X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fcmd.c;h=5a122d8d3c8dcf7e7f361b34f69f46c4680c0d22;hp=01efcaf05371fc178e14f13ff24c2aa1368053d8;hb=1f4962443;hpb=b178f53e90c376dd44b020535c32649edef8f80e diff --git a/src/bin/lttng-sessiond/cmd.c b/src/bin/lttng-sessiond/cmd.c index 01efcaf05..5a122d8d3 100644 --- a/src/bin/lttng-sessiond/cmd.c +++ b/src/bin/lttng-sessiond/cmd.c @@ -22,6 +22,7 @@ #include #include #include +#include #include #include @@ -32,6 +33,8 @@ #include #include #include +#include +#include #include #include #include @@ -67,6 +70,16 @@ /* Sleep for 100ms between each check for the shm path's deletion. */ #define SESSION_DESTROY_SHM_PATH_CHECK_DELAY_US 100000 +struct cmd_destroy_session_reply_context { + int reply_sock_fd; + bool implicit_rotation_on_destroy; + /* + * Indicates whether or not an error occurred while launching the + * destruction of a session. + */ + enum lttng_error_code destruction_status; +}; + static enum lttng_error_code wait_on_path(void *path); /* @@ -99,7 +112,7 @@ static uint64_t relayd_net_seq_idx; static int validate_ust_event_name(const char *); static int cmd_enable_event_internal(struct ltt_session *session, - struct lttng_domain *domain, + const struct lttng_domain *domain, char *channel_name, struct lttng_event *event, char *filter_expression, struct lttng_filter_bytecode *filter, @@ -904,6 +917,10 @@ static enum lttng_error_code add_uri_to_consumer( ret = lttng_strncpy(consumer->dst.session_root_path, uri->dst.path, sizeof(consumer->dst.session_root_path)); + if (ret) { + ret_code = LTTNG_ERR_FATAL; + goto error; + } consumer->type = CONSUMER_DST_LOCAL; break; } @@ -1029,7 +1046,11 @@ static enum lttng_error_code send_consumer_relayd_socket( struct lttng_uri *relayd_uri, struct consumer_output *consumer, struct consumer_socket *consumer_sock, - char *session_name, char *hostname, int session_live_timer) + const char *session_name, const char *hostname, + const char *base_path, int session_live_timer, + const uint64_t *current_chunk_id, + time_t session_creation_time, + bool session_name_contains_creation_time) { int ret; struct lttcomm_relayd_sock *rsock = NULL; @@ -1057,7 +1078,9 @@ static enum lttng_error_code send_consumer_relayd_socket( /* Send relayd socket to consumer. */ ret = consumer_send_relayd_socket(consumer_sock, rsock, consumer, relayd_uri->stype, session_id, - session_name, hostname, session_live_timer); + session_name, hostname, base_path, + session_live_timer, current_chunk_id, + session_creation_time, session_name_contains_creation_time); if (ret < 0) { status = LTTNG_ERR_ENABLE_CONSUMER_FAIL; goto close_sock; @@ -1103,8 +1126,10 @@ relayd_comm_error: 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) + struct consumer_socket *sock, const char *session_name, + const char *hostname, const char *base_path, int session_live_timer, + const uint64_t *current_chunk_id, time_t session_creation_time, + bool session_name_contains_creation_time) { enum lttng_error_code status = LTTNG_OK; @@ -1115,7 +1140,9 @@ static enum lttng_error_code send_consumer_relayd_sockets( if (!sock->control_sock_sent) { status = send_consumer_relayd_socket(session_id, &consumer->dst.net.control, consumer, sock, - session_name, hostname, session_live_timer); + session_name, hostname, base_path, session_live_timer, + current_chunk_id, session_creation_time, + session_name_contains_creation_time); if (status != LTTNG_OK) { goto error; } @@ -1125,7 +1152,9 @@ static enum lttng_error_code send_consumer_relayd_sockets( if (!sock->data_sock_sent) { status = send_consumer_relayd_socket(session_id, &consumer->dst.net.data, consumer, sock, - session_name, hostname, session_live_timer); + session_name, hostname, base_path, session_live_timer, + current_chunk_id, session_creation_time, + session_name_contains_creation_time); if (status != LTTNG_OK) { goto error; } @@ -1147,14 +1176,28 @@ int cmd_setup_relayd(struct ltt_session *session) struct ltt_kernel_session *ksess; struct consumer_socket *socket; struct lttng_ht_iter iter; + LTTNG_OPTIONAL(uint64_t) current_chunk_id = {}; - assert(session); + assert(session); usess = session->ust_session; ksess = session->kernel_session; DBG("Setting relayd for session %s", session->name); + if (session->current_trace_chunk) { + enum lttng_trace_chunk_status status = lttng_trace_chunk_get_id( + session->current_trace_chunk, ¤t_chunk_id.value); + + if (status == LTTNG_TRACE_CHUNK_STATUS_OK) { + current_chunk_id.is_set = true; + } else { + ERR("Failed to get current trace chunk id"); + ret = LTTNG_ERR_UNK; + goto error; + } + } + rcu_read_lock(); if (usess && usess->consumer && usess->consumer->type == CONSUMER_DST_NET @@ -1166,7 +1209,11 @@ int cmd_setup_relayd(struct ltt_session *session) ret = send_consumer_relayd_sockets(LTTNG_DOMAIN_UST, session->id, usess->consumer, socket, session->name, session->hostname, - session->live_timer); + session->base_path, + session->live_timer, + current_chunk_id.is_set ? ¤t_chunk_id.value : NULL, + session->creation_time, + session->name_contains_creation_time); pthread_mutex_unlock(socket->lock); if (ret != LTTNG_OK) { goto error; @@ -1188,7 +1235,11 @@ int cmd_setup_relayd(struct ltt_session *session) ret = send_consumer_relayd_sockets(LTTNG_DOMAIN_KERNEL, session->id, ksess->consumer, socket, session->name, session->hostname, - session->live_timer); + session->base_path, + session->live_timer, + current_chunk_id.is_set ? ¤t_chunk_id.value : NULL, + session->creation_time, + session->name_contains_creation_time); pthread_mutex_unlock(socket->lock); if (ret != LTTNG_OK) { goto error; @@ -1210,7 +1261,7 @@ error: /* * Start a kernel session by opening all necessary streams. */ -static int start_kernel_session(struct ltt_kernel_session *ksess, int wpipe) +static int start_kernel_session(struct ltt_kernel_session *ksess) { int ret; struct ltt_kernel_channel *kchan; @@ -1262,7 +1313,7 @@ static int start_kernel_session(struct ltt_kernel_session *ksess, int wpipe) } /* Quiescent wait after starting trace */ - kernel_wait_quiescent(wpipe); + kernel_wait_quiescent(); ksess->active = 1; @@ -1294,7 +1345,7 @@ int cmd_disable_channel(struct ltt_session *session, goto error; } - kernel_wait_quiescent(kernel_tracer_fd); + kernel_wait_quiescent(); break; } case LTTNG_DOMAIN_UST: @@ -1352,7 +1403,7 @@ int cmd_track_pid(struct ltt_session *session, enum lttng_domain_type domain, goto error; } - kernel_wait_quiescent(kernel_tracer_fd); + kernel_wait_quiescent(); break; } case LTTNG_DOMAIN_UST: @@ -1403,7 +1454,7 @@ int cmd_untrack_pid(struct ltt_session *session, enum lttng_domain_type domain, goto error; } - kernel_wait_quiescent(kernel_tracer_fd); + kernel_wait_quiescent(); break; } case LTTNG_DOMAIN_UST: @@ -1436,27 +1487,29 @@ error: * The wpipe arguments is used as a notifier for the kernel thread. */ int cmd_enable_channel(struct ltt_session *session, - struct lttng_domain *domain, struct lttng_channel *attr, int wpipe) + const struct lttng_domain *domain, const struct lttng_channel *_attr, int wpipe) { int ret; struct ltt_ust_session *usess = session->ust_session; struct lttng_ht *chan_ht; size_t len; + struct lttng_channel attr; assert(session); - assert(attr); + assert(_attr); assert(domain); - len = lttng_strnlen(attr->name, sizeof(attr->name)); + attr = *_attr; + len = lttng_strnlen(attr.name, sizeof(attr.name)); /* Validate channel name */ - if (attr->name[0] == '.' || - memchr(attr->name, '/', len) != NULL) { + if (attr.name[0] == '.' || + memchr(attr.name, '/', len) != NULL) { ret = LTTNG_ERR_INVALID_CHANNEL_NAME; goto end; } - DBG("Enabling channel %s for session %s", attr->name, session->name); + DBG("Enabling channel %s for session %s", attr.name, session->name); rcu_read_lock(); @@ -1475,21 +1528,21 @@ int cmd_enable_channel(struct ltt_session *session, * beacons for inactive streams. */ if (session->live_timer > 0) { - attr->attr.live_timer_interval = session->live_timer; - attr->attr.switch_timer_interval = 0; + attr.attr.live_timer_interval = session->live_timer; + attr.attr.switch_timer_interval = 0; } /* Check for feature support */ switch (domain->type) { case LTTNG_DOMAIN_KERNEL: { - if (kernel_supports_ring_buffer_snapshot_sample_positions(kernel_tracer_fd) != 1) { + if (kernel_supports_ring_buffer_snapshot_sample_positions() != 1) { /* Sampling position of buffer is not supported */ WARN("Kernel tracer does not support buffer monitoring. " "Setting the monitor interval timer to 0 " "(disabled) for channel '%s' of session '%s'", - attr-> name, session->name); - lttng_channel_set_monitor_timer_interval(attr, 0); + attr.name, session->name); + lttng_channel_set_monitor_timer_interval(&attr, 0); } break; } @@ -1514,16 +1567,16 @@ int cmd_enable_channel(struct ltt_session *session, { struct ltt_kernel_channel *kchan; - kchan = trace_kernel_get_channel_by_name(attr->name, + kchan = trace_kernel_get_channel_by_name(attr.name, session->kernel_session); if (kchan == NULL) { if (session->snapshot.nb_output > 0 || session->snapshot_mode) { /* Enforce mmap output for snapshot sessions. */ - attr->attr.output = LTTNG_EVENT_MMAP; + attr.attr.output = LTTNG_EVENT_MMAP; } - ret = channel_kernel_create(session->kernel_session, attr, wpipe); - if (attr->name[0] != '\0') { + ret = channel_kernel_create(session->kernel_session, &attr, wpipe); + if (attr.name[0] != '\0') { session->kernel_session->has_non_default_channel = 1; } } else { @@ -1534,7 +1587,7 @@ int cmd_enable_channel(struct ltt_session *session, goto error; } - kernel_wait_quiescent(kernel_tracer_fd); + kernel_wait_quiescent(); break; } case LTTNG_DOMAIN_UST: @@ -1553,19 +1606,19 @@ int cmd_enable_channel(struct ltt_session *session, * adhered to. */ if (domain->type == LTTNG_DOMAIN_JUL) { - if (strncmp(attr->name, DEFAULT_JUL_CHANNEL_NAME, + if (strncmp(attr.name, DEFAULT_JUL_CHANNEL_NAME, LTTNG_SYMBOL_NAME_LEN)) { ret = LTTNG_ERR_INVALID_CHANNEL_NAME; goto error; } } else if (domain->type == LTTNG_DOMAIN_LOG4J) { - if (strncmp(attr->name, DEFAULT_LOG4J_CHANNEL_NAME, + if (strncmp(attr.name, DEFAULT_LOG4J_CHANNEL_NAME, LTTNG_SYMBOL_NAME_LEN)) { ret = LTTNG_ERR_INVALID_CHANNEL_NAME; goto error; } } else if (domain->type == LTTNG_DOMAIN_PYTHON) { - if (strncmp(attr->name, DEFAULT_PYTHON_CHANNEL_NAME, + if (strncmp(attr.name, DEFAULT_PYTHON_CHANNEL_NAME, LTTNG_SYMBOL_NAME_LEN)) { ret = LTTNG_ERR_INVALID_CHANNEL_NAME; goto error; @@ -1574,10 +1627,10 @@ int cmd_enable_channel(struct ltt_session *session, chan_ht = usess->domain_global.channels; - uchan = trace_ust_find_channel_by_name(chan_ht, attr->name); + uchan = trace_ust_find_channel_by_name(chan_ht, attr.name); if (uchan == NULL) { - ret = channel_ust_create(usess, attr, domain->buf_type); - if (attr->name[0] != '\0') { + ret = channel_ust_create(usess, &attr, domain->buf_type); + if (attr.name[0] != '\0') { usess->has_non_default_channel = 1; } } else { @@ -1590,7 +1643,7 @@ int cmd_enable_channel(struct ltt_session *session, goto error; } - if (ret == LTTNG_OK && attr->attr.output != LTTNG_EVENT_MMAP) { + if (ret == LTTNG_OK && attr.attr.output != LTTNG_EVENT_MMAP) { session->has_non_mmap_channel = true; } error: @@ -1603,11 +1656,11 @@ end: * Command LTTNG_DISABLE_EVENT processed by the client thread. */ int cmd_disable_event(struct ltt_session *session, - enum lttng_domain_type domain, char *channel_name, - struct lttng_event *event) + enum lttng_domain_type domain, const char *channel_name, + const struct lttng_event *event) { int ret; - char *event_name; + const char *event_name; DBG("Disable event command for event \'%s\'", event->name); @@ -1669,7 +1722,7 @@ int cmd_disable_event(struct ltt_session *session, goto error_unlock; } - kernel_wait_quiescent(kernel_tracer_fd); + kernel_wait_quiescent(); break; } case LTTNG_DOMAIN_UST: @@ -1780,7 +1833,7 @@ error: * Command LTTNG_ADD_CONTEXT processed by the client thread. */ int cmd_add_context(struct ltt_session *session, enum lttng_domain_type domain, - char *channel_name, struct lttng_event_context *ctx, int kwpipe) + char *channel_name, const struct lttng_event_context *ctx, int kwpipe) { int ret, chan_kern_created = 0, chan_ust_created = 0; char *app_ctx_provider_name = NULL, *app_ctx_name = NULL; @@ -1951,7 +2004,7 @@ end: * enable the events through which all "agent" events are funeled. */ static int _cmd_enable_event(struct ltt_session *session, - struct lttng_domain *domain, + const struct lttng_domain *domain, char *channel_name, struct lttng_event *event, char *filter_expression, struct lttng_filter_bytecode *filter, @@ -2119,7 +2172,7 @@ static int _cmd_enable_event(struct ltt_session *session, goto error; } - kernel_wait_quiescent(kernel_tracer_fd); + kernel_wait_quiescent(); break; } case LTTNG_DOMAIN_UST: @@ -2348,7 +2401,8 @@ error: * Command LTTNG_ENABLE_EVENT processed by the client thread. * We own filter, exclusion, and filter_expression. */ -int cmd_enable_event(struct ltt_session *session, struct lttng_domain *domain, +int cmd_enable_event(struct ltt_session *session, + const struct lttng_domain *domain, char *channel_name, struct lttng_event *event, char *filter_expression, struct lttng_filter_bytecode *filter, @@ -2365,7 +2419,7 @@ int cmd_enable_event(struct ltt_session *session, struct lttng_domain *domain, * reserved names. */ static int cmd_enable_event_internal(struct ltt_session *session, - struct lttng_domain *domain, + const struct lttng_domain *domain, char *channel_name, struct lttng_event *event, char *filter_expression, struct lttng_filter_bytecode *filter, @@ -2387,7 +2441,7 @@ ssize_t cmd_list_tracepoints(enum lttng_domain_type domain, switch (domain) { case LTTNG_DOMAIN_KERNEL: - nb_events = kernel_list_events(kernel_tracer_fd, events); + nb_events = kernel_list_events(events); if (nb_events < 0) { ret = LTTNG_ERR_KERN_LIST_FAIL; goto error; @@ -2507,106 +2561,6 @@ error: return -ret; } -static -int domain_mkdir(const struct consumer_output *output, - const struct ltt_session *session, - uid_t uid, gid_t gid) -{ - struct consumer_socket *socket; - struct lttng_ht_iter iter; - int ret; - char path[LTTNG_PATH_MAX]; - - if (!output || !output->socks) { - ERR("No consumer output found"); - ret = -1; - goto end; - } - - ret = snprintf(path, sizeof(path), "%s/%s%s", - session_get_base_path(session), - output->chunk_path, - output->domain_subdir); - if (ret < 0 || ret >= LTTNG_PATH_MAX) { - ERR("Failed to format path new chunk domain path"); - ret = -1; - goto end; - } - - DBG("Domain mkdir %s for session %" PRIu64, path, session->id); - rcu_read_lock(); - /* - * We have to iterate to find a socket, but we only need to send the - * rename command to one consumer, so we break after the first one. - */ - cds_lfht_for_each_entry(output->socks->ht, &iter.iter, socket, node.node) { - pthread_mutex_lock(socket->lock); - ret = consumer_mkdir(socket, session->id, output, path, uid, gid); - pthread_mutex_unlock(socket->lock); - if (ret) { - ERR("Failed to create directory at \"%s\"", path); - ret = -1; - goto end_unlock; - } - break; - } - - ret = 0; - -end_unlock: - rcu_read_unlock(); -end: - return ret; -} - -static -int session_mkdir(const struct ltt_session *session) -{ - int ret; - struct consumer_output *output; - uid_t uid; - gid_t gid; - - /* - * Unsupported feature in lttng-relayd before 2.11, not an error since it - * is only needed for session rotation and the user will get an error - * on rotate. - */ - if (session->consumer->type == CONSUMER_DST_NET && - session->consumer->relay_major_version == 2 && - session->consumer->relay_minor_version < 11) { - ret = 0; - goto end; - } - - if (session->kernel_session) { - output = session->kernel_session->consumer; - uid = session->kernel_session->uid; - gid = session->kernel_session->gid; - ret = domain_mkdir(output, session, uid, gid); - if (ret) { - ERR("Mkdir kernel"); - goto end; - } - } - - if (session->ust_session) { - output = session->ust_session->consumer; - uid = session->ust_session->uid; - gid = session->ust_session->gid; - ret = domain_mkdir(output, session, uid, gid); - if (ret) { - ERR("Mkdir UST"); - goto end; - } - } - - ret = 0; - -end: - return ret; -} - /* * Command LTTNG_START_TRACE processed by the client thread. * @@ -2614,10 +2568,12 @@ end: */ int cmd_start_trace(struct ltt_session *session) { - int ret; + enum lttng_error_code ret; unsigned long nb_chan = 0; struct ltt_kernel_session *ksession; struct ltt_ust_session *usess; + const bool session_rotated_after_last_stop = + session->rotated_after_last_stop; assert(session); @@ -2631,6 +2587,22 @@ int cmd_start_trace(struct ltt_session *session) goto error; } + if (session->rotation_state == LTTNG_ROTATION_STATE_ONGOING && + !session->current_trace_chunk) { + /* + * A rotation was launched while the session was stopped and + * it has not been completed yet. It is not possible to start + * the session since starting the session here would require a + * rotation from "NULL" to a new trace chunk. That rotation + * would overlap with the ongoing rotation, which is not + * supported. + */ + WARN("Refusing to start session \"%s\" as a rotation launched after the last \"stop\" is still ongoing", + session->name); + ret = LTTNG_ERR_ROTATION_PENDING; + goto error; + } + /* * Starting a session without channel is useless since after that it's not * possible to enable channel thus inform the client. @@ -2646,32 +2618,52 @@ int cmd_start_trace(struct ltt_session *session) goto error; } - /* - * Record the timestamp of the first time the session is started for - * an eventual session rotation call. - */ - if (!session->has_been_started) { - session->current_chunk_start_ts = time(NULL); - if (session->current_chunk_start_ts == (time_t) -1) { - PERROR("Failed to retrieve the \"%s\" session's start time", + session->active = 1; + session->rotated_after_last_stop = false; + if (session->output_traces && !session->current_trace_chunk) { + if (!session->has_been_started) { + struct lttng_trace_chunk *trace_chunk; + + DBG("Creating initial trace chunk of session \"%s\"", session->name); - ret = LTTNG_ERR_FATAL; - goto error; - } - if (!session->snapshot_mode && session->output_traces) { - ret = session_mkdir(session); - if (ret) { - ERR("Failed to create the session directories"); + trace_chunk = session_create_new_trace_chunk( + session, NULL, NULL, NULL); + if (!trace_chunk) { ret = LTTNG_ERR_CREATE_DIR_FAIL; goto error; } + assert(!session->current_trace_chunk); + ret = session_set_trace_chunk(session, trace_chunk, + NULL); + lttng_trace_chunk_put(trace_chunk); + if (ret) { + ret = LTTNG_ERR_CREATE_TRACE_CHUNK_FAIL_CONSUMER; + goto error; + } + } else { + DBG("Rotating session \"%s\" from its current \"NULL\" trace chunk to a new chunk", + session->name); + /* + * Rotate existing streams into the new chunk. + * This is a "quiet" rotation has no client has + * explicitly requested this operation. + * + * There is also no need to wait for the rotation + * to complete as it will happen immediately. No data + * was produced as the session was stopped, so the + * rotation should happen on reception of the command. + */ + ret = cmd_rotate_session(session, NULL, true); + if (ret != LTTNG_OK) { + goto error; + } } } /* Kernel tracing */ if (ksession != NULL) { DBG("Start kernel tracing session %s", session->name); - ret = start_kernel_session(ksession, kernel_tracer_fd); + ret = start_kernel_session(ksession); if (ret != LTTNG_OK) { goto error; } @@ -2679,17 +2671,14 @@ int cmd_start_trace(struct ltt_session *session) /* Flag session that trace should start automatically */ if (usess) { - ret = ust_app_start_trace_all(usess); - if (ret < 0) { + int int_ret = ust_app_start_trace_all(usess); + + if (int_ret < 0) { ret = LTTNG_ERR_UST_START_FAIL; goto error; } } - /* Flag this after a successful start. */ - session->has_been_started = 1; - session->active = 1; - /* * Clear the flag that indicates that a rotation was done while the * session was stopped. @@ -2697,9 +2686,10 @@ int cmd_start_trace(struct ltt_session *session) session->rotated_after_last_stop = false; if (session->rotate_timer_period) { - ret = timer_session_rotation_schedule_timer_start(session, - session->rotate_timer_period); - if (ret < 0) { + int int_ret = timer_session_rotation_schedule_timer_start( + session, session->rotate_timer_period); + + if (int_ret < 0) { ERR("Failed to enable rotate timer"); ret = LTTNG_ERR_UNK; goto error; @@ -2709,6 +2699,15 @@ int cmd_start_trace(struct ltt_session *session) ret = LTTNG_OK; error: + if (ret == LTTNG_OK) { + /* Flag this after a successful start. */ + session->has_been_started |= 1; + } else { + session->active = 0; + /* Restore initial state on error. */ + session->rotated_after_last_stop = + session_rotated_after_last_stop; + } return ret; } @@ -2746,13 +2745,14 @@ int cmd_stop_trace(struct ltt_session *session) goto error; } - kernel_wait_quiescent(kernel_tracer_fd); + kernel_wait_quiescent(); /* Flush metadata after stopping (if exists) */ if (ksession->metadata_stream_fd >= 0) { ret = kernel_metadata_flush_buffer(ksession->metadata_stream_fd); if (ret < 0) { ERR("Kernel metadata flush failed"); + error_occurred = true; } } @@ -2761,6 +2761,7 @@ int cmd_stop_trace(struct ltt_session *session) ret = kernel_flush_buffer(kchan); if (ret < 0) { ERR("Kernel flush buffer error"); + error_occurred = true; } } @@ -2785,6 +2786,45 @@ error: return ret; } +/* + * Set the base_path of the session only if subdir of a control uris is set. + * Return LTTNG_OK on success, otherwise LTTNG_ERR_*. + */ +static int set_session_base_path_from_uris(struct ltt_session *session, + size_t nb_uri, + struct lttng_uri *uris) +{ + int ret; + size_t i; + + for (i = 0; i < nb_uri; i++) { + if (uris[i].stype != LTTNG_STREAM_CONTROL || + uris[i].subdir[0] == '\0') { + /* Not interested in these URIs */ + continue; + } + + if (session->base_path != NULL) { + free(session->base_path); + session->base_path = NULL; + } + + /* Set session base_path */ + session->base_path = strdup(uris[i].subdir); + if (!session->base_path) { + PERROR("Failed to copy base path \"%s\" to session \"%s\"", + uris[i].subdir, session->name); + ret = LTTNG_ERR_NOMEM; + goto error; + } + DBG2("Setting base path \"%s\" for session \"%s\"", + session->base_path, session->name); + } + ret = LTTNG_OK; +error: + return ret; +} + /* * Command LTTNG_SET_CONSUMER_URI processed by the client thread. */ @@ -2805,11 +2845,20 @@ int cmd_set_consumer_uri(struct ltt_session *session, size_t nb_uri, goto error; } + /* + * Set the session base path if any. This is done inside + * cmd_set_consumer_uri to preserve backward compatibility of the + * previous session creation api vs the session descriptor api. + */ + ret = set_session_base_path_from_uris(session, nb_uri, uris); + if (ret != LTTNG_OK) { + goto error; + } + /* Set the "global" consumer URIs */ for (i = 0; i < nb_uri; i++) { - ret = add_uri_to_consumer(session, - session->consumer, - &uris[i], LTTNG_DOMAIN_NONE); + ret = add_uri_to_consumer(session, session->consumer, &uris[i], + LTTNG_DOMAIN_NONE); if (ret != LTTNG_OK) { goto error; } @@ -2962,6 +3011,7 @@ enum lttng_error_code cmd_create_session_from_descriptor( ret_code = LTTNG_ERR_INVALID; goto end; } + ret_code = session_create(session_name, creds->uid, creds->gid, &new_session); if (ret_code != LTTNG_OK) { @@ -3102,26 +3152,136 @@ error: return ret_code; } +static +void cmd_destroy_session_reply(const struct ltt_session *session, + void *_reply_context) +{ + int ret; + ssize_t comm_ret; + const struct cmd_destroy_session_reply_context *reply_context = + _reply_context; + struct lttng_dynamic_buffer payload; + struct lttcomm_session_destroy_command_header cmd_header; + struct lttng_trace_archive_location *location = NULL; + struct lttcomm_lttng_msg llm = { + .cmd_type = LTTNG_DESTROY_SESSION, + .ret_code = reply_context->destruction_status, + .pid = UINT32_MAX, + .cmd_header_size = + sizeof(struct lttcomm_session_destroy_command_header), + .data_size = 0, + }; + size_t payload_size_before_location; + + lttng_dynamic_buffer_init(&payload); + + ret = lttng_dynamic_buffer_append(&payload, &llm, sizeof(llm)); + if (ret) { + ERR("Failed to append session destruction message"); + goto error; + } + + cmd_header.rotation_state = + (int32_t) (reply_context->implicit_rotation_on_destroy ? + session->rotation_state : + LTTNG_ROTATION_STATE_NO_ROTATION); + ret = lttng_dynamic_buffer_append(&payload, &cmd_header, + sizeof(cmd_header)); + if (ret) { + ERR("Failed to append session destruction command header"); + goto error; + } + + if (!reply_context->implicit_rotation_on_destroy) { + DBG("No implicit rotation performed during the destruction of session \"%s\", sending reply", + session->name); + goto send_reply; + } + if (session->rotation_state != LTTNG_ROTATION_STATE_COMPLETED) { + DBG("Rotation state of session \"%s\" is not \"completed\", sending session destruction reply", + session->name); + goto send_reply; + } + + location = session_get_trace_archive_location(session); + if (!location) { + ERR("Failed to get the location of the trace archive produced during the destruction of session \"%s\"", + session->name); + goto error; + } + + payload_size_before_location = payload.size; + comm_ret = lttng_trace_archive_location_serialize(location, + &payload); + if (comm_ret < 0) { + ERR("Failed to serialize the location of the trace archive produced during the destruction of session \"%s\"", + session->name); + goto error; + } + /* Update the message to indicate the location's length. */ + ((struct lttcomm_lttng_msg *) payload.data)->data_size = + payload.size - payload_size_before_location; +send_reply: + comm_ret = lttcomm_send_unix_sock(reply_context->reply_sock_fd, + payload.data, payload.size); + if (comm_ret != (ssize_t) payload.size) { + ERR("Failed to send result of the destruction of session \"%s\" to client", + session->name); + } +error: + ret = close(reply_context->reply_sock_fd); + if (ret) { + PERROR("Failed to close client socket in deferred session destroy reply"); + } + lttng_dynamic_buffer_reset(&payload); + free(_reply_context); +} + /* * Command LTTNG_DESTROY_SESSION processed by the client thread. * * Called with session lock held. */ int cmd_destroy_session(struct ltt_session *session, - struct notification_thread_handle *notification_thread_handle) + struct notification_thread_handle *notification_thread_handle, + int *sock_fd) { int ret; + enum lttng_error_code destruction_last_error = LTTNG_OK; + struct cmd_destroy_session_reply_context *reply_context = NULL; + + if (sock_fd) { + reply_context = zmalloc(sizeof(*reply_context)); + if (!reply_context) { + ret = LTTNG_ERR_NOMEM; + goto end; + } + reply_context->reply_sock_fd = *sock_fd; + } /* Safety net */ assert(session); - DBG("Begin destroy session %s (id %" PRIu64 ")", session->name, session->id); + DBG("Begin destroy session %s (id %" PRIu64 ")", session->name, + session->id); + if (session->active) { + DBG("Session \"%s\" is active, attempting to stop it before destroying it", + session->name); + ret = cmd_stop_trace(session); + if (ret != LTTNG_OK && ret != LTTNG_ERR_TRACE_ALREADY_STOPPED) { + /* Carry on with the destruction of the session. */ + ERR("Failed to stop session \"%s\" as part of its destruction: %s", + session->name, lttng_strerror(-ret)); + destruction_last_error = ret; + } + } 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); + destruction_last_error = LTTNG_ERR_TIMER_STOP_ERROR; } } @@ -3130,26 +3290,37 @@ int cmd_destroy_session(struct ltt_session *session, session->rotate_size = 0; } - 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); - } + if (session->most_recent_chunk_id.is_set && + session->most_recent_chunk_id.value != 0 && + session->current_trace_chunk && session->output_traces) { + /* + * Perform a last rotation on destruction if rotations have + * occurred during the session's lifetime. + */ + ret = cmd_rotate_session(session, NULL, false); + if (ret != LTTNG_OK) { + ERR("Failed to perform an implicit rotation as part of the destruction of session \"%s\": %s", + session->name, lttng_strerror(-ret)); + destruction_last_error = -ret; + } + if (reply_context) { + reply_context->implicit_rotation_on_destroy = true; + } + } else if (session->has_been_started && session->current_trace_chunk) { + /* + * The user has not triggered a session rotation. However, to + * ensure all data has been consumed, the session is rotated + * to a 'null' trace chunk before it is destroyed. + * + * This is a "quiet" rotation meaning that no notification is + * emitted and no renaming of the current trace chunk takes + * place. + */ + ret = cmd_rotate_session(session, NULL, true); + if (ret != LTTNG_OK) { + ERR("Failed to perform a quiet rotation as part of the destruction of session \"%s\": %s", + session->name, lttng_strerror(-ret)); + destruction_last_error = -ret; } } @@ -3213,8 +3384,20 @@ int cmd_destroy_session(struct ltt_session *session, * _at least_ up to the point when that reference is released. */ session_destroy(session); - ret = LTTNG_OK; - + if (reply_context) { + reply_context->destruction_status = destruction_last_error; + ret = session_add_destroy_notifier(session, + cmd_destroy_session_reply, + (void *) reply_context); + if (ret) { + ret = LTTNG_ERR_FATAL; + goto end; + } else { + *sock_fd = -1; + } + } + ret = LTTNG_OK; +end: return ret; } @@ -3633,7 +3816,7 @@ error: * Return LTTNG_OK on success or else a LTTNG_ERR code. */ int cmd_snapshot_add_output(struct ltt_session *session, - struct lttng_snapshot_output *output, uint32_t *id) + const struct lttng_snapshot_output *output, uint32_t *id) { int ret; struct snapshot_output *new_output; @@ -3701,7 +3884,7 @@ error: * Return LTTNG_OK on success or else a LTTNG_ERR code. */ int cmd_snapshot_del_output(struct ltt_session *session, - struct lttng_snapshot_output *output) + const struct lttng_snapshot_output *output) { int ret; struct snapshot_output *sout = NULL; @@ -4162,37 +4345,66 @@ end: * Return LTTNG_OK on success or a LTTNG_ERR code. */ static enum lttng_error_code set_relayd_for_snapshot( - struct consumer_output *consumer, - struct snapshot_output *snap_output, - struct ltt_session *session) + struct consumer_output *output, + const struct ltt_session *session) { enum lttng_error_code status = LTTNG_OK; struct lttng_ht_iter iter; struct consumer_socket *socket; + LTTNG_OPTIONAL(uint64_t) current_chunk_id = {}; + const char *base_path; - assert(consumer); - assert(snap_output); + assert(output); assert(session); DBG2("Set relayd object from snapshot output"); + if (session->current_trace_chunk) { + enum lttng_trace_chunk_status chunk_status = + lttng_trace_chunk_get_id( + session->current_trace_chunk, + ¤t_chunk_id.value); + + if (chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK) { + current_chunk_id.is_set = true; + } else { + ERR("Failed to get current trace chunk id"); + status = LTTNG_ERR_UNK; + goto error; + } + } + /* Ignore if snapshot consumer output is not network. */ - if (snap_output->consumer->type != CONSUMER_DST_NET) { + if (output->type != CONSUMER_DST_NET) { goto error; } + /* + * The snapshot record URI base path overrides the session + * base path. + */ + if (output->dst.net.control.subdir[0] != '\0') { + base_path = output->dst.net.control.subdir; + } else { + base_path = session->base_path; + } + /* * For each consumer socket, create and send the relayd object of the * snapshot output. */ rcu_read_lock(); - cds_lfht_for_each_entry(snap_output->consumer->socks->ht, &iter.iter, + cds_lfht_for_each_entry(output->socks->ht, &iter.iter, socket, node.node) { pthread_mutex_lock(socket->lock); status = send_consumer_relayd_sockets(0, session->id, - snap_output->consumer, socket, + output, socket, session->name, session->hostname, - session->live_timer); + base_path, + session->live_timer, + current_chunk_id.is_set ? ¤t_chunk_id.value : NULL, + session->creation_time, + session->name_contains_creation_time); pthread_mutex_unlock(socket->lock); if (status != LTTNG_OK) { rcu_read_unlock(); @@ -4210,44 +4422,20 @@ error: * * Return LTTNG_OK on success or a LTTNG_ERR code. */ -static enum lttng_error_code record_kernel_snapshot(struct ltt_kernel_session *ksess, - struct snapshot_output *output, struct ltt_session *session, +static enum lttng_error_code record_kernel_snapshot( + struct ltt_kernel_session *ksess, + const struct consumer_output *output, + const 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) { - status = LTTNG_ERR_NOMEM; - goto error; - } - - status = set_relayd_for_snapshot(ksess->consumer, output, session); - if (status != LTTNG_OK) { - goto error_snapshot; - } - - status = kernel_snapshot_record(ksess, output, wait, nb_packets_per_stream); - if (status != LTTNG_OK) { - goto error_snapshot; - } - - goto end; - -error_snapshot: - /* Clean up copied sockets so this output can use some other later on. */ - consumer_destroy_output_sockets(output->consumer); -error: -end: + status = kernel_snapshot_record( + ksess, output, wait, nb_packets_per_stream); return status; } @@ -4257,55 +4445,31 @@ end: * Returns LTTNG_OK on success or a LTTNG_ERR error code. */ static enum lttng_error_code record_ust_snapshot(struct ltt_ust_session *usess, - struct snapshot_output *output, struct ltt_session *session, + const struct consumer_output *output, + const struct ltt_session *session, int wait, uint64_t nb_packets_per_stream) { - int ret; enum lttng_error_code status; assert(usess); assert(output); assert(session); - /* - * Copy UST session sockets so we can communicate with the right - * consumer for the snapshot record command. - */ - ret = consumer_copy_sockets(output->consumer, usess->consumer); - if (ret < 0) { - status = LTTNG_ERR_NOMEM; - goto error; - } - - status = set_relayd_for_snapshot(usess->consumer, output, session); - if (status != LTTNG_OK) { - goto error_snapshot; - } - - status = ust_app_snapshot_record(usess, output, wait, nb_packets_per_stream); - if (status != LTTNG_OK) { - goto error_snapshot; - } - - goto end; - -error_snapshot: - /* Clean up copied sockets so this output can use some other later on. */ - consumer_destroy_output_sockets(output->consumer); -error: -end: + status = ust_app_snapshot_record( + usess, output, wait, nb_packets_per_stream); return status; } static -uint64_t get_session_size_one_more_packet_per_stream(struct ltt_session *session, - uint64_t cur_nr_packets) +uint64_t get_session_size_one_more_packet_per_stream( + const struct ltt_session *session, uint64_t cur_nr_packets) { uint64_t tot_size = 0; if (session->kernel_session) { struct ltt_kernel_channel *chan; - struct ltt_kernel_session *ksess = session->kernel_session; + const struct ltt_kernel_session *ksess = + session->kernel_session; cds_list_for_each_entry(chan, &ksess->channel_list.head, list) { if (cur_nr_packets >= chan->channel->attr.num_subbuf) { @@ -4321,7 +4485,7 @@ uint64_t get_session_size_one_more_packet_per_stream(struct ltt_session *session } if (session->ust_session) { - struct ltt_ust_session *usess = session->ust_session; + const struct ltt_ust_session *usess = session->ust_session; tot_size += ust_app_get_size_one_more_packet_per_stream(usess, cur_nr_packets); @@ -4351,7 +4515,8 @@ uint64_t get_session_size_one_more_packet_per_stream(struct ltt_session *session * in between this call and actually grabbing data. */ static -int64_t get_session_nb_packets_per_stream(struct ltt_session *session, uint64_t max_size) +int64_t get_session_nb_packets_per_stream(const struct ltt_session *session, + uint64_t max_size) { int64_t size_left; uint64_t cur_nb_packets = 0; @@ -4364,8 +4529,8 @@ int64_t get_session_nb_packets_per_stream(struct ltt_session *session, uint64_t for (;;) { uint64_t one_more_packet_tot_size; - one_more_packet_tot_size = get_session_size_one_more_packet_per_stream(session, - cur_nb_packets); + one_more_packet_tot_size = get_session_size_one_more_packet_per_stream( + session, cur_nb_packets); if (!one_more_packet_tot_size) { /* We are already grabbing all packets. */ break; @@ -4383,6 +4548,161 @@ int64_t get_session_nb_packets_per_stream(struct ltt_session *session, uint64_t return cur_nb_packets; } +static +enum lttng_error_code snapshot_record(struct ltt_session *session, + const struct snapshot_output *snapshot_output, int wait) +{ + int64_t nb_packets_per_stream; + char snapshot_chunk_name[LTTNG_NAME_MAX]; + int ret; + enum lttng_error_code ret_code = LTTNG_OK; + struct lttng_trace_chunk *snapshot_trace_chunk; + struct consumer_output *original_ust_consumer_output = NULL; + struct consumer_output *original_kernel_consumer_output = NULL; + struct consumer_output *snapshot_ust_consumer_output = NULL; + struct consumer_output *snapshot_kernel_consumer_output = NULL; + + ret = snprintf(snapshot_chunk_name, sizeof(snapshot_chunk_name), + "%s-%s-%" PRIu64, + snapshot_output->name, + snapshot_output->datetime, + snapshot_output->nb_snapshot); + if (ret < 0 || ret >= sizeof(snapshot_chunk_name)) { + ERR("Failed to format snapshot name"); + ret_code = LTTNG_ERR_INVALID; + goto error; + } + DBG("Recording snapshot \"%s\" for session \"%s\" with chunk name \"%s\"", + snapshot_output->name, session->name, + snapshot_chunk_name); + if (!session->kernel_session && !session->ust_session) { + ERR("Failed to record snapshot as no channels exist"); + ret_code = LTTNG_ERR_NO_CHANNEL; + goto error; + } + + if (session->kernel_session) { + original_kernel_consumer_output = + session->kernel_session->consumer; + snapshot_kernel_consumer_output = + consumer_copy_output(snapshot_output->consumer); + strcpy(snapshot_kernel_consumer_output->chunk_path, + snapshot_chunk_name); + ret = consumer_copy_sockets(snapshot_kernel_consumer_output, + original_kernel_consumer_output); + if (ret < 0) { + ERR("Failed to copy consumer sockets from snapshot output configuration"); + ret_code = LTTNG_ERR_NOMEM; + goto error; + } + ret_code = set_relayd_for_snapshot( + snapshot_kernel_consumer_output, session); + if (ret_code != LTTNG_OK) { + ERR("Failed to setup relay daemon for kernel tracer snapshot"); + goto error; + } + session->kernel_session->consumer = + snapshot_kernel_consumer_output; + } + if (session->ust_session) { + original_ust_consumer_output = session->ust_session->consumer; + snapshot_ust_consumer_output = + consumer_copy_output(snapshot_output->consumer); + strcpy(snapshot_ust_consumer_output->chunk_path, + snapshot_chunk_name); + ret = consumer_copy_sockets(snapshot_ust_consumer_output, + original_ust_consumer_output); + if (ret < 0) { + ERR("Failed to copy consumer sockets from snapshot output configuration"); + ret_code = LTTNG_ERR_NOMEM; + goto error; + } + ret_code = set_relayd_for_snapshot( + snapshot_ust_consumer_output, session); + if (ret_code != LTTNG_OK) { + ERR("Failed to setup relay daemon for userspace tracer snapshot"); + goto error; + } + session->ust_session->consumer = + snapshot_ust_consumer_output; + } + + snapshot_trace_chunk = session_create_new_trace_chunk(session, + snapshot_kernel_consumer_output ?: + snapshot_ust_consumer_output, + consumer_output_get_base_path( + snapshot_output->consumer), + snapshot_chunk_name); + if (!snapshot_trace_chunk) { + ERR("Failed to create temporary trace chunk to record a snapshot of session \"%s\"", + session->name); + ret_code = LTTNG_ERR_CREATE_DIR_FAIL; + goto error; + } + assert(!session->current_trace_chunk); + ret = session_set_trace_chunk(session, snapshot_trace_chunk, NULL); + lttng_trace_chunk_put(snapshot_trace_chunk); + snapshot_trace_chunk = NULL; + if (ret) { + ERR("Failed to set temporary trace chunk to record a snapshot of session \"%s\"", + session->name); + ret_code = LTTNG_ERR_CREATE_TRACE_CHUNK_FAIL_CONSUMER; + goto error; + } + + nb_packets_per_stream = get_session_nb_packets_per_stream(session, + snapshot_output->max_size); + if (nb_packets_per_stream < 0) { + ret_code = LTTNG_ERR_MAX_SIZE_INVALID; + goto error; + } + + if (session->kernel_session) { + ret_code = record_kernel_snapshot(session->kernel_session, + snapshot_kernel_consumer_output, session, + wait, nb_packets_per_stream); + if (ret_code != LTTNG_OK) { + goto error; + } + } + + if (session->ust_session) { + ret_code = record_ust_snapshot(session->ust_session, + snapshot_ust_consumer_output, session, + wait, nb_packets_per_stream); + if (ret_code != LTTNG_OK) { + goto error; + } + } + + if (session_close_trace_chunk( + session, session->current_trace_chunk, NULL, NULL)) { + /* + * Don't goto end; make sure the chunk is closed for the session + * to allow future snapshots. + */ + ERR("Failed to close snapshot trace chunk of session \"%s\"", + session->name); + ret_code = LTTNG_ERR_CLOSE_TRACE_CHUNK_FAIL_CONSUMER; + } + if (session_set_trace_chunk(session, NULL, NULL)) { + ERR("Failed to release the current trace chunk of session \"%s\"", + session->name); + ret_code = LTTNG_ERR_UNK; + } +error: + if (original_ust_consumer_output) { + session->ust_session->consumer = original_ust_consumer_output; + } + if (original_kernel_consumer_output) { + session->kernel_session->consumer = + original_kernel_consumer_output; + } + consumer_output_put(snapshot_ust_consumer_output); + consumer_output_put(snapshot_kernel_consumer_output); + return ret_code; +} + /* * Command LTTNG_SNAPSHOT_RECORD from lib lttng ctl. * @@ -4392,14 +4712,13 @@ int64_t get_session_nb_packets_per_stream(struct ltt_session *session, uint64_t * Return LTTNG_OK on success or else a LTTNG_ERR code. */ int cmd_snapshot_record(struct ltt_session *session, - struct lttng_snapshot_output *output, int wait) + const struct lttng_snapshot_output *output, int wait) { 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; char datetime[16]; + struct snapshot_output *tmp_output = NULL; assert(session); assert(output); @@ -4431,11 +4750,17 @@ int cmd_snapshot_record(struct ltt_session *session, /* Use temporary output for the session. */ if (*output->ctrl_url != '\0') { + tmp_output = snapshot_output_alloc(); + if (!tmp_output) { + cmd_ret = LTTNG_ERR_NOMEM; + goto error; + } + ret = snapshot_output_init(session, output->max_size, output->name, output->ctrl_url, output->data_url, session->consumer, - &tmp_output, NULL); + tmp_output, NULL); if (ret < 0) { if (ret == -ENOMEM) { cmd_ret = LTTNG_ERR_NOMEM; @@ -4445,41 +4770,14 @@ int cmd_snapshot_record(struct ltt_session *session, goto error; } /* Use the global session count for the temporary snapshot. */ - tmp_output.nb_snapshot = session->snapshot.nb_snapshot; + tmp_output->nb_snapshot = session->snapshot.nb_snapshot; /* Use the global datetime */ - memcpy(tmp_output.datetime, datetime, sizeof(datetime)); - use_tmp_output = 1; - } - - if (use_tmp_output) { - int64_t nb_packets_per_stream; - - nb_packets_per_stream = get_session_nb_packets_per_stream(session, - tmp_output.max_size); - if (nb_packets_per_stream < 0) { - cmd_ret = LTTNG_ERR_MAX_SIZE_INVALID; + memcpy(tmp_output->datetime, datetime, sizeof(datetime)); + cmd_ret = snapshot_record(session, tmp_output, wait); + if (cmd_ret != LTTNG_OK) { goto error; } - - if (session->kernel_session) { - cmd_ret = record_kernel_snapshot(session->kernel_session, - &tmp_output, session, - wait, nb_packets_per_stream); - if (cmd_ret != LTTNG_OK) { - goto error; - } - } - - if (session->ust_session) { - cmd_ret = record_ust_snapshot(session->ust_session, - &tmp_output, session, - wait, nb_packets_per_stream); - if (cmd_ret != LTTNG_OK) { - goto error; - } - } - snapshot_success = 1; } else { struct snapshot_output *sout; @@ -4488,58 +4786,38 @@ int cmd_snapshot_record(struct ltt_session *session, rcu_read_lock(); cds_lfht_for_each_entry(session->snapshot.output_ht->ht, &iter.iter, sout, node.node) { - int64_t nb_packets_per_stream; + struct snapshot_output output_copy; /* - * Make a local copy of the output and assign the possible - * temporary value given by the caller. + * Make a local copy of the output and override output + * parameters with those provided as part of the + * command. */ - memset(&tmp_output, 0, sizeof(tmp_output)); - memcpy(&tmp_output, sout, sizeof(tmp_output)); + memcpy(&output_copy, sout, sizeof(output_copy)); if (output->max_size != (uint64_t) -1ULL) { - tmp_output.max_size = output->max_size; + output_copy.max_size = output->max_size; } - nb_packets_per_stream = get_session_nb_packets_per_stream(session, - tmp_output.max_size); - if (nb_packets_per_stream < 0) { - cmd_ret = LTTNG_ERR_MAX_SIZE_INVALID; - rcu_read_unlock(); - goto error; - } + output_copy.nb_snapshot = session->snapshot.nb_snapshot; + memcpy(output_copy.datetime, datetime, + sizeof(datetime)); /* Use temporary name. */ if (*output->name != '\0') { - if (lttng_strncpy(tmp_output.name, output->name, - sizeof(tmp_output.name))) { + if (lttng_strncpy(output_copy.name, + output->name, + sizeof(output_copy.name))) { cmd_ret = LTTNG_ERR_INVALID; rcu_read_unlock(); goto error; } } - tmp_output.nb_snapshot = session->snapshot.nb_snapshot; - memcpy(tmp_output.datetime, datetime, sizeof(datetime)); - - if (session->kernel_session) { - cmd_ret = record_kernel_snapshot(session->kernel_session, - &tmp_output, session, - wait, nb_packets_per_stream); - if (cmd_ret != LTTNG_OK) { - rcu_read_unlock(); - goto error; - } - } - - if (session->ust_session) { - cmd_ret = record_ust_snapshot(session->ust_session, - &tmp_output, session, - wait, nb_packets_per_stream); - if (cmd_ret != LTTNG_OK) { - rcu_read_unlock(); - goto error; - } + cmd_ret = snapshot_record(session, &output_copy, wait); + if (cmd_ret != LTTNG_OK) { + rcu_read_unlock(); + goto error; } snapshot_success = 1; } @@ -4553,6 +4831,9 @@ int cmd_snapshot_record(struct ltt_session *session, } error: + if (tmp_output) { + snapshot_output_destroy(tmp_output); + } return cmd_ret; } @@ -4588,19 +4869,17 @@ int cmd_set_session_shm_path(struct ltt_session *session, * 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) + struct lttng_rotate_session_return *rotate_return, + bool quiet_rotation) { int ret; + uint64_t ongoing_rotation_chunk_id; enum lttng_error_code cmd_ret = LTTNG_OK; - size_t strf_ret; - struct tm *timeinfo; - char datetime[21]; - time_t now; - /* - * 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; + struct lttng_trace_chunk *chunk_being_archived = NULL; + struct lttng_trace_chunk *new_trace_chunk = NULL; + enum lttng_trace_chunk_status chunk_status; + bool failed_to_rotate = false; + enum lttng_error_code rotation_fail_code = LTTNG_OK; assert(session); @@ -4609,15 +4888,20 @@ int cmd_rotate_session(struct ltt_session *session, goto end; } - if (session->live_timer || !session->output_traces) { + /* + * Explicit rotation is not supported for live sessions. + * However, live sessions can perform a quiet rotation on + * destroy. + * Rotation is not supported for snapshot traces (no output). + */ + if ((!quiet_rotation && session->live_timer) || + !session->output_traces) { cmd_ret = LTTNG_ERR_ROTATION_NOT_AVAILABLE; goto end; } - /* - * Unsupported feature in lttng-relayd before 2.11. - */ - if (session->consumer->type == CONSUMER_DST_NET && + /* Unsupported feature in lttng-relayd before 2.11. */ + if (!quiet_rotation && session->consumer->type == CONSUMER_DST_NET && (session->consumer->relay_major_version == 2 && session->consumer->relay_minor_version < 11)) { cmd_ret = LTTNG_ERR_ROTATION_NOT_AVAILABLE_RELAY; @@ -4642,156 +4926,68 @@ int cmd_rotate_session(struct ltt_session *session, goto end; } - /* Special case for the first rotation. */ - 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. */ - base_path = session_get_base_path(session); - assert(base_path); + session->rotation_state = LTTNG_ROTATION_STATE_ONGOING; - 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"); - cmd_ret = LTTNG_ERR_UNK; - goto end; - } - } else { - /* - * The currently active tracing path is now the folder we - * want to rotate. - */ - 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 the active tracing path to the current rotate path"); - cmd_ret = LTTNG_ERR_UNK; - goto end; + if (session->active) { + new_trace_chunk = session_create_new_trace_chunk(session, NULL, + NULL, NULL); + if (!new_trace_chunk) { + cmd_ret = LTTNG_ERR_CREATE_DIR_FAIL; + goto error; } - } - DBG("Current rotate path %s", session->rotation_chunk.current_rotate_path); - - /* - * Channels created after this point will belong to the next - * archive id. - */ - session->current_archive_id++; - - now = time(NULL); - if (now == (time_t) -1) { - 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"); - cmd_ret = LTTNG_ERR_UNK; - goto end; - } - 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"); - cmd_ret = LTTNG_ERR_UNK; - goto end; - } - - /* Current chunk directory, ex: 20170922-111754-42 */ - ret = snprintf(session->consumer->chunk_path, - sizeof(session->consumer->chunk_path), - "%s-%" PRIu64, datetime, - session->current_archive_id + 1); - if (ret < 0 || ret >= sizeof(session->consumer->chunk_path)) { - ERR("Failed to format the new chunk's directory in rotate session command"); - cmd_ret = LTTNG_ERR_UNK; - goto error; - } + } /* - * The active path for the next rotation/destroy. - * Ex: ~/lttng-traces/auto-20170922-111748/20170922-111754-42 + * The current trace chunk becomes the chunk being archived. + * + * After this point, "chunk_being_archived" must absolutely + * be closed on the consumer(s), otherwise it will never be + * cleaned-up, which will result in a leak. */ - ret = snprintf(session->rotation_chunk.active_tracing_path, - sizeof(session->rotation_chunk.active_tracing_path), - "%s/%s", - session_get_base_path(session), - session->consumer->chunk_path); - if (ret < 0 || ret >= sizeof(session->rotation_chunk.active_tracing_path)) { - ERR("Failed to format active tracing path in rotate session command"); - cmd_ret = LTTNG_ERR_UNK; + ret = session_set_trace_chunk(session, new_trace_chunk, + &chunk_being_archived); + if (ret) { + cmd_ret = LTTNG_ERR_CREATE_TRACE_CHUNK_FAIL_CONSUMER; goto error; } - /* - * 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; + assert(chunk_being_archived); + chunk_status = lttng_trace_chunk_get_id(chunk_being_archived, + &ongoing_rotation_chunk_id); + assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK); if (session->kernel_session) { - ret = lttng_strncpy( - session->kernel_session->consumer->chunk_path, - session->consumer->chunk_path, - sizeof(session->kernel_session->consumer->chunk_path)); - if (ret) { - ERR("Failed to copy current chunk directory to kernel session"); - cmd_ret = LTTNG_ERR_UNK; - goto error; - } - /* - * Create the new chunk folder, before the rotation begins so we - * don't race with the consumer/tracer activity. - */ - ret = domain_mkdir(session->kernel_session->consumer, session, - session->kernel_session->uid, - session->kernel_session->gid); - if (ret) { - cmd_ret = LTTNG_ERR_CREATE_DIR_FAIL; - goto error; - } cmd_ret = kernel_rotate_session(session); if (cmd_ret != LTTNG_OK) { - goto error; + failed_to_rotate = true; + rotation_fail_code = cmd_ret; } } if (session->ust_session) { - ret = lttng_strncpy( - session->ust_session->consumer->chunk_path, - session->consumer->chunk_path, - sizeof(session->ust_session->consumer->chunk_path)); - if (ret) { - ERR("Failed to copy current chunk directory to userspace session"); - cmd_ret = LTTNG_ERR_UNK; - goto error; - } - ret = domain_mkdir(session->ust_session->consumer, session, - session->ust_session->uid, - session->ust_session->gid); - if (ret) { - cmd_ret = LTTNG_ERR_CREATE_DIR_FAIL; - goto error; - } cmd_ret = ust_app_rotate_session(session); if (cmd_ret != LTTNG_OK) { - goto error; + failed_to_rotate = true; + rotation_fail_code = cmd_ret; } } + ret = session_close_trace_chunk(session, chunk_being_archived, + quiet_rotation ? + NULL : + &((enum lttng_trace_chunk_command_type){ + LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED}), + session->last_chunk_path); + if (ret) { + cmd_ret = LTTNG_ERR_CLOSE_TRACE_CHUNK_FAIL_CONSUMER; + goto error; + } + + if (failed_to_rotate) { + cmd_ret = rotation_fail_code; + goto error; + } + + session->quiet_rotation = quiet_rotation; ret = timer_session_rotation_pending_check_start(session, DEFAULT_ROTATE_PENDING_TIMER); if (ret) { @@ -4804,29 +5000,33 @@ int cmd_rotate_session(struct ltt_session *session, } if (rotate_return) { - rotate_return->rotation_id = session->current_archive_id; + rotate_return->rotation_id = ongoing_rotation_chunk_id; } - 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; + session->chunk_being_archived = chunk_being_archived; + chunk_being_archived = NULL; + if (!quiet_rotation) { + ret = notification_thread_command_session_rotation_ongoing( + notification_thread_handle, + session->name, session->uid, session->gid, + ongoing_rotation_chunk_id); + 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); + session->name, ongoing_rotation_chunk_id); end: + lttng_trace_chunk_put(new_trace_chunk); + lttng_trace_chunk_put(chunk_being_archived); 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)) { + LTTNG_ROTATION_STATE_ERROR)) { ERR("Failed to reset rotation state of session \"%s\"", session->name); } @@ -4838,35 +5038,62 @@ error: * * Check if the session has finished its rotation. * - * Return 0 on success or else a LTTNG_ERR code. + * Return LTTNG_OK on success or else an LTTNG_ERR code. */ 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); + enum lttng_error_code cmd_ret = LTTNG_OK; + enum lttng_rotation_state rotation_state; DBG("Cmd rotate_get_info session %s, rotation id %" PRIu64, session->name, - session->current_archive_id); + session->most_recent_chunk_id.value); - if (session->current_archive_id != rotation_id) { - info_return->status = (int32_t) LTTNG_ROTATION_STATE_EXPIRED; - ret = LTTNG_OK; - goto end; + if (session->chunk_being_archived) { + enum lttng_trace_chunk_status chunk_status; + uint64_t chunk_id; + + chunk_status = lttng_trace_chunk_get_id( + session->chunk_being_archived, + &chunk_id); + assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK); + + rotation_state = rotation_id == chunk_id ? + LTTNG_ROTATION_STATE_ONGOING : + LTTNG_ROTATION_STATE_EXPIRED; + } else { + if (session->last_archived_chunk_id.is_set && + rotation_id != session->last_archived_chunk_id.value) { + rotation_state = LTTNG_ROTATION_STATE_EXPIRED; + } else { + rotation_state = session->rotation_state; + } } - switch (session->rotation_state) { + switch (rotation_state) { + case LTTNG_ROTATION_STATE_NO_ROTATION: + DBG("Reporting that no rotation has occurred within the lifetime of session \"%s\"", + session->name); + goto end; + case LTTNG_ROTATION_STATE_EXPIRED: + DBG("Reporting that the rotation state of rotation id %" PRIu64 " of session \"%s\" has expired", + rotation_id, session->name); + break; case LTTNG_ROTATION_STATE_ONGOING: - DBG("Reporting that rotation id %" PRIu64 " of session %s is still pending", + DBG("Reporting that rotation id %" PRIu64 " of session \"%s\" is still pending", rotation_id, session->name); break; case LTTNG_ROTATION_STATE_COMPLETED: { + int fmt_ret; + char *chunk_path; char *current_tracing_path_reply; size_t current_tracing_path_reply_len; + DBG("Reporting that rotation id %" PRIu64 " of session \"%s\" is completed", + rotation_id, session->name); + switch (session_get_consumer_destination_type(session)) { case CONSUMER_DST_LOCAL: current_tracing_path_reply = @@ -4875,8 +5102,21 @@ int cmd_rotate_get_info(struct ltt_session *session, sizeof(info_return->location.local.absolute_path); info_return->location_type = (int8_t) LTTNG_TRACE_ARCHIVE_LOCATION_TYPE_LOCAL; + fmt_ret = asprintf(&chunk_path, + "%s/" DEFAULT_ARCHIVED_TRACE_CHUNKS_DIRECTORY "/%s", + session_get_base_path(session), + session->last_archived_chunk_name); + if (fmt_ret == -1) { + PERROR("Failed to format the path of the last archived trace chunk"); + info_return->status = LTTNG_ROTATION_STATUS_ERROR; + cmd_ret = LTTNG_ERR_UNK; + goto end; + } break; case CONSUMER_DST_NET: + { + uint16_t ctrl_port, data_port; + current_tracing_path_reply = info_return->location.relay.relative_path; current_tracing_path_reply_len = @@ -4885,49 +5125,58 @@ int cmd_rotate_get_info(struct ltt_session *session, info_return->location.relay.protocol = (int8_t) LTTNG_TRACE_ARCHIVE_LOCATION_RELAY_PROTOCOL_TYPE_TCP; - ret = lttng_strncpy(info_return->location.relay.host, + fmt_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"); + if (fmt_ret) { + ERR("Failed to copy host name to rotate_get_info reply"); info_return->status = LTTNG_ROTATION_STATUS_ERROR; - ret = -LTTNG_ERR_UNK; + cmd_ret = LTTNG_ERR_SET_URL; goto end; } - session_get_net_consumer_ports(session, - &info_return->location.relay.ports.control, - &info_return->location.relay.ports.data); + session_get_net_consumer_ports(session, &ctrl_port, &data_port); + info_return->location.relay.ports.control = ctrl_port; + info_return->location.relay.ports.data = data_port; info_return->location_type = (int8_t) LTTNG_TRACE_ARCHIVE_LOCATION_TYPE_RELAY; + chunk_path = strdup(session->last_chunk_path); + if (!chunk_path) { + ERR("Failed to allocate the path of the last archived trace chunk"); + info_return->status = LTTNG_ROTATION_STATUS_ERROR; + cmd_ret = LTTNG_ERR_UNK; + goto end; + } break; + } default: abort(); } - ret = lttng_strncpy(current_tracing_path_reply, - session->rotation_chunk.current_rotate_path, - current_tracing_path_reply_len); - if (ret) { - ERR("Failed to copy current tracing path to rotate_get_info reply"); + + fmt_ret = lttng_strncpy(current_tracing_path_reply, + chunk_path, current_tracing_path_reply_len); + free(chunk_path); + if (fmt_ret) { + ERR("Failed to copy path of the last archived trace chunk to rotate_get_info reply"); info_return->status = LTTNG_ROTATION_STATUS_ERROR; - ret = -LTTNG_ERR_UNK; + cmd_ret = LTTNG_ERR_UNK; goto end; } break; } case LTTNG_ROTATION_STATE_ERROR: - DBG("Reporting that an error occurred during rotation %" PRIu64 " of session %s", + 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; + cmd_ret = LTTNG_OK; end: - return ret; + info_return->status = (int32_t) rotation_state; + return cmd_ret; } /*