Merge branch '2790-properly-handle-oversized-messages-in-isc_nm_send-v9_16' into 'v9_16'

Handle `UV_EMSGSIZE` in the uv_udp_send() callback (v9.16)

See merge request isc-projects/bind9!5226

CHANGES

@@ -1,3 +1,9 @@
+5664.	[func]		Handle a UDP sending error on UDP messages larger
+			than the path MTU; in such a case an empty response is
+			sent back with the TC (TrunCated) bit set. Re-enable
+			setting the DF (Don't Fragment) flag on outgoing
+			UDP sockets. [GL #2790]
+
 5662.	[bug]		Views with recursion disabled are now configured with a
 			default cache size of 2 MB, unless "max-cache-size" is
 			explicitly set. This prevents cache RBT hash tables from

doc/notes/notes-current.rst

@@ -34,7 +34,10 @@ Removed Features
 Feature Changes
 ~~~~~~~~~~~~~~~
 
-- None.
+- IP fragmentation on outgoing UDP sockets has been disabled.  Errors from
+  sending DNS messages larger than the specified path MTU are properly handled;
+  ``named`` now sends back empty DNS messages with the TC (TrunCated) bit set,
+  forcing the DNS client to fall back to TCP.  :gl:`#2790`
 
 Bug Fixes
 ~~~~~~~~~

lib/isc/netmgr/netmgr.c

@@ -2944,22 +2944,14 @@ isc__nm_socket_dontfrag(uv_os_sock_t fd, sa_family_t sa_family) {
 	 */
 	if (sa_family == AF_INET6) {
 #if defined(IPV6_DONTFRAG)
-		if (setsockopt_off(fd, IPPROTO_IPV6, IPV6_DONTFRAG) == -1) {
+		if (setsockopt_on(fd, IPPROTO_IPV6, IPV6_DONTFRAG) == -1) {
 			return (ISC_R_FAILURE);
 		} else {
 			return (ISC_R_SUCCESS);
 		}
-#elif defined(IPV6_MTU_DISCOVER) && defined(IP_PMTUDISC_OMIT)
+#elif defined(IPV6_MTU_DISCOVER)
 		if (setsockopt(fd, IPPROTO_IPV6, IPV6_MTU_DISCOVER,
-			       &(int){ IP_PMTUDISC_OMIT }, sizeof(int)) == -1)
-		{
-			return (ISC_R_FAILURE);
-		} else {
-			return (ISC_R_SUCCESS);
-		}
-#elif defined(IPV6_MTU_DISCOVER) && defined(IP_PMTUDISC_DONT)
-		if (setsockopt(fd, IPPROTO_IPV6, IPV6_MTU_DISCOVER,
-			       &(int){ IP_PMTUDISC_DONT }, sizeof(int)) == -1)
+			       &(int){ IP_PMTUDISC_DO }, sizeof(int)) == -1)
 		{
 			return (ISC_R_FAILURE);
 		} else {
@@ -2970,22 +2962,14 @@ isc__nm_socket_dontfrag(uv_os_sock_t fd, sa_family_t sa_family) {
 #endif
 	} else if (sa_family == AF_INET) {
 #if defined(IP_DONTFRAG)
-		if (setsockopt_off(fd, IPPROTO_IP, IP_DONTFRAG) == -1) {
+		if (setsockopt_on(fd, IPPROTO_IP, IP_DONTFRAG) == -1) {
 			return (ISC_R_FAILURE);
 		} else {
 			return (ISC_R_SUCCESS);
 		}
-#elif defined(IP_MTU_DISCOVER) && defined(IP_PMTUDISC_OMIT)
+#elif defined(IP_MTU_DISCOVER)
 		if (setsockopt(fd, IPPROTO_IP, IP_MTU_DISCOVER,
-			       &(int){ IP_PMTUDISC_OMIT }, sizeof(int)) == -1)
-		{
-			return (ISC_R_FAILURE);
-		} else {
-			return (ISC_R_SUCCESS);
-		}
-#elif defined(IP_MTU_DISCOVER) && defined(IP_PMTUDISC_DONT)
-		if (setsockopt(fd, IPPROTO_IP, IP_MTU_DISCOVER,
-			       &(int){ IP_PMTUDISC_DONT }, sizeof(int)) == -1)
+			       &(int){ IP_PMTUDISC_DO }, sizeof(int)) == -1)
 		{
 			return (ISC_R_FAILURE);
 		} else {

lib/isc/netmgr/uverr2result.c

@@ -85,6 +85,8 @@ isc___nm_uverr2result(int uverr, bool dolog, const char *file,
 		return (ISC_R_CANCELED);
 	case UV_EOF:
 		return (ISC_R_EOF);
+	case UV_EMSGSIZE:
+		return (ISC_R_MAXSIZE);
 	default:
 		if (dolog) {
 			UNEXPECTED_ERROR(

lib/ns/client.c

@@ -279,13 +279,33 @@ client_senddone(isc_nmhandle_t *handle, isc_result_t result, void *cbarg) {
 	REQUIRE(client->sendhandle == handle);
 
 	CTRACE("senddone");
+
+	/*
+	 * Set sendhandle to NULL, but don't detach it immediately, in
+	 * case we need to retry the send. If we do resend, then
+	 * sendhandle will be reattached. Whether or not we resend,
+	 * we will then detach the handle from *this* send by detaching
+	 * 'handle' directly below.
+	 */
+	client->sendhandle = NULL;
+
 	if (result != ISC_R_SUCCESS) {
-		ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
-			      NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(3),
-			      "send failed: %s", isc_result_totext(result));
+		if (!TCP_CLIENT(client) && result == ISC_R_MAXSIZE) {
+			ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
+				      NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(3),
+				      "send exceeded maximum size: truncating");
+			client->query.attributes &= ~NS_QUERYATTR_ANSWERED;
+			client->rcode_override = dns_rcode_noerror;
+			ns_client_error(client, ISC_R_MAXSIZE);
+		} else {
+			ns_client_log(client, DNS_LOGCATEGORY_SECURITY,
+				      NS_LOGMODULE_CLIENT, ISC_LOG_DEBUG(3),
+				      "send failed: %s",
+				      isc_result_totext(result));
+		}
 	}
 
-	isc_nmhandle_detach(&client->sendhandle);
+	isc_nmhandle_detach(&handle);
 }
 
 static void
@@ -721,8 +741,9 @@ ns_client_dropport(in_port_t port) {
 
 void
 ns_client_error(ns_client_t *client, isc_result_t result) {
+	dns_message_t *message = NULL;
 	dns_rcode_t rcode;
-	dns_message_t *message;
+	bool trunc = false;
 
 	REQUIRE(NS_CLIENT_VALID(client));
 
@@ -736,6 +757,10 @@ ns_client_error(ns_client_t *client, isc_result_t result) {
 		rcode = (dns_rcode_t)(client->rcode_override & 0xfff);
 	}
 
+	if (result == ISC_R_MAXSIZE) {
+		trunc = true;
+	}
+
 #if NS_CLIENT_DROPPORT
 	/*
 	 * Don't send FORMERR to ports on the drop port list.
@@ -832,7 +857,11 @@ ns_client_error(ns_client_t *client, isc_result_t result) {
 			return;
 		}
 	}
+
 	message->rcode = rcode;
+	if (trunc) {
+		message->flags |= DNS_MESSAGEFLAG_TC;
+	}
 
 	if (rcode == dns_rcode_formerr) {
 		/*