ZFS: Panic when attempting to delete certain data

Andriy Gapon avg at FreeBSD.org
Wed Nov 28 13:50:13 UTC 2012


on 28/11/2012 15:41 Josh Beard said the following:
> 
> 
> On Wed, Nov 28, 2012 at 1:47 AM, Andriy Gapon <avg at freebsd.org
> <mailto:avg at freebsd.org>> wrote:
> 
>     on 27/11/2012 23:47 Josh Beard said the following:
>     > Thanks!  Here we go:
>     >
>     > (kgdb) frame 7
>     > #7  0xffffffff80ebd45a in zfs_freebsd_remove (ap=Variable "ap" is not available.
>     > ) at
>     >
>     /usr/src/sys/modules/zfs/../../cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_vnops.c:1855
>     > 1855                    dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
>     > (kgdb) list
>     > 1850                &xattr_obj, sizeof (xattr_obj));
>     > 1851            if (error == 0 && xattr_obj) {
>     > 1852                    error = zfs_zget(zfsvfs, xattr_obj, &xzp);
>     > 1853                    ASSERT3U(error, ==, 0);
>     > 1854                    dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
>     > 1855                    dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
>     > 1856            }
>     > 1857
>     > 1858            mutex_enter(&zp->z_lock);
>     > 1859            if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
> 
>     That's what I suspected.
> 
>     > (kgdb) frame 7
>     > #7  0xffffffff80ebd45a in zfs_freebsd_remove (ap=Variable "ap" is not available.
>     > ) at
>     >
>     /usr/src/sys/modules/zfs/../../cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_vnops.c:1855
>     > 1855                    dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
>     > (kgdb) info local
>     > No locals.
>     > (kgdb)
> 
>     A little bit unfortunate.
> 
>     > # ls -Pi /DSDK12_NHR.pax.gz  (symlink to ../Archive.pax.gz)
>     > 249868
>     >
>     ./Imaging/Packages/DSDK12_NHR_2012-02-23.pkg/Contents/Resources/DSDK12_NHR.pax.gz
>     >
>     > # zdb -ddddd store/tdxs1 249868
>     > Dataset store/tdxs1 [ZPL], ID 109, cr_txg 35014, 1.33T, 1106389 objects, rootbp
>     > DVA[0]=<0:8000204400:400> DVA[1]=<0:30800644400:400> [L0 DMU objset] fletcher4
>     > lzjb LE contiguous unique double size=800L/200P birth=1167710L/1167710P
>     > fill=1106389 cksum=1966704b59:757ae6cb615:134bfd597bca9:254b2ee348393d
>     >
>     >     Object  lvl   iblk   dblk  dsize  lsize   %full  type
>     >     249868    1    16K    512      0    512    0.00  ZFS plain file
>     >                                         201   bonus  System attributes
>     >         dnode flags: USERUSED_ACCOUNTED
>     >         dnode maxblkid: 0
>     >         path
>     >
>      /tech/2012-09-14-01-00/Imaging/Packages/DSDK12_NHR_2012-02-23.pkg/Contents/Resources/DSDK12_NHR.pax.gz
>     >         uid     300002
>     >         gid     80
>     >         atime   Tue Nov 27 14:43:00 2012
>     >         mtime   Thu Feb 23 08:59:21 2012
>     >         ctime   Fri Sep 14 01:12:37 2012
>     >         crtime  Fri Sep 14 01:11:50 2012
>     >         gen     81430
>     >         mode    120755
>     >         size    17
>     >         parent  249866
>     >         links   1
>     >         pflags  40800000104
>     >         xattr   230
> 
> 
> # zdb -ddddd store/tdxs1 230
> Dataset store/tdxs1 [ZPL], ID 109, cr_txg 35014, 1.33T, 1106389 objects, rootbp
> DVA[0]=<0:8000284800:400> DVA[1]=<0:308006b3800:400> [L0 DMU objset] fletcher4
> lzjb LE contiguous unique double size=800L/200P birth=1167748L/1167748P
> fill=1106389 cksum=16e1e08cb8:70a50f1ec5a:13419c71d1cda:260fbed28af8f5
> 
>     Object  lvl   iblk   dblk  dsize  lsize   %full  type
> zdb: dmu_bonus_hold(230) failed, errno 2
>  

errno 2 is ENOENT, so all pieces match.
The file appears to have extended attributes, but they do not actually exist.
If your ZFS module was built with debug, then ASSERT3U(error, ==, 0) would be
triggered.

This is something that I said earlier in the private conversation that I mentioned:

Andriy Gapon said:
> The relevant commits are r240632 and r240345.  I can't recall when I MFC-ed them
> to stable/9.  Most likely they are not in 9.1 releng branch.
[snip]
> I also don't have a good advice on how to fix the existing corruption.
> I'd probably go with using something like tar/cpio/pax to move data to fresh
> filesystem.  Please be sure to use latest stable/9 or 8 to not run into the
> issue again.
>
> zfs send/recv won't help, it would mindlessly replicate the corrupted attributes.

To this I might add that the bugs were not FreeBSD-specifc and they may be still
present in ZFS upstream and other ZFS ports.

Thank you for the debugging information.
-- 
Andriy Gapon


More information about the freebsd-fs mailing list