XFS internal error when making hard link on full fs.

XFS internal error when making hard link on full fs.

[Michael Morrison]

Got the following when I tried to make a hard link on a full filesystem:
I'm running Linux kernel 2.6.18.  I'm  unable to try a newer kernel at
the present time.
The application was properly given ENOSPC in errno when the link call
failed.

The filesystem is built on a linux software raid0 with 8 drives using
the following command lines:

/sbin/mdadm --create /dev/md0 --level=0 \
    --chunk=64   \
    --raid-devices=8  \
    /dev/sda /dev/sde \
    /dev/sdb /dev/sdf \
    /dev/sdc /dev/sdg \
    /dev/sdd /dev/sdh

/sbin/mkfs -t xfs -f -d unwritten=0 -l version=2,sunit=128,internal 
/dev/md0
/bin/mount -t xfs /dev/md0 /mnt/dmag -o noatime,nodiratime,logbufs=4

Filesystem "md0": XFS internal error xfs_trans_cancel at line 1138 of
file fs/xfs/xfs_trans.c.  Caller 0xc02d53a5
 [<c02cb987>] xfs_trans_cancel+0x108/0x14f
 [<c02d53a5>] xfs_link+0x40f/0x585
 [<c02d53a5>] xfs_link+0x40f/0x585
 [<c04937ab>] _spin_unlock+0xd/0x21
 [<c02e19d5>] xfs_vn_link+0x64/0xd3
 [<c0181b66>] mntput_no_expire+0x1c/0x75
 [<c017de95>] __d_lookup+0x8f/0x13b
 [<c016e2b2>] vfs_stat+0x1f/0x23
 [<c0172788>] cached_lookup+0x23/0x85
 [<c01725de>] permission+0x85/0xaa
 [<c0175e24>] vfs_link+0xc7/0x183
 [<c0176008>] sys_linkat+0x128/0x14a
 [<c0176059>] sys_link+0x2f/0x33
 [<c0102ecd>] sysenter_past_esp+0x56/0x79
xfs_force_shutdown(md0,0x8) called from line 1139 of file
fs/xfs/xfs_trans.c.  Return address = 0xc02cb9ad
Filesystem "md0": Corruption of in-memory data detected.  Shutting down
filesystem: md0
Please umount the filesystem, and rectify the problem(s)


Unmounting the fs and running xfs_check did not produce any output.  The
filesystem seems happy after mounting it again.

Re: XFS internal error when making hard link on full fs.

[David Chinner]

On Thu, Jul 19, 2007 at 08:00:33PM -0700, Michael Morrison wrote:
> Got the following when I tried to make a hard link on a full filesystem:
> I'm running Linux kernel 2.6.18.  I'm  unable to try a newer kernel at
> the present time.
> The application was properly given ENOSPC in errno when the link call
> failed.

*nod*

That problem. If only I could reproduce it....

> Filesystem "md0": XFS internal error xfs_trans_cancel at line 1138 of
> file fs/xfs/xfs_trans.c.  Caller 0xc02d53a5
>  [<c02cb987>] xfs_trans_cancel+0x108/0x14f
>  [<c02d53a5>] xfs_link+0x40f/0x585
>  [<c02d53a5>] xfs_link+0x40f/0x585
>  [<c04937ab>] _spin_unlock+0xd/0x21
>  [<c02e19d5>] xfs_vn_link+0x64/0xd3
>  [<c0181b66>] mntput_no_expire+0x1c/0x75
>  [<c017de95>] __d_lookup+0x8f/0x13b
>  [<c016e2b2>] vfs_stat+0x1f/0x23
>  [<c0172788>] cached_lookup+0x23/0x85
>  [<c01725de>] permission+0x85/0xaa
>  [<c0175e24>] vfs_link+0xc7/0x183
>  [<c0176008>] sys_linkat+0x128/0x14a
>  [<c0176059>] sys_link+0x2f/0x33
>  [<c0102ecd>] sysenter_past_esp+0x56/0x79
> xfs_force_shutdown(md0,0x8) called from line 1139 of file
> fs/xfs/xfs_trans.c.  Return address = 0xc02cb9ad
> Filesystem "md0": Corruption of in-memory data detected.  Shutting down
> filesystem: md0
> Please umount the filesystem, and rectify the problem(s)

Yeah, it shut down due to cancelling a dirty transaction. Basically,
we've seen that we are at ENOSPC and tried to do a link without a
space reservation. We call xfs_dir_canenter() to determine if this
is possible or not. If it is possible, we then call xfs_dir_createname()
to create the entry. At that point, if we try to do an allocation we'll
get ENOSPC and the transaction will be dirty. It then gets cancelled and
we shutdown.

The problem is that for some reason we are needing to do an allocation
that the check function is not picking up. xfs_create() has a similar
problem and I haven't been able to get to the bottom of the problem
yet as no-one can reproduce this easily.

> Unmounting the fs and running xfs_check did not produce any output.  The
> filesystem seems happy after mounting it again.

That's normal - it's an in-core error and not something on disk.

Cheers,

Dave.
-- 
Dave Chinner
Principal Engineer
SGI Australian Software Group