On Mon, Jan 22, 2024 at 05:49:01PM +0800, Peter Xu wrote:
> On Fri, Jan 19, 2024 at 08:39:18PM -0300, Fabiano Rosas wrote:
> > We're currently allowing the process_incoming_migration_bh bottom-half
> > to run without holding a reference to the 'current_migration' object,
> > which leads to a segmentation fault if the BH is still live after
> > migration_shutdown() has dropped the last reference to
> > current_migration.
> >
> > In my system the bug manifests as migrate_multifd() returning true
> > when it shouldn't and multifd_load_shutdown() calling
> > multifd_recv_terminate_threads() which crashes due to an uninitialized
> > multifd_recv_state.
> >
> > Fix the issue by holding a reference to the object when scheduling the
> > BH and dropping it before returning from the BH. The same is already
> > done for the cleanup_bh at migrate_fd_cleanup_schedule().
> >
> > Signed-off-by: Fabiano Rosas <faro...@suse.de>
> > ---
> > migration/migration.c | 2 ++
> > 1 file changed, 2 insertions(+)
> >
> > diff --git a/migration/migration.c b/migration/migration.c
> > index 219447dea1..cf17b68e57 100644
> > --- a/migration/migration.c
> > +++ b/migration/migration.c
> > @@ -648,6 +648,7 @@ static void process_incoming_migration_bh(void *opaque)
> > MIGRATION_STATUS_COMPLETED);
> > qemu_bh_delete(mis->bh);
> > migration_incoming_state_destroy();
> > + object_unref(OBJECT(migrate_get_current()));
> > }
> >
> > static void coroutine_fn
> > @@ -713,6 +714,7 @@ process_incoming_migration_co(void *opaque)
> > }
> >
> > mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
> > + object_ref(OBJECT(migrate_get_current()));
> > qemu_bh_schedule(mis->bh);
> > return;
> > fail:
> > --
> > 2.35.3
> >
>
> I know I missed something, but I'd better ask: use-after-free needs to
> happen only after migration_shutdown() / qemu_cleanup(), am I right?
>
> If so, shouldn't qemu_main_loop() already returned? Then how could any BH
> keep running (including migration's) without qemu_main_loop()?
Hmm, I saw a pretty old stack mentioned in commit fd392cfa8e6:
Original output:
qemu-system-x86_64: terminating on signal 15 from pid 31980 (<unknown
process>)
=================================================================
==31958==ERROR: AddressSanitizer: heap-use-after-free on address
0x61900001d210
at pc 0x555558a535ca bp 0x7fffffffb190 sp 0x7fffffffb188
READ of size 8 at 0x61900001d210 thread T0 (qemu-vm-0)
#0 0x555558a535c9 in migrate_fd_cleanup migration/migration.c:1502:23
#1 0x5555594fde0a in aio_bh_call util/async.c:90:5
#2 0x5555594fe522 in aio_bh_poll util/async.c:118:13
#3 0x555559524783 in aio_poll util/aio-posix.c:725:17
#4 0x555559504fb3 in aio_wait_bh_oneshot util/aio-wait.c:71:5
#5 0x5555573bddf6 in virtio_blk_data_plane_stop
hw/block/dataplane/virtio-blk.c:282:5
#6 0x5555589d5c09 in virtio_bus_stop_ioeventfd
hw/virtio/virtio-bus.c:246:9
#7 0x5555589e9917 in virtio_pci_stop_ioeventfd
hw/virtio/virtio-pci.c:287:5
#8 0x5555589e22bf in virtio_pci_vmstate_change
hw/virtio/virtio-pci.c:1072:9
#9 0x555557628931 in virtio_vmstate_change hw/virtio/virtio.c:2257:9
#10 0x555557c36713 in vm_state_notify vl.c:1605:9
#11 0x55555716ef53 in do_vm_stop cpus.c:1074:9
#12 0x55555716eeff in vm_shutdown cpus.c:1092:12
#13 0x555557c4283e in main vl.c:4617:5
#14 0x7fffdfdb482f in __libc_start_main
(/lib/x86_64-linux-gnu/libc.so.6+0x2082f)
#15 0x555556ecb118 in _start
(x86_64-softmmu/qemu-system-x86_64+0x1977118)
Would that be the same case that you mentioned here? As vm_shutdown() is
indeed after migration_shutdown().
Even if so, two follow up comments..
(1) How did it help if process_incoming_migration_bh() took ref on
MigrationState*? I didn't even see it touching the object (instead, it
uses the incoming object)?
(2) This is what I'm just wondering.. on whether we should clear
current_migration to NULL in migration_shutdown() after we unref it.
Maybe it'll make such issues abort in an even clearer way.
--
Peter Xu
