X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fcmd.c;h=0330186209268cb63b49175e08d4a4102d07e96a;hp=a2af4c38aa40a9a040ecb5e911408367d9ef2213;hb=92fe5ca1528abfa90cc26f767d274dbff61ffc61;hpb=d88744a44aa5f2ca90ab87946692b9eed3120641 diff --git a/src/bin/lttng-sessiond/cmd.c b/src/bin/lttng-sessiond/cmd.c index a2af4c38a..033018620 100644 --- a/src/bin/lttng-sessiond/cmd.c +++ b/src/bin/lttng-sessiond/cmd.c @@ -21,6 +21,7 @@ #include #include #include +#include #include #include @@ -37,6 +38,8 @@ #include #include #include +#include +#include #include #include "channel.h" @@ -47,17 +50,42 @@ #include "kernel-consumer.h" #include "lttng-sessiond.h" #include "utils.h" -#include "syscall.h" +#include "lttng-syscall.h" #include "agent.h" #include "buffer-registry.h" #include "notification-thread.h" #include "notification-thread-commands.h" #include "rotate.h" #include "rotation-thread.h" -#include "sessiond-timer.h" +#include "timer.h" +#include "agent-thread.h" #include "cmd.h" +/* Sleep for 100ms between each check for the shm path's deletion. */ +#define SESSION_DESTROY_SHM_PATH_CHECK_DELAY_US 100000 + +static enum lttng_error_code wait_on_path(void *path); + +/* + * Command completion handler that is used by the destroy command + * when a session that has a non-default shm_path is being destroyed. + * + * See comment in cmd_destroy_session() for the rationale. + */ +static struct destroy_completion_handler { + struct cmd_completion_handler handler; + char shm_path[member_sizeof(struct ltt_session, shm_path)]; +} destroy_completion_handler = { + .handler = { + .run = wait_on_path, + .data = destroy_completion_handler.shm_path + }, + .shm_path = { 0 }, +}; + +static struct cmd_completion_handler *current_completion_handler; + /* * Used to keep a unique index for each relayd socket created where this value * is associated with streams on the consumer so it can match the right relayd @@ -357,9 +385,13 @@ end: } } -static void increment_extended_len(const char *filter_expression, - struct lttng_event_exclusion *exclusion, size_t *extended_len) +static int increment_extended_len(const char *filter_expression, + struct lttng_event_exclusion *exclusion, + const struct lttng_userspace_probe_location *probe_location, + size_t *extended_len) { + int ret = 0; + *extended_len += sizeof(struct lttcomm_event_extended_header); if (filter_expression) { @@ -369,14 +401,31 @@ static void increment_extended_len(const char *filter_expression, if (exclusion) { *extended_len += exclusion->count * LTTNG_SYMBOL_NAME_LEN; } + + if (probe_location) { + ret = lttng_userspace_probe_location_serialize(probe_location, + NULL, NULL); + if (ret < 0) { + goto end; + } + *extended_len += ret; + } + ret = 0; +end: + return ret; } -static void append_extended_info(const char *filter_expression, - struct lttng_event_exclusion *exclusion, void **extended_at) +static int append_extended_info(const char *filter_expression, + struct lttng_event_exclusion *exclusion, + struct lttng_userspace_probe_location *probe_location, + void **extended_at) { - struct lttcomm_event_extended_header extended_header; + int ret = 0; size_t filter_len = 0; size_t nb_exclusions = 0; + size_t userspace_probe_location_len = 0; + struct lttng_dynamic_buffer location_buffer; + struct lttcomm_event_extended_header extended_header; if (filter_expression) { filter_len = strlen(filter_expression) + 1; @@ -386,9 +435,21 @@ static void append_extended_info(const char *filter_expression, nb_exclusions = exclusion->count; } + if (probe_location) { + lttng_dynamic_buffer_init(&location_buffer); + ret = lttng_userspace_probe_location_serialize(probe_location, + &location_buffer, NULL); + if (ret < 0) { + ret = -1; + goto end; + } + userspace_probe_location_len = location_buffer.size; + } + /* Set header fields */ extended_header.filter_len = filter_len; extended_header.nb_exclusions = nb_exclusions; + extended_header.userspace_probe_location_len = userspace_probe_location_len; /* Copy header */ memcpy(*extended_at, &extended_header, sizeof(extended_header)); @@ -407,6 +468,15 @@ static void append_extended_info(const char *filter_expression, memcpy(*extended_at, &exclusion->names, len); *extended_at += len; } + + if (probe_location) { + memcpy(*extended_at, location_buffer.data, location_buffer.size); + *extended_at += location_buffer.size; + lttng_dynamic_buffer_reset(&location_buffer); + } + ret = 0; +end: + return ret; } /* @@ -420,7 +490,7 @@ static int list_lttng_agent_events(struct agent *agt, int i = 0, ret = 0; unsigned int nb_event = 0; struct agent_event *event; - struct lttng_event *tmp_events; + struct lttng_event *tmp_events = NULL; struct lttng_ht_iter iter; size_t extended_len = 0; void *extended_at; @@ -448,8 +518,13 @@ static int list_lttng_agent_events(struct agent *agt, */ rcu_read_lock(); cds_lfht_for_each_entry(agt->events->ht, &iter.iter, event, node.node) { - increment_extended_len(event->filter_expression, NULL, + ret = increment_extended_len(event->filter_expression, NULL, NULL, &extended_len); + if (ret) { + DBG("Error computing the length of extended info message"); + ret = -LTTNG_ERR_FATAL; + goto error; + } } rcu_read_unlock(); @@ -474,17 +549,25 @@ static int list_lttng_agent_events(struct agent *agt, i++; /* Append extended info */ - append_extended_info(event->filter_expression, NULL, + ret = append_extended_info(event->filter_expression, NULL, NULL, &extended_at); + if (ret) { + DBG("Error appending extended info message"); + ret = -LTTNG_ERR_FATAL; + goto error; + } } - rcu_read_unlock(); *events = tmp_events; ret = nb_event; - -error: assert(nb_event == i); + +end: + rcu_read_unlock(); return ret; +error: + free(tmp_events); + goto end; } /* @@ -533,8 +616,13 @@ static int list_lttng_ust_global_events(char *channel_name, continue; } - increment_extended_len(uevent->filter_expression, - uevent->exclusion, &extended_len); + ret = increment_extended_len(uevent->filter_expression, + uevent->exclusion, NULL, &extended_len); + if (ret) { + DBG("Error computing the length of extended info message"); + ret = -LTTNG_ERR_FATAL; + goto end; + } } if (nb_event == 0) { /* All events are internal, skip. */ @@ -594,8 +682,13 @@ static int list_lttng_ust_global_events(char *channel_name, i++; /* Append extended info */ - append_extended_info(uevent->filter_expression, - uevent->exclusion, &extended_at); + ret = append_extended_info(uevent->filter_expression, + uevent->exclusion, NULL, &extended_at); + if (ret) { + DBG("Error appending extended info message"); + ret = -LTTNG_ERR_FATAL; + goto end; + } } ret = nb_event; @@ -637,14 +730,20 @@ static int list_lttng_kernel_events(char *channel_name, /* Compute required extended infos size */ cds_list_for_each_entry(event, &kchan->events_list.head, list) { - increment_extended_len(event->filter_expression, NULL, + ret = increment_extended_len(event->filter_expression, NULL, + event->userspace_probe_location, &extended_len); + if (ret) { + DBG("Error computing the length of extended info message"); + ret = -LTTNG_ERR_FATAL; + goto error; + } } *total_size = nb_event * sizeof(struct lttng_event) + extended_len; *events = zmalloc(*total_size); if (*events == NULL) { - ret = LTTNG_ERR_FATAL; + ret = -LTTNG_ERR_FATAL; goto error; } @@ -673,6 +772,9 @@ static int list_lttng_kernel_events(char *channel_name, memcpy(&(*events)[i].attr.probe, &event->event->u.kprobe, sizeof(struct lttng_kernel_kprobe)); break; + case LTTNG_KERNEL_UPROBE: + (*events)[i].type = LTTNG_EVENT_USERSPACE_PROBE; + break; case LTTNG_KERNEL_FUNCTION: (*events)[i].type = LTTNG_EVENT_FUNCTION; memcpy(&((*events)[i].attr.ftrace), &event->event->u.ftrace, @@ -685,14 +787,21 @@ static int list_lttng_kernel_events(char *channel_name, (*events)[i].type = LTTNG_EVENT_SYSCALL; break; case LTTNG_KERNEL_ALL: + /* fall-through. */ + default: assert(0); break; } i++; /* Append extended info */ - append_extended_info(event->filter_expression, NULL, - &extended_at); + ret = append_extended_info(event->filter_expression, NULL, + event->userspace_probe_location, &extended_at); + if (ret) { + DBG("Error appending extended info message"); + ret = -LTTNG_ERR_FATAL; + goto error; + } } end: @@ -845,19 +954,20 @@ error: * Create a socket to the relayd using the URI. * * On success, the relayd_sock pointer is set to the created socket. - * Else, it's stays untouched and a lttcomm error code is returned. + * Else, it remains untouched and an LTTng error code is returned. */ -static int create_connect_relayd(struct lttng_uri *uri, +static enum lttng_error_code create_connect_relayd(struct lttng_uri *uri, struct lttcomm_relayd_sock **relayd_sock, struct consumer_output *consumer) { int ret; + enum lttng_error_code status = LTTNG_OK; struct lttcomm_relayd_sock *rsock; rsock = lttcomm_alloc_relayd_sock(uri, RELAYD_VERSION_COMM_MAJOR, RELAYD_VERSION_COMM_MINOR); if (!rsock) { - ret = LTTNG_ERR_FATAL; + status = LTTNG_ERR_FATAL; goto error; } @@ -871,7 +981,7 @@ static int create_connect_relayd(struct lttng_uri *uri, health_poll_exit(); if (ret < 0) { ERR("Unable to reach lttng-relayd"); - ret = LTTNG_ERR_RELAYD_CONNECT_FAIL; + status = LTTNG_ERR_RELAYD_CONNECT_FAIL; goto free_sock; } @@ -882,10 +992,11 @@ static int create_connect_relayd(struct lttng_uri *uri, /* Check relayd version */ ret = relayd_version_check(rsock); if (ret == LTTNG_ERR_RELAYD_VERSION_FAIL) { + status = LTTNG_ERR_RELAYD_VERSION_FAIL; goto close_sock; } else if (ret < 0) { ERR("Unable to reach lttng-relayd"); - ret = LTTNG_ERR_RELAYD_CONNECT_FAIL; + status = LTTNG_ERR_RELAYD_CONNECT_FAIL; goto close_sock; } consumer->relay_major_version = rsock->major; @@ -895,13 +1006,13 @@ static int create_connect_relayd(struct lttng_uri *uri, } else { /* Command is not valid */ ERR("Relayd invalid stream type: %d", uri->stype); - ret = LTTNG_ERR_INVALID; + status = LTTNG_ERR_INVALID; goto close_sock; } *relayd_sock = rsock; - return LTTNG_OK; + return status; close_sock: /* The returned value is not useful since we are on an error path. */ @@ -909,26 +1020,30 @@ close_sock: free_sock: free(rsock); error: - return ret; + return status; } /* * Connect to the relayd using URI and send the socket to the right consumer. * * The consumer socket lock must be held by the caller. + * + * Returns LTTNG_OK on success or an LTTng error code on failure. */ -static int send_consumer_relayd_socket(enum lttng_domain_type domain, - unsigned int session_id, struct lttng_uri *relayd_uri, +static enum lttng_error_code send_consumer_relayd_socket( + unsigned int session_id, + struct lttng_uri *relayd_uri, struct consumer_output *consumer, struct consumer_socket *consumer_sock, char *session_name, char *hostname, int session_live_timer) { int ret; struct lttcomm_relayd_sock *rsock = NULL; + enum lttng_error_code status; /* Connect to relayd and make version check if uri is the control. */ - ret = create_connect_relayd(relayd_uri, &rsock, consumer); - if (ret != LTTNG_OK) { + status = create_connect_relayd(relayd_uri, &rsock, consumer); + if (status != LTTNG_OK) { goto relayd_comm_error; } assert(rsock); @@ -950,7 +1065,7 @@ static int send_consumer_relayd_socket(enum lttng_domain_type domain, relayd_uri->stype, session_id, session_name, hostname, session_live_timer); if (ret < 0) { - ret = LTTNG_ERR_ENABLE_CONSUMER_FAIL; + status = LTTNG_ERR_ENABLE_CONSUMER_FAIL; goto close_sock; } @@ -961,15 +1076,13 @@ static int send_consumer_relayd_socket(enum lttng_domain_type domain, consumer_sock->data_sock_sent = 1; } - ret = LTTNG_OK; - /* * Close socket which was dup on the consumer side. The session daemon does * NOT keep track of the relayd socket(s) once transfer to the consumer. */ close_sock: - if (ret != LTTNG_OK) { + if (status != LTTNG_OK) { /* * The consumer output for this session should not be used anymore * since the relayd connection failed thus making any tracing or/and @@ -981,7 +1094,7 @@ close_sock: free(rsock); relayd_comm_error: - return ret; + return status; } /* @@ -990,39 +1103,42 @@ relayd_comm_error: * session. * * The consumer socket lock must be held by the caller. + * + * Returns LTTNG_OK, or an LTTng error code on failure. */ -static int send_consumer_relayd_sockets(enum lttng_domain_type domain, +static enum lttng_error_code send_consumer_relayd_sockets( + enum lttng_domain_type domain, unsigned int session_id, struct consumer_output *consumer, struct consumer_socket *sock, char *session_name, char *hostname, int session_live_timer) { - int ret = LTTNG_OK; + enum lttng_error_code status = LTTNG_OK; assert(consumer); assert(sock); /* Sending control relayd socket. */ if (!sock->control_sock_sent) { - ret = send_consumer_relayd_socket(domain, session_id, + status = send_consumer_relayd_socket(session_id, &consumer->dst.net.control, consumer, sock, session_name, hostname, session_live_timer); - if (ret != LTTNG_OK) { + if (status != LTTNG_OK) { goto error; } } /* Sending data relayd socket. */ if (!sock->data_sock_sent) { - ret = send_consumer_relayd_socket(domain, session_id, + status = send_consumer_relayd_socket(session_id, &consumer->dst.net.data, consumer, sock, session_name, hostname, session_live_timer); - if (ret != LTTNG_OK) { + if (status != LTTNG_OK) { goto error; } } error: - return ret; + return status; } /* @@ -1152,7 +1268,7 @@ static int start_kernel_session(struct ltt_kernel_session *ksess, int wpipe) } /* Quiescent wait after starting trace */ - kernel_wait_quiescent(kernel_tracer_fd); + kernel_wait_quiescent(wpipe); ksess->active = 1; @@ -1384,9 +1500,15 @@ int cmd_enable_channel(struct ltt_session *session, break; } case LTTNG_DOMAIN_UST: + break; case LTTNG_DOMAIN_JUL: case LTTNG_DOMAIN_LOG4J: case LTTNG_DOMAIN_PYTHON: + if (!agent_tracing_is_enabled()) { + DBG("Attempted to enable a channel in an agent domain but the agent thread is not running"); + ret = LTTNG_ERR_AGENT_TRACING_DISABLED; + goto error; + } break; default: ret = LTTNG_ERR_UNKNOWN_DOMAIN; @@ -1963,6 +2085,7 @@ static int _cmd_enable_event(struct ltt_session *session, break; } case LTTNG_EVENT_PROBE: + case LTTNG_EVENT_USERSPACE_PROBE: case LTTNG_EVENT_FUNCTION: case LTTNG_EVENT_FUNCTION_ENTRY: case LTTNG_EVENT_TRACEPOINT: @@ -2060,8 +2183,7 @@ static int _cmd_enable_event(struct ltt_session *session, ret = validate_ust_event_name(event->name); if (ret) { WARN("Userspace event name %s failed validation.", - event->name ? - event->name : "NULL"); + event->name); ret = LTTNG_ERR_INVALID_EVENT_NAME; goto error; } @@ -2094,6 +2216,12 @@ static int _cmd_enable_event(struct ltt_session *session, assert(usess); + if (!agent_tracing_is_enabled()) { + DBG("Attempted to enable an event in an agent domain but the agent thread is not running"); + ret = LTTNG_ERR_AGENT_TRACING_DISABLED; + goto error; + } + agt = trace_ust_find_agent(usess, domain->type); if (!agt) { agt = agent_create(domain->type); @@ -2556,12 +2684,6 @@ int cmd_start_trace(struct ltt_session *session) /* Flag session that trace should start automatically */ if (usess) { - /* - * Even though the start trace might fail, flag this session active so - * other application coming in are started by default. - */ - usess->active = 1; - ret = ust_app_start_trace_all(usess); if (ret < 0) { ret = LTTNG_ERR_UST_START_FAIL; @@ -2579,51 +2701,19 @@ int cmd_start_trace(struct ltt_session *session) */ session->rotated_after_last_stop = false; - ret = LTTNG_OK; - -error: - return ret; -} - -static -int rename_active_chunk(struct ltt_session *session) -{ - int ret; - - session->rotate_count++; - - /* - * The currently active tracing path is now the folder we - * want to rename. - */ - ret = lttng_strncpy(session->rotation_chunk.current_rotate_path, - session->rotation_chunk.active_tracing_path, - sizeof(session->rotation_chunk.current_rotate_path)); - if (ret) { - ERR("Failed to copy active tracing path"); - goto end; + if (session->rotate_timer_period) { + ret = timer_session_rotation_schedule_timer_start(session, + session->rotate_timer_period); + if (ret < 0) { + ERR("Failed to enable rotate timer"); + ret = LTTNG_ERR_UNK; + goto error; + } } - ret = rename_complete_chunk(session, time(NULL)); - if (ret < 0) { - ERR("Failed to rename current rotate path"); - goto end; - } + ret = LTTNG_OK; - /* - * We just renamed, the folder, we didn't do an actual rotation, so - * the active tracing path is now the renamed folder and we have to - * restore the rotate count. - */ - ret = lttng_strncpy(session->rotation_chunk.active_tracing_path, - session->rotation_chunk.current_rotate_path, - sizeof(session->rotation_chunk.active_tracing_path)); - if (ret) { - ERR("Failed to rename active session chunk tracing path"); - goto end; - } -end: - session->rotate_count--; +error: return ret; } @@ -2636,7 +2726,7 @@ int cmd_stop_trace(struct ltt_session *session) struct ltt_kernel_channel *kchan; struct ltt_kernel_session *ksession; struct ltt_ust_session *usess; - bool error_occured = false; + bool error_occurred = false; assert(session); @@ -2651,21 +2741,6 @@ int cmd_stop_trace(struct ltt_session *session) goto error; } - if (session->rotate_relay_pending_timer_enabled) { - sessiond_timer_rotate_pending_stop(session); - } - - if (session->rotate_count > 0 && !session->rotate_pending) { - ret = rename_active_chunk(session); - if (ret) { - /* - * This error should not prevent the user from stopping - * the session. However, it will be reported at the end. - */ - error_occured = true; - } - } - /* Kernel tracer */ if (ksession && ksession->active) { DBG("Stop kernel tracing"); @@ -2700,12 +2775,6 @@ int cmd_stop_trace(struct ltt_session *session) } if (usess && usess->active) { - /* - * Even though the stop trace might fail, flag this session inactive so - * other application coming in are not started by default. - */ - usess->active = 0; - ret = ust_app_stop_trace_all(usess); if (ret < 0) { ret = LTTNG_ERR_UST_STOP_FAIL; @@ -2715,7 +2784,7 @@ int cmd_stop_trace(struct ltt_session *session) /* Flag inactive after a successful stop. */ session->active = 0; - ret = !error_occured ? LTTNG_OK : LTTNG_ERR_UNK; + ret = !error_occurred ? LTTNG_OK : LTTNG_ERR_UNK; error: return ret; @@ -2803,39 +2872,31 @@ int cmd_create_session_uri(char *name, struct lttng_uri *uris, size_t nb_uri, lttng_sock_cred *creds, unsigned int live_timer) { int ret; - struct ltt_session *session; + struct ltt_session *session = NULL; assert(name); assert(creds); - /* - * Verify if the session already exist - * - * XXX: There is no need for the session lock list here since the caller - * (process_client_msg) is holding it. We might want to change that so a - * single command does not lock the entire session list. - */ + /* Check if the session already exists. */ + session_lock_list(); session = session_find_by_name(name); + session_unlock_list(); if (session != NULL) { ret = LTTNG_ERR_EXIST_SESS; - goto find_error; + goto end; } /* Create tracing session in the registry */ ret = session_create(name, LTTNG_SOCK_GET_UID_CRED(creds), LTTNG_SOCK_GET_GID_CRED(creds)); if (ret != LTTNG_OK) { - goto session_error; + goto end; } - /* - * Get the newly created session pointer back - * - * XXX: There is no need for the session lock list here since the caller - * (process_client_msg) is holding it. We might want to change that so a - * single command does not lock the entire session list. - */ + /* Get the newly created session pointer back. */ + session_lock_list(); session = session_find_by_name(name); + session_unlock_list(); assert(session); session->live_timer = live_timer; @@ -2843,13 +2904,13 @@ int cmd_create_session_uri(char *name, struct lttng_uri *uris, session->consumer = consumer_create_output(CONSUMER_DST_LOCAL); if (session->consumer == NULL) { ret = LTTNG_ERR_FATAL; - goto consumer_error; + goto end; } if (uris) { ret = cmd_set_consumer_uri(session, nb_uri, uris); if (ret != LTTNG_OK) { - goto consumer_error; + goto end; } session->output_traces = 1; } else { @@ -2859,12 +2920,13 @@ int cmd_create_session_uri(char *name, struct lttng_uri *uris, session->consumer->enabled = 1; - return LTTNG_OK; - -consumer_error: - session_destroy(session); -session_error: -find_error: + ret = LTTNG_OK; +end: + if (session) { + session_lock_list(); + session_put(session); + session_unlock_list(); + } return ret; } @@ -2875,7 +2937,7 @@ int cmd_create_session_snapshot(char *name, struct lttng_uri *uris, size_t nb_uri, lttng_sock_cred *creds) { int ret; - struct ltt_session *session; + struct ltt_session *session = NULL; struct snapshot_output *new_output = NULL; assert(name); @@ -2887,11 +2949,13 @@ int cmd_create_session_snapshot(char *name, struct lttng_uri *uris, */ ret = cmd_create_session_uri(name, NULL, 0, creds, 0); if (ret != LTTNG_OK) { - goto error; + goto end; } /* Get the newly created session pointer back. This should NEVER fail. */ + session_lock_list(); session = session_find_by_name(name); + session_unlock_list(); assert(session); /* Flag session for snapshot mode. */ @@ -2899,6 +2963,7 @@ int cmd_create_session_snapshot(char *name, struct lttng_uri *uris, /* Skip snapshot output creation if no URI is given. */ if (nb_uri == 0) { + /* Not an error. */ goto end; } @@ -2923,14 +2988,18 @@ int cmd_create_session_snapshot(char *name, struct lttng_uri *uris, snapshot_add_output(&session->snapshot, new_output); rcu_read_unlock(); -end: - return LTTNG_OK; + ret = LTTNG_OK; + goto end; error_snapshot: snapshot_output_destroy(new_output); error_snapshot_alloc: - session_destroy(session); -error: +end: + if (session) { + session_lock_list(); + session_put(session); + session_unlock_list(); + } return ret; } @@ -2939,61 +3008,113 @@ error: * * Called with session lock held. */ -int cmd_destroy_session(struct ltt_session *session, int wpipe) +int cmd_destroy_session(struct ltt_session *session, + struct notification_thread_handle *notification_thread_handle) { int ret; - struct ltt_ust_session *usess; - struct ltt_kernel_session *ksess; /* Safety net */ assert(session); - usess = session->ust_session; - ksess = session->kernel_session; - DBG("Begin destroy session %s (id %" PRIu64 ")", session->name, session->id); - if (session->rotate_relay_pending_timer_enabled) { - sessiond_timer_rotate_pending_stop(session); + if (session->rotation_schedule_timer_enabled) { + if (timer_session_rotation_schedule_timer_stop( + session)) { + ERR("Failed to stop the \"rotation schedule\" timer of session %s", + session->name); + } } - /* - * The rename of the current chunk is performed at stop, but if we rotated - * the session after the previous stop command, we need to rename the - * new (and empty) chunk that was started in between. - */ - if (session->rotated_after_last_stop) { - rename_active_chunk(session); + if (session->rotate_size) { + unsubscribe_session_consumed_size_rotation(session, notification_thread_handle); + session->rotate_size = 0; } - /* Clean kernel session teardown */ - kernel_destroy_session(ksess); - - /* UST session teardown */ - if (usess) { - /* Close any relayd session */ - consumer_output_send_destroy_relayd(usess->consumer); - - /* Destroy every UST application related to this session. */ - ret = ust_app_destroy_trace_all(usess); - if (ret) { - ERR("Error in ust_app_destroy_trace_all"); + if (session->current_archive_id != 0) { + if (!session->rotated_after_last_stop) { + ret = cmd_rotate_session(session, NULL); + if (ret != LTTNG_OK) { + ERR("Failed to perform an implicit rotation as part of the rotation: %s", lttng_strerror(-ret)); + } + } else { + /* + * Rename the active chunk to ensure it has a name + * of the form ts_begin-ts_end-id. + * + * Note that no trace data has been produced since + * the last rotation; the directory should be + * removed. + */ + ret = rename_active_chunk(session); + if (ret) { + ERR("Failed to rename active chunk during the destruction of session \"%s\"", + session->name); + } } + } - /* Clean up the rest. */ - trace_ust_destroy_session(usess); + if (session->shm_path[0]) { + /* + * When a session is created with an explicit shm_path, + * the consumer daemon will create its shared memory files + * at that location and will *not* unlink them. This is normal + * as the intention of that feature is to make it possible + * to retrieve the content of those files should a crash occur. + * + * To ensure the content of those files can be used, the + * sessiond daemon will replicate the content of the metadata + * cache in a metadata file. + * + * On clean-up, it is expected that the consumer daemon will + * unlink the shared memory files and that the session daemon + * will unlink the metadata file. Then, the session's directory + * in the shm path can be removed. + * + * Unfortunately, a flaw in the design of the sessiond's and + * consumerd's tear down of channels makes it impossible to + * determine when the sessiond _and_ the consumerd have both + * destroyed their representation of a channel. For one, the + * unlinking, close, and rmdir happen in deferred 'call_rcu' + * callbacks in both daemons. + * + * However, it is also impossible for the sessiond to know when + * the consumer daemon is done destroying its channel(s) since + * it occurs as a reaction to the closing of the channel's file + * descriptor. There is no resulting communication initiated + * from the consumerd to the sessiond to confirm that the + * operation is completed (and was successful). + * + * Until this is all fixed, the session daemon checks for the + * removal of the session's shm path which makes it possible + * to safely advertise a session as having been destroyed. + * + * Prior to this fix, it was not possible to reliably save + * a session making use of the --shm-path option, destroy it, + * and load it again. This is because the creation of the + * session would fail upon seeing the session's shm path + * already in existence. + * + * Note that none of the error paths in the check for the + * directory's existence return an error. This is normal + * as there isn't much that can be done. The session will + * be destroyed properly, except that we can't offer the + * guarantee that the same session can be re-created. + */ + current_completion_handler = &destroy_completion_handler.handler; + ret = lttng_strncpy(destroy_completion_handler.shm_path, + session->shm_path, + sizeof(destroy_completion_handler.shm_path)); + assert(!ret); } /* - * Must notify the kernel thread here to update it's poll set in order to - * remove the channel(s)' fd just destroyed. + * The session is destroyed. However, note that the command context + * still holds a reference to the session, thus delaying its destruction + * _at least_ up to the point when that reference is released. */ - ret = notify_thread_pipe(wpipe); - if (ret < 0) { - PERROR("write kernel poll pipe"); - } - - ret = session_destroy(session); + session_destroy(session); + ret = LTTNG_OK; return ret; } @@ -3298,10 +3419,15 @@ void cmd_list_lttng_sessions(struct lttng_session *sessions, uid_t uid, * the buffer. */ cds_list_for_each_entry(session, &list->head, list) { + if (!session_get(session)) { + continue; + } /* * Only list the sessions the user can control. */ - if (!session_access_ok(session, uid, gid)) { + if (!session_access_ok(session, uid, gid) || + session->destroyed) { + session_put(session); continue; } @@ -3319,6 +3445,7 @@ void cmd_list_lttng_sessions(struct lttng_session *sessions, uid_t uid, } if (ret < 0) { PERROR("snprintf session path"); + session_put(session); continue; } @@ -3328,6 +3455,7 @@ void cmd_list_lttng_sessions(struct lttng_session *sessions, uid_t uid, sessions[i].snapshot_mode = session->snapshot_mode; sessions[i].live_timer_interval = session->live_timer; i++; + session_put(session); } } @@ -3366,10 +3494,8 @@ int cmd_data_pending(struct ltt_session *session) } } - /* - * A rotation is still pending, we have to wait. - */ - if (session->rotate_pending) { + /* A rotation is still pending, we have to wait. */ + if (session->rotation_state == LTTNG_ROTATION_STATE_ONGOING) { DBG("Rotate still pending for session %s", session->name); ret = 1; goto error; @@ -3647,10 +3773,12 @@ static int clear_metadata_file(int fd) { int ret; + off_t lseek_ret; - ret = lseek(fd, 0, SEEK_SET); - if (ret < 0) { + lseek_ret = lseek(fd, 0, SEEK_SET); + if (lseek_ret < 0) { PERROR("lseek"); + ret = -1; goto end; } @@ -3923,12 +4051,14 @@ end: * Send relayd sockets from snapshot output to consumer. Ignore request if the * snapshot output is *not* set with a remote destination. * - * Return 0 on success or a LTTNG_ERR code. + * Return LTTNG_OK on success or a LTTNG_ERR code. */ -static int set_relayd_for_snapshot(struct consumer_output *consumer, - struct snapshot_output *snap_output, struct ltt_session *session) +static enum lttng_error_code set_relayd_for_snapshot( + struct consumer_output *consumer, + struct snapshot_output *snap_output, + struct ltt_session *session) { - int ret = LTTNG_OK; + enum lttng_error_code status = LTTNG_OK; struct lttng_ht_iter iter; struct consumer_socket *socket; @@ -3951,12 +4081,12 @@ static int set_relayd_for_snapshot(struct consumer_output *consumer, cds_lfht_for_each_entry(snap_output->consumer->socks->ht, &iter.iter, socket, node.node) { pthread_mutex_lock(socket->lock); - ret = send_consumer_relayd_sockets(0, session->id, + status = send_consumer_relayd_sockets(0, session->id, snap_output->consumer, socket, session->name, session->hostname, session->live_timer); pthread_mutex_unlock(socket->lock); - if (ret != LTTNG_OK) { + if (status != LTTNG_OK) { rcu_read_unlock(); goto error; } @@ -3964,7 +4094,7 @@ static int set_relayd_for_snapshot(struct consumer_output *consumer, rcu_read_unlock(); error: - return ret; + return status; } /* @@ -3972,38 +4102,37 @@ error: * * Return LTTNG_OK on success or a LTTNG_ERR code. */ -static int record_kernel_snapshot(struct ltt_kernel_session *ksess, +static enum lttng_error_code record_kernel_snapshot(struct ltt_kernel_session *ksess, struct snapshot_output *output, struct ltt_session *session, int wait, uint64_t nb_packets_per_stream) { int ret; + enum lttng_error_code status; assert(ksess); assert(output); assert(session); - /* * Copy kernel session sockets so we can communicate with the right * consumer for the snapshot record command. */ ret = consumer_copy_sockets(output->consumer, ksess->consumer); if (ret < 0) { - ret = LTTNG_ERR_NOMEM; + status = LTTNG_ERR_NOMEM; goto error; } - ret = set_relayd_for_snapshot(ksess->consumer, output, session); - if (ret != LTTNG_OK) { + status = set_relayd_for_snapshot(ksess->consumer, output, session); + if (status != LTTNG_OK) { goto error_snapshot; } - ret = kernel_snapshot_record(ksess, output, wait, nb_packets_per_stream); - if (ret != LTTNG_OK) { + status = kernel_snapshot_record(ksess, output, wait, nb_packets_per_stream); + if (status != LTTNG_OK) { goto error_snapshot; } - ret = LTTNG_OK; goto end; error_snapshot: @@ -4011,19 +4140,20 @@ error_snapshot: consumer_destroy_output_sockets(output->consumer); error: end: - return ret; + return status; } /* * Record a UST snapshot. * - * Return 0 on success or a LTTNG_ERR error code. + * Returns LTTNG_OK on success or a LTTNG_ERR error code. */ -static int record_ust_snapshot(struct ltt_ust_session *usess, +static enum lttng_error_code record_ust_snapshot(struct ltt_ust_session *usess, struct snapshot_output *output, struct ltt_session *session, int wait, uint64_t nb_packets_per_stream) { int ret; + enum lttng_error_code status; assert(usess); assert(output); @@ -4035,35 +4165,28 @@ static int record_ust_snapshot(struct ltt_ust_session *usess, */ ret = consumer_copy_sockets(output->consumer, usess->consumer); if (ret < 0) { - ret = LTTNG_ERR_NOMEM; + status = LTTNG_ERR_NOMEM; goto error; } - ret = set_relayd_for_snapshot(usess->consumer, output, session); - if (ret != LTTNG_OK) { + status = set_relayd_for_snapshot(usess->consumer, output, session); + if (status != LTTNG_OK) { goto error_snapshot; } - ret = ust_app_snapshot_record(usess, output, wait, nb_packets_per_stream); - if (ret < 0) { - switch (-ret) { - case EINVAL: - ret = LTTNG_ERR_INVALID; - break; - default: - ret = LTTNG_ERR_SNAPSHOT_FAIL; - break; - } + status = ust_app_snapshot_record(usess, output, wait, nb_packets_per_stream); + if (status != LTTNG_OK) { goto error_snapshot; } - ret = LTTNG_OK; + goto end; error_snapshot: /* Clean up copied sockets so this output can use some other later on. */ consumer_destroy_output_sockets(output->consumer); error: - return ret; +end: + return status; } static @@ -4163,7 +4286,8 @@ int64_t get_session_nb_packets_per_stream(struct ltt_session *session, uint64_t int cmd_snapshot_record(struct ltt_session *session, struct lttng_snapshot_output *output, int wait) { - int ret = LTTNG_OK; + enum lttng_error_code cmd_ret = LTTNG_OK; + int ret; unsigned int use_tmp_output = 0; struct snapshot_output tmp_output; unsigned int snapshot_success = 0; @@ -4178,7 +4302,7 @@ int cmd_snapshot_record(struct ltt_session *session, ret = utils_get_current_time_str("%Y%m%d-%H%M%S", datetime, sizeof(datetime)); if (!ret) { - ret = LTTNG_ERR_INVALID; + cmd_ret = LTTNG_ERR_INVALID; goto error; } @@ -4187,13 +4311,13 @@ int cmd_snapshot_record(struct ltt_session *session, * set in no output mode. */ if (session->output_traces) { - ret = LTTNG_ERR_NOT_SNAPSHOT_SESSION; + cmd_ret = LTTNG_ERR_NOT_SNAPSHOT_SESSION; goto error; } /* The session needs to be started at least once. */ if (!session->has_been_started) { - ret = LTTNG_ERR_START_SESSION_ONCE; + cmd_ret = LTTNG_ERR_START_SESSION_ONCE; goto error; } @@ -4204,9 +4328,9 @@ int cmd_snapshot_record(struct ltt_session *session, &tmp_output, NULL); if (ret < 0) { if (ret == -ENOMEM) { - ret = LTTNG_ERR_NOMEM; + cmd_ret = LTTNG_ERR_NOMEM; } else { - ret = LTTNG_ERR_INVALID; + cmd_ret = LTTNG_ERR_INVALID; } goto error; } @@ -4224,24 +4348,24 @@ int cmd_snapshot_record(struct ltt_session *session, nb_packets_per_stream = get_session_nb_packets_per_stream(session, tmp_output.max_size); if (nb_packets_per_stream < 0) { - ret = LTTNG_ERR_MAX_SIZE_INVALID; + cmd_ret = LTTNG_ERR_MAX_SIZE_INVALID; goto error; } if (session->kernel_session) { - ret = record_kernel_snapshot(session->kernel_session, + cmd_ret = record_kernel_snapshot(session->kernel_session, &tmp_output, session, wait, nb_packets_per_stream); - if (ret != LTTNG_OK) { + if (cmd_ret != LTTNG_OK) { goto error; } } if (session->ust_session) { - ret = record_ust_snapshot(session->ust_session, + cmd_ret = record_ust_snapshot(session->ust_session, &tmp_output, session, wait, nb_packets_per_stream); - if (ret != LTTNG_OK) { + if (cmd_ret != LTTNG_OK) { goto error; } } @@ -4270,7 +4394,7 @@ int cmd_snapshot_record(struct ltt_session *session, nb_packets_per_stream = get_session_nb_packets_per_stream(session, tmp_output.max_size); if (nb_packets_per_stream < 0) { - ret = LTTNG_ERR_MAX_SIZE_INVALID; + cmd_ret = LTTNG_ERR_MAX_SIZE_INVALID; rcu_read_unlock(); goto error; } @@ -4279,7 +4403,7 @@ int cmd_snapshot_record(struct ltt_session *session, if (*output->name != '\0') { if (lttng_strncpy(tmp_output.name, output->name, sizeof(tmp_output.name))) { - ret = LTTNG_ERR_INVALID; + cmd_ret = LTTNG_ERR_INVALID; rcu_read_unlock(); goto error; } @@ -4289,20 +4413,20 @@ int cmd_snapshot_record(struct ltt_session *session, memcpy(tmp_output.datetime, datetime, sizeof(datetime)); if (session->kernel_session) { - ret = record_kernel_snapshot(session->kernel_session, + cmd_ret = record_kernel_snapshot(session->kernel_session, &tmp_output, session, wait, nb_packets_per_stream); - if (ret != LTTNG_OK) { + if (cmd_ret != LTTNG_OK) { rcu_read_unlock(); goto error; } } if (session->ust_session) { - ret = record_ust_snapshot(session->ust_session, + cmd_ret = record_ust_snapshot(session->ust_session, &tmp_output, session, wait, nb_packets_per_stream); - if (ret != LTTNG_OK) { + if (cmd_ret != LTTNG_OK) { rcu_read_unlock(); goto error; } @@ -4315,11 +4439,11 @@ int cmd_snapshot_record(struct ltt_session *session, if (snapshot_success) { session->snapshot.nb_snapshot++; } else { - ret = LTTNG_ERR_SNAPSHOT_FAIL; + cmd_ret = LTTNG_ERR_SNAPSHOT_FAIL; } error: - return ret; + return cmd_ret; } /* @@ -4351,29 +4475,33 @@ int cmd_set_session_shm_path(struct ltt_session *session, * Ask the consumer to rotate the session output directory. * The session lock must be held. * - * Return LTTNG_OK on success or else a LTTNG_ERR code. + * Returns LTTNG_OK on success or else a negative LTTng error code. */ int cmd_rotate_session(struct ltt_session *session, struct lttng_rotate_session_return *rotate_return) { int ret; + enum lttng_error_code cmd_ret = LTTNG_OK; size_t strf_ret; struct tm *timeinfo; - char datetime[16]; + char datetime[21]; time_t now; - bool ust_active = false; + /* + * Used to roll-back timestamps in case of failure to launch the + * rotation. + */ + time_t original_last_chunk_start_ts, original_current_chunk_start_ts; assert(session); if (!session->has_been_started) { - ret = -LTTNG_ERR_START_SESSION_ONCE; - goto error; + cmd_ret = LTTNG_ERR_START_SESSION_ONCE; + goto end; } - if (session->live_timer || session->snapshot_mode || - !session->output_traces) { - ret = -LTTNG_ERR_ROTATION_NOT_AVAILABLE; - goto error; + if (session->live_timer || !session->output_traces) { + cmd_ret = LTTNG_ERR_ROTATION_NOT_AVAILABLE; + goto end; } /* @@ -4382,14 +4510,15 @@ int cmd_rotate_session(struct ltt_session *session, if (session->consumer->type == CONSUMER_DST_NET && (session->consumer->relay_major_version == 2 && session->consumer->relay_minor_version < 11)) { - ret = -LTTNG_ERR_ROTATION_NOT_AVAILABLE; - goto error; + cmd_ret = LTTNG_ERR_ROTATION_NOT_AVAILABLE_RELAY; + goto end; } - if (session->rotate_pending || session->rotate_pending_relay) { - ret = -LTTNG_ERR_ROTATION_PENDING; - DBG("Rotate already in progress"); - goto error; + if (session->rotation_state == LTTNG_ROTATION_STATE_ONGOING) { + DBG("Refusing to launch a rotation; a rotation is already in progress for session %s", + session->name); + cmd_ret = LTTNG_ERR_ROTATION_PENDING; + goto end; } /* @@ -4399,30 +4528,26 @@ int cmd_rotate_session(struct ltt_session *session, if (session->rotated_after_last_stop) { DBG("Session \"%s\" was already rotated after stop, refusing rotation", session->name); - ret = -LTTNG_ERR_ROTATION_MULTIPLE_AFTER_STOP; - goto error; + cmd_ret = LTTNG_ERR_ROTATION_MULTIPLE_AFTER_STOP; + goto end; } /* Special case for the first rotation. */ - if (session->rotate_count == 0) { + if (session->current_archive_id == 0) { const char *base_path = NULL; + assert(session->kernel_session || session->ust_session); /* Either one of the two sessions is enough to get the root path. */ - if (session->kernel_session) { - base_path = session_get_base_path(session); - } else if (session->ust_session) { - base_path = session_get_base_path(session); - } else { - assert(0); - } + base_path = session_get_base_path(session); assert(base_path); + ret = lttng_strncpy(session->rotation_chunk.current_rotate_path, base_path, sizeof(session->rotation_chunk.current_rotate_path)); if (ret) { ERR("Failed to copy session base path to current rotation chunk path"); - ret = -LTTNG_ERR_UNK; - goto error; + cmd_ret = LTTNG_ERR_UNK; + goto end; } } else { /* @@ -4434,40 +4559,54 @@ int cmd_rotate_session(struct ltt_session *session, sizeof(session->rotation_chunk.current_rotate_path)); if (ret) { ERR("Failed to copy the active tracing path to the current rotate path"); - ret = -LTTNG_ERR_UNK; - goto error; + cmd_ret = LTTNG_ERR_UNK; + goto end; } } DBG("Current rotate path %s", session->rotation_chunk.current_rotate_path); - session->rotate_count++; - session->rotate_pending = true; - session->rotation_status = LTTNG_ROTATION_STATUS_STARTED; - /* - * Create the path name for the next chunk. + * Channels created after this point will belong to the next + * archive id. */ + session->current_archive_id++; + now = time(NULL); if (now == (time_t) -1) { - ret = -LTTNG_ERR_ROTATION_NOT_AVAILABLE; - goto error; + cmd_ret = LTTNG_ERR_UNK; + goto end; } + + /* Sample chunk bounds for roll-back in case of error. */ + original_last_chunk_start_ts = session->last_chunk_start_ts; + original_current_chunk_start_ts = session->current_chunk_start_ts; + session->last_chunk_start_ts = session->current_chunk_start_ts; session->current_chunk_start_ts = now; timeinfo = localtime(&now); if (!timeinfo) { PERROR("Failed to sample local time in rotate session command"); - ret = -LTTNG_ERR_UNK; - goto error; + cmd_ret = LTTNG_ERR_UNK; + goto end; } - strf_ret = strftime(datetime, sizeof(datetime), "%Y%m%d-%H%M%S", + strf_ret = strftime(datetime, sizeof(datetime), "%Y%m%dT%H%M%S%z", timeinfo); if (!strf_ret) { ERR("Failed to format local time timestamp in rotate session command"); - ret = -LTTNG_ERR_UNK; - goto error; + cmd_ret = LTTNG_ERR_UNK; + goto end; } + + /* + * A rotation has a local step even if the destination is a relay + * daemon; the buffers must be consumed by the consumer daemon. + */ + session->rotation_pending_local = true; + session->rotation_pending_relay = + session_get_consumer_destination_type(session) == CONSUMER_DST_NET; + session->rotation_state = LTTNG_ROTATION_STATE_ONGOING; + if (session->kernel_session) { /* * The active path for the next rotation/destroy. @@ -4477,10 +4616,10 @@ int cmd_rotate_session(struct ltt_session *session, sizeof(session->rotation_chunk.active_tracing_path), "%s/%s-%" PRIu64, session_get_base_path(session), - datetime, session->rotate_count + 1); + datetime, session->current_archive_id + 1); if (ret < 0 || ret == sizeof(session->rotation_chunk.active_tracing_path)) { ERR("Failed to format active kernel tracing path in rotate session command"); - ret = -LTTNG_ERR_UNK; + cmd_ret = LTTNG_ERR_UNK; goto error; } /* @@ -4490,10 +4629,10 @@ int cmd_rotate_session(struct ltt_session *session, ret = snprintf(session->kernel_session->consumer->chunk_path, sizeof(session->kernel_session->consumer->chunk_path), "/%s-%" PRIu64, datetime, - session->rotate_count + 1); + session->current_archive_id + 1); if (ret < 0 || ret == sizeof(session->kernel_session->consumer->chunk_path)) { ERR("Failed to format the kernel consumer's sub-directory in rotate session command"); - ret = -LTTNG_ERR_UNK; + cmd_ret = LTTNG_ERR_UNK; goto error; } /* @@ -4505,11 +4644,12 @@ int cmd_rotate_session(struct ltt_session *session, session->kernel_session->gid); if (ret) { ERR("Failed to create kernel session tracing path at %s", - session->kernel_session->chunk_path); + session->kernel_session->consumer->chunk_path); + cmd_ret = LTTNG_ERR_CREATE_DIR_FAIL; goto error; } - ret = kernel_rotate_session(session); - if (ret != LTTNG_OK) { + cmd_ret = kernel_rotate_session(session); + if (cmd_ret != LTTNG_OK) { goto error; } } @@ -4517,18 +4657,18 @@ int cmd_rotate_session(struct ltt_session *session, ret = snprintf(session->rotation_chunk.active_tracing_path, PATH_MAX, "%s/%s-%" PRIu64, session_get_base_path(session), - datetime, session->rotate_count + 1); + datetime, session->current_archive_id + 1); if (ret < 0) { ERR("Failed to format active UST tracing path in rotate session command"); - ret = -LTTNG_ERR_UNK; + cmd_ret = LTTNG_ERR_UNK; goto error; } ret = snprintf(session->ust_session->consumer->chunk_path, PATH_MAX, "/%s-%" PRIu64, datetime, - session->rotate_count + 1); + session->current_archive_id + 1); if (ret < 0) { ERR("Failed to format the UST consumer's sub-directory in rotate session command"); - ret = -LTTNG_ERR_UNK; + cmd_ret = LTTNG_ERR_UNK; goto error; } /* @@ -4538,155 +4678,343 @@ int cmd_rotate_session(struct ltt_session *session, ret = domain_mkdir(session->ust_session->consumer, session, session->ust_session->uid, session->ust_session->gid); - ret = ust_app_rotate_session(session, &ust_active); - if (ret != LTTNG_OK) { + if (ret) { + cmd_ret = LTTNG_ERR_CREATE_DIR_FAIL; goto error; } - /* - * Handle the case where we did not start a rotation on any channel. - * The consumer will never wake up the rotation thread to perform the - * rename, so we have to do it here while we hold the session and - * session_list locks. - */ - if (!session->kernel_session && !ust_active) { - ret = rename_complete_chunk(session, now); - if (ret < 0) { - ERR("Failed to rename completed rotation chunk"); - goto end; - } - session->rotate_pending = false; - session->rotation_status = LTTNG_ROTATION_STATUS_COMPLETED; + cmd_ret = ust_app_rotate_session(session); + if (cmd_ret != LTTNG_OK) { + goto error; } } + ret = timer_session_rotation_pending_check_start(session, + DEFAULT_ROTATE_PENDING_TIMER); + if (ret) { + cmd_ret = LTTNG_ERR_UNK; + goto error; + } + if (!session->active) { session->rotated_after_last_stop = true; } if (rotate_return) { - (*rotate_return)->rotate_id = session->rotate_count; - (*rotate_return)->status = LTTNG_ROTATION_STATUS_STARTED; + rotate_return->rotation_id = session->current_archive_id; } - - DBG("Cmd rotate session %s, rotate_id %" PRIu64 " completed", session->name, - session->rotate_count); - ret = LTTNG_OK; - - goto end; - -error: - if (rotate_return) { - (*rotate_return)->status = LTTNG_ROTATION_STATUS_ERROR; + ret = notification_thread_command_session_rotation_ongoing( + notification_thread_handle, + session->name, session->uid, session->gid, + session->current_archive_id - 1); + if (ret != LTTNG_OK) { + ERR("Failed to notify notification thread that a session rotation is ongoing for session %s", + session->name); + cmd_ret = ret; } + + DBG("Cmd rotate session %s, archive_id %" PRIu64 " sent", + session->name, session->current_archive_id - 1); end: + ret = (cmd_ret == LTTNG_OK) ? cmd_ret : -((int) cmd_ret); return ret; +error: + session->last_chunk_start_ts = original_last_chunk_start_ts; + session->current_archive_id = original_current_chunk_start_ts; + if (session_reset_rotation_state(session, + LTTNG_ROTATION_STATE_NO_ROTATION)) { + ERR("Failed to reset rotation state of session \"%s\"", + session->name); + } + goto end; } /* - * Command LTTNG_ROTATE_PENDING from the lttng-ctl library. + * Command LTTNG_ROTATION_GET_INFO from the lttng-ctl library. * * Check if the session has finished its rotation. * * Return 0 on success or else a LTTNG_ERR code. */ -int cmd_rotate_pending(struct ltt_session *session, - struct lttng_rotate_pending_return **pending_return, - uint64_t rotate_id) +int cmd_rotate_get_info(struct ltt_session *session, + struct lttng_rotation_get_info_return *info_return, + uint64_t rotation_id) { int ret; assert(session); - DBG("Cmd rotate pending session %s, rotate_id %" PRIu64, session->name, - session->rotate_count); - - *pending_return = zmalloc(sizeof(struct lttng_rotate_pending_return)); - if (!*pending_return) { - ret = -ENOMEM; - goto end; - } + DBG("Cmd rotate_get_info session %s, rotation id %" PRIu64, session->name, + session->current_archive_id); - if (session->rotate_count != rotate_id) { - (*pending_return)->status = LTTNG_ROTATION_STATUS_EXPIRED; + if (session->current_archive_id != rotation_id) { + info_return->status = (int32_t) LTTNG_ROTATION_STATE_EXPIRED; ret = LTTNG_OK; goto end; } - if (session->rotation_status == LTTNG_ROTATION_STATUS_ERROR) { - DBG("An error occurred during rotation"); - (*pending_return)->status = LTTNG_ROTATION_STATUS_ERROR; - /* Rotate with a relay */ - } else if (session->rotate_pending_relay) { - DBG("Session %s, rotate_id %" PRIu64 " still pending", - session->name, session->rotate_count); - (*pending_return)->status = LTTNG_ROTATION_STATUS_STARTED; - } else if (session->rotate_pending) { - DBG("Session %s, rotate_id %" PRIu64 " still pending", - session->name, session->rotate_count); - (*pending_return)->status = LTTNG_ROTATION_STATUS_STARTED; - } else { - DBG("Session %s, rotate_id %" PRIu64 " finished", - session->name, session->rotate_count); - (*pending_return)->status = LTTNG_ROTATION_STATUS_COMPLETED; - ret = lttng_strncpy((*pending_return)->output_path, + switch (session->rotation_state) { + case LTTNG_ROTATION_STATE_ONGOING: + DBG("Reporting that rotation id %" PRIu64 " of session %s is still pending", + rotation_id, session->name); + break; + case LTTNG_ROTATION_STATE_COMPLETED: + { + char *current_tracing_path_reply; + size_t current_tracing_path_reply_len; + + switch (session_get_consumer_destination_type(session)) { + case CONSUMER_DST_LOCAL: + current_tracing_path_reply = + info_return->location.local.absolute_path; + current_tracing_path_reply_len = + sizeof(info_return->location.local.absolute_path); + info_return->location_type = + (int8_t) LTTNG_TRACE_ARCHIVE_LOCATION_TYPE_LOCAL; + break; + case CONSUMER_DST_NET: + current_tracing_path_reply = + info_return->location.relay.relative_path; + current_tracing_path_reply_len = + sizeof(info_return->location.relay.relative_path); + /* Currently the only supported relay protocol. */ + info_return->location.relay.protocol = + (int8_t) LTTNG_TRACE_ARCHIVE_LOCATION_RELAY_PROTOCOL_TYPE_TCP; + + ret = lttng_strncpy(info_return->location.relay.host, + session_get_net_consumer_hostname(session), + sizeof(info_return->location.relay.host)); + if (ret) { + ERR("Failed to host name to rotate_get_info reply"); + info_return->status = LTTNG_ROTATION_STATUS_ERROR; + ret = -LTTNG_ERR_UNK; + goto end; + } + + session_get_net_consumer_ports(session, + &info_return->location.relay.ports.control, + &info_return->location.relay.ports.data); + info_return->location_type = + (int8_t) LTTNG_TRACE_ARCHIVE_LOCATION_TYPE_RELAY; + break; + default: + abort(); + } + ret = lttng_strncpy(current_tracing_path_reply, session->rotation_chunk.current_rotate_path, - sizeof((*pending_return)->output_path)); + current_tracing_path_reply_len); if (ret) { - ERR("Failed to copy active tracing path to rotate pending command reply"); - (*pending_return)->status = LTTNG_ROTATION_STATUS_ERROR; - ret = -1; + ERR("Failed to copy current tracing path to rotate_get_info reply"); + info_return->status = LTTNG_ROTATION_STATUS_ERROR; + ret = -LTTNG_ERR_UNK; goto end; } + + break; + } + case LTTNG_ROTATION_STATE_ERROR: + DBG("Reporting that an error occurred during rotation %" PRIu64 " of session %s", + rotation_id, session->name); + break; + default: + abort(); } + info_return->status = (int32_t) session->rotation_state; ret = LTTNG_OK; - - goto end; - end: return ret; } /* - * Command ROTATE_GET_CURRENT_PATH from the lttng-ctl library. + * Command LTTNG_ROTATION_SET_SCHEDULE from the lttng-ctl library. * * Configure the automatic rotation parameters. - * Set to -1ULL to disable them. + * 'activate' to true means activate the rotation schedule type with 'new_value'. + * 'activate' to false means deactivate the rotation schedule and validate that + * 'new_value' has the same value as the currently active value. * - * Return LTTNG_OK on success or else a LTTNG_ERR code. + * Return 0 on success or else a positive LTTNG_ERR code. */ -int cmd_rotate_get_current_path(struct ltt_session *session, - struct lttng_rotate_get_current_path **get_return) +int cmd_rotation_set_schedule(struct ltt_session *session, + bool activate, enum lttng_rotation_schedule_type schedule_type, + uint64_t new_value, + struct notification_thread_handle *notification_thread_handle) { int ret; + uint64_t *parameter_value; + + assert(session); + + DBG("Cmd rotate set schedule session %s", session->name); - *get_return = zmalloc(sizeof(struct lttng_rotate_get_current_path)); - if (!*get_return) { - ret = -ENOMEM; + if (session->live_timer || !session->output_traces) { + DBG("Failing ROTATION_SET_SCHEDULE command as the rotation feature is not available for this session"); + ret = LTTNG_ERR_ROTATION_NOT_AVAILABLE; goto end; } - if (session->rotate_count == 0) { - (*get_return)->status = LTTNG_ROTATION_STATUS_NO_ROTATION; - } else { - (*get_return)->status = session->rotation_status; - ret = lttng_strncpy((*get_return)->output_path, - session->rotation_chunk.current_rotate_path, - sizeof((*get_return)->output_path)); - if (ret) { - ERR("Failed to copy trace output path to rotate get current path command reply"); - ret = -1; + switch (schedule_type) { + case LTTNG_ROTATION_SCHEDULE_TYPE_SIZE_THRESHOLD: + parameter_value = &session->rotate_size; + break; + case LTTNG_ROTATION_SCHEDULE_TYPE_PERIODIC: + parameter_value = &session->rotate_timer_period; + if (new_value >= UINT_MAX) { + DBG("Failing ROTATION_SET_SCHEDULE command as the value requested for a periodic rotation schedule is invalid: %" PRIu64 " > %u (UINT_MAX)", + new_value, UINT_MAX); + ret = LTTNG_ERR_INVALID; goto end; } + break; + default: + WARN("Failing ROTATION_SET_SCHEDULE command on unknown schedule type"); + ret = LTTNG_ERR_INVALID; + goto end; + } + + /* Improper use of the API. */ + if (new_value == -1ULL) { + WARN("Failing ROTATION_SET_SCHEDULE command as the value requested is -1"); + ret = LTTNG_ERR_INVALID; + goto end; + } + + /* + * As indicated in struct ltt_session's comments, a value of == 0 means + * this schedule rotation type is not in use. + * + * Reject the command if we were asked to activate a schedule that was + * already active. + */ + if (activate && *parameter_value != 0) { + DBG("Failing ROTATION_SET_SCHEDULE (activate) command as the schedule is already active"); + ret = LTTNG_ERR_ROTATION_SCHEDULE_SET; + goto end; + } + + /* + * Reject the command if we were asked to deactivate a schedule that was + * not active. + */ + if (!activate && *parameter_value == 0) { + DBG("Failing ROTATION_SET_SCHEDULE (deactivate) command as the schedule is already inactive"); + ret = LTTNG_ERR_ROTATION_SCHEDULE_NOT_SET; + goto end; + } + + /* + * Reject the command if we were asked to deactivate a schedule that + * doesn't exist. + */ + if (!activate && *parameter_value != new_value) { + DBG("Failing ROTATION_SET_SCHEDULE (deactivate) command as an inexistant schedule was provided"); + ret = LTTNG_ERR_ROTATION_SCHEDULE_NOT_SET; + goto end; + } + + *parameter_value = activate ? new_value : 0; + + switch (schedule_type) { + case LTTNG_ROTATION_SCHEDULE_TYPE_PERIODIC: + if (activate && session->active) { + /* + * Only start the timer if the session is active, + * otherwise it will be started when the session starts. + */ + ret = timer_session_rotation_schedule_timer_start( + session, new_value); + if (ret) { + ERR("Failed to enable session rotation timer in ROTATION_SET_SCHEDULE command"); + ret = LTTNG_ERR_UNK; + goto end; + } + } else { + ret = timer_session_rotation_schedule_timer_stop( + session); + if (ret) { + ERR("Failed to disable session rotation timer in ROTATION_SET_SCHEDULE command"); + ret = LTTNG_ERR_UNK; + goto end; + } + } + break; + case LTTNG_ROTATION_SCHEDULE_TYPE_SIZE_THRESHOLD: + if (activate) { + ret = subscribe_session_consumed_size_rotation(session, + new_value, notification_thread_handle); + if (ret) { + ERR("Failed to enable consumed-size notification in ROTATION_SET_SCHEDULE command"); + ret = LTTNG_ERR_UNK; + goto end; + } + } else { + ret = unsubscribe_session_consumed_size_rotation(session, + notification_thread_handle); + if (ret) { + ERR("Failed to disable consumed-size notification in ROTATION_SET_SCHEDULE command"); + ret = LTTNG_ERR_UNK; + goto end; + } + + } + break; + default: + /* Would have been caught before. */ + abort(); } ret = LTTNG_OK; + goto end; + end: return ret; } +/* Wait for a given path to be removed before continuing. */ +static enum lttng_error_code wait_on_path(void *path_data) +{ + const char *shm_path = path_data; + + DBG("Waiting for the shm path at %s to be removed before completing session destruction", + shm_path); + while (true) { + int ret; + struct stat st; + + ret = stat(shm_path, &st); + if (ret) { + if (errno != ENOENT) { + PERROR("stat() returned an error while checking for the existence of the shm path"); + } else { + DBG("shm path no longer exists, completing the destruction of session"); + } + break; + } else { + if (!S_ISDIR(st.st_mode)) { + ERR("The type of shm path %s returned by stat() is not a directory; aborting the wait for shm path removal", + shm_path); + break; + } + } + usleep(SESSION_DESTROY_SHM_PATH_CHECK_DELAY_US); + } + return LTTNG_OK; +} + +/* + * Returns a pointer to a handler to run on completion of a command. + * Returns NULL if no handler has to be run for the last command executed. + */ +const struct cmd_completion_handler *cmd_pop_completion_handler(void) +{ + struct cmd_completion_handler *handler = current_completion_handler; + + current_completion_handler = NULL; + return handler; +} + /* * Init command subsystem. */