X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fmain.c;h=3e04f963329e6603c859164f1ee4ff4207a8e118;hp=5205984461a759074c945575c05ebf497554a0b0;hb=c5d350b2620b89d5d488e7bf15e3357ba2b831e8;hpb=72453c8c97897cb5b5e0aabcbd42a8994fab66b2 diff --git a/src/bin/lttng-sessiond/main.c b/src/bin/lttng-sessiond/main.c index 520598446..3e04f9633 100644 --- a/src/bin/lttng-sessiond/main.c +++ b/src/bin/lttng-sessiond/main.c @@ -18,6 +18,7 @@ */ #define _GNU_SOURCE +#define _LGPL_SOURCE #include #include #include @@ -46,6 +47,7 @@ #include #include #include +#include #include #include "lttng-sessiond.h" @@ -66,7 +68,10 @@ #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" @@ -76,12 +81,13 @@ static int tracing_group_name_override; static char *opt_pidfile; static int opt_sig_parent; static int opt_verbose_consumer; -static int opt_daemon; +static int opt_daemon, opt_background; static int opt_no_kernel; -static int is_root; /* Set to 1 if the daemon is running as root */ +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; @@ -139,6 +145,7 @@ static const struct option long_options[] = { { "consumerd64-path", 1, 0, 't' }, { "consumerd64-libdir", 1, 0, 'T' }, { "daemonize", 0, 0, 'd' }, + { "background", 0, 0, 'b' }, { "sig-parent", 0, 0, 'S' }, { "help", 0, 0, 'h' }, { "group", 1, 0, 'g' }, @@ -148,8 +155,11 @@ 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 } }; @@ -197,14 +207,17 @@ 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 * wakers / 1 waiter implemented and detailed in futex.c/.h * - * The thread_manage_apps and thread_dispatch_ust_registration interact with - * this queue and the wait/wake scheme. + * The thread_registration_apps and thread_dispatch_ust_registration uses this + * queue along with the wait/wake scheme. The thread_manage_apps receives down + * the line new application socket and monitors it for any I/O error or clean + * close that triggers an unregistration of the application. */ static struct ust_cmd_queue ust_cmd_queue; @@ -279,11 +292,49 @@ 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 + * sessiond_notify_ready(). + */ +#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 */ +LTTNG_HIDDEN +void sessiond_notify_ready(void) +{ + if (uatomic_sub_return(<tng_sessiond_ready, 1) == 0) { + /* + * Notify parent pid that we are ready to accept command + * for client side. This ppid is the one from the + * external process that spawned us. + */ + if (opt_sig_parent) { + kill(ppid, SIGUSR1); + } + + /* + * Notify the parent of the fork() process that we are + * ready. + */ + if (opt_daemon || opt_background) { + kill(child_ppid, SIGUSR1); + } + } +} + static void setup_consumerd_path(void) { @@ -464,6 +515,27 @@ 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 */ @@ -502,7 +574,7 @@ 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); @@ -545,14 +617,6 @@ 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 */ @@ -580,6 +644,7 @@ static void cleanup(void) } DBG("Unloading kernel modules"); modprobe_remove_lttng_all(); + free(syscall_table); } close_consumer_sockets(); @@ -608,6 +673,43 @@ 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", @@ -1015,7 +1117,7 @@ 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; @@ -1116,7 +1218,6 @@ restart: } health_code_update(); - if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) { /* Connect both socket, command and metadata. */ consumer_data->cmd_sock = @@ -1174,6 +1275,15 @@ 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(); @@ -1196,12 +1306,12 @@ 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; - } + /* + * 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 */ @@ -1230,11 +1340,8 @@ 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(); } @@ -1273,13 +1380,13 @@ error: } consumer_data->cmd_sock = -1; } - if (*consumer_data->metadata_sock.fd_ptr >= 0) { + if (consumer_data->metadata_sock.fd_ptr && + *consumer_data->metadata_sock.fd_ptr >= 0) { ret = close(*consumer_data->metadata_sock.fd_ptr); if (ret) { PERROR("close"); } } - if (sock >= 0) { ret = close(sock); if (ret) { @@ -1293,9 +1400,10 @@ error: pthread_mutex_unlock(&consumer_data->lock); /* Cleanup metadata socket mutex. */ - pthread_mutex_destroy(consumer_data->metadata_sock.lock); - free(consumer_data->metadata_sock.lock); - + if (consumer_data->metadata_sock.lock) { + pthread_mutex_destroy(consumer_data->metadata_sock.lock); + free(consumer_data->metadata_sock.lock); + } lttng_poll_clean(&events); error_poll: if (err) { @@ -1587,7 +1695,7 @@ 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 = { @@ -1596,6 +1704,10 @@ static void *thread_dispatch_ust_registration(void *data) health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH); + if (testpoint(sessiond_thread_app_reg_dispatch)) { + goto error_testpoint; + } + health_code_update(); CDS_INIT_LIST_HEAD(&wait_queue.head); @@ -1621,7 +1733,7 @@ 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 */ @@ -1800,6 +1912,7 @@ error: free(wait_node); } +error_testpoint: DBG("Dispatch thread dying"); if (err) { health_error(); @@ -1974,11 +2087,11 @@ 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); } @@ -1988,11 +2101,6 @@ static void *thread_registration_apps(void *data) exit: error: - if (err) { - health_error(); - ERR("Health error occurred in %s", __func__); - } - /* Notify that the registration thread is gone */ notify_ust_apps(0); @@ -2017,6 +2125,10 @@ error_listen: error_create_poll: error_testpoint: DBG("UST Registration thread cleanup complete"); + if (err) { + health_error(); + ERR("Health error occurred in %s", __func__); + } health_unregister(health_sessiond); return NULL; @@ -2113,19 +2225,23 @@ static int spawn_consumer_thread(struct consumer_data *consumer_data) if (ret != 0) { errno = ret; if (ret == ETIMEDOUT) { + int pth_ret; + /* * Call has timed out so we kill the kconsumerd_thread and return * an error. */ ERR("Condition timed out. The consumer thread was never ready." " Killing it"); - ret = pthread_cancel(consumer_data->thread); - if (ret < 0) { + pth_ret = pthread_cancel(consumer_data->thread); + if (pth_ret < 0) { PERROR("pthread_cancel consumer thread"); } } else { PERROR("pthread_cond_wait failed consumer thread"); } + /* Caller is expecting a negative value on failure. */ + ret = -1; goto error; } @@ -2186,9 +2302,12 @@ 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: /* @@ -2211,10 +2330,11 @@ static pid_t spawn_consumerd(struct consumer_data *consumer_data) consumer_to_use = consumerd32_bin; } else { DBG("Could not find any valid consumerd executable"); + ret = -EINVAL; break; } DBG("Using kernel consumer at: %s", consumer_to_use); - execl(consumer_to_use, + ret = execl(consumer_to_use, "lttng-consumerd", verbosity, "-k", "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path, "--consumerd-err-sock", consumer_data->err_unix_sock_path, @@ -2262,9 +2382,6 @@ static pid_t spawn_consumerd(struct consumer_data *consumer_data) if (consumerd64_libdir[0] != '\0') { free(tmpnew); } - if (ret) { - goto error; - } break; } case LTTNG_CONSUMER32_UST: @@ -2308,9 +2425,6 @@ static pid_t spawn_consumerd(struct consumer_data *consumer_data) if (consumerd32_libdir[0] != '\0') { free(tmpnew); } - if (ret) { - goto error; - } break; } default: @@ -2318,8 +2432,9 @@ static pid_t spawn_consumerd(struct consumer_data *consumer_data) exit(EXIT_FAILURE); } if (errno != 0) { - PERROR("kernel start consumer exec"); + PERROR("Consumer execl()"); } + /* Reaching this point, we got a failure on our execl(). */ exit(EXIT_FAILURE); } else if (pid > 0) { ret = pid; @@ -2488,6 +2603,8 @@ 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_PYTHON: case LTTNG_DOMAIN_UST: DBG3("Copying tracing session consumer output in UST session"); if (session->ust_session->consumer) { @@ -2532,6 +2649,8 @@ static int create_ust_session(struct ltt_session *session, switch (domain->type) { case LTTNG_DOMAIN_JUL: + case LTTNG_DOMAIN_LOG4J: + case LTTNG_DOMAIN_PYTHON: case LTTNG_DOMAIN_UST: break; default: @@ -2678,6 +2797,7 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, case LTTNG_SNAPSHOT_DEL_OUTPUT: case LTTNG_SNAPSHOT_LIST_OUTPUT: case LTTNG_SNAPSHOT_RECORD: + case LTTNG_SAVE_SESSION: need_domain = 0; break; default: @@ -2717,6 +2837,7 @@ 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 */ @@ -2735,7 +2856,9 @@ 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; break; default: @@ -2757,6 +2880,39 @@ 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_PYTHON: + 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; } @@ -2824,6 +2980,8 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, break; case LTTNG_DOMAIN_JUL: + case LTTNG_DOMAIN_LOG4J: + case LTTNG_DOMAIN_PYTHON: case LTTNG_DOMAIN_UST: { if (!ust_app_supported()) { @@ -2877,6 +3035,7 @@ 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) { @@ -2916,6 +3075,8 @@ 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_PYTHON: case LTTNG_DOMAIN_UST: if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) { ret = LTTNG_ERR_NO_USTCONSUMERD; @@ -2976,17 +3137,11 @@ skip_domain: } case LTTNG_DISABLE_EVENT: { + /* 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.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: @@ -2999,6 +3154,7 @@ skip_domain: { struct lttng_event_exclusion *exclusion = NULL; struct lttng_filter_bytecode *bytecode = NULL; + char *filter_expression = NULL; /* Handle exclusion events and receive it from the client. */ if (cmd_ctx->lsm->u.enable.exclusion_count > 0) { @@ -3024,18 +3180,52 @@ skip_domain: } } + /* Get filter expression from client. */ + if (cmd_ctx->lsm->u.enable.expression_len > 0) { + size_t expression_len = + cmd_ctx->lsm->u.enable.expression_len; + + if (expression_len > LTTNG_FILTER_MAX_LEN) { + ret = LTTNG_ERR_FILTER_INVAL; + free(exclusion); + goto error; + } + + filter_expression = zmalloc(expression_len); + if (!filter_expression) { + free(exclusion); + ret = LTTNG_ERR_FILTER_NOMEM; + goto error; + } + + /* Receive var. len. data */ + DBG("Receiving var len filter's expression from client ..."); + ret = lttcomm_recv_unix_sock(sock, filter_expression, + expression_len); + if (ret <= 0) { + DBG("Nothing recv() from client car len data... continuing"); + *sock_error = 1; + free(filter_expression); + free(exclusion); + ret = LTTNG_ERR_FILTER_INVAL; + goto error; + } + } + /* Handle filter and get bytecode from client. */ if (cmd_ctx->lsm->u.enable.bytecode_len > 0) { size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len; 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; @@ -3047,6 +3237,7 @@ 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; @@ -3054,6 +3245,7 @@ skip_domain: } if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) { + free(filter_expression); free(bytecode); free(exclusion); ret = LTTNG_ERR_FILTER_INVAL; @@ -3063,25 +3255,19 @@ skip_domain: ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain, cmd_ctx->lsm->u.enable.channel_name, - &cmd_ctx->lsm->u.enable.event, bytecode, exclusion, + &cmd_ctx->lsm->u.enable.event, + filter_expression, bytecode, exclusion, 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, 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; @@ -3112,8 +3298,10 @@ 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; @@ -3140,6 +3328,37 @@ 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; @@ -3260,7 +3479,7 @@ skip_domain: case LTTNG_LIST_DOMAINS: { ssize_t nb_dom; - struct lttng_domain *domains; + struct lttng_domain *domains = NULL; nb_dom = cmd_list_domains(cmd_ctx->session, &domains); if (nb_dom < 0) { @@ -3287,7 +3506,7 @@ 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); @@ -3532,6 +3751,12 @@ skip_domain: free(uris); break; } + case LTTNG_SAVE_SESSION: + { + ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr, + &cmd_ctx->creds); + break; + } default: ret = LTTNG_ERR_UND; break; @@ -3630,6 +3855,8 @@ static void *thread_manage_health(void *data) goto error; } + sessiond_notify_ready(); + while (1) { DBG("Health check ready"); @@ -3694,7 +3921,7 @@ restart: rcu_thread_online(); - reply.ret_code = 0; + memset(&reply, 0, sizeof(reply)); for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) { /* * health_check_state returns 0 if health is @@ -3780,17 +4007,11 @@ static void *thread_manage_clients(void *data) goto error; } - /* - * Notify parent pid that we are ready to accept command for client side. - * This ppid is the one from the external process that spawned us. - */ - if (opt_sig_parent) { - kill(ppid, SIGUSR1); - } - - /* Notify the parent of the fork() process that we are ready. */ - if (opt_daemon) { - kill(child_ppid, SIGUSR1); + 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. */ @@ -4015,6 +4236,7 @@ static void usage(void) 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"); @@ -4023,8 +4245,11 @@ 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"); } /* @@ -4037,6 +4262,17 @@ 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); @@ -4053,8 +4289,24 @@ static int set_option(int opt, const char *arg, const char *optname) case 'd': opt_daemon = 1; break; + case 'b': + 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(); @@ -4092,10 +4344,15 @@ 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) { @@ -4105,48 +4362,124 @@ 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; } + 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); + } +end: return ret; } @@ -4641,9 +4974,27 @@ 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]; @@ -4651,122 +5002,50 @@ 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; } /* - * Daemonize this process by forking and making the parent wait for the child - * to signal it indicating readiness. Once received, the parent successfully - * quits. - * - * The child process undergoes the same action that daemon(3) does meaning - * setsid, chdir, and dup /dev/null into 0, 1 and 2. - * - * Return 0 on success else -1 on error. + * 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 int daemonize(void) +static void start_load_session_thread(void) { int ret; - pid_t pid; - - /* Get parent pid of this process. */ - child_ppid = getppid(); - - pid = fork(); - if (pid < 0) { - PERROR("fork"); - goto error; - } else if (pid == 0) { - int fd; - pid_t sid; - - /* Child */ - - /* - * Get the newly created parent pid so we can signal that process when - * we are ready to operate. - */ - child_ppid = getppid(); - - sid = setsid(); - if (sid < 0) { - PERROR("setsid"); - goto error; - } - - /* Try to change directory to /. If we can't well at least notify. */ - ret = chdir("/"); - if (ret < 0) { - PERROR("chdir"); - } - - fd = open(_PATH_DEVNULL, O_RDWR, 0); - if (fd < 0) { - PERROR("open %s", _PATH_DEVNULL); - /* Let 0, 1 and 2 open since we can't bind them to /dev/null. */ - } else { - (void) dup2(fd, STDIN_FILENO); - (void) dup2(fd, STDOUT_FILENO); - (void) dup2(fd, STDERR_FILENO); - if (fd > 2) { - ret = close(fd); - if (ret < 0) { - PERROR("close"); - } - } - } - goto end; - } else { - /* Parent */ - /* - * Waiting for child to notify this parent that it can exit. Note that - * sleep() is interrupted before the 1 second delay as soon as the - * signal is received, so it will not cause visible delay for the - * user. - */ - while (!CMM_LOAD_SHARED(recv_child_signal)) { - int status; - pid_t ret; - - /* - * Check if child exists without blocking. If so, we have to stop - * this parent process and return an error. - */ - ret = waitpid(pid, &status, WNOHANG); - if (ret < 0 || (ret != 0 && WIFEXITED(status))) { - /* The child exited somehow or was not valid. */ - goto error; - } - sleep(1); - } + /* 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; + } - /* - * From this point on, the parent can exit and the child is now an - * operationnal session daemon ready to serve clients and applications. - */ - exit(EXIT_SUCCESS); + ret = pthread_detach(load_session_thread); + if (ret != 0) { + PERROR("pthread_detach load_session_thread"); } -end: - return 0; + /* Everything went well so don't cleanup anything. */ -error: - return -1; +error_create: + /* The cleanup() function will destroy the load_info data. */ + return; } /* @@ -4802,10 +5081,11 @@ int main(int argc, char **argv) } /* Daemonize */ - if (opt_daemon) { + if (opt_daemon || opt_background) { int i; - ret = daemonize(); + ret = lttng_daemonize(&child_ppid, &recv_child_signal, + !opt_background); if (ret < 0) { goto error; } @@ -4916,6 +5196,11 @@ 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; @@ -4965,8 +5250,8 @@ int main(int argc, char **argv) */ ust_app_ht_alloc(); - /* Initialize JUL domain subsystem. */ - if ((ret = jul_init()) < 0) { + /* Initialize agent domain subsystem. */ + if ((ret = agent_setup()) < 0) { /* ENOMEM at this point. */ goto error; } @@ -4988,6 +5273,13 @@ 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 */ @@ -5048,7 +5340,7 @@ 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 @@ -5070,13 +5362,18 @@ 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. @@ -5139,16 +5436,16 @@ 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 */ @@ -5160,7 +5457,12 @@ 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"); @@ -5169,13 +5471,13 @@ 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");