Messages in this thread | | | Date | Tue, 12 Jun 2012 19:46:18 -0400 | From | Mark Lord <> | Subject | Re: mount.nfs: cannot allocate memory |
| |
On 12-06-12 07:28 PM, Myklebust, Trond wrote: > On Tue, 2012-06-12 at 19:08 -0400, Mark Lord wrote: >> On 12-06-12 06:58 PM, Mark Lord wrote: >>> Adding Ben Hutchings to CC: as he seems to have looked into this before. >>> >>>> On 12-06-12 06:44 PM, Mark Lord wrote: >>>>> I've been seeing these messages on my AMD Fusion server >>>>> running linux-3.3.7-64bit. Does this ring any bells for anyone else? >>>>> The system is NOT low on memory. >>>>> >>>>> I'm building/installing 3.4.2 now to see if it behaves any better. >>>>> >>>>> [962841.265658] mount.nfs: page allocation failure: order:4, mode:0xc0d0 >>>>> [962841.265674] Pid: 32116, comm: mount.nfs Not tainted 3.3.7 #2 >>>>> [962841.265680] Call Trace: >>>>> [962841.265700] [<ffffffff81079363>] ? warn_alloc_failed+0x11a/0x12d >>>>> [962841.265713] [<ffffffff8107b904>] ? __alloc_pages_nodemask+0x6c0/0x702 >>>>> [962841.265725] [<ffffffff8114fcb8>] ? timerqueue_del+0x53/0x63 >>>>> [962841.265758] [<ffffffff8107b9ba>] ? __get_free_pages+0x10/0x3f >>>>> [962841.265805] [<ffffffffa01ea33d>] ? nfs_idmap_new+0x28/0xde [nfs] >>>>> [962841.265836] [<ffffffffa01c79c9>] ? nfs4_init_client+0x74/0x12a [nfs] >> ... >>>> Looks like an old bug, perhaps: >>>> https://bugzilla.redhat.com/show_bug.cgi?id=593035 >>>> https://bugzilla.redhat.com/show_bug.cgi?id=728003 >>>> >>>> I wonder what the underlying cause is? >>>> And I also wonder if the NFS code should be more clever >>>> about handing order-4 allocation failures. >>> >>> More references: >>> http://www.mail-archive.com/debian-kernel@lists.debian.org/msg71281.html >>> https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1004619 >>> http://codemonkey.org.uk/2012/02/17/fedora-16-kernel-bugzilla-status-report-20120210-20120217/ >>> https://lkml.org/lkml/2011/2/24/73 >> >> Ahh.. this one seems to explain it, kind of: >> http://permalink.gmane.org/gmane.linux.nfs/41686 >> >> So in 3.3.7, I didn't have CONFIG_NFS_USE_NEW_IDMAPPER enabled, >> and the kernel seems to be buggy without that. >> >> In 3.4.2, that option has disappeared entirely, >> most likely because it is now the default behaviour. Right? > > Yes, but we also added some patches to Linux 3.4 in order to fix this > bug. > > Please see commit d073e9b541e1ac3f52d72c3a153855d9a9ee3278 (NFSv4: > Reduce the footprint of the idmapper) and commit > 685f50f9188ac1e8244d0340a9d6ea36b6136cec (NFSv4: Further reduce the > footprint of the idmapper)
Ah, okay. Good work, guye.
Cheers
| |