Merge branch 'bridge_tcp_user_timeout' of git://github.com/abiliojr/mosquitto into abiliojr-bridge_tcp_user_timeout

man/mosquitto.conf.5.xml

@@ -1697,7 +1697,6 @@ openssl dhparam -out dhparam.pem 2048</programlisting>
 						<replaceable>mqttv311</replaceable>.</para>
 				</listitem>
 			</varlistentry>
-
 			<varlistentry>
 				<term><option>bridge_tcp_keepalive</option> <replaceable>idle</replaceable> <replaceable>interval</replaceable> <replaceable>counter</replaceable></term>
 				<listitem>
@@ -1714,6 +1713,29 @@ openssl dhparam -out dhparam.pem 2048</programlisting>
 					<para>Disabled by default.</para>
 				</listitem>
 			</varlistentry>
+			<varlistentry>
+				<term><option>bridge_tcp_user_timeout</option> <replaceable>timeout</replaceable></term>
+				<listitem>
+					<para>
+						It specifies the maximum amount of time (in milliseconds) that
+						transmitted data may remain unacknowledged at TCP level.
+						Popular Linux distributions seem to set this time to "up to 20
+						minutes with system defaults" (from Linux tcp man page).
+						The default time is related to tcp_retries2.
+						Reducing this value helps detecting dropped connections faster.
+					</para>
+					<para>
+						Be aware that when used in combination with TCP Keepalive, it might
+						change the latter's behavior.
+					</para>
+					<para>
+						You can find more details about this setting at:
+						<ulink url="https://blog.cloudflare.com/when-tcp-sockets-refuse-to-die/"/>
+					<para>
+						Only available on Linux.
+					</para>
+				</listitem>
+			</varlistentry>
 			<varlistentry>
 				<term><option>cleansession</option> [ true | false ]</term>
 				<listitem>

mosquitto.conf

@@ -723,6 +723,16 @@
 # Disabled by default.
 #bridge_tcp_keepalive <idle> <interval> <counter>
 
+# Change the maximum amount of time (in milliseconds) that transmitted data
+# may remain unacknowledged at TCP level.
+# Popular Linux distributions seem to set this time to "up to 20 minutes with
+# system defaults" (from Linux tcp man page). Reducing this value helps
+# detecting dropped connections faster.
+# When used together with TCP Keepalive, it might change it's behavior.
+# More details at: https://blog.cloudflare.com/when-tcp-sockets-refuse-to-die/
+# Only available on Linux.
+#bridge_tcp_user_timeout <ms>
+
 # Set the clientid to use on the local broker. If not defined, this defaults to
 # 'local.<clientid>'. If you are bridging a broker to itself, it is important
 # that local_clientid and clientid do not match.

src/bridge.c

@@ -198,6 +198,19 @@ static int bridge__set_tcp_keepalive(struct mosquitto *context)
 	return MOSQ_ERR_SUCCESS;
 }
 
+#ifdef WITH_TCP_USER_TIMEOUT
+static int bridge__set_tcp_user_timeout(struct mosquitto *context) {
+	int timeout = context->bridge->tcp_user_timeout;
+	if(timeout >= 0) {
+		if(setsockopt(context->sock, IPPROTO_TCP, TCP_USER_TIMEOUT, (char *)&timeout, sizeof(timeout))) {
+			return MOSQ_ERR_UNKNOWN;
+		}
+	}
+
+	return MOSQ_ERR_SUCCESS;
+}
+#endif
+
 #if defined(__GLIBC__) && defined(WITH_ADNS)
 static int bridge__connect_step1(struct mosquitto *context)
 {
@@ -373,6 +386,9 @@ int bridge__connect_step3(struct mosquitto *context)
 	}
 
 	if (bridge__set_tcp_keepalive(context) != MOSQ_ERR_SUCCESS) return MOSQ_ERR_UNKNOWN;
+#ifdef WITH_TCP_USER_TIMEOUT
+	if(bridge__set_tcp_user_timeout(context)) return MOSQ_ERR_UNKNOWN;
+#endif
 
 	if(context->bridge->max_topic_alias != 0){
 		topic_alias_max.next = NULL;
@@ -526,6 +542,9 @@ int bridge__connect(struct mosquitto *context)
 	HASH_ADD(hh_sock, db.contexts_by_sock, sock, sizeof(context->sock), context);
 
 	if (bridge__set_tcp_keepalive(context) != MOSQ_ERR_SUCCESS) return MOSQ_ERR_UNKNOWN;
+#ifdef WITH_TCP_USER_TIMEOUT
+	if(bridge__set_tcp_user_timeout(context)) return MOSQ_ERR_UNKNOWN;
+#endif
 
 	if(context->bridge->max_topic_alias){
 		topic_alias_max.next = NULL;

src/conf.c

@@ -1273,6 +1273,22 @@ static int config__read_file_core(struct mosquitto__config *config, bool reload,
 					cur_bridge->tcp_keepalive_counter = (unsigned int)tmp_int;
 #else
 					log__printf(NULL, MOSQ_LOG_WARNING, "Warning: Bridge support not available.");
+#endif
+				}else if(!strcmp(token, "bridge_tcp_user_timeout")){
+#if defined(WITH_BRIDGE) && defined(WITH_TCP_USER_TIMEOUT)
+					if(!cur_bridge){
+						log__printf(NULL, MOSQ_LOG_ERR, "Error: Invalid bridge configuration.");
+						return MOSQ_ERR_INVAL;
+					}
+
+					if(conf__parse_int(&token, "bridge_tcp_user_timeout", &tmp_int, &saveptr)) return MOSQ_ERR_INVAL;
+					if(tmp_int < 0) {
+						log__printf(NULL, MOSQ_LOG_ERR, "Error: invalid TCP user timeout value.");
+						return MOSQ_ERR_INVAL;
+					}
+					cur_bridge->tcp_user_timeout = (unsigned int) tmp_int;
+#else
+					log__printf(NULL, MOSQ_LOG_WARNING, "Warning: Bridge and/or TCP user timeout support not available.");
 #endif
 				}else if(!strcmp(token, "bridge_tls_version")){
 #if defined(WITH_BRIDGE) && defined(WITH_TLS)
@@ -1416,6 +1432,9 @@ static int config__read_file_core(struct mosquitto__config *config, bool reload,
 						cur_bridge->clean_start_local = -1;
 						cur_bridge->reload_type = brt_lazy;
 						cur_bridge->max_topic_alias = 10;
+#ifdef WITH_TCP_USER_TIMEOUT
+						cur_bridge->tcp_user_timeout = -1;
+#endif
 					}else{
 						log__printf(NULL, MOSQ_LOG_ERR, "Error: Empty connection value in configuration.");
 						return MOSQ_ERR_INVAL;

src/mosquitto_broker_internal.h

@@ -30,6 +30,10 @@ Contributors:
 #  endif
 #endif
 
+#ifdef __linux__
+#define WITH_TCP_USER_TIMEOUT
+#endif
+
 #include "mosquitto_internal.h"
 #include "mosquitto_broker.h"
 #include "mosquitto_plugin.h"
@@ -533,6 +537,9 @@ struct mosquitto__bridge{
 	unsigned int tcp_keepalive_idle;
 	unsigned int tcp_keepalive_interval;
 	unsigned int tcp_keepalive_counter;
+#ifdef WITH_TCP_USER_TIMEOUT
+	unsigned int tcp_user_timeout;
+#endif
 	struct mosquitto__bridge_topic *topics;
 	int topic_count;
 	bool topic_remapping;