On Mon, Feb 24, 2003 at 05:58:27PM -0800, LA Walsh wrote:
> Redid the mess w/debug and an 'strace' dumping off to a synchronous NFS
> mounted
> partition while "sync" was executing in 'forloop' on 2ndary console.
>
> this time instead of doing dismount on my root FS, it only dumped me into
> KDB...
> such a deal!...
Hmmm -- that's the idea with debugging enabled.
> too bad I don't have serial console setup....did a continue
> after it died though, and did get more output....
>
> strace output is attached, ishtar (system under te^h^hstress) syslog says:
>
> Feb 24 17:37:05 ishtar su: (to backup) root on /dev/tty1
> Feb 24 17:37:05 ishtar kernel: XFS assertion failed:
> INT_GET(agi->agi_unlinked[bucket_index], ARCH_CONVERT) != NULLAGINO, file:
> xfs_inode.c, line: 1937
OK, so this is in a completely different part of the code to
where we were looking before... ugh, not useful. I think we
will have to get you to run repair on this, so we can start
at least with no known corruption problems. (to run repair
on your root you need to boot off a CD, floppy, or some other
alternate root).
To answer the question Chris was asking, for this case at least,
yes xfsdump is doing a bulkstat at this point...
> ...
> Feb 24 17:37:05 ishtar kernel: [xfs_bulkstat_one+833/1584]
> xfs_bulkstat_one+0x341/0x630 [kernel]
> Feb 24 17:37:05 ishtar kernel: [<c021be31>] xfs_bulkstat_one+0x341/0x630
> [kernel]
> Feb 24 17:37:05 ishtar kernel: [xfs_bulkstat_single+85/320]
> xfs_bulkstat_single+0x55/0x140 [kernel]
> Feb 24 17:37:05 ishtar kernel: [<c021cd35>] xfs_bulkstat_single+0x55/0x140
> [kernel]
> Feb 24 17:37:05 ishtar kernel: [xfs_ioc_bulkstat+450/528]
> xfs_ioc_bulkstat+0x1c2/0x210 [kernel]
> Feb 24 17:37:05 ishtar kernel: [<c024d422>] xfs_ioc_bulkstat+0x1c2/0x210
> [kernel]
> Feb 24 17:37:05 ishtar kernel: [xfs_ioctl+1593/2016] xfs_ioctl+0x639/0x7e0
> [kernel]
> Feb 24 17:37:05 ishtar kernel: [<c024cfb9>] xfs_ioctl+0x639/0x7e0 [kernel]
> ...
>
> ----
> I take it that this isn't normal, expected, behavior in a modern file
> system? :-)
>
With debug enabled, basically all bets are off and we're actively
trying to crash the system - since your fs had known corruption,
this was always going to happen. Guess I should have spelled that
out a bit more... ;)
cheers.
--
Nathan
|