Be much more paranoid about backends, now that they can go away.

git-svn-id: svn+ssh://projects.linpro.no/svn/varnish/trunk@1917 d4fa192b-c00b-0410-8231-f00ffab90ce4

diff --git a/varnish-cache/bin/varnishd/cache_center.c b/varnish-cache/bin/varnishd/cache_center.c
index 3650e3e7864bde1dfcaac85a18e95f8800fd5a2d..09f07edca00eaa21c0273fa3b6bf689b3017c86a 100644 (file)
--- a/varnish-cache/bin/varnishd/cache_center.c
+++ b/varnish-cache/bin/varnishd/cache_center.c
@@ -297,7 +297,9 @@ cnt_fetch(struct sess *sp)
        int i;
 
        AN(sp->bereq);
+       CHECK_OBJ_NOTNULL(sp->backend, BACKEND_MAGIC);
        i = Fetch(sp);
+       CHECK_OBJ_NOTNULL(sp->backend, BACKEND_MAGIC);
 
        /* Experimental. Set time for last check of backend health.
         * If the backend replied with 200, it is obviously up and running,
@@ -426,6 +428,7 @@ cnt_hit(struct sess *sp)
 
        /* Experimental. Reduce health parameter of backend towards zero
         * if it has been more than a minute since it was checked. */
+       CHECK_OBJ_NOTNULL(sp->backend, BACKEND_MAGIC);
        time_diff = TIM_mono() - sp->backend->last_check;
        minutes = time_diff / 60;
        if (minutes > sp->backend->minute_limit) {
@@ -759,7 +762,9 @@ cnt_recv(struct sess *sp)
        http_DoConnection(sp);
 
        /* By default we use the first backend */
+       AZ(sp->backend);
        sp->backend = sp->vcl->backend[0];
+       CHECK_OBJ_NOTNULL(sp->backend, BACKEND_MAGIC);
 
        /* XXX: Handle TRACE & OPTIONS of Max-Forwards = 0 */
 
@@ -817,6 +822,8 @@ CNT_Session(struct sess *sp)
                        CHECK_OBJ(w->nobj, OBJECT_MAGIC);
                if (w->nobjhead != NULL)
                        CHECK_OBJ(w->nobjhead, OBJHEAD_MAGIC);
+               if (sp->backend != NULL)
+                       CHECK_OBJ(sp->backend, BACKEND_MAGIC);
 
                switch (sp->step) {
 #define STEP(l,u) case STP_##u: done = cnt_##l(sp); break;