linux-ext4 - Re: Excessive stall times on ext4 in 3.9-rc2

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20130412094532.GH11656@suse.de>
Date:	Fri, 12 Apr 2013 10:45:32 +0100
From:	Mel Gorman <mgorman@...e.de>
To:	Theodore Ts'o <tytso@....edu>, linux-ext4@...r.kernel.org,
	LKML <linux-kernel@...r.kernel.org>,
	Linux-MM <linux-mm@...ck.org>, Jiri Slaby <jslaby@...e.cz>
Subject: Re: Excessive stall times on ext4 in 3.9-rc2

On Thu, Apr 11, 2013 at 02:35:12PM -0400, Theodore Ts'o wrote:
> On Thu, Apr 11, 2013 at 06:04:02PM +0100, Mel Gorman wrote:
> > > If we're stalling on lock_buffer(), that implies that buffer was being
> > > written, and for some reason it was taking a very long time to
> > > complete.
> > > 
> > 
> > Yes.
> > 
> > > It might be worthwhile to put a timestamp in struct dm_crypt_io, and
> > > record the time when a particular I/O encryption/decryption is getting
> > > queued to the kcryptd workqueues, and when they finally squirt out.
> > > 
> > 
> > That somewhat assumes that dm_crypt was at fault which is not unreasonable
> > but I was skeptical as the workload on dm_crypt was opening a maildir
> > and mostly reads.
> 
> Hmm... well, I've reviewed all of the places in the ext4 and jbd2
> layer where we call lock_buffer(), and with one exception[1] we're not
> holding the the bh locked any longer than necessary.  There are a few
> places where we grab a spinlock or two before we can do what we need
> to do and then release the lock'ed buffer head, but the only time we
> hold the bh locked for long periods of time is when we submit metadata
> blocks for I/O.
> 

Yeah, ok. This is not the answer I was hoping for but it's the answer I
expected.

> Could you code which checks the hold time of lock_buffer(), measuing
> from when the lock is successfully grabbed, to see if you can see if I
> missed some code path in ext4 or jbd2 where the bh is locked and then
> there is some call to some function which needs to block for some
> random reason?
>
> What I'd suggest is putting a timestamp in buffer_head
> structure, which is set by lock_buffer once it is successfully grabbed
> the lock, and then in unlock_buffer(), if it is held for more than a
> second or some such, to dump out the stack trace.
> 

I can do that but the results might lack meaning. What I could do instead
is use a variation of the page owner tracking patch (current iteration at
https://lkml.org/lkml/2012/12/7/487) to record a stack trace in lock_buffer
and dump it from jbd2/transaction.c if it stalls for too long. I'll report
if I find something useful.

> Because at this point, either I'm missing something or I'm beginning
> to suspect that your hard drive (or maybe something the block layer?)
> is simply taking a long time to service an I/O request. 

It could be because the drive is a piece of crap.

-- 
Mel Gorman
SUSE Labs
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html