The persistent stevedore is not even able to allocate an object without

and objcore, so return NULL if asked to.

Add a fallback in cnt_error() to attempt the Transient storage if we
hit this case.  There already is a similar retry for the regular
object allocation attempt.

Fixes	#915

bin/varnishd/cache_center.c

@@ -415,6 +415,9 @@ cnt_error(struct sess *sp)
 		EXP_Clr(&w->exp);
 		sp->obj = STV_NewObject(sp, NULL, 1024, &w->exp,
 		     params->http_max_hdr);
+		if (sp->obj == NULL) 
+			sp->obj = STV_NewObject(sp, TRANSIENT_STORAGE,
+			    1024, &w->exp, params->http_max_hdr);
 		if (sp->obj == NULL) {
 			sp->doclose = "Out of objects";
 			sp->step = STP_DONE;

bin/varnishd/storage_persistent.c

@@ -474,6 +474,8 @@ smp_allocobj(struct stevedore *stv, struct sess *sp, unsigned ltot,
 	struct objcore *oc;
 	unsigned objidx;
 
+	if (sp->objcore == NULL)
+		return (NULL);		/* from cnt_error */
 	CAST_OBJ_NOTNULL(sc, stv->priv, SMP_SC_MAGIC);
 	AN(sp->objcore);
 	AN(sp->wrk->exp.ttl > 0.);

bin/varnishtest/tests/r00915.vtc

+varnishtest "error object allocation with persistent"
+
+server s1 {
+	rxreq 
+	txresp 
+} -start
+
+shell "rm -f ${tmpdir}/_.per"
+
+varnish v1 \
+	-arg "-pdiag_bitmap=0x20000" \
+	-storage "-spersistent,${tmpdir}/_.per,10m" \
+	-vcl+backend { 
+
+	sub vcl_fetch {
+		error(751);
+	}
+} -start 
+
+client c1 {
+	txreq -url "/"
+	rxresp
+	expect resp.status == 751
+} -run