Crash in jbd2_chksum due to null journal->j_chksum_driver

[Nikolay Borisov]

Hello, 

Today a colleague was testing something and while doing so he observed 
the following crash: 

jbd2_journal_bmap: journal block not found at offset 67 on dm-26-8
Aborting journal on device dm-26-8.
BUG: unable to handle kernel NULL pointer dereference at           (null)
IP: [<ffffffff812b12eb>] jbd2_superblock_csum+0x2b/0x80
PGD 3fcef54067 PUD 3fce84e067 PMD 0 
Oops: 0000 [#1] SMP 
Modules linked in: act_police cls_basic sch_ingress veth dm_snapshot openvswitch gre vxlan ip_tunnel xt_owner xt_conntrack iptable_mangle xt_nat iptable_nat nf_conntrack_ipv4 nf_defrag_ipv4 nf_nat_ipv4 nf_nat xt_CT nf_conntrack iptable_raw ib_ipoib rdma_ucm ib_ucm ib_uverbs ib_umad rdma_cm ib_cm iw_cm ib_sa ib_mad ib_core ib_addr ipv6 ext2 dm_thin_pool dm_bio_prison dm_persistent_data dm_bufio dm_mirror dm_region_hash dm_log ses enclosure igb i2c_algo_bit x86_pkg_temp_thermal crc32_pclmul i2c_i801 lpc_ich mfd_core ioapic ioatdma dca shpchp ipmi_devintf ipmi_si ipmi_msghandler
CPU: 0 PID: 12059 Comm: jbd2/dm-26-8 Not tainted 3.12.47-clouder1 #1
Hardware name: Supermicro X10DRi/X10DRi, BIOS 1.1 04/14/2015
task: ffff883f904958b0 ti: ffff883fce4d8000 task.ti: ffff883fce4d8000
RIP: 0010:[<ffffffff812b12eb>]  [<ffffffff812b12eb>] jbd2_superblock_csum+0x2b/0x80
RSP: 0018:ffff883fce4d9a58  EFLAGS: 00010282
RAX: 0000000000000000 RBX: ffff883f8dd77000 RCX: 0000000000000006
RDX: 0000000000000000 RSI: ffff883f8dd77000 RDI: ffff883fa0fc6800
RBP: ffff883fce4d9a88 R08: 0000000000000000 R09: 0000000000000000
R10: 0000000000000001 R11: 0000000000000000 R12: 00000000f0459c0b
R13: 0000000000000411 R14: ffff883f8dd77000 R15: 00000000560bb55d
FS:  0000000000000000(0000) GS:ffff881fffa00000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000000000000 CR3: 0000003fd145d000 CR4: 00000000001407f0
Stack:
 ffffffff81e07402 ffff883fa0fc6800 00000000fffffffb ffff883fce4d9b90
 ffff883f8dd77000 ffff883fa0fc6800 ffff883fce4d9aa8 ffffffff812b1369
 0000000000000010 ffff883f90c772d8 ffff883fce4d9ae8 ffffffff812b1455
Call Trace:
 [<ffffffff812b1369>] jbd2_superblock_csum_set+0x29/0x40
 [<ffffffff812b1455>] jbd2_write_superblock+0x85/0x1b0
 [<ffffffff812b1b70>] jbd2_journal_update_sb_errno+0x50/0x60
 [<ffffffff812b1bd0>] __journal_abort_soft+0x50/0x60
 [<ffffffff812b1c80>] jbd2_journal_bmap+0x90/0xa0
 [<ffffffff812b1ec7>] jbd2_journal_next_log_block+0x77/0x80
 [<ffffffff812b1ef3>] jbd2_journal_get_descriptor_buffer+0x23/0xb0
 [<ffffffff812aa02c>] journal_submit_commit_record+0x7c/0x1e0
 [<ffffffff812abade>] jbd2_journal_commit_transaction+0x194e/0x1d20
 [<ffffffff812b062f>] kjournald2+0xef/0x2b0
 [<ffffffff810aef00>] ? wake_up_bit+0x40/0x40
 [<ffffffff812b0540>] ? commit_timeout+0x10/0x10
 [<ffffffff810ae48e>] kthread+0xce/0xe0
 [<ffffffff810ae3c0>] ? kthread_freezable_should_stop+0x80/0x80
 [<ffffffff816571c8>] ret_from_fork+0x58/0x90
 [<ffffffff810ae3c0>] ? kthread_freezable_should_stop+0x80/0x80
Code: 55 48 89 e5 41 54 53 48 83 ec 20 0f 1f 44 00 00 44 8b a6 fc 00 00 00 48 89 f3 c7 86 fc 00 00 00 00 00 00 00 48 8b 87 d0 04 00 00 <83> 38 04 77 39 48 89 45 d0 c7 45 d8 00 00 00 00 48 8d 7d d0 c7 
RIP  [<ffffffff812b12eb>] jbd2_superblock_csum+0x2b/0x80
 RSP <ffff883fce4d9a58>
CR2: 0000000000000000
---[ end trace e1bd94031f410b71 ]---

The ffffffff812b12eb address actually is jbd2_chksum and the 
instruction where the deference is happening in 
crypto_shash_descsize(), essentially journal->j_chksum_driver is NULL. 

Now, how we got ourselves in this situation - we have an lvm thin 
volume with ext4 fs and a container started from it,
then, while the container is running we invoke the following 
command to scrub its contents:

openssl enc -aes-256-ctr -pass pass:"$(dd if=/dev/urandom bs=128 count=1 2>/dev/null | base64)" -nosalt </dev/zero | dd bs=64K of=/dev/volumegroupname/volumename


And then when we try to umount the volume we get the aforementioned 
crash. Naturally, because we overwrite the on-disk contents jbd2_journal_bmap 
fails which triggers the journal abort which wants to update the on-disk
errno, which naturally triggers a superblock checksum regeneration
and this goes BOOM. 

I looked around the code but couldn't figure out a code path
which allows the checksum driver to become null at runtime.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/