jfs breakage in 2.5.60
Running fsx & fsstress in parallel gets a load of oopsen in the logs, here's the first one..
Dave
Feb 12 04:30:16 mesh kernel: BUG at fs/jfs/jfs_logmgr.c:1438 assert(log->cqueue.head == NULL) Feb 12 04:30:16 mesh kernel: ------------[ cut here ]------------ Feb 12 04:30:16 mesh kernel: kernel BUG at fs/jfs/jfs_logmgr.c:1438! Feb 12 04:30:16 mesh kernel: invalid operand: 0000 Feb 12 04:30:16 mesh kernel: CPU: 0 Feb 12 04:30:16 mesh kernel: EIP: 0060:[<c02963ee>] Not tainted Feb 12 04:30:16 mesh kernel: EFLAGS: 00010296 Feb 12 04:30:16 mesh kernel: EIP is at jfs_flush_journal+0x1de/0x2b0 Feb 12 04:30:16 mesh kernel: eax: 00000044 ebx: 00000320 ecx: c3772680 edx: c0578678 Feb 12 04:30:16 mesh kernel: esi: c3742000 edi: c72fea00 ebp: c3743f70 esp: c3743f44 Feb 12 04:30:16 mesh kernel: ds: 007b es: 007b ss: 0068 Feb 12 04:30:16 mesh kernel: Process fsstress (pid: 1427, threadinfo=c3742000 task=c3773980) Feb 12 04:30:16 mesh kernel: Stack: c0518308 c0518735 0000059e c0518764 c10bbeb8 c3742000 c1287994 c72feae4 Feb 12 04:30:16 mesh kernel: c72fec40 c72fec00 00000001 c3743f80 c02774d5 c72fea00 00000001 c3743fb0 Feb 12 04:30:16 mesh kernel: c017052e c72fec00 00000001 c1287994 c72fec00 4000988c c3743fb0 c0193125 Feb 12 04:30:16 mesh kernel: Call Trace: Feb 12 04:30:16 mesh kernel: [<c02774d5>] jfs_sync_fs+0x25/0x30 Feb 12 04:30:16 mesh kernel: [<c017052e>] sync_filesystems+0x2fe/0x450 Feb 12 04:30:16 mesh kernel: [<c0193125>] sync_inodes+0x25/0xb0 Feb 12 04:30:16 mesh kernel: [<c01694fb>] sys_sync+0x3b/0x50 Feb 12 04:30:16 mesh kernel: [<c010a5f7>] syscall_call+0x7/0xb Feb 12 04:30:16 mesh kernel: Feb 12 04:30:16 mesh kernel: Code: 0f 0b 9e 05 35 87 51 c0 eb 8c e8 93 8b e8 ff e9 15 ff ff ff
Signature · 2 lines
| Dave Jones. http://www.codemonkey.org.uk | SuSE Labs
Quoted 4 lines · Dave Jones
Running fsx & fsstress in parallel gets a load of oopsen in the logs, here's the first one..
Dave
Thanks Dave,
I think I know what the problem is here. I tried to adapt some code I used at unmount for the sync_fs call, but I screwed up. The routine is waiting until all the journal writes have completed but it doesn't anticipate (or prevent) new activity. The code times out after a while because a BUG at unmount time is easier to track down than a hang.
I'll work on a real patch, but this should work in the mean time. (Pardon the compiler warning.)
===== fs/jfs/super.c 1.33 vs edited =====
fs/jfs/super.c +0 -1
@@ -396,7 +396,6 @@ .write_inode = jfs_write_inode, .delete_inode = jfs_delete_inode, .put_super = jfs_put_super, - .sync_fs = jfs_sync_fs, .write_super_lockfs = jfs_write_super_lockfs, .unlockfs = jfs_unlockfs, .statfs = jfs_statfs,
Signature · 2 lines
David Kleikamp IBM Linux Technology Center