[v9_9] Prevent dig INSIST failures and hangs in some failure modes

4756.	[bug]		Interrupting dig could lead to an INSIST failure after
			certain errors were encountered while querying a host
			whose name resolved to more than one address.  Change
			4537 increased the odds of triggering this issue by
			causing dig to hang indefinitely when certain error
			paths were evaluated.  dig now also retries TCP queries
			(once) if the server gracefully closes the connection
			before sending a response. [RT #42832, #45159]

(cherry picked from commit 1ddb1cd440)

CHANGES

@@ -1,3 +1,12 @@
+4756.	[bug]		Interrupting dig could lead to an INSIST failure after
+			certain errors were encountered while querying a host
+			whose name resolved to more than one address.  Change
+			4537 increased the odds of triggering this issue by
+			causing dig to hang indefinitely when certain error
+			paths were evaluated.  dig now also retries TCP queries
+			(once) if the server gracefully closes the connection
+			before sending a response. [RT #42832, #45159]
+
 4754.	[bug]		dns_zone_setview needs a two stage commit to properly
 			handle errors. [RT #45841]
 

bin/dig/dighost.c

@@ -837,6 +837,7 @@ make_empty_lookup(void) {
 	looknew->new_search = ISC_FALSE;
 	looknew->done_as_is = ISC_FALSE;
 	looknew->need_search = ISC_FALSE;
+	looknew->eoferr = 0;
 	dns_fixedname_init(&looknew->fdomain);
 	ISC_LINK_INIT(looknew, link);
 	ISC_LIST_INIT(looknew->q);
@@ -917,6 +918,7 @@ clone_lookup(dig_lookup_t *lookold, isc_boolean_t servers) {
 	looknew->tsigctx = NULL;
 	looknew->need_search = lookold->need_search;
 	looknew->done_as_is = lookold->done_as_is;
+	looknew->eoferr = lookold->eoferr;
 
 	dns_name_copy(dns_fixedname_name(&lookold->fdomain),
 		      dns_fixedname_name(&looknew->fdomain), NULL);
@@ -2824,7 +2826,7 @@ tcp_length_done(isc_task_t *task, isc_event_t *event) {
 	isc_buffer_t *b = NULL;
 	isc_result_t result;
 	dig_query_t *query = NULL;
-	dig_lookup_t *l;
+	dig_lookup_t *l, *n;
 	isc_uint16_t length;
 
 	REQUIRE(event->ev_type == ISC_SOCKEVENT_RECVDONE);
@@ -2859,13 +2861,20 @@ tcp_length_done(isc_task_t *task, isc_event_t *event) {
 				    sizeof(sockstr));
 		printf(";; communications error to %s: %s\n",
 		       sockstr, isc_result_totext(sevent->result));
+		if (keep != NULL)
+			isc_socket_detach(&keep);
 		l = query->lookup;
 		isc_socket_detach(&query->sock);
 		sockcount--;
 		debug("sockcount=%d", sockcount);
 		INSIST(sockcount >= 0);
+		if (sevent->result == ISC_R_EOF && l->eoferr == 0U) {
+			n = requeue_lookup(l, ISC_TRUE);
+			n->eoferr++;
+		}
 		isc_event_free(&event);
 		clear_query(query);
+		cancel_lookup(l);
 		check_next_lookup(l);
 		UNLOCK_LOOKUP;
 		return;
@@ -3273,13 +3282,20 @@ recv_done(isc_task_t *task, isc_event_t *event) {
 		} else {
 			printf(";; communications error: %s\n",
 			       isc_result_totext(sevent->result));
+			if (keep != NULL)
+				isc_socket_detach(&keep);
 			isc_socket_detach(&query->sock);
 			sockcount--;
 			debug("sockcount=%d", sockcount);
 			INSIST(sockcount >= 0);
 		}
+		if (sevent->result == ISC_R_EOF && l->eoferr == 0U) {
+			n = requeue_lookup(l, ISC_TRUE);
+			n->eoferr++;
+		}
 		isc_event_free(&event);
 		clear_query(query);
+		cancel_lookup(l);
 		check_next_lookup(l);
 		UNLOCK_LOOKUP;
 		return;
@@ -3341,6 +3357,7 @@ recv_done(isc_task_t *task, isc_event_t *event) {
 			if (fail) {
 				isc_event_free(&event);
 				clear_query(query);
+				cancel_lookup(l);
 				check_next_lookup(l);
 				UNLOCK_LOOKUP;
 				return;
@@ -3452,6 +3469,7 @@ recv_done(isc_task_t *task, isc_event_t *event) {
 			if (l->tcp_mode) {
 				isc_event_free(&event);
 				clear_query(query);
+				cancel_lookup(l);
 				check_next_lookup(l);
 				UNLOCK_LOOKUP;
 				return;

bin/dig/include/dig/dig.h

@@ -183,6 +183,7 @@ isc_boolean_t	sigchase;
 	isc_buffer_t *querysig;
 	isc_uint32_t msgcounter;
 	dns_fixedname_t fdomain;
+	unsigned int eoferr;
 };
 
 /*% The dig_query structure */