[Linux-cluster] Problem with GFS2 - Kernel Panic

[Theophanis Kontogiannis]

Hello all,

 

I have a two node cluster with DRBD running in Primary/Primary.

Both nodes are running:

 

·         Kernel 2.6.18-92.1.6.el5.centos.plus

·         GFS2 fsck 0.1.44

·         cman_tool 2.0.84

·         Cluster LVM daemon version: 2.02.32-RHEL5 (2008-03-04)

Protocol version:           0.2.1

·         DRBD Version: 8.2.6 (api:88)

 

 

After a corruption (which was the result of combining updating and rebooting
with the FS mounted, network interruption during the reboot and like issues,
I keep on getting the following on one node:

 

Jun 30 00:13:40 tweety1 clurgmgrd[5283]: <notice> stop on script "BOINC"
returned 1 (generic error)

Jun 30 00:13:40 tweety1 clurgmgrd[5283]: <info> Services Initialized

Jun 30 00:13:40 tweety1 clurgmgrd[5283]: <info> State change: Local UP

Jun 30 00:13:45 tweety1 clurgmgrd[5283]: <notice> Starting stopped service
service:BOINC-t1

Jun 30 00:13:45 tweety1 kernel: GFS2: fsid=tweety:gfs2-00.0: fatal: invalid
metadata block

Jun 30 00:13:45 tweety1 kernel: GFS2: fsid=tweety:gfs2-00.0:   bh = 21879736
(magic number)

Jun 30 00:13:45 tweety1 kernel: GFS2: fsid=tweety:gfs2-00.0:   function =
gfs2_meta_indirect_buffer, file = fs/gfs2/meta_io.c, line = 332

Jun 30 00:13:45 tweety1 kernel: GFS2: fsid=tweety:gfs2-00.0: about to
withdraw this file system

Jun 30 00:13:45 tweety1 kernel: GFS2: fsid=tweety:gfs2-00.0: telling LM to
withdraw

Jun 30 00:13:46 tweety1 clurgmgrd[5283]: <notice> Service service:BOINC-t1
started

Jun 30 00:13:46 tweety1 kernel: GFS2: fsid=tweety:gfs2-00.0: withdrawn

Jun 30 00:13:46 tweety1 kernel:

Jun 30 00:13:46 tweety1 kernel: Call Trace:

Jun 30 00:13:46 tweety1 kernel:  [<ffffffff88629146>]
:gfs2:gfs2_lm_withdraw+0xc1/0xd0

Jun 30 00:13:46 tweety1 kernel:  [<ffffffff800639de>]
__wait_on_bit+0x60/0x6e

Jun 30 00:13:46 tweety1 kernel:  [<ffffffff80014eec>] sync_buffer+0x0/0x3f

Jun 30 00:13:46 tweety1 kernel:  [<ffffffff80063a58>]
out_of_line_wait_on_bit+0x6c/0x78

Jun 30 00:13:46 tweety1 kernel:  [<ffffffff8009d1bb>]
wake_bit_function+0x0/0x23

Jun 30 00:13:46 tweety1 kernel:  [<ffffffff8863af7f>]
:gfs2:gfs2_meta_check_ii+0x2c/0x38

Jun 30 00:13:46 tweety1 kernel:  [<ffffffff8862ca06>]
:gfs2:gfs2_meta_indirect_buffer+0x104/0x15e

Jun 30 00:13:46 tweety1 kernel:  [<ffffffff8862795a>]
:gfs2:gfs2_inode_refresh+0x22/0x2ca

Jun 30 00:13:46 tweety1 kernel:  [<ffffffff8009d1bb>]
wake_bit_function+0x0/0x23

Jun 30 00:13:46 tweety1 kernel:  [<ffffffff88626d9c>]
:gfs2:inode_go_lock+0x29/0x57

Jun 30 00:13:47 tweety1 kernel:  [<ffffffff88625f04>]
:gfs2:glock_wait_internal+0x1d4/0x23f

Jun 30 00:13:47 tweety1 kernel:  [<ffffffff8862611d>]
:gfs2:gfs2_glock_nq+0x1ae/0x1d4

Jun 30 00:13:47 tweety1 kernel:  [<ffffffff88632053>]
:gfs2:gfs2_lookup+0x58/0xa7

Jun 30 00:13:47 tweety1 kernel:  [<ffffffff8863204b>]
:gfs2:gfs2_lookup+0x50/0xa7

Jun 30 00:13:47 tweety1 kernel:  [<ffffffff80022663>] d_alloc+0x174/0x1a9

Jun 30 00:13:47 tweety1 kernel:  [<ffffffff8000cbb4>] do_lookup+0xd3/0x1d4

Jun 30 00:13:47 tweety1 kernel:  [<ffffffff80009f73>]
__link_path_walk+0xa01/0xf42

Jun 30 00:13:47 tweety1 kernel:  [<ffffffff8861fd37>]
:gfs2:compare_dents+0x0/0x57

Jun 30 00:13:47 tweety1 kernel:  [<ffffffff8000e782>]
link_path_walk+0x5c/0xe5

Jun 30 00:13:47 tweety1 kernel:  [<ffffffff88624d6f>]
:gfs2:gfs2_glock_put+0x26/0x133

 

 

After that, the machine freezes completely. The only way to recover is to
power-cycle / reset.

 

"gfs2-fsck -vy /dev/mapper/vg0-data0" ends (not terminates, it just look
like it finishes) with: 

 

Pass5 complete

Writing changes to disk

gfs2_fsck: buffer still held for block: 21875415 (0x14dcad7)

 

After remounting the file system and having a service start (that has its
files on this gfs2 filesystem), the kernel again crasses with the same
message and the node freezes up.

 

Unfortunately due to bad handling, I failed to DRBD invalidate the
problematic node, and instead of making it sync target (which theoretically
would solve the problem, since the good node, would sync the bad node).

Instead I made the bad node, sync source and now both nodes have the same
issue L

 

 

Any ideas of how can I resolve this issue?

 

Sincerely,

 

Theophanis Kontogiannis

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/linux-cluster/attachments/20080630/a18cf3fb/attachment.htm>