On Sat, Mar 03, 2012 at 12:37:07PM +0100, Rafa?? Kupka wrote:
> On Fri, 2012-03-02 at 05:43 -0500, Christoph Hellwig wrote:
> > From the trace this looks like we got incorrect results from xfs_imap,
> > which got some fairly significant changes between the two kernels
> > you tested. Can you check if 2.6.36 shows this issue? It's the first
> > kernel that had these big changes in inode lookup.
> I've tested more kernels and here are results (linus git tree, no
> additional patches):
> - v2.6.36 -- good
> - v2.6.38 -- good
> - v2.6.39 -- bad, I/O error log entry and locked processes:
> - git revision 243b422af9ea9af4ead07a8ad54c90d4f9b6081a -- good
> It's one patch before "xfs: preallocation transactions do not need to be
> - git revision c5850150d0b9ae16840c5d9846f9d5a759996a15:
> Merge branch 'for-linus' of git://oss.sgi.com/xfs/xfs
> * 'for-linus' of git://oss.sgi.com/xfs/xfs:
> xfs: stop using the page cache to back the buffer cache
> xfs: register the inode cache shrinker before quotachecks
> xfs: xfs_trans_read_buf() should return an error on failure
This is the only patch in that area. Give me a bit more time to
figure out how it could related to your issues.
> Hung xfsdump (only?) without any additional log entries
> (CONFIG_XFS_DEBUG is set):
> I don't know if it's the same bug - there is no assertion nor I/O error
> log entry...
It's the same code path at least..