NetBSD Problem Report #13863

Received: (qmail 8882 invoked from network); 4 Sep 2001 06:38:56 -0000
Message-Id: <20010904064402.1340.qmail@eros.endersgame.net>
Date: 4 Sep 2001 06:44:02 -0000
From: dive@endersgame.net
Reply-To: dive@endersgame.net
To: gnats-bugs@gnats.netbsd.org
Subject: writing to an shlight-mounted filesystem causes total NFS hang
X-Send-Pr-Version: 3.95

>Number:         13863
>Category:       pkg
>Synopsis:       writing to an shlight-mounted filesystem hangs and kills NFS
>Confidential:   no
>Severity:       serious
>Priority:       high
>Responsible:    pkg-manager
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Tue Sep 04 06:39:01 +0000 2001
>Closed-Date:    
>Last-Modified:  Tue Dec 21 09:37:01 +0000 2010
>Originator:     Sean Davis
>Release:        20010904
>Organization:
EGNet
>Environment:
System: NetBSD eros.endersgame.net 1.5X NetBSD 1.5X (EROS) #128: Mon Sep 3 22:24:52 EDT 2001 dive@eros.endersgame.net:/work/src/sys/arch/i386/compile/EROS i386
Architecture: i386
Machine: i386
>Description:
Creating/writing to a file on a filesystem mounted via the shlight program more
often than not causes the process doing the writing to hang and not respond to
kill -9, and freezes up all NFS activity, causing df, mount, and anything that
tries to access that filesystem to hang as well. The only way to recover from
this is to reboot, and three out of the four times this has happened, I ended up
having to drop into DDB and reboot with the sync command, since shutdown -r now
got as far as "syncing disks..." and stuck.

>How-To-Repeat:
shlight //foo/share /mnt/foo
cd /mnt/foo
bonnie	(or try to create a tar archive, or use pax, those are the three i've
tried so far and all three hung it. ls works, FWIW)

>Fix:
unknown
>Release-Note:
>Audit-Trail:
State-Changed-From-To: open->closed 
State-Changed-By: wiz 
State-Changed-When: Sun Mar 24 07:43:24 PST 2002 
State-Changed-Why:  
Submitter reports problem is fixed. 
State-Changed-From-To: closed->feedback 
State-Changed-By: wiz 
State-Changed-When: Sun Mar 24 07:43:53 PST 2002 
State-Changed-Why:  
Sorry, last message was incorrect. 
MAINTAINER says package should work; could you please try again? 
Thanks. 

From: Sean Davis <dive-nb@endersgame.net>
To: gnats-bugs@gnats.netbsd.org, netbsd-bugs@netbsd.org, wiz@netbsd.org
Cc:  
Subject: pkg/13863
Date: Mon, 8 Apr 2002 02:46:01 -0400

 It still causes the problems outlined in the PR. I just tried it. The
 commandline was:

 shlight //bugger/share /smb

 I was able to create a file (echo foo > foo), dd 5k to a file (dd
 if=/dev/zero of=foo bs=1024 count=5), but when I tried to make an archive in
 the shlight-mounted directory with pax, it hung, just as the PR describes.

 descolada# pax -vwPXf /smb/desco-root.tar /
 pax: Removing leading / from absolute path names in the archive

 usr
 etc
 etc/fstab
 etc/defaults
 etc/defaults/daily.conf
 etc/defaults/monthly.conf
 etc/defaults/rc.conf^C

 I hit control C when it stuck at rc.conf, but it didn't exit, and is still
 sitting there. I expect I'll have to reboot it. df, mount, etc all do the
 same thing, and cannot be killed even by root.

 The machine is running NetBSD current i386, and the windows box is running
 Windows ME.

 -- 
 /~\ The ASCII                         Sean Davis
 \ / Ribbon Campaign                    aka dive
  X  Against HTML
 / \ Email!               http://eros.endersgame.net:8000/~dive

From: Sean Davis <dive-nb@endersgame.net>
To: gnats-bugs@gnats.netbsd.org, wiz@netbsd.org
Cc:  
Subject: pkg/13863
Date: Mon, 8 Apr 2002 02:51:04 -0400

 It hung on shutdown -r now, also. After waiting a while, I dropped into DDB.
 Here's the traceback I got:
 db> t
 cpu_Debugger(c02f5f00,c3ed8ac8,c3ed8ac8,c0102efe,c02f5ec0) at
 cpu_Debugger+0x4
 comintr(c02ed400) at comintr+0xf4
 Xintr4() at Xintr4+0x7e
 --- interrupt ---
 idle(c3ed8ac8,1f4,c011f83c,c3ed8ac8) at idle+0x1b
 bpendtsleep(c0345608,18,c01c5f0e,1f4,0) at bpendtsleep
 sbwait(c0345608,c0134e74,8,202,23) at sbwait+0x39
 soreceive(c03455c4,0,c3faacec,c3faad58,0) at soreceive+0x2a0
 nfs_receive(c02f1e80,c3faad54,c3faad58,0,c02ea100) at nfs_receive+0x45d
 nfs_reply(c02f1e80,0,3b9aca00,0,c3faae70) at nfs_reply+0x59
 nfs_request(c3ef943c,c02ead00,7,0,c0358e80) at nfs_request+0x3c6
 nfs_writerpc(c3ef943c,c3faaf0c,c3faaefc,c3faaf00,c02bf8a4) at
 nfs_writerpc+0x6ec

 nfs_doio(c030c1a0,0,0,0) at nfs_doio+0x5d9
 nfssvc_iod(c3ed8ac8,0,c01008f4,c3ed8ac8,0,0,0,c0100337,c3ed8ac8,c028,0,0,c010033
 1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,c3facff0,10,0,0,0,0,21d000,0,0,0
 ,0,0,0,c3facea0,c3faceec,0,0,0,0,0,0,0,0,18,2980000) at nfssvc_iod+0x136
 start_nfsio(c3ed8ac8) at start_nfsio+0x14
 db> 


 -- 
 /~\ The ASCII                         Sean Davis
 \ / Ribbon Campaign                    aka dive
  X  Against HTML
 / \ Email!               http://eros.endersgame.net:8000/~dive

From: Sean Davis <dive@endersgame.net>
To: gnats-bugs@netbsd.org
Cc:  
Subject: pkg/13863
Date: Sun, 5 May 2002 19:07:44 -0400

 Please close this PR. The problem still exists, and in my tests still
 happens, but this PR has been open for a while with no change. Perhaps mark
 the package as BROKEN? It does work, just not correctly... dunno.

 -- 
 /~\ The ASCII                         Sean Davis
 \ / Ribbon Campaign                    aka dive
  X  Against HTML
 / \ Email!               http://eros.endersgame.net:8000/~dive
State-Changed-From-To: feedback->open 
State-Changed-By: wiz 
State-Changed-When: Wed May 8 04:07:58 PDT 2002 
State-Changed-Why:  
Feedback supplied, problem still there. 
Sorry that noone found the cause for your problem yet, but I haven't given 
up hope yet that someone will, so I'll keep the PR open for now. 
State-Changed-From-To: open->feedback
State-Changed-By: asau@NetBSD.org
State-Changed-When: Fri, 19 Feb 2010 12:09:30 +0000
State-Changed-Why:
Reading this 8 years later I don't understand what the problem was.


From: David Holland <dholland-pbugs@netbsd.org>
To: gnats-bugs@NetBSD.org
Cc: asau@NetBSD.org, dive@endersgame.net
Subject: Re: pkg/13863 (writing to an shlight-mounted filesystem hangs and
	kills NFS)
Date: Sat, 20 Feb 2010 08:39:23 +0000

 On Fri, Feb 19, 2010 at 12:09:31PM +0000, asau@NetBSD.org wrote:
  > Synopsis: writing to an shlight-mounted filesystem hangs and kills NFS
  > 
  > State-Changed-From-To: open->feedback
  > State-Changed-By: asau@NetBSD.org
  > State-Changed-When: Fri, 19 Feb 2010 12:09:30 +0000
  > State-Changed-Why:
  > Reading this 8 years later I don't understand what the problem was.

 It sounds to me like a generic usermode-filesystem type of deadlock.

 How does the package work? It's far too old to be puffs; is it a
 user-level loopback NFS server?

 -- 
 David A. Holland
 dholland@netbsd.org

State-Changed-From-To: feedback->closed
State-Changed-By: asau@NetBSD.org
State-Changed-When: Tue, 14 Dec 2010 18:33:29 +0000
State-Changed-Why:
It isn't clear what the problem was.
It isn't clear what package is affected, and there was no feedback
enough to derive necessary information.
If you still have the problem, reply to reopen or file another PR.


From: David Holland <dholland-pbugs@netbsd.org>
To: gnats-bugs@NetBSD.org
Cc: Aleksej Saushev <asau@netbsd.org>
Subject: Re: pkg/13863 (writing to an shlight-mounted filesystem hangs and
 kills NFS)
Date: Tue, 21 Dec 2010 08:40:00 +0000

 On Tue, Dec 14, 2010 at 06:33:30PM +0000, asau@NetBSD.org wrote:
  > Synopsis: writing to an shlight-mounted filesystem hangs and kills NFS
  > 
  > State-Changed-From-To: feedback->closed
  > State-Changed-By: asau@NetBSD.org
  > State-Changed-When: Tue, 14 Dec 2010 18:33:29 +0000
  > State-Changed-Why:
  > It isn't clear what the problem was.
  > It isn't clear what package is affected, and there was no feedback
  > enough to derive necessary information.
  > If you still have the problem, reply to reopen or file another PR.

 The package is net/sharity-light, a loopback NFS server; the problem
 is that it deadlocks/ed the kernel; and I'm sure the problem is easy
 to reproduce, or it would be if (1) it hasn't gone away and (2) the
 package even compiled these days.

 There does not seem to be a fuse-smbfs or other more modern non-kernel
 smbfs in pkgsrc, though, so I'm not sure the package should be
 abandoned entirely.

 Plus, NetBSD shouldn't fall over just because someone tries to do a
 loopback NFS mount.

 -- 
 David A. Holland
 dholland@netbsd.org

State-Changed-From-To: closed->open
State-Changed-By: asau@NetBSD.org
State-Changed-When: Tue, 21 Dec 2010 09:37:01 +0000
State-Changed-Why:
Suddenly, feedback arrived.


>Unformatted:

Home	PR Database Search