Re: panic: mutex Giant owned at nfs_syscalls.c:556

pluknet wrote:
> On 21/02/2008, Kris Kennaway <kris_at_freebsd.org> wrote:
>> pluknet wrote:
>>  > I got this assertion while attempting to remove file on nfs mounted
>>  > ffs filesystem.
>>  > NFS client on 7.0-PRERELEASE and NFS server on 8-CURRENT.
>>  >
>>  > FreeBSD 7.0-PRERELEASE #1: Wed Feb  6 18:09:18 MSK 2008
>>  > FreeBSD 8.0-CURRENT #9: Fri Feb 15 14:31:07 MSK 2008
>>  >
>>  > Unread portion of the kernel message buffer:
>>  > panic: mutex Giant owned at
>>  > /usr/src/sys/modules/nfsserver/../../nfsserver/nfs_syscalls.c:556
>>  > KDB: enter: panic
>>  > exclusive sleep mutex nfsd_mtx r = 0 (0xc41d1660) locked _at_
>>  > /usr/src/sys/modules/nfsserver/../../nfsserver/nfs_syscalls.c:501
>>  > exclusive sleep mutex Giant r = 0 (0xc07e6410) locked _at_
>>  > /usr/src/sys/kern/vfs_lookup.c:663
>>  > ...
>>  > #9  0xc053959d in panic (fmt=0xc076181d "mutex %s owned at %s:%d")
>>  >     at /usr/src/sys/kern/kern_shutdown.c:555
>>  > #10 0xc052adf7 in _mtx_assert (m=0xc07e6410, what=0,
>>  >     file=0xc41cb7b2
>>  > "/usr/src/sys/modules/nfsserver/../../nfsserver/nfs_syscalls.c",
>>  > line=556) at /usr/src/sys/kern/kern_mutex.c:652
>>  > #11 0xc41c9e82 in nfssvc (td=0xc2e68000, uap=0xd600dcfc)
>>  >     at /usr/src/sys/modules/nfsserver/../../nfsserver/nfs_syscalls.c:556
>>  > #12 0xc0727903 in syscall (frame=0xd600dd38)
>>  >     at /usr/src/sys/i386/i386/trap.c:1034
>>  > #13 0xc0711630 in Xint0x80_syscall () at /usr/src/sys/i386/i386/exception.s:203
>>  > ---Type <return> to continue, or q <return> to quit---
>>  > #14 0x00000033 in ?? ()
>>  >
>>  > Looks somewhat strange because nfs_syscalls.c:556 is not in nfssvc(),
>>  > it's in nfssvc_nfsd().
>>  > Kernel and world synchronized on 8-CUR though.
>>  >
>>
>>
>> This has been reported a couple of times before but you are the first to
>>  provide the WITNESS data, which was necessary to proceed.  Thanks :)
>>
>>  So there are two questions:
>>
>>  1) Why is Giant being acquired at all?  vfs_lookup.c:663 is a
>>  VFS_LOCK_GIANT indicating that you are doing a lookup on a non-mpsafe
>>  filesystem.  What filesystems are being exported by the server?
>>  Previously people claimed not to be exporting any filesystems that
>>  should be marked !mpsafe, which was a puzzle.
>>
> 
> I'm sorry for confusing you. :/ Actually 1 msdosfs filesystem is being exported.
> BTW panic occurred on UP kernel.
> 
>>  2) Missed VFS_UNLOCK_GIANT somewhere, or broken assert.  The above trace
>>  should be enough to diagnose though.
>>
> 
> I will try to reproduce and investigate this (in a week).

Thanks, it makes sense that it should be acquiring Giant then.

/*
  * Whatever happened to the SoC project that was supposed to work on
  * making msdosfs mpsafe?
  */

Hopefully it should be a straightforward fix to track down which code 
path is missing the VFS_UNLOCK_GIANT() to reach the above stack trace, 
or to reproduce.

Kris