Re: virtio-scsi and another complex AioContext issue

[Stefan Hajnoczi]

On Thu, Jun 11, 2020 at 10:36:22AM +0200, Sergio Lopez wrote:
> Hi,
> 
> While debugging BZ#1844343, I managed to reproduce the issue which
> leads to crash with a backtrace like this one:
> 
> <---- snip ---->
> Thread 2 (Thread 0x7fe208463f00 (LWP 1659571)):
> #0  0x00007fe2033b78ed in __lll_lock_wait () at /lib64/libpthread.so.0
> #1  0x00007fe2033b0bd4 in pthread_mutex_lock () at /lib64/libpthread.so.0
> #2  0x0000560caa8f1e6d in qemu_mutex_lock_impl
>     (mutex=0x560cacc68a10, file=0x560caaa9797f "util/async.c", line=521) at 
> util/qemu-thread-posix.c:78
> #3  0x0000560caa82414d in bdrv_set_aio_context_ignore
>     (bs=bs@entry=0x560cacc73570, 
> new_context=new_context@entry=0x560cacc5fed0, 
> ignore=ignore@entry=0x7ffe388b1cc0) at block.c:6192
> #4  0x0000560caa824503 in bdrv_child_try_set_aio_context
>     (bs=bs@entry=0x560cacc73570, ctx=0x560cacc5fed0, ignore_child=<optimized 
> out>, errp=<optimized out>)
>     at block.c:6272
> #5  0x0000560caa859e6b in blk_do_set_aio_context
>     (blk=0x560cacecf370, new_context=0x560cacc5fed0, 
> update_root_node=update_root_node@entry=true, errp=errp@entry=0x0) at 
> block/block-backend.c:1989
> #6  0x0000560caa85c501 in blk_set_aio_context
>     (blk=<optimized out>, new_context=<optimized out>, errp=errp@entry=0x0) 
> at block/block-backend.c:2010
> #7  0x0000560caa61db30 in virtio_scsi_hotunplug
>     (hotplug_dev=0x560cadaafbd0, dev=0x560cacec1210, errp=0x7ffe388b1d80)
>     at 
> /usr/src/debug/qemu-kvm-4.2.0-22.module+el8.2.1+6758+cb8d64c2.x86_64/hw/scsi/virtio-scsi.c:869
> #8  0x0000560caa6ccd1e in qdev_unplug (dev=0x560cacec1210, 
> errp=errp@entry=0x7ffe388b1db8)
>     at qdev-monitor.c:872
> #9  0x0000560caa6ccd9e in qmp_device_del (id=<optimized out>, 
> errp=errp@entry=0x7ffe388b1db8)
>     at qdev-monitor.c:884
> #10 0x0000560caa7ec4d3 in qmp_marshal_device_del
>     (args=<optimized out>, ret=<optimized out>, errp=0x7ffe388b1e18) at 
> qapi/qapi-commands-qdev.c:99
> #11 0x0000560caa8a45ec in do_qmp_dispatch
>     (errp=0x7ffe388b1e10, allow_oob=<optimized out>, request=<optimized out>, 
> cmds=0x560cab1928a0 <qmp_commands>) at qapi/qmp-dispatch.c:132
> #12 0x0000560caa8a45ec in qmp_dispatch
>     (cmds=0x560cab1928a0 <qmp_commands>, request=<optimized out>, 
> allow_oob=<optimized out>)
>     at qapi/qmp-dispatch.c:175
> #13 0x0000560caa7c2521 in monitor_qmp_dispatch (mon=0x560cacca2f00, 
> req=<optimized out>)
>     at monitor/qmp.c:145
> #14 0x0000560caa7c2bba in monitor_qmp_bh_dispatcher (data=<optimized out>) at 
> monitor/qmp.c:234
> #15 0x0000560caa8ec716 in aio_bh_call (bh=0x560cacbd80e0) at util/async.c:117
> #16 0x0000560caa8ec716 in aio_bh_poll (ctx=ctx@entry=0x560cacbd6da0) at 
> util/async.c:117
> #17 0x0000560caa8efb04 in aio_dispatch (ctx=0x560cacbd6da0) at 
> util/aio-posix.c:459
> #18 0x0000560caa8ec5f2 in aio_ctx_dispatch
>     (source=<optimized out>, callback=<optimized out>, user_data=<optimized 
> out>) at util/async.c:260
> #19 0x00007fe2078d167d in g_main_context_dispatch () at 
> /lib64/libglib-2.0.so.0
> #20 0x0000560caa8eebb8 in glib_pollfds_poll () at util/main-loop.c:219
> #21 0x0000560caa8eebb8 in os_host_main_loop_wait (timeout=<optimized out>) at 
> util/main-loop.c:242
> #22 0x0000560caa8eebb8 in main_loop_wait (nonblocking=<optimized out>) at 
> util/main-loop.c:518
> #23 0x0000560caa6cfe51 in main_loop () at vl.c:1828
> #24 0x0000560caa57b322 in main (argc=<optimized out>, argv=<optimized out>, 
> envp=<optimized out>)
>     at vl.c:4504
> 
> Thread 1 (Thread 0x7fe1fb059700 (LWP 1659573)):
> #0  0x00007fe20301b70f in raise () at /lib64/libc.so.6
> #1  0x00007fe203005b25 in abort () at /lib64/libc.so.6
> #2  0x00007fe2030059f9 in _nl_load_domain.cold.0 () at /lib64/libc.so.6
> #3  0x00007fe203013cc6 in .annobin_assert.c_end () at /lib64/libc.so.6
> #4  0x0000560caa85bfe4 in blk_get_aio_context (blk=0x560cacecf370) at 
> block/block-backend.c:1968
> #5  0x0000560caa85bfe4 in blk_get_aio_context (blk=0x560cacecf370) at 
> block/block-backend.c:1962
> #6  0x0000560caa61d79c in virtio_scsi_ctx_check (s=0x560cadaafbd0, 
> s=0x560cadaafbd0, d=0x560cacec1210)
>     at 
> /usr/src/debug/qemu-kvm-4.2.0-22.module+el8.2.1+6758+cb8d64c2.x86_64/hw/scsi/virtio-scsi.c:250
> #7  0x0000560caa61d79c in virtio_scsi_handle_cmd_req_prepare 
> (req=0x7fe1ec013880, s=0x560cadaafbd0)
>     at 
> /usr/src/debug/qemu-kvm-4.2.0-22.module+el8.2.1+6758+cb8d64c2.x86_64/hw/scsi/virtio-scsi.c:569
> #8  0x0000560caa61d79c in virtio_scsi_handle_cmd_vq
>     (s=s@entry=0x560cadaafbd0, vq=vq@entry=0x7fe1f82ac140)
>     at 
> /usr/src/debug/qemu-kvm-4.2.0-22.module+el8.2.1+6758+cb8d64c2.x86_64/hw/scsi/virtio-scsi.c:612
> #9  0x0000560caa61e48e in virtio_scsi_data_plane_handle_cmd (vdev=<optimized 
> out>, vq=0x7fe1f82ac140)
>     at 
> /usr/src/debug/qemu-kvm-4.2.0-22.module+el8.2.1+6758+cb8d64c2.x86_64/hw/scsi/virtio-scsi-dataplane.c:60
> #10 0x0000560caa62bfbe in virtio_queue_notify_aio_vq (vq=<optimized out>)
>     at 
> /usr/src/debug/qemu-kvm-4.2.0-22.module+el8.2.1+6758+cb8d64c2.x86_64/hw/virtio/virtio.c:2243
> #11 0x0000560caa8ef046 in run_poll_handlers_once
>     (ctx=ctx@entry=0x560cacc689b0, timeout=timeout@entry=0x7fe1fb058658) at 
> util/aio-posix.c:517
> #12 0x0000560caa8efbc5 in try_poll_mode (timeout=0x7fe1fb058658, 
> ctx=0x560cacc689b0)
>     at util/aio-posix.c:607
> #13 0x0000560caa8efbc5 in aio_poll (ctx=0x560cacc689b0, 
> blocking=blocking@entry=true)
>     at util/aio-posix.c:639
> #14 0x0000560caa6ca7f4 in iothread_run (opaque=0x560cacc1f000) at 
> iothread.c:75
> #15 0x0000560caa8f1d84 in qemu_thread_start (args=0x560cacc666f0) at 
> util/qemu-thread-posix.c:519
> #16 0x00007fe2033ae2de in start_thread () at /lib64/libpthread.so.0
> #17 0x00007fe2030dfe83 in clone () at /lib64/libc.so.6
> <---- snip ---->
> 
> Both the code path initiated by virtio_scsi_data_plane_cmd_vq() and
> the one coming down from virtio_scsi_hotunplug() should be protected
> by virtio_scsi_acquire(), but this can still happen because the latter
> works by acquiring the AioContext pointed by s->ctx, and we have this
> in bdrv_set_aio_context_ignore():
> 
>    6140 void bdrv_set_aio_context_ignore(BlockDriverState *bs,
>    6141                                  AioContext *new_context, GSList 
> **ignore)
>    6142 {
> (...)
>    6179     /*
>    6180      * If this function was recursively called from
>    6181      * bdrv_set_aio_context_ignore(), there may be nodes in the
>    6182      * subtree that have not yet been moved to the new AioContext.
>    6183      * Release the old one so bdrv_drained_end() can poll them.
>    6184      */
>    6185     if (qemu_get_aio_context() != old_context) {
>    6186         aio_context_release(old_context);
>    6187     }
> 
> My first thought here is that given the context and the apparent
> expectations of where and how virtio_scsi_acquire() is used, we should
> probably use an independent lock on VirtIOSCSI instead of acquiring
> the AioContext.
> 
> What do you think?

The virtio_scsi_ctx_check() assertion seems reasonable. virtio-scsi
shouldn't be processing requests while AioContexts are changing.

I'm surprised that aio_disable_external() doesn't already protect this
code path?

Stefan

signature.asc
Description: PGP signature