X-Git-Url: http://git.lttng.org/?a=blobdiff_plain;f=libust%2Flttng-ust-comm.c;h=3327e10b827e4a5255bd2f1792bdc95bd74bc6fb;hb=d4419b81b243bc3a6bdd4a09b3ca2216d044a1c7;hp=a9f5797df671cc8af4ff5e2f01f4408ab0143360;hpb=11ff9c7d186237bf04d14ca05b2071bc0f12cae3;p=lttng-ust.git

diff --git a/libust/lttng-ust-comm.c b/libust/lttng-ust-comm.c
index a9f5797d..3327e10b 100644
--- a/libust/lttng-ust-comm.c
+++ b/libust/lttng-ust-comm.c
@@ -19,23 +19,44 @@
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
+#define _LGPL_SOURCE
 #include <sys/types.h>
 #include <sys/socket.h>
+#include <sys/prctl.h>
+#include <sys/mman.h>
+#include <sys/stat.h>
+#include <sys/types.h>
+#include <sys/wait.h>
+#include <fcntl.h>
 #include <unistd.h>
 #include <errno.h>
-#include <ust/lttng-ust-abi.h>
-#include <lttng-ust-comm.h>
-#include <ust/usterr-signal-safe.h>
 #include <pthread.h>
 #include <semaphore.h>
 #include <time.h>
 #include <assert.h>
+#include <signal.h>
+#include <urcu/uatomic.h>
+#include <urcu/futex.h>
+
+#include <lttng-ust-comm.h>
+#include <ust/lttng-events.h>
+#include <ust/usterr-signal-safe.h>
+#include <ust/lttng-ust-abi.h>
+#include <ust/tracepoint.h>
+#include <ust/tracepoint-internal.h>
+#include <ust/ust.h>
+#include "ltt-tracer-core.h"
+
+/*
+ * Has lttng ust comm constructor been called ?
+ */
+static int initialized;
 
 /*
- * communication thread mutex. Held when handling a command, also held
- * by fork() to deal with removal of threads, and by exit path.
+ * The ust_lock/ust_unlock lock is used as a communication thread mutex.
+ * Held when handling a command, also held by fork() to deal with
+ * removal of threads, and by exit path.
  */
-static pthread_mutex_t lttng_ust_comm_mutex = PTHREAD_MUTEX_INITIALIZER;
 
 /* Should the ust comm thread quit ? */
 static int lttng_ust_comm_should_quit;
@@ -51,44 +72,86 @@ static int lttng_ust_comm_should_quit;
  *   daemon problems).
  */
 static sem_t constructor_wait;
+/*
+ * Doing this for both the global and local sessiond.
+ */
+static int sem_count = { 2 };
 
 /*
  * Info about socket and associated listener thread.
  */
 struct sock_info {
 	const char *name;
-	char sock_path[PATH_MAX];
-	int socket;
 	pthread_t ust_listener;	/* listener thread */
 	int root_handle;
+	int constructor_sem_posted;
+	int allowed;
+	int global;
+
+	char sock_path[PATH_MAX];
+	int socket;
+
+	char wait_shm_path[PATH_MAX];
+	char *wait_shm_mmap;
 };
 
 /* Socket from app (connect) to session daemon (listen) for communication */
 struct sock_info global_apps = {
 	.name = "global",
+	.global = 1,
+
+	.root_handle = -1,
+	.allowed = 1,
+
 	.sock_path = DEFAULT_GLOBAL_APPS_UNIX_SOCK,
 	.socket = -1,
-	.root_handle = -1,
+
+	.wait_shm_path = DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH,
 };
 
 /* TODO: allow global_apps_sock_path override */
 
 struct sock_info local_apps = {
 	.name = "local",
-	.socket = -1,
+	.global = 0,
 	.root_handle = -1,
+	.allowed = 0,	/* Check setuid bit first */
+
+	.socket = -1,
 };
 
+static int wait_poll_fallback;
+
+extern void ltt_ring_buffer_client_overwrite_init(void);
+extern void ltt_ring_buffer_client_discard_init(void);
+extern void ltt_ring_buffer_metadata_client_init(void);
+extern void ltt_ring_buffer_client_overwrite_exit(void);
+extern void ltt_ring_buffer_client_discard_exit(void);
+extern void ltt_ring_buffer_metadata_client_exit(void);
+
 static
-int setup_local_apps_socket(void)
+int setup_local_apps(void)
 {
 	const char *home_dir;
+	uid_t uid;
 
+	uid = getuid();
+	/*
+	 * Disallow per-user tracing for setuid binaries.
+	 */
+	if (uid != geteuid()) {
+		local_apps.allowed = 0;
+		return 0;
+	} else {
+		local_apps.allowed = 1;
+	}
 	home_dir = (const char *) getenv("HOME");
 	if (!home_dir)
 		return -ENOENT;
 	snprintf(local_apps.sock_path, PATH_MAX,
 		 DEFAULT_HOME_APPS_UNIX_SOCK, home_dir);
+	snprintf(local_apps.wait_shm_path, PATH_MAX,
+		 DEFAULT_HOME_APPS_WAIT_SHM_PATH, uid);
 	return 0;
 }
 
@@ -96,30 +159,41 @@ static
 int register_app_to_sessiond(int socket)
 {
 	ssize_t ret;
+	int prctl_ret;
 	struct {
 		uint32_t major;
 		uint32_t minor;
 		pid_t pid;
+		pid_t ppid;
 		uid_t uid;
+		gid_t gid;
+		char name[16];	/* process name */
 	} reg_msg;
 
 	reg_msg.major = LTTNG_UST_COMM_VERSION_MAJOR;
 	reg_msg.minor = LTTNG_UST_COMM_VERSION_MINOR;
 	reg_msg.pid = getpid();
+	reg_msg.ppid = getppid();
 	reg_msg.uid = getuid();
+	reg_msg.gid = getgid();
+	prctl_ret = prctl(PR_GET_NAME, (unsigned long) reg_msg.name, 0, 0, 0);
+	if (prctl_ret) {
+		ERR("Error executing prctl");
+		return -errno;
+	}
 
-	ret = lttcomm_send_unix_sock(socket, &reg_msg, sizeof(reg_msg));
+	ret = ustcomm_send_unix_sock(socket, &reg_msg, sizeof(reg_msg));
 	if (ret >= 0 && ret != sizeof(reg_msg))
 		return -EIO;
 	return ret;
 }
 
 static
-int send_reply(int sock, struct lttcomm_ust_reply *lur)
+int send_reply(int sock, struct ustcomm_ust_reply *lur)
 {
 	ssize_t len;
 
-	len = lttcomm_send_unix_sock(sock, lur, sizeof(*lur));
+	len = ustcomm_send_unix_sock(sock, lur, sizeof(*lur));
 	switch (len) {
 	case sizeof(*lur):
 		DBG("message successfully sent");
@@ -137,24 +211,34 @@ int send_reply(int sock, struct lttcomm_ust_reply *lur)
 }
 
 static
-int handle_register_done(void)
+int handle_register_done(struct sock_info *sock_info)
 {
 	int ret;
 
-	ret = sem_post(&constructor_wait);
-	assert(!ret);
+	if (sock_info->constructor_sem_posted)
+		return 0;
+	sock_info->constructor_sem_posted = 1;
+	if (uatomic_read(&sem_count) <= 0) {
+		return 0;
+	}
+	ret = uatomic_add_return(&sem_count, -1);
+	if (ret == 0) {
+		ret = sem_post(&constructor_wait);
+		assert(!ret);
+	}
 	return 0;
 }
 
 static
 int handle_message(struct sock_info *sock_info,
-		int sock, struct lttcomm_ust_msg *lum)
+		int sock, struct ustcomm_ust_msg *lum)
 {
 	int ret = 0;
-	const struct objd_ops *ops;
-	struct lttcomm_ust_reply lur;
+	const struct lttng_ust_objd_ops *ops;
+	struct ustcomm_ust_reply lur;
+	int shm_fd, wait_fd;
 
-	pthread_mutex_lock(&lttng_ust_comm_mutex);
+	ust_lock();
 
 	memset(&lur, 0, sizeof(lur));
 
@@ -172,7 +256,7 @@ int handle_message(struct sock_info *sock_info,
 	switch (lum->cmd) {
 	case LTTNG_UST_REGISTER_DONE:
 		if (lum->handle == LTTNG_UST_ROOT_HANDLE)
-			ret = handle_register_done();
+			ret = handle_register_done(sock_info);
 		else
 			ret = -EINVAL;
 		break;
@@ -180,7 +264,7 @@ int handle_message(struct sock_info *sock_info,
 		if (lum->handle == LTTNG_UST_ROOT_HANDLE)
 			ret = -EPERM;
 		else
-			ret = objd_unref(lum->handle);
+			ret = lttng_ust_objd_unref(lum->handle);
 		break;
 	default:
 		if (ops->cmd)
@@ -196,13 +280,59 @@ end:
 	lur.cmd = lum->cmd;
 	lur.ret_val = ret;
 	if (ret >= 0) {
-		lur.ret_code = LTTCOMM_OK;
+		lur.ret_code = USTCOMM_OK;
 	} else {
-		lur.ret_code = LTTCOMM_SESSION_FAIL;
+		//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 = lum->u.stream.memory_map_size;
+		shm_fd = lum->u.stream.shm_fd;
+		wait_fd = lum->u.stream.wait_fd;
+		break;
+	case LTTNG_UST_METADATA:
+	case LTTNG_UST_CHANNEL:
+		lur.u.channel.memory_map_size = lum->u.channel.memory_map_size;
+		shm_fd = lum->u.channel.shm_fd;
+		wait_fd = lum->u.channel.wait_fd;
+		break;
+	case LTTNG_UST_VERSION:
+		lur.u.version = lum->u.version;
+		break;
 	}
 	ret = send_reply(sock, &lur);
+	if (ret < 0) {
+		perror("error sending reply");
+		goto error;
+	}
 
-	pthread_mutex_unlock(&lttng_ust_comm_mutex);
+	if ((lum->cmd == LTTNG_UST_STREAM
+	     || lum->cmd == LTTNG_UST_CHANNEL
+	     || lum->cmd == LTTNG_UST_METADATA)
+			&& lur.ret_code == USTCOMM_OK) {
+		/* 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;
+		}
+		ret = ustcomm_send_fds_unix_sock(sock,
+			&wait_fd, &wait_fd,
+			1, sizeof(int));
+		if (ret < 0) {
+			perror("send wait_fd");
+			goto error;
+		}
+	}
+error:
+	ust_unlock();
 	return ret;
 }
 
@@ -214,17 +344,237 @@ void cleanup_sock_info(struct sock_info *sock_info)
 	if (sock_info->socket != -1) {
 		ret = close(sock_info->socket);
 		if (ret) {
-			ERR("Error closing local apps socket");
+			ERR("Error closing apps socket");
 		}
 		sock_info->socket = -1;
 	}
 	if (sock_info->root_handle != -1) {
-		ret = objd_unref(sock_info->root_handle);
+		ret = lttng_ust_objd_unref(sock_info->root_handle);
 		if (ret) {
 			ERR("Error unref root handle");
 		}
 		sock_info->root_handle = -1;
 	}
+	sock_info->constructor_sem_posted = 0;
+	if (sock_info->wait_shm_mmap) {
+		ret = munmap(sock_info->wait_shm_mmap, sysconf(_SC_PAGE_SIZE));
+		if (ret) {
+			ERR("Error unmapping wait shm");
+		}
+		sock_info->wait_shm_mmap = NULL;
+	}
+}
+
+/*
+ * Using fork to set umask in the child process (not multi-thread safe).
+ * We deal with the shm_open vs ftruncate race (happening when the
+ * sessiond owns the shm and does not let everybody modify it, to ensure
+ * safety against shm_unlink) by simply letting the mmap fail and
+ * retrying after a few seconds.
+ * For global shm, everybody has rw access to it until the sessiond
+ * starts.
+ */
+static
+int get_wait_shm(struct sock_info *sock_info, size_t mmap_size)
+{
+	int wait_shm_fd, ret;
+	pid_t pid;
+
+	/*
+	 * Try to open read-only.
+	 */
+	wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
+	if (wait_shm_fd >= 0) {
+		goto end;
+	} else if (wait_shm_fd < 0 && errno != ENOENT) {
+		/*
+		 * Real-only open did not work, and it's not because the
+		 * entry was not present. It's a failure that prohibits
+		 * using shm.
+		 */
+		ERR("Error opening shm %s", sock_info->wait_shm_path);
+		goto end;
+	}
+	/*
+	 * If the open failed because the file did not exist, try
+	 * creating it ourself.
+	 */
+	pid = fork();
+	if (pid > 0) {
+		int status;
+
+		/*
+		 * Parent: wait for child to return, in which case the
+		 * shared memory map will have been created.
+		 */
+		pid = wait(&status);
+		if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
+			wait_shm_fd = -1;
+			goto end;
+		}
+		/*
+		 * Try to open read-only again after creation.
+		 */
+		wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
+		if (wait_shm_fd < 0) {
+			/*
+			 * Real-only open did not work. It's a failure
+			 * that prohibits using shm.
+			 */
+			ERR("Error opening shm %s", sock_info->wait_shm_path);
+			goto end;
+		}
+		goto end;
+	} else if (pid == 0) {
+		int create_mode;
+
+		/* Child */
+		create_mode = S_IRUSR | S_IWUSR | S_IRGRP;
+		if (sock_info->global)
+			create_mode |= S_IROTH | S_IWGRP | S_IWOTH;
+		/*
+		 * We're alone in a child process, so we can modify the
+		 * process-wide umask.
+		 */
+		umask(~create_mode);
+		/*
+		 * Try creating shm (or get rw access).
+		 * We don't do an exclusive open, because we allow other
+		 * processes to create+ftruncate it concurrently.
+		 */
+		wait_shm_fd = shm_open(sock_info->wait_shm_path,
+				O_RDWR | O_CREAT, create_mode);
+		if (wait_shm_fd >= 0) {
+			ret = ftruncate(wait_shm_fd, mmap_size);
+			if (ret) {
+				PERROR("ftruncate");
+				exit(EXIT_FAILURE);
+			}
+			exit(EXIT_SUCCESS);
+		}
+		/*
+		 * For local shm, we need to have rw access to accept
+		 * opening it: this means the local sessiond will be
+		 * able to wake us up. For global shm, we open it even
+		 * if rw access is not granted, because the root.root
+		 * sessiond will be able to override all rights and wake
+		 * us up.
+		 */
+		if (!sock_info->global && errno != EACCES) {
+			ERR("Error opening shm %s", sock_info->wait_shm_path);
+			exit(EXIT_FAILURE);
+		}
+		/*
+		 * The shm exists, but we cannot open it RW. Report
+		 * success.
+		 */
+		exit(EXIT_SUCCESS);
+	} else {
+		return -1;
+	}
+end:
+	if (wait_shm_fd >= 0 && !sock_info->global) {
+		struct stat statbuf;
+
+		/*
+		 * Ensure that our user is the owner of the shm file for
+		 * local shm. If we do not own the file, it means our
+		 * sessiond will not have access to wake us up (there is
+		 * probably a rogue process trying to fake our
+		 * sessiond). Fallback to polling method in this case.
+		 */
+		ret = fstat(wait_shm_fd, &statbuf);
+		if (ret) {
+			PERROR("fstat");
+			goto error_close;
+		}
+		if (statbuf.st_uid != getuid())
+			goto error_close;
+	}
+	return wait_shm_fd;
+
+error_close:
+	ret = close(wait_shm_fd);
+	if (ret) {
+		PERROR("Error closing fd");
+	}
+	return -1;
+}
+
+static
+char *get_map_shm(struct sock_info *sock_info)
+{
+	size_t mmap_size = sysconf(_SC_PAGE_SIZE);
+	int wait_shm_fd, ret;
+	char *wait_shm_mmap;
+
+	wait_shm_fd = get_wait_shm(sock_info, mmap_size);
+	if (wait_shm_fd < 0) {
+		goto error;
+	}
+	wait_shm_mmap = mmap(NULL, mmap_size, PROT_READ,
+		  MAP_SHARED, wait_shm_fd, 0);
+	/* close shm fd immediately after taking the mmap reference */
+	ret = close(wait_shm_fd);
+	if (ret) {
+		PERROR("Error closing fd");
+	}
+	if (wait_shm_mmap == MAP_FAILED) {
+		DBG("mmap error (can be caused by race with sessiond). Fallback to poll mode.");
+		goto error;
+	}
+	return wait_shm_mmap;
+
+error:
+	return NULL;
+}
+
+static
+void wait_for_sessiond(struct sock_info *sock_info)
+{
+	int ret;
+
+	ust_lock();
+	if (lttng_ust_comm_should_quit) {
+		goto quit;
+	}
+	if (wait_poll_fallback) {
+		goto error;
+	}
+	if (!sock_info->wait_shm_mmap) {
+		sock_info->wait_shm_mmap = get_map_shm(sock_info);
+		if (!sock_info->wait_shm_mmap)
+			goto error;
+	}
+	ust_unlock();
+
+	DBG("Waiting for %s apps sessiond", sock_info->name);
+	/* Wait for futex wakeup */
+	if (uatomic_read((int32_t *) sock_info->wait_shm_mmap) == 0) {
+		ret = futex_async((int32_t *) sock_info->wait_shm_mmap,
+			FUTEX_WAIT, 0, NULL, NULL, 0);
+		if (ret < 0) {
+			if (errno == EFAULT) {
+				wait_poll_fallback = 1;
+				WARN(
+"Linux kernels 2.6.33 to 3.0 (with the exception of stable versions) "
+"do not support FUTEX_WAKE on read-only memory mappings correctly. "
+"Please upgrade your kernel "
+"(fix is commit 9ea71503a8ed9184d2d0b8ccc4d269d05f7940ae in Linux kernel "
+"mainline). LTTng-UST will use polling mode fallback.");
+			}
+			PERROR("futex");
+		}
+	}
+	return;
+
+quit:
+	ust_unlock();
+	return;
+
+error:
+	ust_unlock();
+	return;
 }
 
 /*
@@ -238,14 +588,29 @@ static
 void *ust_listener_thread(void *arg)
 {
 	struct sock_info *sock_info = arg;
-	int sock, ret;
+	int sock, ret, prev_connect_failed = 0, has_waited = 0;
 
 	/* Restart trying to connect to the session daemon */
 restart:
-	pthread_mutex_lock(&lttng_ust_comm_mutex);
+	if (prev_connect_failed) {
+		/* Wait for sessiond availability with pipe */
+		wait_for_sessiond(sock_info);
+		if (has_waited) {
+			has_waited = 0;
+			/*
+			 * Sleep for 5 seconds before retrying after a
+			 * sequence of failure / wait / failure. This
+			 * deals with a killed or broken session daemon.
+			 */
+			sleep(5);
+		}
+		has_waited = 1;
+		prev_connect_failed = 0;
+	}
+	ust_lock();
 
 	if (lttng_ust_comm_should_quit) {
-		pthread_mutex_unlock(&lttng_ust_comm_mutex);
+		ust_unlock();
 		goto quit;
 	}
 
@@ -257,20 +622,18 @@ restart:
 		sock_info->socket = -1;
 	}
 
-	/* Check for sessiond availability with pipe TODO */
-
 	/* Register */
-	ret = lttcomm_connect_unix_sock(sock_info->sock_path);
+	ret = ustcomm_connect_unix_sock(sock_info->sock_path);
 	if (ret < 0) {
 		ERR("Error connecting to %s apps socket", sock_info->name);
+		prev_connect_failed = 1;
 		/*
 		 * If we cannot find the sessiond daemon, don't delay
 		 * constructor execution.
 		 */
-		ret = handle_register_done();
+		ret = handle_register_done(sock_info);
 		assert(!ret);
-		pthread_mutex_unlock(&lttng_ust_comm_mutex);
-		sleep(5);
+		ust_unlock();
 		goto restart;
 	}
 
@@ -282,9 +645,9 @@ restart:
 	 */
 	if (sock_info->root_handle == -1) {
 		ret = lttng_abi_create_root_handle();
-		if (ret) {
+		if (ret < 0) {
 			ERR("Error creating root handle");
-			pthread_mutex_unlock(&lttng_ust_comm_mutex);
+			ust_unlock();
 			goto quit;
 		}
 		sock_info->root_handle = ret;
@@ -293,23 +656,23 @@ restart:
 	ret = register_app_to_sessiond(sock);
 	if (ret < 0) {
 		ERR("Error registering to %s apps socket", sock_info->name);
+		prev_connect_failed = 1;
 		/*
 		 * If we cannot register to the sessiond daemon, don't
 		 * delay constructor execution.
 		 */
-		ret = handle_register_done();
+		ret = handle_register_done(sock_info);
 		assert(!ret);
-		pthread_mutex_unlock(&lttng_ust_comm_mutex);
-		sleep(5);
+		ust_unlock();
 		goto restart;
 	}
-	pthread_mutex_unlock(&lttng_ust_comm_mutex);
+	ust_unlock();
 
 	for (;;) {
 		ssize_t len;
-		struct lttcomm_ust_msg lum;
+		struct ustcomm_ust_msg lum;
 
-		len = lttcomm_recv_unix_sock(sock, &lum, sizeof(lum));
+		len = ustcomm_recv_unix_sock(sock, &lum, sizeof(lum));
 		switch (len) {
 		case 0:	/* orderly shutdown */
 			DBG("%s ltt-sessiond has performed an orderly shutdown\n", sock_info->name);
@@ -339,30 +702,44 @@ quit:
 	return NULL;
 }
 
+/*
+ * Return values: -1: don't wait. 0: wait forever. 1: timeout wait.
+ */
 static
 int get_timeout(struct timespec *constructor_timeout)
 {
-	struct timespec constructor_delay =
-		{
-			.tv_sec = LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_S,
-		  	.tv_nsec = LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_NS,
-		};
-	struct timespec realtime;
+	long constructor_delay_ms = LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS;
+	char *str_delay;
 	int ret;
 
-	ret = clock_gettime(CLOCK_REALTIME, &realtime);
-	if (ret)
-		return ret;
+	str_delay = getenv("UST_REGISTER_TIMEOUT");
+	if (str_delay) {
+		constructor_delay_ms = strtol(str_delay, NULL, 10);
+	}
+
+	switch (constructor_delay_ms) {
+	case -1:/* fall-through */
+	case 0:
+		return constructor_delay_ms;
+	default:
+		break;
+	}
 
-	constructor_timeout->tv_sec =
-		realtime.tv_sec + constructor_delay.tv_sec;
-	constructor_timeout->tv_nsec =
-		constructor_delay.tv_nsec + realtime.tv_nsec;
+	/*
+	 * If we are unable to find the current time, don't wait.
+	 */
+	ret = clock_gettime(CLOCK_REALTIME, constructor_timeout);
+	if (ret) {
+		return -1;
+	}
+	constructor_timeout->tv_sec += constructor_delay_ms / 1000UL;
+	constructor_timeout->tv_nsec +=
+		(constructor_delay_ms % 1000UL) * 1000000UL;
 	if (constructor_timeout->tv_nsec >= 1000000000UL) {
 		constructor_timeout->tv_sec++;
 		constructor_timeout->tv_nsec -= 1000000000UL;
 	}
-	return 0;
+	return 1;
 }
 
 /*
@@ -371,47 +748,91 @@ int get_timeout(struct timespec *constructor_timeout)
  */
 /* TODO */
 
-void __attribute__((constructor)) lttng_ust_comm_init(void)
+void __attribute__((constructor)) lttng_ust_init(void)
 {
 	struct timespec constructor_timeout;
+	int timeout_mode;
 	int ret;
 
+	if (uatomic_xchg(&initialized, 1) == 1)
+		return;
+
+	/*
+	 * We want precise control over the order in which we construct
+	 * our sub-libraries vs starting to receive commands from
+	 * sessiond (otherwise leading to errors when trying to create
+	 * sessiond before the init functions are completed).
+	 */
 	init_usterr();
+	init_tracepoint();
+	ltt_ring_buffer_metadata_client_init();
+	ltt_ring_buffer_client_overwrite_init();
+	ltt_ring_buffer_client_discard_init();
 
-	ret = get_timeout(&constructor_timeout);
-	assert(!ret);
+	timeout_mode = get_timeout(&constructor_timeout);
 
-	ret = sem_init(&constructor_wait, 0, 2);
+	ret = sem_init(&constructor_wait, 0, 0);
 	assert(!ret);
 
-	ret = setup_local_apps_socket();
+	ret = setup_local_apps();
 	if (ret) {
-		ERR("Error setting up to local apps socket");
+		ERR("Error setting up to local apps");
 	}
-
-	/*
-	 * Wait for the pthread cond to let us continue to main program
-	 * execution. Hold mutex across thread creation, so we start
-	 * waiting for the condition before the threads can signal its
-	 * completion.
-	 */
-	pthread_mutex_lock(&lttng_ust_comm_mutex);
-	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);
 
-	ret = sem_timedwait(&constructor_wait, &constructor_timeout);
-	if (ret < 0 && errno == ETIMEDOUT) {
-		ERR("Timed out waiting for ltt-sessiond");
+	if (local_apps.allowed) {
+		ret = pthread_create(&global_apps.ust_listener, NULL,
+				ust_listener_thread, &global_apps);
 	} else {
+		handle_register_done(&local_apps);
+	}
+
+	switch (timeout_mode) {
+	case 1:	/* timeout wait */
+		do {
+			ret = sem_timedwait(&constructor_wait,
+					&constructor_timeout);
+		} while (ret < 0 && errno == EINTR);
+		if (ret < 0 && errno == ETIMEDOUT) {
+			ERR("Timed out waiting for ltt-sessiond");
+		} else {
+			assert(!ret);
+		}
+		break;
+	case -1:/* wait forever */
+		do {
+			ret = sem_wait(&constructor_wait);
+		} while (ret < 0 && errno == EINTR);
 		assert(!ret);
+		break;
+	case 0:	/* no timeout */
+		break;
 	}
-	pthread_mutex_unlock(&lttng_ust_comm_mutex);
+}
 
+static
+void lttng_ust_cleanup(int exiting)
+{
+	cleanup_sock_info(&global_apps);
+	if (local_apps.allowed) {
+		cleanup_sock_info(&local_apps);
+	}
+	lttng_ust_abi_exit();
+	lttng_ust_events_exit();
+	ltt_ring_buffer_client_discard_exit();
+	ltt_ring_buffer_client_overwrite_exit();
+	ltt_ring_buffer_metadata_client_exit();
+	exit_tracepoint();
+	if (!exiting) {
+		/* Reinitialize values for fork */
+		sem_count = 2;
+		lttng_ust_comm_should_quit = 0;
+		initialized = 0;
+	}
 }
 
-void __attribute__((destructor)) lttng_ust_comm_exit(void)
+void __attribute__((destructor)) lttng_ust_exit(void)
 {
 	int ret;
 
@@ -426,26 +847,89 @@ void __attribute__((destructor)) lttng_ust_comm_exit(void)
 	 * mutexes to ensure it is not in a mutex critical section when
 	 * pthread_cancel is later called.
 	 */
-	pthread_mutex_lock(&lttng_ust_comm_mutex);
+	ust_lock();
 	lttng_ust_comm_should_quit = 1;
-	pthread_mutex_unlock(&lttng_ust_comm_mutex);
+	ust_unlock();
 
-#if 0
 	ret = pthread_cancel(global_apps.ust_listener);
 	if (ret) {
 		ERR("Error cancelling global ust listener thread");
 	}
-#endif //0
+	if (local_apps.allowed) {
+		ret = pthread_cancel(local_apps.ust_listener);
+		if (ret) {
+			ERR("Error cancelling local ust listener thread");
+		}
+	}
+	lttng_ust_cleanup(1);
+}
 
-	cleanup_sock_info(&global_apps);
+/*
+ * We exclude the worker threads across fork and clone (except
+ * CLONE_VM), because these system calls only keep the forking thread
+ * running in the child.  Therefore, we don't want to call fork or clone
+ * in the middle of an tracepoint or ust tracing state modification.
+ * Holding this mutex protects these structures across fork and clone.
+ */
+void ust_before_fork(ust_fork_info_t *fork_info)
+{
+	/*
+	 * Disable signals. This is to avoid that the child intervenes
+	 * before it is properly setup for tracing. It is safer to
+	 * disable all signals, because then we know we are not breaking
+	 * anything by restoring the original mask.
+         */
+	sigset_t all_sigs;
+	int ret;
 
-	ret = pthread_cancel(local_apps.ust_listener);
-	if (ret) {
-		ERR("Error cancelling local ust listener thread");
+	/* Disable signals */
+	sigfillset(&all_sigs);
+	ret = sigprocmask(SIG_BLOCK, &all_sigs, &fork_info->orig_sigs);
+	if (ret == -1) {
+		PERROR("sigprocmask");
+	}
+	ust_lock();
+	rcu_bp_before_fork();
+}
+
+static void ust_after_fork_common(ust_fork_info_t *fork_info)
+{
+	int ret;
+
+	DBG("process %d", getpid());
+	ust_unlock();
+	/* Restore signals */
+	ret = sigprocmask(SIG_SETMASK, &fork_info->orig_sigs, NULL);
+	if (ret == -1) {
+		PERROR("sigprocmask");
 	}
+}
 
-	cleanup_sock_info(&local_apps);
+void ust_after_fork_parent(ust_fork_info_t *fork_info)
+{
+	DBG("process %d", getpid());
+	rcu_bp_after_fork_parent();
+	/* Release mutexes and reenable signals */
+	ust_after_fork_common(fork_info);
+}
 
-	lttng_ust_abi_exit();
-	ltt_events_exit();
+/*
+ * After fork, in the child, we need to cleanup all the leftover state,
+ * except the worker thread which already magically disappeared thanks
+ * to the weird Linux fork semantics. After tyding up, we call
+ * lttng_ust_init() again to start over as a new PID.
+ *
+ * This is meant for forks() that have tracing in the child between the
+ * fork and following exec call (if there is any).
+ */
+void ust_after_fork_child(ust_fork_info_t *fork_info)
+{
+	DBG("process %d", getpid());
+	/* Release urcu mutexes */
+	rcu_bp_after_fork_child();
+	lttng_ust_cleanup(0);
+	lttng_context_vtid_reset();
+	/* Release mutexes and reenable signals */
+	ust_after_fork_common(fork_info);
+	lttng_ust_init();
 }