Re: RELENG_5 panic [nic: _mtx_lock_sleep: recursed on non-recursivemutex nfsd_mtx]

From: Robert Watson <rwatson_at_freebsd.org>
Date: Mon, 18 Oct 2004 05:22:01 -0400 (EDT)
On Mon, 18 Oct 2004, Wilkinson, Alex wrote:

> 3. I upgraded to:
> 
>    FreeBSD 5.3-BETA7 #3: Sun Oct 17 18:51:58 CST 2004
> 
>    and the panic hasn't happened since.
> 
> Something else to note is that IPv6 was not compiled into the kernel.

Hmm.  There was a bug in nfs_serv.c corrected on 2004/08/25 that corrected
an incorrect lock pair, but I don't know of any related bugs corrrected
since then.  If it's possible to get back to a point where this can be
reproduced, configuring your kernel to run with KTR and to use KTR_LOCK
would be quite helpful  You can find a bit of information on using KTR
here:

    http://www.watson.org/~robert/freebsd/netperf/ktr/

Basically, once it panics, you can use the KTR-related commands in DDB to
inspect the KTR buffer, and I believe also extract the KTR log from a core
dump.  Using the set of KTR_COMPILE flags I have on that web page would be
helpeful.  Basically, this generates a trace of locking, context switch,
interrupt, callout, memory allocation, and system call events.  The
problem likely occurred pretty immediately prior to the panic.

If this is timing-related, it could be KTR disrupts the timing
sufficiently to prevent it.  More likely, it went away because the work
load on the NFS server changed, or the directory layout changed, or some
other factor that caused the passage through the NFS code to change.

Thanks!

Robert N M Watson             FreeBSD Core Team, TrustedBSD Projects
robert_at_fledge.watson.org      Principal Research Scientist, McAfee Research


> 
>  - aW
> 
> 
> 
> 	0n Sun, Oct 17, 2004 at 10:19:34AM -0400, Robert Watson wrote: 
> 	
> 	On Fri, 15 Oct 2004, Wilkinson, Alex wrote:
> 	
> 	> Hi all,
> 	> 
> 	> I currently get a panic with "nfs_server_enable=YES" in /etc/rc.conf.
> 	> 
> 	> OS: FreeBSD 5.3-BETA4 #2: Tue Sep 14 13:55:30 UTC 2004
> 	> 
> 	> Backtrace
> 	> ---------
> 	> 
> 	> panic: _mtx_lock_sleep: recursed on non-recursive mutex nfsd_mtx _at_ /usr /src/sys/nfsserver/nfs_serv.c:1947
> 	
> 	Is the NFS server code compiled into your kernel, or is it getting loaded
> 	as a module?  Do you have any other NFS-related entries in /etc/rc.conf?
> 	Could you show the output of "show locks" and "show witness" with witness
> 	compiled into the kernel?
> 	
> 	Thanks!
> 	
> 	Robert N M Watson             FreeBSD Core Team, TrustedBSD Projects
> 	robert_at_fledge.watson.org      Principal Research Scientist, McAfee Research
> 	
> 	
> 	> cpuid = 0
> 	> KDB: enter: panic
> 	> [thread 100074]
> 	> Stopped at      kdb_enter+0x32: leave
> 	> db> tr
> 	> kdb_enter(c068ba66,0,c068aed8,dd2ed90c,c1b6b340) at kdb_enter+0x32
> 	> panic(c068aed8,c069885f,c0698617,79b,c0698617) at panic+0x1b0
> 	> _mtx_lock_sleep(c071c940,c1b6b340,0,c0698617,79b) at _mtx_lock_sleep+0x16e
> 	> _mtx_lock_flags(c071c940,0,c0698617,79b,0) at _mtx_lock_flags+0xb0
> 	> nfsrv_create(c1eb6800,c1b98800,c1b6b340,dd2edc8c,0) at nfsrv_create+0x8c4
> 	> nfssvc(c1b6b340,dd2edd14,8,0,2) at nfssvc+0x6ea
> 	> syscall(2f,2f,2f,0,0) at syscall+0x13b
> 	> Xint0x80_syscall() at Xint0x80_syscall+0x1f
> 	> --- syscall (155, FreeBSD ELF32, nfssvc), eip = 0x280c60bf, esp = 0xbfbfeb1c, eb p = 0xbfbfeb38 ---
> 	> 
> 	> 
> 	>  - aW
> 	> _______________________________________________
> 	> freebsd-current_at_freebsd.org mailing list
> 	> http://lists.freebsd.org/mailman/listinfo/freebsd-current
> 	> To unsubscribe, send any mail to "freebsd-current-unsubscribe_at_freebsd.org"
> 	> 
> 	
> _______________________________________________
> freebsd-current_at_freebsd.org mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-current
> To unsubscribe, send any mail to "freebsd-current-unsubscribe_at_freebsd.org"
> 
Received on Mon Oct 18 2004 - 07:22:10 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:38:18 UTC