Merge branch 'hotfix/44-proper-cla-rate-limiting'

See !148

components/cla/posix/cla_bibe.c

@@ -61,7 +61,6 @@ struct bibe_contact_parameters {
 	const char *partner_eid;
 
 	bool in_contact;
-	int connect_attempt;
 
 	int socket;
 };
@@ -111,7 +110,6 @@ static void bibe_link_management_task(void *p)
 		if (param->socket > 0) {
 			// NOTE the following function releases and re-locks sem
 			handle_established_connection(param);
-			param->connect_attempt = 0;
 			param->socket = -1;
 		} else {
 			if (param->cla_sock_addr[0] == '\0') {
@@ -121,31 +119,17 @@ static void bibe_link_management_task(void *p)
 			ASSERT(param->socket < 0);
 			hal_semaphore_release(param->param_semphr);
 
+			if (cla_tcp_rate_limit_connection_attempts(
+					&param->config->base))
+				break;
 			const int socket = cla_tcp_connect_to_cla_addr(
 				param->cla_sock_addr, // only used by us
 				NULL
 			);
 
 			hal_semaphore_take_blocking(param->param_semphr);
-			if (socket < 0) {
-				if (++param->connect_attempt >
-						CLA_TCP_MAX_RETRY_ATTEMPTS) {
-					LOG_WARN("BIBE: Final retry failed.");
-					break;
-				}
-				LOGF_INFO(
-					"BIBE: Delayed retry %d of %d in %d ms",
-					param->connect_attempt,
-					CLA_TCP_MAX_RETRY_ATTEMPTS,
-					CLA_TCP_RETRY_INTERVAL_MS
-				);
-				hal_semaphore_release(param->param_semphr);
-				hal_task_delay(CLA_TCP_RETRY_INTERVAL_MS);
-				hal_semaphore_take_blocking(
-					param->param_semphr
-				);
+			if (socket < 0)
 				continue;
-			}
 			param->socket = socket;
 
 			const struct aap_message register_bibe = {
@@ -171,22 +155,7 @@ static void bibe_link_management_task(void *p)
 			if (wsp.errno_) {
 				LOG_ERRNO("BIBE", "send()", wsp.errno_);
 				close(param->socket);
-				if (++param->connect_attempt >
-						CLA_TCP_MAX_RETRY_ATTEMPTS) {
-					LOG_WARN("BIBE: Final retry failed.");
-					break;
-				}
-				LOGF_INFO(
-					"BIBE: Delayed retry %d of %d in %d ms",
-					param->connect_attempt,
-					CLA_TCP_MAX_RETRY_ATTEMPTS,
-					CLA_TCP_RETRY_INTERVAL_MS
-				);
-				hal_semaphore_release(param->param_semphr);
-				hal_task_delay(CLA_TCP_RETRY_INTERVAL_MS);
-				hal_semaphore_take_blocking(
-					param->param_semphr
-				);
+				param->socket = -1;
 				continue;
 			}
 			LOGF_INFO(
@@ -236,7 +205,6 @@ static void launch_connection_management_task(
 	}
 
 	contact_params->config = bibe_config;
-	contact_params->connect_attempt = 0;
 
 	char *const cla_sock_addr = cla_get_connect_addr(cla_addr, CLA_NAME);
 

components/cla/posix/cla_mtcp.c

@@ -52,7 +52,6 @@ struct mtcp_contact_parameters {
 
 	bool is_outgoing;
 	bool in_contact;
-	int connect_attempt;
 
 	int socket;
 };
@@ -103,7 +102,6 @@ static void mtcp_link_management_task(void *p)
 		if (param->socket > 0) {
 			// NOTE the following function releases and re-locks sem
 			handle_established_connection(param);
-			param->connect_attempt = 0;
 			param->socket = -1;
 		} else {
 			if (param->cla_sock_addr[0] == '\0') {
@@ -113,31 +111,17 @@ static void mtcp_link_management_task(void *p)
 			ASSERT(param->socket < 0);
 			hal_semaphore_release(param->param_semphr);
 
+			if (cla_tcp_rate_limit_connection_attempts(
+					&param->config->base))
+				break;
 			const int socket = cla_tcp_connect_to_cla_addr(
 				param->cla_sock_addr, // only used by us
 				NULL
 			);
 
 			hal_semaphore_take_blocking(param->param_semphr);
-			if (socket < 0) {
-				if (++param->connect_attempt >
-						CLA_TCP_MAX_RETRY_ATTEMPTS) {
-					LOG_WARN("MTCP: Final retry failed.");
-					break;
-				}
-				LOGF_INFO(
-					"MTCP: Delayed retry %d of %d in %d ms",
-					param->connect_attempt,
-					CLA_TCP_MAX_RETRY_ATTEMPTS,
-					CLA_TCP_RETRY_INTERVAL_MS
-				);
-				hal_semaphore_release(param->param_semphr);
-				hal_task_delay(CLA_TCP_RETRY_INTERVAL_MS);
-				hal_semaphore_take_blocking(
-					param->param_semphr
-				);
+			if (socket < 0)
 				continue;
-			}
 			param->socket = socket;
 			LOGF_INFO(
 				"MTCP: Connected successfully to \"%s\"",
@@ -184,7 +168,6 @@ static void launch_connection_management_task(
 	}
 
 	contact_params->config = mtcp_config;
-	contact_params->connect_attempt = 0;
 
 	if (sock < 0) {
 		contact_params->cla_sock_addr = cla_get_connect_addr(

components/cla/posix/cla_smtcp.c

@@ -40,9 +40,6 @@ static void smtcp_link_creation_task(void *param)
 		smtcp_config,
 		sizeof(struct mtcp_link)
 	);
-
-	// Should never get here.
-	abort();
 }
 
 static enum ud3tn_result smtcp_launch(struct cla_config *const config)

components/cla/posix/cla_tcp_common.c

@@ -7,6 +7,7 @@
 #include "platform/hal_io.h"
 #include "platform/hal_semaphore.h"
 #include "platform/hal_task.h"
+#include "platform/hal_time.h"
 
 #include "ud3tn/common.h"
 #include "ud3tn/result.h"
@@ -40,6 +41,8 @@ enum ud3tn_result cla_tcp_config_init(
 		return UD3TN_FAIL;
 
 	config->socket = -1;
+	config->last_connection_attempt_ms = 0;
+	config->last_connection_attempt_no = 1;
 
 	return UD3TN_OK;
 }
@@ -294,14 +297,14 @@ void cla_tcp_single_connect_task(struct cla_tcp_single_config *config,
 			config->service
 		);
 
+		if (cla_tcp_rate_limit_connection_attempts(&config->base))
+			break;
 		if (cla_tcp_connect(&config->base,
 				    config->node, config->service) != UD3TN_OK) {
 			LOGF_WARN(
-				"TCP: CLA \"%s\": Connection failed, will retry in %d ms as long as a contact is ongoing.",
-				config->base.base.vtable->cla_name_get(),
-				CLA_TCP_RETRY_INTERVAL_MS
+				"TCP: CLA \"%s\": Connection failed, will retry as long as a contact is ongoing.",
+				config->base.base.vtable->cla_name_get()
 			);
-			hal_task_delay(CLA_TCP_RETRY_INTERVAL_MS);
 		} else {
 			handle_established_connection(config, NULL,
 						      config->base.socket,
@@ -359,11 +362,58 @@ void cla_tcp_single_link_creation_task(struct cla_tcp_single_config *config,
 				config->node,
 				config->service
 			);
-			abort();
-			return;
+		} else {
+			cla_tcp_single_listen_task(config, struct_size);
 		}
-		cla_tcp_single_listen_task(config, struct_size);
 	}
+
+	LOGF_ERROR(
+		"TCP: CLA \"%s\" link task terminated.",
+		config->base.base.vtable->cla_name_get()
+	);
+	if (CLA_TCP_ABORT_ON_LINK_TASK_TERMINATION)
+		abort();
+}
+
+int cla_tcp_rate_limit_connection_attempts(struct cla_tcp_config *config)
+{
+	const uint64_t rt_limit_ms = CLA_TCP_RETRY_INTERVAL_MS;
+	const int rt_max_attempts = CLA_TCP_MAX_RETRY_ATTEMPTS;
+	const uint64_t now_ms = hal_time_get_timestamp_ms();
+
+	if (config->last_connection_attempt_ms + rt_limit_ms > now_ms) {
+		if (rt_max_attempts <= 0) {
+			LOGF_WARN(
+				"TCP: Last connection attempt was less than %llu ms ago, delaying next attempt.",
+				rt_limit_ms
+			);
+		} else {
+			if (config->last_connection_attempt_no >=
+					rt_max_attempts) {
+				LOGF_WARN(
+					"TCP: Final retry %d of %d failed.",
+					config->last_connection_attempt_no,
+					rt_max_attempts
+				);
+				return -1;
+			}
+			LOGF_WARN(
+				"TCP: Last connection attempt %d of %d was less than %llu ms ago, delaying next attempt.",
+				config->last_connection_attempt_no,
+				rt_max_attempts,
+				rt_limit_ms
+			);
+			config->last_connection_attempt_no++;
+		}
+		hal_task_delay(rt_limit_ms);
+		config->last_connection_attempt_ms = (
+			hal_time_get_timestamp_ms()
+		);
+	} else {
+		config->last_connection_attempt_ms = now_ms;
+		config->last_connection_attempt_no = 1;
+	}
+	return 0;
 }
 
 struct cla_tx_queue cla_tcp_single_get_tx_queue(

components/cla/posix/cla_tcp_util.c

@@ -2,6 +2,7 @@
 #include "cla/posix/cla_tcp_util.h"
 
 #include "platform/hal_io.h"
+#include "platform/posix/socket_util.h"
 
 #include "ud3tn/common.h"
 

components/cla/posix/cla_tcpclv3.c

@@ -71,8 +71,6 @@ struct tcpclv3_contact_parameters {
 	char *eid;
 	char *cla_addr;
 
-	int connect_attempt;
-
 	int socket;
 
 	enum TCPCLV3_STATE state;
@@ -288,30 +286,18 @@ static void tcpclv3_link_management_task(void *p)
 
 			hal_semaphore_release(param->param_semphr);
 
+			if (cla_tcp_rate_limit_connection_attempts(
+					&param->config->base))
+				break;
 			const int socket = cla_tcp_connect_to_cla_addr(
 				param->cla_addr,
 				"4556"
 			);
 
 			hal_semaphore_take_blocking(param->param_semphr);
 
-			if (socket < 0) {
-				if (++param->connect_attempt >
-						CLA_TCP_MAX_RETRY_ATTEMPTS) {
-					LOG_WARN("TCPCLv3: Final retry failed.");
-					break;
-				}
-				LOGF_INFO("TCPCLv3: Delayed retry %d of %d in %d ms",
-				     param->connect_attempt,
-				     CLA_TCP_MAX_RETRY_ATTEMPTS,
-				     CLA_TCP_RETRY_INTERVAL_MS);
-				hal_semaphore_release(param->param_semphr);
-				hal_task_delay(CLA_TCP_RETRY_INTERVAL_MS);
-				hal_semaphore_take_blocking(
-					param->param_semphr
-				);
+			if (socket < 0)
 				continue;
-			}
 			LOGF_INFO("TCPCLv3: Connected successfully to \"%s\"",
 			     param->cla_addr);
 			param->socket = socket;
@@ -329,7 +315,6 @@ static void tcpclv3_link_management_task(void *p)
 				break;
 			}
 			param->state = TCPCLV3_CONNECTING;
-			param->connect_attempt = 0;
 		} else {
 			// TCPCLV3_INACTIVE, TCPCLV3_ESTABLISHED
 			// should never happen as we are not created or wait
@@ -386,7 +371,6 @@ static void launch_connection_management_task(
 	}
 
 	contact_params->config = tcpclv3_config;
-	contact_params->connect_attempt = 0;
 
 	if (sock < 0) {
 		ASSERT(eid && cla_addr);

components/cla/posix/cla_tcpspp.c

@@ -81,9 +81,6 @@ static void tcpspp_link_creation_task(void *param)
 		&tcpspp_config->base,
 		sizeof(struct cla_tcp_link)
 	);
-
-	// Should never get here.
-	abort();
 }
 
 static enum ud3tn_result tcpspp_launch(struct cla_config *const config)

components/platform/posix/socket_util.c

@@ -71,14 +71,18 @@ int poll_recv_timeout(const int socket_fd, const int timeout)
 			LOG_ERRNO("Socket Util", "poll()", err);
 			return -1;
 		}
-		if ((pollfd[0].revents & POLLERR)) {
+		if (pollfd[0].revents & POLLERR) {
 			LOG_WARN("Socket Util: Socket error (e.g. TCP RST) detected.");
 			return -1;
 		}
 		if (pollfd[0].revents & POLLHUP) {
 			LOG_INFO("Socket Util: The peer closed the connection.");
 			return -1;
 		}
+		if (pollfd[0].revents & POLLNVAL) {
+			LOG_INFO("Socket Util: Connection was not open.");
+			return -1;
+		}
 		if (pollfd[0].revents & POLLIN)
 			return 1;
 		return 0;

config.mk.example

@@ -130,12 +130,16 @@
 # Setting this can help when broken records may be part of the incoming data.
 #CPPFLAGS += -DCLA_RX_READ_TIMEOUT=0
 
+# Whether to abort() uD3TN in case a TCP CLA terminates finally.
+#CPPFLAGS += -DCLA_TCP_ABORT_ON_LINK_TASK_TERMINATION=0
+
 # Whether to set SO_REUSEPORT on listening TCP sockets. Note that this may have
 # security implications and it is Linux-/BSD-specific.
 #CPPFLAGS += -DCLA_TCP_ALLOW_REUSE_PORT=0
 
 # Maximum number of attempts to create a connection on contact start.
-#CPPFLAGS += -DCLA_TCP_MAX_RETRY_ATTEMPTS=10
+# The default value of 0 means infinite.
+#CPPFLAGS += -DCLA_TCP_MAX_RETRY_ATTEMPTS=0
 
 # The length of the listen backlog for multi-connection TCP CLAs.
 #CPPFLAGS += -DCLA_TCP_MULTI_BACKLOG=64