X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fmain.c;h=9b6f70cd08134acfa45afa1d970735d3d2ee6a4c;hp=2582e886d1e1a5f5352a6f91b8bd82b2ea914c7b;hb=287a512f2b868c16132b894a8143a5faf0379ca8;hpb=a433283e401f556b0eb87814475fc84b11fc9666 diff --git a/src/bin/lttng-sessiond/main.c b/src/bin/lttng-sessiond/main.c index 2582e886d..9b6f70cd0 100644 --- a/src/bin/lttng-sessiond/main.c +++ b/src/bin/lttng-sessiond/main.c @@ -17,7 +17,6 @@ * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. */ -#define _GNU_SOURCE #define _LGPL_SOURCE #include #include @@ -38,7 +37,7 @@ #include #include #include -#include +#include #include #include @@ -49,7 +48,7 @@ #include #include #include -#include +#include #include "lttng-sessiond.h" #include "buffer-registry.h" @@ -72,12 +71,24 @@ #include "agent-thread.h" #include "save.h" #include "load-session-thread.h" +#include "notification-thread.h" +#include "notification-thread-commands.h" #include "syscall.h" +#include "agent.h" +#include "ht-cleanup.h" #define CONSUMERD_FILE "lttng-consumerd" +static const char *help_msg = +#ifdef LTTNG_EMBED_HELP +#include +#else +NULL +#endif +; + const char *progname; -static const char *tracing_group_name = DEFAULT_TRACING_GROUP; +const char *tracing_group_name = DEFAULT_TRACING_GROUP; static int tracing_group_name_override; static char *opt_pidfile; static int opt_sig_parent; @@ -103,6 +114,7 @@ static struct consumer_data kconsumer_data = { .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH, .err_sock = -1, .cmd_sock = -1, + .channel_monitor_pipe = -1, .pid_mutex = PTHREAD_MUTEX_INITIALIZER, .lock = PTHREAD_MUTEX_INITIALIZER, .cond = PTHREAD_COND_INITIALIZER, @@ -114,6 +126,7 @@ static struct consumer_data ustconsumer64_data = { .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, .err_sock = -1, .cmd_sock = -1, + .channel_monitor_pipe = -1, .pid_mutex = PTHREAD_MUTEX_INITIALIZER, .lock = PTHREAD_MUTEX_INITIALIZER, .cond = PTHREAD_COND_INITIALIZER, @@ -125,6 +138,7 @@ static struct consumer_data ustconsumer32_data = { .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, .err_sock = -1, .cmd_sock = -1, + .channel_monitor_pipe = -1, .pid_mutex = PTHREAD_MUTEX_INITIALIZER, .lock = PTHREAD_MUTEX_INITIALIZER, .cond = PTHREAD_COND_INITIALIZER, @@ -190,7 +204,6 @@ static int kernel_poll_pipe[2] = { -1, -1 }; * for all threads when receiving an event on the pipe. */ static int thread_quit_pipe[2] = { -1, -1 }; -static int ht_cleanup_quit_pipe[2] = { -1, -1 }; /* * This pipe is used to inform the thread managing application communication @@ -211,6 +224,7 @@ static pthread_t health_thread; static pthread_t ht_cleanup_thread; static pthread_t agent_reg_thread; static pthread_t load_session_thread; +static pthread_t notification_thread; /* * UST registration command queue. This queue is tied with a futex and uses a N @@ -305,14 +319,26 @@ const char * const config_section_name = "sessiond"; /* Load session thread information to operate. */ struct load_session_thread_data *load_info; +/* Notification thread handle. */ +struct notification_thread_handle *notification_thread_handle; + +/* Global hash tables */ +struct lttng_ht *agent_apps_ht_by_sock = NULL; + /* - * Whether sessiond is ready for commands/health check requests. + * Whether sessiond is ready for commands/notification channel/health check + * requests. * NR_LTTNG_SESSIOND_READY must match the number of calls to * sessiond_notify_ready(). */ -#define NR_LTTNG_SESSIOND_READY 3 +#define NR_LTTNG_SESSIOND_READY 4 int lttng_sessiond_ready = NR_LTTNG_SESSIOND_READY; +int sessiond_check_thread_quit_pipe(int fd, uint32_t events) +{ + return (fd == thread_quit_pipe[0] && (events & LPOLLIN)) ? 1 : 0; +} + /* Notify parents that we are ready for cmd and health check */ LTTNG_HIDDEN void sessiond_notify_ready(void) @@ -419,47 +445,6 @@ int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size) return __sessiond_set_thread_pollset(events, size, thread_quit_pipe); } -/* - * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set. - */ -int sessiond_set_ht_cleanup_thread_pollset(struct lttng_poll_event *events, - size_t size) -{ - return __sessiond_set_thread_pollset(events, size, - ht_cleanup_quit_pipe); -} - -static -int __sessiond_check_thread_quit_pipe(int fd, uint32_t events, int a_pipe) -{ - if (fd == a_pipe && (events & LPOLLIN)) { - return 1; - } - return 0; -} - -/* - * Check if the thread quit pipe was triggered. - * - * Return 1 if it was triggered else 0; - */ -int sessiond_check_thread_quit_pipe(int fd, uint32_t events) -{ - return __sessiond_check_thread_quit_pipe(fd, events, - thread_quit_pipe[0]); -} - -/* - * Check if the ht_cleanup thread quit pipe was triggered. - * - * Return 1 if it was triggered else 0; - */ -int sessiond_check_ht_cleanup_quit(int fd, uint32_t events) -{ - return __sessiond_check_thread_quit_pipe(fd, events, - ht_cleanup_quit_pipe[0]); -} - /* * Init thread quit pipe. * @@ -492,11 +477,6 @@ static int init_thread_quit_pipe(void) return __init_thread_quit_pipe(thread_quit_pipe); } -static int init_ht_cleanup_quit_pipe(void) -{ - return __init_thread_quit_pipe(ht_cleanup_quit_pipe); -} - /* * Stop all threads by closing the thread quit pipe. */ @@ -559,6 +539,24 @@ static void close_consumer_sockets(void) PERROR("UST consumerd64 cmd_sock close"); } } + if (kconsumer_data.channel_monitor_pipe >= 0) { + ret = close(kconsumer_data.channel_monitor_pipe); + if (ret < 0) { + PERROR("kernel consumer channel monitor pipe close"); + } + } + if (ustconsumer32_data.channel_monitor_pipe >= 0) { + ret = close(ustconsumer32_data.channel_monitor_pipe); + if (ret < 0) { + PERROR("UST consumerd32 channel monitor pipe close"); + } + } + if (ustconsumer64_data.channel_monitor_pipe >= 0) { + ret = close(ustconsumer64_data.channel_monitor_pipe); + if (ret < 0) { + PERROR("UST consumerd64 channel monitor pipe close"); + } + } } /* @@ -582,6 +580,33 @@ static int generate_lock_file_path(char *path, size_t len) return ret; } +/* + * Wait on consumer process termination. + * + * Need to be called with the consumer data lock held or from a context + * ensuring no concurrent access to data (e.g: cleanup). + */ +static void wait_consumer(struct consumer_data *consumer_data) +{ + pid_t ret; + int status; + + if (consumer_data->pid <= 0) { + return; + } + + DBG("Waiting for complete teardown of consumerd (PID: %d)", + consumer_data->pid); + ret = waitpid(consumer_data->pid, &status, 0); + if (ret == -1) { + PERROR("consumerd waitpid pid: %d", consumer_data->pid) + } else if (!WIFEXITED(status)) { + ERR("consumerd termination with error: %d", + WEXITSTATUS(ret)); + } + consumer_data->pid = 0; +} + /* * Cleanup the session daemon's data structures. */ @@ -676,6 +701,10 @@ static void sessiond_cleanup(void) } } + wait_consumer(&kconsumer_data); + wait_consumer(&ustconsumer64_data); + wait_consumer(&ustconsumer32_data); + DBG("Cleaning up all agent apps"); agent_app_ht_clean(); @@ -765,11 +794,7 @@ static void sessiond_cleanup_options(void) free(kmod_probes_list); free(kmod_extra_probes_list); - /* */ - DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm" - "Matthew, BEET driven development works!%c[%dm", - 27, 1, 31, 27, 0, 27, 1, 33, 27, 0); - /* */ + run_as_destroy_worker(); } /* @@ -835,34 +860,57 @@ error: * right amount of memory and copying the original information from the lsm * structure. * - * Return total size of the buffer pointed by buf. + * Return 0 on success, negative value on error. */ -static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size) +static int setup_lttng_msg(struct command_ctx *cmd_ctx, + const void *payload_buf, size_t payload_len, + const void *cmd_header_buf, size_t cmd_header_len) { - int ret, buf_size; + int ret = 0; + const size_t header_len = sizeof(struct lttcomm_lttng_msg); + const size_t cmd_header_offset = header_len; + const size_t payload_offset = cmd_header_offset + cmd_header_len; + const size_t total_msg_size = header_len + cmd_header_len + payload_len; - buf_size = size; + cmd_ctx->llm = zmalloc(total_msg_size); - cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size); if (cmd_ctx->llm == NULL) { PERROR("zmalloc"); ret = -ENOMEM; - goto error; + goto end; } /* Copy common data */ cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type; cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid; + cmd_ctx->llm->cmd_header_size = cmd_header_len; + cmd_ctx->llm->data_size = payload_len; + cmd_ctx->lttng_msg_size = total_msg_size; - cmd_ctx->llm->data_size = size; - cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size; + /* Copy command header */ + if (cmd_header_len) { + memcpy(((uint8_t *) cmd_ctx->llm) + cmd_header_offset, cmd_header_buf, + cmd_header_len); + } - return buf_size; + /* Copy payload */ + if (payload_len) { + memcpy(((uint8_t *) cmd_ctx->llm) + payload_offset, payload_buf, + payload_len); + } -error: +end: return ret; } +/* + * Version of setup_lttng_msg() without command header. + */ +static int setup_lttng_msg_no_cmd_header(struct command_ctx *cmd_ctx, + void *payload_buf, size_t payload_len) +{ + return setup_lttng_msg(cmd_ctx, payload_buf, payload_len, NULL, 0); +} /* * Update the kernel poll set of all channel fd available over all tracing * session. Add the wakeup pipe at the end of the set. @@ -1129,31 +1177,33 @@ static void *thread_manage_kernel(void *data) } /* Check for data on kernel pipe */ - if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) { - (void) lttng_read(kernel_poll_pipe[0], - &tmp, 1); - /* - * Ret value is useless here, if this pipe gets any actions an - * update is required anyway. - */ - update_poll_flag = 1; - continue; - } else { - /* - * New CPU detected by the kernel. Adding kernel stream to - * kernel session and updating the kernel consumer - */ - if (revents & LPOLLIN) { + if (revents & LPOLLIN) { + if (pollfd == kernel_poll_pipe[0]) { + (void) lttng_read(kernel_poll_pipe[0], + &tmp, 1); + /* + * Ret value is useless here, if this pipe gets any actions an + * update is required anyway. + */ + update_poll_flag = 1; + continue; + } else { + /* + * New CPU detected by the kernel. Adding kernel stream to + * kernel session and updating the kernel consumer + */ ret = update_kernel_stream(&kconsumer_data, pollfd); if (ret < 0) { continue; } break; - /* - * TODO: We might want to handle the LPOLLERR | LPOLLHUP - * and unregister kernel stream at this point. - */ } + } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { + update_poll_flag = 1; + continue; + } else { + ERR("Unexpected poll events %u for sock %d", revents, pollfd); + goto error; } } } @@ -1209,9 +1259,13 @@ static void *thread_manage_consumer(void *data) enum lttcomm_return_code code; struct lttng_poll_event events; struct consumer_data *consumer_data = data; + struct consumer_socket *cmd_socket_wrapper = NULL; DBG("[thread] Manage consumer started"); + rcu_register_thread(); + rcu_thread_online(); + health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER); health_code_update(); @@ -1280,9 +1334,14 @@ restart: /* Event on the registration socket */ if (pollfd == consumer_data->err_sock) { - if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { + if (revents & LPOLLIN) { + continue; + } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { ERR("consumer err socket poll error"); goto error; + } else { + ERR("Unexpected poll events %u for sock %d", revents, pollfd); + goto error; } } } @@ -1310,40 +1369,43 @@ restart: } health_code_update(); - 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); - 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; - } - 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 { + if (code != LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) { ERR("consumer error when waiting for SOCK_READY : %s", lttcomm_get_readable_code(-code)); goto error; } - /* Remove the consumerd error sock since we've established a connexion */ + /* Connect both command and metadata sockets. */ + consumer_data->cmd_sock = + lttcomm_connect_unix_sock( + consumer_data->cmd_unix_sock_path); + 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"); + goto error; + } + pthread_mutex_init(consumer_data->metadata_sock.lock, NULL); + + DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock); + DBG("Consumer metadata socket ready (fd: %d)", + consumer_data->metadata_fd); + + /* + * Remove the consumerd error sock since we've established a connection. + */ ret = lttng_poll_del(&events, consumer_data->err_sock); if (ret < 0) { goto error; @@ -1364,6 +1426,27 @@ restart: health_code_update(); + /* + * Transfer the write-end of the channel monitoring pipe to the + * by issuing a SET_CHANNEL_MONITOR_PIPE command. + */ + cmd_socket_wrapper = consumer_allocate_socket(&consumer_data->cmd_sock); + if (!cmd_socket_wrapper) { + goto error; + } + + ret = consumer_send_channel_monitor_pipe(cmd_socket_wrapper, + consumer_data->channel_monitor_pipe); + if (ret) { + goto error; + } + /* Discard the socket wrapper as it is no longer needed. */ + consumer_destroy_socket(cmd_socket_wrapper); + cmd_socket_wrapper = NULL; + + /* The thread is completely initialized, signal that it is ready. */ + signal_consumer_condition(consumer_data, 1); + /* Infinite blocking call, waiting for transmission */ restart_poll: while (1) { @@ -1412,7 +1495,8 @@ restart_poll: if (pollfd == sock) { /* Event on the consumerd socket */ - if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { + if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP) + && !(revents & LPOLLIN)) { ERR("consumer err socket second poll error"); goto error; } @@ -1430,6 +1514,11 @@ restart_poll: goto exit; } else if (pollfd == consumer_data->metadata_fd) { + if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP) + && !(revents & LPOLLIN)) { + ERR("consumer err metadata socket second poll error"); + goto error; + } /* UST metadata requests */ ret = ust_consumer_metadata_request( &consumer_data->metadata_sock); @@ -1493,7 +1582,6 @@ 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. */ @@ -1502,6 +1590,10 @@ error: free(consumer_data->metadata_sock.lock); } lttng_poll_clean(&events); + + if (cmd_socket_wrapper) { + consumer_destroy_socket(cmd_socket_wrapper); + } error_poll: if (err) { health_error(); @@ -1510,6 +1602,9 @@ error_poll: health_unregister(health_sessiond); DBG("consumer thread cleanup completed"); + rcu_thread_offline(); + rcu_unregister_thread(); + return NULL; } @@ -1595,10 +1690,7 @@ static void *thread_manage_apps(void *data) /* Inspect the apps cmd pipe */ if (pollfd == apps_cmd_pipe[0]) { - if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { - ERR("Apps command pipe error"); - goto error; - } else if (revents & LPOLLIN) { + if (revents & LPOLLIN) { int sock; /* Empty pipe */ @@ -1611,9 +1703,8 @@ static void *thread_manage_apps(void *data) health_code_update(); /* - * We only monitor the error events of the socket. This - * thread does not handle any incoming data from UST - * (POLLIN). + * Since this is a command socket (write then read), + * we only monitor the error events of the socket. */ ret = lttng_poll_add(&events, sock, LPOLLERR | LPOLLHUP | LPOLLRDHUP); @@ -1622,6 +1713,12 @@ static void *thread_manage_apps(void *data) } DBG("Apps with sock %d added to poll set", sock); + } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { + ERR("Apps command pipe error"); + goto error; + } else { + ERR("Unknown poll events %u for sock %d", revents, pollfd); + goto error; } } else { /* @@ -1637,6 +1734,9 @@ static void *thread_manage_apps(void *data) /* Socket closed on remote end. */ ust_app_unregister(pollfd); + } else { + ERR("Unexpected poll events %u for sock %d", revents, pollfd); + goto error; } } @@ -1777,7 +1877,16 @@ static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue) wait_queue->count--; ust_app_destroy(wait_node->app); free(wait_node); + /* + * Silence warning of use-after-free in + * cds_list_for_each_entry_safe which uses + * __typeof__(*wait_node). + */ + wait_node = NULL; break; + } else { + ERR("Unexpected poll events %u for sock %d", revents, pollfd); + goto error; } } } @@ -1811,6 +1920,8 @@ static void *thread_dispatch_ust_registration(void *data) .count = 0, }; + rcu_register_thread(); + health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH); if (testpoint(sessiond_thread_app_reg_dispatch)) { @@ -1979,7 +2090,7 @@ static void *thread_dispatch_ust_registration(void *data) * Don't care about return value. Let the manage apps threads * handle app unregistration upon socket close. */ - (void) ust_app_register_done(app->sock); + (void) ust_app_register_done(app); /* * Even if the application socket has been closed, send the app @@ -2021,6 +2132,22 @@ error: free(wait_node); } + /* Empty command queue. */ + for (;;) { + /* Dequeue command for registration */ + node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail); + if (node == NULL) { + break; + } + ust_cmd = caa_container_of(node, struct ust_command, node); + ret = close(ust_cmd->sock); + if (ret < 0) { + PERROR("close ust sock exit dispatch %d", ust_cmd->sock); + } + lttng_fd_put(LTTNG_FD_APPS, 1); + free(ust_cmd); + } + error_testpoint: DBG("Dispatch thread dying"); if (err) { @@ -2028,6 +2155,7 @@ error_testpoint: ERR("Health error occurred in %s", __func__); } health_unregister(health_sessiond); + rcu_unregister_thread(); return NULL; } @@ -2122,10 +2250,7 @@ static void *thread_registration_apps(void *data) /* Event on the registration socket */ if (pollfd == apps_sock) { - if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { - ERR("Register apps socket poll error"); - goto error; - } else if (revents & LPOLLIN) { + if (revents & LPOLLIN) { sock = lttcomm_accept_unix_sock(apps_sock); if (sock < 0) { goto error; @@ -2138,10 +2263,12 @@ static void *thread_registration_apps(void *data) * lttcomm_setsockopt_snd_timeout expect msec as * parameter. */ - (void) lttcomm_setsockopt_rcv_timeout(sock, - app_socket_timeout * 1000); - (void) lttcomm_setsockopt_snd_timeout(sock, - app_socket_timeout * 1000); + if (app_socket_timeout >= 0) { + (void) lttcomm_setsockopt_rcv_timeout(sock, + app_socket_timeout * 1000); + (void) lttcomm_setsockopt_snd_timeout(sock, + app_socket_timeout * 1000); + } /* * Set the CLOEXEC flag. Return code is useless because @@ -2153,6 +2280,10 @@ static void *thread_registration_apps(void *data) ust_cmd = zmalloc(sizeof(struct ust_command)); if (ust_cmd == NULL) { PERROR("ust command zmalloc"); + ret = close(sock); + if (ret) { + PERROR("close"); + } goto error; } @@ -2208,6 +2339,12 @@ static void *thread_registration_apps(void *data) * barrier with the exchange in cds_wfcq_enqueue. */ futex_nto1_wake(&ust_cmd_queue.futex); + } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { + ERR("Register apps socket poll error"); + goto error; + } else { + ERR("Unexpected poll events %u for sock %d", revents, pollfd); + goto error; } } } @@ -2257,7 +2394,12 @@ static int spawn_consumer_thread(struct consumer_data *consumer_data) int ret, clock_ret; struct timespec timeout; - /* Make sure we set the readiness flag to 0 because we are NOT ready */ + /* + * Make sure we set the readiness flag to 0 because we are NOT ready. + * This access to consumer_thread_is_ready does not need to be + * protected by consumer_data.cond_mutex (yet) since the consumer + * management thread has not been started at this point. + */ consumer_data->consumer_thread_is_ready = 0; /* Setup pthread condition */ @@ -2287,8 +2429,8 @@ static int spawn_consumer_thread(struct consumer_data *consumer_data) goto error; } - ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer, - consumer_data); + ret = pthread_create(&consumer_data->thread, default_pthread_attr(), + thread_manage_consumer, consumer_data); if (ret) { errno = ret; PERROR("pthread_create consumer"); @@ -2300,7 +2442,7 @@ static int spawn_consumer_thread(struct consumer_data *consumer_data) pthread_mutex_lock(&consumer_data->cond_mutex); /* Get time for sem_timedwait absolute timeout */ - clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout); + clock_ret = lttng_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 @@ -2709,7 +2851,7 @@ static int copy_session_consumer(int domain, struct ltt_session *session) * domain. */ if (session->kernel_session->consumer) { - consumer_destroy_output(session->kernel_session->consumer); + consumer_output_put(session->kernel_session->consumer); } session->kernel_session->consumer = consumer_copy_output(session->consumer); @@ -2723,7 +2865,7 @@ static int copy_session_consumer(int domain, struct ltt_session *session) 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); + consumer_output_put(session->ust_session->consumer); } session->ust_session->consumer = consumer_copy_output(session->consumer); @@ -2843,7 +2985,7 @@ static int create_kernel_session(struct ltt_session *session) session->kernel_session->consumer->dst.trace_path, S_IRWXU | S_IRWXG, session->uid, session->gid); if (ret < 0) { - if (ret != -EEXIST) { + if (errno != EEXIST) { ERR("Trace directory creation error"); goto error; } @@ -2905,6 +3047,8 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, DBG("Processing client command %d", cmd_ctx->lsm->cmd_type); + assert(!rcu_read_ongoing()); + *sock_error = 0; switch (cmd_ctx->lsm->cmd_type) { @@ -2923,6 +3067,10 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, case LTTNG_SNAPSHOT_RECORD: case LTTNG_SAVE_SESSION: case LTTNG_SET_SESSION_SHM_PATH: + case LTTNG_REGENERATE_METADATA: + case LTTNG_REGENERATE_STATEDUMP: + case LTTNG_REGISTER_TRIGGER: + case LTTNG_UNREGISTER_TRIGGER: need_domain = 0; break; default: @@ -2964,10 +3112,11 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, case LTTNG_LIST_EVENTS: case LTTNG_LIST_SYSCALLS: case LTTNG_LIST_TRACKER_PIDS: + case LTTNG_DATA_PENDING: break; default: /* Setup lttng message with no payload */ - ret = setup_lttng_msg(cmd_ctx, 0); + ret = setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0); if (ret < 0) { /* This label does not try to unlock the session */ goto init_setup_error; @@ -2979,12 +3128,13 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, 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: case LTTNG_LIST_SYSCALLS: case LTTNG_LIST_TRACEPOINT_FIELDS: case LTTNG_SAVE_SESSION: + case LTTNG_REGISTER_TRIGGER: + case LTTNG_UNREGISTER_TRIGGER: need_tracing_session = 0; break; default: @@ -3200,6 +3350,8 @@ skip_domain: if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE || cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) { switch (cmd_ctx->lsm->domain.type) { + case LTTNG_DOMAIN_NONE: + break; case LTTNG_DOMAIN_JUL: case LTTNG_DOMAIN_LOG4J: case LTTNG_DOMAIN_PYTHON: @@ -3215,6 +3367,9 @@ skip_domain: goto error; } break; + default: + ret = LTTNG_ERR_UNKNOWN_DOMAIN; + goto error; } } @@ -3250,9 +3405,74 @@ skip_domain: switch (cmd_ctx->lsm->cmd_type) { case LTTNG_ADD_CONTEXT: { - ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type, + /* + * An LTTNG_ADD_CONTEXT command might have a supplementary + * payload if the context being added is an application context. + */ + if (cmd_ctx->lsm->u.context.ctx.ctx == + LTTNG_EVENT_CONTEXT_APP_CONTEXT) { + char *provider_name = NULL, *context_name = NULL; + size_t provider_name_len = + cmd_ctx->lsm->u.context.provider_name_len; + size_t context_name_len = + cmd_ctx->lsm->u.context.context_name_len; + + if (provider_name_len == 0 || context_name_len == 0) { + /* + * Application provider and context names MUST + * be provided. + */ + ret = -LTTNG_ERR_INVALID; + goto error; + } + + provider_name = zmalloc(provider_name_len + 1); + if (!provider_name) { + ret = -LTTNG_ERR_NOMEM; + goto error; + } + cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name = + provider_name; + + context_name = zmalloc(context_name_len + 1); + if (!context_name) { + ret = -LTTNG_ERR_NOMEM; + goto error_add_context; + } + cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name = + context_name; + + ret = lttcomm_recv_unix_sock(sock, provider_name, + provider_name_len); + if (ret < 0) { + goto error_add_context; + } + + ret = lttcomm_recv_unix_sock(sock, context_name, + context_name_len); + if (ret < 0) { + goto error_add_context; + } + } + + /* + * cmd_add_context assumes ownership of the provider and context + * names. + */ + ret = cmd_add_context(cmd_ctx->session, + cmd_ctx->lsm->domain.type, cmd_ctx->lsm->u.context.channel_name, - &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]); + &cmd_ctx->lsm->u.context.ctx, + kernel_poll_pipe[1]); + + cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name = NULL; + cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name = NULL; +error_add_context: + free(cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name); + free(cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name); + if (ret < 0) { + goto error; + } break; } case LTTNG_DISABLE_CHANNEL: @@ -3263,8 +3483,34 @@ skip_domain: } case LTTNG_DISABLE_EVENT: { + + /* + * FIXME: handle filter; for now we just receive the filter's + * bytecode along with the filter expression which are sent by + * liblttng-ctl and discard them. + * + * This fixes an issue where the client may block while sending + * the filter payload and encounter an error because the session + * daemon closes the socket without ever handling this data. + */ + size_t count = cmd_ctx->lsm->u.disable.expression_len + + cmd_ctx->lsm->u.disable.bytecode_len; + + if (count) { + char data[LTTNG_FILTER_MAX_LEN]; + + DBG("Discarding disable event command payload of size %zu", count); + while (count) { + ret = lttcomm_recv_unix_sock(sock, data, + count > sizeof(data) ? sizeof(data) : count); + if (ret < 0) { + goto error; + } + + count -= (size_t) ret; + } + } /* FIXME: passing packed structure to non-packed pointer */ - /* TODO: handle filter */ ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type, cmd_ctx->lsm->u.disable.channel_name, &cmd_ctx->lsm->u.disable.event); @@ -3272,8 +3518,11 @@ skip_domain: } case LTTNG_ENABLE_CHANNEL: { + cmd_ctx->lsm->u.channel.chan.attr.extended.ptr = + (struct lttng_channel_extended *) &cmd_ctx->lsm->u.channel.extended; ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain, - &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]); + &cmd_ctx->lsm->u.channel.chan, + kernel_poll_pipe[1]); break; } case LTTNG_TRACK_PID: @@ -3418,18 +3667,14 @@ skip_domain: * Setup lttng message with payload size set to the event list size in * bytes and then copy list into the llm payload. */ - ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events); + ret = setup_lttng_msg_no_cmd_header(cmd_ctx, events, + sizeof(struct lttng_event) * nb_events); + free(events); + if (ret < 0) { - free(events); goto setup_error; } - /* Copy event list into message payload */ - memcpy(cmd_ctx->llm->payload, events, - sizeof(struct lttng_event) * nb_events); - - free(events); - ret = LTTNG_OK; break; } @@ -3452,19 +3697,14 @@ skip_domain: * Setup lttng message with payload size set to the event list size in * bytes and then copy list into the llm payload. */ - ret = setup_lttng_msg(cmd_ctx, + ret = setup_lttng_msg_no_cmd_header(cmd_ctx, fields, sizeof(struct lttng_event_field) * nb_fields); + free(fields); + if (ret < 0) { - free(fields); goto setup_error; } - /* Copy event list into message payload */ - memcpy(cmd_ctx->llm->payload, fields, - sizeof(struct lttng_event_field) * nb_fields); - - free(fields); - ret = LTTNG_OK; break; } @@ -3484,18 +3724,14 @@ skip_domain: * Setup lttng message with payload size set to the event list size in * bytes and then copy list into the llm payload. */ - ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events); + ret = setup_lttng_msg_no_cmd_header(cmd_ctx, events, + sizeof(struct lttng_event) * nb_events); + free(events); + if (ret < 0) { - free(events); goto setup_error; } - /* Copy event list into message payload */ - memcpy(cmd_ctx->llm->payload, events, - sizeof(struct lttng_event) * nb_events); - - free(events); - ret = LTTNG_OK; break; } @@ -3516,18 +3752,14 @@ skip_domain: * Setup lttng message with payload size set to the event list size in * bytes and then copy list into the llm payload. */ - ret = setup_lttng_msg(cmd_ctx, sizeof(int32_t) * nr_pids); + ret = setup_lttng_msg_no_cmd_header(cmd_ctx, pids, + sizeof(int32_t) * nr_pids); + free(pids); + if (ret < 0) { - free(pids); goto setup_error; } - /* Copy event list into message payload */ - memcpy(cmd_ctx->llm->payload, pids, - sizeof(int) * nr_pids); - - free(pids); - ret = LTTNG_OK; break; } @@ -3641,46 +3873,38 @@ skip_domain: goto error; } - ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain)); + ret = setup_lttng_msg_no_cmd_header(cmd_ctx, domains, + nb_dom * sizeof(struct lttng_domain)); + free(domains); + if (ret < 0) { - free(domains); goto setup_error; } - /* Copy event list into message payload */ - memcpy(cmd_ctx->llm->payload, domains, - nb_dom * sizeof(struct lttng_domain)); - - free(domains); - ret = LTTNG_OK; break; } case LTTNG_LIST_CHANNELS: { - int nb_chan; + ssize_t payload_size; struct lttng_channel *channels = NULL; - nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type, + payload_size = cmd_list_channels(cmd_ctx->lsm->domain.type, cmd_ctx->session, &channels); - if (nb_chan < 0) { + if (payload_size < 0) { /* Return value is a negative lttng_error_code. */ - ret = -nb_chan; + ret = -payload_size; goto error; } - ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel)); + ret = setup_lttng_msg_no_cmd_header(cmd_ctx, channels, + payload_size); + free(channels); + if (ret < 0) { - free(channels); goto setup_error; } - /* Copy event list into message payload */ - memcpy(cmd_ctx->llm->payload, channels, - nb_chan * sizeof(struct lttng_channel)); - - free(channels); - ret = LTTNG_OK; break; } @@ -3688,61 +3912,68 @@ skip_domain: { ssize_t nb_event; struct lttng_event *events = NULL; + struct lttcomm_event_command_header cmd_header; + size_t total_size; + + memset(&cmd_header, 0, sizeof(cmd_header)); + /* Extended infos are included at the end of events */ + nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, + cmd_ctx->session, cmd_ctx->lsm->u.list.channel_name, + &events, &total_size); - 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)); + cmd_header.nb_events = nb_event; + ret = setup_lttng_msg(cmd_ctx, events, total_size, + &cmd_header, sizeof(cmd_header)); + free(events); + if (ret < 0) { - free(events); goto setup_error; } - /* Copy event list into message payload */ - memcpy(cmd_ctx->llm->payload, events, - nb_event * sizeof(struct lttng_event)); - - free(events); - ret = LTTNG_OK; break; } case LTTNG_LIST_SESSIONS: { unsigned int nr_sessions; + void *sessions_payload; + size_t payload_len; session_lock_list(); nr_sessions = lttng_sessions_count( LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds), LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds)); + payload_len = sizeof(struct lttng_session) * nr_sessions; + sessions_payload = zmalloc(payload_len); - ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions); - if (ret < 0) { + if (!sessions_payload) { session_unlock_list(); + ret = -ENOMEM; goto setup_error; } - /* Filled the session array */ - cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload), + cmd_list_lttng_sessions(sessions_payload, LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds), LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds)); - session_unlock_list(); + ret = setup_lttng_msg_no_cmd_header(cmd_ctx, sessions_payload, + payload_len); + free(sessions_payload); + + if (ret < 0) { + goto setup_error; + } + ret = LTTNG_OK; break; } - case LTTNG_CALIBRATE: - { - ret = cmd_calibrate(cmd_ctx->lsm->domain.type, - &cmd_ctx->lsm->u.calibrate); - break; - } case LTTNG_REGISTER_CONSUMER: { struct consumer_data *cdata; @@ -3762,7 +3993,44 @@ skip_domain: } case LTTNG_DATA_PENDING: { - ret = cmd_data_pending(cmd_ctx->session); + int pending_ret; + uint8_t pending_ret_byte; + + pending_ret = cmd_data_pending(cmd_ctx->session); + + /* + * FIXME + * + * This function may returns 0 or 1 to indicate whether or not + * there is data pending. In case of error, it should return an + * LTTNG_ERR code. However, some code paths may still return + * a nondescript error code, which we handle by returning an + * "unknown" error. + */ + if (pending_ret == 0 || pending_ret == 1) { + /* + * ret will be set to LTTNG_OK at the end of + * this function. + */ + } else if (pending_ret < 0) { + ret = LTTNG_ERR_UNK; + goto setup_error; + } else { + ret = pending_ret; + goto setup_error; + } + + pending_ret_byte = (uint8_t) pending_ret; + + /* 1 byte to return whether or not data is pending */ + ret = setup_lttng_msg_no_cmd_header(cmd_ctx, + &pending_ret_byte, 1); + + if (ret < 0) { + goto setup_error; + } + + ret = LTTNG_OK; break; } case LTTNG_SNAPSHOT_ADD_OUTPUT: @@ -3775,13 +4043,13 @@ skip_domain: goto error; } - ret = setup_lttng_msg(cmd_ctx, sizeof(reply)); + ret = setup_lttng_msg_no_cmd_header(cmd_ctx, &reply, + 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; } @@ -3802,20 +4070,15 @@ skip_domain: goto error; } - ret = setup_lttng_msg(cmd_ctx, + assert((nb_output > 0 && outputs) || nb_output == 0); + ret = setup_lttng_msg_no_cmd_header(cmd_ctx, outputs, nb_output * sizeof(struct lttng_snapshot_output)); + free(outputs); + 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; } @@ -3916,6 +4179,28 @@ skip_domain: cmd_ctx->lsm->u.set_shm_path.shm_path); break; } + case LTTNG_REGENERATE_METADATA: + { + ret = cmd_regenerate_metadata(cmd_ctx->session); + break; + } + case LTTNG_REGENERATE_STATEDUMP: + { + ret = cmd_regenerate_statedump(cmd_ctx->session); + break; + } + case LTTNG_REGISTER_TRIGGER: + { + ret = cmd_register_trigger(cmd_ctx, sock, + notification_thread_handle); + break; + } + case LTTNG_UNREGISTER_TRIGGER: + { + ret = cmd_unregister_trigger(cmd_ctx, sock, + notification_thread_handle); + break; + } default: ret = LTTNG_ERR_UND; break; @@ -3924,7 +4209,7 @@ skip_domain: error: if (cmd_ctx->llm == NULL) { DBG("Missing llm structure. Allocating one."); - if (setup_lttng_msg(cmd_ctx, 0) < 0) { + if (setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0) < 0) { goto setup_error; } } @@ -3938,6 +4223,7 @@ setup_error: session_unlock_list(); } init_setup_error: + assert(!rcu_read_ongoing()); return ret; } @@ -3963,7 +4249,6 @@ static void *thread_manage_health(void *data) sock = lttcomm_create_unix_sock(health_unix_sock_path); if (sock < 0) { ERR("Unable to create health check Unix socket"); - ret = -1; goto error; } @@ -3974,7 +4259,6 @@ static void *thread_manage_health(void *data) if (ret < 0) { ERR("Unable to set group on %s", health_unix_sock_path); PERROR("chown"); - ret = -1; goto error; } @@ -3983,7 +4267,6 @@ static void *thread_manage_health(void *data) if (ret < 0) { ERR("Unable to set permissions on %s", health_unix_sock_path); PERROR("chmod"); - ret = -1; goto error; } } @@ -4053,9 +4336,14 @@ restart: /* Event on the registration socket */ if (pollfd == sock) { - if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { + if (revents & LPOLLIN) { + continue; + } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { ERR("Health socket poll error"); goto error; + } else { + ERR("Unexpected poll events %u for sock %d", revents, pollfd); + goto error; } } } @@ -4079,7 +4367,6 @@ restart: if (ret) { PERROR("close"); } - new_sock = -1; continue; } @@ -4108,7 +4395,6 @@ restart: if (ret) { PERROR("close"); } - new_sock = -1; } exit: @@ -4126,7 +4412,7 @@ error: } lttng_poll_clean(&events); - + stop_threads(); rcu_unregister_thread(); return NULL; } @@ -4230,9 +4516,14 @@ static void *thread_manage_clients(void *data) /* Event on the registration socket */ if (pollfd == client_sock) { - if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { + if (revents & LPOLLIN) { + continue; + } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { ERR("Client socket poll error"); goto error; + } else { + ERR("Unexpected poll events %u for sock %d", revents, pollfd); + goto error; } } } @@ -4329,9 +4620,10 @@ static void *thread_manage_clients(void *data) health_code_update(); - DBG("Sending response (size: %d, retcode: %s)", + DBG("Sending response (size: %d, retcode: %s (%d))", cmd_ctx->lttng_msg_size, - lttng_strerror(-cmd_ctx->llm->ret_code)); + lttng_strerror(-cmd_ctx->llm->ret_code), + cmd_ctx->llm->ret_code); ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size); if (ret < 0) { ERR("Failed to send data back to client"); @@ -4406,43 +4698,6 @@ error_create_poll: return NULL; } - -/* - * usage function on stderr - */ -static void usage(void) -{ - fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname); - fprintf(stderr, " -h, --help Display this usage.\n"); - fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n"); - fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n"); - fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n"); - fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n"); - fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n"); - fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n"); - fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n"); - fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n"); - fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n"); - fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n"); - fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n"); - fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n"); - fprintf(stderr, " -d, --daemonize Start as a daemon.\n"); - fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n"); - fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n"); - fprintf(stderr, " -V, --version Show version number.\n"); - fprintf(stderr, " -S, --sig-parent Send SIGUSR1 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"); - fprintf(stderr, " --agent-tcp-port Agent registration TCP port\n"); - fprintf(stderr, " -f --config PATH Load daemon configuration file\n"); - fprintf(stderr, " -l --load PATH Load session configuration\n"); - fprintf(stderr, " --kmod-probes Specify kernel module probes to load\n"); - fprintf(stderr, " --extra-kmod-probes Specify extra kernel module probes to load\n"); -} - static int string_match(const char *str1, const char *str2) { return (str1 && str2) && !strcmp(str1, str2); @@ -4458,18 +4713,11 @@ static int set_option(int opt, const char *arg, const char *optname) { int ret = 0; - if (arg && arg[0] == '\0') { - /* - * This only happens if the value is read from daemon config - * file. This means the option requires an argument and the - * configuration file contains a line such as: - * my_option = - */ - ret = -EINVAL; - goto end; - } - if (string_match(optname, "client-sock") || opt == 'c') { + if (!arg || *arg == '\0') { + ret = -EINVAL; + goto end; + } if (lttng_is_setuid_setgid()) { WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", "-c, --client-sock"); @@ -4477,6 +4725,10 @@ static int set_option(int opt, const char *arg, const char *optname) snprintf(client_unix_sock_path, PATH_MAX, "%s", arg); } } else if (string_match(optname, "apps-sock") || opt == 'a') { + if (!arg || *arg == '\0') { + ret = -EINVAL; + goto end; + } if (lttng_is_setuid_setgid()) { WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", "-a, --apps-sock"); @@ -4488,6 +4740,10 @@ static int set_option(int opt, const char *arg, const char *optname) } else if (string_match(optname, "background") || opt == 'b') { opt_background = 1; } else if (string_match(optname, "group") || opt == 'g') { + if (!arg || *arg == '\0') { + ret = -EINVAL; + goto end; + } if (lttng_is_setuid_setgid()) { WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", "-g, --group"); @@ -4508,14 +4764,22 @@ static int set_option(int opt, const char *arg, const char *optname) tracing_group_name_override = 1; } } else if (string_match(optname, "help") || opt == 'h') { - usage(); - exit(EXIT_SUCCESS); + ret = utils_show_help(8, "lttng-sessiond", help_msg); + if (ret) { + ERR("Cannot show --help for `lttng-sessiond`"); + perror("exec"); + } + exit(ret ? EXIT_FAILURE : EXIT_SUCCESS); } else if (string_match(optname, "version") || opt == 'V') { fprintf(stdout, "%s\n", VERSION); exit(EXIT_SUCCESS); } else if (string_match(optname, "sig-parent") || opt == 'S') { opt_sig_parent = 1; } else if (string_match(optname, "kconsumerd-err-sock")) { + if (!arg || *arg == '\0') { + ret = -EINVAL; + goto end; + } if (lttng_is_setuid_setgid()) { WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", "--kconsumerd-err-sock"); @@ -4523,6 +4787,10 @@ static int set_option(int opt, const char *arg, const char *optname) snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", arg); } } else if (string_match(optname, "kconsumerd-cmd-sock")) { + if (!arg || *arg == '\0') { + ret = -EINVAL; + goto end; + } if (lttng_is_setuid_setgid()) { WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", "--kconsumerd-cmd-sock"); @@ -4530,6 +4798,10 @@ static int set_option(int opt, const char *arg, const char *optname) snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", arg); } } else if (string_match(optname, "ustconsumerd64-err-sock")) { + if (!arg || *arg == '\0') { + ret = -EINVAL; + goto end; + } if (lttng_is_setuid_setgid()) { WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", "--ustconsumerd64-err-sock"); @@ -4537,6 +4809,10 @@ static int set_option(int opt, const char *arg, const char *optname) snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", arg); } } else if (string_match(optname, "ustconsumerd64-cmd-sock")) { + if (!arg || *arg == '\0') { + ret = -EINVAL; + goto end; + } if (lttng_is_setuid_setgid()) { WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", "--ustconsumerd64-cmd-sock"); @@ -4544,6 +4820,10 @@ static int set_option(int opt, const char *arg, const char *optname) snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", arg); } } else if (string_match(optname, "ustconsumerd32-err-sock")) { + if (!arg || *arg == '\0') { + ret = -EINVAL; + goto end; + } if (lttng_is_setuid_setgid()) { WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", "--ustconsumerd32-err-sock"); @@ -4551,6 +4831,10 @@ static int set_option(int opt, const char *arg, const char *optname) snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", arg); } } else if (string_match(optname, "ustconsumerd32-cmd-sock")) { + if (!arg || *arg == '\0') { + ret = -EINVAL; + goto end; + } if (lttng_is_setuid_setgid()) { WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", "--ustconsumerd32-cmd-sock"); @@ -4577,9 +4861,13 @@ static int set_option(int opt, const char *arg, const char *optname) if (arg) { opt_verbose_consumer = config_parse_value(arg); } else { - opt_verbose_consumer += 1; + opt_verbose_consumer++; } } else if (string_match(optname, "consumerd32-path")) { + if (!arg || *arg == '\0') { + ret = -EINVAL; + goto end; + } if (lttng_is_setuid_setgid()) { WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", "--consumerd32-path"); @@ -4595,6 +4883,10 @@ static int set_option(int opt, const char *arg, const char *optname) consumerd32_bin_override = 1; } } else if (string_match(optname, "consumerd32-libdir")) { + if (!arg || *arg == '\0') { + ret = -EINVAL; + goto end; + } if (lttng_is_setuid_setgid()) { WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", "--consumerd32-libdir"); @@ -4610,6 +4902,10 @@ static int set_option(int opt, const char *arg, const char *optname) consumerd32_libdir_override = 1; } } else if (string_match(optname, "consumerd64-path")) { + if (!arg || *arg == '\0') { + ret = -EINVAL; + goto end; + } if (lttng_is_setuid_setgid()) { WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", "--consumerd64-path"); @@ -4625,6 +4921,10 @@ static int set_option(int opt, const char *arg, const char *optname) consumerd64_bin_override = 1; } } else if (string_match(optname, "consumerd64-libdir")) { + if (!arg || *arg == '\0') { + ret = -EINVAL; + goto end; + } if (lttng_is_setuid_setgid()) { WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", "--consumerd64-libdir"); @@ -4640,6 +4940,10 @@ static int set_option(int opt, const char *arg, const char *optname) consumerd64_libdir_override = 1; } } else if (string_match(optname, "pidfile") || opt == 'p') { + if (!arg || *arg == '\0') { + ret = -EINVAL; + goto end; + } if (lttng_is_setuid_setgid()) { WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", "-p, --pidfile"); @@ -4652,16 +4956,16 @@ static int set_option(int opt, const char *arg, const char *optname) } } } else if (string_match(optname, "agent-tcp-port")) { + if (!arg || *arg == '\0') { + ret = -EINVAL; + goto end; + } if (lttng_is_setuid_setgid()) { WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", "--agent-tcp-port"); } else { unsigned long v; - if (!arg) { - ret = -EINVAL; - goto end; - } errno = 0; v = strtoul(arg, NULL, 0); if (errno != 0 || !isdigit(arg[0])) { @@ -4676,6 +4980,10 @@ static int set_option(int opt, const char *arg, const char *optname) DBG3("Agent TCP port set to non default: %u", agent_tcp_port); } } else if (string_match(optname, "load") || opt == 'l') { + if (!arg || *arg == '\0') { + ret = -EINVAL; + goto end; + } if (lttng_is_setuid_setgid()) { WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", "-l, --load"); @@ -4688,6 +4996,10 @@ static int set_option(int opt, const char *arg, const char *optname) } } } else if (string_match(optname, "kmod-probes")) { + if (!arg || *arg == '\0') { + ret = -EINVAL; + goto end; + } if (lttng_is_setuid_setgid()) { WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", "--kmod-probes"); @@ -4700,6 +5012,10 @@ static int set_option(int opt, const char *arg, const char *optname) } } } else if (string_match(optname, "extra-kmod-probes")) { + if (!arg || *arg == '\0') { + ret = -EINVAL; + goto end; + } if (lttng_is_setuid_setgid()) { WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", "--extra-kmod-probes"); @@ -4742,7 +5058,7 @@ end: /* * config_entry_handler_cb used to handle options read from a config file. - * See config_entry_handler_cb comment in common/config/config.h for the + * See config_entry_handler_cb comment in common/config/session-config.h for the * return value conventions. */ static int config_entry_handler(const struct config_entry *entry, void *unused) @@ -5134,9 +5450,6 @@ error: static void sighandler(int sig) { switch (sig) { - case SIGPIPE: - DBG("SIGPIPE caught"); - return; case SIGINT: DBG("SIGINT caught"); stop_threads(); @@ -5168,9 +5481,10 @@ static int set_signal_handler(void) return ret; } - sa.sa_handler = sighandler; sa.sa_mask = sigset; sa.sa_flags = 0; + + sa.sa_handler = sighandler; if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) { PERROR("sigaction"); return ret; @@ -5181,12 +5495,13 @@ static int set_signal_handler(void) return ret; } - if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) { + if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) { PERROR("sigaction"); return ret; } - if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) { + sa.sa_handler = SIG_IGN; + if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) { PERROR("sigaction"); return ret; } @@ -5198,14 +5513,14 @@ static int set_signal_handler(void) /* * Set open files limit to unlimited. This daemon can open a large number of - * file descriptors in order to consumer multiple kernel traces. + * file descriptors in order to consume multiple kernel traces. */ static void set_ulimit(void) { int ret; struct rlimit lim; - /* The kernel does not allowed an infinite limit for open files */ + /* The kernel does not allow an infinite limit for open files */ lim.rlim_cur = 65535; lim.rlim_max = 65535; @@ -5226,7 +5541,10 @@ static int write_pidfile(void) assert(rundir); if (opt_pidfile) { - strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path)); + if (lttng_strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path))) { + ret = -1; + goto error; + } } else { /* Build pidfile path from rundir and opt_pidfile. */ ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/" @@ -5297,9 +5615,10 @@ int main(int argc, char **argv) int ret = 0, retval = 0; void *status; const char *home_path, *env_app_timeout; - - /* Initialize agent apps ht global variable */ - agent_apps_ht_by_sock = NULL; + struct lttng_pipe *ust32_channel_monitor_pipe = NULL, + *ust64_channel_monitor_pipe = NULL, + *kernel_channel_monitor_pipe = NULL; + bool notification_thread_running = false; init_kernel_workarounds(); @@ -5354,6 +5673,10 @@ int main(int argc, char **argv) } } + if (run_as_create_worker(argv[0]) < 0) { + goto exit_create_run_as_worker_cleanup; + } + /* * Starting from here, we can create threads. This needs to be after * lttng_daemonize due to RCU. @@ -5370,29 +5693,8 @@ int main(int argc, char **argv) goto exit_health_sessiond_cleanup; } - if (init_ht_cleanup_quit_pipe()) { - retval = -1; - goto exit_ht_cleanup_quit_pipe; - } - - /* Setup the thread ht_cleanup communication pipe. */ - if (utils_create_pipe_cloexec(ht_cleanup_pipe)) { - retval = -1; - goto exit_ht_cleanup_pipe; - } - - /* Set up max poll set size */ - if (lttng_poll_set_max_size()) { - retval = -1; - goto exit_set_max_size; - } - /* Create thread to clean up RCU hash tables */ - ret = pthread_create(&ht_cleanup_thread, NULL, - thread_ht_cleanup, (void *) NULL); - if (ret) { - errno = ret; - PERROR("pthread_create ht_cleanup"); + if (init_ht_cleanup_thread(&ht_cleanup_thread)) { retval = -1; goto exit_ht_cleanup; } @@ -5475,6 +5777,19 @@ int main(int argc, char **argv) kconsumer_data.err_unix_sock_path); DBG2("Kernel consumer cmd path: %s", kconsumer_data.cmd_unix_sock_path); + kernel_channel_monitor_pipe = lttng_pipe_open(0); + if (!kernel_channel_monitor_pipe) { + ERR("Failed to create kernel consumer channel monitor pipe"); + retval = -1; + goto exit_init_data; + } + kconsumer_data.channel_monitor_pipe = + lttng_pipe_release_writefd( + kernel_channel_monitor_pipe); + if (kconsumer_data.channel_monitor_pipe < 0) { + retval = -1; + goto exit_init_data; + } } else { home_path = utils_get_home_dir(); if (home_path == NULL) { @@ -5579,6 +5894,18 @@ int main(int argc, char **argv) ustconsumer32_data.err_unix_sock_path); DBG2("UST consumer 32 bits cmd path: %s", ustconsumer32_data.cmd_unix_sock_path); + ust32_channel_monitor_pipe = lttng_pipe_open(0); + if (!ust32_channel_monitor_pipe) { + ERR("Failed to create 32-bit user space consumer channel monitor pipe"); + retval = -1; + goto exit_init_data; + } + ustconsumer32_data.channel_monitor_pipe = lttng_pipe_release_writefd( + ust32_channel_monitor_pipe); + if (ustconsumer32_data.channel_monitor_pipe < 0) { + retval = -1; + goto exit_init_data; + } /* 64 bits consumerd path setup */ ret = snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, @@ -5600,6 +5927,18 @@ int main(int argc, char **argv) ustconsumer64_data.err_unix_sock_path); DBG2("UST consumer 64 bits cmd path: %s", ustconsumer64_data.cmd_unix_sock_path); + ust64_channel_monitor_pipe = lttng_pipe_open(0); + if (!ust64_channel_monitor_pipe) { + ERR("Failed to create 64-bit user space consumer channel monitor pipe"); + retval = -1; + goto exit_init_data; + } + ustconsumer64_data.channel_monitor_pipe = lttng_pipe_release_writefd( + ust64_channel_monitor_pipe); + if (ustconsumer64_data.channel_monitor_pipe < 0) { + retval = -1; + goto exit_init_data; + } /* * See if daemon already exist. @@ -5759,8 +6098,8 @@ int main(int argc, char **argv) } load_info->path = opt_load_session_path; - /* Create health-check thread */ - ret = pthread_create(&health_thread, NULL, + /* Create health-check thread. */ + ret = pthread_create(&health_thread, default_pthread_attr(), thread_manage_health, (void *) NULL); if (ret) { errno = ret; @@ -5769,86 +6108,118 @@ int main(int argc, char **argv) goto exit_health; } + /* notification_thread_data acquires the pipes' read side. */ + notification_thread_handle = notification_thread_handle_create( + ust32_channel_monitor_pipe, + ust64_channel_monitor_pipe, + kernel_channel_monitor_pipe); + if (!notification_thread_handle) { + retval = -1; + ERR("Failed to create notification thread shared data"); + stop_threads(); + goto exit_notification; + } + + /* Create notification thread. */ + ret = pthread_create(¬ification_thread, default_pthread_attr(), + thread_notification, notification_thread_handle); + if (ret) { + errno = ret; + PERROR("pthread_create notification"); + retval = -1; + stop_threads(); + goto exit_notification; + } + notification_thread_running = true; + /* Create thread to manage the client socket */ - ret = pthread_create(&client_thread, NULL, + ret = pthread_create(&client_thread, default_pthread_attr(), thread_manage_clients, (void *) NULL); if (ret) { errno = ret; PERROR("pthread_create clients"); retval = -1; + stop_threads(); goto exit_client; } /* Create thread to dispatch registration */ - ret = pthread_create(&dispatch_thread, NULL, + ret = pthread_create(&dispatch_thread, default_pthread_attr(), thread_dispatch_ust_registration, (void *) NULL); if (ret) { errno = ret; PERROR("pthread_create dispatch"); retval = -1; + stop_threads(); goto exit_dispatch; } /* Create thread to manage application registration. */ - ret = pthread_create(®_apps_thread, NULL, + ret = pthread_create(®_apps_thread, default_pthread_attr(), thread_registration_apps, (void *) NULL); if (ret) { errno = ret; PERROR("pthread_create registration"); retval = -1; + stop_threads(); goto exit_reg_apps; } /* Create thread to manage application socket */ - ret = pthread_create(&apps_thread, NULL, + ret = pthread_create(&apps_thread, default_pthread_attr(), thread_manage_apps, (void *) NULL); if (ret) { errno = ret; PERROR("pthread_create apps"); retval = -1; + stop_threads(); goto exit_apps; } /* Create thread to manage application notify socket */ - ret = pthread_create(&apps_notify_thread, NULL, + ret = pthread_create(&apps_notify_thread, default_pthread_attr(), ust_thread_manage_notify, (void *) NULL); if (ret) { errno = ret; PERROR("pthread_create notify"); retval = -1; + stop_threads(); goto exit_apps_notify; } /* Create agent registration thread. */ - ret = pthread_create(&agent_reg_thread, NULL, + ret = pthread_create(&agent_reg_thread, default_pthread_attr(), agent_thread_manage_registration, (void *) NULL); if (ret) { errno = ret; PERROR("pthread_create agent"); retval = -1; + stop_threads(); goto exit_agent_reg; } /* 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, + ret = pthread_create(&kernel_thread, default_pthread_attr(), thread_manage_kernel, (void *) NULL); if (ret) { errno = ret; PERROR("pthread_create kernel"); retval = -1; + stop_threads(); goto exit_kernel; } } /* Create session loading thread. */ - ret = pthread_create(&load_session_thread, NULL, thread_load_session, - load_info); + ret = pthread_create(&load_session_thread, default_pthread_attr(), + thread_load_session, load_info); if (ret) { errno = ret; PERROR("pthread_create load_session_thread"); retval = -1; + stop_threads(); goto exit_load_session; } @@ -5907,6 +6278,10 @@ exit_apps: } exit_reg_apps: + /* + * Join dispatch thread after joining reg_apps_thread to ensure + * we don't leak applications in the queue. + */ ret = pthread_join(dispatch_thread, &status); if (ret) { errno = ret; @@ -5921,52 +6296,72 @@ exit_dispatch: PERROR("pthread_join"); retval = -1; } -exit_client: +exit_client: +exit_notification: ret = pthread_join(health_thread, &status); if (ret) { errno = ret; PERROR("pthread_join health thread"); retval = -1; } -exit_health: +exit_health: exit_init_data: + /* + * Wait for all pending call_rcu work to complete before tearing + * down data structures. call_rcu worker may be trying to + * perform lookups in those structures. + */ + rcu_barrier(); /* * sessiond_cleanup() is called when no other thread is running, except * the ht_cleanup thread, which is needed to destroy the hash tables. */ rcu_thread_online(); sessiond_cleanup(); - rcu_thread_offline(); - rcu_unregister_thread(); - ret = notify_thread_pipe(ht_cleanup_quit_pipe[1]); - if (ret < 0) { - ERR("write error on ht_cleanup quit pipe"); - retval = -1; + /* + * Ensure all prior call_rcu are done. call_rcu callbacks may push + * hash tables to the ht_cleanup thread. Therefore, we ensure that + * the queue is empty before shutting down the clean-up thread. + */ + rcu_barrier(); + + /* + * The teardown of the notification system is performed after the + * session daemon's teardown in order to allow it to be notified + * of the active session and channels at the moment of the teardown. + */ + if (notification_thread_handle) { + if (notification_thread_running) { + notification_thread_command_quit( + notification_thread_handle); + ret = pthread_join(notification_thread, &status); + if (ret) { + errno = ret; + PERROR("pthread_join notification thread"); + retval = -1; + } + } + notification_thread_handle_destroy(notification_thread_handle); } - ret = pthread_join(ht_cleanup_thread, &status); + rcu_thread_offline(); + rcu_unregister_thread(); + + ret = fini_ht_cleanup_thread(&ht_cleanup_thread); if (ret) { - errno = ret; - PERROR("pthread_join ht cleanup thread"); retval = -1; } + lttng_pipe_destroy(ust32_channel_monitor_pipe); + lttng_pipe_destroy(ust64_channel_monitor_pipe); + lttng_pipe_destroy(kernel_channel_monitor_pipe); exit_ht_cleanup: -exit_set_max_size: - - utils_close_pipe(ht_cleanup_pipe); -exit_ht_cleanup_pipe: - - /* - * Close the ht_cleanup quit pipe. - */ - utils_close_pipe(ht_cleanup_quit_pipe); -exit_ht_cleanup_quit_pipe: health_app_destroy(health_sessiond); exit_health_sessiond_cleanup: +exit_create_run_as_worker_cleanup: exit_options: sessiond_cleanup_options();