X-Git-Url: http://git.efficios.com/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fmain.c;h=deb2edf9ba1e5189cd1d9e40934abfc3da1e43ab;hp=25023ada36fa7e0735ca8cbde6bc7606d3bca6c7;hb=ab0ee2ca29394fa28284b94cc3c598960660404f;hpb=a5f0d7f3a5d8083ea0c313e1b17c935077b69b62 diff --git a/src/bin/lttng-sessiond/main.c b/src/bin/lttng-sessiond/main.c index 25023ada3..deb2edf9b 100644 --- a/src/bin/lttng-sessiond/main.c +++ b/src/bin/lttng-sessiond/main.c @@ -37,6 +37,7 @@ #include #include #include +#include #include #include @@ -72,6 +73,7 @@ #include "load-session-thread.h" #include "syscall.h" #include "agent.h" +#include "ht-cleanup.h" #define CONSUMERD_FILE "lttng-consumerd" @@ -189,7 +191,6 @@ 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 @@ -315,6 +316,11 @@ struct lttng_ht *agent_apps_ht_by_sock = NULL; #define NR_LTTNG_SESSIOND_READY 3 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) @@ -421,47 +427,6 @@ 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. * @@ -494,11 +459,6 @@ 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. */ @@ -800,12 +760,6 @@ static void sessiond_cleanup_options(void) free(kmod_extra_probes_list); run_as_destroy_worker(); - - /* */ - 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); - /* */ } /* @@ -1397,7 +1351,6 @@ restart: 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); @@ -1902,6 +1855,8 @@ 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)) { @@ -2135,6 +2090,7 @@ error_testpoint: ERR("Health error occurred in %s", __func__); } health_unregister(health_sessiond); + rcu_unregister_thread(); return NULL; } @@ -2242,10 +2198,12 @@ 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 @@ -2371,7 +2329,12 @@ 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 */ @@ -2401,8 +2364,8 @@ 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"); @@ -2414,7 +2377,7 @@ 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 @@ -3039,7 +3002,8 @@ 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_METADATA_REGENERATE: + case LTTNG_REGENERATE_METADATA: + case LTTNG_REGENERATE_STATEDUMP: need_domain = 0; break; default: @@ -3081,6 +3045,7 @@ 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 */ @@ -3096,7 +3061,6 @@ 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: @@ -3938,12 +3902,6 @@ error_add_context: 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; @@ -4149,9 +4107,14 @@ error_add_context: cmd_ctx->lsm->u.set_shm_path.shm_path); break; } - case LTTNG_METADATA_REGENERATE: + case LTTNG_REGENERATE_METADATA: + { + ret = cmd_regenerate_metadata(cmd_ctx->session); + break; + } + case LTTNG_REGENERATE_STATEDUMP: { - ret = cmd_metadata_regenerate(cmd_ctx->session); + ret = cmd_regenerate_statedump(cmd_ctx->session); break; } default: @@ -4202,7 +4165,6 @@ 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; } @@ -4213,7 +4175,6 @@ 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; } @@ -4222,7 +4183,6 @@ 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; } } @@ -4323,7 +4283,6 @@ restart: if (ret) { PERROR("close"); } - new_sock = -1; continue; } @@ -4352,7 +4311,6 @@ restart: if (ret) { PERROR("close"); } - new_sock = -1; } exit: @@ -4370,7 +4328,7 @@ error: } lttng_poll_clean(&events); - + stop_threads(); rcu_unregister_thread(); return NULL; } @@ -5408,9 +5366,6 @@ error: static void sighandler(int sig) { switch (sig) { - case SIGPIPE: - DBG("SIGPIPE caught"); - return; case SIGINT: DBG("SIGINT caught"); stop_threads(); @@ -5442,9 +5397,10 @@ 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; @@ -5455,12 +5411,13 @@ 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; } @@ -5472,14 +5429,14 @@ 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; @@ -5648,29 +5605,8 @@ 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; } @@ -6037,8 +5973,8 @@ 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; @@ -6048,7 +5984,7 @@ int main(int argc, char **argv) } /* 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; @@ -6058,7 +5994,7 @@ int main(int argc, char **argv) } /* 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; @@ -6068,7 +6004,7 @@ int main(int argc, char **argv) } /* 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; @@ -6078,7 +6014,7 @@ int main(int argc, char **argv) } /* 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; @@ -6088,7 +6024,7 @@ int main(int argc, char **argv) } /* 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; @@ -6098,7 +6034,7 @@ int main(int argc, char **argv) } /* 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; @@ -6110,7 +6046,7 @@ int main(int argc, char **argv) /* 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; @@ -6121,8 +6057,8 @@ int main(int argc, char **argv) } /* 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"); @@ -6214,6 +6150,12 @@ exit_client: 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. @@ -6223,38 +6165,24 @@ exit_init_data: 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(); - ret = pthread_join(ht_cleanup_thread, &status); + ret = fini_ht_cleanup_thread(&ht_cleanup_thread); if (ret) { - errno = ret; - PERROR("pthread_join ht cleanup thread"); retval = -1; } 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: - /* Ensure all prior call_rcu are done. */ - rcu_barrier(); - sessiond_cleanup_options(); exit_set_signal_handler: