util/async: print leaked BH name when AioContext finalizes

BHs must be deleted before the AioContext is finalized. If not, it's a
bug and probably indicates that some part of the program still expects
the BH to run in the future. That can lead to memory leaks, inconsistent
state, or just hangs.

Unfortunately the assert(flags & BH_DELETED) call in aio_ctx_finalize()
is difficult to debug because the assertion failure contains no
information about the BH!

Use the QEMUBH name field added in the previous patch to show a useful
error when a leaked BH is detected.

Suggested-by: Eric Ernst <eric.g.ernst@gmail.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-Id: <20210414200247.917496-3-stefanha@redhat.com>

diff --git a/util/async.c b/util/async.c
index 9a668996b8f0b2dc7a6a2f0d5af672875f20919a..9a415913190ecf13527730d3d8441979a81554ff 100644 (file)
--- a/util/async.c
+++ b/util/async.c
@@ -344,8 +344,20 @@ aio_ctx_finalize(GSource     *source)
     assert(QSIMPLEQ_EMPTY(&ctx->bh_slice_list));
 
     while ((bh = aio_bh_dequeue(&ctx->bh_list, &flags))) {
-        /* qemu_bh_delete() must have been called on BHs in this AioContext */
-        assert(flags & BH_DELETED);
+        /*
+         * qemu_bh_delete() must have been called on BHs in this AioContext. In
+         * many cases memory leaks, hangs, or inconsistent state occur when a
+         * BH is leaked because something still expects it to run.
+         *
+         * If you hit this, fix the lifecycle of the BH so that
+         * qemu_bh_delete() and any associated cleanup is called before the
+         * AioContext is finalized.
+         */
+        if (unlikely(!(flags & BH_DELETED))) {
+            fprintf(stderr, "%s: BH '%s' leaked, aborting...\n",
+                    __func__, bh->name);
+            abort();
+        }
 
         g_free(bh);
     }