Structure needs cleaning? (take #2)

[Justin Piszcz]

Hi RUMI,

1. Have you run memtest86 for a few passes?

2. Have you run a short+long test against the drive?
    smartctl -t short /dev/sda  # wait 10min
    smartctl -t long # wait until its done
    show smartctl -a /dev/sda output <- report this back

3. Is this the first time this has happened?

4. Is your system on a UPS?

5. How do you mount your XFS partition?
    a. Do you use any special parameters?

Justin.


On Wed, 2 Sep 2009, Eric Sandeen wrote:

> RUMI Szabolcs wrote:
>> Hi!
>>
>> Well, what could be the reason? I mean, there was no hardware failure,
>> no crash, no reboot, no errors in the disk's SMART error log, no nothing.
>> What I did was that I've extracted and deleted the rather huge OpenOffice
>> source tree several times (sometimes with overwriting) and finally it
>> ended up with these undeletable files and xfs errors. Is it considered
>> normal for xfs to get messed up like that under such load?
>
> No, not normal.
>
> It found the text "0xAAAA0000,6,0x" on disk in an area where it expected
> to find valid filesystem metadata.
>
> Corruption could come from anywhere - an xfs bug, some other bug, bad
> memory, bad cables, neon death rays from space, writing directly to the
> disk, who knows.  Awfully hard to track down a one-off occurrence like
> this, I'm afraid.
>
>> Thanks,
>> Sab
>>
>>
>>
>> The xfs_repair output and xfs_info output is included below:
>>
>> # xfs_repair -v /dev/sda10
>
> ...
>
>> Phase 6 - check inode connectivity...
>>         - resetting contents of realtime bitmap and summary inodes
>>         - traversing filesystem ...
>>         - agno = 0
>>         - agno = 1
>>         - agno = 2
>>         - agno = 3
>>         - agno = 4
>>         - agno = 5
>>         - agno = 6
>>         - agno = 7
>> leaf block 8388608 for directory inode 4051737 bad header
>> rebuilding directory inode 4051737
>> leaf block 8388608 for directory inode 4053318 bad header
>> rebuilding directory inode 4053318
>
> ...
>
> above is the problem, properly found & repaired.
>
> -Eric
>
> _______________________________________________
> xfs mailing list
> xfs at oss.sgi.com
> http://oss.sgi.com/mailman/listinfo/xfs
>