X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fcmd.c;h=50bacf0c236dc3ea4917a6f4100a80aa39b98623;hp=bdc6db89d5e9d303417f992ce2d12dbd4087ed7c;hb=d5a1b7aa06b4c924b1cd30623758343c74ecab5c;hpb=b019620d62206d12b962d99bb1b02186209caa9d diff --git a/src/bin/lttng-sessiond/cmd.c b/src/bin/lttng-sessiond/cmd.c index bdc6db89d..50bacf0c2 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: @@ -917,8 +1026,8 @@ error: * * The consumer socket lock must be held by the caller. */ -static int send_consumer_relayd_socket(enum lttng_domain_type domain, - unsigned int session_id, struct lttng_uri *relayd_uri, +static int 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) @@ -1003,7 +1112,7 @@ static int send_consumer_relayd_sockets(enum lttng_domain_type domain, /* Sending control relayd socket. */ if (!sock->control_sock_sent) { - ret = send_consumer_relayd_socket(domain, session_id, + ret = send_consumer_relayd_socket(session_id, &consumer->dst.net.control, consumer, sock, session_name, hostname, session_live_timer); if (ret != LTTNG_OK) { @@ -1013,7 +1122,7 @@ static int send_consumer_relayd_sockets(enum lttng_domain_type domain, /* Sending data relayd socket. */ if (!sock->data_sock_sent) { - ret = send_consumer_relayd_socket(domain, session_id, + ret = send_consumer_relayd_socket(session_id, &consumer->dst.net.data, consumer, sock, session_name, hostname, session_live_timer); if (ret != LTTNG_OK) { @@ -1152,7 +1261,7 @@ 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 +1493,15 @@ 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 +2078,7 @@ 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 +2176,7 @@ 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 +2209,12 @@ 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); @@ -2580,7 +2701,7 @@ int cmd_start_trace(struct ltt_session *session) session->rotated_after_last_stop = false; if (session->rotate_timer_period) { - ret = sessiond_rotate_timer_start(session, + ret = timer_session_rotation_schedule_timer_start(session, session->rotate_timer_period); if (ret < 0) { ERR("Failed to enable rotate timer"); @@ -2595,48 +2716,6 @@ 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; - } - - ret = rename_complete_chunk(session, time(NULL)); - if (ret < 0) { - ERR("Failed to rename current rotate path"); - goto end; - } - - /* - * 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--; - return ret; -} - /* * Command LTTNG_STOP_TRACE processed by the client thread. */ @@ -2661,15 +2740,22 @@ 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_timer_enabled) { - sessiond_rotate_timer_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); + } } - if (session->rotate_count > 0 && !session->rotate_pending) { + /* + * A rotation is still ongoing. The check timer will continue to wait + * for the rotation to complete. When the rotation finally completes, + * a check will be performed to rename the "active" chunk to the + * expected "timestamp_begin-timestamp_end" format. + */ + if (session->current_archive_id > 0 && + session->rotation_state != LTTNG_ROTATION_STATE_ONGOING) { ret = rename_active_chunk(session); if (ret) { /* @@ -2968,12 +3054,19 @@ int cmd_destroy_session(struct ltt_session *session, int wpipe, 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_pending_check_timer_enabled) { + if (timer_session_rotation_pending_check_stop(session)) { + ERR("Failed to stop the \"rotation pending check\" timer of session %s", + session->name); + } } - if (session->rotate_timer_enabled) { - sessiond_rotate_timer_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); + } } if (session->rotate_size) { @@ -3017,6 +3110,59 @@ int cmd_destroy_session(struct ltt_session *session, int wpipe, PERROR("write kernel poll pipe"); } + 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); + } ret = session_destroy(session); return ret; @@ -3390,10 +3536,8 @@ 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; @@ -3671,10 +3815,12 @@ 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; } @@ -4375,28 +4521,33 @@ 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[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; + 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; + cmd_ret = LTTNG_ERR_ROTATION_NOT_AVAILABLE; goto end; } @@ -4406,13 +4557,14 @@ 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; + 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"); + 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; } @@ -4423,29 +4575,25 @@ 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; + 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; + cmd_ret = LTTNG_ERR_UNK; goto end; } } else { @@ -4458,40 +4606,54 @@ 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; + 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_state = LTTNG_ROTATION_STATE_ONGOING; - /* - * 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; + cmd_ret = LTTNG_ERR_ROTATION_NOT_AVAILABLE; 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; + 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"); - ret = -LTTNG_ERR_UNK; + 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. @@ -4501,11 +4663,11 @@ 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; - goto end; + cmd_ret = LTTNG_ERR_UNK; + goto error; } /* * The sub-directory for the consumer @@ -4514,11 +4676,11 @@ 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; - goto end; + cmd_ret = LTTNG_ERR_UNK; + goto error; } /* * Create the new chunk folder, before the rotation begins so we don't @@ -4530,32 +4692,31 @@ int cmd_rotate_session(struct ltt_session *session, if (ret) { ERR("Failed to create kernel session tracing path at %s", session->kernel_session->consumer->chunk_path); - ret = -LTTNG_ERR_CREATE_DIR_FAIL; - goto end; + cmd_ret = LTTNG_ERR_CREATE_DIR_FAIL; + goto error; } - ret = kernel_rotate_session(session); - if (ret != LTTNG_OK) { - ret = -ret; - goto end; + cmd_ret = kernel_rotate_session(session); + if (cmd_ret != LTTNG_OK) { + goto error; } } if (session->ust_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; - goto end; + 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; - goto end; + cmd_ret = LTTNG_ERR_UNK; + goto error; } /* * Create the new chunk folder, before the rotation begins so we don't @@ -4565,44 +4726,58 @@ int cmd_rotate_session(struct ltt_session *session, session->ust_session->uid, session->ust_session->gid); if (ret) { - ret = -LTTNG_ERR_CREATE_DIR_FAIL; - goto end; - } - ret = ust_app_rotate_session(session, &ust_active); - if (ret != LTTNG_OK) { - goto end; + 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. + * TODO: ust_app_rotate_session must be adapted to return + * an lttng_error_code, like its kernel counterpart. */ - 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_state = LTTNG_ROTATION_STATE_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->rotation_id = session->rotate_count; + rotate_return->rotation_id = session->current_archive_id; } - DBG("Cmd rotate session %s, rotate_id %" PRIu64 " sent", session->name, - session->rotate_count); - ret = LTTNG_OK; + 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; } /* @@ -4621,9 +4796,9 @@ int cmd_rotate_get_info(struct ltt_session *session, assert(session); DBG("Cmd rotate_get_info session %s, rotation id %" PRIu64, session->name, - session->rotate_count); + session->current_archive_id); - if (session->rotate_count != rotation_id) { + if (session->current_archive_id != rotation_id) { info_return->status = (int32_t) LTTNG_ROTATION_STATE_EXPIRED; ret = LTTNG_OK; goto end; @@ -4635,16 +4810,59 @@ int cmd_rotate_get_info(struct ltt_session *session, rotation_id, session->name); break; case LTTNG_ROTATION_STATE_COMPLETED: - ret = lttng_strncpy(info_return->path, + { + 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(info_return->path)); + current_tracing_path_reply_len); if (ret) { - ERR("Failed to copy active tracing path to rotate_get_info reply"); + 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); @@ -4663,15 +4881,19 @@ end: * 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 0 on success or else an LTTNG_ERR code. + * Return 0 on success or else a positive LTTNG_ERR code. */ int cmd_rotation_set_schedule(struct ltt_session *session, - uint64_t timer_us, uint64_t size, + 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); @@ -4679,71 +4901,119 @@ int cmd_rotation_set_schedule(struct ltt_session *session, if (session->live_timer || session->snapshot_mode || !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; } - /* Trying to override an already active timer. */ - if (timer_us && timer_us != -1ULL && session->rotate_timer_period) { - ret = LTTNG_ERR_ROTATION_TIMER_SET; + 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; - /* Trying to disable an inactive timer. */ - } else if (timer_us == -1ULL && !session->rotate_timer_period) { - ret = LTTNG_ERR_ROTATION_NO_TIMER_SET; + } + + /* + * 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; } - if (size && size != -1ULL && session->rotate_size) { - ret = LTTNG_ERR_ROTATION_SIZE_SET; + /* + * 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; - } else if (size == -1ULL && !session->rotate_size) { - ret = LTTNG_ERR_ROTATION_NO_SIZE_SET; + } + + /* + * 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; } - if (timer_us && !session->rotate_timer_period) { - if (timer_us > UINT_MAX) { - ret = LTTNG_ERR_INVALID; - goto end; - } + *parameter_value = activate ? new_value : 0; - session->rotate_timer_period = timer_us; - /* - * Only start the timer if the session is active, otherwise - * it will be started when the session starts. - */ - if (session->active) { - ret = sessiond_rotate_timer_start(session, timer_us); + 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 enable rotate timer"); + ERR("Failed to disable session rotation timer in ROTATION_SET_SCHEDULE command"); ret = LTTNG_ERR_UNK; goto end; } } - } else if (timer_us == -1ULL && session->rotate_timer_period > 0) { - sessiond_rotate_timer_stop(session); - session->rotate_timer_period = 0; - } - - if (size > 0) { - if (size == -1ULL) { - ret = unsubscribe_session_consumed_size_rotation(session, - notification_thread_handle); + 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; } - session->rotate_size = 0; } else { - ret = subscribe_session_consumed_size_rotation(session, - size, notification_thread_handle); + ret = unsubscribe_session_consumed_size_rotation(session, + notification_thread_handle); if (ret) { - PERROR("Subscribe to session usage"); + ERR("Failed to disable consumed-size notification in ROTATION_SET_SCHEDULE command"); ret = LTTNG_ERR_UNK; goto end; } - session->rotate_size = size; + } + break; + default: + /* Would have been caught before. */ + abort(); } ret = LTTNG_OK; @@ -4754,56 +5024,47 @@ end: return ret; } -/* - * Command ROTATE_GET_CURRENT_PATH from the lttng-ctl library. - * - * Configure the automatic rotation parameters. - * Set to -1ULL to disable them. - * - * Return LTTNG_OK on success or else a LTTNG_ERR code. - */ -int cmd_session_get_current_output(struct ltt_session *session, - struct lttng_session_get_current_output_return *output_return) +/* Wait for a given path to be removed before continuing. */ +static enum lttng_error_code wait_on_path(void *path_data) { - int ret; - const char *path; + const char *shm_path = path_data; - if (!session->snapshot_mode) { - if (session->rotate_count == 0) { - if (session->kernel_session) { - path = session_get_base_path(session); - } else if (session->ust_session) { - path = session_get_base_path(session); + 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 { - abort(); + DBG("shm path no longer exists, completing the destruction of session"); } - assert(path); + break; } else { - path = session->rotation_chunk.active_tracing_path; + 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; + } } - } else { - /* - * A snapshot session does not have a "current" trace archive - * location. - */ - path = ""; + usleep(SESSION_DESTROY_SHM_PATH_CHECK_DELAY_US); } + return LTTNG_OK; +} - DBG("Cmd get current output for session %s, returning %s", - session->name, path); - - ret = lttng_strncpy(output_return->path, - path, - sizeof(output_return->path)); - if (ret) { - ERR("Failed to copy trace output path to session get current output command reply"); - ret = -LTTNG_ERR_UNK; - goto end; - } +/* + * 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; - ret = LTTNG_OK; -end: - return ret; + current_completion_handler = NULL; + return handler; } /*