res_pjsip: Failover when server is not available

Previously Asterisk did not properly failover to the next resolved DNS
address when a endpoint could not be reached. With this patch, and while
using res_pjsip, SIP requests (both in/out of dialog) now attempt to use
the next address in the list of resolved addresses until a proper response
is received or no more addresses are left.

ASTERISK-25076 #close
Reported by: Joshua Colp

Change-Id: Ief14f4ebd82474881f72f4538f4577f30af2a764

include/asterisk/res_pjsip.h

@@ -2044,4 +2044,13 @@ unsigned int ast_sip_get_max_initial_qualify_time(void);
 const char *ast_sip_get_contact_status_label(const enum ast_sip_contact_status_type status);
 const char *ast_sip_get_contact_short_status_label(const enum ast_sip_contact_status_type status);
 
+/*!
+ * \brief Set a request to use the next value in the list of resolved addresses.
+ *
+ * \param tdata the tx data from the original request
+ * \retval 0 No more addresses to try
+ * \retval 1 The request was successfully re-intialized
+ */
+int ast_sip_failover_request(pjsip_tx_data *tdata);
+
 #endif /* _RES_PJSIP_H */

res/res_pjsip.c

@@ -3135,19 +3135,88 @@ static pj_status_t endpt_send_request(struct ast_sip_endpoint *endpoint,
 	return ret_val;
 }
 
+int ast_sip_failover_request(pjsip_tx_data *tdata)
+{
+	pjsip_via_hdr *via;
+
+	if (tdata->dest_info.cur_addr == tdata->dest_info.addr.count - 1) {
+		/* No more addresses to try */
+		return 0;
+	}
+
+	/* Try next address */
+	++tdata->dest_info.cur_addr;
+
+	via = (pjsip_via_hdr*)pjsip_msg_find_hdr(tdata->msg, PJSIP_H_VIA, NULL);
+	via->branch_param.slen = 0;
+
+	pjsip_tx_data_invalidate_msg(tdata);
+
+	return 1;
+}
+
+static void send_request_cb(void *token, pjsip_event *e);
+
+static int check_request_status(struct send_request_data *req_data, pjsip_event *e)
+{
+	struct ast_sip_endpoint *endpoint;
+	pjsip_transaction *tsx;
+	pjsip_tx_data *tdata;
+	int res = 0;
+
+	if (!(endpoint = ao2_bump(req_data->endpoint))) {
+		return 0;
+	}
+
+	tsx = e->body.tsx_state.tsx;
+
+	switch (tsx->status_code) {
+	case 401:
+	case 407:
+		/* Resend the request with a challenge response if we are challenged. */
+		res = ++req_data->challenge_count < MAX_RX_CHALLENGES /* Not in a challenge loop */
+			&& !ast_sip_create_request_with_auth(&endpoint->outbound_auths,
+				e->body.tsx_state.src.rdata, tsx->last_tx, &tdata);
+		break;
+	case 408:
+	case 503:
+		if ((res = ast_sip_failover_request(tsx->last_tx))) {
+			tdata = tsx->last_tx;
+			/*
+			 * Bump the ref since it will be on a new transaction and
+			 * we don't want it to go away along with the old transaction.
+			 */
+			pjsip_tx_data_add_ref(tdata);
+		}
+		break;
+	}
+
+	if (res) {
+		res = endpt_send_request(endpoint, tdata, -1,
+					 req_data, send_request_cb) == PJ_SUCCESS;
+	}
+
+	ao2_ref(endpoint, -1);
+	return res;
+}
+
 static void send_request_cb(void *token, pjsip_event *e)
 {
 	struct send_request_data *req_data = token;
-	pjsip_transaction *tsx;
 	pjsip_rx_data *challenge;
-	pjsip_tx_data *tdata;
 	struct ast_sip_supplement *supplement;
-	struct ast_sip_endpoint *endpoint;
-	int res;
 
 	switch(e->body.tsx_state.type) {
 	case PJSIP_EVENT_TRANSPORT_ERROR:
 	case PJSIP_EVENT_TIMER:
+		/*
+		 * Check the request status on transport error or timeout. A transport
+		 * error can occur when a TCP socket closes and that can be the result
+		 * of a 503. Also we may need to failover on a timeout (408).
+		 */
+		if (check_request_status(req_data, e)) {
+			return;
+		}
 		break;
 	case PJSIP_EVENT_RX_MSG:
 		challenge = e->body.tsx_state.src.rdata;
@@ -3166,20 +3235,9 @@ static void send_request_cb(void *token, pjsip_event *e)
 		}
 		AST_RWLIST_UNLOCK(&supplements);
 
-		/* Resend the request with a challenge response if we are challenged. */
-		tsx = e->body.tsx_state.tsx;
-		endpoint = ao2_bump(req_data->endpoint);
-		res = (tsx->status_code == 401 || tsx->status_code == 407)
-			&& endpoint
-			&& ++req_data->challenge_count < MAX_RX_CHALLENGES /* Not in a challenge loop */
-			&& !ast_sip_create_request_with_auth(&endpoint->outbound_auths,
-				challenge, tsx->last_tx, &tdata)
-			&& endpt_send_request(endpoint, tdata, -1, req_data, send_request_cb)
-				== PJ_SUCCESS;
-		ao2_cleanup(endpoint);
-		if (res) {
+		if (check_request_status(req_data, e)) {
 			/*
-			 * Request with challenge response sent.
+			 * Request with challenge response or failover sent.
 			 * Passed our req_data ref to the new request.
 			 */
 			return;

res/res_pjsip_session.c

@@ -2267,6 +2267,29 @@ static int session_end(struct ast_sip_session *session)
 	return 0;
 }
 
+static int check_request_status(pjsip_inv_session *inv, pjsip_event *e)
+{
+	struct ast_sip_session *session = inv->mod_data[session_module.id];
+	pjsip_transaction *tsx = e->body.tsx_state.tsx;
+
+	if (tsx->status_code != 503 && tsx->status_code != 408) {
+		return 0;
+	}
+
+	if (!ast_sip_failover_request(tsx->last_tx)) {
+		return 0;
+	}
+
+	pjsip_inv_uac_restart(inv, PJ_FALSE);
+	/*
+	 * Bump the ref since it will be on a new transaction and
+	 * we don't want it to go away along with the old transaction.
+	 */
+	pjsip_tx_data_add_ref(tsx->last_tx);
+	ast_sip_session_send_request(session, tsx->last_tx);
+	return 1;
+}
+
 static void session_inv_on_state_changed(pjsip_inv_session *inv, pjsip_event *e)
 {
 	struct ast_sip_session *session = inv->mod_data[session_module.id];
@@ -2299,11 +2322,20 @@ static void session_inv_on_state_changed(pjsip_inv_session *inv, pjsip_event *e)
 			handle_outgoing(session, e->body.tsx_state.src.tdata);
 			break;
 		case PJSIP_EVENT_RX_MSG:
-			handle_incoming(session, e->body.tsx_state.src.rdata, type,
-					AST_SIP_SESSION_BEFORE_MEDIA);
+			if (!check_request_status(inv, e)) {
+				handle_incoming(session, e->body.tsx_state.src.rdata, type,
+						AST_SIP_SESSION_BEFORE_MEDIA);
+			}
 			break;
 		case PJSIP_EVENT_TRANSPORT_ERROR:
 		case PJSIP_EVENT_TIMER:
+			/*
+			 * Check the request status on transport error or timeout. A transport
+			 * error can occur when a TCP socket closes and that can be the result
+			 * of a 503. Also we may need to failover on a timeout (408).
+			 */
+			check_request_status(inv, e);
+			break;
 		case PJSIP_EVENT_USER:
 		case PJSIP_EVENT_UNKNOWN:
 		case PJSIP_EVENT_TSX_STATE: