Merge branch '3128-dig-does-not-recover-from-a-isc_nm_udpconnect-failure' into 'main'

After dig request errors, try to use other servers when they exist

Closes #3128

See merge request isc-projects/bind9!5967

CHANGES

@@ -1,3 +1,7 @@
+5832.	[bug]		When timing-out or having other types of socket errors
+			during a query, dig wasn't trying to perform the lookup
+			using other servers, in case they exist. [GL #3128]
+
 5831.	[bug]		When resending a UDP request in the result of a timeout,
 			the recv_done() function in dighost.c was prepending
 			the new query into the loookup's queries list instead

bin/dig/dighost.c

@@ -3280,9 +3280,9 @@ tcp_connected(isc_nmhandle_t *handle, isc_result_t eresult, void *arg) {
 		}
 
 		if (l->retries > 1) {
+			l->retries--;
 			debug("making new TCP request, %d tries left",
 			      l->retries);
-			l->retries--;
 			requeue_lookup(l, true);
 			next = NULL;
 		} else if ((l->current_query != NULL) &&
@@ -3622,49 +3622,118 @@ recv_done(isc_nmhandle_t *handle, isc_result_t eresult, isc_region_t *region,
 		TIME_NOW(&query->time_recv);
 	}
 
-	if (eresult == ISC_R_TIMEDOUT && !l->tcp_mode && l->retries > 1) {
-		dig_query_t *newq = NULL;
-
-		l->retries--;
-		debug("making new UDP request, %d tries left", l->retries);
-		newq = new_query(l, query->servname, query->userarg);
-
-		ISC_LIST_INSERTAFTER(l->q, query, newq, link);
-		if (l->current_query == query) {
-			query_detach(&l->current_query);
-		}
-		start_udp(newq);
-		goto detach_query;
-	}
-
 	if ((!l->pending && !l->ns_search_only) || atomic_load(&cancel_now)) {
 		debug("no longer pending.  Got %s", isc_result_totext(eresult));
 
 		goto next_lookup;
 	}
 
-	if (eresult != ISC_R_SUCCESS) {
-		char sockstr[ISC_SOCKADDR_FORMATSIZE];
-		isc_sockaddr_format(&query->sockaddr, sockstr, sizeof(sockstr));
-
 	if (eresult == ISC_R_TIMEDOUT) {
-			if (l->retries > 1) {
+		if (l->retries > 1 && !l->tcp_mode) {
+			dig_query_t *newq = NULL;
+
+			/*
+			 * For UDP, insert a copy of the current query just
+			 * after itself in the list, and start it to retry the
+			 * request.
+			 */
+			newq = new_query(l, query->servname, query->userarg);
+			ISC_LIST_INSERTAFTER(l->q, query, newq, link);
+			if (l->current_query == query) {
+				query_detach(&l->current_query);
+			}
+			if (l->current_query == NULL) {
+				l->retries--;
+				debug("making new UDP request, %d tries left",
+				      l->retries);
+				start_udp(newq);
+			}
+
+			goto detach_query;
+		} else if (l->retries > 1 && l->tcp_mode) {
+			/*
+			 * For TCP, we have to requeue the whole lookup, see
+			 * the comments above the start_tcp() function.
+			 */
+			l->retries--;
 			debug("making new TCP request, %d tries left",
 			      l->retries);
-				l->retries--;
 			requeue_lookup(l, true);
+
+			if (keep != NULL) {
+				isc_nmhandle_detach(&keep);
+			}
+
+			goto cancel_lookup;
 		} else {
+			dig_query_t *next = ISC_LIST_NEXT(query, link);
+
+			/*
+			 * No retries left, go to the next query, if there is
+			 * one.
+			 */
+			if (next != NULL) {
+				if (l->current_query == query) {
+					query_detach(&l->current_query);
+				}
+				if (l->current_query == NULL) {
+					debug("starting next query %p", next);
+					if (l->tcp_mode) {
+						start_tcp(next);
+					} else {
+						start_udp(next);
+					}
+				}
+				goto detach_query;
+			}
+
+			/*
+			 * Otherwise, print the cmdline and an error message,
+			 * and cancel the lookup.
+			 */
 			printf("%s", l->cmdline);
 			dighost_error("connection timed out; "
 				      "no servers could be reached\n");
 			if (exitcode < 9) {
 				exitcode = 9;
 			}
+
+			if (keep != NULL) {
+				isc_nmhandle_detach(&keep);
 			}
+
+			goto cancel_lookup;
+		}
+	} else if (eresult != ISC_R_SUCCESS) {
+		dig_query_t *next = ISC_LIST_NEXT(query, link);
+		char sockstr[ISC_SOCKADDR_FORMATSIZE];
+		isc_sockaddr_format(&query->sockaddr, sockstr, sizeof(sockstr));
+
+		/*
+		 * There was a communication error with the current query,
+		 * go to the next query, if there is one.
+		 */
+		if (next != NULL) {
+			if (l->current_query == query) {
+				query_detach(&l->current_query);
+			}
+			if (l->current_query == NULL) {
+				debug("starting next query %p", next);
+				if (l->tcp_mode) {
+					start_tcp(next);
 				} else {
-			dighost_error("communications error to %s: %s\n",
+					start_udp(next);
-				      sockstr, isc_result_totext(eresult));
 				}
+			}
+			goto detach_query;
+		}
+
+		/*
+		 * Otherwise, print an error message and cancel the
+		 * lookup.
+		 */
+		dighost_error("communications error to %s: %s\n", sockstr,
+			      isc_result_totext(eresult));
 
 		if (keep != NULL) {
 			isc_nmhandle_detach(&keep);

bin/tests/system/digdelv/tests.sh

@@ -1014,6 +1014,24 @@ if [ -x "$DIG" ] ; then
   grep "status: SERVFAIL" dig.out.test$n > /dev/null || ret=1
   if [ $ret -ne 0 ]; then echo_i "failed"; fi
   status=$((status+ret))
+
+  n=$((n+1))
+  echo_i "check that dig tries the next server after a connection error ($n)"
+  ret=0
+  dig_with_opts -d @10.53.0.99 @10.53.0.3 a.example > dig.out.test$n 2>&1 || ret=1
+  grep "status: NOERROR" dig.out.test$n > /dev/null || ret=1
+  if [ $ret -ne 0 ]; then echo_i "failed"; fi
+  status=$((status+ret))
+
+  n=$((n+1))
+  echo_i "check that dig tries the next server after timeouts ($n)"
+  # Ask ans4 to not respond to queries
+  echo "//" | sendcmd 10.53.0.4
+  ret=0
+  dig_with_opts -d @10.53.0.4 @10.53.0.3 a.example > dig.out.test$n 2>&1 || ret=1
+  grep "status: NOERROR" dig.out.test$n > /dev/null || ret=1
+  if [ $ret -ne 0 ]; then echo_i "failed"; fi
+  status=$((status+ret))
 else
   echo_i "$DIG is needed, so skipping these dig tests"
 fi