From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from ipmail05.adl6.internode.on.net ([150.101.137.143]:26674 "EHLO ipmail05.adl6.internode.on.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750938AbcKHGPz (ORCPT ); Tue, 8 Nov 2016 01:15:55 -0500 Date: Tue, 8 Nov 2016 17:15:39 +1100 From: Dave Chinner Subject: Re: [PATCH 1/3] xfs: fix the AG loop in xfs_alloc_vextent for busy extents Message-ID: <20161108061539.GB28922@dastard> References: <1476735753-5861-1-git-send-email-hch@lst.de> <1476735753-5861-2-git-send-email-hch@lst.de> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <1476735753-5861-2-git-send-email-hch@lst.de> Sender: linux-xfs-owner@vger.kernel.org List-ID: List-Id: xfs To: Christoph Hellwig Cc: linux-xfs@vger.kernel.org, michaelcallahan@fb.com On Mon, Oct 17, 2016 at 10:22:31PM +0200, Christoph Hellwig wrote: > If we have too many busy extents, or just enough to make our wanted > allocation impossible we will never move on to another AG in > xfs_alloc_vextent currently. Change the loop exit condition to keep > looking for an AG if we can't allocate an extent. > > For the single AG cases we don't need to change anything as the higher > layers need to handle this case. > > Signed-off-by: Christoph Hellwig > --- > fs/xfs/libxfs/xfs_alloc.c | 11 ++++++++--- > 1 file changed, 8 insertions(+), 3 deletions(-) > > diff --git a/fs/xfs/libxfs/xfs_alloc.c b/fs/xfs/libxfs/xfs_alloc.c > index effb64c..0a04bec 100644 > --- a/fs/xfs/libxfs/xfs_alloc.c > +++ b/fs/xfs/libxfs/xfs_alloc.c > @@ -2753,13 +2753,18 @@ xfs_alloc_vextent( > trace_xfs_alloc_vextent_nofix(args); > goto error0; > } > + > /* > - * If we get a buffer back then the allocation will fly. > + * If we get a buffer back then the allocation will fly, > + * unless there are busy extents in the way. > */ > if (args->agbp) { > - if ((error = xfs_alloc_ag_vextent(args))) > + error = xfs_alloc_ag_vextent(args); > + if (error) > goto error0; > - break; > + if (args->agbno != NULLAGBLOCK) > + break; > + xfs_trans_brelse(args->tp, args->agbp); > } > > trace_xfs_alloc_vextent_loopfailed(args); Here's the problem: At ENOSPC, we dirty AG headers fixing freelists, then fail to allocate inodes, resulting in a shutdown - a regression that xfs/076 trips over quite regularly on my test machines. XFS (pmem1): Internal error xfs_trans_cancel at line 983 of file fs/xfs/xfs_trans.c. Caller xfs_create+0x498/0x780 CPU: 3 PID: 19236 Comm: touch Not tainted 4.9.0-rc4-dgc+ #1019 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Debian-1.8.2-1 04/01/2014 ffffc900144dbb28 ffffffff81822480 ffff880239580500 ffff88023abd8000 ffffc900144dbb40 ffffffff8152340c ffffffff81534708 ffffc900144dbb80 ffffffff8154570b 0000000100000000 ffff88032e9ff800 ffff88023abd8000 Call Trace: [] dump_stack+0x63/0x83 [] xfs_error_report+0x3c/0x40 [] ? xfs_create+0x498/0x780 [] xfs_trans_cancel+0x12b/0x150 [] xfs_create+0x498/0x780 [] xfs_generic_create+0x1df/0x2c0 [] xfs_vn_create+0x16/0x20 [] path_openat+0x1312/0x13e0 [] ? unlock_page+0x36/0x40 [] do_filp_open+0x7e/0xe0 [] ? getname_flags+0x4f/0x1f0 [] ? kmem_cache_alloc+0x42/0x180 [] ? _raw_spin_unlock+0x10/0x30 [] ? __alloc_fd+0xb2/0x160 [] do_sys_open+0x123/0x200 [] SyS_open+0x1e/0x20 [] entry_SYSCALL_64_fastpath+0x1a/0xa9 You're going to have to rethink this one, Christoph. Cheers, Dave. > -- > 2.1.4 > > -- > To unsubscribe from this list: send the line "unsubscribe linux-xfs" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html > -- Dave Chinner david@fromorbit.com