Re: XFS crashed twice, once in 2.6.16.20, next in 2.6.17, reproducable

[Nathan Scott <nathans@xxxxxxx>]

On Sat, Jul 01, 2006 at 04:02:35PM +0200, Adam Sj?gren wrote:
> On Thu, 22 Jun 2006 12:56:40 +1000, Nathan wrote:
> The subsequent xfs_repair run gave different, expected I guess, output
> (it noticed the modified directory inode and stuff), but at the end of
> Phase 7, it said:
> 
>  cache_purge: share on cache 0x100930b0 left 1 nodes!?
>  cache_purge: share on cache 0x100930b0 left 1 nodes!?
>  done

Thats some diagnostic stuff for us developers.  Don't worry about
it, its just telling us we have inode refcounting issues which we
will have to resolve for a parallel xfs_repair.

> ...
> I proceeded to run xfs_repair. In Phase 3 under '- agno 0' it said:
> 
>  bad dir magic number 0x30 in inode 180 bno = 8388608
> 
> It continued and in Phase 6 after saying '- traversing filesystem
> starting at / ...' it said:
> 
>  rebuilding directory inode 128

Thats your root directory.  Its being rebuilt due to a lost+found
directory being earlier unlinked, most likely.

> And a flurry of "disconnected inode [number], moving to lost+found"
> followed.

Yeah, anything that was in lost+found after the first run, gets
put back there each time.

> Any hints on how to fully repair the filesystem, in place?

It sounds like your filesystem is repaired at this point - you
need to fix up (remove) lost+found, and you should get no issues
reported thereafter.

cheers.

-- 
Nathan