X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fnotification-thread-events.c;h=823f2a7eb3b2d53a4642844738ebd808445e2f26;hp=81fba4e6912253c7daf8b825594bf3106cc93361;hb=b9a8d78fefbc856370939d9eb553d6e9c1fcc86a;hpb=e7c93cf951a3a2316cb66ca240127d3d23280e29 diff --git a/src/bin/lttng-sessiond/notification-thread-events.c b/src/bin/lttng-sessiond/notification-thread-events.c index 81fba4e69..823f2a7eb 100644 --- a/src/bin/lttng-sessiond/notification-thread-events.c +++ b/src/bin/lttng-sessiond/notification-thread-events.c @@ -318,6 +318,38 @@ int match_session(struct cds_lfht_node *node, const void *key) return !strcmp(session_info->name, name); } +static +const char *notification_command_type_str( + enum notification_thread_command_type type) +{ + switch (type) { + case NOTIFICATION_COMMAND_TYPE_REGISTER_TRIGGER: + return "REGISTER_TRIGGER"; + case NOTIFICATION_COMMAND_TYPE_UNREGISTER_TRIGGER: + return "UNREGISTER_TRIGGER"; + case NOTIFICATION_COMMAND_TYPE_ADD_CHANNEL: + return "ADD_CHANNEL"; + case NOTIFICATION_COMMAND_TYPE_REMOVE_CHANNEL: + return "REMOVE_CHANNEL"; + case NOTIFICATION_COMMAND_TYPE_SESSION_ROTATION_ONGOING: + return "SESSION_ROTATION_ONGOING"; + case NOTIFICATION_COMMAND_TYPE_SESSION_ROTATION_COMPLETED: + return "SESSION_ROTATION_COMPLETED"; + case NOTIFICATION_COMMAND_TYPE_ADD_TRACER_EVENT_SOURCE: + return "ADD_TRACER_EVENT_SOURCE"; + case NOTIFICATION_COMMAND_TYPE_REMOVE_TRACER_EVENT_SOURCE: + return "REMOVE_TRACER_EVENT_SOURCE"; + case NOTIFICATION_COMMAND_TYPE_LIST_TRIGGERS: + return "LIST_TRIGGERS"; + case NOTIFICATION_COMMAND_TYPE_QUIT: + return "QUIT"; + case NOTIFICATION_COMMAND_TYPE_CLIENT_COMMUNICATION_UPDATE: + return "CLIENT_COMMUNICATION_UPDATE"; + default: + abort(); + } +} + /* * Match trigger based on name and credentials only. * Name duplication is NOT allowed for the same uid. @@ -1609,7 +1641,7 @@ int handle_notification_thread_command_add_channel( DBG("[notification-thread] Adding channel %s from session %s, channel key = %" PRIu64 " in %s domain", channel_name, session_name, channel_key_int, - channel_domain == LTTNG_DOMAIN_KERNEL ? "kernel" : "user space"); + lttng_domain_type_str(channel_domain)); CDS_INIT_LIST_HEAD(&trigger_list); @@ -1710,7 +1742,7 @@ int handle_notification_thread_command_remove_channel( struct channel_info *channel_info; DBG("[notification-thread] Removing channel key = %" PRIu64 " in %s domain", - channel_key, domain == LTTNG_DOMAIN_KERNEL ? "kernel" : "user space"); + channel_key, lttng_domain_type_str(domain)); rcu_read_lock(); @@ -1921,7 +1953,6 @@ int handle_notification_thread_command_add_tracer_event_source( goto end; } - CDS_INIT_LIST_HEAD(&element->node); element->fd = tracer_event_source_fd; element->domain = domain_type; @@ -1956,6 +1987,7 @@ int handle_notification_thread_command_remove_tracer_event_source( enum lttng_error_code *_cmd_result) { int ret = 0; + bool found = false; enum lttng_error_code cmd_result = LTTNG_OK; struct notification_event_tracer_event_source_element *source_element = NULL, *tmp; @@ -1969,11 +2001,19 @@ int handle_notification_thread_command_remove_tracer_event_source( tracer_event_source_fd, lttng_domain_type_str(source_element->domain)); cds_list_del(&source_element->node); + found = true; break; } - /* It should always be found. */ - assert(source_element); + if (!found) { + /* + * This is temporarily allowed since the poll activity set is + * not properly cleaned-up for the moment. This is adressed in + * an upcoming fix. + */ + source_element = NULL; + goto end; + } if (!source_element->is_fd_in_poll_set) { /* Skip the poll set removal. */ @@ -1994,6 +2034,8 @@ int handle_notification_thread_command_remove_tracer_event_source( goto end; } + source_element->is_fd_in_poll_set = false; + end: free(source_element); *_cmd_result = cmd_result; @@ -2555,11 +2597,15 @@ int handle_notification_thread_command_register_trigger( ret = evaluate_session_condition_for_client(condition, state, &evaluation, &object_uid, &object_gid); + LTTNG_OPTIONAL_SET(&object_creds.uid, object_uid); + LTTNG_OPTIONAL_SET(&object_creds.gid, object_gid); break; case LTTNG_OBJECT_TYPE_CHANNEL: ret = evaluate_channel_condition_for_client(condition, state, &evaluation, &object_uid, &object_gid); + LTTNG_OPTIONAL_SET(&object_creds.uid, object_uid); + LTTNG_OPTIONAL_SET(&object_creds.gid, object_gid); break; case LTTNG_OBJECT_TYPE_NONE: ret = 0; @@ -2575,9 +2621,6 @@ int handle_notification_thread_command_register_trigger( goto error_put_client_list; } - LTTNG_OPTIONAL_SET(&object_creds.uid, object_uid); - LTTNG_OPTIONAL_SET(&object_creds.gid, object_gid); - DBG("Newly registered trigger's condition evaluated to %s", evaluation ? "true" : "false"); if (!evaluation) { @@ -2660,7 +2703,7 @@ void free_notification_trigger_tokens_ht_element_rcu(struct rcu_head *node) static int handle_notification_thread_command_unregister_trigger( struct notification_thread_state *state, - struct lttng_trigger *trigger, + const struct lttng_trigger *trigger, enum lttng_error_code *_cmd_reply) { struct cds_lfht_iter iter; @@ -2668,7 +2711,7 @@ int handle_notification_thread_command_unregister_trigger( struct lttng_channel_trigger_list *trigger_list; struct notification_client_list *client_list; struct lttng_trigger_ht_element *trigger_ht_element = NULL; - struct lttng_condition *condition = lttng_trigger_get_condition( + const struct lttng_condition *condition = lttng_trigger_get_const_condition( trigger); enum lttng_error_code cmd_reply; @@ -2779,21 +2822,22 @@ int handle_notification_thread_command( struct notification_thread_command, cmd_list_node); cds_list_del(&cmd->cmd_list_node); pthread_mutex_unlock(&handle->cmd_queue.lock); + + DBG("[notification-thread] Received `%s` command", + notification_command_type_str(cmd->type)); switch (cmd->type) { case NOTIFICATION_COMMAND_TYPE_REGISTER_TRIGGER: - DBG("[notification-thread] Received register trigger command"); - ret = handle_notification_thread_command_register_trigger( - state, cmd->parameters.trigger, + ret = handle_notification_thread_command_register_trigger(state, + cmd->parameters.register_trigger.trigger, &cmd->reply_code); break; case NOTIFICATION_COMMAND_TYPE_UNREGISTER_TRIGGER: - DBG("[notification-thread] Received unregister trigger command"); ret = handle_notification_thread_command_unregister_trigger( - state, cmd->parameters.trigger, + state, + cmd->parameters.unregister_trigger.trigger, &cmd->reply_code); break; case NOTIFICATION_COMMAND_TYPE_ADD_CHANNEL: - DBG("[notification-thread] Received add channel command"); ret = handle_notification_thread_command_add_channel( state, cmd->parameters.add_channel.session.name, @@ -2806,7 +2850,6 @@ int handle_notification_thread_command( &cmd->reply_code); break; case NOTIFICATION_COMMAND_TYPE_REMOVE_CHANNEL: - DBG("[notification-thread] Received remove channel command"); ret = handle_notification_thread_command_remove_channel( state, cmd->parameters.remove_channel.key, cmd->parameters.remove_channel.domain, @@ -2814,9 +2857,6 @@ int handle_notification_thread_command( break; case NOTIFICATION_COMMAND_TYPE_SESSION_ROTATION_ONGOING: case NOTIFICATION_COMMAND_TYPE_SESSION_ROTATION_COMPLETED: - DBG("[notification-thread] Received session rotation %s command", - cmd->type == NOTIFICATION_COMMAND_TYPE_SESSION_ROTATION_ONGOING ? - "ongoing" : "completed"); ret = handle_notification_thread_command_session_rotation( state, cmd->type, @@ -2855,7 +2895,6 @@ int handle_notification_thread_command( break; } case NOTIFICATION_COMMAND_TYPE_QUIT: - DBG("[notification-thread] Received quit command"); cmd->reply_code = LTTNG_OK; ret = 1; goto end; @@ -4122,6 +4161,199 @@ end: return ret; } +static struct lttng_event_notifier_notification *receive_notification( + int notification_pipe_read_fd, enum lttng_domain_type domain) +{ + int ret; + uint64_t token; + struct lttng_event_notifier_notification *notification = NULL; + void *reception_buffer; + size_t reception_size; + + struct lttng_ust_event_notifier_notification ust_notification; + struct lttng_kernel_event_notifier_notification kernel_notification; + + /* Init lttng_event_notifier_notification */ + switch(domain) { + case LTTNG_DOMAIN_UST: + reception_buffer = (void *) &ust_notification; + reception_size = sizeof(ust_notification); + break; + case LTTNG_DOMAIN_KERNEL: + reception_buffer = (void *) &kernel_notification; + reception_size = sizeof(kernel_notification); + break; + default: + abort(); + } + + /* + * The monitoring pipe only holds messages smaller than PIPE_BUF, + * ensuring that read/write of tracer notifications are atomic. + */ + ret = lttng_read(notification_pipe_read_fd, reception_buffer, + reception_size); + if (ret != reception_size) { + PERROR("Failed to read from event source notification pipe: fd = %d, size to read = %zu, ret = %d", + notification_pipe_read_fd, reception_size, ret); + ret = -1; + goto end; + } + + switch(domain) { + case LTTNG_DOMAIN_UST: + token = ust_notification.token; + break; + case LTTNG_DOMAIN_KERNEL: + token = kernel_notification.token; + break; + default: + abort(); + } + + notification = lttng_event_notifier_notification_create( + token, domain); +end: + return notification; +} + +int handle_notification_thread_event_notification(struct notification_thread_state *state, + int pipe, + enum lttng_domain_type domain) +{ + int ret; + enum lttng_trigger_status trigger_status; + struct cds_lfht_node *node; + struct cds_lfht_iter iter; + struct notification_trigger_tokens_ht_element *element; + struct lttng_evaluation *evaluation = NULL; + struct lttng_event_notifier_notification *notification = NULL; + enum action_executor_status executor_status; + struct notification_client_list *client_list = NULL; + const char *trigger_name; + + notification = receive_notification(pipe, domain); + if (notification == NULL) { + ERR("[notification-thread] Error receiving notification from tracer (fd = %i, domain = %s)", + pipe, lttng_domain_type_str(domain)); + ret = -1; + goto end; + } + + /* Find triggers associated with this token. */ + rcu_read_lock(); + cds_lfht_lookup(state->trigger_tokens_ht, + hash_key_u64(¬ification->tracer_token, lttng_ht_seed), + match_trigger_token, ¬ification->tracer_token, &iter); + node = cds_lfht_iter_get_node(&iter); + if (caa_unlikely(!node)) { + /* + * This is not an error, slow consumption of the tracer + * notifications can lead to situations where a trigger is + * removed but we still get tracer notifications matching a + * trigger that no longer exists. + */ + ret = 0; + goto end_unlock; + } + + element = caa_container_of(node, + struct notification_trigger_tokens_ht_element, + node); + + if (!lttng_trigger_should_fire(element->trigger)) { + ret = 0; + goto end_unlock; + } + + lttng_trigger_fire(element->trigger); + + trigger_status = lttng_trigger_get_name(element->trigger, &trigger_name); + assert(trigger_status == LTTNG_TRIGGER_STATUS_OK); + + evaluation = lttng_evaluation_event_rule_create( + trigger_name); + if (evaluation == NULL) { + ERR("[notification-thread] Failed to create event rule hit evaluation while creating and enqueuing action executor job"); + ret = -1; + goto end_unlock; + } + + client_list = get_client_list_from_condition(state, + lttng_trigger_get_const_condition(element->trigger)); + executor_status = action_executor_enqueue(state->executor, + element->trigger, evaluation, NULL, client_list); + switch (executor_status) { + case ACTION_EXECUTOR_STATUS_OK: + ret = 0; + break; + case ACTION_EXECUTOR_STATUS_OVERFLOW: + { + struct notification_client_list_element *client_list_element, + *tmp; + + /* + * Not a fatal error; this is expected and simply means the + * executor has too much work queued already. + */ + ret = 0; + + /* No clients subscribed to notifications for this trigger. */ + if (!client_list) { + break; + } + + /* Warn clients that a notification (or more) was dropped. */ + pthread_mutex_lock(&client_list->lock); + cds_list_for_each_entry_safe(client_list_element, tmp, + &client_list->list, node) { + enum client_transmission_status transmission_status; + struct notification_client *client = + client_list_element->client; + + pthread_mutex_lock(&client->lock); + ret = client_notification_overflow(client); + if (ret) { + /* Fatal error. */ + goto next_client; + } + + transmission_status = + client_flush_outgoing_queue(client); + ret = client_handle_transmission_status( + client, transmission_status, state); + if (ret) { + /* Fatal error. */ + goto next_client; + } +next_client: + pthread_mutex_unlock(&client->lock); + if (ret) { + break; + } + } + + pthread_mutex_unlock(&client_list->lock); + break; + } + case ACTION_EXECUTOR_STATUS_ERROR: + /* Fatal error, shut down everything. */ + ERR("Fatal error encoutered while enqueuing action to the action executor"); + ret = -1; + goto end_unlock; + default: + /* Unhandled error. */ + abort(); + } + +end_unlock: + lttng_event_notifier_notification_destroy(notification); + notification_client_list_put(client_list); + rcu_read_unlock(); +end: + return ret; +} + int handle_notification_thread_channel_sample( struct notification_thread_state *state, int pipe, enum lttng_domain_type domain) @@ -4175,8 +4407,7 @@ int handle_notification_thread_channel_sample( */ DBG("[notification-thread] Received a sample for an unknown channel from consumerd, key = %" PRIu64 " in %s domain", latest_sample.key.key, - domain == LTTNG_DOMAIN_KERNEL ? "kernel" : - "user space"); + lttng_domain_type_str(domain)); goto end_unlock; } channel_info = caa_container_of(node, struct channel_info,