tree fd9ec57c3e0413828b1e0862d155d3093f3f0384
parent b34cdf32256e59e0183fe0f7ee5b5f4865d9ccc0
author Michael Qiu <qiudayu@huayun.com> 1612320059 +0800
committer Kevin Wolf <kwolf@redhat.com> 1613398214 +0100

blockjob: Fix crash with IOthread when block commit after snapshot

Currently, if guest has workloads, IO thread will acquire aio_context
lock before do io_submit, it leads to segmentfault when do block commit
after snapshot. Just like below:

Program received signal SIGSEGV, Segmentation fault.

[Switching to Thread 0x7f7c7d91f700 (LWP 99907)]
0x00005576d0f65aab in bdrv_mirror_top_pwritev at ../block/mirror.c:1437
1437    ../block/mirror.c: No such file or directory.
(gdb) p s->job
$17 = (MirrorBlockJob *) 0x0
(gdb) p s->stop
$18 = false

Call trace of IO thread:
0  0x00005576d0f65aab in bdrv_mirror_top_pwritev at ../block/mirror.c:1437
1  0x00005576d0f7f3ab in bdrv_driver_pwritev at ../block/io.c:1174
2  0x00005576d0f8139d in bdrv_aligned_pwritev at ../block/io.c:1988
3  0x00005576d0f81b65 in bdrv_co_pwritev_part at ../block/io.c:2156
4  0x00005576d0f8e6b7 in blk_do_pwritev_part at ../block/block-backend.c:1260
5  0x00005576d0f8e84d in blk_aio_write_entry at ../block/block-backend.c:1476
...

Switch to qemu main thread:
0  0x00007f903be704ed in __lll_lock_wait at
/lib/../lib64/libpthread.so.0
1  0x00007f903be6bde6 in _L_lock_941 at /lib/../lib64/libpthread.so.0
2  0x00007f903be6bcdf in pthread_mutex_lock at
/lib/../lib64/libpthread.so.0
3  0x0000564b21456889 in qemu_mutex_lock_impl at
../util/qemu-thread-posix.c:79
4  0x0000564b213af8a5 in block_job_add_bdrv at ../blockjob.c:224
5  0x0000564b213b00ad in block_job_create at ../blockjob.c:440
6  0x0000564b21357c0a in mirror_start_job at ../block/mirror.c:1622
7  0x0000564b2135a9af in commit_active_start at ../block/mirror.c:1867
8  0x0000564b2133d132 in qmp_block_commit at ../blockdev.c:2768
9  0x0000564b2141fef3 in qmp_marshal_block_commit at
qapi/qapi-commands-block-core.c:346
10 0x0000564b214503c9 in do_qmp_dispatch_bh at
../qapi/qmp-dispatch.c:110
11 0x0000564b21451996 in aio_bh_poll at ../util/async.c:164
12 0x0000564b2146018e in aio_dispatch at ../util/aio-posix.c:381
13 0x0000564b2145187e in aio_ctx_dispatch at ../util/async.c:306
14 0x00007f9040239049 in g_main_context_dispatch at
/lib/../lib64/libglib-2.0.so.0
15 0x0000564b21447368 in main_loop_wait at ../util/main-loop.c:232
16 0x0000564b21447368 in main_loop_wait at ../util/main-loop.c:255
17 0x0000564b21447368 in main_loop_wait at ../util/main-loop.c:531
18 0x0000564b212304e1 in qemu_main_loop at ../softmmu/runstate.c:721
19 0x0000564b20f7975e in main at ../softmmu/main.c:50

In IO thread when do bdrv_mirror_top_pwritev, the job is NULL, and stop field
is false, this means the MirrorBDSOpaque "s" object has not been initialized
yet, and this object is initialized by block_job_create(), but the initialize
process is stuck in acquiring the lock.

In this situation, IO thread come to bdrv_mirror_top_pwritev(),which means that
mirror-top node is already inserted into block graph, but its bs->opaque->job
is not initialized.

The root cause is that qemu main thread do release/acquire when hold the lock,
at the same time, IO thread get the lock after release stage, and the crash
occured.

Actually, in this situation, job->job.aio_context will not equal to
qemu_get_aio_context(), and will be the same as bs->aio_context,
thus, no need to release the lock, becasue bdrv_root_attach_child()
will not change the context.

This patch fix this issue.

Fixes: 132ada80 "block: Adjust AioContexts when attaching nodes"

Signed-off-by: Michael Qiu <qiudayu@huayun.com>
Message-Id: <20210203024059.52683-1-08005325@163.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>