block: Walk bs->children carefully in bdrv_drain_recurse

The recursive bdrv_drain_recurse may run a block job completion BH that
drops nodes. The coming changes will make that more likely and use-after-free
would happen without this patch

Stash the bs pointer and use bdrv_ref/bdrv_unref in addition to
QLIST_FOREACH_SAFE to prevent such a case from happening.

Since bdrv_unref accesses global state that is not protected by the AioContext
lock, we cannot use bdrv_ref/bdrv_unref unconditionally.  Fortunately the
protection is not needed in IOThread because only main loop can modify a graph
with the AioContext lock held.

Signed-off-by: Fam Zheng <famz@redhat.com>
Message-Id: <20170418143044.12187-2-famz@redhat.com>
Reviewed-by: Jeff Cody <jcody@redhat.com>
Tested-by: Jeff Cody <jcody@redhat.com>
Signed-off-by: Fam Zheng <famz@redhat.com>

1 files changed, 20 insertions, 3 deletions

diff --git a/block/io.c b/block/io.c
index 8706bfa578..a7142e00e8 100644
--- a/block/io.c
+++ b/block/io.c
@@ -158,7 +158,7 @@ bool bdrv_requests_pending(BlockDriverState *bs)
 
 static bool bdrv_drain_recurse(BlockDriverState *bs)
 {
-    BdrvChild *child;
+    BdrvChild *child, *tmp;
     bool waited;
 
     waited = BDRV_POLL_WHILE(bs, atomic_read(&bs->in_flight) > 0);
@@ -167,8 +167,25 @@ static bool bdrv_drain_recurse(BlockDriverState *bs)
         bs->drv->bdrv_drain(bs);
     }
 
-    QLIST_FOREACH(child, &bs->children, next) {
-        waited |= bdrv_drain_recurse(child->bs);
+    QLIST_FOREACH_SAFE(child, &bs->children, next, tmp) {
+        BlockDriverState *bs = child->bs;
+        bool in_main_loop =
+            qemu_get_current_aio_context() == qemu_get_aio_context();
+        assert(bs->refcnt > 0);
+        if (in_main_loop) {
+            /* In case the recursive bdrv_drain_recurse processes a
+             * block_job_defer_to_main_loop BH and modifies the graph,
+             * let's hold a reference to bs until we are done.
+             *
+             * IOThread doesn't have such a BH, and it is not safe to call
+             * bdrv_unref without BQL, so skip doing it there.
+             */
+            bdrv_ref(bs);
+        }
+        waited |= bdrv_drain_recurse(bs);
+        if (in_main_loop) {
+            bdrv_unref(bs);
+        }
     }
 
     return waited;