Return-Path: <linux-nfs-owner@vger.kernel.org>
Received: from fieldses.org ([174.143.236.118]:35933 "EHLO fieldses.org"
	rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP
	id S932141Ab1BCSOF (ORCPT <rfc822;linux-nfs@vger.kernel.org>);
	Thu, 3 Feb 2011 13:14:05 -0500
Date: Thu, 3 Feb 2011 13:14:03 -0500
To: Nikolaus Rath <Nikolaus@rath.org>
Cc: linux-nfs@vger.kernel.org
Subject: Re: Client hangs in __mutex_lock_slowpath, network connection
 maxed out
Message-ID: <20110203181403.GA28234@fieldses.org>
References: <4D4AEE47.8080504@rath.org>
Content-Type: text/plain; charset=us-ascii
In-Reply-To: <4D4AEE47.8080504@rath.org>
From: "J. Bruce Fields" <bfields@fieldses.org>
Sender: linux-nfs-owner@vger.kernel.org
List-ID: <linux-nfs.vger.kernel.org>
MIME-Version: 1.0

On Thu, Feb 03, 2011 at 01:04:55PM -0500, Nikolaus Rath wrote:
> Hello,
> 
> I'm mounting an NFS4 volume from a kernel 2.6.32-27 server in a
> 2.6.37-020637rc2 client.
> Very often, when there is significant IO on the mounted volume, the
> client hangs. The process accessing the mountpoint is stuck in D
> state. Any more attempts to access the mountpoint get stuck in D state
> as well. Shortly after, the kernel reports the task as hanging:
> 
> [ 3120.620125] Call Trace:
> [ 3120.620138]  [<c05ec44d>] ? _raw_spin_lock+0xd/0x10
> [ 3120.620146]  [<c05eb4b6>] __mutex_lock_slowpath+0xc6/0x120
> [ 3120.620152]  [<c05eb1c5>] mutex_lock+0x25/0x40
> [ 3120.620158]  [<c022e3b3>] do_lookup+0xc3/0x120
> [ 3120.620164]  [<c022fbb2>] link_path_walk+0x402/0x9b0
> [ 3120.620171]  [<c0167b5a>] ? queue_work+0x1a/0x20
> 
> Additional tasks hang at a slightly different point:
> 
> [ 3120.620304]  [<f90b863d>] ? nfs_access_add_cache+0xcd/0x120 [nfs]
> [ 3120.620326]  [<f90b8709>] ? nfs_do_access+0x79/0xc0 [nfs]
> [ 3120.620333]  [<c05eb4b6>] __mutex_lock_slowpath+0xc6/0x120
> [ 3120.620339]  [<c05eb1c5>] mutex_lock+0x25/0x40
> [ 3120.620344]  [<c022e3b3>] do_lookup+0xc3/0x120
> [ 3120.620350]  [<c022fbb2>] link_path_walk+0x402/0x9b0
> [ 3120.620356]  [<c0230697>] path_walk+0x47/0xa0
> [ 3120.620361]  [<c0230819>] do_path_lookup+0x59/0xb0
> [ 3120.620367]  [<c02311ff>] user_path_at+0x3f/0x70
> [ 3120.620374]  [<c02005b8>] ? do_linear_fault+0x58/0x70
> [ 3120.620380]  [<c020133a>] ? handle_mm_fault+0xda/0x230
> [ 3120.620387]  [<c0228899>] vfs_fstatat+0x49/0x70
> 
> 
> Finally, the client saturates the 100 MBit connection with sending and
> receiving from the NFS server. I'm pretty sure this is garbage data,
> even if the client would copy the entire mountpoint there wouldn't be
> that much data to copy.

Could you take a look at that traffic with wireshark (or send the raw
packet dump data) and see what that traffic consists of?

For example it may be repeating the same few operations, in which case
there's probably some kind of infinite loop.

--b.