NetBSD Problem Report #42737

From riastradh@joule.t.continuation.org  Thu Feb  4 05:21:43 2010
Return-Path: <riastradh@joule.t.continuation.org>
Received: from mail.netbsd.org (mail.netbsd.org [204.152.190.11])
	by www.NetBSD.org (Postfix) with ESMTP id EBF6E63B886
	for <gnats-bugs@gnats.NetBSD.org>; Thu,  4 Feb 2010 05:21:42 +0000 (UTC)
Message-Id: <20100204052141.1EAA7120D@joule.t.continuation.org>
Date: Thu,  4 Feb 2010 05:21:41 +0000 (UTC)
From: Taylor R Campbell <campbell@mumble.net>
Reply-To: Taylor R Campbell <campbell@mumble.net>
To: gnats-bugs@gnats.NetBSD.org
Subject: sbp temporary detachment feature causes deadlocks
X-Send-Pr-Version: 3.95

>Number:         42737
>Category:       kern
>Synopsis:       sbp temporary detachment feature causes deadlocks
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    kern-bug-people
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Thu Feb 04 05:25:00 +0000 2010
>Last-Modified:  Thu Feb 04 05:40:03 +0000 2010
>Originator:     Taylor R Campbell <campbell@mumble.net>
>Release:        NetBSD 5.0_STABLE
>Organization:
>Environment:
NetBSD ... 5.0_STABLE NetBSD 5.0_STABLE (APPLE_UFS_EI) #2: Fri Jul 31 20:44:51 UTC 2009  riastradh@... i386
Architecture: i386
Machine: i386
>Description:

	Several days ago, I plugged a FireWire device (specifically, a
	MacBook with a hosed disk, booted into target disk mode) into
	a machine running NetBSD 5.0_STABLE.  I performed various
	diagnostics upon it, and mounted some some file systems from
	it using RUMP, which I fully unmounted before unplugging the
	device.  I did not run `fwctl -r' after unplugging the device,
	however, which apparently one is supposed to do.

	Today, I noticed that I hadn't gotten daily and security
	reports from that machine, which turns out to be because the
	processes generating them are waiting for `disklabel sd0',
	which is hung on a mutex.  I tried running `file -s /dev/sd0a'
	which hung too, unkillably.

	My kernel is configured just like GENERIC with the following
	two options added:

options 	FFS_EI		# FFS Endian Independent support
options 	APPLE_UFS	# Apple UFS support in FFS

	Whether or not I ought to have run `fwctl -r' after unplugging
	the device, NetBSD shouldn't wedge up like this.

>How-To-Repeat:

	I have not yet attempted to reproduce this; I shall do so once
	I have built a kernel with LOCKDEBUG enabled in order to
	diagnose the problem better in case I can reproduce it.

>Fix:

	I do not yet have a fix, but shall endeavour to seek one once
	I can reproduce the problem and debug the kernel symbolically.

>Audit-Trail:
From: Taylor R Campbell <campbell@mumble.net>
To: gnats-bugs@NetBSD.org
Cc: 
Subject: Re: kern/42737: sbp temporary detachment feature causes deadlocks
Date: Thu, 4 Feb 2010 00:35:55 -0500

 Forgot to add: I entered ddb and ran `t/t 0t87' to get a trace for
 process 87, one of the `disklabel sd0' processes, and copied the
 output by hand.  (No doubt I made some errors in transcription.)

 db{0}> t/t 0t87
 trace: pid 87 lid 1 at 0xcbfa19ac
 sleepq_block(0,0,c0a26161,c0a882fc,ca95f4d0,cbfd2458,0,cd4b0a7c,75,40) at n=
 etbsd:sleepq_block+0xeb
 turnstile_block(cd4b0a64,1,cebab400,c0a882fc,cc561e4c,1,cbfa1a4c,c04cd27d,2=
 0000020,cd4b0a64) at netbsd:turnstile_block+0x1a5
 mutex_vector_enter(cebab400,0,0,1,d0818e64,cebab400,cbfa1aec,c03d8988,d0818=
 f04,3) at netbsd:mutex_vector_enter+0x398
 sdopen(d03,1,2000,cbfd22a0,2000,1,6,d0818e64,cbb94268,0) at netbsd:sdopen+0=
 x8c
 cdev_open(d03,1,2000,cbfd22a0,4,1a0,0,5,cbfd22a0,d03) at netbsd:cdev_open+0=
 x99
 spec_open(cbfa1b88,3,1,c048b237,6,c0826f60,d0818e64,1,d1f05840,0) at netbsd=
 :spec_open+0x165
 VOP_OPEN(d0818e64,1,d1f05840,1,cc465ba4,0,0,d0818e64,cbfd22a0,d1f05840) at =
 netbsd:VOP_OPEN+0x6c
 vn_open(cbfa1c78,1,40,c0474d12,c0b3bac0,cd7e996c,0,c04752d9,cbfa1cdc,cbb25c=
 00) at netbsd:vn_open+0x259
 sys_open(cbfd22a0,cbfa1d00,cbfa1d28,8052640,0,64,8052640,0,0,bbbf6983) at n=
 etbsd:sys_open+0xc0
 syscall(cbfa1d48,b3,ab,1f,1f,8052640,64,bfbfe948,bbbe68b8,bbbd5840) at netb=
 sd:syscall+0xbd

Home	PR Database Search