Bart:

After researching the problem you described below, I have determined that a
kernel bug with dcache processing is the culprit, affecting OrangeFS as
well as other filesystems.  The bug is first introduced in RHEL 5.8,
linux-2.6.18-308.1.1.el5.x86_64 and persists throughout the remaining 5.8
life cycle (linux-2.6.18-308.20.1.el5.x86_64).  The initial release of 5.8,
which uses linux-2.6.18-308.el5.x86_64 works correctly.  The first release
of RHEL 6.0, which uses linux-2.6.32-71.18.1, also works correctly.

I have tried to narrow down which vanilla linux version contains the
correction to this problem and have determined that somewhere between
2.6.25 and 2.6.27 is where the fix occurred.  I found bug reports about
similar kernel panics surrounding slightly different scenarios where a fix
was available in 2.6.25.  In addition, I found where major dcache changes
were made starting in 2.6.27.  I haven't actually tried running these
particular linux versions to verify, but the research seems sound.

If you find out which vanilla linux version has the correction in it,
please let us know.

Thanks,
Becky Ligon

On Tue, Oct 30, 2012 at 5:41 PM, Bart Taylor <bat...@gmail.com> wrote:

>
> Has anyone else experienced consistent kernel panics on unmount?
>
> I frequently get a kernel panic when I unmount a file system from a RHEL5
> x86_64 client (kernel = 2.6.18-308.el5). That is the latest RHEL5 kernel. I
> tested this against a stock Orangefs 2.8.6 build and also against older
> PVFS 2.8 and PVFS 2.6 builds; a panic results in all cases. I also tried an
> errata build kernel=2.6.18-308.8.1.el5.
>
> Here is my test case:
>
> Setup:
> [root@client1 ~]# mkdir /mnt/orangefs/foo
> [root@client1 ~]# echo "bar" > /mnt/orangefs/foo/bar
>
> Trigger a panic:
> [root@client1 ~]# mount -t pvfs2 tcp://server2:3334/pvfs2-fs
> /mnt/orangefs/
> [root@client1 ~]# cp /mnt/orangefs/foo/bar /mnt/orangefs/
> [root@client1 ~]# umount /mnt/orangefs/
>
>
> Below is the backtrace I get from the core file. The exception is always
> in shrink_dcache_for_unmount via the pvfs2_kill_sb call.
>
> Bart.
>
>
>
> crash> bt
> PID: 30232  TASK: ffff8101bbf220c0  CPU: 0   COMMAND: "umount"
>  #0 [ffff81019eb77b60] crash_kexec at ffffffff800b099c
>  #1 [ffff81019eb77c20] __die at ffffffff80065137
>  #2 [ffff81019eb77c60] die at ffffffff8006c789
>  #3 [ffff81019eb77c90] do_invalid_op at ffffffff8006cd49
>  #4 [ffff81019eb77d50] error_exit at ffffffff8005dde9
>     [exception RIP: shrink_dcache_for_umount_subtree+375]
>     RIP: ffffffff800ef567  RSP: ffff81019eb77e08  RFLAGS: 00010286
>     RAX: 000000000000005d  RBX: ffff8101a32029c0  RCX: ffffffff80323028
>     RDX: ffffffff80323028  RSI: 0000000000000000  RDI: ffffffff80323020
>     RBP: ffff81019eb4e9d0   R8: ffffffff80323028   R9: 000000000000003e
>     R10: ffff81019eb77aa8  R11: 0000000000000080  R12: ffff8101a3202a20
>     R13: 0000000000000000  R14: 0000000000000000  R15: 00000000125aa580
>     ORIG_RAX: ffffffffffffffff  CS: 0010  SS: 0018
>  #5 [ffff81019eb77e30] shrink_dcache_for_umount at ffffffff800efa83
>  #6 [ffff81019eb77e40] generic_shutdown_super at ffffffff800e774d
>  #7 [ffff81019eb77e60] kill_anon_super at ffffffff800e787b
>  #8 [ffff81019eb77e70] pvfs2_kill_sb at ffffffff88725b8a [pvfs2]
>  #9 [ffff81019eb77e90] deactivate_super at ffffffff800e792c
> #10 [ffff81019eb77eb0] sys_umount at ffffffff800f1e8b
> #11 [ffff81019eb77f80] system_call at ffffffff8005d116
>     RIP: 00000036b80d4a67  RSP: 00007fffc15c4ee8  RFLAGS: 00010246
>     RAX: 00000000000000a6  RBX: ffffffff8005d116  RCX: 0000000000406d3f
>     RDX: 0000000000000000  RSI: 0000000000000000  RDI: 00000000125ab490
>     RBP: 00000000125ab4f0   R8: 0000000000000004   R9: 0000000000000001
>     R10: 0000000000000008  R11: 0000000000000246  R12: 00000000125ab460
>     R13: 00000000125ab490  R14: 00000000125ab4f0  R15: 0000000000000000
>     ORIG_RAX: 00000000000000a6  CS: 0033  SS: 002b
>
>
> _______________________________________________
> Pvfs2-developers mailing list
> Pvfs2-developers@beowulf-underground.org
> http://www.beowulf-underground.org/mailman/listinfo/pvfs2-developers
>
>


-- 
Becky Ligon
OrangeFS Support and Development
Omnibond Systems
Anderson, South Carolina
_______________________________________________
Pvfs2-developers mailing list
Pvfs2-developers@beowulf-underground.org
http://www.beowulf-underground.org/mailman/listinfo/pvfs2-developers

Reply via email to