kyber: fix out of bounds access when preempted
authorOmar Sandoval <osandov@fb.com>
Tue, 11 May 2021 00:05:35 +0000 (17:05 -0700)
committerJens Axboe <axboe@kernel.dk>
Tue, 11 May 2021 14:12:14 +0000 (08:12 -0600)
commitefed9a3337e341bd0989161b97453b52567bc59d
treefb410f141f98f62e9f0ad6687b0508dea415720d
parent63c8af5687f6b1b70e9458cac1ffb25e86db1695
kyber: fix out of bounds access when preempted

__blk_mq_sched_bio_merge() gets the ctx and hctx for the current CPU and
passes the hctx to ->bio_merge(). kyber_bio_merge() then gets the ctx
for the current CPU again and uses that to get the corresponding Kyber
context in the passed hctx. However, the thread may be preempted between
the two calls to blk_mq_get_ctx(), and the ctx returned the second time
may no longer correspond to the passed hctx. This "works" accidentally
most of the time, but it can cause us to read garbage if the second ctx
came from an hctx with more ctx's than the first one (i.e., if
ctx->index_hw[hctx->type] > hctx->nr_ctx).

This manifested as this UBSAN array index out of bounds error reported
by Jakub:

UBSAN: array-index-out-of-bounds in ../kernel/locking/qspinlock.c:130:9
index 13106 is out of range for type 'long unsigned int [128]'
Call Trace:
 dump_stack+0xa4/0xe5
 ubsan_epilogue+0x5/0x40
 __ubsan_handle_out_of_bounds.cold.13+0x2a/0x34
 queued_spin_lock_slowpath+0x476/0x480
 do_raw_spin_lock+0x1c2/0x1d0
 kyber_bio_merge+0x112/0x180
 blk_mq_submit_bio+0x1f5/0x1100
 submit_bio_noacct+0x7b0/0x870
 submit_bio+0xc2/0x3a0
 btrfs_map_bio+0x4f0/0x9d0
 btrfs_submit_data_bio+0x24e/0x310
 submit_one_bio+0x7f/0xb0
 submit_extent_page+0xc4/0x440
 __extent_writepage_io+0x2b8/0x5e0
 __extent_writepage+0x28d/0x6e0
 extent_write_cache_pages+0x4d7/0x7a0
 extent_writepages+0xa2/0x110
 do_writepages+0x8f/0x180
 __writeback_single_inode+0x99/0x7f0
 writeback_sb_inodes+0x34e/0x790
 __writeback_inodes_wb+0x9e/0x120
 wb_writeback+0x4d2/0x660
 wb_workfn+0x64d/0xa10
 process_one_work+0x53a/0xa80
 worker_thread+0x69/0x5b0
 kthread+0x20b/0x240
 ret_from_fork+0x1f/0x30

Only Kyber uses the hctx, so fix it by passing the request_queue to
->bio_merge() instead. BFQ and mq-deadline just use that, and Kyber can
map the queues itself to avoid the mismatch.

Fixes: a6088845c2bf ("block: kyber: make kyber more friendly with merging")
Reported-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Link: https://lore.kernel.org/r/c7598605401a48d5cfeadebb678abd10af22b83f.1620691329.git.osandov@fb.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
block/bfq-iosched.c
block/blk-mq-sched.c
block/kyber-iosched.c
block/mq-deadline.c
include/linux/elevator.h