X-Git-Url: http://git.lttng.org/?a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fmain.c;h=c5a2ca64e1b6dcc3f4b2c64bc2f0a68a0aaf2b53;hb=0825d4beedb0dbe7bc8abb779c34e80a9a1bc658;hp=2f90b3b30fcd132ce4e9c6597b33dffbed1b03d2;hpb=b22c5da8183071fe98110672b24d8a5ddfdd4511;p=lttng-tools.git diff --git a/src/bin/lttng-sessiond/main.c b/src/bin/lttng-sessiond/main.c index 2f90b3b30..c5a2ca64e 100644 --- a/src/bin/lttng-sessiond/main.c +++ b/src/bin/lttng-sessiond/main.c @@ -67,8 +67,10 @@ #include "health-sessiond.h" #include "testpoint.h" #include "ust-thread.h" -#include "jul-thread.h" +#include "agent-thread.h" #include "save.h" +#include "load-session-thread.h" +#include "syscall.h" #define CONSUMERD_FILE "lttng-consumerd" @@ -80,9 +82,11 @@ static int opt_sig_parent; static int opt_verbose_consumer; static int opt_daemon, opt_background; static int opt_no_kernel; +static char *opt_load_session_path; static pid_t ppid; /* Parent PID for --sig-parent option */ static pid_t child_ppid; /* Internal parent PID use with daemonize. */ static char *rundir; +static int lockfile_fd = -1; /* Set to 1 when a SIGUSR1 signal is received. */ static int recv_child_signal; @@ -150,8 +154,11 @@ static const struct option long_options[] = { { "verbose-consumer", 0, 0, 'Z' }, { "no-kernel", 0, 0, 'N' }, { "pidfile", 1, 0, 'p' }, - { "jul-tcp-port", 1, 0, 'J' }, + { "agent-tcp-port", 1, 0, 'J' }, { "config", 1, 0, 'f' }, + { "load", 1, 0, 'l' }, + { "kmod-probes", 1, 0, 'P' }, + { "extra-kmod-probes", 1, 0, 'e' }, { NULL, 0, 0, 0 } }; @@ -199,7 +206,8 @@ static pthread_t kernel_thread; static pthread_t dispatch_thread; static pthread_t health_thread; static pthread_t ht_cleanup_thread; -static pthread_t jul_reg_thread; +static pthread_t agent_reg_thread; +static pthread_t load_session_thread; /* * UST registration command queue. This queue is tied with a futex and uses a N @@ -283,25 +291,28 @@ long page_size; /* Application health monitoring */ struct health_app *health_sessiond; -/* JUL TCP port for registration. Used by the JUL thread. */ -unsigned int jul_tcp_port = DEFAULT_JUL_TCP_PORT; +/* Agent TCP port for registration. Used by the agent thread. */ +unsigned int agent_tcp_port = DEFAULT_AGENT_TCP_PORT; /* Am I root or not. */ int is_root; /* Set to 1 if the daemon is running as root */ const char * const config_section_name = "sessiond"; +/* Load session thread information to operate. */ +struct load_session_thread_data *load_info; + /* * Whether sessiond is ready for commands/health check requests. * NR_LTTNG_SESSIOND_READY must match the number of calls to - * lttng_sessiond_notify_ready(). + * sessiond_notify_ready(). */ -#define NR_LTTNG_SESSIOND_READY 2 +#define NR_LTTNG_SESSIOND_READY 3 int lttng_sessiond_ready = NR_LTTNG_SESSIOND_READY; /* Notify parents that we are ready for cmd and health check */ -static -void lttng_sessiond_notify_ready(void) +LTTNG_HIDDEN +void sessiond_notify_ready(void) { if (uatomic_sub_return(<tng_sessiond_ready, 1) == 0) { /* @@ -503,6 +514,27 @@ static void close_consumer_sockets(void) } } +/* + * Generate the full lock file path using the rundir. + * + * Return the snprintf() return value thus a negative value is an error. + */ +static int generate_lock_file_path(char *path, size_t len) +{ + int ret; + + assert(path); + assert(rundir); + + /* Build lockfile path from rundir. */ + ret = snprintf(path, len, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE, rundir); + if (ret < 0) { + PERROR("snprintf lockfile path"); + } + + return ret; +} + /* * Cleanup the daemon */ @@ -541,7 +573,7 @@ static void cleanup(void) (void) unlink(path); snprintf(path, PATH_MAX, "%s/%s", rundir, - DEFAULT_LTTNG_SESSIOND_JULPORT_FILE); + DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE); DBG("Removing %s", path); (void) unlink(path); @@ -584,14 +616,6 @@ static void cleanup(void) DBG("Removing directory %s", path); (void) rmdir(path); - /* - * We do NOT rmdir rundir because there are other processes - * using it, for instance lttng-relayd, which can start in - * parallel with this teardown. - */ - - free(rundir); - DBG("Cleaning up all sessions"); /* Destroy session list mutex */ @@ -605,6 +629,9 @@ static void cleanup(void) } } + DBG("Cleaning up all agent apps"); + agent_app_ht_clean(); + DBG("Closing all UST sockets"); ust_app_clean_list(); buffer_reg_destroy_registries(); @@ -619,6 +646,7 @@ static void cleanup(void) } DBG("Unloading kernel modules"); modprobe_remove_lttng_all(); + free(syscall_table); } close_consumer_sockets(); @@ -647,6 +675,43 @@ static void cleanup(void) free(opt_pidfile); } + if (opt_load_session_path) { + free(opt_load_session_path); + } + + if (load_info) { + load_session_destroy_data(load_info); + free(load_info); + } + + /* + * Cleanup lock file by deleting it and finaly closing it which will + * release the file system lock. + */ + if (lockfile_fd >= 0) { + char lockfile_path[PATH_MAX]; + + ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path)); + if (ret > 0) { + ret = remove(lockfile_path); + if (ret < 0) { + PERROR("remove lock file"); + } + ret = close(lockfile_fd); + if (ret < 0) { + PERROR("close lock file"); + } + } + } + + /* + * We do NOT rmdir rundir because there are other processes + * using it, for instance lttng-relayd, which can start in + * parallel with this teardown. + */ + + free(rundir); + /* */ DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm" "Matthew, BEET driven development works!%c[%dm", @@ -940,12 +1005,14 @@ static void *thread_manage_kernel(void *data) update_poll_flag = 0; } - DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events)); + DBG("Thread kernel polling"); /* Poll infinite value of time */ restart: health_poll_entry(); ret = lttng_poll_wait(&events, -1); + DBG("Thread kernel return from poll on %d fds", + LTTNG_POLL_GETNB(&events)); health_poll_exit(); if (ret < 0) { /* @@ -971,6 +1038,11 @@ static void *thread_manage_kernel(void *data) health_code_update(); + if (!revents) { + /* No activity for this FD (poll implementation). */ + continue; + } + /* Thread quit pipe has been closed. Killing thread. */ ret = sessiond_check_thread_quit_pipe(pollfd, revents); if (ret) { @@ -979,31 +1051,33 @@ 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; } } } @@ -1054,7 +1128,7 @@ static void signal_consumer_condition(struct consumer_data *data, int state) */ static void *thread_manage_consumer(void *data) { - int sock = -1, i, ret, pollfd, err = -1; + int sock = -1, i, ret, pollfd, err = -1, should_quit = 0; uint32_t revents, nb_fd; enum lttcomm_return_code code; struct lttng_poll_event events; @@ -1062,6 +1136,9 @@ static void *thread_manage_consumer(void *data) DBG("[thread] Manage consumer started"); + rcu_register_thread(); + rcu_thread_online(); + health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER); health_code_update(); @@ -1116,6 +1193,11 @@ restart: health_code_update(); + if (!revents) { + /* No activity for this FD (poll implementation). */ + continue; + } + /* Thread quit pipe has been closed. Killing thread. */ ret = sessiond_check_thread_quit_pipe(pollfd, revents); if (ret) { @@ -1125,9 +1207,14 @@ 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; } } } @@ -1212,6 +1299,15 @@ restart: /* Infinite blocking call, waiting for transmission */ restart_poll: while (1) { + health_code_update(); + + /* Exit the thread because the thread quit pipe has been triggered. */ + if (should_quit) { + /* Not a health error. */ + err = 0; + goto exit; + } + health_poll_entry(); ret = lttng_poll_wait(&events, -1); health_poll_exit(); @@ -1234,16 +1330,22 @@ restart_poll: health_code_update(); - /* Thread quit pipe has been closed. Killing thread. */ - ret = sessiond_check_thread_quit_pipe(pollfd, revents); - if (ret) { - err = 0; - goto exit; + if (!revents) { + /* No activity for this FD (poll implementation). */ + continue; } + /* + * Thread quit pipe has been triggered, flag that we should stop + * but continue the current loop to handle potential data from + * consumer. + */ + should_quit = sessiond_check_thread_quit_pipe(pollfd, revents); + 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; } @@ -1261,6 +1363,11 @@ 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); @@ -1268,11 +1375,8 @@ restart_poll: ERR("Handling metadata request"); goto error; } - break; - } else { - ERR("Unknown pollfd"); - goto error; } + /* No need for an else branch all FDs are tested prior. */ } health_code_update(); } @@ -1344,6 +1448,9 @@ error_poll: health_unregister(health_sessiond); DBG("consumer thread cleanup completed"); + rcu_thread_offline(); + rcu_unregister_thread(); + return NULL; } @@ -1387,12 +1494,14 @@ static void *thread_manage_apps(void *data) health_code_update(); while (1) { - DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events)); + DBG("Apps thread polling"); /* Inifinite blocking call, waiting for transmission */ restart: health_poll_entry(); ret = lttng_poll_wait(&events, -1); + DBG("Apps thread return from poll on %d fds", + LTTNG_POLL_GETNB(&events)); health_poll_exit(); if (ret < 0) { /* @@ -1413,6 +1522,11 @@ static void *thread_manage_apps(void *data) health_code_update(); + if (!revents) { + /* No activity for this FD (poll implementation). */ + continue; + } + /* Thread quit pipe has been closed. Killing thread. */ ret = sessiond_check_thread_quit_pipe(pollfd, revents); if (ret) { @@ -1422,10 +1536,7 @@ 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 */ @@ -1438,9 +1549,8 @@ 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); @@ -1449,6 +1559,12 @@ 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 { /* @@ -1464,6 +1580,9 @@ 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; } } @@ -1591,6 +1710,11 @@ static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue) uint32_t revents = LTTNG_POLL_GETEV(&events, i); int pollfd = LTTNG_POLL_GETFD(&events, i); + if (!revents) { + /* No activity for this FD (poll implementation). */ + continue; + } + cds_list_for_each_entry_safe(wait_node, tmp_wait_node, &wait_queue->head, head) { if (pollfd == wait_node->app->sock && @@ -1600,6 +1724,9 @@ static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue) ust_app_destroy(wait_node->app); free(wait_node); break; + } else { + ERR("Unexpected poll events %u for sock %d", revents, pollfd); + goto error; } } } @@ -1626,7 +1753,7 @@ error_create: static void *thread_dispatch_ust_registration(void *data) { int ret, err = -1; - struct cds_wfq_node *node; + struct cds_wfcq_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 = { @@ -1664,7 +1791,7 @@ static void *thread_dispatch_ust_registration(void *data) health_code_update(); /* Dequeue command for registration */ - node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue); + node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail); if (node == NULL) { DBG("Woken up but nothing in the UST command queue"); /* Continue thread execution */ @@ -1801,7 +1928,7 @@ 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 @@ -1843,6 +1970,22 @@ 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) { @@ -1930,6 +2073,11 @@ static void *thread_registration_apps(void *data) revents = LTTNG_POLL_GETEV(&events, i); pollfd = LTTNG_POLL_GETFD(&events, i); + if (!revents) { + /* No activity for this FD (poll implementation). */ + continue; + } + /* Thread quit pipe has been closed. Killing thread. */ ret = sessiond_check_thread_quit_pipe(pollfd, revents); if (ret) { @@ -1939,10 +2087,7 @@ 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; @@ -1970,6 +2115,10 @@ 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; } @@ -2018,13 +2167,19 @@ static void *thread_registration_apps(void *data) * Lock free enqueue the registration request. The red pill * has been taken! This apps will be part of the *system*. */ - cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node); + cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node); /* * Wake the registration queue futex. Implicit memory - * barrier with the exchange in cds_wfq_enqueue. + * 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; } } } @@ -2233,9 +2388,12 @@ static pid_t spawn_consumerd(struct consumer_data *consumer_data) */ if (opt_verbose_consumer) { verbosity = "--verbose"; - } else { + } else if (lttng_opt_quiet) { verbosity = "--quiet"; + } else { + verbosity = ""; } + switch (consumer_data->type) { case LTTNG_CONSUMER_KERNEL: /* @@ -2522,7 +2680,7 @@ 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); @@ -2531,10 +2689,11 @@ static int copy_session_consumer(int domain, struct ltt_session *session) dir_name = DEFAULT_KERNEL_TRACE_DIR; break; case LTTNG_DOMAIN_JUL: + case LTTNG_DOMAIN_LOG4J: 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); @@ -2575,6 +2734,7 @@ static int create_ust_session(struct ltt_session *session, switch (domain->type) { case LTTNG_DOMAIN_JUL: + case LTTNG_DOMAIN_LOG4J: case LTTNG_DOMAIN_UST: break; default: @@ -2643,7 +2803,7 @@ 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; } @@ -2761,6 +2921,7 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, case LTTNG_LIST_DOMAINS: case LTTNG_LIST_CHANNELS: case LTTNG_LIST_EVENTS: + case LTTNG_LIST_SYSCALLS: break; default: /* Setup lttng message with no payload */ @@ -2779,6 +2940,7 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, case LTTNG_CALIBRATE: case LTTNG_LIST_SESSIONS: case LTTNG_LIST_TRACEPOINTS: + case LTTNG_LIST_SYSCALLS: case LTTNG_LIST_TRACEPOINT_FIELDS: case LTTNG_SAVE_SESSION: need_tracing_session = 0; @@ -2802,6 +2964,38 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, break; } + /* + * Commands that need a valid session but should NOT create one if none + * exists. Instead of creating one and destroying it when the command is + * handled, process that right before so we save some round trip in useless + * code path. + */ + switch (cmd_ctx->lsm->cmd_type) { + case LTTNG_DISABLE_CHANNEL: + case LTTNG_DISABLE_EVENT: + switch (cmd_ctx->lsm->domain.type) { + case LTTNG_DOMAIN_KERNEL: + if (!cmd_ctx->session->kernel_session) { + ret = LTTNG_ERR_NO_CHANNEL; + goto error; + } + break; + case LTTNG_DOMAIN_JUL: + case LTTNG_DOMAIN_LOG4J: + case LTTNG_DOMAIN_UST: + if (!cmd_ctx->session->ust_session) { + ret = LTTNG_ERR_NO_CHANNEL; + goto error; + } + break; + default: + ret = LTTNG_ERR_UNKNOWN_DOMAIN; + goto error; + } + default: + break; + } + if (!need_domain) { goto skip_domain; } @@ -2869,6 +3063,7 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, break; case LTTNG_DOMAIN_JUL: + case LTTNG_DOMAIN_LOG4J: case LTTNG_DOMAIN_UST: { if (!ust_app_supported()) { @@ -2922,6 +3117,7 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, } /* 32-bit */ + pthread_mutex_lock(&ustconsumer32_data.pid_mutex); if (consumerd32_bin[0] != '\0' && ustconsumer32_data.pid == 0 && cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) { @@ -2961,6 +3157,7 @@ skip_domain: cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) { switch (cmd_ctx->lsm->domain.type) { case LTTNG_DOMAIN_JUL: + case LTTNG_DOMAIN_LOG4J: case LTTNG_DOMAIN_UST: if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) { ret = LTTNG_ERR_NO_USTCONSUMERD; @@ -3021,17 +3218,37 @@ 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 */ ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type, cmd_ctx->lsm->u.disable.channel_name, - cmd_ctx->lsm->u.disable.name); - break; - } - case LTTNG_DISABLE_ALL_EVENT: - { - DBG("Disabling all events"); - - ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type, - cmd_ctx->lsm->u.disable.channel_name); + &cmd_ctx->lsm->u.disable.event); break; } case LTTNG_ENABLE_CHANNEL: @@ -3108,12 +3325,14 @@ skip_domain: if (bytecode_len > LTTNG_FILTER_MAX_LEN) { ret = LTTNG_ERR_FILTER_INVAL; + free(filter_expression); free(exclusion); goto error; } bytecode = zmalloc(bytecode_len); if (!bytecode) { + free(filter_expression); free(exclusion); ret = LTTNG_ERR_FILTER_NOMEM; goto error; @@ -3125,6 +3344,7 @@ skip_domain: if (ret <= 0) { DBG("Nothing recv() from client car len data... continuing"); *sock_error = 1; + free(filter_expression); free(bytecode); free(exclusion); ret = LTTNG_ERR_FILTER_INVAL; @@ -3132,6 +3352,7 @@ skip_domain: } if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) { + free(filter_expression); free(bytecode); free(exclusion); ret = LTTNG_ERR_FILTER_INVAL; @@ -3146,22 +3367,14 @@ skip_domain: 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, - cmd_ctx->lsm->u.enable.channel_name, - cmd_ctx->lsm->u.enable.event.type, NULL, NULL, - kernel_poll_pipe[1]); - break; - } case LTTNG_LIST_TRACEPOINTS: { struct lttng_event *events; ssize_t nb_events; + session_lock_list(); nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events); + session_unlock_list(); if (nb_events < 0) { /* Return value is a negative lttng_error_code. */ ret = -nb_events; @@ -3192,8 +3405,10 @@ skip_domain: struct lttng_event_field *fields; ssize_t nb_fields; + session_lock_list(); nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type, &fields); + session_unlock_list(); if (nb_fields < 0) { /* Return value is a negative lttng_error_code. */ ret = -nb_fields; @@ -3220,6 +3435,37 @@ skip_domain: ret = LTTNG_OK; break; } + case LTTNG_LIST_SYSCALLS: + { + struct lttng_event *events; + ssize_t nb_events; + + nb_events = cmd_list_syscalls(&events); + if (nb_events < 0) { + /* Return value is a negative lttng_error_code. */ + ret = -nb_events; + goto error; + } + + /* + * 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); + 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; + } case LTTNG_SET_CONSUMER_URI: { size_t nb_uri, len; @@ -3250,31 +3496,12 @@ skip_domain: goto error; } - ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session, - nb_uri, uris); + ret = cmd_set_consumer_uri(cmd_ctx->session, nb_uri, uris); + free(uris); if (ret != LTTNG_OK) { - free(uris); goto error; } - /* - * XXX: 0 means that this URI should be applied on the session. Should - * be a DOMAIN enuam. - */ - 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_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session, - nb_uri, uris); - } else if (cmd_ctx->session->kernel_session && - cmd_ctx->session->kernel_session->consumer) { - ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL, - cmd_ctx->session, nb_uri, uris); - } - } - - free(uris); break; } @@ -3367,7 +3594,7 @@ skip_domain: case LTTNG_LIST_CHANNELS: { int nb_chan; - struct lttng_channel *channels; + struct lttng_channel *channels = NULL; nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type, cmd_ctx->session, &channels); @@ -3716,7 +3943,7 @@ static void *thread_manage_health(void *data) goto error; } - lttng_sessiond_notify_ready(); + sessiond_notify_ready(); while (1) { DBG("Health check ready"); @@ -3741,6 +3968,11 @@ restart: revents = LTTNG_POLL_GETEV(&events, i); pollfd = LTTNG_POLL_GETFD(&events, i); + if (!revents) { + /* No activity for this FD (poll implementation). */ + continue; + } + /* Thread quit pipe has been closed. Killing thread. */ ret = sessiond_check_thread_quit_pipe(pollfd, revents); if (ret) { @@ -3750,9 +3982,14 @@ 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; } } } @@ -3868,7 +4105,12 @@ static void *thread_manage_clients(void *data) goto error; } - lttng_sessiond_notify_ready(); + sessiond_notify_ready(); + ret = sem_post(&load_info->message_thread_ready); + if (ret) { + PERROR("sem_post message_thread_ready"); + goto error; + } /* This testpoint is after we signal readiness to the parent. */ if (testpoint(sessiond_thread_manage_clients)) { @@ -3908,6 +4150,11 @@ static void *thread_manage_clients(void *data) health_code_update(); + if (!revents) { + /* No activity for this FD (poll implementation). */ + continue; + } + /* Thread quit pipe has been closed. Killing thread. */ ret = sessiond_check_thread_quit_pipe(pollfd, revents); if (ret) { @@ -3917,9 +4164,14 @@ 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; } } } @@ -4101,8 +4353,11 @@ static void usage(void) 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, " --jul-tcp-port JUL application registration TCP port\n"); - fprintf(stderr, " -f --config Load daemon configuration file\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"); } /* @@ -4115,6 +4370,17 @@ 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; + } + switch (opt) { case 0: fprintf(stderr, "option %s", optname); @@ -4135,7 +4401,20 @@ static int set_option(int opt, const char *arg, const char *optname) opt_background = 1; break; case 'g': + /* + * If the override option is set, the pointer points to a + * *non* const thus freeing it even though the variable type is + * set to const. + */ + if (tracing_group_name_override) { + free((void *) tracing_group_name); + } tracing_group_name = strdup(arg); + if (!tracing_group_name) { + perror("strdup"); + ret = -ENOMEM; + } + tracing_group_name_override = 1; break; case 'h': usage(); @@ -4173,10 +4452,15 @@ static int set_option(int opt, const char *arg, const char *optname) case 'v': /* Verbose level can increase using multiple -v */ if (arg) { + /* Value obtained from config file */ lttng_opt_verbose = config_parse_value(arg); } else { - lttng_opt_verbose += 1; + /* -v used on command line */ + lttng_opt_verbose++; } + /* Clamp value to [0, 3] */ + lttng_opt_verbose = lttng_opt_verbose < 0 ? 0 : + (lttng_opt_verbose <= 3 ? lttng_opt_verbose : 3); break; case 'Z': if (arg) { @@ -4186,48 +4470,125 @@ static int set_option(int opt, const char *arg, const char *optname) } break; case 'u': + if (consumerd32_bin_override) { + free((void *) consumerd32_bin); + } consumerd32_bin = strdup(arg); + if (!consumerd32_bin) { + perror("strdup"); + ret = -ENOMEM; + } consumerd32_bin_override = 1; break; case 'U': + if (consumerd32_libdir_override) { + free((void *) consumerd32_libdir); + } consumerd32_libdir = strdup(arg); + if (!consumerd32_libdir) { + perror("strdup"); + ret = -ENOMEM; + } consumerd32_libdir_override = 1; break; case 't': + if (consumerd64_bin_override) { + free((void *) consumerd64_bin); + } consumerd64_bin = strdup(arg); + if (!consumerd64_bin) { + perror("strdup"); + ret = -ENOMEM; + } consumerd64_bin_override = 1; break; case 'T': + if (consumerd64_libdir_override) { + free((void *) consumerd64_libdir); + } consumerd64_libdir = strdup(arg); + if (!consumerd64_libdir) { + perror("strdup"); + ret = -ENOMEM; + } consumerd64_libdir_override = 1; break; case 'p': + free(opt_pidfile); opt_pidfile = strdup(arg); + if (!opt_pidfile) { + perror("strdup"); + ret = -ENOMEM; + } break; - case 'J': /* JUL TCP port. */ + case 'J': /* Agent TCP port. */ { unsigned long v; errno = 0; v = strtoul(arg, NULL, 0); if (errno != 0 || !isdigit(arg[0])) { - ERR("Wrong value in --jul-tcp-port parameter: %s", arg); + ERR("Wrong value in --agent-tcp-port parameter: %s", arg); return -1; } if (v == 0 || v >= 65535) { - ERR("Port overflow in --jul-tcp-port parameter: %s", arg); + ERR("Port overflow in --agent-tcp-port parameter: %s", arg); return -1; } - jul_tcp_port = (uint32_t) v; - DBG3("JUL TCP port set to non default: %u", jul_tcp_port); + agent_tcp_port = (uint32_t) v; + DBG3("Agent TCP port set to non default: %u", agent_tcp_port); break; } + case 'l': + free(opt_load_session_path); + opt_load_session_path = strdup(arg); + if (!opt_load_session_path) { + perror("strdup"); + ret = -ENOMEM; + } + break; + case 'P': /* probe modules list */ + free(kmod_probes_list); + kmod_probes_list = strdup(arg); + if (!kmod_probes_list) { + perror("strdup"); + ret = -ENOMEM; + } + break; + case 'e': + free(kmod_extra_probes_list); + kmod_extra_probes_list = strdup(arg); + if (!kmod_extra_probes_list) { + perror("strdup"); + ret = -ENOMEM; + } + break; + case 'f': + /* This is handled in set_options() thus silent break. */ + break; default: /* Unknown option or other error. * Error is printed by getopt, just return */ ret = -1; } +end: + if (ret == -EINVAL) { + const char *opt_name = "unknown"; + int i; + + for (i = 0; i < sizeof(long_options) / sizeof(struct option); + i++) { + if (opt == long_options[i].val) { + opt_name = long_options[i].name; + break; + } + } + + WARN("Invalid argument provided for option \"%s\", using default value.", + opt_name); + } + return ret; } @@ -4700,7 +5061,10 @@ static void 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/" @@ -4722,9 +5086,27 @@ error: } /* - * Write JUL TCP port using the rundir. + * Create lockfile using the rundir and return its fd. + */ +static int create_lockfile(void) +{ + int ret; + char lockfile_path[PATH_MAX]; + + ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path)); + if (ret < 0) { + goto error; + } + + ret = utils_create_lock_file(lockfile_path); +error: + return ret; +} + +/* + * Write agent TCP port using the rundir. */ -static void write_julport(void) +static void write_agent_port(void) { int ret; char path[PATH_MAX]; @@ -4732,23 +5114,52 @@ static void write_julport(void) assert(rundir); ret = snprintf(path, sizeof(path), "%s/" - DEFAULT_LTTNG_SESSIOND_JULPORT_FILE, rundir); + DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE, rundir); if (ret < 0) { - PERROR("snprintf julport path"); + PERROR("snprintf agent port path"); goto error; } /* - * Create TCP JUL port file in rundir. Return value is of no importance. + * Create TCP agent port 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(jul_tcp_port, path); + (void) utils_create_pid_file(agent_tcp_port, path); error: return; } +/* + * Start the load session thread and dettach from it so the main thread can + * continue. This does not return a value since whatever the outcome, the main + * thread will continue. + */ +static void start_load_session_thread(void) +{ + int ret; + + /* Create session loading thread. */ + ret = pthread_create(&load_session_thread, NULL, thread_load_session, + load_info); + if (ret != 0) { + PERROR("pthread_create load_session_thread"); + goto error_create; + } + + ret = pthread_detach(load_session_thread); + if (ret != 0) { + PERROR("pthread_detach load_session_thread"); + } + + /* Everything went well so don't cleanup anything. */ + +error_create: + /* The cleanup() function will destroy the load_info data. */ + return; +} + /* * main */ @@ -4758,6 +5169,9 @@ int main(int argc, char **argv) void *status; const char *home_path, *env_app_timeout; + /* Initialize agent apps ht global variable */ + agent_apps_ht_by_sock = NULL; + init_kernel_workarounds(); rcu_register_thread(); @@ -4811,6 +5225,10 @@ int main(int argc, char **argv) if (is_root) { rundir = strdup(DEFAULT_LTTNG_RUNDIR); + if (!rundir) { + ret = -ENOMEM; + goto error; + } /* Create global run dir with root access */ ret = create_lttng_rundir(rundir); @@ -4897,6 +5315,11 @@ int main(int argc, char **argv) } } + lockfile_fd = create_lockfile(); + if (lockfile_fd < 0) { + goto error; + } + /* Set consumer initial state */ kernel_consumerd_state = CONSUMER_STOPPED; ust_consumerd_state = CONSUMER_STOPPED; @@ -4940,20 +5363,24 @@ int main(int argc, char **argv) goto error; } + /* After this point, we can safely call cleanup() with "goto exit" */ + /* * Init UST app hash table. Alloc hash table before this point since * cleanup() can get called after that point. */ ust_app_ht_alloc(); - /* Initialize JUL domain subsystem. */ - if ((ret = jul_init()) < 0) { - /* ENOMEM at this point. */ - goto error; + /* + * Initialize agent app hash table. We allocate the hash table here + * since cleanup() can get called after this point. + */ + if (agent_app_ht_alloc()) { + ERR("Failed to allocate Agent app hash table"); + ret = -1; + goto exit; } - /* After this point, we can safely call cleanup() with "goto exit" */ - /* * These actions must be executed as root. We do that *after* setting up * the sockets path because we MUST make the check for another daemon using @@ -4969,6 +5396,13 @@ int main(int argc, char **argv) /* Setup kernel tracer */ if (!opt_no_kernel) { init_kernel_tracer(); + if (kernel_tracer_fd >= 0) { + ret = syscall_init_table(); + if (ret < 0) { + ERR("Unable to populate syscall table. Syscall tracing" + " won't work for this session daemon."); + } + } } /* Set ulimit for open files */ @@ -5029,7 +5463,7 @@ int main(int argc, char **argv) buffer_reg_init_pid_registry(); /* Init UST command queue. */ - cds_wfq_init(&ust_cmd_queue.queue); + cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail); /* * Get session list pointer. This pointer MUST NOT be free(). This list is @@ -5051,13 +5485,18 @@ int main(int argc, char **argv) } write_pidfile(); - write_julport(); + write_agent_port(); /* Initialize communication library */ lttcomm_init(); /* This is to get the TCP timeout value. */ lttcomm_inet_init(); + if (load_session_init_data(&load_info) < 0) { + goto exit; + } + load_info->path = opt_load_session_path; + /* * Initialize the health check subsystem. This call should set the * appropriate time values. @@ -5120,16 +5559,16 @@ int main(int argc, char **argv) ret = pthread_create(&apps_notify_thread, NULL, ust_thread_manage_notify, (void *) NULL); if (ret != 0) { - PERROR("pthread_create apps"); + PERROR("pthread_create notify"); goto exit_apps_notify; } - /* Create JUL registration thread. */ - ret = pthread_create(&jul_reg_thread, NULL, - jul_thread_manage_registration, (void *) NULL); + /* Create agent registration thread. */ + ret = pthread_create(&agent_reg_thread, NULL, + agent_thread_manage_registration, (void *) NULL); if (ret != 0) { - PERROR("pthread_create apps"); - goto exit_jul_reg; + PERROR("pthread_create agent"); + goto exit_agent_reg; } /* Don't start this thread if kernel tracing is not requested nor root */ @@ -5141,7 +5580,12 @@ int main(int argc, char **argv) PERROR("pthread_create kernel"); goto exit_kernel; } + } + /* Load possible session(s). */ + start_load_session_thread(); + + if (is_root && !opt_no_kernel) { ret = pthread_join(kernel_thread, &status); if (ret != 0) { PERROR("pthread_join"); @@ -5150,13 +5594,13 @@ int main(int argc, char **argv) } exit_kernel: - ret = pthread_join(jul_reg_thread, &status); + ret = pthread_join(agent_reg_thread, &status); if (ret != 0) { - PERROR("pthread_join JUL"); + PERROR("pthread_join agent"); goto error; /* join error, exit without cleanup */ } -exit_jul_reg: +exit_agent_reg: ret = pthread_join(apps_notify_thread, &status); if (ret != 0) { PERROR("pthread_join apps notify"); @@ -5234,6 +5678,7 @@ exit: cleanup(); rcu_thread_offline(); rcu_unregister_thread(); + rcu_barrier(); if (!ret) { exit(EXIT_SUCCESS); }