Revert "block: Add warning for bi_next not NULL in bio_endio()"
authorBart Van Assche <bart.vanassche@wdc.com>
Tue, 19 Jun 2018 17:26:40 +0000 (10:26 -0700)
committerJens Axboe <axboe@kernel.dk>
Tue, 19 Jun 2018 17:59:47 +0000 (11:59 -0600)
commit9c24c10a2c1e1bb478b6bb70612d9e885aee044f
treed81a89febf73385e45e3c4c538ee3f7580e120ed
parent0cc61e64e21cfc24fa0d938fd148aba4a595163b
Revert "block: Add warning for bi_next not NULL in bio_endio()"

Commit 0ba99ca4838b ("block: Add warning for bi_next not NULL in
bio_endio()") breaks the dm driver. end_clone_bio() detects whether
or not a bio is the last bio associated with a request by checking
the .bi_next field. Commit 0ba99ca4838b clears that field before
end_clone_bio() has had a chance to inspect that field. Hence revert
commit 0ba99ca4838b.

This patch avoids that KASAN reports the following complaint when
running the srp-test software (srp-test/run_tests -c -d -r 10 -t 02-mq):

==================================================================
BUG: KASAN: use-after-free in bio_advance+0x11b/0x1d0
Read of size 4 at addr ffff8801300e06d0 by task ksoftirqd/0/9

CPU: 0 PID: 9 Comm: ksoftirqd/0 Not tainted 4.18.0-rc1-dbg+ #1
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.0.0-prebuilt.qemu-project.org 04/01/2014
Call Trace:
 dump_stack+0xa4/0xf5
 print_address_description+0x6f/0x270
 kasan_report+0x241/0x360
 __asan_load4+0x78/0x80
 bio_advance+0x11b/0x1d0
 blk_update_request+0xa7/0x5b0
 scsi_end_request+0x56/0x320 [scsi_mod]
 scsi_io_completion+0x7d6/0xb20 [scsi_mod]
 scsi_finish_command+0x1c0/0x280 [scsi_mod]
 scsi_softirq_done+0x19a/0x230 [scsi_mod]
 blk_mq_complete_request+0x160/0x240
 scsi_mq_done+0x50/0x1a0 [scsi_mod]
 srp_recv_done+0x515/0x1330 [ib_srp]
 __ib_process_cq+0xa0/0xf0 [ib_core]
 ib_poll_handler+0x38/0xa0 [ib_core]
 irq_poll_softirq+0xe8/0x1f0
 __do_softirq+0x128/0x60d
 run_ksoftirqd+0x3f/0x60
 smpboot_thread_fn+0x352/0x460
 kthread+0x1c1/0x1e0
 ret_from_fork+0x24/0x30

Allocated by task 1918:
 save_stack+0x43/0xd0
 kasan_kmalloc+0xad/0xe0
 kasan_slab_alloc+0x11/0x20
 kmem_cache_alloc+0xfe/0x350
 mempool_alloc_slab+0x15/0x20
 mempool_alloc+0xfb/0x270
 bio_alloc_bioset+0x244/0x350
 submit_bh_wbc+0x9c/0x2f0
 __block_write_full_page+0x299/0x5a0
 block_write_full_page+0x16b/0x180
 blkdev_writepage+0x18/0x20
 __writepage+0x42/0x80
 write_cache_pages+0x376/0x8a0
 generic_writepages+0xbe/0x110
 blkdev_writepages+0xe/0x10
 do_writepages+0x9b/0x180
 __filemap_fdatawrite_range+0x178/0x1c0
 file_write_and_wait_range+0x59/0xc0
 blkdev_fsync+0x46/0x80
 vfs_fsync_range+0x66/0x100
 do_fsync+0x3d/0x70
 __x64_sys_fsync+0x21/0x30
 do_syscall_64+0x77/0x230
 entry_SYSCALL_64_after_hwframe+0x49/0xbe

Freed by task 9:
 save_stack+0x43/0xd0
 __kasan_slab_free+0x137/0x190
 kasan_slab_free+0xe/0x10
 kmem_cache_free+0xd3/0x380
 mempool_free_slab+0x17/0x20
 mempool_free+0x63/0x160
 bio_free+0x81/0xa0
 bio_put+0x59/0x60
 end_bio_bh_io_sync+0x5d/0x70
 bio_endio+0x1a7/0x360
 blk_update_request+0xd0/0x5b0
 end_clone_bio+0xa3/0xd0 [dm_mod]
 bio_endio+0x1a7/0x360
 blk_update_request+0xd0/0x5b0
 scsi_end_request+0x56/0x320 [scsi_mod]
 scsi_io_completion+0x7d6/0xb20 [scsi_mod]
 scsi_finish_command+0x1c0/0x280 [scsi_mod]
 scsi_softirq_done+0x19a/0x230 [scsi_mod]
 blk_mq_complete_request+0x160/0x240
 scsi_mq_done+0x50/0x1a0 [scsi_mod]
 srp_recv_done+0x515/0x1330 [ib_srp]
 __ib_process_cq+0xa0/0xf0 [ib_core]
 ib_poll_handler+0x38/0xa0 [ib_core]
 irq_poll_softirq+0xe8/0x1f0
 __do_softirq+0x128/0x60d

The buggy address belongs to the object at ffff8801300e0640
 which belongs to the cache bio-0 of size 200
The buggy address is located 144 bytes inside of
 200-byte region [ffff8801300e0640ffff8801300e0708)
The buggy address belongs to the page:
page:ffffea0004c03800 count:1 mapcount:0 mapping:ffff88015a563a00 index:0x0 compound_mapcount: 0
flags: 0x8000000000008100(slab|head)
raw: 8000000000008100 dead000000000100 dead000000000200 ffff88015a563a00
raw: 0000000000000000 0000000000330033 00000001ffffffff 0000000000000000
page dumped because: kasan: bad access detected

Memory state around the buggy address:
 ffff8801300e0580: fb fb fb fb fb fb fb fb fb fc fc fc fc fc fc fc
 ffff8801300e0600: fc fc fc fc fc fc fc fc fb fb fb fb fb fb fb fb
>ffff8801300e0680: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
                                                 ^
 ffff8801300e0700: fb fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
 ffff8801300e0780: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
==================================================================

Cc: Kent Overstreet <kent.overstreet@gmail.com>
Fixes: 0ba99ca4838b ("block: Add warning for bi_next not NULL in bio_endio()")
Acked-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
block/bio.c
block/blk-core.c