dialog: add options_ping_max_retries parameter for OPTIONS ping retries

modules/dialog/README.md

@@ -446,6 +446,32 @@ modparam("dialog", "options_ping_interval", 20)
 ```
 
 
+#### options_ping_max_retries (integer)
+
+
+The number of additional in-dialog OPTIONS pings to send when the
+previous OPTIONS transaction times out, before declaring the dialog leg
+unreachable. Retries are sent immediately after each transaction timeout,
+without waiting for the next `options_ping_interval` tick. The retry counter
+is maintained independently for each dialog leg and is reset by any final
+response other than 408 or 481. A 481 response is considered definitive and
+is not retried.
+
+The maximum time needed to detect an unreachable peer is approximately
+`(options_ping_max_retries + 1) * fr_timeout`, where `fr_timeout` is configured
+by the "tm" module.
+
+
+*Default value is "0" (no additional retries).*
+
+
+```opensips title="Set options_ping_max_retries parameter"
+...
+modparam("dialog", "options_ping_max_retries", 2)
+...
+```
+
+
 #### reinvite_ping_interval (integer)
 
 

modules/dialog/dialog.c

@@ -72,6 +72,7 @@ static char* profiles_nv_s = NULL;
 int dlg_bulk_del_no = 1; /* delete one by one */
 int seq_match_mode = SEQ_MATCH_FALLBACK;
 int options_ping_interval = 30;      /* seconds */
+int options_ping_max_retries = 0;    /* extra OPTIONS pings to send on timeout before declaring leg dead */
 int reinvite_ping_interval = 300;    /* seconds */
 int dlg_del_delay = 0;               /* in seconds, default off */
 str dlg_extra_hdrs = {NULL,0};
@@ -310,6 +311,7 @@ static const param_export_t mod_params[]={
 	{ "rr_param",              STR_PARAM, &rr_param.s               },
 	{ "default_timeout",       INT_PARAM, &default_timeout          },
 	{ "options_ping_interval", INT_PARAM, &options_ping_interval    },
+	{ "options_ping_max_retries", INT_PARAM, &options_ping_max_retries },
 	{ "reinvite_ping_interval",INT_PARAM, &reinvite_ping_interval   },
 	{ "delete_delay",          INT_PARAM, &dlg_del_delay            },
 	{ "dlg_extra_hdrs",        STR_PARAM, &dlg_extra_hdrs.s         },
@@ -805,6 +807,11 @@ static int mod_init(void)
 		return -1;
 	}
 
+	if (options_ping_max_retries < 0) {
+		LM_ERR("Negative options_ping_max_retries not accepted!!\n");
+		return -1;
+	}
+
 	/* update the len of the extra headers */
 	if (dlg_extra_hdrs.s)
 		dlg_extra_hdrs.len = strlen(dlg_extra_hdrs.s);

modules/dialog/dlg_hash.h

@@ -113,6 +113,7 @@ struct dlg_leg {
 	struct dlg_leg_cseq_map *cseq_maps; /* used when translating ACKs */
 	char reply_received;
 	char reinvite_confirmed;
+	unsigned char ping_retries;	/* failed OPTIONS pings since last success */
 	const struct socket_info *bind_addr;
 };
 

modules/dialog/dlg_timer.c

@@ -38,6 +38,13 @@ struct dlg_reinvite_ping_timer *reinvite_ping_timer=0;
 str options_str=str_init("OPTIONS");
 str invite_str=str_init("INVITE");
 
+extern int options_ping_max_retries;
+
+/* forward declarations for OPTIONS ping retry from inside the reply callback */
+void reply_from_caller(struct cell* t, int type, struct tmcb_params* ps);
+void reply_from_callee(struct cell* t, int type, struct tmcb_params* ps);
+void unref_dlg_cb(void *dlg);
+
 /* for the dlg timer, there are 3 possible states :
  * prev=next=0 -> dialog not in timer list
  * prev=0 -> dialog expired
@@ -765,6 +772,36 @@ int dlg_handle_seq_reply(struct dlg_cell *dlg, struct sip_msg* rpl,
 	LM_DBG("Status Code received =  [%d]\n", statuscode);
 
 	if (rpl == FAKED_REPLY || statuscode == 408) {
+		/* OPTIONS ping timeout: optionally retry immediately before giving up.
+		 * Re-INVITE ping path is unchanged. */
+		if (!is_reinvite_rpl &&
+		    dlg->legs[leg].ping_retries < (unsigned char)options_ping_max_retries) {
+			dlg_request_callback *cb;
+
+			dlg->legs[leg].ping_retries++;
+			LM_INFO("OPTIONS ping timeout on %s leg, retry %u/%d, "
+			        "ci: [%.*s]\n",
+			        leg == DLG_CALLER_LEG ? "caller" : "callee",
+			        dlg->legs[leg].ping_retries, options_ping_max_retries,
+			        dlg->callid.len, dlg->callid.s);
+
+			/* clear the gate so a new in-flight ping can be tracked */
+			*ping_status = DLG_PING_SUCCESS;
+
+			cb = (leg == DLG_CALLER_LEG) ? reply_from_caller : reply_from_callee;
+			ref_dlg(dlg, 1);
+			if (send_leg_msg(dlg, &options_str, other_leg(dlg, leg), leg,
+			        0, 0, cb, dlg, unref_dlg_cb,
+			        &dlg->legs[leg].reply_received) < 0) {
+				LM_ERR("failed to send OPTIONS retry on %s leg\n",
+				       leg == DLG_CALLER_LEG ? "caller" : "callee");
+				unref_dlg(dlg, 1);
+				/* leave ping_retries as-is; the next interval-driven ping
+				 * will continue from the current retry count */
+			}
+			return 0;
+		}
+
 		/* timeout occurred, nothing else to do now
 		 * next time timer fires, it will detect ping reply was not received
 		 */
@@ -788,6 +825,8 @@ int dlg_handle_seq_reply(struct dlg_cell *dlg, struct sip_msg* rpl,
 	}
 
 	*ping_status = DLG_PING_SUCCESS;
+	if (!is_reinvite_rpl)
+		dlg->legs[leg].ping_retries = 0;
 	if (is_reinvite_rpl && statuscode < 300 && send_leg_msg(dlg, &ack,
 			other_leg(dlg, leg), leg, NULL, NULL, NULL, NULL, NULL, NULL) < 0)
 		LM_ERR("cannot send ACK message!\n");