sessiond: snapshot: set trace chunk to NULL before closing it
[lttng-tools.git] / src / bin / lttng-sessiond / cmd.c
index a17cc678ecab2363a95166b82f1c794a15c8ff96..53850f74aae784b97a668f30c429c92cca8858be 100644 (file)
@@ -998,6 +998,8 @@ static enum lttng_error_code create_connect_relayd(struct lttng_uri *uri,
 
        /* Create socket for control stream. */
        if (uri->stype == LTTNG_STREAM_CONTROL) {
+               uint64_t result_flags;
+
                DBG3("Creating relayd stream socket from URI");
 
                /* Check relayd version */
@@ -1012,6 +1014,16 @@ static enum lttng_error_code create_connect_relayd(struct lttng_uri *uri,
                }
                consumer->relay_major_version = rsock->major;
                consumer->relay_minor_version = rsock->minor;
+               ret = relayd_get_configuration(rsock, 0,
+                               &result_flags);
+               if (ret < 0) {
+                       ERR("Unable to get relayd configuration");
+                       status = LTTNG_ERR_RELAYD_CONNECT_FAIL;
+                       goto close_sock;
+               }
+               if (result_flags & LTTCOMM_RELAYD_CONFIGURATION_FLAG_CLEAR_ALLOWED) {
+                       consumer->relay_allows_clear = true;
+               }
        } else if (uri->stype == LTTNG_STREAM_DATA) {
                DBG3("Creating relayd data socket from URI");
        } else {
@@ -1185,6 +1197,7 @@ int cmd_setup_relayd(struct ltt_session *session)
 
        DBG("Setting relayd for session %s", session->name);
 
+       rcu_read_lock();
        if (session->current_trace_chunk) {
                enum lttng_trace_chunk_status status = lttng_trace_chunk_get_id(
                                session->current_trace_chunk, &current_chunk_id.value);
@@ -1198,8 +1211,6 @@ int cmd_setup_relayd(struct ltt_session *session)
                }
        }
 
-       rcu_read_lock();
-
        if (usess && usess->consumer && usess->consumer->type == CONSUMER_DST_NET
                        && usess->consumer->enabled) {
                /* For each consumer socket, send relayd sockets */
@@ -1225,6 +1236,8 @@ int cmd_setup_relayd(struct ltt_session *session)
                        usess->consumer->relay_major_version;
                session->consumer->relay_minor_version =
                        usess->consumer->relay_minor_version;
+               session->consumer->relay_allows_clear =
+                       usess->consumer->relay_allows_clear;
        }
 
        if (ksess && ksess->consumer && ksess->consumer->type == CONSUMER_DST_NET
@@ -1251,6 +1264,8 @@ int cmd_setup_relayd(struct ltt_session *session)
                        ksess->consumer->relay_major_version;
                session->consumer->relay_minor_version =
                        ksess->consumer->relay_minor_version;
+               session->consumer->relay_allows_clear =
+                       ksess->consumer->relay_allows_clear;
        }
 
 error:
@@ -1261,7 +1276,7 @@ error:
 /*
  * Start a kernel session by opening all necessary streams.
  */
-static int start_kernel_session(struct ltt_kernel_session *ksess)
+int start_kernel_session(struct ltt_kernel_session *ksess)
 {
        int ret;
        struct ltt_kernel_channel *kchan;
@@ -1323,6 +1338,53 @@ error:
        return ret;
 }
 
+int stop_kernel_session(struct ltt_kernel_session *ksess)
+{
+       struct ltt_kernel_channel *kchan;
+       bool error_occurred = false;
+       int ret;
+
+       if (!ksess || !ksess->active) {
+               return LTTNG_OK;
+       }
+       DBG("Stopping kernel tracing");
+
+       ret = kernel_stop_session(ksess);
+       if (ret < 0) {
+               ret = LTTNG_ERR_KERN_STOP_FAIL;
+               goto error;
+       }
+
+       kernel_wait_quiescent();
+
+       /* Flush metadata after stopping (if exists) */
+       if (ksess->metadata_stream_fd >= 0) {
+               ret = kernel_metadata_flush_buffer(ksess->metadata_stream_fd);
+               if (ret < 0) {
+                       ERR("Kernel metadata flush failed");
+                       error_occurred = true;
+               }
+       }
+
+       /* Flush all buffers after stopping */
+       cds_list_for_each_entry(kchan, &ksess->channel_list.head, list) {
+               ret = kernel_flush_buffer(kchan);
+               if (ret < 0) {
+                       ERR("Kernel flush buffer error");
+                       error_occurred = true;
+               }
+       }
+
+       ksess->active = 0;
+       if (error_occurred) {
+               ret = LTTNG_ERR_UNK;
+       } else {
+               ret = LTTNG_OK;
+       }
+error:
+       return ret;
+}
+
 /*
  * Command LTTNG_DISABLE_CHANNEL processed by the client thread.
  */
@@ -2572,6 +2634,10 @@ 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;
+       const bool session_cleared_after_last_stop =
+                       session->cleared_after_last_stop;
 
        assert(session);
 
@@ -2585,6 +2651,22 @@ 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.
@@ -2600,21 +2682,47 @@ int cmd_start_trace(struct ltt_session *session)
                goto error;
        }
 
+       session->active = 1;
+       session->rotated_after_last_stop = false;
+       session->cleared_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,
+                                       LTTNG_TRACE_CHUNK_COMMAND_TYPE_NO_OPERATION);
+                       if (ret != LTTNG_OK) {
+                               goto error;
+                       }
                }
        }
 
@@ -2637,10 +2745,6 @@ 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.
@@ -2661,6 +2765,17 @@ 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;
+               session->cleared_after_last_stop =
+                               session_cleared_after_last_stop;
+       }
        return ret;
 }
 
@@ -2670,14 +2785,12 @@ error:
 int cmd_stop_trace(struct ltt_session *session)
 {
        int ret;
-       struct ltt_kernel_channel *kchan;
        struct ltt_kernel_session *ksession;
        struct ltt_ust_session *usess;
-       bool error_occurred = false;
 
        assert(session);
 
-       DBG("Begin stop session %s (id %" PRIu64 ")", session->name, session->id);
+       DBG("Begin stop session \"%s\" (id %" PRIu64 ")", session->name, session->id);
        /* Short cut */
        ksession = session->kernel_session;
        usess = session->ust_session;
@@ -2688,39 +2801,9 @@ int cmd_stop_trace(struct ltt_session *session)
                goto error;
        }
 
-       /* Kernel tracer */
-       if (ksession && ksession->active) {
-               DBG("Stop kernel tracing");
-
-               ret = kernel_stop_session(ksession);
-               if (ret < 0) {
-                       ret = LTTNG_ERR_KERN_STOP_FAIL;
-                       goto error;
-               }
-
-               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;
-                       }
-               }
-
-               /* Flush all buffers after stopping */
-               cds_list_for_each_entry(kchan, &ksession->channel_list.head, list) {
-                       ret = kernel_flush_buffer(kchan);
-                       if (ret < 0) {
-                               ERR("Kernel flush buffer error");
-                               error_occurred = true;
-                       }
-               }
-
-               ksession->active = 0;
-               DBG("Kernel session stopped %s (id %" PRIu64 ")", session->name,
-                               session->id);
+       ret = stop_kernel_session(ksession);
+       if (ret != LTTNG_OK) {
+               goto error;
        }
 
        if (usess && usess->active) {
@@ -2731,14 +2814,55 @@ int cmd_stop_trace(struct ltt_session *session)
                }
        }
 
+       DBG("Completed stop session \"%s\" (id %" PRIu64 ")", session->name,
+                       session->id);
        /* Flag inactive after a successful stop. */
        session->active = 0;
-       ret = !error_occurred ? LTTNG_OK : LTTNG_ERR_UNK;
+       ret = LTTNG_OK;
 
 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.
  */
@@ -2759,11 +2883,20 @@ 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;
                }
@@ -2894,7 +3027,6 @@ 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) {
@@ -2917,13 +3049,9 @@ 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;
        }
@@ -3200,14 +3328,13 @@ int cmd_destroy_session(struct ltt_session *session,
                session->rotate_size = 0;
        }
 
-       if (session->most_recent_chunk_id.is_set &&
-                       session->most_recent_chunk_id.value != 0 &&
-                       session->current_trace_chunk && session->output_traces) {
+       if (session->rotated && 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);
+               ret = cmd_rotate_session(session, NULL, false,
+                       LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED);
                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));
@@ -3226,7 +3353,8 @@ int cmd_destroy_session(struct ltt_session *session,
                 * emitted and no renaming of the current trace chunk takes
                 * place.
                 */
-               ret = cmd_rotate_session(session, NULL, true);
+               ret = cmd_rotate_session(session, NULL, true,
+                       LTTNG_TRACE_CHUNK_COMMAND_TYPE_NO_OPERATION);
                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));
@@ -4585,8 +4713,14 @@ enum lttng_error_code snapshot_record(struct ltt_session *session,
                }
        }
 
-       if (session_close_trace_chunk(
-                           session, session->current_trace_chunk, NULL, NULL)) {
+       if (session_set_trace_chunk(session, NULL, &snapshot_trace_chunk)) {
+               ERR("Failed to release the current trace chunk of session \"%s\"",
+                               session->name);
+               ret_code = LTTNG_ERR_UNK;
+       }
+
+       if (session_close_trace_chunk(session, snapshot_trace_chunk,
+                       LTTNG_TRACE_CHUNK_COMMAND_TYPE_NO_OPERATION, NULL)) {
                /*
                 * Don't goto end; make sure the chunk is closed for the session
                 * to allow future snapshots.
@@ -4595,11 +4729,6 @@ enum lttng_error_code snapshot_record(struct ltt_session *session,
                                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;
@@ -4780,7 +4909,8 @@ int cmd_set_session_shm_path(struct ltt_session *session,
  */
 int cmd_rotate_session(struct ltt_session *session,
                struct lttng_rotate_session_return *rotate_return,
-               bool quiet_rotation)
+               bool quiet_rotation,
+               enum lttng_trace_chunk_command_type command)
 {
        int ret;
        uint64_t ongoing_rotation_chunk_id;
@@ -4788,6 +4918,8 @@ 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);
 
@@ -4816,6 +4948,12 @@ int cmd_rotate_session(struct ltt_session *session,
                goto end;
        }
 
+       /* Unsupported feature in lttng-modules before 2.8 (lack of sequence number). */
+       if (session->kernel_session && !kernel_supports_ring_buffer_packet_sequence_number()) {
+               cmd_ret = LTTNG_ERR_ROTATION_NOT_AVAILABLE_KERNEL;
+               goto end;
+       }
+
        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);
@@ -4834,7 +4972,16 @@ int cmd_rotate_session(struct ltt_session *session,
                goto end;
        }
 
-       session->rotation_state = LTTNG_ROTATION_STATE_ONGOING;
+       /*
+        * After a stop followed by a clear, disallow following rotations a they would
+        * generate empty chunks.
+        */
+       if (session->cleared_after_last_stop) {
+               DBG("Session \"%s\" was already cleared after stop, refusing rotation",
+                               session->name);
+               cmd_ret = LTTNG_ERR_ROTATION_AFTER_STOP_CLEAR;
+               goto end;
+       }
 
        if (session->active) {
                new_trace_chunk = session_create_new_trace_chunk(session, NULL,
@@ -4845,7 +4992,13 @@ 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) {
@@ -4853,35 +5006,53 @@ int cmd_rotate_session(struct ltt_session *session,
                goto error;
        }
 
-       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) {
                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) {
+                       failed_to_rotate = true;
+                       rotation_fail_code = cmd_ret;
+               }
+       }
+
+       if (!session->active) {
+               session->rotated_after_last_stop = true;
+       }
+
+       if (!chunk_being_archived) {
+               DBG("Rotating session \"%s\" from a \"NULL\" trace chunk to a new trace chunk, skipping completion check",
+                               session->name);
+               if (failed_to_rotate) {
+                       cmd_ret = rotation_fail_code;
                        goto error;
                }
+               cmd_ret = LTTNG_OK;
+               goto end;
        }
 
+       session->rotation_state = LTTNG_ROTATION_STATE_ONGOING;
+       chunk_status = lttng_trace_chunk_get_id(chunk_being_archived,
+                       &ongoing_rotation_chunk_id);
+       assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK);
+
        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);
+               command, 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);
@@ -4890,10 +5061,6 @@ int cmd_rotate_session(struct ltt_session *session,
                goto error;
        }
 
-       if (!session->active) {
-               session->rotated_after_last_stop = true;
-       }
-
        if (rotate_return) {
                rotate_return->rotation_id = ongoing_rotation_chunk_id;
        }
@@ -4968,7 +5135,7 @@ 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:
This page took 0.030343 seconds and 4 git commands to generate.