Fix: sessiond: no rotation performed from null chunk to new chunk
[lttng-tools.git] / src / bin / lttng-sessiond / cmd.c
index 10fd45757397257d84a4da8c48de3385964c9fcd..5a122d8d3c8dcf7e7f361b34f69f46c4680c0d22 100644 (file)
 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);
@@ -107,7 +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,
@@ -912,6 +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;
        }
@@ -1478,27 +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();
 
@@ -1517,8 +1528,8 @@ 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 */
@@ -1530,8 +1541,8 @@ int cmd_enable_channel(struct ltt_session *session,
                        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;
        }
@@ -1556,16 +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 {
@@ -1595,19 +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;
@@ -1616,10 +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 {
@@ -1632,7 +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:
@@ -1645,11 +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);
 
@@ -1822,7 +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;
@@ -1993,7 +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,
@@ -2390,7 +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,
@@ -2407,7 +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,
@@ -2560,6 +2572,8 @@ int cmd_start_trace(struct ltt_session *session)
        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);
 
@@ -2573,6 +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.
@@ -2588,21 +2618,45 @@ int cmd_start_trace(struct ltt_session *session)
                goto error;
        }
 
+       session->active = 1;
+       session->rotated_after_last_stop = false;
        if (session->output_traces && !session->current_trace_chunk) {
-               struct lttng_trace_chunk *trace_chunk;
+               if (!session->has_been_started) {
+                       struct lttng_trace_chunk *trace_chunk;
 
-               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;
+                       DBG("Creating initial trace chunk of session \"%s\"",
+                                       session->name);
+                       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;
+                       }
                }
        }
 
@@ -2625,10 +2679,6 @@ int cmd_start_trace(struct ltt_session *session)
                }
        }
 
-       /* 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.
@@ -2649,6 +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;
 }
 
@@ -2727,6 +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.
  */
@@ -2747,11 +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;
                }
@@ -2882,7 +2989,6 @@ enum lttng_error_code cmd_create_session_from_descriptor(
        const char *session_name;
        struct ltt_session *new_session = NULL;
        enum lttng_session_descriptor_status descriptor_status;
-       const char *base_path;
 
        session_lock_list();
        if (home_path) {
@@ -2905,13 +3011,9 @@ enum lttng_error_code cmd_create_session_from_descriptor(
                ret_code = LTTNG_ERR_INVALID;
                goto end;
        }
-       ret = lttng_session_descriptor_get_base_path(descriptor, &base_path);
-       if (ret) {
-               ret_code = LTTNG_ERR_INVALID;
-               goto end;
-       }
+
        ret_code = session_create(session_name, creds->uid, creds->gid,
-                       base_path, &new_session);
+                       &new_session);
        if (ret_code != LTTNG_OK) {
                goto end;
        }
@@ -3063,7 +3165,7 @@ void cmd_destroy_session_reply(const struct ltt_session *session,
        struct lttng_trace_archive_location *location = NULL;
        struct lttcomm_lttng_msg llm = {
                .cmd_type = LTTNG_DESTROY_SESSION,
-               .ret_code = LTTNG_OK,
+               .ret_code = reply_context->destruction_status,
                .pid = UINT32_MAX,
                .cmd_header_size =
                        sizeof(struct lttcomm_session_destroy_command_header),
@@ -3145,6 +3247,7 @@ int cmd_destroy_session(struct ltt_session *session,
                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) {
@@ -3169,6 +3272,7 @@ int cmd_destroy_session(struct ltt_session *session,
                        /* 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;
                }
        }
 
@@ -3177,6 +3281,7 @@ int cmd_destroy_session(struct ltt_session *session,
                                session)) {
                        ERR("Failed to stop the \"rotation schedule\" timer of session %s",
                                        session->name);
+                       destruction_last_error = LTTNG_ERR_TIMER_STOP_ERROR;
                }
        }
 
@@ -3196,6 +3301,7 @@ int cmd_destroy_session(struct ltt_session *session,
                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;
@@ -3214,6 +3320,7 @@ int cmd_destroy_session(struct ltt_session *session,
                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;
                }
        }
 
@@ -3278,6 +3385,7 @@ int cmd_destroy_session(struct ltt_session *session,
         */
        session_destroy(session);
        if (reply_context) {
+               reply_context->destruction_status = destruction_last_error;
                ret = session_add_destroy_notifier(session,
                                cmd_destroy_session_reply,
                                (void *) reply_context);
@@ -3708,7 +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;
@@ -3776,7 +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;
@@ -4604,7 +4712,7 @@ error:
  * 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;
@@ -4770,6 +4878,8 @@ int cmd_rotate_session(struct ltt_session *session,
        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);
 
@@ -4827,7 +4937,13 @@ int cmd_rotate_session(struct ltt_session *session,
                }
         }
 
-        /* The current trace chunk becomes the chunk being archived. */
+       /*
+        * 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 = session_set_trace_chunk(session, new_trace_chunk,
                        &chunk_being_archived);
        if (ret) {
@@ -4843,13 +4959,15 @@ int cmd_rotate_session(struct ltt_session *session,
        if (session->kernel_session) {
                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) {
                cmd_ret = ust_app_rotate_session(session);
                if (cmd_ret != LTTNG_OK) {
-                       goto error;
+                       failed_to_rotate = true;
+                       rotation_fail_code = cmd_ret;
                }
        }
 
@@ -4864,6 +4982,11 @@ int cmd_rotate_session(struct ltt_session *session,
                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);
@@ -4950,7 +5073,7 @@ int cmd_rotate_get_info(struct ltt_session *session,
 
        switch (rotation_state) {
        case LTTNG_ROTATION_STATE_NO_ROTATION:
-               DBG("Reporting that no rotation has occured within the lifetime of session \"%s\"",
+               DBG("Reporting that no rotation has occurred within the lifetime of session \"%s\"",
                                session->name);
                goto end;
        case LTTNG_ROTATION_STATE_EXPIRED:
@@ -4991,6 +5114,9 @@ int cmd_rotate_get_info(struct ltt_session *session,
                        }
                        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 =
@@ -5009,9 +5135,9 @@ int cmd_rotate_get_info(struct ltt_session *session,
                                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);
@@ -5022,6 +5148,7 @@ int cmd_rotate_get_info(struct ltt_session *session,
                                goto end;
                        }
                        break;
+               }
                default:
                        abort();
                }
This page took 0.031607 seconds and 4 git commands to generate.