When doing the conversion to danpb's new memory API, a small bug was

introduced into the qemudNetworkIfaceConnect() function.  In particular, there
is a call:

    if (VIR_ALLOC_N(vm->tapfds, vm->ntapfds+2) < 0)
        goto no_memory;

However, the tapfds structure is used to track *all* of the tap fds, and is
called once for each network that is being attached to the domain.  VIR_ALLOC_N
maps to calloc().  So the first network would work just fine, but if you had
more than one network, subsequent calls to this function would blow away the
stored fd's that were already there and fill them all in with zeros.  This
causes multiple problems, from the qemu domains not starting properly to
improper cleanup on shutdown.  The attached patch just changes the VIR_ALLOC_N()
to a VIR_REALLOC_N(), and everything is happy again.

Signed-off-by: Chris Lalancette <clalance@redhat.com>

diff --git a/src/qemu_conf.c b/src/qemu_conf.c
index f671578e24d98bfcec8defb012a17de0cab5b89a..17f0162cd9efb7706c5e3e10ad1d4896267e0896 100644 (file)
--- a/src/qemu_conf.c
+++ b/src/qemu_conf.c
@@ -2317,7 +2317,7 @@ qemudNetworkIfaceConnect(virConnectPtr conn,
     if (!(retval = strdup(tapfdstr)))
         goto no_memory;
 
-    if (VIR_ALLOC_N(vm->tapfds, vm->ntapfds+2) < 0)
+    if (VIR_REALLOC_N(vm->tapfds, vm->ntapfds+2) < 0)
         goto no_memory;
 
     vm->tapfds[vm->ntapfds++] = tapfd;