xfs: _{attr,data}_map_shared should take ILOCK_EXCL until iread_extents is completely...
authorDarrick J. Wong <djwong@kernel.org>
Wed, 12 Apr 2023 05:49:10 +0000 (15:49 +1000)
committerDave Chinner <dchinner@redhat.com>
Wed, 12 Apr 2023 05:49:10 +0000 (15:49 +1000)
commitc95356ca884885db702670e24933ee7f2b9f1754
tree554e9ec9c18210b9adbf1911cc41027254b5617c
parent4b827b3f305d1fcf837265f1e12acc22ee84327c
xfs: _{attr,data}_map_shared should take ILOCK_EXCL until iread_extents is completely done

While fuzzing the data fork extent count on a btree-format directory
with xfs/375, I observed the following (excerpted) splat:

XFS: Assertion failed: xfs_isilocked(ip, XFS_ILOCK_EXCL), file: fs/xfs/libxfs/xfs_bmap.c, line: 1208
------------[ cut here ]------------
WARNING: CPU: 0 PID: 43192 at fs/xfs/xfs_message.c:104 assfail+0x46/0x4a [xfs]
Call Trace:
 <TASK>
 xfs_iread_extents+0x1af/0x210 [xfs 09f66509ece4938760fac7de64732a0cbd3e39cd]
 xchk_dir_walk+0xb8/0x190 [xfs 09f66509ece4938760fac7de64732a0cbd3e39cd]
 xchk_parent_count_parent_dentries+0x41/0x80 [xfs 09f66509ece4938760fac7de64732a0cbd3e39cd]
 xchk_parent_validate+0x199/0x2e0 [xfs 09f66509ece4938760fac7de64732a0cbd3e39cd]
 xchk_parent+0xdf/0x130 [xfs 09f66509ece4938760fac7de64732a0cbd3e39cd]
 xfs_scrub_metadata+0x2b8/0x730 [xfs 09f66509ece4938760fac7de64732a0cbd3e39cd]
 xfs_scrubv_metadata+0x38b/0x4d0 [xfs 09f66509ece4938760fac7de64732a0cbd3e39cd]
 xfs_ioc_scrubv_metadata+0x111/0x160 [xfs 09f66509ece4938760fac7de64732a0cbd3e39cd]
 xfs_file_ioctl+0x367/0xf50 [xfs 09f66509ece4938760fac7de64732a0cbd3e39cd]
 __x64_sys_ioctl+0x82/0xa0
 do_syscall_64+0x2b/0x80
 entry_SYSCALL_64_after_hwframe+0x46/0xb0

The cause of this is a race condition in xfs_ilock_data_map_shared,
which performs an unlocked access to the data fork to guess which lock
mode it needs:

Thread 0                          Thread 1

xfs_need_iread_extents
<observe no iext tree>
xfs_ilock(..., ILOCK_EXCL)
xfs_iread_extents
<observe no iext tree>
<check ILOCK_EXCL>
<load bmbt extents into iext>
<notice iext size doesn't
 match nextents>
                                  xfs_need_iread_extents
                                  <observe iext tree>
                                  xfs_ilock(..., ILOCK_SHARED)
<tear down iext tree>
xfs_iunlock(..., ILOCK_EXCL)
                                  xfs_iread_extents
                                  <observe no iext tree>
                                  <check ILOCK_EXCL>
                                  *BOOM*

Fix this race by adding a flag to the xfs_ifork structure to indicate
that we have not yet read in the extent records and changing the
predicate to look at the flag state, not if_height.  The memory barrier
ensures that the flag will not be set until the very end of the
function.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
fs/xfs/libxfs/xfs_bmap.c
fs/xfs/libxfs/xfs_inode_fork.c
fs/xfs/libxfs/xfs_inode_fork.h