[Ocfs2-users] kernel BUG at fs/inode.c:250

[Mark Fasheh]

On Tue, May 22, 2007 at 12:24:09AM +0200, Yassine Diallo wrote:
> Hello,
> 
> I've installed the latest available version of ocfs2 1.2.5.1 by building the
> source since no rpm was available for my distro ( suse 10.2) running on kernel
> 2.6.18.2-34.
> 
> I managed to compile and install the source. The whole ocfs2 stack is running
> on my two small nodes and I manage to touch / cp / rm files and makes them
> visible on both nodes. However if I start to launch the Oracle clusterware
> installation , the OUI get stuck during the process of creating the OCR files.
> I receive the following message in dmesg on one of the two nodes :
> 
> kernel BUG at fs/inode.c:250!
> invalid opcode: 0000 [#1]
> SMP
> last sysfs file: /devices/pci0000:00/0000:00:19.0/0000:07:11.0/fw-host0/
> 0030bd0623001554/0030bd0623001554-0/version
> Modules linked in: ocfs2 ocfs2_dlmfs ocfs2_dlm ocfs2_nodemanager configfs ipv6
> snd_pcm_oss snd_mixer_oss snd_seq snd_seq_device button battery ac apparmor
> aamatch_pcre loop dm_mod sbp2 usb_storage usbhid r8169 ohci1394 sk98lin
> ieee1394 i2c_ali1535 i2c_ali15x3 snd_hda_intel snd_hda_codec i2c_core snd_pcm
> snd_timer ide_cd cdrom snd soundcore snd_page_alloc shpchp pci_hotplug ati_agp
> agpgart ehci_hcd ohci_hcd usbcore parport_pc lp parport ext3 mbcache jbd edd sg
> fan sata_uli libata alim15x3 thermal processor sd_mod scsi_mod ide_disk
> ide_core
> CPU:    1
> EIP:    0060:[<c017a130>]    Tainted: G     U VLI
> EFLAGS: 00010202   (2.6.18.2-34-default #1)
> EIP is at clear_inode+0x1c/0xec
> eax: c1edd388   ebx: c1edd388   ecx: ef5fb830   edx: ee58c000
> esi: f0f1da94   edi: ef284000   ebp: 00000000   esp: ee58de3c
> ds: 007b   es: 007b   ss: 0068
> Process ocfs2vote-1 (pid: 4996, ti=ee58c000 task=ef5fb830 task.ti=ee58c000)
> Stack: d54c04c0 c1edd388 f0f1da94 f0f1e37d ee58de48 ef5fb830 c44fec00 00000038
>        00000002 2464e6d4 00000000 46521229 0006eb1b 46521229 0006eb1c c1edd388
>        0006eb21 00000000 00000000 00000038 00000001 00000000 c1edd388 f0f1da94
> Call Trace:
>  [<f0f1da94>] ocfs2_delete_inode+0x0/0x94e [ocfs2]
>  [<f0f1e37d>] ocfs2_delete_inode+0x8e9/0x94e [ocfs2]
>  [<f0f1da94>] ocfs2_delete_inode+0x0/0x94e [ocfs2]
>  [<c017a2a5>] generic_delete_inode+0xa5/0x10f
>  [<f0f1d3e3>] ocfs2_drop_inode+0x123/0x178 [ocfs2]
>  [<f0e8978d>] o2net_send_message+0x31/0x36 [ocfs2_nodemanager]
>  [<c01bf87a>] _atomic_dec_and_lock+0x2a/0x44
>  [<c0179d3d>] iput+0x63/0x65
>  [<f0f3d373>] ocfs2_process_vote+0xcbe/0xcea [ocfs2]
>  [<c02a43fe>] schedule+0x8aa/0x962
>  [<f0f100d3>] ocfs2_unblock_meta+0x540/0x756 [ocfs2]
>  [<f0f3e27b>] ocfs2_vote_thread+0x1f6/0x409 [ocfs2]
>  [<c01314f2>] autoremove_wake_function+0x0/0x35
>  [<f0f3e085>] ocfs2_vote_thread+0x0/0x409 [ocfs2]
>  [<c0131420>] kthread+0xc0/0xec
>  [<c0131360>] kthread+0x0/0xec
>  [<c0102005>] kernel_thread_helper+0x5/0xb
> 
> I *believe* that it comes back to the following line in the ocfs source file fs
> /inode.c
> 
> #ifdef INODE_HAS_BLKSIZE
>         inode->i_blksize = (u32)osb->s_clustersize;
> #endif

You're looking at the wrong file - you need to be looking at "fs/inode.c" in
your kernel tree.

The stack indicates you crashed in clear_inode(). For folks that build ocfs2
1.2 from source, that's usually because some backwards compat code isn't
turned on. How exactly did you build? Did you use "make", or did you use one
of the spec files in the ocfs2 distribution?

Play with the "GENERIC_DELETE_INODE_NOT_TRUNCATES" directive to make - I
believe in your case you need to make sure it's off.
	--Mark

--
Mark Fasheh
Senior Software Developer, Oracle
mark.fasheh at oracle.com