zfs issue - disappearing data

[Mike Carlson]

We had a critical issue with a zfs server that exports shares via samba 
(3.5) last night

system info:
uname -a

    FreeBSD zfs-1.discdrive.bayphoto.com 9.1-RELEASE FreeBSD 9.1-RELEASE
    #0 r243825: Tue Dec  4 09:23:10 UTC 2012
    root at farrell.cse.buffalo.edu:/usr/obj/usr/src/sys/GENERIC  amd64

zpool history:

    History for 'data':
    2013-02-25.17:11:37 zpool create data raidz /dev/gpt/disk1.nop
    /dev/gpt/disk2.nop /dev/gpt/disk3.nop /dev/gpt/disk4.nop
    2013-02-25.17:11:41 zpool add data raidz /dev/gpt/disk5.nop
    /dev/gpt/disk6.nop /dev/gpt/disk7.nop /dev/gpt/disk8.nop
    2013-02-25.17:11:47 zpool add data raidz /dev/gpt/disk9.nop
    /dev/gpt/disk10.nop /dev/gpt/disk11.nop /dev/gpt/disk12.nop
    2013-02-25.17:11:53 zpool add data raidz /dev/gpt/disk13.nop
    /dev/gpt/disk14.nop /dev/gpt/disk15.nop /dev/gpt/disk16.nop
    2013-02-25.17:11:57 zpool add data raidz /dev/gpt/disk17.nop
    /dev/gpt/disk18.nop /dev/gpt/disk19.nop /dev/gpt/disk20.nop
    2013-02-25.17:12:02 zpool add data raidz /dev/gpt/disk21.nop
    /dev/gpt/disk22.nop /dev/gpt/disk23.nop /dev/gpt/disk24.nop
    2013-02-25.17:12:08 zpool add data spare /dev/gpt/disk25.nop
    /dev/gpt/disk26.nop
    2013-02-25.17:12:15 zpool add data log /dev/gpt/log.nop
    2013-02-25.17:12:19 zfs set checksum=fletcher4 data
    2013-02-25.17:12:22 zfs set compression=lzjb data
    2013-02-25.17:12:25 zfs set aclmode=passthrough data
    2013-02-25.17:12:30 zfs set aclinherit=passthrough data
    2013-02-25.17:13:25 zpool export data
    2013-02-25.17:15:33 zpool import -d /dev/gpt data
    2013-03-01.12:31:58 zpool add data cache /dev/gpt/cache.nop
    2013-03-15.12:22:22 zfs create data/XML_WORKFLOW
    2013-03-27.12:05:42 zfs create data/IMAGEQUIX
    2013-03-27.13:32:54 zfs create data/ROES_ORDERS
    2013-03-27.13:32:59 zfs create data/ROES_PRINTABLES
    2013-03-27.13:33:21 zfs destroy data/ROES_PRINTABLES
    2013-03-27.13:33:26 zfs create data/ROES_PRINTABLE

We had a file structure drop off:

/data/XML_WORKFLOW/XML_ORDERS/

around 5/2/2012 @ 17:00

In that directory, there were a few thousand directories (containing 
images and a couple metadata text/xml files)

What is odd, is doing a du -h in the parent XML_WORKFLOW directory, only 
reports ~150MB:

      # find . -type f |wc -l
    86
    # du -sh .
    130M .


however, df reports 1.5GB:

    # df -h .
    Filesystem Size Used Avail Capacity Mounted on
    data/XML_WORKFLOW 28T 1.5G 28T 0% /data/XML_WORKFLOW

zdb -d shows:

    # zdb -d data/XML_WORKFLOW
    Dataset data/XML_WORKFLOW [ZPL], ID 139, cr_txg 339633, 1.53G,
    212812 objects

Digging further into zdb, the path is missing for most of those objects:

    # zdb -ddddd data/XML_WORKFLOW 635248
    Dataset data/XML_WORKFLOW [ZPL], ID 139, cr_txg 339633, 1.53G,
    212812 objects, rootbp DVA[0]=<5:b274264000:2000>
    DVA[1]=<0:b4d81a8000:2000> [L0 DMU objset] fletcher4 lzjb LE
    contiguous unique double size=800L/200P birth=1202311L/1202311P
    fill=212812 cksum=16d24fb5aa:6c2e0aff6bc:129af90fe2eff:2612f938c5292b

    Object lvl iblk dblk dsize lsize %full type
    635248 1 16K 512 6.00K 512 100.00 ZFS plain file
    168 bonus System attributes
    dnode flags: USED_BYTES USERUSED_ACCOUNTED
    dnode maxblkid: 0
    path ???<object#635248>
    uid 11258
    gid 10513
    atime Thu May 2 17:31:26 2013
    mtime Thu May 2 17:31:26 2013
    ctime Thu May 2 17:31:26 2013
    crtime Thu May 2 17:13:58 2013
    gen 1197180
    mode 100600
    size 52
    parent 635247
    links 1
    pflags 40800000005
    Indirect blocks:
    0 L0 3:a9da05a000:2000 200L/200P F=1 B=1197391/1197391

    segment [0000000000000000, 0000000000000200) size 512

The application that writes to this volume runs on a windows client, so 
far, it has exhibited identical behavior across two zfs servers, but not 
on a generic windows server 2003 network share.

The question is, what is happening to the data. Is it a samba issue? Is 
it ZFS? I've enabled the samba full_audit module to track file 
deletions, so I should have more information on that side.

If anyone has seen similar behavior please let me know

Mike C
-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 6054 bytes
Desc: S/MIME Cryptographic Signature
URL: <http://lists.freebsd.org/pipermail/freebsd-fs/attachments/20130503/1fb4aa8a/attachment.bin>