Re: XFS: Assertion failed: xfs_dir2_sf_lookup(args) == ENOENT, file:fs/xfs/xfs_dir2_sf.c, line: 358

[Dave Chinner]

On Thu, Jul 11, 2013 at 10:39:30PM -0400, Dave Jones wrote:
> Just saw this during boot after an unclean shutdown. It hung afterwards.
> 
> [   97.162665] XFS: Assertion failed: xfs_dir2_sf_lookup(args) == ENOENT, file: fs/xfs/xfs_dir2_sf.c, line: 358
....
> [   97.173730]  [<ffffffffa0076953>] xfs_dir2_sf_addname+0x43/0x760 [xfs]
> [   97.173743]  [<ffffffffa0067cfc>] xfs_dir_createname+0x15c/0x1b0 [xfs]
> [   97.173754]  [<ffffffffa002f2dc>] xfs_create+0x4cc/0x710 [xfs]
> [   97.173764]  [<ffffffffa00278ca>] xfs_vn_mknod+0x9a/0x1c0 [xfs]
> [   97.173773]  [<ffffffffa0027a03>] xfs_vn_create+0x13/0x20 [xfs]
> [   97.173776]  [<ffffffff811d100d>] vfs_create+0x9d/0x100
> [   97.173778]  [<ffffffff811d1995>] do_last+0x925/0xe00
> [   97.173780]  [<ffffffff811d1f2e>] path_openat+0xbe/0x6f0
> [   97.173783]  [<ffffffff8109e33f>] ? local_clock+0x3f/0x50
> [   97.173785]  [<ffffffff811e1b5f>] ? __alloc_fd+0xaf/0x200
> [   97.173787]  [<ffffffff811d2c3a>] do_filp_open+0x3a/0x90
> [   97.173789]  [<ffffffff811e1b5f>] ? __alloc_fd+0xaf/0x200
> [   97.173790]  [<ffffffff811c0ddb>] do_sys_open+0x10b/0x200
> [   97.173792]  [<ffffffff81010578>] ? syscall_trace_enter+0x18/0x290
> [   97.173794]  [<ffffffff811c0eee>] SyS_open+0x1e/0x20
> 
> This trace repeated a few times, then the same assertion was triggered from sys_renameat.

That's rather curious. What this means is that there is either an
EIO or EEXIST error being returned from xfs_dir2_sf_lookup() when a
we're about to add the new entry. There are two things here - EIO
can only be returned if a shutdown has occurred - are there any
signs of a shutdown in the logs? If there is a shutdown in progress,
then this is just unlucky to shutdown with an inode in an
inconsistent state in memory that triggers this validity check
failure.

And EEXIST means that the initial lookup of the name during the open
failed to find the entry we are now trying to create. i.e. the
initial path walk failed to do the correct lookup on the directory,
and so never got down to xfs_dir2_sf_lookup() to find the directory
entry (perhaps a problem with a cached negative dentry?). Hence it
was decided during the open(O_CREATE) call that the directory entry
needed to be created, we get down to XFS to create it, and then get
EEXIST because the name already exists...

So, it's not clear what has caused this yet. Is it reproducable? If
would be good to get a trace of lookup vs addname events from XFS,
too (i.e. all the xfs_dir* and xfs_da* events) so we can see if the
correct lookups were done prior to the failing addname operation...

Cheers,

Dave.
-- 
Dave Chinner
david@xxxxxxxxxxxxx
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/