Re: [Qemu-devel] [PATCH for-2.4 0/2] AioContext: fix deadlock after aio_context_acquire() race

[Paolo Bonzini <pbonzini@redhat.com>]

On 28/07/2015 12:58, Cornelia Huck wrote:
> > > Thanks.  I understand how to reproduce it now: use -drive aio=threads
> > > and do I/O during managedsave.
> > >
> > > I suspect there are more cases of this.  We need to clean it up during QEMU 2.5.
> > >
> > > For now let's continue leaking these BHs as we've always done.
> > 
> > Actually, this case can be fixed in the patch by moving
> > thread_pool_free() before the BH cleanup loop.
>
> Tried that, may have done it wrong, because the assertion still hits.

If you're doing savevm with a dataplane disk as the destination, that 
cannot work; savevm doesn't attempt to acquire the AioContext so it is 
not thread safe.

An even simpler reproducer for this bug, however, is to hot-unplug a 
disk created with x-data-plane.  It also shows another bug, fixed by 
this patch:

diff --git a/hw/block/dataplane/virtio-blk.c b/hw/block/dataplane/virtio-blk.c
index 3db139b..6106e46 100644
--- a/hw/block/dataplane/virtio-blk.c
+++ b/hw/block/dataplane/virtio-blk.c
@@ -223,8 +223,8 @@ void virtio_blk_data_plane_destroy(VirtIOBlockDataPlane *s)
     virtio_blk_data_plane_stop(s);
     blk_op_unblock_all(s->conf->conf.blk, s->blocker);
     error_free(s->blocker);
-    object_unref(OBJECT(s->iothread));
     qemu_bh_delete(s->bh);
+    object_unref(OBJECT(s->iothread));
     g_free(s);
 }
 
which I'll formally send shortly.

I would prefer to fix them all in 2.4 and risk regressions, because the
bugs are use-after-frees, i.e. pretty bad.

Paolo