X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fagent-thread.c;h=575f6aee934a504ddb6248d4e87610d622d0959c;hp=6790a63c642bd78bb37484a6c6af0b1af39081c7;hb=1f4962443f25c371e4b54e97f9eb867d67cbf88e;hpb=fefd409b002735b415c5f653cdb2587be454f145

diff --git a/src/bin/lttng-sessiond/agent-thread.c b/src/bin/lttng-sessiond/agent-thread.c
index 6790a63c6..575f6aee9 100644
--- a/src/bin/lttng-sessiond/agent-thread.c
+++ b/src/bin/lttng-sessiond/agent-thread.c
@@ -15,7 +15,7 @@
  * Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  */
 
-#define _GNU_SOURCE
+#define _LGPL_SOURCE
 #include <assert.h>
 
 #include <common/common.h>
@@ -27,9 +27,18 @@
 
 #include "fd-limit.h"
 #include "agent-thread.h"
+#include "agent.h"
 #include "lttng-sessiond.h"
 #include "session.h"
 #include "utils.h"
+#include "thread.h"
+
+struct thread_notifiers {
+	struct lttng_pipe *quit_pipe;
+	sem_t ready;
+};
+
+static int agent_tracing_enabled = -1;
 
 /*
  * Note that there is not port here. It's set after this URI is parsed so we
@@ -56,67 +65,27 @@ static void update_agent_app(struct agent_app *app)
 
 	session_lock_list();
 	cds_list_for_each_entry_safe(session, stmp, &list->head, list) {
+		if (!session_get(session)) {
+			continue;
+		}
+
 		session_lock(session);
 		if (session->ust_session) {
 			struct agent *agt;
 
+			rcu_read_lock();
 			agt = trace_ust_find_agent(session->ust_session, app->domain);
 			if (agt) {
 				agent_update(agt, app->sock->fd);
 			}
+			rcu_read_unlock();
 		}
 		session_unlock(session);
+		session_put(session);
 	}
 	session_unlock_list();
 }
 
-/*
- * Destroy a agent application by socket.
- */
-static void destroy_agent_app(int sock)
-{
-	struct agent_app *app;
-
-	assert(sock >= 0);
-
-	/*
-	 * Not finding an application is a very important error that should NEVER
-	 * happen. The hash table deletion is ONLY done through this call even on
-	 * thread cleanup.
-	 */
-	rcu_read_lock();
-	app = agent_find_app_by_sock(sock);
-	assert(app);
-	rcu_read_unlock();
-
-	/* RCU read side lock is taken in this function call. */
-	agent_delete_app(app);
-
-	/* The application is freed in a RCU call but the socket is closed here. */
-	agent_destroy_app(app);
-}
-
-/*
- * Cleanup remaining agent apps in the hash table. This should only be called in
- * the exit path of the thread.
- */
-static void clean_agent_apps_ht(void)
-{
-	struct lttng_ht_node_ulong *node;
-	struct lttng_ht_iter iter;
-
-	DBG3("[agent-thread] Cleaning agent apps ht");
-
-	rcu_read_lock();
-	cds_lfht_for_each_entry(agent_apps_ht_by_sock->ht, &iter.iter, node, node) {
-		struct agent_app *app;
-
-		app = caa_container_of(node, struct agent_app, node);
-		destroy_agent_app(app->sock->fd);
-	}
-	rcu_read_unlock();
-}
-
 /*
  * Create and init socket from uri.
  */
@@ -125,6 +94,8 @@ static struct lttcomm_sock *init_tcp_socket(void)
 	int ret;
 	struct lttng_uri *uri = NULL;
 	struct lttcomm_sock *sock = NULL;
+	unsigned int port;
+	bool bind_succeeded = false;
 
 	/*
 	 * This should never fail since the URI is hardcoded and the port is set
@@ -132,8 +103,8 @@ static struct lttcomm_sock *init_tcp_socket(void)
 	 */
 	ret = uri_parse(default_reg_uri, &uri);
 	assert(ret);
-	assert(agent_tcp_port);
-	uri->port = agent_tcp_port;
+	assert(config.agent_tcp_port.begin > 0);
+	uri->port = config.agent_tcp_port.begin;
 
 	sock = lttcomm_alloc_sock_from_uri(uri);
 	uri_free(uri);
@@ -147,11 +118,43 @@ static struct lttcomm_sock *init_tcp_socket(void)
 		goto error;
 	}
 
-	ret = sock->ops->bind(sock);
-	if (ret < 0) {
-		WARN("Another session daemon is using this agent port. Agent support "
-				"will be deactivated to prevent interfering with the tracing.");
-		goto error;
+	for (port = config.agent_tcp_port.begin;
+			port <= config.agent_tcp_port.end; port++) {
+		ret = lttcomm_sock_set_port(sock, (uint16_t) port);
+		if (ret) {
+			ERR("[agent-thread] Failed to set port %u on socket",
+					port);
+			goto error;
+		}
+		DBG3("[agent-thread] Trying to bind on port %u", port);
+		ret = sock->ops->bind(sock);
+		if (!ret) {
+			bind_succeeded = true;
+			break;
+		}
+
+		if (errno == EADDRINUSE) {
+			DBG("Failed to bind to port %u since it is already in use",
+					port);
+		} else {
+			PERROR("Failed to bind to port %u", port);
+			goto error;
+		}
+	}
+
+	if (!bind_succeeded) {
+		if (config.agent_tcp_port.begin == config.agent_tcp_port.end) {
+			WARN("Another process is already using the agent port %i. "
+					"Agent support will be deactivated.",
+					config.agent_tcp_port.begin);
+			goto error;
+		} else {
+			WARN("All ports in the range [%i, %i] are already in use. "
+					"Agent support will be deactivated.",
+					config.agent_tcp_port.begin,
+					config.agent_tcp_port.end);
+			goto error;
+		}
 	}
 
 	ret = sock->ops->listen(sock, -1);
@@ -160,7 +163,7 @@ static struct lttcomm_sock *init_tcp_socket(void)
 	}
 
 	DBG("[agent-thread] Listening on TCP port %u and socket %d",
-			agent_tcp_port, sock->fd);
+			port, sock->fd);
 
 	return sock;
 
@@ -176,9 +179,19 @@ error:
  */
 static void destroy_tcp_socket(struct lttcomm_sock *sock)
 {
+	int ret;
+	uint16_t port;
+
 	assert(sock);
 
-	DBG3("[agent-thread] Destroy TCP socket on port %u", agent_tcp_port);
+	ret = lttcomm_sock_get_port(sock, &port);
+	if (ret) {
+		ERR("[agent-thread] Failed to get port of agent TCP socket");
+		port = 0;
+	}
+
+	DBG3("[agent-thread] Destroy TCP socket on port %" PRIu16,
+			port);
 
 	/* This will return gracefully if fd is invalid. */
 	sock->ops->close(sock);
@@ -198,6 +211,7 @@ static int handle_registration(struct lttcomm_sock *reg_sock,
 {
 	int ret;
 	pid_t pid;
+	uint32_t major_version, minor_version;
 	ssize_t size;
 	enum lttng_domain_type domain;
 	struct agent_app *app;
@@ -214,11 +228,23 @@ static int handle_registration(struct lttcomm_sock *reg_sock,
 
 	size = new_sock->ops->recvmsg(new_sock, &msg, sizeof(msg), 0);
 	if (size < sizeof(msg)) {
-		ret = -errno;
+		ret = -EINVAL;
 		goto error_socket;
 	}
 	domain = be32toh(msg.domain);
 	pid = be32toh(msg.pid);
+	major_version = be32toh(msg.major_version);
+	minor_version = be32toh(msg.minor_version);
+
+	/* Test communication protocol version of the registring agent. */
+	if (major_version != AGENT_MAJOR_VERSION) {
+		ret = -EINVAL;
+		goto error_socket;
+	}
+	if (minor_version != AGENT_MINOR_VERSION) {
+		ret = -EINVAL;
+		goto error_socket;
+	}
 
 	DBG2("[agent-thread] New registration for pid %d domain %d on socket %d",
 			pid, domain, new_sock->fd);
@@ -254,15 +280,51 @@ error:
 	return ret;
 }
 
+bool agent_tracing_is_enabled(void)
+{
+	int enabled;
+
+	enabled = uatomic_read(&agent_tracing_enabled);
+	assert(enabled != -1);
+	return enabled == 1;
+}
+
+/*
+ * Write agent TCP port using the rundir.
+ */
+static int write_agent_port(uint16_t port)
+{
+	return utils_create_pid_file((pid_t) port,
+			config.agent_port_file_path.value);
+}
+
+static
+void mark_thread_as_ready(struct thread_notifiers *notifiers)
+{
+	DBG("Marking agent management thread as ready");
+	sem_post(&notifiers->ready);
+}
+
+static
+void wait_until_thread_is_ready(struct thread_notifiers *notifiers)
+{
+	DBG("Waiting for agent management thread to be ready");
+	sem_wait(&notifiers->ready);
+	DBG("Agent management thread is ready");
+}
+
 /*
  * This thread manage application notify communication.
  */
-void *agent_thread_manage_registration(void *data)
+static void *thread_agent_management(void *data)
 {
 	int i, ret, pollfd;
 	uint32_t revents, nb_fd;
 	struct lttng_poll_event events;
 	struct lttcomm_sock *reg_sock;
+	struct thread_notifiers *notifiers = data;
+	const int quit_pipe_read_fd = lttng_pipe_get_readfd(
+			notifiers->quit_pipe);
 
 	DBG("[agent-thread] Manage agent application registration.");
 
@@ -272,18 +334,45 @@ void *agent_thread_manage_registration(void *data)
 	/* Agent initialization call MUST be called before starting the thread. */
 	assert(agent_apps_ht_by_sock);
 
-	/* Create pollset with size 2, quit pipe and socket. */
-	ret = sessiond_set_thread_pollset(&events, 2);
+	/* Create pollset with size 2, quit pipe and registration socket. */
+	ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
 	if (ret < 0) {
 		goto error_poll_create;
 	}
 
+	ret = lttng_poll_add(&events, quit_pipe_read_fd,
+			LPOLLIN | LPOLLERR);
+	if (ret < 0) {
+		goto error_tcp_socket;
+	}
+
 	reg_sock = init_tcp_socket();
-	if (!reg_sock) {
+	if (reg_sock) {
+		uint16_t port;
+
+		assert(lttcomm_sock_get_port(reg_sock, &port) == 0);
+
+		ret = write_agent_port(port);
+		if (ret) {
+			ERR("[agent-thread] Failed to create agent port file: agent tracing will be unavailable");
+			/* Don't prevent the launch of the sessiond on error. */
+			mark_thread_as_ready(notifiers);
+			goto error;
+		}
+	} else {
+		/* Don't prevent the launch of the sessiond on error. */
+		mark_thread_as_ready(notifiers);
 		goto error_tcp_socket;
 	}
 
-	/* Add create valid TCP socket to poll set. */
+	/*
+	 * Signal that the agent thread is ready. The command thread
+	 * may start to query whether or not agent tracing is enabled.
+	 */
+	uatomic_set(&agent_tracing_enabled, 1);
+	mark_thread_as_ready(notifiers);
+
+	/* Add TCP socket to poll set. */
 	ret = lttng_poll_add(&events, reg_sock->fd,
 			LPOLLIN | LPOLLERR | LPOLLHUP | LPOLLRDHUP);
 	if (ret < 0) {
@@ -291,12 +380,13 @@ void *agent_thread_manage_registration(void *data)
 	}
 
 	while (1) {
-		DBG3("[agent-thread] Manage agent polling on %d fds",
-				LTTNG_POLL_GETNB(&events));
+		DBG3("[agent-thread] Manage agent polling");
 
 		/* Inifinite blocking call, waiting for transmission */
 restart:
 		ret = lttng_poll_wait(&events, -1);
+		DBG3("[agent-thread] Manage agent return from poll on %d fds",
+				LTTNG_POLL_GETNB(&events));
 		if (ret < 0) {
 			/*
 			 * Restart interrupted system call.
@@ -315,44 +405,30 @@ restart:
 			pollfd = LTTNG_POLL_GETFD(&events, i);
 
 			/* Thread quit pipe has been closed. Killing thread. */
-			ret = sessiond_check_thread_quit_pipe(pollfd, revents);
-			if (ret) {
+			if (pollfd == quit_pipe_read_fd) {
 				goto exit;
 			}
 
-			/*
-			 * Check first if this is a POLLERR since POLLIN is also included
-			 * in an error value thus checking first.
-			 */
-			if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
-				/* Removing from the poll set */
-				ret = lttng_poll_del(&events, pollfd);
-				if (ret < 0) {
-					goto error;
-				}
-
-				destroy_agent_app(pollfd);
-			} else if (revents & (LPOLLIN)) {
+			if (revents & LPOLLIN) {
 				int new_fd;
 				struct agent_app *app = NULL;
 
-				/* Pollin event of agent app socket should NEVER happen. */
 				assert(pollfd == reg_sock->fd);
-
 				new_fd = handle_registration(reg_sock, &app);
 				if (new_fd < 0) {
-					WARN("[agent-thread] agent registration failed. Ignoring.");
-					/* Somehow the communication failed. Just continue. */
 					continue;
 				}
 				/* Should not have a NULL app on success. */
 				assert(app);
 
-				/* Only add poll error event to only detect shutdown. */
+				/*
+				 * Since this is a command socket (write then read),
+				 * only add poll error event to only detect shutdown.
+				 */
 				ret = lttng_poll_add(&events, new_fd,
 						LPOLLERR | LPOLLHUP | LPOLLRDHUP);
 				if (ret < 0) {
-					destroy_agent_app(new_fd);
+					agent_destroy_app_by_sock(new_fd);
 					continue;
 				}
 
@@ -360,10 +436,26 @@ restart:
 				update_agent_app(app);
 
 				/* On failure, the poll will detect it and clean it up. */
-				(void) agent_send_registration_done(app);
+				ret = agent_send_registration_done(app);
+				if (ret < 0) {
+					/* Removing from the poll set */
+					ret = lttng_poll_del(&events, new_fd);
+					if (ret < 0) {
+						goto error;
+					}
+					agent_destroy_app_by_sock(new_fd);
+					continue;
+				}
+			} else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
+				/* Removing from the poll set */
+				ret = lttng_poll_del(&events, pollfd);
+				if (ret < 0) {
+					goto error;
+				}
+				agent_destroy_app_by_sock(pollfd);
 			} else {
-				ERR("Unknown poll events %u for sock %d", revents, pollfd);
-				continue;
+				ERR("Unexpected poll events %u for sock %d", revents, pollfd);
+				goto error;
 			}
 		}
 	}
@@ -376,14 +468,58 @@ error:
 error_tcp_socket:
 	lttng_poll_clean(&events);
 error_poll_create:
-	DBG("[agent-thread] is cleaning up and stopping.");
-
-	if (agent_apps_ht_by_sock) {
-		clean_agent_apps_ht();
-		lttng_ht_destroy(agent_apps_ht_by_sock);
-	}
-
+	uatomic_set(&agent_tracing_enabled, 0);
+	DBG("[agent-thread] Cleaning up and stopping.");
 	rcu_thread_offline();
 	rcu_unregister_thread();
 	return NULL;
 }
+
+static bool shutdown_agent_management_thread(void *data)
+{
+	struct thread_notifiers *notifiers = data;
+	const int write_fd = lttng_pipe_get_writefd(notifiers->quit_pipe);
+
+	return notify_thread_pipe(write_fd) == 1;
+}
+
+static void cleanup_agent_management_thread(void *data)
+{
+	struct thread_notifiers *notifiers = data;
+
+	lttng_pipe_destroy(notifiers->quit_pipe);
+	sem_destroy(&notifiers->ready);
+	free(notifiers);
+}
+
+bool launch_agent_management_thread(void)
+{
+	struct thread_notifiers *notifiers;
+	struct lttng_thread *thread;
+
+	notifiers = zmalloc(sizeof(*notifiers));
+	if (!notifiers) {
+		goto error_alloc;
+	}
+
+	sem_init(&notifiers->ready, 0, 0);
+	notifiers->quit_pipe = lttng_pipe_open(FD_CLOEXEC);
+	if (!notifiers->quit_pipe) {
+		goto error;
+	}
+	thread = lttng_thread_create("Agent management",
+			thread_agent_management,
+			shutdown_agent_management_thread,
+			cleanup_agent_management_thread,
+			notifiers);
+	if (!thread) {
+		goto error;
+	}
+	wait_until_thread_is_ready(notifiers);
+	lttng_thread_put(thread);
+	return true;
+error:
+	cleanup_agent_management_thread(notifiers);
+error_alloc:
+	return false;
+}