Bug 11805

Created attachment 18395 [details]
Output from dmesg

Looks like allocating a buffer in xlog_alloc_log() via xfs_buf_get_noaddr()
is failing and we are not checking the return value before trying to
lock the buffer.

That implies you have not enough memory available to mount the filesystem.
The only thing I can see that would have caused this was the call to
alloc_pages(GFP_KERNEL) failed.

As it is, this is not a regression in XFS - we've never checked the return
value to xfs_buf_get_noaddr(). That is easily fixed - I'll have a patch in
a few minutes.

That being said - your kernel reports itself as:

2.6.27maluta-05577-g0cfd810-dirty #10

which, IIRC, indicates that you've built it from a git tree that
contains local modifications (the -dirty bit). Have you modified the
kernel you are running and if so, could those modifications be responsible
for lack of memory in the machine?

Yes. I modified the kernel, as long as I no the patch didn't affect memory. In fact the patch affects DMI (used to fix a bug on Lguest) http://marc.info/?l=linux-kernel&m=122445958110927&w=2

Created attachment 18397 [details]
check for memory allocation failure in xlog_alloc_log

This patch should fix the mount ENOMEM problem. Can you try it?

Fixed. 
Thanks!

On Sunday, 26 of October 2008, Dave Chinner wrote:
> On Sat, Oct 25, 2008 at 10:06:44PM +0200, Rafael J. Wysocki wrote:
> > This message has been generated automatically as a part of a report
> > of recent regressions.
> > 
> > The following bug entry is on the current list of known regressions
> > from 2.6.27.  Please verify if it still should be listed and let me know
> > (either way).
> > 
> > 
> > Bug-Entry   : http://bugzilla.kernel.org/show_bug.cgi?id=11805
> > Subject             : mounting XFS produces a segfault
> > Submitter   : Tiago Maluta <maluta_tiago@yahoo.com.br>
> > Date                : 2008-10-21 18:00 (5 days old)
> 
> Ah - this was reported as a 2.6.26 -> 2.6.27 regression, not a
> .27->.28-rcX regression.
> 
> Even so, it's not obviously an XFS regression as the problem is
> that alloc_pages(GFP_KERNEL) is the new failure on .27. The fact
> that XFS never handled the allocation failure is not a new bug
> or regression - it has never caught failures during log
> allocation...
> 
> So really, if you want to look for a regression here, it is the
> change of behaviour in the VM leading to a memory allocation failure
> where it has never, ever previously failed...

Handled-By : Dave Chinner <dgc@sgi.com>
Patch : http://bugzilla.kernel.org/attachment.cgi?id=18397&action=view

Fixed by commit 8f330f5149ef41ff943b04d914406cc417f62784 .