Allow nested qemu_bh_poll() after BH deletion

Without this, qemu segfaults when a BH handler first deletes its BH and
then calls another function which involves a nested qemu_bh_poll() call.

This can be reproduced by generating an I/O error (e.g. with blkdebug) on
an IDE device and using rerror/werror=stop to stop the VM. When continuing
the VM, qemu segfaults.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>

async.c

diff --git a/async.c b/async.c
index 57ac3a8..fd313df 100644
--- a/async.c
+++ b/async.c
@@ -137,11 +137,12 @@
 
 int qemu_bh_poll(void)
 {
-    QEMUBH *bh, **bhp;
+    QEMUBH *bh, **bhp, *next;
     int ret;
 
     ret = 0;
-    for (bh = async_context->first_bh; bh; bh = bh->next) {
+    for (bh = async_context->first_bh; bh; bh = next) {
+        next = bh->next;
         if (!bh->deleted && bh->scheduled) {
             bh->scheduled = 0;
             if (!bh->idle)