From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail.jrs-s.net ([173.230.137.22]:57507 "EHLO mail.jrs-s.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751246AbaAMTDx (ORCPT ); Mon, 13 Jan 2014 14:03:53 -0500 Message-ID: <52D43898.5080801@jrs-s.net> Date: Mon, 13 Jan 2014 14:03:52 -0500 From: Jim Salter MIME-Version: 1.0 To: dsterba@suse.cz, linux-btrfs , jbacik@fb.com Subject: Re: btrfs send: page allocation failure References: <52D3E308.7060309@jrs-s.net> <20140113182312.GR6498@twin.jikos.cz> <52D4326B.6040009@jrs-s.net> <20140113185636.GU6498@suse.cz> In-Reply-To: <20140113185636.GU6498@suse.cz> Content-Type: text/plain; charset=ISO-8859-1; format=flowed Sender: linux-btrfs-owner@vger.kernel.org List-ID: OK, thanks. If kernel memory fragmentation is a big factor, that would also explain why it succeeds after a reboot but does not succeed after weeks of uptime... On 01/13/2014 01:56 PM, David Sterba wrote: > On Mon, Jan 13, 2014 at 01:37:31PM -0500, Jim Salter wrote: >> What makes you believe that? The bug filed there appears to be related to >> defragging, which I am not doing either manually or automatically. > The quota groups are on and the symptoms match the known problem when > there are lots of backrefs. > krealloc is on the stack although not mentioned explicitly, called > through ulist_add_merge. The allocation order is 6 from the OOM report, > means that somebody wanted a large congiguous chunk of memory, which is > what the krealloc did and failed. Kernel memory is fragmented and this > kind of allocations is hard to satisfy in the long run. > > david