From: Jan Kara Subject: Re: ext4_da_release_space:1333: ext4_da_release_space: ino 12, to_free 1 with only 0 reserved data blocks Date: Wed, 31 Jul 2013 20:32:00 +0200 Message-ID: <20130731183200.GA28018@quack.suse.cz> References: <20130716202533.GA16061@redhat.com> <20130717125322.GA29294@quack.suse.cz> <20130717145218.GA27731@redhat.com> <20130717213850.GA5025@quack.suse.cz> <20130731163901.GA10973@redhat.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Jan Kara , Linux Kernel , linux-ext4@vger.kernel.org To: Dave Jones Return-path: Content-Disposition: inline In-Reply-To: <20130731163901.GA10973@redhat.com> Sender: linux-kernel-owner@vger.kernel.org List-Id: linux-ext4.vger.kernel.org On Wed 31-07-13 12:39:01, Dave Jones wrote: > On Wed, Jul 17, 2013 at 11:38:50PM +0200, Jan Kara wrote: > > On Wed 17-07-13 10:52:18, Dave Jones wrote: > > > On Wed, Jul 17, 2013 at 02:53:22PM +0200, Jan Kara wrote: > > > > On Tue 16-07-13 16:25:33, Dave Jones wrote: > > > > > I've seen this happen a few times this week.. > > > > Thanks for report! Was this when fuzzing or just normal desktop load? > > > > What is inode with inode number 12 on your filesystem sdb1? What IO happens > > > > to it? Apparently some delalloc accounting went wrong somewhere and it's > > > > searching for a needle in a haystack unless we have more details... > > > > > > It was running this.. https://github.com/kernelslacker/io-tests/blob/master/setup.sh > > > in a loop. After about 6 hours, that fell out. It made it all the way through > > > every test a few times, which is odd, as the test should be fairly deterministic. > > > Ah, I wasn't capturing the fsx seed. I'll do that on the next run. > > So inode 12 was likely the file used by fsx. OK. Looking at the script > > link, fsx is run as: > > /usr/local/bin/fsx -N 250000 -S0 foo & > > so the seed is always 0. So it is a deterministic test and there must be > > some race with writeback or something that is rarely triggered. Drat. > > > > Umm, looking at the filesystems this tests, ext4 with 1 KB blocksize is > > likely the config which hits this (the accounting is more complex there) so > > it might be interesting to concentrace on this one. > > I just retried these tests, and hit this again. > I can confirm it happens on the 1k block filesystem. Great, thanks for confirmation. I've just managed to reproduce the problem yesterday so I'm debugging it... Honza -- Jan Kara SUSE Labs, CR