the 'fs' part sounds like it's a filesystem
flag. I don't have a better idea for now.

Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: rename btrfs_update_iflags to reflect which flags it touches
David Sterba [Mon, 26 Mar 2018 16:40:21 +0000 (18:40 +0200)]
btrfs: rename btrfs_update_iflags to reflect which flags it touches

The btrfs inode flag flavour is now simply called 'inode flags' and the
vfs inode are i_flags.

Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: use common variable for fs_devices in btrfs_destroy_dev_replace_tgtdev
Anand Jain [Thu, 12 Apr 2018 02:29:38 +0000 (10:29 +0800)]
btrfs: use common variable for fs_devices in btrfs_destroy_dev_replace_tgtdev

Use a local btrfs_fs_devices variable to access the structure.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: drop uuid_mutex in btrfs_destroy_dev_replace_tgtdev
Anand Jain [Thu, 12 Apr 2018 02:29:37 +0000 (10:29 +0800)]
btrfs: drop uuid_mutex in btrfs_destroy_dev_replace_tgtdev

Delete the uuid_mutex lock here as this thread accesses the
btrfs_fs_devices::devices only (counters or called functions do a list
traversal). And the device_list_mutex lock is already taken.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ update changelog ]
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: drop uuid_mutex in btrfs_dev_replace_finishing
Anand Jain [Thu, 12 Apr 2018 02:29:36 +0000 (10:29 +0800)]
btrfs: drop uuid_mutex in btrfs_dev_replace_finishing

btrfs_dev_replace_finishing updates devices (soruce and target) which
are within the btrfs_fs_devices::devices or withint the cloned seed
devices (btrfs_fs_devices::seed::devices), so we don't need the global
uuid_mutex.

The device replace context is also locked by its own locks.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: replace uuid_mutex by device_list_mutex in btrfs_open_devices
Anand Jain [Thu, 12 Apr 2018 02:29:34 +0000 (10:29 +0800)]
btrfs: replace uuid_mutex by device_list_mutex in btrfs_open_devices

btrfs_open_devices() is using the uuid_mutex, but as btrfs_open_devices
is just limited to openning all the devices under for given fsid, so we
don't need uuid_mutex.

Instead it should hold the device_list_mutex as it updates the members
of the btrfs_fs_devices and btrfs_device and not the whole fs_devs list.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ update changelog ]
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: document uuid_mutex uasge in read_chunk_tree
Anand Jain [Thu, 12 Apr 2018 02:29:32 +0000 (10:29 +0800)]
btrfs: document uuid_mutex uasge in read_chunk_tree

read_chunk_tree() calls read_one_dev(), but for seed device we have
to search the fs_uuids list, so we need the uuid_mutex. Add a comment
comment, so that we can improve this part.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: use existing cur_devices, cleanup btrfs_rm_device
Anand Jain [Thu, 12 Apr 2018 02:29:31 +0000 (10:29 +0800)]
btrfs: use existing cur_devices, cleanup btrfs_rm_device

Instead of de-referencing the device->fs_devices use cur_devices
which points to the same fs_devices and does not change.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: reduce uuid_mutex critical section while scanning devices
Anand Jain [Thu, 12 Apr 2018 02:29:24 +0000 (10:29 +0800)]
btrfs: reduce uuid_mutex critical section while scanning devices

The generic block device lookup or cleanup does not need the uuid mutex,
that's only for the device_list_add.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ update changelog ]
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Unexport and rename btrfs_invalidate_inodes
Nikolay Borisov [Fri, 27 Apr 2018 11:36:24 +0000 (14:36 +0300)]
btrfs: Unexport and rename btrfs_invalidate_inodes

This function is no longer used outside of inode.c so just make it
static. At the same time give a more becoming name, since it's not
really invalidating the inodes but just calling d_prune_alias. Last,
but not least - move the function above the sole caller to avoid
introducing yet-another-pointless forward declaration.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: replace waitqueue_actvie with cond_wake_up
David Sterba [Mon, 26 Feb 2018 15:15:17 +0000 (16:15 +0100)]
btrfs: replace waitqueue_actvie with cond_wake_up

Use the wrappers and reduce the amount of low-level details about the
waitqueue management.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: add barriers to btrfs_sync_log before log_commit_wait wakeups
David Sterba [Tue, 24 Apr 2018 12:53:56 +0000 (14:53 +0200)]
btrfs: add barriers to btrfs_sync_log before log_commit_wait wakeups

Currently the code assumes that there's an implied barrier by the
sequence of code preceding the wakeup, namely the mutex unlock.

As Nikolay pointed out:

I think this is wrong (not your code) but the original assumption that
the RELEASE semantics provided by mutex_unlock is sufficient.
According to memory-barriers.txt:

Section 'LOCK ACQUISITION FUNCTIONS' states:

 (2) RELEASE operation implication:

     Memory operations issued before the RELEASE will be completed before the
     RELEASE operation has completed.

     Memory operations issued after the RELEASE *may* be completed before the
     RELEASE operation has completed.

(I've bolded the may portion)

The example given there:

As an example, consider the following:

    *A = a;
    *B = b;
    ACQUIRE
    *C = c;
    *D = d;
    RELEASE
    *E = e;
    *F = f;

The following sequence of events is acceptable:

    ACQUIRE, {*F,*A}, *E, {*C,*D}, *B, RELEASE

So if we assume that *C is modifying the flag which the waitqueue is checking,
and *E is the actual wakeup, then those accesses can be re-ordered...

IMHO this code should be considered broken...
---

To be on the safe side, add the barriers. The synchronization logic
around log using the mutexes and several other threads does not make it
easy to reason for/against the barrier.

CC: Nikolay Borisov <nborisov@suse.com>
Link: https://lkml.kernel.org/r/6ee068d8-1a69-3728-00d1-d86293d43c9f@suse.com
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: introduce conditional wakeup helpers
David Sterba [Mon, 26 Feb 2018 14:43:18 +0000 (15:43 +0100)]
btrfs: introduce conditional wakeup helpers

Add convenience wrappers for the waitqueue management that involves
memory barriers to prevent deadlocks. The helpers will let us remove
barriers and the necessary comments in several places.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: qgroup: Finish rescan when hit the last leaf of extent tree
Qu Wenruo [Mon, 14 May 2018 01:38:13 +0000 (09:38 +0800)]
btrfs: qgroup: Finish rescan when hit the last leaf of extent tree

Under the following case, qgroup rescan can double account cowed tree
blocks:

In this case, extent tree only has one tree block.

-
| transid=5 last committed=4
| btrfs_qgroup_rescan_worker()
| |- btrfs_start_transaction()
| |  transid = 5
| |- qgroup_rescan_leaf()
|    |- btrfs_search_slot_for_read() on extent tree
|       Get the only extent tree block from commit root (transid = 4).
|       Scan it, set qgroup_rescan_progress to the last
|       EXTENT/META_ITEM + 1
|       now qgroup_rescan_progress = A + 1.
|
| fs tree get CoWed, new tree block is at A + 16K
| transid 5 get committed
-
| transid=6 last committed=5
| btrfs_qgroup_rescan_worker()
| btrfs_qgroup_rescan_worker()
| |- btrfs_start_transaction()
| |  transid = 5
| |- qgroup_rescan_leaf()
|    |- btrfs_search_slot_for_read() on extent tree
|       Get the only extent tree block from commit root (transid = 5).
|       scan it using qgroup_rescan_progress (A + 1).
|       found new tree block beyong A, and it's fs tree block,
|       account it to increase qgroup numbers.
-

In above case, tree block A, and tree block A + 16K get accounted twice,
while qgroup rescan should stop when it already reach the last leaf,
other than continue using its qgroup_rescan_progress.

Such case could happen by just looping btrfs/017 and with some
possibility it can hit such double qgroup accounting problem.

Fix it by checking the path to determine if we should finish qgroup
rescan, other than relying on next loop to exit.

Reported-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: qgroup: Search commit root for rescan to avoid missing extent
Qu Wenruo [Mon, 14 May 2018 01:38:12 +0000 (09:38 +0800)]
btrfs: qgroup: Search commit root for rescan to avoid missing extent

When doing qgroup rescan using the following script (modified from
btrfs/017 test case), we can sometimes hit qgroup corruption.

------
umount $dev &> /dev/null
umount $mnt &> /dev/null

mkfs.btrfs -f -n 64k $dev
mount $dev $mnt

extent_size=8192

xfs_io -f -d -c "pwrite 0 $extent_size" $mnt/foo > /dev/null
btrfs subvolume snapshot $mnt $mnt/snap

xfs_io -f -c "reflink $mnt/foo" $mnt/foo-reflink > /dev/null
xfs_io -f -c "reflink $mnt/foo" $mnt/snap/foo-reflink > /dev/null
xfs_io -f -c "reflink $mnt/foo" $mnt/snap/foo-reflink2 > /dev/unll
btrfs quota enable $mnt

 # -W is the new option to only wait rescan while not starting new one
btrfs quota rescan -W $mnt
btrfs qgroup show -prce $mnt
umount $mnt

 # Need to patch btrfs-progs to report qgroup mismatch as error
btrfs check $dev || _fail
------

For fast machine, we can hit some corruption which missed accounting
tree blocks:
------
qgroupid         rfer         excl     max_rfer     max_excl parent  child
--------         ----         ----     --------     -------- ------  -----
0/5           8.00KiB        0.00B         none         none ---     ---
0/257         8.00KiB        0.00B         none         none ---     ---
------

This is due to the fact that we're always searching commit root for
btrfs_find_all_roots() at qgroup_rescan_leaf(), but the leaf we get is
from current transaction, not commit root.

And if our tree blocks get modified in current transaction, we won't
find any owner in commit root, thus causing the corruption.

Fix it by searching commit root for extent tree for
qgroup_rescan_leaf().

Reported-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: take the last remnants of ->d_fsdata use out
Al Viro [Sun, 13 May 2018 18:03:18 +0000 (19:03 +0100)]
btrfs: take the last remnants of ->d_fsdata use out

[spotted while going through ->d_fsdata handling around d_splice_alias();
don't really care which tree that goes through]

The only thing even looking at ->d_fsdata in there (since 2012)
had been kfree(dentry->d_fsdata) in btrfs_dentry_delete().  Which,
incidentally, is all btrfs_dentry_delete() does.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Do super block verification before writing it to disk
Qu Wenruo [Fri, 11 May 2018 05:35:27 +0000 (13:35 +0800)]
btrfs: Do super block verification before writing it to disk

There are already 2 reports about strangely corrupted super blocks,
where csum still matches but extra garbage gets slipped into super block.

The corruption would looks like:
------
superblock: bytenr=65536, device=/dev/sdc1
---------------------------------------------------------
csum_type               41700 (INVALID)
csum                    0x3b252d3a [match]
bytenr                  65536
flags                   0x1
                        ( WRITTEN )
magic                   _BHRfS_M [match]
...
incompat_flags          0x5b22400000000169
                        ( MIXED_BACKREF |
                          COMPRESS_LZO |
                          BIG_METADATA |
                          EXTENDED_IREF |
                          SKINNY_METADATA |
                          unknown flag: 0x5b22400000000000 )
...
------
Or
------
superblock: bytenr=65536, device=/dev/mapper/x
---------------------------------------------------------
csum_type              35355 (INVALID)
csum_size              32
csum                   0xf0dbeddd [match]
bytenr                 65536
flags                  0x1
                       ( WRITTEN )
magic                  _BHRfS_M [match]
...
incompat_flags         0x176d200000000169
                       ( MIXED_BACKREF |
                         COMPRESS_LZO |
                         BIG_METADATA |
                         EXTENDED_IREF |
                         SKINNY_METADATA |
                         unknown flag: 0x176d200000000000 )
------

Obviously, csum_type and incompat_flags get some garbage, but its csum
still matches, which means kernel calculates the csum based on corrupted
super block memory.
And after manually fixing these values, the filesystem is completely
healthy without any problem exposed by btrfs check.

Although the cause is still unknown, at least detect it and prevent further
corruption.

Both reports have same symptoms, there's an overwrite on offset 192 of
the superblock, by 4 bytes. The superblock structure is not allocated or
freed and stays in the memory for the whole filesystem lifetime, so it's
not a use-after-free kind of error on someone else's leaked page.

As a vague point for the problable cause is mentioning of other system
freezing related to graphic card drivers.

Reported-by: Ken Swenson <flat@imo.uto.moe>
Reported-by: Ben Parsons <9parsonsb@gmail.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ add brief analysis of the reports ]
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Refactor btrfs_check_super_valid
Qu Wenruo [Fri, 11 May 2018 05:35:26 +0000 (13:35 +0800)]
btrfs: Refactor btrfs_check_super_valid

Refactor btrfs_check_super_valid:

1) Rename it to btrfs_validate_mount_super()
   Now it's more obvious when the function should be called.

2) Extract core check routine into validate_super()
   Later write time check can reuse it, and if needed, we could also
   use validate_super() to check each super block.

3) Add more comments about btrfs_validate_mount_super()
   Mostly about what it doesn't check and when it should be called.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ rename to validate_super ]
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Move btrfs_check_super_valid() to avoid forward declaration
Qu Wenruo [Fri, 11 May 2018 05:35:25 +0000 (13:35 +0800)]
btrfs: Move btrfs_check_super_valid() to avoid forward declaration

Move btrfs_check_super_valid() before its single caller to avoid forward
declaration.

Though such code motion is not recommended as it pollutes git history,
in this case the following patches would need to add new forward
declarations for static functions that we want to avoid.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Remove fs_info argument from populate_free_space_tree
Nikolay Borisov [Thu, 10 May 2018 12:44:56 +0000 (15:44 +0300)]
btrfs: Remove fs_info argument from populate_free_space_tree

This function always takes a transaction handle which contains a
reference to the fs_info. Use that and remove the extra argument.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Remove fs_info argument from add_to_free_space_tree
Nikolay Borisov [Thu, 10 May 2018 12:44:55 +0000 (15:44 +0300)]
btrfs: Remove fs_info argument from add_to_free_space_tree

This function takes a transaction handle which already contains a
reference to the fs_info. So use it and remove the extra function
argument.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Remove fs_info argument from remove_from_free_space_tree
Nikolay Borisov [Thu, 10 May 2018 12:44:54 +0000 (15:44 +0300)]
btrfs: Remove fs_info argument from remove_from_free_space_tree

This function alreay takes a transaction handle which holds a reference
to the fs_info. Use that and remove the extra argument.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Remove fs_info argument from __remove_from_free_space_tree
Nikolay Borisov [Thu, 10 May 2018 12:44:53 +0000 (15:44 +0300)]
btrfs: Remove fs_info argument from __remove_from_free_space_tree

This function takes a transaction handle which holds a reference to
fs_info. So use that and remove the extra argument.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>