X-Git-Url: http://git.lttng.org/?a=blobdiff_plain;f=liblttng-ust%2Flttng-ust-comm.c;h=8c864f8ce595621434cfac4a08dec1aa705d5671;hb=3b8b68e73ec9b2b3cf550048046d3f7f69050688;hp=798c10a36c47bc1e7257227ba57dc46945fac853;hpb=3ad5f707e8784c29c14bcf46b4b2ea9753679ad2;p=lttng-ust.git

diff --git a/liblttng-ust/lttng-ust-comm.c b/liblttng-ust/lttng-ust-comm.c
index 798c10a3..8c864f8c 100644
--- a/liblttng-ust/lttng-ust-comm.c
+++ b/liblttng-ust/lttng-ust-comm.c
@@ -46,6 +46,7 @@
 #include <usterr-signal-safe.h>
 #include "tracepoint-internal.h"
 #include "ltt-tracer-core.h"
+#include "../libringbuffer/tlsfixup.h"
 
 /*
  * Has lttng ust comm constructor been called ?
@@ -77,6 +78,12 @@ static sem_t constructor_wait;
  */
 static int sem_count = { 2 };
 
+/*
+ * Counting nesting within lttng-ust. Used to ensure that calling fork()
+ * from liblttng-ust does not execute the pre/post fork handlers.
+ */
+static int __thread lttng_ust_nest_count;
+
 /*
  * Info about socket and associated listener thread.
  */
@@ -289,28 +296,30 @@ end:
 		//lur.ret_code = USTCOMM_SESSION_FAIL;
 		lur.ret_code = ret;
 	}
-	switch (lum->cmd) {
-	case LTTNG_UST_STREAM:
-		/*
-		 * Special-case reply to send stream info.
-		 * Use lum.u output.
-		 */
-		lur.u.stream.memory_map_size = *args.stream.memory_map_size;
-		shm_fd = *args.stream.shm_fd;
-		wait_fd = *args.stream.wait_fd;
-		break;
-	case LTTNG_UST_METADATA:
-	case LTTNG_UST_CHANNEL:
-		lur.u.channel.memory_map_size = *args.channel.memory_map_size;
-		shm_fd = *args.channel.shm_fd;
-		wait_fd = *args.channel.wait_fd;
-		break;
-	case LTTNG_UST_TRACER_VERSION:
-		lur.u.version = lum->u.version;
-		break;
-	case LTTNG_UST_TRACEPOINT_LIST_GET:
-		memcpy(&lur.u.tracepoint, &lum->u.tracepoint, sizeof(lur.u.tracepoint));
-		break;
+	if (ret >= 0) {
+		switch (lum->cmd) {
+		case LTTNG_UST_STREAM:
+			/*
+			 * Special-case reply to send stream info.
+			 * Use lum.u output.
+			 */
+			lur.u.stream.memory_map_size = *args.stream.memory_map_size;
+			shm_fd = *args.stream.shm_fd;
+			wait_fd = *args.stream.wait_fd;
+			break;
+		case LTTNG_UST_METADATA:
+		case LTTNG_UST_CHANNEL:
+			lur.u.channel.memory_map_size = *args.channel.memory_map_size;
+			shm_fd = *args.channel.shm_fd;
+			wait_fd = *args.channel.wait_fd;
+			break;
+		case LTTNG_UST_TRACER_VERSION:
+			lur.u.version = lum->u.version;
+			break;
+		case LTTNG_UST_TRACEPOINT_LIST_GET:
+			memcpy(&lur.u.tracepoint, &lum->u.tracepoint, sizeof(lur.u.tracepoint));
+			break;
+		}
 	}
 	ret = send_reply(sock, &lur);
 	if (ret < 0) {
@@ -322,14 +331,20 @@ end:
 	     || lum->cmd == LTTNG_UST_CHANNEL
 	     || lum->cmd == LTTNG_UST_METADATA)
 			&& lur.ret_code == USTCOMM_OK) {
+		int sendret = 0;
+
 		/* we also need to send the file descriptors. */
 		ret = ustcomm_send_fds_unix_sock(sock,
 			&shm_fd, &shm_fd,
 			1, sizeof(int));
 		if (ret < 0) {
 			perror("send shm_fd");
-			goto error;
+			sendret = ret;
 		}
+		/*
+		 * The sessiond expects 2 file descriptors, even upon
+		 * error.
+		 */
 		ret = ustcomm_send_fds_unix_sock(sock,
 			&wait_fd, &wait_fd,
 			1, sizeof(int));
@@ -337,6 +352,10 @@ end:
 			perror("send wait_fd");
 			goto error;
 		}
+		if (sendret) {
+			ret = sendret;
+			goto error;
+		}
 	}
 	/*
 	 * We still have the memory map reference, and the fds have been
@@ -388,12 +407,12 @@ error:
 }
 
 static
-void cleanup_sock_info(struct sock_info *sock_info)
+void cleanup_sock_info(struct sock_info *sock_info, int exiting)
 {
 	int ret;
 
 	if (sock_info->socket != -1) {
-		ret = close(sock_info->socket);
+		ret = ustcomm_close_unix_sock(sock_info->socket);
 		if (ret) {
 			ERR("Error closing apps socket");
 		}
@@ -407,7 +426,13 @@ void cleanup_sock_info(struct sock_info *sock_info)
 		sock_info->root_handle = -1;
 	}
 	sock_info->constructor_sem_posted = 0;
-	if (sock_info->wait_shm_mmap) {
+	/*
+	 * wait_shm_mmap is used by listener threads outside of the
+	 * ust lock, so we cannot tear it down ourselves, because we
+	 * cannot join on these threads. Leave this task to the OS
+	 * process exit.
+	 */
+	if (!exiting && sock_info->wait_shm_mmap) {
 		ret = munmap(sock_info->wait_shm_mmap, sysconf(_SC_PAGE_SIZE));
 		if (ret) {
 			ERR("Error unmapping wait shm");
@@ -450,7 +475,9 @@ int get_wait_shm(struct sock_info *sock_info, size_t mmap_size)
 	 * If the open failed because the file did not exist, try
 	 * creating it ourself.
 	 */
+	lttng_ust_nest_count++;
 	pid = fork();
+	lttng_ust_nest_count--;
 	if (pid > 0) {
 		int status;
 
@@ -667,7 +694,7 @@ restart:
 	}
 
 	if (sock_info->socket != -1) {
-		ret = close(sock_info->socket);
+		ret = ustcomm_close_unix_sock(sock_info->socket);
 		if (ret) {
 			ERR("Error closing %s apps socket", sock_info->name);
 		}
@@ -818,12 +845,22 @@ int get_timeout(struct timespec *constructor_timeout)
 void __attribute__((constructor)) lttng_ust_init(void)
 {
 	struct timespec constructor_timeout;
+	sigset_t sig_all_blocked, orig_parent_mask;
 	int timeout_mode;
 	int ret;
 
 	if (uatomic_xchg(&initialized, 1) == 1)
 		return;
 
+	/*
+	 * Fixup interdependency between TLS fixup mutex (which happens
+	 * to be the dynamic linker mutex) and ust_lock, taken within
+	 * the ust lock.
+	 */
+	lttng_fixup_event_tls();
+	lttng_fixup_ringbuffer_tls();
+	lttng_fixup_vtid_tls();
+
 	/*
 	 * We want precise control over the order in which we construct
 	 * our sub-libraries vs starting to receive commands from
@@ -845,16 +882,39 @@ void __attribute__((constructor)) lttng_ust_init(void)
 	if (ret) {
 		ERR("Error setting up to local apps");
 	}
-	ret = pthread_create(&local_apps.ust_listener, NULL,
-			ust_listener_thread, &local_apps);
 
+	/* A new thread created by pthread_create inherits the signal mask
+	 * from the parent. To avoid any signal being received by the
+	 * listener thread, we block all signals temporarily in the parent,
+	 * while we create the listener thread.
+	 */
+	sigfillset(&sig_all_blocked);
+	ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_parent_mask);
+	if (ret) {
+		ERR("pthread_sigmask: %s", strerror(ret));
+	}
+
+	ret = pthread_create(&global_apps.ust_listener, NULL,
+			ust_listener_thread, &global_apps);
+	if (ret) {
+		ERR("pthread_create global: %s", strerror(ret));
+	}
 	if (local_apps.allowed) {
-		ret = pthread_create(&global_apps.ust_listener, NULL,
-				ust_listener_thread, &global_apps);
+		ret = pthread_create(&local_apps.ust_listener, NULL,
+				ust_listener_thread, &local_apps);
+		if (ret) {
+			ERR("pthread_create local: %s", strerror(ret));
+		}
 	} else {
 		handle_register_done(&local_apps);
 	}
 
+	/* Restore original signal mask in parent */
+	ret = pthread_sigmask(SIG_SETMASK, &orig_parent_mask, NULL);
+	if (ret) {
+		ERR("pthread_sigmask: %s", strerror(ret));
+	}
+
 	switch (timeout_mode) {
 	case 1:	/* timeout wait */
 		do {
@@ -881,10 +941,17 @@ void __attribute__((constructor)) lttng_ust_init(void)
 static
 void lttng_ust_cleanup(int exiting)
 {
-	cleanup_sock_info(&global_apps);
+	cleanup_sock_info(&global_apps, exiting);
 	if (local_apps.allowed) {
-		cleanup_sock_info(&local_apps);
+		cleanup_sock_info(&local_apps, exiting);
 	}
+	/*
+	 * The teardown in this function all affect data structures
+	 * accessed under the UST lock by the listener thread. This
+	 * lock, along with the lttng_ust_comm_should_quit flag, ensure
+	 * that none of these threads are accessing this data at this
+	 * point.
+	 */
 	lttng_ust_abi_exit();
 	lttng_ust_events_exit();
 	ltt_ring_buffer_client_discard_exit();
@@ -918,16 +985,27 @@ void __attribute__((destructor)) lttng_ust_exit(void)
 	lttng_ust_comm_should_quit = 1;
 	ust_unlock();
 
+	/* cancel threads */
 	ret = pthread_cancel(global_apps.ust_listener);
 	if (ret) {
-		ERR("Error cancelling global ust listener thread");
+		ERR("Error cancelling global ust listener thread: %s",
+			strerror(ret));
 	}
 	if (local_apps.allowed) {
 		ret = pthread_cancel(local_apps.ust_listener);
 		if (ret) {
-			ERR("Error cancelling local ust listener thread");
+			ERR("Error cancelling local ust listener thread: %s",
+				strerror(ret));
 		}
 	}
+	/*
+	 * Do NOT join threads: use of sys_futex makes it impossible to
+	 * join the threads without using async-cancel, but async-cancel
+	 * is delivered by a signal, which could hit the target thread
+	 * anywhere in its code path, including while the ust_lock() is
+	 * held, causing a deadlock for the other thread. Let the OS
+	 * cleanup the threads if there are stalled in a syscall.
+	 */
 	lttng_ust_cleanup(1);
 }
 
@@ -949,6 +1027,8 @@ void ust_before_fork(sigset_t *save_sigset)
 	sigset_t all_sigs;
 	int ret;
 
+	if (lttng_ust_nest_count)
+		return;
 	/* Disable signals */
 	sigfillset(&all_sigs);
 	ret = sigprocmask(SIG_BLOCK, &all_sigs, save_sigset);
@@ -974,6 +1054,8 @@ static void ust_after_fork_common(sigset_t *restore_sigset)
 
 void ust_after_fork_parent(sigset_t *restore_sigset)
 {
+	if (lttng_ust_nest_count)
+		return;
 	DBG("process %d", getpid());
 	rcu_bp_after_fork_parent();
 	/* Release mutexes and reenable signals */
@@ -991,6 +1073,8 @@ void ust_after_fork_parent(sigset_t *restore_sigset)
  */
 void ust_after_fork_child(sigset_t *restore_sigset)
 {
+	if (lttng_ust_nest_count)
+		return;
 	DBG("process %d", getpid());
 	/* Release urcu mutexes */
 	rcu_bp_after_fork_child();