Snapshot corruption on 6.1/amd64

[James Lauser]

Thanks a bunch for the suggestion.  I upgraded overnight, and now all  
seems to be working.


--  James L. Lauser
     james at jlauser.net

+--------------+
| +--------+   +-+
| | +   \  | | | |
| |   -  | | | | |
| | +   /  |   | |
| +--------+ + +-+
+--------------+


On Sep 22, 2006, at 0:15 , Kostik Belousov wrote:

> On Thu, Sep 21, 2006 at 03:43:03PM -0500, Eric Anderson wrote:
>> On 09/21/06 14:59, James Lauser wrote:
>>> Hello.
>>>
>>> I've been having some trouble with snapshots on my FreeBSD 6.1/amd64
>>> system.  Basically, I have this system set up with a 3ware RAID card
>>> and several disks, and use it to collect backups from my other
>>> FreeBSD server (sparc64) and three Macs via rsync..
>>>
>>> Every night, I have a script generate a snapshot of the RAID's file
>>> system, and those snapshots are kept on the system for one week
>>> before being removed (i.e. there are always 7 snapshots present on
>>> the system), so I can recover files that were accidentally  
>>> removed or
>>> changed.
>>>
>>> The problem is that when a large number of files are removed or
>>> changed on the file system, the corresponding files in the snapshot
>>> get corrupted.  This, obviously, makes the snapshots quite useless.
>>>
>>> After some searching, I've found a bug report filed last year that
>>> describes this problem exactly, though the log of that report does
>>> not suggest that anything has been done with it.  That report is at
>>> http://www.freebsd.org/cgi/query-pr.cgi?pr=kern/90512
>>>
>>> Any help would be greatly appreciated.  A test I ran showing the
>>> problem, plus the output of uname -a and dmesg.boot is attached.
>>>
>>> Thanks in advance.
>>
>>
>>> Sledge# cd /raid
>>> Sledge# touch foo
>>> Sledge# ls -l foo
>>> -rw-r--r--  1 root  wheel  0 Sep 18 14:07 foo
>>> Sledge# mksnap_ffs /raid /raid/.snap/snap
>>> Sledge# rm foo
>>> Sledge# mdconfig -a -t vnode -f /raid/.snap/snap -u 4
>>> WARNING: opening backing store: /raid/.snap/snap readonly
>>> Sledge# mount -r /dev/md4 /mnt
>>> Sledge# cd /mnt
>>> Sledge# ls -l foo
>>> ls: foo: Bad file descriptor
>>> Sledge#
>>> Sledge#
>>> Sledge#
>>> Sledge# uname -a
>>> FreeBSD Sledge.jlauser.net 6.1-RELEASE-p6 FreeBSD 6.1-RELEASE-p6  
>>> #4: Wed
>>> Sep  6 23:30:56 EDT 2006
>>> root at Sledge.jlauser.net:/usr/obj/usr/src/sys/GENERIC  amd64
>>
>> Hmm.. Interesting, it seems to work fine for me:
>>
>> [root at neutrino /tmp]# dd if=/dev/zero of=TESTDISK bs=1m count=100
>> 100+0 records in
>> 100+0 records out
>> 104857600 bytes transferred in 3.390829 secs (30923882 bytes/sec)
>> [root at neutrino /tmp]# mdconfig -a -t vnode -f ./TESTDISK
>> md0
>> [root at neutrino /tmp]# newfs -U /dev/md0
>> /dev/md0: 100.0MB (204800 sectors) block size 16384, fragment size  
>> 2048
>>         using 4 cylinder groups of 25.02MB, 1601 blks, 3264 inodes.
>>         with soft updates
>> super-block backups (for fsck -b #) at:
>>  160, 51392, 102624, 153856
>> [root at neutrino /tmp]# mount /dev/md0 /mnt
>> [root at neutrino /tmp]# touch /mnt/foo
>> [root at neutrino /tmp]# mksnap_ffs /mnt/ /mnt/.snap/snap
>> [root at neutrino /tmp]# mdconfig -a -t vnode -f /mnt/.snap/snap
>> WARNING: opening backing store: /mnt/.snap/snap readonly
>> md1
>> [root at neutrino /tmp]# mount -r /dev/md1 /mnt2
>> [root at neutrino /tmp]# cd /mnt2
>> [root at neutrino /mnt2]# ls -l foo
>> -rw-r--r--  1 root  wheel  0 Sep 21 15:37 foo
>> [root at neutrino /mnt2]#
> I think you have i386 system ?
>
> James, look at the PR/100365. Supposed fix is MFCed. Original reporter
> said that this changed nothing for him. I have not much time lately to
> look at this problem, but would like to get additional data points.
>
> BTW, use of snapshots with stock 6.1 is not very attractive idea,  
> better
> to update to the 6-STABLE (many important fixes in that area were  
> made).