Fix: leak of consumer_output when using an explicit snapshot output
[lttng-tools.git] / src / bin / lttng-sessiond / cmd.c
index b5bde4a9db637ba25d1246c55b2e8bdfcf595c0f..b76941648bd50d4a8ddafa6bc2afa8460b8e74b5 100644 (file)
@@ -34,6 +34,7 @@
 #include <common/dynamic-buffer.h>
 #include <common/buffer-view.h>
 #include <common/trace-chunk.h>
+#include <lttng/location-internal.h>
 #include <lttng/trigger/trigger-internal.h>
 #include <lttng/condition/condition.h>
 #include <lttng/action/action.h>
 /* Sleep for 100ms between each check for the shm path's deletion. */
 #define SESSION_DESTROY_SHM_PATH_CHECK_DELAY_US 100000
 
+struct cmd_destroy_session_reply_context {
+       int reply_sock_fd;
+       bool implicit_rotation_on_destroy;
+};
+
 static enum lttng_error_code wait_on_path(void *path);
 
 /*
@@ -1032,7 +1038,9 @@ static enum lttng_error_code send_consumer_relayd_socket(
                struct consumer_output *consumer,
                struct consumer_socket *consumer_sock,
                const char *session_name, const char *hostname,
-               int session_live_timer)
+               int session_live_timer,
+               const uint64_t *current_chunk_id,
+               time_t session_creation_time)
 {
        int ret;
        struct lttcomm_relayd_sock *rsock = NULL;
@@ -1060,7 +1068,8 @@ static enum lttng_error_code send_consumer_relayd_socket(
        /* Send relayd socket to consumer. */
        ret = consumer_send_relayd_socket(consumer_sock, rsock, consumer,
                        relayd_uri->stype, session_id,
-                       session_name, hostname, session_live_timer);
+                       session_name, hostname, session_live_timer,
+                       current_chunk_id, session_creation_time);
        if (ret < 0) {
                status = LTTNG_ERR_ENABLE_CONSUMER_FAIL;
                goto close_sock;
@@ -1107,7 +1116,8 @@ static enum lttng_error_code send_consumer_relayd_sockets(
                enum lttng_domain_type domain,
                unsigned int session_id, struct consumer_output *consumer,
                struct consumer_socket *sock, const char *session_name,
-               const char *hostname, int session_live_timer)
+               const char *hostname, int session_live_timer,
+               const uint64_t *current_chunk_id, time_t session_creation_time)
 {
        enum lttng_error_code status = LTTNG_OK;
 
@@ -1118,7 +1128,8 @@ static enum lttng_error_code send_consumer_relayd_sockets(
        if (!sock->control_sock_sent) {
                status = send_consumer_relayd_socket(session_id,
                                &consumer->dst.net.control, consumer, sock,
-                               session_name, hostname, session_live_timer);
+                               session_name, hostname, session_live_timer,
+                               current_chunk_id, session_creation_time);
                if (status != LTTNG_OK) {
                        goto error;
                }
@@ -1128,7 +1139,8 @@ static enum lttng_error_code send_consumer_relayd_sockets(
        if (!sock->data_sock_sent) {
                status = send_consumer_relayd_socket(session_id,
                                &consumer->dst.net.data, consumer, sock,
-                               session_name, hostname, session_live_timer);
+                               session_name, hostname, session_live_timer,
+                               current_chunk_id, session_creation_time);
                if (status != LTTNG_OK) {
                        goto error;
                }
@@ -1150,14 +1162,28 @@ int cmd_setup_relayd(struct ltt_session *session)
        struct ltt_kernel_session *ksess;
        struct consumer_socket *socket;
        struct lttng_ht_iter iter;
+        LTTNG_OPTIONAL(uint64_t) current_chunk_id = {};
 
-       assert(session);
+        assert(session);
 
        usess = session->ust_session;
        ksess = session->kernel_session;
 
        DBG("Setting relayd for session %s", session->name);
 
+       if (session->current_trace_chunk) {
+               enum lttng_trace_chunk_status status = lttng_trace_chunk_get_id(
+                               session->current_trace_chunk, &current_chunk_id.value);
+
+               if (status == LTTNG_TRACE_CHUNK_STATUS_OK) {
+                       current_chunk_id.is_set = true;
+               } else {
+                       ERR("Failed to get current trace chunk id");
+                       ret = LTTNG_ERR_UNK;
+                       goto error;
+               }
+       }
+
        rcu_read_lock();
 
        if (usess && usess->consumer && usess->consumer->type == CONSUMER_DST_NET
@@ -1169,7 +1195,9 @@ int cmd_setup_relayd(struct ltt_session *session)
                        ret = send_consumer_relayd_sockets(LTTNG_DOMAIN_UST, session->id,
                                        usess->consumer, socket,
                                        session->name, session->hostname,
-                                       session->live_timer);
+                                       session->live_timer,
+                                       current_chunk_id.is_set ? &current_chunk_id.value : NULL,
+                                       session->creation_time);
                        pthread_mutex_unlock(socket->lock);
                        if (ret != LTTNG_OK) {
                                goto error;
@@ -1191,7 +1219,9 @@ int cmd_setup_relayd(struct ltt_session *session)
                        ret = send_consumer_relayd_sockets(LTTNG_DOMAIN_KERNEL, session->id,
                                        ksess->consumer, socket,
                                        session->name, session->hostname,
-                                       session->live_timer);
+                                       session->live_timer,
+                                       current_chunk_id.is_set ? &current_chunk_id.value : NULL,
+                                       session->creation_time);
                        pthread_mutex_unlock(socket->lock);
                        if (ret != LTTNG_OK) {
                                goto error;
@@ -3003,20 +3033,127 @@ error:
        return ret_code;
 }
 
+static
+void cmd_destroy_session_reply(const struct ltt_session *session,
+               void *_reply_context)
+{
+       int ret;
+       ssize_t comm_ret;
+       const struct cmd_destroy_session_reply_context *reply_context =
+                       _reply_context;
+       struct lttng_dynamic_buffer payload;
+       struct lttcomm_session_destroy_command_header cmd_header;
+       struct lttng_trace_archive_location *location = NULL;
+       struct lttcomm_lttng_msg llm = {
+               .cmd_type = LTTNG_DESTROY_SESSION,
+               .ret_code = LTTNG_OK,
+               .pid = UINT32_MAX,
+               .cmd_header_size =
+                       sizeof(struct lttcomm_session_destroy_command_header),
+               .data_size = 0,
+       };
+       size_t payload_size_before_location;
+
+       lttng_dynamic_buffer_init(&payload);
+
+       ret = lttng_dynamic_buffer_append(&payload, &llm, sizeof(llm));
+        if (ret) {
+               ERR("Failed to append session destruction message");
+               goto error;
+        }
+
+       cmd_header.rotation_state =
+                       (int32_t) (reply_context->implicit_rotation_on_destroy ?
+                               session->rotation_state :
+                               LTTNG_ROTATION_STATE_NO_ROTATION);
+       ret = lttng_dynamic_buffer_append(&payload, &cmd_header,
+                       sizeof(cmd_header));
+       if (ret) {
+               ERR("Failed to append session destruction command header");
+               goto error;
+       }
+
+       if (!reply_context->implicit_rotation_on_destroy) {
+               DBG("No implicit rotation performed during the destruction of session \"%s\", sending reply",
+                               session->name);
+               goto send_reply;
+       }
+       if (session->rotation_state != LTTNG_ROTATION_STATE_COMPLETED) {
+               DBG("Rotation state of session \"%s\" is not \"completed\", sending session destruction reply",
+                               session->name);
+               goto send_reply;
+       }
+
+       location = session_get_trace_archive_location(session);
+       if (!location) {
+               ERR("Failed to get the location of the trace archive produced during the destruction of session \"%s\"",
+                               session->name);
+               goto error;
+       }
+
+       payload_size_before_location = payload.size;
+       comm_ret = lttng_trace_archive_location_serialize(location,
+                       &payload);
+       if (comm_ret < 0) {
+               ERR("Failed to serialize the location of the trace archive produced during the destruction of session \"%s\"",
+                               session->name);
+               goto error;
+       }
+       /* Update the message to indicate the location's length. */
+       ((struct lttcomm_lttng_msg *) payload.data)->data_size =
+                       payload.size - payload_size_before_location;
+send_reply:
+       comm_ret = lttcomm_send_unix_sock(reply_context->reply_sock_fd,
+                       payload.data, payload.size);
+       if (comm_ret != (ssize_t) payload.size) {
+               ERR("Failed to send result of the destruction of session \"%s\" to client",
+                               session->name);
+       }
+error:
+       ret = close(reply_context->reply_sock_fd);
+       if (ret) {
+               PERROR("Failed to close client socket in deferred session destroy reply");
+       }
+       lttng_dynamic_buffer_reset(&payload);
+       free(_reply_context);
+}
+
 /*
  * Command LTTNG_DESTROY_SESSION processed by the client thread.
  *
  * Called with session lock held.
  */
 int cmd_destroy_session(struct ltt_session *session,
-               struct notification_thread_handle *notification_thread_handle)
+               struct notification_thread_handle *notification_thread_handle,
+               int *sock_fd)
 {
        int ret;
+       struct cmd_destroy_session_reply_context *reply_context = NULL;
+
+       if (sock_fd) {
+               reply_context = zmalloc(sizeof(*reply_context));
+               if (!reply_context) {
+                       ret = LTTNG_ERR_NOMEM;
+                       goto end;
+               }
+               reply_context->reply_sock_fd = *sock_fd;
+       }
 
        /* Safety net */
        assert(session);
 
-       DBG("Begin destroy session %s (id %" PRIu64 ")", session->name, session->id);
+       DBG("Begin destroy session %s (id %" PRIu64 ")", session->name,
+                       session->id);
+       if (session->active) {
+               DBG("Session \"%s\" is active, attempting to stop it before destroying it",
+                               session->name);
+               ret = cmd_stop_trace(session);
+               if (ret != LTTNG_OK && ret != LTTNG_ERR_TRACE_ALREADY_STOPPED) {
+                       /* Carry on with the destruction of the session. */
+                       ERR("Failed to stop session \"%s\" as part of its destruction: %s",
+                                       session->name, lttng_strerror(-ret));
+               }
+       }
 
        if (session->rotation_schedule_timer_enabled) {
                if (timer_session_rotation_schedule_timer_stop(
@@ -3039,7 +3176,10 @@ int cmd_destroy_session(struct ltt_session *session,
                        ERR("Failed to perform an implicit rotation as part of the destruction of session \"%s\": %s",
                                        session->name, lttng_strerror(-ret));
                }
-       }
+                if (reply_context) {
+                       reply_context->implicit_rotation_on_destroy = true;
+                }
+        }
 
        if (session->shm_path[0]) {
                /*
@@ -3101,8 +3241,19 @@ int cmd_destroy_session(struct ltt_session *session,
         * _at least_ up to the point when that reference is released.
         */
        session_destroy(session);
-       ret = LTTNG_OK;
-
+       if (reply_context) {
+               ret = session_add_destroy_notifier(session,
+                               cmd_destroy_session_reply,
+                               (void *) reply_context);
+               if (ret) {
+                       ret = LTTNG_ERR_FATAL;
+                       goto end;
+               } else {
+                       *sock_fd = -1;
+               }
+        }
+        ret = LTTNG_OK;
+end:
        return ret;
 }
 
@@ -4057,6 +4208,7 @@ static enum lttng_error_code set_relayd_for_snapshot(
        enum lttng_error_code status = LTTNG_OK;
        struct lttng_ht_iter iter;
        struct consumer_socket *socket;
+       LTTNG_OPTIONAL(uint64_t) current_chunk_id = {};
 
        assert(consumer);
        assert(snap_output);
@@ -4064,6 +4216,19 @@ static enum lttng_error_code set_relayd_for_snapshot(
 
        DBG2("Set relayd object from snapshot output");
 
+       if (session->current_trace_chunk) {
+               enum lttng_trace_chunk_status status = lttng_trace_chunk_get_id(
+                               session->current_trace_chunk, &current_chunk_id.value);
+
+               if (status == LTTNG_TRACE_CHUNK_STATUS_OK) {
+                       current_chunk_id.is_set = true;
+               } else {
+                       ERR("Failed to get current trace chunk id");
+                       status = LTTNG_ERR_UNK;
+                       goto error;
+               }
+       }
+
        /* Ignore if snapshot consumer output is not network. */
        if (snap_output->consumer->type != CONSUMER_DST_NET) {
                goto error;
@@ -4080,7 +4245,9 @@ static enum lttng_error_code set_relayd_for_snapshot(
                status = send_consumer_relayd_sockets(0, session->id,
                                snap_output->consumer, socket,
                                session->name, session->hostname,
-                               session->live_timer);
+                               session->live_timer,
+                               current_chunk_id.is_set ? &current_chunk_id.value : NULL,
+                               session->creation_time);
                pthread_mutex_unlock(socket->lock);
                if (status != LTTNG_OK) {
                        rcu_read_unlock();
@@ -4341,7 +4508,8 @@ enum lttng_error_code snapshot_record(struct ltt_session *session,
                }
        }
 
-       if (session_close_trace_chunk(session, session->current_trace_chunk)) {
+       if (session_close_trace_chunk(
+                           session, session->current_trace_chunk, NULL)) {
                /*
                 * Don't goto end; make sure the chunk is closed for the session
                 * to allow future snapshots.
@@ -4372,10 +4540,9 @@ int cmd_snapshot_record(struct ltt_session *session,
 {
        enum lttng_error_code cmd_ret = LTTNG_OK;
        int ret;
-       unsigned int use_tmp_output = 0;
-       struct snapshot_output tmp_output;
        unsigned int snapshot_success = 0;
        char datetime[16];
+       struct snapshot_output *tmp_output = NULL;
 
        assert(session);
        assert(output);
@@ -4407,11 +4574,17 @@ int cmd_snapshot_record(struct ltt_session *session,
 
        /* Use temporary output for the session. */
        if (*output->ctrl_url != '\0') {
+               tmp_output = snapshot_output_alloc();
+               if (!tmp_output) {
+                       cmd_ret = LTTNG_ERR_NOMEM;
+                       goto error;
+               }
+
                ret = snapshot_output_init(session, output->max_size,
                                output->name,
                                output->ctrl_url, output->data_url,
                                session->consumer,
-                               &tmp_output, NULL);
+                               tmp_output, NULL);
                if (ret < 0) {
                        if (ret == -ENOMEM) {
                                cmd_ret = LTTNG_ERR_NOMEM;
@@ -4421,15 +4594,11 @@ int cmd_snapshot_record(struct ltt_session *session,
                        goto error;
                }
                /* Use the global session count for the temporary snapshot. */
-               tmp_output.nb_snapshot = session->snapshot.nb_snapshot;
+               tmp_output->nb_snapshot = session->snapshot.nb_snapshot;
 
                /* Use the global datetime */
-               memcpy(tmp_output.datetime, datetime, sizeof(datetime));
-               use_tmp_output = 1;
-       }
-
-       if (use_tmp_output) {
-               cmd_ret = snapshot_record(session, &tmp_output, wait);
+               memcpy(tmp_output->datetime, datetime, sizeof(datetime));
+               cmd_ret = snapshot_record(session, tmp_output, wait);
                if (cmd_ret != LTTNG_OK) {
                        goto error;
                }
@@ -4441,30 +4610,35 @@ int cmd_snapshot_record(struct ltt_session *session,
                rcu_read_lock();
                cds_lfht_for_each_entry(session->snapshot.output_ht->ht,
                                &iter.iter, sout, node.node) {
+                       struct snapshot_output output_copy;
+
                        /*
-                        * Make a local copy of the output and assign the
-                        * possible temporary value given by the caller.
+                        * Make a local copy of the output and override output
+                        * parameters with those provided as part of the
+                        * command.
                         */
-                       memcpy(&tmp_output, sout, sizeof(tmp_output));
+                       memcpy(&output_copy, sout, sizeof(output_copy));
 
                        if (output->max_size != (uint64_t) -1ULL) {
-                               tmp_output.max_size = output->max_size;
+                               output_copy.max_size = output->max_size;
                        }
 
-                       tmp_output.nb_snapshot = session->snapshot.nb_snapshot;
-                       memcpy(tmp_output.datetime, datetime, sizeof(datetime));
+                       output_copy.nb_snapshot = session->snapshot.nb_snapshot;
+                       memcpy(output_copy.datetime, datetime,
+                                       sizeof(datetime));
 
                        /* Use temporary name. */
                        if (*output->name != '\0') {
-                               if (lttng_strncpy(tmp_output.name, output->name,
-                                               sizeof(tmp_output.name))) {
+                               if (lttng_strncpy(output_copy.name,
+                                               output->name,
+                                               sizeof(output_copy.name))) {
                                        cmd_ret = LTTNG_ERR_INVALID;
                                        rcu_read_unlock();
                                        goto error;
                                }
                        }
 
-                       cmd_ret = snapshot_record(session, &tmp_output, wait);
+                       cmd_ret = snapshot_record(session, &output_copy, wait);
                        if (cmd_ret != LTTNG_OK) {
                                rcu_read_unlock();
                                goto error;
@@ -4481,6 +4655,9 @@ int cmd_snapshot_record(struct ltt_session *session,
        }
 
 error:
+       if (tmp_output) {
+               snapshot_output_destroy(tmp_output);
+       }
        return cmd_ret;
 }
 
@@ -4587,14 +4764,6 @@ int cmd_rotate_session(struct ltt_session *session,
                        &ongoing_rotation_chunk_id);
        assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK);
 
-       chunk_status = lttng_trace_chunk_set_close_command(
-                       chunk_being_archived,
-                       LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED);
-       if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
-               cmd_ret = LTTNG_ERR_FATAL;
-               goto error;
-       }
-
        if (session->kernel_session) {
                cmd_ret = kernel_rotate_session(session);
                if (cmd_ret != LTTNG_OK) {
@@ -4608,7 +4777,9 @@ int cmd_rotate_session(struct ltt_session *session,
                }
        }
 
-       ret = session_close_trace_chunk(session, chunk_being_archived);
+       ret = session_close_trace_chunk(session, chunk_being_archived,
+                       &((enum lttng_trace_chunk_command_type) {
+                                       LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED}));
        if (ret) {
                cmd_ret = LTTNG_ERR_CLOSE_TRACE_CHUNK_FAIL_CONSUMER;
                goto error;
This page took 0.030207 seconds and 4 git commands to generate.