X-Git-Url: http://git.efficios.com/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fmain.c;h=550aa1b83f68d65f01e936d1e81fb5308b0b2b4e;hp=d0b1e17e544ef21b110381ea266f5037172e80f0;hb=55d097957f5bb8138959ad2202a40d85d49f029e;hpb=8d3113b21fe5ed84af58fbfdff9ad7742db507d5 diff --git a/src/bin/lttng-sessiond/main.c b/src/bin/lttng-sessiond/main.c index d0b1e17e5..550aa1b83 100644 --- a/src/bin/lttng-sessiond/main.c +++ b/src/bin/lttng-sessiond/main.c @@ -21,11 +21,11 @@ #include #include #include -#include #include #include #include #include +#include #include #include #include @@ -38,7 +38,6 @@ #include #include -#include #include #include #include @@ -47,7 +46,9 @@ #include #include "lttng-sessiond.h" +#include "buffer-registry.h" #include "channel.h" +#include "cmd.h" #include "consumer.h" #include "context.h" #include "event.h" @@ -59,19 +60,18 @@ #include "ust-consumer.h" #include "utils.h" #include "fd-limit.h" -#include "filter.h" -#include "health.h" +#include "health-sessiond.h" +#include "testpoint.h" +#include "ust-thread.h" #define CONSUMERD_FILE "lttng-consumerd" /* Const values */ -const char default_home_dir[] = DEFAULT_HOME_DIR; const char default_tracing_group[] = DEFAULT_TRACING_GROUP; -const char default_ust_sock_dir[] = DEFAULT_UST_SOCK_DIR; -const char default_global_apps_pipe[] = DEFAULT_GLOBAL_APPS_PIPE; const char *progname; const char *opt_tracing_group; +static const char *opt_pidfile; static int opt_sig_parent; static int opt_verbose_consumer; static int opt_daemon; @@ -80,7 +80,10 @@ static int is_root; /* Set to 1 if the daemon is running as root */ static pid_t ppid; /* Parent PID for --sig-parent option */ static char *rundir; -/* Consumer daemon specific control data */ +/* + * Consumer daemon specific control data. Every value not initialized here is + * set to 0 by the static definition. + */ static struct consumer_data kconsumer_data = { .type = LTTNG_CONSUMER_KERNEL, .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH, @@ -89,6 +92,8 @@ static struct consumer_data kconsumer_data = { .cmd_sock = -1, .pid_mutex = PTHREAD_MUTEX_INITIALIZER, .lock = PTHREAD_MUTEX_INITIALIZER, + .cond = PTHREAD_COND_INITIALIZER, + .cond_mutex = PTHREAD_MUTEX_INITIALIZER, }; static struct consumer_data ustconsumer64_data = { .type = LTTNG_CONSUMER64_UST, @@ -98,6 +103,8 @@ static struct consumer_data ustconsumer64_data = { .cmd_sock = -1, .pid_mutex = PTHREAD_MUTEX_INITIALIZER, .lock = PTHREAD_MUTEX_INITIALIZER, + .cond = PTHREAD_COND_INITIALIZER, + .cond_mutex = PTHREAD_MUTEX_INITIALIZER, }; static struct consumer_data ustconsumer32_data = { .type = LTTNG_CONSUMER32_UST, @@ -107,6 +114,8 @@ static struct consumer_data ustconsumer32_data = { .cmd_sock = -1, .pid_mutex = PTHREAD_MUTEX_INITIALIZER, .lock = PTHREAD_MUTEX_INITIALIZER, + .cond = PTHREAD_COND_INITIALIZER, + .cond_mutex = PTHREAD_MUTEX_INITIALIZER, }; /* Shared between threads */ @@ -124,7 +133,7 @@ static char health_unix_sock_path[PATH_MAX]; /* Sockets and FDs */ static int client_sock = -1; static int apps_sock = -1; -static int kernel_tracer_fd = -1; +int kernel_tracer_fd = -1; static int kernel_poll_pipe[2] = { -1, -1 }; /* @@ -139,13 +148,17 @@ static int thread_quit_pipe[2] = { -1, -1 }; */ static int apps_cmd_pipe[2] = { -1, -1 }; +int apps_cmd_notify_pipe[2] = { -1, -1 }; + /* Pthread, Mutexes and Semaphores */ static pthread_t apps_thread; +static pthread_t apps_notify_thread; static pthread_t reg_apps_thread; static pthread_t client_thread; static pthread_t kernel_thread; static pthread_t dispatch_thread; static pthread_t health_thread; +static pthread_t ht_cleanup_thread; /* * UST registration command queue. This queue is tied with a futex and uses a N @@ -176,6 +189,8 @@ static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN; static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR; static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR; +static const char *module_proc_lttng = "/proc/lttng"; + /* * Consumer daemon state which is changed when spawning it, killing it or in * case of a fatal error. @@ -211,20 +226,15 @@ static enum consumerd_state ust_consumerd_state; static enum consumerd_state kernel_consumerd_state; /* - * Used to keep a unique index for each relayd socket created where this value - * is associated with streams on the consumer so it can match the right relayd - * to send to. - * - * This value should be incremented atomically for safety purposes and future - * possible concurrent access. + * Socket timeout for receiving and sending in seconds. */ -static unsigned int relayd_net_seq_idx; +static int app_socket_timeout; + +/* Set in main() with the current page size. */ +long page_size; -/* Used for the health monitoring of the session daemon. See health.h */ -struct health_state health_thread_cmd; -struct health_state health_thread_app_manage; -struct health_state health_thread_app_reg; -struct health_state health_thread_kernel; +/* Application health monitoring */ +struct health_app *health_sessiond; static void setup_consumerd_path(void) @@ -278,15 +288,11 @@ void setup_consumerd_path(void) /* * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set. */ -static int create_thread_poll_set(struct lttng_poll_event *events, - unsigned int size) +int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size) { int ret; - if (events == NULL || size == 0) { - ret = -1; - goto error; - } + assert(events); ret = lttng_poll_create(events, size, LTTNG_CLOEXEC); if (ret < 0) { @@ -294,7 +300,7 @@ static int create_thread_poll_set(struct lttng_poll_event *events, } /* Add quit pipe */ - ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN); + ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR); if (ret < 0) { goto error; } @@ -310,7 +316,7 @@ error: * * Return 1 if it was triggered else 0; */ -static int check_thread_quit_pipe(int fd, uint32_t events) +int sessiond_check_thread_quit_pipe(int fd, uint32_t events) { if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) { return 1; @@ -366,127 +372,67 @@ error: } /* - * Complete teardown of a kernel session. This free all data structure related - * to a kernel session and update counter. + * Stop all threads by closing the thread quit pipe. */ -static void teardown_kernel_session(struct ltt_session *session) +static void stop_threads(void) { int ret; - struct lttng_ht_iter iter; - struct ltt_kernel_session *ksess; - struct consumer_socket *socket; - - if (!session->kernel_session) { - DBG3("No kernel session when tearing down session"); - return; - } - - ksess = session->kernel_session; - DBG("Tearing down kernel session"); - - /* - * Destroy relayd associated with the session consumer. This action is - * valid since in order to destroy a session we must acquire the session - * lock. This means that there CAN NOT be stream(s) being sent to a - * consumer since this action also requires the session lock at any time. - * - * At this point, we are sure that not streams data will be lost after this - * command is issued. - */ - if (ksess->consumer && ksess->consumer->type == CONSUMER_DST_NET) { - cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter, socket, - node.node) { - ret = consumer_send_destroy_relayd(socket, ksess->consumer); - if (ret < 0) { - ERR("Unable to send destroy relayd command to consumer"); - /* Continue since we MUST delete everything at this point. */ - } - } - } - - /* - * If a custom kernel consumer was registered, close the socket before - * tearing down the complete kernel session structure - */ - cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter, socket, - node.node) { - if (socket->fd != kconsumer_data.cmd_sock) { - rcu_read_lock(); - consumer_del_socket(socket, ksess->consumer); - lttcomm_close_unix_sock(socket->fd); - consumer_destroy_socket(socket); - rcu_read_unlock(); - } + /* Stopping all threads */ + DBG("Terminating all threads"); + ret = notify_thread_pipe(thread_quit_pipe[1]); + if (ret < 0) { + ERR("write error on thread quit pipe"); } - trace_kernel_destroy_session(ksess); + /* Dispatch thread */ + CMM_STORE_SHARED(dispatch_thread_exit, 1); + futex_nto1_wake(&ust_cmd_queue.futex); } /* - * Complete teardown of all UST sessions. This will free everything on his path - * and destroy the core essence of all ust sessions :) + * Close every consumer sockets. */ -static void teardown_ust_session(struct ltt_session *session) +static void close_consumer_sockets(void) { int ret; - struct lttng_ht_iter iter; - struct ltt_ust_session *usess; - struct consumer_socket *socket; - if (!session->ust_session) { - DBG3("No UST session when tearing down session"); - return; + if (kconsumer_data.err_sock >= 0) { + ret = close(kconsumer_data.err_sock); + if (ret < 0) { + PERROR("kernel consumer err_sock close"); + } } - usess = session->ust_session; - - DBG("Tearing down UST session(s)"); - - /* - * Destroy relayd associated with the session consumer. This action is - * valid since in order to destroy a session we must acquire the session - * lock. This means that there CAN NOT be stream(s) being sent to a - * consumer since this action also requires the session lock at any time. - * - * At this point, we are sure that no data will be lost after this command - * is issued. - */ - if (usess->consumer && usess->consumer->type == CONSUMER_DST_NET) { - cds_lfht_for_each_entry(usess->consumer->socks->ht, &iter.iter, socket, - node.node) { - ret = consumer_send_destroy_relayd(socket, usess->consumer); - if (ret < 0) { - ERR("Unable to send destroy relayd command to consumer"); - /* Continue since we MUST delete everything at this point. */ - } + if (ustconsumer32_data.err_sock >= 0) { + ret = close(ustconsumer32_data.err_sock); + if (ret < 0) { + PERROR("UST consumerd32 err_sock close"); } } - - ret = ust_app_destroy_trace_all(usess); - if (ret) { - ERR("Error in ust_app_destroy_trace_all"); + if (ustconsumer64_data.err_sock >= 0) { + ret = close(ustconsumer64_data.err_sock); + if (ret < 0) { + PERROR("UST consumerd64 err_sock close"); + } } - - trace_ust_destroy_session(usess); -} - -/* - * Stop all threads by closing the thread quit pipe. - */ -static void stop_threads(void) -{ - int ret; - - /* Stopping all threads */ - DBG("Terminating all threads"); - ret = notify_thread_pipe(thread_quit_pipe[1]); - if (ret < 0) { - ERR("write error on thread quit pipe"); + if (kconsumer_data.cmd_sock >= 0) { + ret = close(kconsumer_data.cmd_sock); + if (ret < 0) { + PERROR("kernel consumer cmd_sock close"); + } + } + if (ustconsumer32_data.cmd_sock >= 0) { + ret = close(ustconsumer32_data.cmd_sock); + if (ret < 0) { + PERROR("UST consumerd32 cmd_sock close"); + } + } + if (ustconsumer64_data.cmd_sock >= 0) { + ret = close(ustconsumer64_data.cmd_sock); + if (ret < 0) { + PERROR("UST consumerd64 cmd_sock close"); + } } - - /* Dispatch thread */ - CMM_STORE_SHARED(dispatch_thread_exit, 1); - futex_nto1_wake(&ust_cmd_queue.futex); } /* @@ -495,11 +441,28 @@ static void stop_threads(void) static void cleanup(void) { int ret; - char *cmd; + char *cmd = NULL; struct ltt_session *sess, *stmp; DBG("Cleaning up"); + /* + * Close the thread quit pipe. It has already done its job, + * since we are now called. + */ + utils_close_pipe(thread_quit_pipe); + + /* + * If opt_pidfile is undefined, the default file will be wiped when + * removing the rundir. + */ + if (opt_pidfile) { + ret = remove(opt_pidfile); + if (ret < 0) { + PERROR("remove pidfile %s", opt_pidfile); + } + } + DBG("Removing %s directory", rundir); ret = asprintf(&cmd, "rm -rf %s", rundir); if (ret < 0) { @@ -512,6 +475,7 @@ static void cleanup(void) ERR("Unable to clean %s", rundir); } free(cmd); + free(rundir); DBG("Cleaning up all sessions"); @@ -522,14 +486,13 @@ static void cleanup(void) /* Cleanup ALL session */ cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) { - teardown_kernel_session(sess); - teardown_ust_session(sess); - free(sess); + cmd_destroy_session(sess, kernel_poll_pipe[1]); } } DBG("Closing all UST sockets"); ust_app_clean_list(); + buffer_reg_destroy_registries(); if (is_root && !opt_no_kernel) { DBG2("Closing kernel fd"); @@ -542,9 +505,8 @@ static void cleanup(void) DBG("Unloading kernel modules"); modprobe_remove_lttng_all(); } - utils_close_pipe(kernel_poll_pipe); - utils_close_pipe(thread_quit_pipe); - utils_close_pipe(apps_cmd_pipe); + + close_consumer_sockets(); /* */ DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm" @@ -561,7 +523,7 @@ static void cleanup(void) static int send_unix_sock(int sock, void *buf, size_t len) { /* Check valid length */ - if (len <= 0) { + if (len == 0) { return -1; } @@ -716,6 +678,8 @@ static int update_kernel_stream(struct consumer_data *consumer_data, int fd) if (ret < 0) { goto error; } + /* Update the stream global counter */ + ksess->stream_count_global += ret; /* * Have we already sent fds to the consumer? If yes, it means @@ -726,20 +690,20 @@ static int update_kernel_stream(struct consumer_data *consumer_data, int fd) struct lttng_ht_iter iter; struct consumer_socket *socket; - + rcu_read_lock(); cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter, socket, node.node) { - /* Code flow error */ - assert(socket->fd >= 0); - pthread_mutex_lock(socket->lock); - ret = kernel_consumer_send_channel_stream(socket->fd, - channel, ksess); + ret = kernel_consumer_send_channel_stream(socket, + channel, ksess, + session->output_traces ? 1 : 0); pthread_mutex_unlock(socket->lock); if (ret < 0) { + rcu_read_unlock(); goto error; } } + rcu_read_unlock(); } goto error; } @@ -756,13 +720,18 @@ error: } /* - * For each tracing session, update newly registered apps. + * For each tracing session, update newly registered apps. The session list + * lock MUST be acquired before calling this. */ static void update_ust_app(int app_sock) { struct ltt_session *sess, *stmp; - session_lock_list(); + /* Consumer is in an ERROR state. Stop any application update. */ + if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) { + /* Stop the update process since the consumer is dead. */ + return; + } /* For all tracing session(s) */ cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) { @@ -772,8 +741,6 @@ static void update_ust_app(int app_sock) } session_unlock(sess); } - - session_unlock_list(); } /* @@ -789,30 +756,44 @@ static void *thread_manage_kernel(void *data) char tmp; struct lttng_poll_event events; - DBG("Thread manage kernel started"); + DBG("[thread] Thread manage kernel started"); - health_code_update(&health_thread_kernel); + health_register(health_sessiond, HEALTH_TYPE_KERNEL); - ret = create_thread_poll_set(&events, 2); - if (ret < 0) { - goto error_poll_create; + /* + * This first step of the while is to clean this structure which could free + * non NULL pointers so initialize it before the loop. + */ + lttng_poll_init(&events); + + if (testpoint(thread_manage_kernel)) { + goto error_testpoint; } - ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN); - if (ret < 0) { - goto error; + health_code_update(); + + if (testpoint(thread_manage_kernel_before_loop)) { + goto error_testpoint; } while (1) { - health_code_update(&health_thread_kernel); + health_code_update(); if (update_poll_flag == 1) { - /* - * Reset number of fd in the poll set. Always 2 since there is the thread - * quit pipe and the kernel pipe. - */ - events.nb_fd = 2; + /* Clean events object. We are about to populate it again. */ + lttng_poll_clean(&events); + + ret = sessiond_set_thread_pollset(&events, 2); + if (ret < 0) { + goto error_poll_create; + } + + ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN); + if (ret < 0) { + goto error; + } + /* This will add the available kernel channel if any. */ ret = update_kernel_poll(&events); if (ret < 0) { goto error; @@ -820,18 +801,13 @@ static void *thread_manage_kernel(void *data) update_poll_flag = 0; } - nb_fd = LTTNG_POLL_GETNB(&events); - - DBG("Thread kernel polling on %d fds", nb_fd); - - /* Zeroed the poll events */ - lttng_poll_reset(&events); + DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events)); /* Poll infinite value of time */ restart: - health_poll_update(&health_thread_kernel); + health_poll_entry(); ret = lttng_poll_wait(&events, -1); - health_poll_update(&health_thread_kernel); + health_poll_exit(); if (ret < 0) { /* * Restart interrupted system call. @@ -847,15 +823,17 @@ static void *thread_manage_kernel(void *data) continue; } + nb_fd = ret; + for (i = 0; i < nb_fd; i++) { /* Fetch once the poll data */ revents = LTTNG_POLL_GETEV(&events, i); pollfd = LTTNG_POLL_GETFD(&events, i); - health_code_update(&health_thread_kernel); + health_code_update(); /* Thread quit pipe has been closed. Killing thread. */ - ret = check_thread_quit_pipe(pollfd, revents); + ret = sessiond_check_thread_quit_pipe(pollfd, revents); if (ret) { err = 0; goto exit; @@ -863,7 +841,13 @@ static void *thread_manage_kernel(void *data) /* Check for data on kernel pipe */ if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) { - ret = read(kernel_poll_pipe[0], &tmp, 1); + do { + ret = read(kernel_poll_pipe[0], &tmp, 1); + } while (ret < 0 && errno == EINTR); + /* + * Ret value is useless here, if this pipe gets any actions an + * update is required anyway. + */ update_poll_flag = 1; continue; } else { @@ -890,15 +874,43 @@ exit: error: lttng_poll_clean(&events); error_poll_create: +error_testpoint: + utils_close_pipe(kernel_poll_pipe); + kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1; if (err) { - health_error(&health_thread_kernel); + health_error(); ERR("Health error occurred in %s", __func__); + WARN("Kernel thread died unexpectedly. " + "Kernel tracing can continue but CPU hotplug is disabled."); } - health_exit(&health_thread_kernel); + health_unregister(health_sessiond); DBG("Kernel thread dying"); return NULL; } +/* + * Signal pthread condition of the consumer data that the thread. + */ +static void signal_consumer_condition(struct consumer_data *data, int state) +{ + pthread_mutex_lock(&data->cond_mutex); + + /* + * The state is set before signaling. It can be any value, it's the waiter + * job to correctly interpret this condition variable associated to the + * consumer pthread_cond. + * + * A value of 0 means that the corresponding thread of the consumer data + * was not started. 1 indicates that the thread has started and is ready + * for action. A negative value means that there was an error during the + * thread bootstrap. + */ + data->consumer_thread_is_ready = state; + (void) pthread_cond_signal(&data->cond); + + pthread_mutex_unlock(&data->cond_mutex); +} + /* * This thread manage the consumer error sent back to the session daemon. */ @@ -912,36 +924,41 @@ static void *thread_manage_consumer(void *data) DBG("[thread] Manage consumer started"); - health_code_update(&consumer_data->health); + health_register(health_sessiond, HEALTH_TYPE_CONSUMER); - ret = lttcomm_listen_unix_sock(consumer_data->err_sock); - if (ret < 0) { - goto error_listen; - } + health_code_update(); /* - * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock. - * Nothing more will be added to this poll set. + * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the + * metadata_sock. Nothing more will be added to this poll set. */ - ret = create_thread_poll_set(&events, 2); + ret = sessiond_set_thread_pollset(&events, 3); if (ret < 0) { goto error_poll; } + /* + * The error socket here is already in a listening state which was done + * just before spawning this thread to avoid a race between the consumer + * daemon exec trying to connect and the listen() call. + */ ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP); if (ret < 0) { goto error; } - nb_fd = LTTNG_POLL_GETNB(&events); - - health_code_update(&consumer_data->health); + health_code_update(); - /* Inifinite blocking call, waiting for transmission */ + /* Infinite blocking call, waiting for transmission */ restart: - health_poll_update(&consumer_data->health); + health_poll_entry(); + + if (testpoint(thread_manage_consumer)) { + goto error; + } + ret = lttng_poll_wait(&events, -1); - health_poll_update(&consumer_data->health); + health_poll_exit(); if (ret < 0) { /* * Restart interrupted system call. @@ -952,15 +969,17 @@ restart: goto error; } + nb_fd = ret; + for (i = 0; i < nb_fd; i++) { /* Fetch once the poll data */ revents = LTTNG_POLL_GETEV(&events, i); pollfd = LTTNG_POLL_GETFD(&events, i); - health_code_update(&consumer_data->health); + health_code_update(); /* Thread quit pipe has been closed. Killing thread. */ - ret = check_thread_quit_pipe(pollfd, revents); + ret = sessiond_check_thread_quit_pipe(pollfd, revents); if (ret) { err = 0; goto exit; @@ -980,7 +999,13 @@ restart: goto error; } - health_code_update(&consumer_data->health); + /* + * Set the CLOEXEC flag. Return code is useless because either way, the + * show must go on. + */ + (void) utils_set_fd_cloexec(sock); + + health_code_update(); DBG2("Receiving code from consumer err_sock"); @@ -991,93 +1016,139 @@ restart: goto error; } - health_code_update(&consumer_data->health); + health_code_update(); - if (code == CONSUMERD_COMMAND_SOCK_READY) { + if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) { + /* Connect both socket, command and metadata. */ consumer_data->cmd_sock = lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path); - if (consumer_data->cmd_sock < 0) { - sem_post(&consumer_data->sem); - PERROR("consumer connect"); + consumer_data->metadata_fd = + lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path); + if (consumer_data->cmd_sock < 0 + || consumer_data->metadata_fd < 0) { + PERROR("consumer connect cmd socket"); + /* On error, signal condition and quit. */ + signal_consumer_condition(consumer_data, -1); + goto error; + } + consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd; + /* Create metadata socket lock. */ + consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t)); + if (consumer_data->metadata_sock.lock == NULL) { + PERROR("zmalloc pthread mutex"); + ret = -1; goto error; } - /* Signal condition to tell that the kconsumerd is ready */ - sem_post(&consumer_data->sem); - DBG("consumer command socket ready"); + pthread_mutex_init(consumer_data->metadata_sock.lock, NULL); + + signal_consumer_condition(consumer_data, 1); + DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock); + DBG("Consumer metadata socket ready (fd: %d)", + consumer_data->metadata_fd); } else { ERR("consumer error when waiting for SOCK_READY : %s", lttcomm_get_readable_code(-code)); goto error; } - /* Remove the kconsumerd error sock since we've established a connexion */ + /* Remove the consumerd error sock since we've established a connexion */ ret = lttng_poll_del(&events, consumer_data->err_sock); if (ret < 0) { goto error; } + /* Add new accepted error socket. */ ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP); if (ret < 0) { goto error; } - health_code_update(&consumer_data->health); + /* Add metadata socket that is successfully connected. */ + ret = lttng_poll_add(&events, consumer_data->metadata_fd, + LPOLLIN | LPOLLRDHUP); + if (ret < 0) { + goto error; + } - /* Update number of fd */ - nb_fd = LTTNG_POLL_GETNB(&events); + health_code_update(); - /* Inifinite blocking call, waiting for transmission */ + /* Infinite blocking call, waiting for transmission */ restart_poll: - health_poll_update(&consumer_data->health); - ret = lttng_poll_wait(&events, -1); - health_poll_update(&consumer_data->health); - if (ret < 0) { - /* - * Restart interrupted system call. - */ - if (errno == EINTR) { - goto restart_poll; + while (1) { + health_poll_entry(); + ret = lttng_poll_wait(&events, -1); + health_poll_exit(); + if (ret < 0) { + /* + * Restart interrupted system call. + */ + if (errno == EINTR) { + goto restart_poll; + } + goto error; } - goto error; - } - for (i = 0; i < nb_fd; i++) { - /* Fetch once the poll data */ - revents = LTTNG_POLL_GETEV(&events, i); - pollfd = LTTNG_POLL_GETFD(&events, i); + nb_fd = ret; - health_code_update(&consumer_data->health); + for (i = 0; i < nb_fd; i++) { + /* Fetch once the poll data */ + revents = LTTNG_POLL_GETEV(&events, i); + pollfd = LTTNG_POLL_GETFD(&events, i); - /* Thread quit pipe has been closed. Killing thread. */ - ret = check_thread_quit_pipe(pollfd, revents); - if (ret) { - err = 0; - goto exit; - } + health_code_update(); - /* Event on the kconsumerd socket */ - if (pollfd == sock) { - if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { - ERR("consumer err socket second poll error"); - goto error; + /* Thread quit pipe has been closed. Killing thread. */ + ret = sessiond_check_thread_quit_pipe(pollfd, revents); + if (ret) { + err = 0; + goto exit; } - } - } - health_code_update(&consumer_data->health); + if (pollfd == sock) { + /* Event on the consumerd socket */ + if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { + ERR("consumer err socket second poll error"); + goto error; + } + health_code_update(); + /* Wait for any kconsumerd error */ + ret = lttcomm_recv_unix_sock(sock, &code, + sizeof(enum lttcomm_return_code)); + if (ret <= 0) { + ERR("consumer closed the command socket"); + goto error; + } - /* Wait for any kconsumerd error */ - ret = lttcomm_recv_unix_sock(sock, &code, - sizeof(enum lttcomm_return_code)); - if (ret <= 0) { - ERR("consumer closed the command socket"); - goto error; - } + ERR("consumer return code : %s", + lttcomm_get_readable_code(-code)); - ERR("consumer return code : %s", lttcomm_get_readable_code(-code)); + goto exit; + } else if (pollfd == consumer_data->metadata_fd) { + /* UST metadata requests */ + ret = ust_consumer_metadata_request( + &consumer_data->metadata_sock); + if (ret < 0) { + ERR("Handling metadata request"); + goto error; + } + break; + } else { + ERR("Unknown pollfd"); + goto error; + } + } + health_code_update(); + } exit: error: + /* + * We lock here because we are about to close the sockets and some other + * thread might be using them so get exclusive access which will abort all + * other consumer command by other threads. + */ + pthread_mutex_lock(&consumer_data->lock); + /* Immediately set the consumerd state to stopped */ if (consumer_data->type == LTTNG_CONSUMER_KERNEL) { uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR); @@ -1094,13 +1165,22 @@ error: if (ret) { PERROR("close"); } + consumer_data->err_sock = -1; } if (consumer_data->cmd_sock >= 0) { ret = close(consumer_data->cmd_sock); if (ret) { PERROR("close"); } + consumer_data->cmd_sock = -1; + } + if (*consumer_data->metadata_sock.fd_ptr >= 0) { + ret = close(*consumer_data->metadata_sock.fd_ptr); + if (ret) { + PERROR("close"); + } } + if (sock >= 0) { ret = close(sock); if (ret) { @@ -1111,15 +1191,19 @@ error: unlink(consumer_data->err_unix_sock_path); unlink(consumer_data->cmd_unix_sock_path); consumer_data->pid = 0; + pthread_mutex_unlock(&consumer_data->lock); + + /* Cleanup metadata socket mutex. */ + pthread_mutex_destroy(consumer_data->metadata_sock.lock); + free(consumer_data->metadata_sock.lock); lttng_poll_clean(&events); error_poll: -error_listen: if (err) { - health_error(&consumer_data->health); + health_error(); ERR("Health error occurred in %s", __func__); } - health_exit(&consumer_data->health); + health_unregister(health_sessiond); DBG("consumer thread cleanup completed"); return NULL; @@ -1132,7 +1216,6 @@ static void *thread_manage_apps(void *data) { int i, ret, pollfd, err = -1; uint32_t revents, nb_fd; - struct ust_command ust_cmd; struct lttng_poll_event events; DBG("[thread] Manage application started"); @@ -1140,9 +1223,15 @@ static void *thread_manage_apps(void *data) rcu_register_thread(); rcu_thread_online(); - health_code_update(&health_thread_app_manage); + health_register(health_sessiond, HEALTH_TYPE_APP_MANAGE); + + if (testpoint(thread_manage_apps)) { + goto error_testpoint; + } + + health_code_update(); - ret = create_thread_poll_set(&events, 2); + ret = sessiond_set_thread_pollset(&events, 2); if (ret < 0) { goto error_poll_create; } @@ -1152,21 +1241,20 @@ static void *thread_manage_apps(void *data) goto error; } - health_code_update(&health_thread_app_manage); - - while (1) { - /* Zeroed the events structure */ - lttng_poll_reset(&events); + if (testpoint(thread_manage_apps_before_loop)) { + goto error; + } - nb_fd = LTTNG_POLL_GETNB(&events); + health_code_update(); - DBG("Apps thread polling on %d fds", nb_fd); + while (1) { + DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events)); /* Inifinite blocking call, waiting for transmission */ restart: - health_poll_update(&health_thread_app_manage); + health_poll_entry(); ret = lttng_poll_wait(&events, -1); - health_poll_update(&health_thread_app_manage); + health_poll_exit(); if (ret < 0) { /* * Restart interrupted system call. @@ -1177,15 +1265,17 @@ static void *thread_manage_apps(void *data) goto error; } + nb_fd = ret; + for (i = 0; i < nb_fd; i++) { /* Fetch once the poll data */ revents = LTTNG_POLL_GETEV(&events, i); pollfd = LTTNG_POLL_GETFD(&events, i); - health_code_update(&health_thread_app_manage); + health_code_update(); /* Thread quit pipe has been closed. Killing thread. */ - ret = check_thread_quit_pipe(pollfd, revents); + ret = sessiond_check_thread_quit_pipe(pollfd, revents); if (ret) { err = 0; goto exit; @@ -1197,64 +1287,45 @@ static void *thread_manage_apps(void *data) ERR("Apps command pipe error"); goto error; } else if (revents & LPOLLIN) { + int sock; + /* Empty pipe */ - ret = read(apps_cmd_pipe[0], &ust_cmd, sizeof(ust_cmd)); - if (ret < 0 || ret < sizeof(ust_cmd)) { + do { + ret = read(apps_cmd_pipe[0], &sock, sizeof(sock)); + } while (ret < 0 && errno == EINTR); + if (ret < 0 || ret < sizeof(sock)) { PERROR("read apps cmd pipe"); goto error; } - health_code_update(&health_thread_app_manage); + health_code_update(); - /* Register applicaton to the session daemon */ - ret = ust_app_register(&ust_cmd.reg_msg, - ust_cmd.sock); - if (ret == -ENOMEM) { + /* + * We only monitor the error events of the socket. This + * thread does not handle any incoming data from UST + * (POLLIN). + */ + ret = lttng_poll_add(&events, sock, + LPOLLERR | LPOLLHUP | LPOLLRDHUP); + if (ret < 0) { goto error; - } else if (ret < 0) { - break; } - health_code_update(&health_thread_app_manage); - /* - * Validate UST version compatibility. + * Set socket timeout for both receiving and ending. + * app_socket_timeout is in seconds, whereas + * lttcomm_setsockopt_rcv_timeout and + * lttcomm_setsockopt_snd_timeout expect msec as + * parameter. */ - ret = ust_app_validate_version(ust_cmd.sock); - if (ret >= 0) { - /* - * Add channel(s) and event(s) to newly registered apps - * from lttng global UST domain. - */ - update_ust_app(ust_cmd.sock); - } - - health_code_update(&health_thread_app_manage); - - ret = ust_app_register_done(ust_cmd.sock); - if (ret < 0) { - /* - * If the registration is not possible, we simply - * unregister the apps and continue - */ - ust_app_unregister(ust_cmd.sock); - } else { - /* - * We just need here to monitor the close of the UST - * socket and poll set monitor those by default. - * Listen on POLLIN (even if we never expect any - * data) to ensure that hangup wakes us. - */ - ret = lttng_poll_add(&events, ust_cmd.sock, LPOLLIN); - if (ret < 0) { - goto error; - } + (void) lttcomm_setsockopt_rcv_timeout(sock, + app_socket_timeout * 1000); + (void) lttcomm_setsockopt_snd_timeout(sock, + app_socket_timeout * 1000); - DBG("Apps with sock %d added to poll set", - ust_cmd.sock); - } + DBG("Apps with sock %d added to poll set", sock); - health_code_update(&health_thread_app_manage); + health_code_update(); break; } @@ -1276,7 +1347,7 @@ static void *thread_manage_apps(void *data) } } - health_code_update(&health_thread_app_manage); + health_code_update(); } } @@ -1284,11 +1355,21 @@ exit: error: lttng_poll_clean(&events); error_poll_create: +error_testpoint: + utils_close_pipe(apps_cmd_pipe); + apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1; + + /* + * We don't clean the UST app hash table here since already registered + * applications can still be controlled so let them be until the session + * daemon dies or the applications stop. + */ + if (err) { - health_error(&health_thread_app_manage); + health_error(); ERR("Health error occurred in %s", __func__); } - health_exit(&health_thread_app_manage); + health_unregister(health_sessiond); DBG("Application communication apps thread cleanup complete"); rcu_thread_offline(); rcu_unregister_thread(); @@ -1296,22 +1377,170 @@ error_poll_create: } /* - * Dispatch request from the registration threads to the application - * communication thread. - */ -static void *thread_dispatch_ust_registration(void *data) + * Send a socket to a thread This is called from the dispatch UST registration + * thread once all sockets are set for the application. + * + * The sock value can be invalid, we don't really care, the thread will handle + * it and make the necessary cleanup if so. + * + * On success, return 0 else a negative value being the errno message of the + * write(). + */ +static int send_socket_to_thread(int fd, int sock) { int ret; + + /* + * It's possible that the FD is set as invalid with -1 concurrently just + * before calling this function being a shutdown state of the thread. + */ + if (fd < 0) { + ret = -EBADF; + goto error; + } + + do { + ret = write(fd, &sock, sizeof(sock)); + } while (ret < 0 && errno == EINTR); + if (ret < 0 || ret != sizeof(sock)) { + PERROR("write apps pipe %d", fd); + if (ret < 0) { + ret = -errno; + } + goto error; + } + + /* All good. Don't send back the write positive ret value. */ + ret = 0; +error: + return ret; +} + +/* + * Sanitize the wait queue of the dispatch registration thread meaning removing + * invalid nodes from it. This is to avoid memory leaks for the case the UST + * notify socket is never received. + */ +static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue) +{ + int ret, nb_fd = 0, i; + unsigned int fd_added = 0; + struct lttng_poll_event events; + struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node; + + assert(wait_queue); + + lttng_poll_init(&events); + + /* Just skip everything for an empty queue. */ + if (!wait_queue->count) { + goto end; + } + + ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC); + if (ret < 0) { + goto error_create; + } + + cds_list_for_each_entry_safe(wait_node, tmp_wait_node, + &wait_queue->head, head) { + assert(wait_node->app); + ret = lttng_poll_add(&events, wait_node->app->sock, + LPOLLHUP | LPOLLERR); + if (ret < 0) { + goto error; + } + + fd_added = 1; + } + + if (!fd_added) { + goto end; + } + + /* + * Poll but don't block so we can quickly identify the faulty events and + * clean them afterwards from the wait queue. + */ + ret = lttng_poll_wait(&events, 0); + if (ret < 0) { + goto error; + } + nb_fd = ret; + + for (i = 0; i < nb_fd; i++) { + /* Get faulty FD. */ + uint32_t revents = LTTNG_POLL_GETEV(&events, i); + int pollfd = LTTNG_POLL_GETFD(&events, i); + + cds_list_for_each_entry_safe(wait_node, tmp_wait_node, + &wait_queue->head, head) { + if (pollfd == wait_node->app->sock && + (revents & (LPOLLHUP | LPOLLERR))) { + cds_list_del(&wait_node->head); + wait_queue->count--; + ust_app_destroy(wait_node->app); + free(wait_node); + break; + } + } + } + + if (nb_fd > 0) { + DBG("Wait queue sanitized, %d node were cleaned up", nb_fd); + } + +end: + lttng_poll_clean(&events); + return; + +error: + lttng_poll_clean(&events); +error_create: + ERR("Unable to sanitize wait queue"); + return; +} + +/* + * Dispatch request from the registration threads to the application + * communication thread. + */ +static void *thread_dispatch_ust_registration(void *data) +{ + int ret, err = -1; struct cds_wfq_node *node; struct ust_command *ust_cmd = NULL; + struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node; + struct ust_reg_wait_queue wait_queue = { + .count = 0, + }; + + health_register(health_sessiond, HEALTH_TYPE_APP_REG_DISPATCH); + + health_code_update(); + + CDS_INIT_LIST_HEAD(&wait_queue.head); DBG("[thread] Dispatch UST command started"); while (!CMM_LOAD_SHARED(dispatch_thread_exit)) { + health_code_update(); + /* Atomically prepare the queue futex */ futex_nto1_prepare(&ust_cmd_queue.futex); do { + struct ust_app *app = NULL; + ust_cmd = NULL; + + /* + * Make sure we don't have node(s) that have hung up before receiving + * the notify socket. This is to clean the list in order to avoid + * memory leaks from notify socket that are never seen. + */ + sanitize_wait_queue(&wait_queue); + + health_code_update(); /* Dequeue command for registration */ node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue); if (node == NULL) { @@ -1328,34 +1557,176 @@ static void *thread_dispatch_ust_registration(void *data) ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid, ust_cmd->sock, ust_cmd->reg_msg.name, ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor); - /* - * Inform apps thread of the new application registration. This - * call is blocking so we can be assured that the data will be read - * at some point in time or wait to the end of the world :) - */ - ret = write(apps_cmd_pipe[1], ust_cmd, - sizeof(struct ust_command)); - if (ret < 0) { - PERROR("write apps cmd pipe"); - if (errno == EBADF) { + + if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) { + wait_node = zmalloc(sizeof(*wait_node)); + if (!wait_node) { + PERROR("zmalloc wait_node dispatch"); + ret = close(ust_cmd->sock); + if (ret < 0) { + PERROR("close ust sock dispatch %d", ust_cmd->sock); + } + lttng_fd_put(1, LTTNG_FD_APPS); + free(ust_cmd); + goto error; + } + CDS_INIT_LIST_HEAD(&wait_node->head); + + /* Create application object if socket is CMD. */ + wait_node->app = ust_app_create(&ust_cmd->reg_msg, + ust_cmd->sock); + if (!wait_node->app) { + ret = close(ust_cmd->sock); + if (ret < 0) { + PERROR("close ust sock dispatch %d", ust_cmd->sock); + } + lttng_fd_put(1, LTTNG_FD_APPS); + free(wait_node); + free(ust_cmd); + continue; + } + /* + * Add application to the wait queue so we can set the notify + * socket before putting this object in the global ht. + */ + cds_list_add(&wait_node->head, &wait_queue.head); + wait_queue.count++; + + free(ust_cmd); + /* + * We have to continue here since we don't have the notify + * socket and the application MUST be added to the hash table + * only at that moment. + */ + continue; + } else { + /* + * Look for the application in the local wait queue and set the + * notify socket if found. + */ + cds_list_for_each_entry_safe(wait_node, tmp_wait_node, + &wait_queue.head, head) { + health_code_update(); + if (wait_node->app->pid == ust_cmd->reg_msg.pid) { + wait_node->app->notify_sock = ust_cmd->sock; + cds_list_del(&wait_node->head); + wait_queue.count--; + app = wait_node->app; + free(wait_node); + DBG3("UST app notify socket %d is set", ust_cmd->sock); + break; + } + } + + /* + * With no application at this stage the received socket is + * basically useless so close it before we free the cmd data + * structure for good. + */ + if (!app) { + ret = close(ust_cmd->sock); + if (ret < 0) { + PERROR("close ust sock dispatch %d", ust_cmd->sock); + } + lttng_fd_put(1, LTTNG_FD_APPS); + } + free(ust_cmd); + } + + if (app) { + /* + * @session_lock_list + * + * Lock the global session list so from the register up to the + * registration done message, no thread can see the application + * and change its state. + */ + session_lock_list(); + rcu_read_lock(); + + /* + * Add application to the global hash table. This needs to be + * done before the update to the UST registry can locate the + * application. + */ + ust_app_add(app); + + /* Set app version. This call will print an error if needed. */ + (void) ust_app_version(app); + + /* Send notify socket through the notify pipe. */ + ret = send_socket_to_thread(apps_cmd_notify_pipe[1], + app->notify_sock); + if (ret < 0) { + rcu_read_unlock(); + session_unlock_list(); + /* + * No notify thread, stop the UST tracing. However, this is + * not an internal error of the this thread thus setting + * the health error code to a normal exit. + */ + err = 0; + goto error; + } + + /* + * Update newly registered application with the tracing + * registry info already enabled information. + */ + update_ust_app(app->sock); + + /* + * Don't care about return value. Let the manage apps threads + * handle app unregistration upon socket close. + */ + (void) ust_app_register_done(app->sock); + + /* + * Even if the application socket has been closed, send the app + * to the thread and unregistration will take place at that + * place. + */ + ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock); + if (ret < 0) { + rcu_read_unlock(); + session_unlock_list(); /* - * We can't inform the application thread to process - * registration. We will exit or else application - * registration will not occur and tracing will never - * start. + * No apps. thread, stop the UST tracing. However, this is + * not an internal error of the this thread thus setting + * the health error code to a normal exit. */ + err = 0; goto error; } + + rcu_read_unlock(); + session_unlock_list(); } - free(ust_cmd); } while (node != NULL); + health_poll_entry(); /* Futex wait on queue. Blocking call on futex() */ futex_nto1_wait(&ust_cmd_queue.futex); + health_poll_exit(); } + /* Normal exit, no error */ + err = 0; error: + /* Clean up wait queue. */ + cds_list_for_each_entry_safe(wait_node, tmp_wait_node, + &wait_queue.head, head) { + cds_list_del(&wait_node->head); + wait_queue.count--; + free(wait_node); + } + DBG("Dispatch thread dying"); + if (err) { + health_error(); + ERR("Health error occurred in %s", __func__); + } + health_unregister(health_sessiond); return NULL; } @@ -1375,6 +1746,12 @@ static void *thread_registration_apps(void *data) DBG("[thread] Manage application registration started"); + health_register(health_sessiond, HEALTH_TYPE_APP_REG); + + if (testpoint(thread_registration_apps)) { + goto error_testpoint; + } + ret = lttcomm_listen_unix_sock(apps_sock); if (ret < 0) { goto error_listen; @@ -1384,7 +1761,7 @@ static void *thread_registration_apps(void *data) * Pass 2 as size here for the thread quit pipe and apps socket. Nothing * more will be added to this poll set. */ - ret = create_thread_poll_set(&events, 2); + ret = sessiond_set_thread_pollset(&events, 2); if (ret < 0) { goto error_create_poll; } @@ -1406,13 +1783,11 @@ static void *thread_registration_apps(void *data) while (1) { DBG("Accepting application registration"); - nb_fd = LTTNG_POLL_GETNB(&events); - /* Inifinite blocking call, waiting for transmission */ restart: - health_poll_update(&health_thread_app_reg); + health_poll_entry(); ret = lttng_poll_wait(&events, -1); - health_poll_update(&health_thread_app_reg); + health_poll_exit(); if (ret < 0) { /* * Restart interrupted system call. @@ -1423,15 +1798,17 @@ static void *thread_registration_apps(void *data) goto error; } + nb_fd = ret; + for (i = 0; i < nb_fd; i++) { - health_code_update(&health_thread_app_reg); + health_code_update(); /* Fetch once the poll data */ revents = LTTNG_POLL_GETEV(&events, i); pollfd = LTTNG_POLL_GETFD(&events, i); /* Thread quit pipe has been closed. Killing thread. */ - ret = check_thread_quit_pipe(pollfd, revents); + ret = sessiond_check_thread_quit_pipe(pollfd, revents); if (ret) { err = 0; goto exit; @@ -1448,6 +1825,12 @@ static void *thread_registration_apps(void *data) goto error; } + /* + * Set the CLOEXEC flag. Return code is useless because + * either way, the show must go on. + */ + (void) utils_set_fd_cloexec(sock); + /* Create UST registration command for enqueuing */ ust_cmd = zmalloc(sizeof(struct ust_command)); if (ust_cmd == NULL) { @@ -1470,16 +1853,12 @@ static void *thread_registration_apps(void *data) sock = -1; continue; } - health_code_update(&health_thread_app_reg); - ret = lttcomm_recv_unix_sock(sock, &ust_cmd->reg_msg, - sizeof(struct ust_register_msg)); - if (ret < 0 || ret < sizeof(struct ust_register_msg)) { - if (ret < 0) { - PERROR("lttcomm_recv_unix_sock register apps"); - } else { - ERR("Wrong size received on apps register"); - } + + health_code_update(); + ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg); + if (ret < 0) { free(ust_cmd); + /* Close socket of the application. */ ret = close(sock); if (ret) { PERROR("close"); @@ -1488,7 +1867,7 @@ static void *thread_registration_apps(void *data) sock = -1; continue; } - health_code_update(&health_thread_app_reg); + health_code_update(); ust_cmd->sock = sock; sock = -1; @@ -1519,10 +1898,9 @@ static void *thread_registration_apps(void *data) exit: error: if (err) { - health_error(&health_thread_app_reg); + health_error(); ERR("Health error occurred in %s", __func__); } - health_exit(&health_thread_app_reg); /* Notify that the registration thread is gone */ notify_ust_apps(0); @@ -1546,7 +1924,9 @@ error_poll_add: lttng_poll_clean(&events); error_listen: error_create_poll: +error_testpoint: DBG("UST Registration thread cleanup complete"); + health_unregister(health_sessiond); return NULL; } @@ -1557,59 +1937,110 @@ error_create_poll: */ static int spawn_consumer_thread(struct consumer_data *consumer_data) { - int ret; + int ret, clock_ret; struct timespec timeout; - timeout.tv_sec = DEFAULT_SEM_WAIT_TIMEOUT; - timeout.tv_nsec = 0; + /* Make sure we set the readiness flag to 0 because we are NOT ready */ + consumer_data->consumer_thread_is_ready = 0; - /* Setup semaphore */ - ret = sem_init(&consumer_data->sem, 0, 0); - if (ret < 0) { - PERROR("sem_init consumer semaphore"); + /* Setup pthread condition */ + ret = pthread_condattr_init(&consumer_data->condattr); + if (ret != 0) { + errno = ret; + PERROR("pthread_condattr_init consumer data"); + goto error; + } + + /* + * Set the monotonic clock in order to make sure we DO NOT jump in time + * between the clock_gettime() call and the timedwait call. See bug #324 + * for a more details and how we noticed it. + */ + ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC); + if (ret != 0) { + errno = ret; + PERROR("pthread_condattr_setclock consumer data"); + goto error; + } + + ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr); + if (ret != 0) { + errno = ret; + PERROR("pthread_cond_init consumer data"); goto error; } - ret = pthread_create(&consumer_data->thread, NULL, - thread_manage_consumer, consumer_data); + ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer, + consumer_data); if (ret != 0) { PERROR("pthread_create consumer"); ret = -1; goto error; } + /* We are about to wait on a pthread condition */ + pthread_mutex_lock(&consumer_data->cond_mutex); + /* Get time for sem_timedwait absolute timeout */ - ret = clock_gettime(CLOCK_REALTIME, &timeout); - if (ret < 0) { - PERROR("clock_gettime spawn consumer"); - /* Infinite wait for the kconsumerd thread to be ready */ - ret = sem_wait(&consumer_data->sem); - } else { - /* Normal timeout if the gettime was successful */ - timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT; - ret = sem_timedwait(&consumer_data->sem, &timeout); + clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout); + /* + * Set the timeout for the condition timed wait even if the clock gettime + * call fails since we might loop on that call and we want to avoid to + * increment the timeout too many times. + */ + timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT; + + /* + * The following loop COULD be skipped in some conditions so this is why we + * set ret to 0 in order to make sure at least one round of the loop is + * done. + */ + ret = 0; + + /* + * Loop until the condition is reached or when a timeout is reached. Note + * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT + * be returned but the pthread_cond(3), from the glibc-doc, says that it is + * possible. This loop does not take any chances and works with both of + * them. + */ + while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) { + if (clock_ret < 0) { + PERROR("clock_gettime spawn consumer"); + /* Infinite wait for the consumerd thread to be ready */ + ret = pthread_cond_wait(&consumer_data->cond, + &consumer_data->cond_mutex); + } else { + ret = pthread_cond_timedwait(&consumer_data->cond, + &consumer_data->cond_mutex, &timeout); + } } - if (ret < 0) { - if (errno == ETIMEDOUT) { + /* Release the pthread condition */ + pthread_mutex_unlock(&consumer_data->cond_mutex); + + if (ret != 0) { + errno = ret; + if (ret == ETIMEDOUT) { /* * Call has timed out so we kill the kconsumerd_thread and return * an error. */ - ERR("The consumer thread was never ready. Killing it"); + ERR("Condition timed out. The consumer thread was never ready." + " Killing it"); ret = pthread_cancel(consumer_data->thread); if (ret < 0) { PERROR("pthread_cancel consumer thread"); } } else { - PERROR("semaphore wait failed consumer thread"); + PERROR("pthread_cond_wait failed consumer thread"); } goto error; } pthread_mutex_lock(&consumer_data->pid_mutex); if (consumer_data->pid == 0) { - ERR("Kconsumerd did not start"); + ERR("Consumerd did not start"); pthread_mutex_unlock(&consumer_data->pid_mutex); goto error; } @@ -1627,10 +2058,10 @@ error: static int join_consumer_thread(struct consumer_data *consumer_data) { void *status; - int ret; /* Consumer pid must be a real one. */ if (consumer_data->pid > 0) { + int ret; ret = kill(consumer_data->pid, SIGTERM); if (ret) { ERR("Error killing consumer daemon"); @@ -1726,6 +2157,7 @@ static pid_t spawn_consumerd(struct consumer_data *consumer_data) ret = putenv(tmpnew); if (ret) { ret = -errno; + free(tmpnew); goto error; } } @@ -1770,6 +2202,7 @@ static pid_t spawn_consumerd(struct consumer_data *consumer_data) ret = putenv(tmpnew); if (ret) { ret = -errno; + free(tmpnew); goto error; } } @@ -1811,6 +2244,16 @@ static int start_consumerd(struct consumer_data *consumer_data) { int ret; + /* + * Set the listen() state on the socket since there is a possible race + * between the exec() of the consumer daemon and this call if place in the + * consumer thread. See bug #366 for more details. + */ + ret = lttcomm_listen_unix_sock(consumer_data->err_sock); + if (ret < 0) { + goto error; + } + pthread_mutex_lock(&consumer_data->pid_mutex); if (consumer_data->pid != 0) { pthread_mutex_unlock(&consumer_data->pid_mutex); @@ -1840,6 +2283,15 @@ end: return 0; error: + /* Cleanup already created sockets on error. */ + if (consumer_data->err_sock >= 0) { + int err; + + err = close(consumer_data->err_sock); + if (err < 0) { + PERROR("close consumer data error socket"); + } + } return ret; } @@ -1851,23 +2303,13 @@ static int check_consumer_health(void) { int ret; - ret = health_check_state(&kconsumer_data.health) && - health_check_state(&ustconsumer32_data.health) && - health_check_state(&ustconsumer64_data.health); + ret = health_check_state(health_sessiond, HEALTH_TYPE_CONSUMER); DBG3("Health consumer check %d", ret); return ret; } -/* - * Check version of the lttng-modules. - */ -static int validate_lttng_modules_version(void) -{ - return kernel_validate_version(kernel_tracer_fd); -} - /* * Setup necessary data for kernel tracer action. */ @@ -1890,7 +2332,7 @@ static int init_kernel_tracer(void) } /* Validate kernel version */ - ret = validate_lttng_modules_version(); + ret = kernel_validate_version(kernel_tracer_fd); if (ret < 0) { goto error_version; } @@ -1910,7 +2352,7 @@ error_version: PERROR("close"); } kernel_tracer_fd = -1; - return LTTCOMM_KERN_VERSION; + return LTTNG_ERR_KERN_VERSION; error_modules: ret = close(kernel_tracer_fd); @@ -1925,2364 +2367,198 @@ error: WARN("No kernel tracer available"); kernel_tracer_fd = -1; if (!is_root) { - return LTTCOMM_NEED_ROOT_SESSIOND; + return LTTNG_ERR_NEED_ROOT_SESSIOND; } else { - return LTTCOMM_KERN_NA; + return LTTNG_ERR_KERN_NA; } } + /* - * Init tracing by creating trace directory and sending fds kernel consumer. + * Copy consumer output from the tracing session to the domain session. The + * function also applies the right modification on a per domain basis for the + * trace files destination directory. + * + * Should *NOT* be called with RCU read-side lock held. */ -static int init_kernel_tracing(struct ltt_kernel_session *session) +static int copy_session_consumer(int domain, struct ltt_session *session) { - int ret = 0; - struct lttng_ht_iter iter; - struct consumer_socket *socket; + int ret; + const char *dir_name; + struct consumer_output *consumer; assert(session); + assert(session->consumer); - if (session->consumer_fds_sent == 0 && session->consumer != NULL) { - cds_lfht_for_each_entry(session->consumer->socks->ht, &iter.iter, - socket, node.node) { - /* Code flow error */ - assert(socket->fd >= 0); - - pthread_mutex_lock(socket->lock); - ret = kernel_consumer_send_session(socket->fd, session); - pthread_mutex_unlock(socket->lock); - if (ret < 0) { - ret = LTTCOMM_KERN_CONSUMER_FAIL; - goto error; - } + switch (domain) { + case LTTNG_DOMAIN_KERNEL: + DBG3("Copying tracing session consumer output in kernel session"); + /* + * XXX: We should audit the session creation and what this function + * does "extra" in order to avoid a destroy since this function is used + * in the domain session creation (kernel and ust) only. Same for UST + * domain. + */ + if (session->kernel_session->consumer) { + consumer_destroy_output(session->kernel_session->consumer); + } + session->kernel_session->consumer = + consumer_copy_output(session->consumer); + /* Ease our life a bit for the next part */ + consumer = session->kernel_session->consumer; + dir_name = DEFAULT_KERNEL_TRACE_DIR; + break; + case LTTNG_DOMAIN_UST: + DBG3("Copying tracing session consumer output in UST session"); + if (session->ust_session->consumer) { + consumer_destroy_output(session->ust_session->consumer); } + session->ust_session->consumer = + consumer_copy_output(session->consumer); + /* Ease our life a bit for the next part */ + consumer = session->ust_session->consumer; + dir_name = DEFAULT_UST_TRACE_DIR; + break; + default: + ret = LTTNG_ERR_UNKNOWN_DOMAIN; + goto error; } + /* Append correct directory to subdir */ + strncat(consumer->subdir, dir_name, + sizeof(consumer->subdir) - strlen(consumer->subdir) - 1); + DBG3("Copy session consumer subdir %s", consumer->subdir); + + ret = LTTNG_OK; + error: return ret; } /* - * Create a socket to the relayd using the URI. + * Create an UST session and add it to the session ust list. * - * On success, the relayd_sock pointer is set to the created socket. - * Else, it is untouched and an lttcomm error code is returned. + * Should *NOT* be called with RCU read-side lock held. */ -static int create_connect_relayd(struct consumer_output *output, - const char *session_name, struct lttng_uri *uri, - struct lttcomm_sock **relayd_sock) +static int create_ust_session(struct ltt_session *session, + struct lttng_domain *domain) { int ret; - struct lttcomm_sock *sock; + struct ltt_ust_session *lus = NULL; - /* Create socket object from URI */ - sock = lttcomm_alloc_sock_from_uri(uri); - if (sock == NULL) { - ret = LTTCOMM_FATAL; - goto error; - } + assert(session); + assert(domain); + assert(session->consumer); - ret = lttcomm_create_sock(sock); - if (ret < 0) { - ret = LTTCOMM_FATAL; + switch (domain->type) { + case LTTNG_DOMAIN_UST: + break; + default: + ERR("Unknown UST domain on create session %d", domain->type); + ret = LTTNG_ERR_UNKNOWN_DOMAIN; goto error; } - /* Connect to relayd so we can proceed with a session creation. */ - ret = relayd_connect(sock); - if (ret < 0) { - ERR("Unable to reach lttng-relayd"); - ret = LTTCOMM_RELAYD_SESSION_FAIL; - goto free_sock; + DBG("Creating UST session"); + + lus = trace_ust_create_session(session->id); + if (lus == NULL) { + ret = LTTNG_ERR_UST_SESS_FAIL; + goto error; } - /* Create socket for control stream. */ - if (uri->stype == LTTNG_STREAM_CONTROL) { - DBG3("Creating relayd stream socket from URI"); - - /* Check relayd version */ - ret = relayd_version_check(sock, LTTNG_UST_COMM_MAJOR, 0); - if (ret < 0) { - ret = LTTCOMM_RELAYD_VERSION_FAIL; - goto close_sock; - } - } else if (uri->stype == LTTNG_STREAM_DATA) { - DBG3("Creating relayd data socket from URI"); - } else { - /* Command is not valid */ - ERR("Relayd invalid stream type: %d", uri->stype); - ret = LTTCOMM_INVALID; - goto close_sock; - } - - *relayd_sock = sock; - - return LTTCOMM_OK; - -close_sock: - if (sock) { - (void) relayd_close(sock); - } -free_sock: - if (sock) { - lttcomm_destroy_sock(sock); - } -error: - return ret; -} - -/* - * Connect to the relayd using URI and send the socket to the right consumer. - */ -static int send_socket_relayd_consumer(int domain, struct ltt_session *session, - struct lttng_uri *relayd_uri, struct consumer_output *consumer, - int consumer_fd) -{ - int ret; - struct lttcomm_sock *sock = NULL; - - /* Set the network sequence index if not set. */ - if (consumer->net_seq_index == -1) { - /* - * Increment net_seq_idx because we are about to transfer the - * new relayd socket to the consumer. - */ - uatomic_inc(&relayd_net_seq_idx); - /* Assign unique key so the consumer can match streams */ - consumer->net_seq_index = uatomic_read(&relayd_net_seq_idx); - } - - /* Connect to relayd and make version check if uri is the control. */ - ret = create_connect_relayd(consumer, session->name, relayd_uri, &sock); - if (ret != LTTCOMM_OK) { - goto close_sock; - } - - /* If the control socket is connected, network session is ready */ - if (relayd_uri->stype == LTTNG_STREAM_CONTROL) { - session->net_handle = 1; - } - - /* Send relayd socket to consumer. */ - ret = consumer_send_relayd_socket(consumer_fd, sock, - consumer, relayd_uri->stype); - if (ret < 0) { - ret = LTTCOMM_ENABLE_CONSUMER_FAIL; - goto close_sock; - } - - ret = LTTCOMM_OK; - - /* - * Close socket which was dup on the consumer side. The session daemon does - * NOT keep track of the relayd socket(s) once transfer to the consumer. - */ - -close_sock: - if (sock) { - (void) relayd_close(sock); - lttcomm_destroy_sock(sock); - } - - return ret; -} - -/* - * Send both relayd sockets to a specific consumer and domain. This is a - * helper function to facilitate sending the information to the consumer for a - * session. - */ -static int send_sockets_relayd_consumer(int domain, - struct ltt_session *session, struct consumer_output *consumer, int fd) -{ - int ret; - - assert(session); - assert(consumer); - - /* Don't resend the sockets to the consumer. */ - if (consumer->dst.net.relayd_socks_sent) { - ret = LTTCOMM_OK; - goto error; - } - - /* Sending control relayd socket. */ - ret = send_socket_relayd_consumer(domain, session, - &consumer->dst.net.control, consumer, fd); - if (ret != LTTCOMM_OK) { - goto error; - } - - /* Sending data relayd socket. */ - ret = send_socket_relayd_consumer(domain, session, - &consumer->dst.net.data, consumer, fd); - if (ret != LTTCOMM_OK) { - goto error; - } - - /* Flag that all relayd sockets were sent to the consumer. */ - consumer->dst.net.relayd_socks_sent = 1; - -error: - return ret; -} - -/* - * Setup relayd connections for a tracing session. First creates the socket to - * the relayd and send them to the right domain consumer. Consumer type MUST be - * network. - */ -static int setup_relayd(struct ltt_session *session) -{ - int ret = LTTCOMM_OK; - struct ltt_ust_session *usess; - struct ltt_kernel_session *ksess; - struct consumer_socket *socket; - struct lttng_ht_iter iter; - - assert(session); - - usess = session->ust_session; - ksess = session->kernel_session; - - DBG2("Setting relayd for session %s", session->name); - - if (usess && usess->consumer && usess->consumer->type == CONSUMER_DST_NET - && usess->consumer->enabled) { - /* For each consumer socket, send relayd sockets */ - cds_lfht_for_each_entry(usess->consumer->socks->ht, &iter.iter, - socket, node.node) { - /* Code flow error */ - assert(socket->fd >= 0); - - pthread_mutex_lock(socket->lock); - send_sockets_relayd_consumer(LTTNG_DOMAIN_UST, session, - usess->consumer, socket->fd); - pthread_mutex_unlock(socket->lock); - if (ret != LTTCOMM_OK) { - goto error; - } - } - } - - if (ksess && ksess->consumer && ksess->consumer->type == CONSUMER_DST_NET - && ksess->consumer->enabled) { - cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter, - socket, node.node) { - /* Code flow error */ - assert(socket->fd >= 0); - - pthread_mutex_lock(socket->lock); - send_sockets_relayd_consumer(LTTNG_DOMAIN_KERNEL, session, - ksess->consumer, socket->fd); - pthread_mutex_unlock(socket->lock); - if (ret != LTTCOMM_OK) { - goto error; - } - } - } - -error: - return ret; -} - -/* - * Set consumer subdirectory using the session name and a generated datetime if - * needed. This is appended to the current subdirectory. - */ -static int set_consumer_subdir(struct consumer_output *consumer, - const char *session_name) -{ - int ret = 0; - unsigned int have_default_name = 0; - char datetime[16], tmp_path[PATH_MAX]; - time_t rawtime; - struct tm *timeinfo; - - assert(consumer); - assert(session_name); - - memset(tmp_path, 0, sizeof(tmp_path)); - - /* Flag if we have a default session. */ - if (strncmp(session_name, DEFAULT_SESSION_NAME "-", - strlen(DEFAULT_SESSION_NAME) + 1) == 0) { - have_default_name = 1; - } else { - /* Get date and time for session path */ - time(&rawtime); - timeinfo = localtime(&rawtime); - strftime(datetime, sizeof(datetime), "%Y%m%d-%H%M%S", timeinfo); - } - - if (have_default_name) { - ret = snprintf(tmp_path, sizeof(tmp_path), - "%s/%s", consumer->subdir, session_name); - } else { - ret = snprintf(tmp_path, sizeof(tmp_path), - "%s/%s-%s/", consumer->subdir, session_name, datetime); - } - if (ret < 0) { - PERROR("snprintf session name date"); - goto error; - } - - strncpy(consumer->subdir, tmp_path, sizeof(consumer->subdir)); - DBG2("Consumer subdir set to %s", consumer->subdir); - -error: - return ret; -} - -/* - * Copy consumer output from the tracing session to the domain session. The - * function also applies the right modification on a per domain basis for the - * trace files destination directory. - */ -static int copy_session_consumer(int domain, struct ltt_session *session) -{ - int ret; - const char *dir_name; - struct consumer_output *consumer; - - assert(session); - assert(session->consumer); - - switch (domain) { - case LTTNG_DOMAIN_KERNEL: - DBG3("Copying tracing session consumer output in kernel session"); - session->kernel_session->consumer = - consumer_copy_output(session->consumer); - /* Ease our life a bit for the next part */ - consumer = session->kernel_session->consumer; - dir_name = DEFAULT_KERNEL_TRACE_DIR; - break; - case LTTNG_DOMAIN_UST: - DBG3("Copying tracing session consumer output in UST session"); - session->ust_session->consumer = - consumer_copy_output(session->consumer); - /* Ease our life a bit for the next part */ - consumer = session->ust_session->consumer; - dir_name = DEFAULT_UST_TRACE_DIR; - break; - default: - ret = LTTCOMM_UNKNOWN_DOMAIN; - goto error; - } - - ret = set_consumer_subdir(session->consumer, session->name); - if (ret < 0) { - ret = LTTCOMM_FATAL; - goto error; - } - - /* Append correct directory to subdir */ - strncat(consumer->subdir, dir_name, sizeof(consumer->subdir)); - DBG3("Copy session consumer subdir %s", consumer->subdir); - - ret = LTTCOMM_OK; - -error: - return ret; -} - -/* - * Create an UST session and add it to the session ust list. - */ -static int create_ust_session(struct ltt_session *session, - struct lttng_domain *domain) -{ - int ret; - struct ltt_ust_session *lus = NULL; - - assert(session); - assert(domain); - assert(session->consumer); - - switch (domain->type) { - case LTTNG_DOMAIN_UST: - break; - default: - ERR("Unknown UST domain on create session %d", domain->type); - ret = LTTCOMM_UNKNOWN_DOMAIN; - goto error; - } - - DBG("Creating UST session"); - - lus = trace_ust_create_session(session->path, session->id, domain); - if (lus == NULL) { - ret = LTTCOMM_UST_SESS_FAIL; - goto error; - } - - lus->uid = session->uid; - lus->gid = session->gid; - session->ust_session = lus; - - /* Copy session output to the newly created UST session */ - ret = copy_session_consumer(domain->type, session); - if (ret != LTTCOMM_OK) { - goto error; - } - - return LTTCOMM_OK; - -error: - free(lus); - session->ust_session = NULL; - return ret; -} - -/* - * Create a kernel tracer session then create the default channel. - */ -static int create_kernel_session(struct ltt_session *session) -{ - int ret; - - DBG("Creating kernel session"); - - ret = kernel_create_session(session, kernel_tracer_fd); - if (ret < 0) { - ret = LTTCOMM_KERN_SESS_FAIL; - goto error; - } - - /* Code flow safety */ - assert(session->kernel_session); - - /* Copy session output to the newly created Kernel session */ - ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session); - if (ret != LTTCOMM_OK) { - goto error; - } - - /* Create directory(ies) on local filesystem. */ - if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL && - strlen(session->kernel_session->consumer->dst.trace_path) > 0) { - ret = run_as_mkdir_recursive( - session->kernel_session->consumer->dst.trace_path, - S_IRWXU | S_IRWXG, session->uid, session->gid); - if (ret < 0) { - if (ret != -EEXIST) { - ERR("Trace directory creation error"); - goto error; - } - } - } - - session->kernel_session->uid = session->uid; - session->kernel_session->gid = session->gid; - - return LTTCOMM_OK; - -error: - trace_kernel_destroy_session(session->kernel_session); - session->kernel_session = NULL; - return ret; -} - -/* - * Check if the UID or GID match the session. Root user has access to all - * sessions. - */ -static int session_access_ok(struct ltt_session *session, uid_t uid, gid_t gid) -{ - if (uid != session->uid && gid != session->gid && uid != 0) { - return 0; - } else { - return 1; - } -} - -/* - * Count number of session permitted by uid/gid. - */ -static unsigned int lttng_sessions_count(uid_t uid, gid_t gid) -{ - unsigned int i = 0; - struct ltt_session *session; - - DBG("Counting number of available session for UID %d GID %d", - uid, gid); - cds_list_for_each_entry(session, &session_list_ptr->head, list) { - /* - * Only list the sessions the user can control. - */ - if (!session_access_ok(session, uid, gid)) { - continue; - } - i++; - } - return i; -} - -/* - * Using the session list, filled a lttng_session array to send back to the - * client for session listing. - * - * The session list lock MUST be acquired before calling this function. Use - * session_lock_list() and session_unlock_list(). - */ -static void list_lttng_sessions(struct lttng_session *sessions, uid_t uid, - gid_t gid) -{ - unsigned int i = 0; - struct ltt_session *session; - - DBG("Getting all available session for UID %d GID %d", - uid, gid); - /* - * Iterate over session list and append data after the control struct in - * the buffer. - */ - cds_list_for_each_entry(session, &session_list_ptr->head, list) { - /* - * Only list the sessions the user can control. - */ - if (!session_access_ok(session, uid, gid)) { - continue; - } - strncpy(sessions[i].path, session->path, PATH_MAX); - sessions[i].path[PATH_MAX - 1] = '\0'; - strncpy(sessions[i].name, session->name, NAME_MAX); - sessions[i].name[NAME_MAX - 1] = '\0'; - sessions[i].enabled = session->enabled; - i++; - } -} - -/* - * Fill lttng_channel array of all channels. - */ -static void list_lttng_channels(int domain, struct ltt_session *session, - struct lttng_channel *channels) -{ - int i = 0; - struct ltt_kernel_channel *kchan; - - DBG("Listing channels for session %s", session->name); - - switch (domain) { - case LTTNG_DOMAIN_KERNEL: - /* Kernel channels */ - if (session->kernel_session != NULL) { - cds_list_for_each_entry(kchan, - &session->kernel_session->channel_list.head, list) { - /* Copy lttng_channel struct to array */ - memcpy(&channels[i], kchan->channel, sizeof(struct lttng_channel)); - channels[i].enabled = kchan->enabled; - i++; - } - } - break; - case LTTNG_DOMAIN_UST: - { - struct lttng_ht_iter iter; - struct ltt_ust_channel *uchan; - - cds_lfht_for_each_entry(session->ust_session->domain_global.channels->ht, - &iter.iter, uchan, node.node) { - strncpy(channels[i].name, uchan->name, LTTNG_SYMBOL_NAME_LEN); - channels[i].attr.overwrite = uchan->attr.overwrite; - channels[i].attr.subbuf_size = uchan->attr.subbuf_size; - channels[i].attr.num_subbuf = uchan->attr.num_subbuf; - channels[i].attr.switch_timer_interval = - uchan->attr.switch_timer_interval; - channels[i].attr.read_timer_interval = - uchan->attr.read_timer_interval; - channels[i].enabled = uchan->enabled; - switch (uchan->attr.output) { - case LTTNG_UST_MMAP: - default: - channels[i].attr.output = LTTNG_EVENT_MMAP; - break; - } - i++; - } - break; - } - default: - break; - } -} - -/* - * Create a list of ust global domain events. - */ -static int list_lttng_ust_global_events(char *channel_name, - struct ltt_ust_domain_global *ust_global, struct lttng_event **events) -{ - int i = 0, ret = 0; - unsigned int nb_event = 0; - struct lttng_ht_iter iter; - struct lttng_ht_node_str *node; - struct ltt_ust_channel *uchan; - struct ltt_ust_event *uevent; - struct lttng_event *tmp; - - DBG("Listing UST global events for channel %s", channel_name); - - rcu_read_lock(); - - lttng_ht_lookup(ust_global->channels, (void *)channel_name, &iter); - node = lttng_ht_iter_get_node_str(&iter); - if (node == NULL) { - ret = -LTTCOMM_UST_CHAN_NOT_FOUND; - goto error; - } - - uchan = caa_container_of(&node->node, struct ltt_ust_channel, node.node); - - nb_event += lttng_ht_get_count(uchan->events); - - if (nb_event == 0) { - ret = nb_event; - goto error; - } - - DBG3("Listing UST global %d events", nb_event); - - tmp = zmalloc(nb_event * sizeof(struct lttng_event)); - if (tmp == NULL) { - ret = -LTTCOMM_FATAL; - goto error; - } - - cds_lfht_for_each_entry(uchan->events->ht, &iter.iter, uevent, node.node) { - strncpy(tmp[i].name, uevent->attr.name, LTTNG_SYMBOL_NAME_LEN); - tmp[i].name[LTTNG_SYMBOL_NAME_LEN - 1] = '\0'; - tmp[i].enabled = uevent->enabled; - switch (uevent->attr.instrumentation) { - case LTTNG_UST_TRACEPOINT: - tmp[i].type = LTTNG_EVENT_TRACEPOINT; - break; - case LTTNG_UST_PROBE: - tmp[i].type = LTTNG_EVENT_PROBE; - break; - case LTTNG_UST_FUNCTION: - tmp[i].type = LTTNG_EVENT_FUNCTION; - break; - } - tmp[i].loglevel = uevent->attr.loglevel; - switch (uevent->attr.loglevel_type) { - case LTTNG_UST_LOGLEVEL_ALL: - tmp[i].loglevel_type = LTTNG_EVENT_LOGLEVEL_ALL; - break; - case LTTNG_UST_LOGLEVEL_RANGE: - tmp[i].loglevel_type = LTTNG_EVENT_LOGLEVEL_RANGE; - break; - case LTTNG_UST_LOGLEVEL_SINGLE: - tmp[i].loglevel_type = LTTNG_EVENT_LOGLEVEL_SINGLE; - break; - } - if (uevent->filter) { - tmp[i].filter = 1; - } - i++; - } - - ret = nb_event; - *events = tmp; - -error: - rcu_read_unlock(); - return ret; -} - -/* - * Fill lttng_event array of all kernel events in the channel. - */ -static int list_lttng_kernel_events(char *channel_name, - struct ltt_kernel_session *kernel_session, struct lttng_event **events) -{ - int i = 0, ret; - unsigned int nb_event; - struct ltt_kernel_event *event; - struct ltt_kernel_channel *kchan; - - kchan = trace_kernel_get_channel_by_name(channel_name, kernel_session); - if (kchan == NULL) { - ret = LTTCOMM_KERN_CHAN_NOT_FOUND; - goto error; - } - - nb_event = kchan->event_count; - - DBG("Listing events for channel %s", kchan->channel->name); - - if (nb_event == 0) { - ret = nb_event; - goto error; - } - - *events = zmalloc(nb_event * sizeof(struct lttng_event)); - if (*events == NULL) { - ret = LTTCOMM_FATAL; - goto error; - } - - /* Kernel channels */ - cds_list_for_each_entry(event, &kchan->events_list.head , list) { - strncpy((*events)[i].name, event->event->name, LTTNG_SYMBOL_NAME_LEN); - (*events)[i].name[LTTNG_SYMBOL_NAME_LEN - 1] = '\0'; - (*events)[i].enabled = event->enabled; - switch (event->event->instrumentation) { - case LTTNG_KERNEL_TRACEPOINT: - (*events)[i].type = LTTNG_EVENT_TRACEPOINT; - break; - case LTTNG_KERNEL_KPROBE: - case LTTNG_KERNEL_KRETPROBE: - (*events)[i].type = LTTNG_EVENT_PROBE; - memcpy(&(*events)[i].attr.probe, &event->event->u.kprobe, - sizeof(struct lttng_kernel_kprobe)); - break; - case LTTNG_KERNEL_FUNCTION: - (*events)[i].type = LTTNG_EVENT_FUNCTION; - memcpy(&((*events)[i].attr.ftrace), &event->event->u.ftrace, - sizeof(struct lttng_kernel_function)); - break; - case LTTNG_KERNEL_NOOP: - (*events)[i].type = LTTNG_EVENT_NOOP; - break; - case LTTNG_KERNEL_SYSCALL: - (*events)[i].type = LTTNG_EVENT_SYSCALL; - break; - case LTTNG_KERNEL_ALL: - assert(0); - break; - } - i++; - } - - return nb_event; - -error: - return ret; -} - - -/* - * Add URI so the consumer output object. Set the correct path depending on the - * domain adding the default trace directory. - */ -static int add_uri_to_consumer(struct consumer_output *consumer, - struct lttng_uri *uri, int domain, const char *session_name) -{ - int ret = LTTCOMM_OK; - const char *default_trace_dir; - - assert(uri); - - if (consumer == NULL) { - DBG("No consumer detected. Don't add URI. Stopping."); - ret = LTTCOMM_NO_CONSUMER; - goto error; - } - - switch (domain) { - case LTTNG_DOMAIN_KERNEL: - default_trace_dir = DEFAULT_KERNEL_TRACE_DIR; - break; - case LTTNG_DOMAIN_UST: - default_trace_dir = DEFAULT_UST_TRACE_DIR; - break; - default: - /* - * This case is possible is we try to add the URI to the global tracing - * session consumer object which in this case there is no subdir. - */ - default_trace_dir = ""; - } - - switch (uri->dtype) { - case LTTNG_DST_IPV4: - case LTTNG_DST_IPV6: - DBG2("Setting network URI to consumer"); - - /* Set URI into consumer output object */ - ret = consumer_set_network_uri(consumer, uri); - if (ret < 0) { - ret = LTTCOMM_FATAL; - goto error; - } - - if (uri->stype == LTTNG_STREAM_CONTROL && strlen(uri->subdir) == 0) { - ret = set_consumer_subdir(consumer, session_name); - if (ret < 0) { - ret = LTTCOMM_FATAL; - goto error; - } - } - - if (uri->stype == LTTNG_STREAM_CONTROL) { - /* On a new subdir, reappend the default trace dir. */ - strncat(consumer->subdir, default_trace_dir, sizeof(consumer->subdir)); - DBG3("Append domain trace name to subdir %s", consumer->subdir); - } - - break; - case LTTNG_DST_PATH: - DBG2("Setting trace directory path from URI to %s", uri->dst.path); - memset(consumer->dst.trace_path, 0, - sizeof(consumer->dst.trace_path)); - strncpy(consumer->dst.trace_path, uri->dst.path, - sizeof(consumer->dst.trace_path)); - /* Append default trace dir */ - strncat(consumer->dst.trace_path, default_trace_dir, - sizeof(consumer->dst.trace_path)); - /* Flag consumer as local. */ - consumer->type = CONSUMER_DST_LOCAL; - break; - } - -error: - return ret; -} - -/* - * Command LTTNG_DISABLE_CHANNEL processed by the client thread. - */ -static int cmd_disable_channel(struct ltt_session *session, - int domain, char *channel_name) -{ - int ret; - struct ltt_ust_session *usess; - - usess = session->ust_session; - - switch (domain) { - case LTTNG_DOMAIN_KERNEL: - { - ret = channel_kernel_disable(session->kernel_session, - channel_name); - if (ret != LTTCOMM_OK) { - goto error; - } - - kernel_wait_quiescent(kernel_tracer_fd); - break; - } - case LTTNG_DOMAIN_UST: - { - struct ltt_ust_channel *uchan; - struct lttng_ht *chan_ht; - - chan_ht = usess->domain_global.channels; - - uchan = trace_ust_find_channel_by_name(chan_ht, channel_name); - if (uchan == NULL) { - ret = LTTCOMM_UST_CHAN_NOT_FOUND; - goto error; - } - - ret = channel_ust_disable(usess, domain, uchan); - if (ret != LTTCOMM_OK) { - goto error; - } - break; - } -#if 0 - case LTTNG_DOMAIN_UST_PID_FOLLOW_CHILDREN: - case LTTNG_DOMAIN_UST_EXEC_NAME: - case LTTNG_DOMAIN_UST_PID: -#endif - default: - ret = LTTCOMM_UNKNOWN_DOMAIN; - goto error; - } - - ret = LTTCOMM_OK; - -error: - return ret; -} - -/* - * Command LTTNG_ENABLE_CHANNEL processed by the client thread. - */ -static int cmd_enable_channel(struct ltt_session *session, - int domain, struct lttng_channel *attr) -{ - int ret; - struct ltt_ust_session *usess = session->ust_session; - struct lttng_ht *chan_ht; - - DBG("Enabling channel %s for session %s", attr->name, session->name); - - switch (domain) { - case LTTNG_DOMAIN_KERNEL: - { - struct ltt_kernel_channel *kchan; - - kchan = trace_kernel_get_channel_by_name(attr->name, - session->kernel_session); - if (kchan == NULL) { - ret = channel_kernel_create(session->kernel_session, - attr, kernel_poll_pipe[1]); - } else { - ret = channel_kernel_enable(session->kernel_session, kchan); - } - - if (ret != LTTCOMM_OK) { - goto error; - } - - kernel_wait_quiescent(kernel_tracer_fd); - break; - } - case LTTNG_DOMAIN_UST: - { - struct ltt_ust_channel *uchan; - - chan_ht = usess->domain_global.channels; - - uchan = trace_ust_find_channel_by_name(chan_ht, attr->name); - if (uchan == NULL) { - ret = channel_ust_create(usess, domain, attr); - } else { - ret = channel_ust_enable(usess, domain, uchan); - } - break; - } -#if 0 - case LTTNG_DOMAIN_UST_PID_FOLLOW_CHILDREN: - case LTTNG_DOMAIN_UST_EXEC_NAME: - case LTTNG_DOMAIN_UST_PID: -#endif - default: - ret = LTTCOMM_UNKNOWN_DOMAIN; - goto error; - } - -error: - return ret; -} - -/* - * Command LTTNG_DISABLE_EVENT processed by the client thread. - */ -static int cmd_disable_event(struct ltt_session *session, int domain, - char *channel_name, char *event_name) -{ - int ret; - - switch (domain) { - case LTTNG_DOMAIN_KERNEL: - { - struct ltt_kernel_channel *kchan; - struct ltt_kernel_session *ksess; - - ksess = session->kernel_session; - - kchan = trace_kernel_get_channel_by_name(channel_name, ksess); - if (kchan == NULL) { - ret = LTTCOMM_KERN_CHAN_NOT_FOUND; - goto error; - } - - ret = event_kernel_disable_tracepoint(ksess, kchan, event_name); - if (ret != LTTCOMM_OK) { - goto error; - } - - kernel_wait_quiescent(kernel_tracer_fd); - break; - } - case LTTNG_DOMAIN_UST: - { - struct ltt_ust_channel *uchan; - struct ltt_ust_session *usess; - - usess = session->ust_session; - - uchan = trace_ust_find_channel_by_name(usess->domain_global.channels, - channel_name); - if (uchan == NULL) { - ret = LTTCOMM_UST_CHAN_NOT_FOUND; - goto error; - } - - ret = event_ust_disable_tracepoint(usess, domain, uchan, event_name); - if (ret != LTTCOMM_OK) { - goto error; - } - - DBG3("Disable UST event %s in channel %s completed", event_name, - channel_name); - break; - } -#if 0 - case LTTNG_DOMAIN_UST_EXEC_NAME: - case LTTNG_DOMAIN_UST_PID: - case LTTNG_DOMAIN_UST_PID_FOLLOW_CHILDREN: -#endif - default: - ret = LTTCOMM_UND; - goto error; - } - - ret = LTTCOMM_OK; - -error: - return ret; -} - -/* - * Command LTTNG_DISABLE_ALL_EVENT processed by the client thread. - */ -static int cmd_disable_event_all(struct ltt_session *session, int domain, - char *channel_name) -{ - int ret; - - switch (domain) { - case LTTNG_DOMAIN_KERNEL: - { - struct ltt_kernel_session *ksess; - struct ltt_kernel_channel *kchan; - - ksess = session->kernel_session; - - kchan = trace_kernel_get_channel_by_name(channel_name, ksess); - if (kchan == NULL) { - ret = LTTCOMM_KERN_CHAN_NOT_FOUND; - goto error; - } - - ret = event_kernel_disable_all(ksess, kchan); - if (ret != LTTCOMM_OK) { - goto error; - } - - kernel_wait_quiescent(kernel_tracer_fd); - break; - } - case LTTNG_DOMAIN_UST: - { - struct ltt_ust_session *usess; - struct ltt_ust_channel *uchan; - - usess = session->ust_session; - - uchan = trace_ust_find_channel_by_name(usess->domain_global.channels, - channel_name); - if (uchan == NULL) { - ret = LTTCOMM_UST_CHAN_NOT_FOUND; - goto error; - } - - ret = event_ust_disable_all_tracepoints(usess, domain, uchan); - if (ret != 0) { - goto error; - } - - DBG3("Disable all UST events in channel %s completed", channel_name); - - break; - } -#if 0 - case LTTNG_DOMAIN_UST_EXEC_NAME: - case LTTNG_DOMAIN_UST_PID: - case LTTNG_DOMAIN_UST_PID_FOLLOW_CHILDREN: -#endif - default: - ret = LTTCOMM_UND; - goto error; - } - - ret = LTTCOMM_OK; - -error: - return ret; -} - -/* - * Command LTTNG_ADD_CONTEXT processed by the client thread. - */ -static int cmd_add_context(struct ltt_session *session, int domain, - char *channel_name, char *event_name, struct lttng_event_context *ctx) -{ - int ret; - - switch (domain) { - case LTTNG_DOMAIN_KERNEL: - /* Add kernel context to kernel tracer */ - ret = context_kernel_add(session->kernel_session, ctx, - event_name, channel_name); - if (ret != LTTCOMM_OK) { - goto error; - } - break; - case LTTNG_DOMAIN_UST: - { - struct ltt_ust_session *usess = session->ust_session; - - ret = context_ust_add(usess, domain, ctx, event_name, channel_name); - if (ret != LTTCOMM_OK) { - goto error; - } - break; - } -#if 0 - case LTTNG_DOMAIN_UST_EXEC_NAME: - case LTTNG_DOMAIN_UST_PID: - case LTTNG_DOMAIN_UST_PID_FOLLOW_CHILDREN: -#endif - default: - ret = LTTCOMM_UND; - goto error; - } - - ret = LTTCOMM_OK; - -error: - return ret; -} - -/* - * Command LTTNG_SET_FILTER processed by the client thread. - */ -static int cmd_set_filter(struct ltt_session *session, int domain, - char *channel_name, char *event_name, - struct lttng_filter_bytecode *bytecode) -{ - int ret; - - switch (domain) { - case LTTNG_DOMAIN_KERNEL: - ret = LTTCOMM_FATAL; - break; - case LTTNG_DOMAIN_UST: - { - struct ltt_ust_session *usess = session->ust_session; - - ret = filter_ust_set(usess, domain, bytecode, event_name, channel_name); - if (ret != LTTCOMM_OK) { - goto error; - } - break; - } -#if 0 - case LTTNG_DOMAIN_UST_EXEC_NAME: - case LTTNG_DOMAIN_UST_PID: - case LTTNG_DOMAIN_UST_PID_FOLLOW_CHILDREN: -#endif - default: - ret = LTTCOMM_UND; - goto error; - } - - ret = LTTCOMM_OK; - -error: - return ret; - -} - -/* - * Command LTTNG_ENABLE_EVENT processed by the client thread. - */ -static int cmd_enable_event(struct ltt_session *session, int domain, - char *channel_name, struct lttng_event *event) -{ - int ret; - struct lttng_channel *attr; - struct ltt_ust_session *usess = session->ust_session; - - switch (domain) { - case LTTNG_DOMAIN_KERNEL: - { - struct ltt_kernel_channel *kchan; - - kchan = trace_kernel_get_channel_by_name(channel_name, - session->kernel_session); - if (kchan == NULL) { - attr = channel_new_default_attr(domain); - if (attr == NULL) { - ret = LTTCOMM_FATAL; - goto error; - } - snprintf(attr->name, NAME_MAX, "%s", channel_name); - - /* This call will notify the kernel thread */ - ret = channel_kernel_create(session->kernel_session, - attr, kernel_poll_pipe[1]); - if (ret != LTTCOMM_OK) { - free(attr); - goto error; - } - free(attr); - } - - /* Get the newly created kernel channel pointer */ - kchan = trace_kernel_get_channel_by_name(channel_name, - session->kernel_session); - if (kchan == NULL) { - /* This sould not happen... */ - ret = LTTCOMM_FATAL; - goto error; - } - - ret = event_kernel_enable_tracepoint(session->kernel_session, kchan, - event); - if (ret != LTTCOMM_OK) { - goto error; - } - - kernel_wait_quiescent(kernel_tracer_fd); - break; - } - case LTTNG_DOMAIN_UST: - { - struct lttng_channel *attr; - struct ltt_ust_channel *uchan; - - /* Get channel from global UST domain */ - uchan = trace_ust_find_channel_by_name(usess->domain_global.channels, - channel_name); - if (uchan == NULL) { - /* Create default channel */ - attr = channel_new_default_attr(domain); - if (attr == NULL) { - ret = LTTCOMM_FATAL; - goto error; - } - snprintf(attr->name, NAME_MAX, "%s", channel_name); - attr->name[NAME_MAX - 1] = '\0'; - - ret = channel_ust_create(usess, domain, attr); - if (ret != LTTCOMM_OK) { - free(attr); - goto error; - } - free(attr); - - /* Get the newly created channel reference back */ - uchan = trace_ust_find_channel_by_name( - usess->domain_global.channels, channel_name); - if (uchan == NULL) { - /* Something is really wrong */ - ret = LTTCOMM_FATAL; - goto error; - } - } - - /* At this point, the session and channel exist on the tracer */ - ret = event_ust_enable_tracepoint(usess, domain, uchan, event); - if (ret != LTTCOMM_OK) { - goto error; - } - break; - } -#if 0 - case LTTNG_DOMAIN_UST_EXEC_NAME: - case LTTNG_DOMAIN_UST_PID: - case LTTNG_DOMAIN_UST_PID_FOLLOW_CHILDREN: -#endif - default: - ret = LTTCOMM_UND; - goto error; - } - - ret = LTTCOMM_OK; - -error: - return ret; -} - -/* - * Command LTTNG_ENABLE_ALL_EVENT processed by the client thread. - */ -static int cmd_enable_event_all(struct ltt_session *session, int domain, - char *channel_name, int event_type) -{ - int ret; - struct ltt_kernel_channel *kchan; - - switch (domain) { - case LTTNG_DOMAIN_KERNEL: - kchan = trace_kernel_get_channel_by_name(channel_name, - session->kernel_session); - if (kchan == NULL) { - /* This call will notify the kernel thread */ - ret = channel_kernel_create(session->kernel_session, NULL, - kernel_poll_pipe[1]); - if (ret != LTTCOMM_OK) { - goto error; - } - - /* Get the newly created kernel channel pointer */ - kchan = trace_kernel_get_channel_by_name(channel_name, - session->kernel_session); - if (kchan == NULL) { - /* This sould not happen... */ - ret = LTTCOMM_FATAL; - goto error; - } - - } - - switch (event_type) { - case LTTNG_EVENT_SYSCALL: - ret = event_kernel_enable_all_syscalls(session->kernel_session, - kchan, kernel_tracer_fd); - break; - case LTTNG_EVENT_TRACEPOINT: - /* - * This call enables all LTTNG_KERNEL_TRACEPOINTS and - * events already registered to the channel. - */ - ret = event_kernel_enable_all_tracepoints(session->kernel_session, - kchan, kernel_tracer_fd); - break; - case LTTNG_EVENT_ALL: - /* Enable syscalls and tracepoints */ - ret = event_kernel_enable_all(session->kernel_session, - kchan, kernel_tracer_fd); - break; - default: - ret = LTTCOMM_KERN_ENABLE_FAIL; - goto error; - } - - /* Manage return value */ - if (ret != LTTCOMM_OK) { - goto error; - } - - kernel_wait_quiescent(kernel_tracer_fd); - break; - case LTTNG_DOMAIN_UST: - { - struct lttng_channel *attr; - struct ltt_ust_channel *uchan; - struct ltt_ust_session *usess = session->ust_session; - - /* Get channel from global UST domain */ - uchan = trace_ust_find_channel_by_name(usess->domain_global.channels, - channel_name); - if (uchan == NULL) { - /* Create default channel */ - attr = channel_new_default_attr(domain); - if (attr == NULL) { - ret = LTTCOMM_FATAL; - goto error; - } - snprintf(attr->name, NAME_MAX, "%s", channel_name); - attr->name[NAME_MAX - 1] = '\0'; - - /* Use the internal command enable channel */ - ret = channel_ust_create(usess, domain, attr); - if (ret != LTTCOMM_OK) { - free(attr); - goto error; - } - free(attr); - - /* Get the newly created channel reference back */ - uchan = trace_ust_find_channel_by_name( - usess->domain_global.channels, channel_name); - if (uchan == NULL) { - /* Something is really wrong */ - ret = LTTCOMM_FATAL; - goto error; - } - } - - /* At this point, the session and channel exist on the tracer */ - - switch (event_type) { - case LTTNG_EVENT_ALL: - case LTTNG_EVENT_TRACEPOINT: - ret = event_ust_enable_all_tracepoints(usess, domain, uchan); - if (ret != LTTCOMM_OK) { - goto error; - } - break; - default: - ret = LTTCOMM_UST_ENABLE_FAIL; - goto error; - } - - /* Manage return value */ - if (ret != LTTCOMM_OK) { - goto error; - } - - break; - } -#if 0 - case LTTNG_DOMAIN_UST_EXEC_NAME: - case LTTNG_DOMAIN_UST_PID: - case LTTNG_DOMAIN_UST_PID_FOLLOW_CHILDREN: -#endif - default: - ret = LTTCOMM_UND; - goto error; - } - - ret = LTTCOMM_OK; - -error: - return ret; -} - -/* - * Command LTTNG_LIST_TRACEPOINTS processed by the client thread. - */ -static ssize_t cmd_list_tracepoints(int domain, struct lttng_event **events) -{ - int ret; - ssize_t nb_events = 0; - - switch (domain) { - case LTTNG_DOMAIN_KERNEL: - nb_events = kernel_list_events(kernel_tracer_fd, events); - if (nb_events < 0) { - ret = LTTCOMM_KERN_LIST_FAIL; - goto error; - } - break; - case LTTNG_DOMAIN_UST: - nb_events = ust_app_list_events(events); - if (nb_events < 0) { - ret = LTTCOMM_UST_LIST_FAIL; - goto error; - } - break; - default: - ret = LTTCOMM_UND; - goto error; - } - - return nb_events; - -error: - /* Return negative value to differentiate return code */ - return -ret; -} - -/* - * Command LTTNG_LIST_TRACEPOINT_FIELDS processed by the client thread. - */ -static ssize_t cmd_list_tracepoint_fields(int domain, - struct lttng_event_field **fields) -{ - int ret; - ssize_t nb_fields = 0; - - switch (domain) { - case LTTNG_DOMAIN_UST: - nb_fields = ust_app_list_event_fields(fields); - if (nb_fields < 0) { - ret = LTTCOMM_UST_LIST_FAIL; - goto error; - } - break; - case LTTNG_DOMAIN_KERNEL: - default: /* fall-through */ - ret = LTTCOMM_UND; - goto error; - } - - return nb_fields; - -error: - /* Return negative value to differentiate return code */ - return -ret; -} - -/* - * Command LTTNG_START_TRACE processed by the client thread. - */ -static int cmd_start_trace(struct ltt_session *session) -{ - int ret; - struct ltt_kernel_session *ksession; - struct ltt_ust_session *usess; - struct ltt_kernel_channel *kchan; - - /* Ease our life a bit ;) */ - ksession = session->kernel_session; - usess = session->ust_session; - - if (session->enabled) { - /* Already started. */ - ret = LTTCOMM_TRACE_ALREADY_STARTED; - goto error; - } - - session->enabled = 1; - - ret = setup_relayd(session); - if (ret != LTTCOMM_OK) { - ERR("Error setting up relayd for session %s", session->name); - goto error; - } - - /* Kernel tracing */ - if (ksession != NULL) { - /* Open kernel metadata */ - if (ksession->metadata == NULL) { - ret = kernel_open_metadata(ksession); - if (ret < 0) { - ret = LTTCOMM_KERN_META_FAIL; - goto error; - } - } - - /* Open kernel metadata stream */ - if (ksession->metadata_stream_fd < 0) { - ret = kernel_open_metadata_stream(ksession); - if (ret < 0) { - ERR("Kernel create metadata stream failed"); - ret = LTTCOMM_KERN_STREAM_FAIL; - goto error; - } - } - - /* For each channel */ - cds_list_for_each_entry(kchan, &ksession->channel_list.head, list) { - if (kchan->stream_count == 0) { - ret = kernel_open_channel_stream(kchan); - if (ret < 0) { - ret = LTTCOMM_KERN_STREAM_FAIL; - goto error; - } - /* Update the stream global counter */ - ksession->stream_count_global += ret; - } - } - - /* Setup kernel consumer socket and send fds to it */ - ret = init_kernel_tracing(ksession); - if (ret < 0) { - ret = LTTCOMM_KERN_START_FAIL; - goto error; - } - - /* This start the kernel tracing */ - ret = kernel_start_session(ksession); - if (ret < 0) { - ret = LTTCOMM_KERN_START_FAIL; - goto error; - } - - /* Quiescent wait after starting trace */ - kernel_wait_quiescent(kernel_tracer_fd); - } - - /* Flag session that trace should start automatically */ - if (usess) { - usess->start_trace = 1; - - ret = ust_app_start_trace_all(usess); - if (ret < 0) { - ret = LTTCOMM_UST_START_FAIL; - goto error; - } - } - - ret = LTTCOMM_OK; - -error: - return ret; -} - -/* - * Command LTTNG_STOP_TRACE processed by the client thread. - */ -static int cmd_stop_trace(struct ltt_session *session) -{ - int ret; - struct ltt_kernel_channel *kchan; - struct ltt_kernel_session *ksession; - struct ltt_ust_session *usess; - - /* Short cut */ - ksession = session->kernel_session; - usess = session->ust_session; - - if (!session->enabled) { - ret = LTTCOMM_TRACE_ALREADY_STOPPED; - goto error; - } - - session->enabled = 0; - - /* Kernel tracer */ - if (ksession != NULL) { - DBG("Stop kernel tracing"); - - /* Flush metadata if exist */ - if (ksession->metadata_stream_fd >= 0) { - ret = kernel_metadata_flush_buffer(ksession->metadata_stream_fd); - if (ret < 0) { - ERR("Kernel metadata flush failed"); - } - } - - /* Flush all buffers before stopping */ - cds_list_for_each_entry(kchan, &ksession->channel_list.head, list) { - ret = kernel_flush_buffer(kchan); - if (ret < 0) { - ERR("Kernel flush buffer error"); - } - } - - ret = kernel_stop_session(ksession); - if (ret < 0) { - ret = LTTCOMM_KERN_STOP_FAIL; - goto error; - } - - kernel_wait_quiescent(kernel_tracer_fd); - } - - if (usess) { - usess->start_trace = 0; - - ret = ust_app_stop_trace_all(usess); - if (ret < 0) { - ret = LTTCOMM_UST_STOP_FAIL; - goto error; - } - } - - ret = LTTCOMM_OK; - -error: - return ret; -} - -/* - * Command LTTNG_SET_CONSUMER_URI processed by the client thread. - */ -static int cmd_set_consumer_uri(int domain, struct ltt_session *session, - size_t nb_uri, struct lttng_uri *uris) -{ - int ret, i; - struct ltt_kernel_session *ksess = session->kernel_session; - struct ltt_ust_session *usess = session->ust_session; - struct consumer_output *consumer = NULL; - - assert(session); - assert(uris); - assert(nb_uri > 0); - - /* Can't enable consumer after session started. */ - if (session->enabled) { - ret = LTTCOMM_TRACE_ALREADY_STARTED; - goto error; - } - - if (!session->start_consumer) { - ret = LTTCOMM_NO_CONSUMER; - goto error; - } - - /* - * This case switch makes sure the domain session has a temporary consumer - * so the URL can be set. - */ - switch (domain) { - case 0: - /* Code flow error. A session MUST always have a consumer object */ - assert(session->consumer); - /* - * The URL will be added to the tracing session consumer instead of a - * specific domain consumer. - */ - consumer = session->consumer; - break; - case LTTNG_DOMAIN_KERNEL: - /* Code flow error if we don't have a kernel session here. */ - assert(ksess); - - /* Create consumer output if none exists */ - consumer = ksess->tmp_consumer; - if (consumer == NULL) { - consumer = consumer_copy_output(ksess->consumer); - if (consumer == NULL) { - ret = LTTCOMM_FATAL; - goto error; - } - /* Trash the consumer subdir, we are about to set a new one. */ - memset(consumer->subdir, 0, sizeof(consumer->subdir)); - ksess->tmp_consumer = consumer; - } - - break; - case LTTNG_DOMAIN_UST: - /* Code flow error if we don't have a kernel session here. */ - assert(usess); - - /* Create consumer output if none exists */ - consumer = usess->tmp_consumer; - if (consumer == NULL) { - consumer = consumer_copy_output(usess->consumer); - if (consumer == NULL) { - ret = LTTCOMM_FATAL; - goto error; - } - /* Trash the consumer subdir, we are about to set a new one. */ - memset(consumer->subdir, 0, sizeof(consumer->subdir)); - usess->tmp_consumer = consumer; - } - - break; - } - - for (i = 0; i < nb_uri; i++) { - struct consumer_socket *socket; - struct lttng_ht_iter iter; - - ret = add_uri_to_consumer(consumer, &uris[i], domain, session->name); - if (ret < 0) { - goto error; - } - - /* - * Don't send relayd socket if URI is NOT remote or if the relayd - * sockets for the session are already sent. - */ - if (uris[i].dtype == LTTNG_DST_PATH || - consumer->dst.net.relayd_socks_sent) { - continue; - } - - /* Try to send relayd URI to the consumer if exist. */ - cds_lfht_for_each_entry(consumer->socks->ht, &iter.iter, - socket, node.node) { - - /* A socket in the HT should never have a negative fd */ - assert(socket->fd >= 0); - - pthread_mutex_lock(socket->lock); - ret = send_socket_relayd_consumer(domain, session, &uris[i], - consumer, socket->fd); - pthread_mutex_unlock(socket->lock); - if (ret != LTTCOMM_OK) { - goto error; - } - } - } - - /* All good! */ - ret = LTTCOMM_OK; - -error: - return ret; -} - - -/* - * Command LTTNG_CREATE_SESSION processed by the client thread. - */ -static int cmd_create_session_uri(char *name, struct lttng_uri *uris, - size_t nb_uri, lttng_sock_cred *creds) -{ - int ret; - char *path = NULL; - struct ltt_session *session; - //struct consumer_output *consumer = NULL; - //struct lttng_uri *ctrl_uri, *data_uri = NULL; - - assert(name); - - /* - * Verify if the session already exist - * - * XXX: There is no need for the session lock list here since the caller - * (process_client_msg) is holding it. We might want to change that so a - * single command does not lock the entire session list. - */ - session = session_find_by_name(name); - if (session != NULL) { - ret = LTTCOMM_EXIST_SESS; - goto find_error; - } - - /* Create tracing session in the registry */ - ret = session_create(name, path, LTTNG_SOCK_GET_UID_CRED(creds), - LTTNG_SOCK_GET_GID_CRED(creds)); - if (ret != LTTCOMM_OK) { - goto session_error; - } - - /* - * Get the newly created session pointer back - * - * XXX: There is no need for the session lock list here since the caller - * (process_client_msg) is holding it. We might want to change that so a - * single command does not lock the entire session list. - */ - session = session_find_by_name(name); - assert(session); - - /* Create default consumer output for the session not yet created. */ - session->consumer = consumer_create_output(CONSUMER_DST_LOCAL); - if (session->consumer == NULL) { - ret = LTTCOMM_FATAL; - goto consumer_error; - } - - /* - * This means that the lttng_create_session call was called with the _path_ - * argument set to NULL. - */ - if (uris == NULL) { - /* - * At this point, we'll skip the consumer URI setup and create a - * session with a NULL path which will flag the session to NOT spawn a - * consumer. - */ - DBG("Create session %s with NO uri, skipping consumer setup", name); - goto end; - } - - session->start_consumer = 1; - - ret = cmd_set_consumer_uri(0, session, nb_uri, uris); - if (ret != LTTCOMM_OK) { - goto consumer_error; - } - - session->consumer->enabled = 1; - -end: - return LTTCOMM_OK; - -consumer_error: - session_destroy(session); -session_error: -find_error: - return ret; -} - -/* - * Command LTTNG_DESTROY_SESSION processed by the client thread. - */ -static int cmd_destroy_session(struct ltt_session *session, char *name) -{ - int ret; - - /* Safety net */ - assert(session); - - /* Clean kernel session teardown */ - teardown_kernel_session(session); - /* UST session teardown */ - teardown_ust_session(session); - - /* - * Must notify the kernel thread here to update it's poll setin order - * to remove the channel(s)' fd just destroyed. - */ - ret = notify_thread_pipe(kernel_poll_pipe[1]); - if (ret < 0) { - PERROR("write kernel poll pipe"); - } - - ret = session_destroy(session); - - return ret; -} - -/* - * Command LTTNG_CALIBRATE processed by the client thread. - */ -static int cmd_calibrate(int domain, struct lttng_calibrate *calibrate) -{ - int ret; - - switch (domain) { - case LTTNG_DOMAIN_KERNEL: - { - struct lttng_kernel_calibrate kcalibrate; - - kcalibrate.type = calibrate->type; - ret = kernel_calibrate(kernel_tracer_fd, &kcalibrate); - if (ret < 0) { - ret = LTTCOMM_KERN_ENABLE_FAIL; - goto error; - } - break; - } - case LTTNG_DOMAIN_UST: - { - struct lttng_ust_calibrate ucalibrate; - - ucalibrate.type = calibrate->type; - ret = ust_app_calibrate_glb(&ucalibrate); - if (ret < 0) { - ret = LTTCOMM_UST_CALIBRATE_FAIL; - goto error; - } - break; - } - default: - ret = LTTCOMM_UND; - goto error; - } - - ret = LTTCOMM_OK; - -error: - return ret; -} - -/* - * Command LTTNG_REGISTER_CONSUMER processed by the client thread. - */ -static int cmd_register_consumer(struct ltt_session *session, int domain, - char *sock_path) -{ - int ret, sock; - struct consumer_socket *socket; - - switch (domain) { - case LTTNG_DOMAIN_KERNEL: - /* Can't register a consumer if there is already one */ - if (session->kernel_session->consumer_fds_sent != 0) { - ret = LTTCOMM_KERN_CONSUMER_FAIL; - goto error; - } - - sock = lttcomm_connect_unix_sock(sock_path); - if (sock < 0) { - ret = LTTCOMM_CONNECT_FAIL; - goto error; - } - - socket = consumer_allocate_socket(sock); - if (socket == NULL) { - ret = LTTCOMM_FATAL; - close(sock); - goto error; - } - - socket->lock = zmalloc(sizeof(pthread_mutex_t)); - if (socket->lock == NULL) { - PERROR("zmalloc pthread mutex"); - ret = LTTCOMM_FATAL; - goto error; - } - pthread_mutex_init(socket->lock, NULL); - - rcu_read_lock(); - consumer_add_socket(socket, session->kernel_session->consumer); - rcu_read_unlock(); - - pthread_mutex_lock(&kconsumer_data.pid_mutex); - kconsumer_data.pid = -1; - pthread_mutex_unlock(&kconsumer_data.pid_mutex); - - break; - default: - /* TODO: Userspace tracing */ - ret = LTTCOMM_UND; - goto error; - } - - ret = LTTCOMM_OK; - -error: - return ret; -} - -/* - * Command LTTNG_LIST_DOMAINS processed by the client thread. - */ -static ssize_t cmd_list_domains(struct ltt_session *session, - struct lttng_domain **domains) -{ - int ret, index = 0; - ssize_t nb_dom = 0; - - if (session->kernel_session != NULL) { - DBG3("Listing domains found kernel domain"); - nb_dom++; - } - - if (session->ust_session != NULL) { - DBG3("Listing domains found UST global domain"); - nb_dom++; - } + lus->uid = session->uid; + lus->gid = session->gid; + lus->output_traces = session->output_traces; + lus->snapshot_mode = session->snapshot_mode; + lus->live_timer_interval = session->live_timer; + session->ust_session = lus; - *domains = zmalloc(nb_dom * sizeof(struct lttng_domain)); - if (*domains == NULL) { - ret = -LTTCOMM_FATAL; + /* Copy session output to the newly created UST session */ + ret = copy_session_consumer(domain->type, session); + if (ret != LTTNG_OK) { goto error; } - if (session->kernel_session != NULL) { - (*domains)[index].type = LTTNG_DOMAIN_KERNEL; - index++; - } - - if (session->ust_session != NULL) { - (*domains)[index].type = LTTNG_DOMAIN_UST; - index++; - } - - return nb_dom; + return LTTNG_OK; error: + free(lus); + session->ust_session = NULL; return ret; } /* - * Command LTTNG_LIST_CHANNELS processed by the client thread. + * Create a kernel tracer session then create the default channel. */ -static ssize_t cmd_list_channels(int domain, struct ltt_session *session, - struct lttng_channel **channels) +static int create_kernel_session(struct ltt_session *session) { int ret; - ssize_t nb_chan = 0; - - switch (domain) { - case LTTNG_DOMAIN_KERNEL: - if (session->kernel_session != NULL) { - nb_chan = session->kernel_session->channel_count; - } - DBG3("Number of kernel channels %zd", nb_chan); - break; - case LTTNG_DOMAIN_UST: - if (session->ust_session != NULL) { - nb_chan = lttng_ht_get_count( - session->ust_session->domain_global.channels); - } - DBG3("Number of UST global channels %zd", nb_chan); - break; - default: - *channels = NULL; - ret = -LTTCOMM_UND; - goto error; - } - - if (nb_chan > 0) { - *channels = zmalloc(nb_chan * sizeof(struct lttng_channel)); - if (*channels == NULL) { - ret = -LTTCOMM_FATAL; - goto error; - } - - list_lttng_channels(domain, session, *channels); - } else { - *channels = NULL; - } - - return nb_chan; - -error: - return ret; -} -/* - * Command LTTNG_LIST_EVENTS processed by the client thread. - */ -static ssize_t cmd_list_events(int domain, struct ltt_session *session, - char *channel_name, struct lttng_event **events) -{ - int ret = 0; - ssize_t nb_event = 0; + DBG("Creating kernel session"); - switch (domain) { - case LTTNG_DOMAIN_KERNEL: - if (session->kernel_session != NULL) { - nb_event = list_lttng_kernel_events(channel_name, - session->kernel_session, events); - } - break; - case LTTNG_DOMAIN_UST: - { - if (session->ust_session != NULL) { - nb_event = list_lttng_ust_global_events(channel_name, - &session->ust_session->domain_global, events); - } - break; - } - default: - ret = -LTTCOMM_UND; + ret = kernel_create_session(session, kernel_tracer_fd); + if (ret < 0) { + ret = LTTNG_ERR_KERN_SESS_FAIL; goto error; } - ret = nb_event; - -error: - return ret; -} - -/* - * Command LTTNG_DISABLE_CONSUMER processed by the client thread. - */ -static int cmd_disable_consumer(int domain, struct ltt_session *session) -{ - int ret; - struct ltt_kernel_session *ksess = session->kernel_session; - struct ltt_ust_session *usess = session->ust_session; - struct consumer_output *consumer; - - assert(session); + /* Code flow safety */ + assert(session->kernel_session); - if (session->enabled) { - /* Can't disable consumer on an already started session */ - ret = LTTCOMM_TRACE_ALREADY_STARTED; + /* Copy session output to the newly created Kernel session */ + ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session); + if (ret != LTTNG_OK) { goto error; } - if (!session->start_consumer) { - ret = LTTCOMM_NO_CONSUMER; - goto error; + /* Create directory(ies) on local filesystem. */ + if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL && + strlen(session->kernel_session->consumer->dst.trace_path) > 0) { + ret = run_as_mkdir_recursive( + session->kernel_session->consumer->dst.trace_path, + S_IRWXU | S_IRWXG, session->uid, session->gid); + if (ret < 0) { + if (ret != -EEXIST) { + ERR("Trace directory creation error"); + goto error; + } + } } - switch (domain) { - case 0: - DBG("Disable tracing session %s consumer", session->name); - consumer = session->consumer; - break; - case LTTNG_DOMAIN_KERNEL: - /* Code flow error if we don't have a kernel session here. */ - assert(ksess); - - DBG("Disabling kernel consumer"); - consumer = ksess->consumer; - - break; - case LTTNG_DOMAIN_UST: - /* Code flow error if we don't have a UST session here. */ - assert(usess); - - DBG("Disabling UST consumer"); - consumer = usess->consumer; - - break; - default: - ret = LTTCOMM_UNKNOWN_DOMAIN; - goto error; - } + session->kernel_session->uid = session->uid; + session->kernel_session->gid = session->gid; + session->kernel_session->output_traces = session->output_traces; + session->kernel_session->snapshot_mode = session->snapshot_mode; - if (consumer) { - consumer->enabled = 0; - /* Success at this point */ - ret = LTTCOMM_OK; - } else { - ret = LTTCOMM_NO_CONSUMER; - } + return LTTNG_OK; error: + trace_kernel_destroy_session(session->kernel_session); + session->kernel_session = NULL; return ret; } /* - * Command LTTNG_ENABLE_CONSUMER processed by the client thread. + * Count number of session permitted by uid/gid. */ -static int cmd_enable_consumer(int domain, struct ltt_session *session) +static unsigned int lttng_sessions_count(uid_t uid, gid_t gid) { - int ret; - struct ltt_kernel_session *ksess = session->kernel_session; - struct ltt_ust_session *usess = session->ust_session; - struct consumer_output *consumer = NULL; - - assert(session); - - /* Can't enable consumer after session started. */ - if (session->enabled) { - ret = LTTCOMM_TRACE_ALREADY_STARTED; - goto error; - } - - if (!session->start_consumer) { - ret = LTTCOMM_NO_CONSUMER; - goto error; - } - - switch (domain) { - case 0: - assert(session->consumer); - consumer = session->consumer; - break; - case LTTNG_DOMAIN_KERNEL: - /* Code flow error if we don't have a kernel session here. */ - assert(ksess); - - /* - * Check if we have already sent fds to the consumer. In that case, - * the enable-consumer command can't be used because a start trace - * had previously occured. - */ - if (ksess->consumer_fds_sent) { - ret = LTTCOMM_ENABLE_CONSUMER_FAIL; - goto error; - } - - consumer = ksess->tmp_consumer; - if (consumer == NULL) { - ret = LTTCOMM_OK; - /* No temp. consumer output exists. Using the current one. */ - DBG3("No temporary consumer. Using default"); - consumer = ksess->consumer; - goto error; - } - - switch (consumer->type) { - case CONSUMER_DST_LOCAL: - DBG2("Consumer output is local. Creating directory(ies)"); - - /* Create directory(ies) */ - ret = run_as_mkdir_recursive(consumer->dst.trace_path, - S_IRWXU | S_IRWXG, session->uid, session->gid); - if (ret < 0) { - if (ret != -EEXIST) { - ERR("Trace directory creation error"); - ret = LTTCOMM_FATAL; - goto error; - } - } - break; - case CONSUMER_DST_NET: - DBG2("Consumer output is network. Validating URIs"); - /* Validate if we have both control and data path set. */ - if (!consumer->dst.net.control_isset) { - ret = LTTCOMM_URL_CTRL_MISS; - goto error; - } - - if (!consumer->dst.net.data_isset) { - ret = LTTCOMM_URL_DATA_MISS; - goto error; - } - - /* Check established network session state */ - if (session->net_handle == 0) { - ret = LTTCOMM_ENABLE_CONSUMER_FAIL; - ERR("Session network handle is not set on enable-consumer"); - goto error; - } - - break; - } - - /* Append default kernel trace dir to subdir */ - strncat(ksess->consumer->subdir, DEFAULT_KERNEL_TRACE_DIR, - sizeof(ksess->consumer->subdir)); - - /* - * @session-lock - * This is race free for now since the session lock is acquired before - * ending up in this function. No other threads can access this kernel - * session without this lock hence freeing the consumer output object - * is valid. - */ - rcu_read_lock(); - consumer_destroy_output(ksess->consumer); - rcu_read_unlock(); - ksess->consumer = consumer; - ksess->tmp_consumer = NULL; - - break; - case LTTNG_DOMAIN_UST: - /* Code flow error if we don't have a UST session here. */ - assert(usess); + unsigned int i = 0; + struct ltt_session *session; + DBG("Counting number of available session for UID %d GID %d", + uid, gid); + cds_list_for_each_entry(session, &session_list_ptr->head, list) { /* - * Check if we have already sent fds to the consumer. In that case, - * the enable-consumer command can't be used because a start trace - * had previously occured. + * Only list the sessions the user can control. */ - if (usess->start_trace) { - ret = LTTCOMM_ENABLE_CONSUMER_FAIL; - goto error; - } - - consumer = usess->tmp_consumer; - if (consumer == NULL) { - ret = LTTCOMM_OK; - /* No temp. consumer output exists. Using the current one. */ - DBG3("No temporary consumer. Using default"); - consumer = usess->consumer; - goto error; - } - - switch (consumer->type) { - case CONSUMER_DST_LOCAL: - DBG2("Consumer output is local. Creating directory(ies)"); - - /* Create directory(ies) */ - ret = run_as_mkdir_recursive(consumer->dst.trace_path, - S_IRWXU | S_IRWXG, session->uid, session->gid); - if (ret < 0) { - if (ret != -EEXIST) { - ERR("Trace directory creation error"); - ret = LTTCOMM_FATAL; - goto error; - } - } - break; - case CONSUMER_DST_NET: - DBG2("Consumer output is network. Validating URIs"); - /* Validate if we have both control and data path set. */ - if (!consumer->dst.net.control_isset) { - ret = LTTCOMM_URL_CTRL_MISS; - goto error; - } - - if (!consumer->dst.net.data_isset) { - ret = LTTCOMM_URL_DATA_MISS; - goto error; - } - - /* Check established network session state */ - if (session->net_handle == 0) { - ret = LTTCOMM_ENABLE_CONSUMER_FAIL; - DBG2("Session network handle is not set on enable-consumer"); - goto error; - } - - if (consumer->net_seq_index == -1) { - ret = LTTCOMM_ENABLE_CONSUMER_FAIL; - DBG2("Network index is not set on the consumer"); - goto error; - } - - break; + if (!session_access_ok(session, uid, gid)) { + continue; } - - /* Append default kernel trace dir to subdir */ - strncat(usess->consumer->subdir, DEFAULT_UST_TRACE_DIR, - sizeof(usess->consumer->subdir)); - - /* - * @session-lock - * This is race free for now since the session lock is acquired before - * ending up in this function. No other threads can access this kernel - * session without this lock hence freeing the consumer output object - * is valid. - */ - rcu_read_lock(); - consumer_destroy_output(usess->consumer); - rcu_read_unlock(); - usess->consumer = consumer; - usess->tmp_consumer = NULL; - - break; - } - - /* Enable it */ - if (consumer) { - consumer->enabled = 1; - /* Success at this point */ - ret = LTTCOMM_OK; - } else { - /* Should not really happend... */ - ret = LTTCOMM_NO_CONSUMER; + i++; } - -error: - return ret; + return i; } /* @@ -4293,11 +2569,13 @@ error: * Return any error encountered or 0 for success. * * "sock" is only used for special-case var. len data. + * + * Should *NOT* be called with RCU read-side lock held. */ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, int *sock_error) { - int ret = LTTCOMM_OK; + int ret = LTTNG_OK; int need_tracing_session = 1; int need_domain; @@ -4307,11 +2585,18 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, switch (cmd_ctx->lsm->cmd_type) { case LTTNG_CREATE_SESSION: + case LTTNG_CREATE_SESSION_SNAPSHOT: + case LTTNG_CREATE_SESSION_LIVE: case LTTNG_DESTROY_SESSION: case LTTNG_LIST_SESSIONS: case LTTNG_LIST_DOMAINS: case LTTNG_START_TRACE: case LTTNG_STOP_TRACE: + case LTTNG_DATA_PENDING: + case LTTNG_SNAPSHOT_ADD_OUTPUT: + case LTTNG_SNAPSHOT_DEL_OUTPUT: + case LTTNG_SNAPSHOT_LIST_OUTPUT: + case LTTNG_SNAPSHOT_RECORD: need_domain = 0; break; default: @@ -4321,9 +2606,9 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, if (opt_no_kernel && need_domain && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) { if (!is_root) { - ret = LTTCOMM_NEED_ROOT_SESSIOND; + ret = LTTNG_ERR_NEED_ROOT_SESSIOND; } else { - ret = LTTCOMM_KERN_NA; + ret = LTTNG_ERR_KERN_NA; } goto error; } @@ -4332,7 +2617,8 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) { pthread_mutex_lock(&kconsumer_data.pid_mutex); if (kconsumer_data.pid > 0) { - ret = LTTCOMM_KERN_CONSUMER_FAIL; + ret = LTTNG_ERR_KERN_CONSUMER_FAIL; + pthread_mutex_unlock(&kconsumer_data.pid_mutex); goto error; } pthread_mutex_unlock(&kconsumer_data.pid_mutex); @@ -4363,6 +2649,8 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, /* Commands that DO NOT need a session. */ switch (cmd_ctx->lsm->cmd_type) { case LTTNG_CREATE_SESSION: + case LTTNG_CREATE_SESSION_SNAPSHOT: + case LTTNG_CREATE_SESSION_LIVE: case LTTNG_CALIBRATE: case LTTNG_LIST_SESSIONS: case LTTNG_LIST_TRACEPOINTS: @@ -4379,12 +2667,7 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, session_lock_list(); cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name); if (cmd_ctx->session == NULL) { - if (cmd_ctx->lsm->session.name != NULL) { - ret = LTTCOMM_SESS_NOT_FOUND; - } else { - /* If no session name specified */ - ret = LTTCOMM_SELECT_SESS; - } + ret = LTTNG_ERR_SESS_NOT_FOUND; goto error; } else { /* Acquire lock for the session */ @@ -4403,7 +2686,7 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, switch (cmd_ctx->lsm->domain.type) { case LTTNG_DOMAIN_KERNEL: if (!is_root) { - ret = LTTCOMM_NEED_ROOT_SESSIOND; + ret = LTTNG_ERR_NEED_ROOT_SESSIOND; goto error; } @@ -4418,7 +2701,7 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, /* Consumer is in an ERROR state. Report back to client */ if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) { - ret = LTTCOMM_NO_KERNCONSUMERD; + ret = LTTNG_ERR_NO_KERNCONSUMERD; goto error; } @@ -4427,7 +2710,7 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, if (cmd_ctx->session->kernel_session == NULL) { ret = create_kernel_session(cmd_ctx->session); if (ret < 0) { - ret = LTTCOMM_KERN_SESS_FAIL; + ret = LTTNG_ERR_KERN_SESS_FAIL; goto error; } } @@ -4435,12 +2718,11 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, /* Start the kernel consumer daemon */ pthread_mutex_lock(&kconsumer_data.pid_mutex); if (kconsumer_data.pid == 0 && - cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER && - cmd_ctx->session->start_consumer) { + cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) { pthread_mutex_unlock(&kconsumer_data.pid_mutex); ret = start_consumerd(&kconsumer_data); if (ret < 0) { - ret = LTTCOMM_KERN_CONSUMER_FAIL; + ret = LTTNG_ERR_KERN_CONSUMER_FAIL; goto error; } uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED); @@ -4462,9 +2744,13 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, break; case LTTNG_DOMAIN_UST: { + if (!ust_app_supported()) { + ret = LTTNG_ERR_NO_UST; + goto error; + } /* Consumer is in an ERROR state. Report back to client */ if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) { - ret = LTTCOMM_NO_USTCONSUMERD; + ret = LTTNG_ERR_NO_USTCONSUMERD; goto error; } @@ -4473,7 +2759,7 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, if (cmd_ctx->session->ust_session == NULL) { ret = create_ust_session(cmd_ctx->session, &cmd_ctx->lsm->domain); - if (ret != LTTCOMM_OK) { + if (ret != LTTNG_OK) { goto error; } } @@ -4483,12 +2769,11 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, pthread_mutex_lock(&ustconsumer64_data.pid_mutex); if (consumerd64_bin[0] != '\0' && ustconsumer64_data.pid == 0 && - cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER && - cmd_ctx->session->start_consumer) { + cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) { pthread_mutex_unlock(&ustconsumer64_data.pid_mutex); ret = start_consumerd(&ustconsumer64_data); if (ret < 0) { - ret = LTTCOMM_UST_CONSUMER64_FAIL; + ret = LTTNG_ERR_UST_CONSUMER64_FAIL; uatomic_set(&ust_consumerd64_fd, -EINVAL); goto error; } @@ -4512,12 +2797,11 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, /* 32-bit */ if (consumerd32_bin[0] != '\0' && ustconsumer32_data.pid == 0 && - cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER && - cmd_ctx->session->start_consumer) { + cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) { pthread_mutex_unlock(&ustconsumer32_data.pid_mutex); ret = start_consumerd(&ustconsumer32_data); if (ret < 0) { - ret = LTTCOMM_UST_CONSUMER32_FAIL; + ret = LTTNG_ERR_UST_CONSUMER32_FAIL; uatomic_set(&ust_consumerd32_fd, -EINVAL); goto error; } @@ -4551,13 +2835,13 @@ skip_domain: switch (cmd_ctx->lsm->domain.type) { case LTTNG_DOMAIN_UST: if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) { - ret = LTTCOMM_NO_USTCONSUMERD; + ret = LTTNG_ERR_NO_USTCONSUMERD; goto error; } break; case LTTNG_DOMAIN_KERNEL: if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) { - ret = LTTCOMM_NO_KERNCONSUMERD; + ret = LTTNG_ERR_NO_KERNCONSUMERD; goto error; } break; @@ -4572,7 +2856,22 @@ skip_domain: if (!session_access_ok(cmd_ctx->session, LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds), LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) { - ret = LTTCOMM_EPERM; + ret = LTTNG_ERR_EPERM; + goto error; + } + } + + /* + * Send relayd information to consumer as soon as we have a domain and a + * session defined. + */ + if (cmd_ctx->session && need_domain) { + /* + * Setup relayd if not done yet. If the relayd information was already + * sent to the consumer, this call will gracefully return. + */ + ret = cmd_setup_relayd(cmd_ctx->session); + if (ret != LTTNG_OK) { goto error; } } @@ -4583,8 +2882,7 @@ skip_domain: { ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type, cmd_ctx->lsm->u.context.channel_name, - cmd_ctx->lsm->u.context.event_name, - &cmd_ctx->lsm->u.context.ctx); + &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]); break; } case LTTNG_DISABLE_CHANNEL: @@ -4608,55 +2906,26 @@ skip_domain: cmd_ctx->lsm->u.disable.channel_name); break; } - case LTTNG_DISABLE_CONSUMER: - { - ret = cmd_disable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session); - break; - } case LTTNG_ENABLE_CHANNEL: { - ret = cmd_enable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type, - &cmd_ctx->lsm->u.channel.chan); - break; - } - case LTTNG_ENABLE_CONSUMER: - { - /* - * XXX: 0 means that this URI should be applied on the session. Should - * be a DOMAIN enuam. - */ - ret = cmd_enable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session); - if (ret != LTTCOMM_OK) { - goto error; - } - - if (cmd_ctx->lsm->domain.type == 0) { - /* Add the URI for the UST session if a consumer is present. */ - if (cmd_ctx->session->ust_session && - cmd_ctx->session->ust_session->consumer) { - ret = cmd_enable_consumer(LTTNG_DOMAIN_UST, cmd_ctx->session); - } else if (cmd_ctx->session->kernel_session && - cmd_ctx->session->kernel_session->consumer) { - ret = cmd_enable_consumer(LTTNG_DOMAIN_KERNEL, - cmd_ctx->session); - } - } + ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain, + &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]); break; } case LTTNG_ENABLE_EVENT: { - ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type, + ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain, cmd_ctx->lsm->u.enable.channel_name, - &cmd_ctx->lsm->u.enable.event); + &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]); break; } case LTTNG_ENABLE_ALL_EVENT: { DBG("Enabling all events"); - ret = cmd_enable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type, + ret = cmd_enable_event_all(cmd_ctx->session, &cmd_ctx->lsm->domain, cmd_ctx->lsm->u.enable.channel_name, - cmd_ctx->lsm->u.enable.event.type); + cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]); break; } case LTTNG_LIST_TRACEPOINTS: @@ -4666,6 +2935,7 @@ skip_domain: nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events); if (nb_events < 0) { + /* Return value is a negative lttng_error_code. */ ret = -nb_events; goto error; } @@ -4686,7 +2956,7 @@ skip_domain: free(events); - ret = LTTCOMM_OK; + ret = LTTNG_OK; break; } case LTTNG_LIST_TRACEPOINT_FIELDS: @@ -4697,6 +2967,7 @@ skip_domain: nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type, &fields); if (nb_fields < 0) { + /* Return value is a negative lttng_error_code. */ ret = -nb_fields; goto error; } @@ -4718,7 +2989,7 @@ skip_domain: free(fields); - ret = LTTCOMM_OK; + ret = LTTNG_OK; break; } case LTTNG_SET_CONSUMER_URI: @@ -4730,13 +3001,13 @@ skip_domain: len = nb_uri * sizeof(struct lttng_uri); if (nb_uri == 0) { - ret = LTTCOMM_INVALID; + ret = LTTNG_ERR_INVALID; goto error; } uris = zmalloc(len); if (uris == NULL) { - ret = LTTCOMM_FATAL; + ret = LTTNG_ERR_FATAL; goto error; } @@ -4746,13 +3017,15 @@ skip_domain: if (ret <= 0) { DBG("No URIs received from client... continuing"); *sock_error = 1; - ret = LTTCOMM_SESSION_FAIL; + ret = LTTNG_ERR_SESSION_FAIL; + free(uris); goto error; } ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session, nb_uri, uris); - if (ret != LTTCOMM_OK) { + if (ret != LTTNG_OK) { + free(uris); goto error; } @@ -4773,6 +3046,8 @@ skip_domain: } } + free(uris); + break; } case LTTNG_START_TRACE: @@ -4796,7 +3071,7 @@ skip_domain: if (nb_uri > 0) { uris = zmalloc(len); if (uris == NULL) { - ret = LTTCOMM_FATAL; + ret = LTTNG_ERR_FATAL; goto error; } @@ -4806,26 +3081,29 @@ skip_domain: if (ret <= 0) { DBG("No URIs received from client... continuing"); *sock_error = 1; - ret = LTTCOMM_SESSION_FAIL; + ret = LTTNG_ERR_SESSION_FAIL; + free(uris); goto error; } if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) { DBG("Creating session with ONE network URI is a bad call"); - ret = LTTCOMM_SESSION_FAIL; + ret = LTTNG_ERR_SESSION_FAIL; + free(uris); goto error; } } ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri, - &cmd_ctx->creds); + &cmd_ctx->creds, 0); + + free(uris); break; } case LTTNG_DESTROY_SESSION: { - ret = cmd_destroy_session(cmd_ctx->session, - cmd_ctx->lsm->session.name); + ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]); /* Set session to NULL so we do not unlock it after free. */ cmd_ctx->session = NULL; @@ -4838,12 +3116,14 @@ skip_domain: nb_dom = cmd_list_domains(cmd_ctx->session, &domains); if (nb_dom < 0) { + /* Return value is a negative lttng_error_code. */ ret = -nb_dom; goto error; } ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain)); if (ret < 0) { + free(domains); goto setup_error; } @@ -4853,7 +3133,7 @@ skip_domain: free(domains); - ret = LTTCOMM_OK; + ret = LTTNG_OK; break; } case LTTNG_LIST_CHANNELS: @@ -4864,12 +3144,14 @@ skip_domain: nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type, cmd_ctx->session, &channels); if (nb_chan < 0) { + /* Return value is a negative lttng_error_code. */ ret = -nb_chan; goto error; } ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel)); if (ret < 0) { + free(channels); goto setup_error; } @@ -4879,7 +3161,7 @@ skip_domain: free(channels); - ret = LTTCOMM_OK; + ret = LTTNG_OK; break; } case LTTNG_LIST_EVENTS: @@ -4890,12 +3172,14 @@ skip_domain: nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session, cmd_ctx->lsm->u.list.channel_name, &events); if (nb_event < 0) { + /* Return value is a negative lttng_error_code. */ ret = -nb_event; goto error; } ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event)); if (ret < 0) { + free(events); goto setup_error; } @@ -4905,7 +3189,7 @@ skip_domain: free(events); - ret = LTTCOMM_OK; + ret = LTTNG_OK; break; } case LTTNG_LIST_SESSIONS: @@ -4924,13 +3208,13 @@ skip_domain: } /* Filled the session array */ - list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload), + cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload), LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds), LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds)); session_unlock_list(); - ret = LTTCOMM_OK; + ret = LTTNG_OK; break; } case LTTNG_CALIBRATE: @@ -4941,49 +3225,207 @@ skip_domain: } case LTTNG_REGISTER_CONSUMER: { + struct consumer_data *cdata; + + switch (cmd_ctx->lsm->domain.type) { + case LTTNG_DOMAIN_KERNEL: + cdata = &kconsumer_data; + break; + default: + ret = LTTNG_ERR_UND; + goto error; + } + ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type, - cmd_ctx->lsm->u.reg.path); + cmd_ctx->lsm->u.reg.path, cdata); break; } - case LTTNG_SET_FILTER: + case LTTNG_ENABLE_EVENT_WITH_FILTER: { struct lttng_filter_bytecode *bytecode; - if (cmd_ctx->lsm->u.filter.bytecode_len > 65336) { - ret = LTTCOMM_FILTER_INVAL; + if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) { + ret = LTTNG_ERR_FILTER_INVAL; + goto error; + } + if (cmd_ctx->lsm->u.enable.bytecode_len == 0) { + ret = LTTNG_ERR_FILTER_INVAL; goto error; } - bytecode = zmalloc(cmd_ctx->lsm->u.filter.bytecode_len); + bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len); if (!bytecode) { - ret = LTTCOMM_FILTER_NOMEM; + ret = LTTNG_ERR_FILTER_NOMEM; goto error; } /* Receive var. len. data */ DBG("Receiving var len data from client ..."); ret = lttcomm_recv_unix_sock(sock, bytecode, - cmd_ctx->lsm->u.filter.bytecode_len); + cmd_ctx->lsm->u.enable.bytecode_len); if (ret <= 0) { DBG("Nothing recv() from client var len data... continuing"); *sock_error = 1; - ret = LTTCOMM_FILTER_INVAL; + ret = LTTNG_ERR_FILTER_INVAL; goto error; } if (bytecode->len + sizeof(*bytecode) - != cmd_ctx->lsm->u.filter.bytecode_len) { + != cmd_ctx->lsm->u.enable.bytecode_len) { free(bytecode); - ret = LTTCOMM_FILTER_INVAL; + ret = LTTNG_ERR_FILTER_INVAL; + goto error; + } + + ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain, + cmd_ctx->lsm->u.enable.channel_name, + &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]); + break; + } + case LTTNG_DATA_PENDING: + { + ret = cmd_data_pending(cmd_ctx->session); + break; + } + case LTTNG_SNAPSHOT_ADD_OUTPUT: + { + struct lttcomm_lttng_output_id reply; + + ret = cmd_snapshot_add_output(cmd_ctx->session, + &cmd_ctx->lsm->u.snapshot_output.output, &reply.id); + if (ret != LTTNG_OK) { + goto error; + } + + ret = setup_lttng_msg(cmd_ctx, sizeof(reply)); + if (ret < 0) { + goto setup_error; + } + + /* Copy output list into message payload */ + memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply)); + ret = LTTNG_OK; + break; + } + case LTTNG_SNAPSHOT_DEL_OUTPUT: + { + ret = cmd_snapshot_del_output(cmd_ctx->session, + &cmd_ctx->lsm->u.snapshot_output.output); + break; + } + case LTTNG_SNAPSHOT_LIST_OUTPUT: + { + ssize_t nb_output; + struct lttng_snapshot_output *outputs = NULL; + + nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs); + if (nb_output < 0) { + ret = -nb_output; goto error; } - ret = cmd_set_filter(cmd_ctx->session, cmd_ctx->lsm->domain.type, - cmd_ctx->lsm->u.filter.channel_name, - cmd_ctx->lsm->u.filter.event_name, - bytecode); + ret = setup_lttng_msg(cmd_ctx, + nb_output * sizeof(struct lttng_snapshot_output)); + if (ret < 0) { + free(outputs); + goto setup_error; + } + + if (outputs) { + /* Copy output list into message payload */ + memcpy(cmd_ctx->llm->payload, outputs, + nb_output * sizeof(struct lttng_snapshot_output)); + free(outputs); + } + + ret = LTTNG_OK; + break; + } + case LTTNG_SNAPSHOT_RECORD: + { + ret = cmd_snapshot_record(cmd_ctx->session, + &cmd_ctx->lsm->u.snapshot_record.output, + cmd_ctx->lsm->u.snapshot_record.wait); + break; + } + case LTTNG_CREATE_SESSION_SNAPSHOT: + { + size_t nb_uri, len; + struct lttng_uri *uris = NULL; + + nb_uri = cmd_ctx->lsm->u.uri.size; + len = nb_uri * sizeof(struct lttng_uri); + + if (nb_uri > 0) { + uris = zmalloc(len); + if (uris == NULL) { + ret = LTTNG_ERR_FATAL; + goto error; + } + + /* Receive variable len data */ + DBG("Waiting for %zu URIs from client ...", nb_uri); + ret = lttcomm_recv_unix_sock(sock, uris, len); + if (ret <= 0) { + DBG("No URIs received from client... continuing"); + *sock_error = 1; + ret = LTTNG_ERR_SESSION_FAIL; + free(uris); + goto error; + } + + if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) { + DBG("Creating session with ONE network URI is a bad call"); + ret = LTTNG_ERR_SESSION_FAIL; + free(uris); + goto error; + } + } + + ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris, + nb_uri, &cmd_ctx->creds); + free(uris); + break; + } + case LTTNG_CREATE_SESSION_LIVE: + { + size_t nb_uri, len; + struct lttng_uri *uris = NULL; + + nb_uri = cmd_ctx->lsm->u.uri.size; + len = nb_uri * sizeof(struct lttng_uri); + + if (nb_uri > 0) { + uris = zmalloc(len); + if (uris == NULL) { + ret = LTTNG_ERR_FATAL; + goto error; + } + + /* Receive variable len data */ + DBG("Waiting for %zu URIs from client ...", nb_uri); + ret = lttcomm_recv_unix_sock(sock, uris, len); + if (ret <= 0) { + DBG("No URIs received from client... continuing"); + *sock_error = 1; + ret = LTTNG_ERR_SESSION_FAIL; + free(uris); + goto error; + } + + if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) { + DBG("Creating session with ONE network URI is a bad call"); + ret = LTTNG_ERR_SESSION_FAIL; + free(uris); + goto error; + } + } + + ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, + nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval); + free(uris); break; } default: - ret = LTTCOMM_UND; + ret = LTTNG_ERR_UND; break; } @@ -5022,6 +3464,9 @@ static void *thread_manage_health(void *data) rcu_register_thread(); + /* We might hit an error path before this is created. */ + lttng_poll_init(&events); + /* Create unix socket */ sock = lttcomm_create_unix_sock(health_unix_sock_path); if (sock < 0) { @@ -5030,6 +3475,12 @@ static void *thread_manage_health(void *data) goto error; } + /* + * Set the CLOEXEC flag. Return code is useless because either way, the + * show must go on. + */ + (void) utils_set_fd_cloexec(sock); + ret = lttcomm_listen_unix_sock(sock); if (ret < 0) { goto error; @@ -5039,7 +3490,7 @@ static void *thread_manage_health(void *data) * Pass 2 as size here for the thread quit pipe and client_sock. Nothing * more will be added to this poll set. */ - ret = create_thread_poll_set(&events, 2); + ret = sessiond_set_thread_pollset(&events, 2); if (ret < 0) { goto error; } @@ -5053,8 +3504,6 @@ static void *thread_manage_health(void *data) while (1) { DBG("Health check ready"); - nb_fd = LTTNG_POLL_GETNB(&events); - /* Inifinite blocking call, waiting for transmission */ restart: ret = lttng_poll_wait(&events, -1); @@ -5068,13 +3517,15 @@ restart: goto error; } + nb_fd = ret; + for (i = 0; i < nb_fd; i++) { /* Fetch once the poll data */ revents = LTTNG_POLL_GETEV(&events, i); pollfd = LTTNG_POLL_GETFD(&events, i); /* Thread quit pipe has been closed. Killing thread. */ - ret = check_thread_quit_pipe(pollfd, revents); + ret = sessiond_check_thread_quit_pipe(pollfd, revents); if (ret) { err = 0; goto exit; @@ -5094,6 +3545,12 @@ restart: goto error; } + /* + * Set the CLOEXEC flag. Return code is useless because either way, the + * show must go on. + */ + (void) utils_set_fd_cloexec(new_sock); + DBG("Receiving data from client for health..."); ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg)); if (ret <= 0) { @@ -5110,30 +3567,42 @@ restart: switch (msg.component) { case LTTNG_HEALTH_CMD: - reply.ret_code = health_check_state(&health_thread_cmd); + reply.ret_code = health_check_state(health_sessiond, HEALTH_TYPE_CMD); break; case LTTNG_HEALTH_APP_MANAGE: - reply.ret_code = health_check_state(&health_thread_app_manage); + reply.ret_code = health_check_state(health_sessiond, HEALTH_TYPE_APP_MANAGE); break; case LTTNG_HEALTH_APP_REG: - reply.ret_code = health_check_state(&health_thread_app_reg); + reply.ret_code = health_check_state(health_sessiond, HEALTH_TYPE_APP_REG); break; case LTTNG_HEALTH_KERNEL: - reply.ret_code = health_check_state(&health_thread_kernel); + reply.ret_code = health_check_state(health_sessiond, HEALTH_TYPE_KERNEL); break; case LTTNG_HEALTH_CONSUMER: reply.ret_code = check_consumer_health(); break; + case LTTNG_HEALTH_HT_CLEANUP: + reply.ret_code = health_check_state(health_sessiond, HEALTH_TYPE_HT_CLEANUP); + break; + case LTTNG_HEALTH_APP_MANAGE_NOTIFY: + reply.ret_code = health_check_state(health_sessiond, HEALTH_TYPE_APP_MANAGE_NOTIFY); + break; + case LTTNG_HEALTH_APP_REG_DISPATCH: + reply.ret_code = health_check_state(health_sessiond, HEALTH_TYPE_APP_REG_DISPATCH); + break; case LTTNG_HEALTH_ALL: reply.ret_code = - health_check_state(&health_thread_app_manage) && - health_check_state(&health_thread_app_reg) && - health_check_state(&health_thread_cmd) && - health_check_state(&health_thread_kernel) && - check_consumer_health(); + health_check_state(health_sessiond, HEALTH_TYPE_APP_MANAGE) && + health_check_state(health_sessiond, HEALTH_TYPE_APP_REG) && + health_check_state(health_sessiond, HEALTH_TYPE_CMD) && + health_check_state(health_sessiond, HEALTH_TYPE_KERNEL) && + check_consumer_health() && + health_check_state(health_sessiond, HEALTH_TYPE_HT_CLEANUP) && + health_check_state(health_sessiond, HEALTH_TYPE_APP_MANAGE_NOTIFY) && + health_check_state(health_sessiond, HEALTH_TYPE_APP_REG_DISPATCH); break; default: - reply.ret_code = LTTCOMM_UND; + reply.ret_code = LTTNG_ERR_UND; break; } @@ -5175,12 +3644,6 @@ error: PERROR("close"); } } - if (new_sock >= 0) { - ret = close(new_sock); - if (ret) { - PERROR("close"); - } - } lttng_poll_clean(&events); @@ -5204,20 +3667,26 @@ static void *thread_manage_clients(void *data) rcu_register_thread(); - health_code_update(&health_thread_cmd); + health_register(health_sessiond, HEALTH_TYPE_CMD); + + if (testpoint(thread_manage_clients)) { + goto error_testpoint; + } + + health_code_update(); ret = lttcomm_listen_unix_sock(client_sock); if (ret < 0) { - goto error; + goto error_listen; } /* * Pass 2 as size here for the thread quit pipe and client_sock. Nothing * more will be added to this poll set. */ - ret = create_thread_poll_set(&events, 2); + ret = sessiond_set_thread_pollset(&events, 2); if (ret < 0) { - goto error; + goto error_create_poll; } /* Add the application registration socket */ @@ -5233,18 +3702,20 @@ static void *thread_manage_clients(void *data) kill(ppid, SIGUSR1); } - health_code_update(&health_thread_cmd); + if (testpoint(thread_manage_clients_before_loop)) { + goto error; + } + + health_code_update(); while (1) { DBG("Accepting client command ..."); - nb_fd = LTTNG_POLL_GETNB(&events); - /* Inifinite blocking call, waiting for transmission */ restart: - health_poll_update(&health_thread_cmd); + health_poll_entry(); ret = lttng_poll_wait(&events, -1); - health_poll_update(&health_thread_cmd); + health_poll_exit(); if (ret < 0) { /* * Restart interrupted system call. @@ -5255,15 +3726,17 @@ static void *thread_manage_clients(void *data) goto error; } + nb_fd = ret; + for (i = 0; i < nb_fd; i++) { /* Fetch once the poll data */ revents = LTTNG_POLL_GETEV(&events, i); pollfd = LTTNG_POLL_GETFD(&events, i); - health_code_update(&health_thread_cmd); + health_code_update(); /* Thread quit pipe has been closed. Killing thread. */ - ret = check_thread_quit_pipe(pollfd, revents); + ret = sessiond_check_thread_quit_pipe(pollfd, revents); if (ret) { err = 0; goto exit; @@ -5280,13 +3753,19 @@ static void *thread_manage_clients(void *data) DBG("Wait for client response"); - health_code_update(&health_thread_cmd); + health_code_update(); sock = lttcomm_accept_unix_sock(client_sock); if (sock < 0) { goto error; } + /* + * Set the CLOEXEC flag. Return code is useless because either way, the + * show must go on. + */ + (void) utils_set_fd_cloexec(sock); + /* Set socket option for credentials retrieval */ ret = lttcomm_setsockopt_creds_unix_sock(sock); if (ret < 0) { @@ -5310,7 +3789,7 @@ static void *thread_manage_clients(void *data) cmd_ctx->llm = NULL; cmd_ctx->session = NULL; - health_code_update(&health_thread_cmd); + health_code_update(); /* * Data is received from the lttng client. The struct @@ -5331,7 +3810,7 @@ static void *thread_manage_clients(void *data) continue; } - health_code_update(&health_thread_cmd); + health_code_update(); // TODO: Validate cmd_ctx including sanity check for // security purpose. @@ -5346,13 +3825,11 @@ static void *thread_manage_clients(void *data) ret = process_client_msg(cmd_ctx, sock, &sock_error); rcu_thread_offline(); if (ret < 0) { - if (sock_error) { - ret = close(sock); - if (ret) { - PERROR("close"); - } - sock = -1; + ret = close(sock); + if (ret) { + PERROR("close"); } + sock = -1; /* * TODO: Inform client somehow of the fatal error. At * this point, ret < 0 means that a zmalloc failed @@ -5364,7 +3841,7 @@ static void *thread_manage_clients(void *data) continue; } - health_code_update(&health_thread_cmd); + health_code_update(); DBG("Sending response (size: %d, retcode: %s)", cmd_ctx->lttng_msg_size, @@ -5383,18 +3860,24 @@ static void *thread_manage_clients(void *data) clean_command_ctx(&cmd_ctx); - health_code_update(&health_thread_cmd); + health_code_update(); } exit: error: - if (err) { - health_error(&health_thread_cmd); - ERR("Health error occurred in %s", __func__); + if (sock >= 0) { + ret = close(sock); + if (ret) { + PERROR("close"); + } } - health_exit(&health_thread_cmd); - DBG("Client thread dying"); + lttng_poll_clean(&events); + clean_command_ctx(&cmd_ctx); + +error_listen: +error_create_poll: +error_testpoint: unlink(client_unix_sock_path); if (client_sock >= 0) { ret = close(client_sock); @@ -5402,15 +3885,15 @@ error: PERROR("close"); } } - if (sock >= 0) { - ret = close(sock); - if (ret) { - PERROR("close"); - } + + if (err) { + health_error(); + ERR("Health error occurred in %s", __func__); } - lttng_poll_clean(&events); - clean_command_ctx(&cmd_ctx); + health_unregister(health_sessiond); + + DBG("Client thread dying"); rcu_unregister_thread(); return NULL; @@ -5442,6 +3925,7 @@ static void usage(void) fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n"); fprintf(stderr, " -q, --quiet No output at all.\n"); fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n"); + fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n"); fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n"); fprintf(stderr, " --no-kernel Disable kernel tracer\n"); } @@ -5475,12 +3959,13 @@ static int parse_args(int argc, char **argv) { "verbose", 0, 0, 'v' }, { "verbose-consumer", 0, 0, 'Z' }, { "no-kernel", 0, 0, 'N' }, + { "pidfile", 1, 0, 'p' }, { NULL, 0, 0, 0 } }; while (1) { int option_index = 0; - c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t", + c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:", long_options, &option_index); if (c == -1) { break; @@ -5557,6 +4042,9 @@ static int parse_args(int argc, char **argv) case 'T': consumerd64_libdir = optarg; break; + case 'p': + opt_pidfile = optarg; + break; default: /* Unknown option or other error. * Error is printed by getopt, just return */ @@ -5587,6 +4075,14 @@ static int init_daemon_socket(void) goto end; } + /* Set the cloexec flag */ + ret = utils_set_fd_cloexec(client_sock); + if (ret < 0) { + ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). " + "Continuing but note that the consumer daemon will have a " + "reference to this socket on exec()", client_sock); + } + /* File permission MUST be 660 */ ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP); if (ret < 0) { @@ -5603,6 +4099,14 @@ static int init_daemon_socket(void) goto end; } + /* Set the cloexec flag */ + ret = utils_set_fd_cloexec(apps_sock); + if (ret < 0) { + ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). " + "Continuing but note that the consumer daemon will have a " + "reference to this socket on exec()", apps_sock); + } + /* File permission MUST be 666 */ ret = chmod(apps_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH); @@ -5612,6 +4116,9 @@ static int init_daemon_socket(void) goto end; } + DBG3("Session daemon client socket %d and application socket %d created", + client_sock, apps_sock); + end: umask(old_umask); return ret; @@ -5645,12 +4152,12 @@ static int set_permissions(char *rundir) ret = allowed_group(); if (ret < 0) { WARN("No tracing group detected"); - ret = 0; - goto end; + /* Setting gid to 0 if no tracing group is found */ + gid = 0; + } else { + gid = ret; } - gid = ret; - /* Set lttng run dir */ ret = chown(rundir, 0, gid); if (ret < 0) { @@ -5658,7 +4165,7 @@ static int set_permissions(char *rundir) PERROR("chown"); } - /* Ensure tracing group can search the run dir */ + /* Ensure all applications and tracing group can search the run dir */ ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH); if (ret < 0) { ERR("Unable to set permissions on %s", rundir); @@ -5695,7 +4202,6 @@ static int set_permissions(char *rundir) DBG("All permissions are set"); -end: return ret; } @@ -5769,6 +4275,16 @@ static int set_consumer_sockets(struct consumer_data *consumer_data, goto error; } + /* + * Set the CLOEXEC flag. Return code is useless because either way, the + * show must go on. + */ + ret = utils_set_fd_cloexec(consumer_data->err_sock); + if (ret < 0) { + PERROR("utils_set_fd_cloexec"); + /* continue anyway */ + } + /* File permission MUST be 660 */ ret = chmod(consumer_data->err_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP); @@ -5864,6 +4380,38 @@ static void set_ulimit(void) } } +/* + * Write pidfile using the rundir and opt_pidfile. + */ +static void write_pidfile(void) +{ + int ret; + char pidfile_path[PATH_MAX]; + + assert(rundir); + + if (opt_pidfile) { + strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path)); + } else { + /* Build pidfile path from rundir and opt_pidfile. */ + ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/" + DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir); + if (ret < 0) { + PERROR("snprintf pidfile path"); + goto error; + } + } + + /* + * Create pid file in rundir. Return value is of no importance. The + * execution will continue even though we are not able to write the file. + */ + (void) utils_create_pid_file(getpid(), pidfile_path); + +error: + return; +} + /* * main */ @@ -5871,7 +4419,7 @@ int main(int argc, char **argv) { int ret = 0; void *status; - const char *home_path; + const char *home_path, *env_app_timeout; init_kernel_workarounds(); @@ -5879,9 +4427,16 @@ int main(int argc, char **argv) setup_consumerd_path(); + page_size = sysconf(_SC_PAGESIZE); + if (page_size < 0) { + PERROR("sysconf _SC_PAGESIZE"); + page_size = LONG_MAX; + WARN("Fallback page size to %ld", page_size); + } + /* Parse arguments */ progname = argv[0]; - if ((ret = parse_args(argc, argv) < 0)) { + if ((ret = parse_args(argc, argv)) < 0) { goto error; } @@ -5958,7 +4513,7 @@ int main(int argc, char **argv) DBG2("Kernel consumer cmd path: %s", kconsumer_data.cmd_unix_sock_path); } else { - home_path = get_home_dir(); + home_path = utils_get_home_dir(); if (home_path == NULL) { /* TODO: Add --socket PATH option */ ERR("Can't get HOME directory for sockets creation."); @@ -5995,7 +4550,7 @@ int main(int argc, char **argv) /* Set global SHM for ust */ if (strlen(wait_shm_path) == 0) { snprintf(wait_shm_path, PATH_MAX, - DEFAULT_HOME_APPS_WAIT_SHM_PATH, geteuid()); + DEFAULT_HOME_APPS_WAIT_SHM_PATH, getuid()); } /* Set health check Unix path */ @@ -6011,6 +4566,7 @@ int main(int argc, char **argv) DBG("Client socket path %s", client_unix_sock_path); DBG("Application socket path %s", apps_unix_sock_path); + DBG("Application wait path %s", wait_shm_path); DBG("LTTng run directory path: %s", rundir); /* 32 bits consumerd path setup */ @@ -6108,7 +4664,14 @@ int main(int argc, char **argv) } /* Setup the kernel pipe for waking up the kernel thread */ - if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) { + if (is_root && !opt_no_kernel) { + if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) { + goto exit; + } + } + + /* Setup the thread ht_cleanup communication pipe. */ + if (utils_create_pipe_cloexec(ht_cleanup_pipe) < 0) { goto exit; } @@ -6117,6 +4680,15 @@ int main(int argc, char **argv) goto exit; } + /* Setup the thread apps notify communication pipe. */ + if (utils_create_pipe_cloexec(apps_cmd_notify_pipe) < 0) { + goto exit; + } + + /* Initialize global buffer per UID and PID registry. */ + buffer_reg_init_uid_registry(); + buffer_reg_init_pid_registry(); + /* Init UST command queue. */ cds_wfq_init(&ust_cmd_queue.queue); @@ -6129,29 +4701,40 @@ int main(int argc, char **argv) /* Set up max poll set size */ lttng_poll_set_max_size(); - /* - * Set network sequence index to 1 for streams to match a relayd socket on - * the consumer side. - */ - uatomic_set(&relayd_net_seq_idx, 1); + cmd_init(); + + /* Check for the application socket timeout env variable. */ + env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV); + if (env_app_timeout) { + app_socket_timeout = atoi(env_app_timeout); + } else { + app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT; + } + + write_pidfile(); - /* Init all health thread counters. */ - health_init(&health_thread_cmd); - health_init(&health_thread_kernel); - health_init(&health_thread_app_manage); - health_init(&health_thread_app_reg); + /* Initialize communication library */ + lttcomm_init(); + /* This is to get the TCP timeout value. */ + lttcomm_inet_init(); /* - * Init health counters of the consumer thread. We do a quick hack here to - * the state of the consumer health is fine even if the thread is not - * started. This is simply to ease our life and has no cost what so ever. + * Initialize the health check subsystem. This call should set the + * appropriate time values. */ - health_init(&kconsumer_data.health); - health_poll_update(&kconsumer_data.health); - health_init(&ustconsumer32_data.health); - health_poll_update(&ustconsumer32_data.health); - health_init(&ustconsumer64_data.health); - health_poll_update(&ustconsumer64_data.health); + health_sessiond = health_app_create(HEALTH_NUM_TYPE); + if (!health_sessiond) { + PERROR("health_app_create error"); + goto exit_health_sessiond_cleanup; + } + + /* Create thread to manage the client socket */ + ret = pthread_create(&ht_cleanup_thread, NULL, + thread_ht_cleanup, (void *) NULL); + if (ret != 0) { + PERROR("pthread_create ht_cleanup"); + goto exit_ht_cleanup; + } /* Create thread to manage the client socket */ ret = pthread_create(&health_thread, NULL, @@ -6193,27 +4776,46 @@ int main(int argc, char **argv) goto exit_apps; } - /* Create kernel thread to manage kernel event */ - ret = pthread_create(&kernel_thread, NULL, - thread_manage_kernel, (void *) NULL); + /* Create thread to manage application notify socket */ + ret = pthread_create(&apps_notify_thread, NULL, + ust_thread_manage_notify, (void *) NULL); if (ret != 0) { - PERROR("pthread_create kernel"); - goto exit_kernel; + PERROR("pthread_create apps"); + goto exit_apps_notify; + } + + /* Don't start this thread if kernel tracing is not requested nor root */ + if (is_root && !opt_no_kernel) { + /* Create kernel thread to manage kernel event */ + ret = pthread_create(&kernel_thread, NULL, + thread_manage_kernel, (void *) NULL); + if (ret != 0) { + PERROR("pthread_create kernel"); + goto exit_kernel; + } + + ret = pthread_join(kernel_thread, &status); + if (ret != 0) { + PERROR("pthread_join"); + goto error; /* join error, exit without cleanup */ + } } - ret = pthread_join(kernel_thread, &status); +exit_kernel: + ret = pthread_join(apps_notify_thread, &status); if (ret != 0) { - PERROR("pthread_join"); + PERROR("pthread_join apps notify"); goto error; /* join error, exit without cleanup */ } -exit_kernel: +exit_apps_notify: ret = pthread_join(apps_thread, &status); if (ret != 0) { - PERROR("pthread_join"); + PERROR("pthread_join apps"); goto error; /* join error, exit without cleanup */ } + exit_apps: ret = pthread_join(reg_apps_thread, &status); if (ret != 0) { @@ -6241,8 +4843,34 @@ exit_dispatch: goto error; /* join error, exit without cleanup */ } + ret = join_consumer_thread(&ustconsumer32_data); + if (ret != 0) { + PERROR("join_consumer ust32"); + goto error; /* join error, exit without cleanup */ + } + + ret = join_consumer_thread(&ustconsumer64_data); + if (ret != 0) { + PERROR("join_consumer ust64"); + goto error; /* join error, exit without cleanup */ + } + exit_client: + ret = pthread_join(health_thread, &status); + if (ret != 0) { + PERROR("pthread_join health thread"); + goto error; /* join error, exit without cleanup */ + } + exit_health: + ret = pthread_join(ht_cleanup_thread, &status); + if (ret != 0) { + PERROR("pthread_join ht cleanup thread"); + goto error; /* join error, exit without cleanup */ + } +exit_ht_cleanup: + health_app_destroy(health_sessiond); +exit_health_sessiond_cleanup: exit: /* * cleanup() is called when no other thread is running.