aoeblk_make_request() chokes on zero-length barriers
Tejun Heo
This is reported on openSUSE 11.2 by Lars Tauber on bnc581271[1] but
upstream code seems to have the same problem. If a bio doesn't have
bi_io_vec, aoeblk_make_request() triggers BUG() but in recent kernels
the block layer will happily send down zero-length barrier bios w/o
bi_io_vec set. These zero-length barriers are used to broadcast
flushes to RAID members and should be translated into FLUSH[_EXT].
[10285.958008] ------------[ cut here ]------------
[10285.958030] kernel BUG at /usr/src/packages/BUILD/kernel-xen-2.6.31.12/linux-2.6.31/drivers/block/aoe/aoeblk.c:177!
[10285.958064] invalid opcode: 0000 [#1] SMP
[10285.958083] last sysfs file: /sys/devices/virtual/block/dm-0/range
[10285.958095] CPU 2
[10285.958104] Modules linked in: nf_conntrack_ipv4 nf_defrag_ipv4 xt_state nf_conntrack xt_physdev iptable_filter ip_tables x_tables nls_utf8 netbk blkbk blkback_pagemap blktap xenbus_be nfs lockd fscache nfs_acl auth_rpcgss sunrpc bridge stp llc fuse loop dm_mod aoe tg3 bnx2 i2c_piix4 sg pcspkr shpchp container i2c_core pci_hotplug button mptctl raid456 raid6_pq async_xor async_memcpy async_tx xor raid0 ohci_hcd ehci_hcd xenblk cdrom xennet edd raid1 fan processor mptsas mptscsih mptbase scsi_transport_sas thermal thermal_sys hwmon
[10285.958279] Pid: 3888, comm: kdmflush Not tainted 2.6.31.12-0.1-xen #1 PRIMERGY BX630 S2
[10285.958296] RIP: e030:[<ffffffffa01ea976>] [<ffffffffa01ea976>] aoeblk_make_request+0x256/0x280 [aoe]
[10285.958330] RSP: e02b:ffff8807a3949ad0 EFLAGS: 00010246
[10285.958341] RAX: 000000000000002c RBX: ffff8807a395df00 RCX: 000000000003ffff
[10285.958353] RDX: 0000000000000000 RSI: 0000000000000001 RDI: ffffffff80801490
[10285.958366] RBP: ffff8807a3949b20 R08: 0000000000000033 R09: 0000000000010643
[10285.958378] R10: 0000000000000005 R11: 0000000000000000 R12: ffff8807a4b0c600
[10285.958390] R13: 0000000000000000 R14: 0000000000000000 R15: 0000000009800810
[10285.958414] FS: 00007fbfb5ca4790(0000) GS:ffffc90000020000(0000) knlGS:0000000000000000
[10285.958429] CS: e033 DS: 0000 ES: 0000 CR0: 000000008005003b
[10285.958440] CR2: 00007fff8852f0a0 CR3: 00000007de18a000 CR4: 0000000000000660
[10285.958455] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[10285.958468] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
[10285.958481] Process kdmflush (pid: 3888, threadinfo ffff8807a3948000, task ffff8806c11cc440)
[10285.958495] Stack:
[10285.958501] ffff8806c11cc440 ffff8807a395df00 0000000000000000 0000000000000000
[10285.958519] <0> 0000000000000000 000000007e5388d1 ffff8807a395df00 ffff8807da980e58
[10285.958539] <0> 0000000000000000 0000000000000000 ffff8807a3949be0 ffffffff802223bb
[10285.958562] Call Trace:
[10285.958605] [<ffffffff802223bb>] generic_make_request+0x19b/0x4c0
[10285.958646] [<ffffffffa01fa215>] __map_bio+0xd5/0x160 [dm_mod]
[10285.958687] [<ffffffffa01fa34b>] __clone_and_map_empty_barrier+0xab/0xf0 [dm_mod]
[10285.958725] [<ffffffffa01fa6af>] __clone_and_map+0x31f/0x330 [dm_mod]
[10285.958761] [<ffffffffa01fbdb7>] __split_and_process_bio+0x117/0x1a0 [dm_mod]
[10285.958801] [<ffffffffa01fbe9f>] dm_flush+0x5f/0x90 [dm_mod]
[10285.958838] [<ffffffffa01fbf08>] process_barrier+0x38/0x150 [dm_mod]
[10285.958876] [<ffffffffa01fc07b>] dm_wq_work+0x5b/0x170 [dm_mod]
[10285.958909] [<ffffffff80069303>] run_workqueue+0x83/0x230
[10285.958930] [<ffffffff80069564>] worker_thread+0xb4/0x140
[10285.958947] [<ffffffff8006f9d6>] kthread+0xb6/0xc0
[10285.958967] [<ffffffff8000d38a>] child_rip+0xa/0x20
[10285.958981] Code: e8 1e a0 31 c0 e8 0c 13 28 e0 48 8b 7d b8 be f4 ff ff ff
e8 8d 1f f6 df e9 5e ff ff ff 48 c7 c7 b0 e5 1e a0 31 c0 e8 eb 12 28 e0 <0f> 0b
eb fe 48 c7 c7 a8 e8 1e a0 31 c0 e8 d9 12 28 e0 0f 0b eb
[10285.959107] RIP [<ffffffffa01ea976>] aoeblk_make_request+0x256/0x280 [aoe]
[10285.959128] RSP <ffff8807a3949ad0>
[10285.960116] ---[ end trace be65f4e90431af85 ]---
Thanks.
--
tejun
[1] https://bugzilla.novell.com/show_bug.cgi?id=581271
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majo...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Ed Cashin
> Hello, Ed L. Cashin.
>
> This is reported on openSUSE 11.2 by Lars Tauber on bnc581271[1] but
> upstream code seems to have the same problem. If a bio doesn't have
> bi_io_vec, aoeblk_make_request() triggers BUG() but in recent kernels
> the block layer will happily send down zero-length barrier bios w/o
> bi_io_vec set.
It looks like the opensuse 11.x kernel doesn't have the
change that made aoe return -EOPNOTSUPP for barriers.
It's commit 18d8217bc441630c3c5ec7416c5a65c69e8a0979,
appearing in v2.6.32-rc1.
I have most of a barrier implementation for aoe done
but got sidetracked. I'd welcome any volunteers who
would like to do testing. Please email me directly.
--
Ed Cashin <eca...@coraid.com>
Tejun Heo
On 03/04/2010 12:30 AM, Ed Cashin wrote:
> It looks like the opensuse 11.x kernel doesn't have the
> change that made aoe return -EOPNOTSUPP for barriers.
>
> It's commit 18d8217bc441630c3c5ec7416c5a65c69e8a0979,
> appearing in v2.6.32-rc1.
Yeap, I was looking at the wrong source tree and thought that the
problem wasn't fixed in mainline yet. Jens pointed it out and I've
committed the fix to suse kernel trees. Jens also forwarded the fix
to -stable.
Thanks.
--
tejun