Map session id of relayd and sessiond in consumer
[lttng-tools.git] / src / common / consumer.c
index 260779ae6d1ace6d9a186c40c2202a654c85dbea..b5e212f27d61c5f08f53ef32fb19ac91c1e55917 100644 (file)
@@ -63,6 +63,20 @@ volatile int consumer_quit;
 static struct lttng_ht *metadata_ht;
 static struct lttng_ht *data_ht;
 
+/*
+ * This hash table contains the mapping between the session id of the sessiond
+ * and the relayd session id. Element of the ht are indexed by sessiond_id.
+ *
+ * Node can be added when a relayd communication is opened in the sessiond
+ * thread.
+ *
+ * Note that a session id of the session daemon is unique to a tracing session
+ * and not to a domain session. However, a domain session has one consumer
+ * which forces the 1-1 mapping between a consumer and a domain session (ex:
+ * UST). This means that we can't have duplicate in this ht.
+ */
+static struct lttng_ht *relayd_session_id_ht;
+
 /*
  * Notify a thread pipe to poll back again. This usually means that some global
  * state has changed so we just send back the thread in a poll wait call.
@@ -127,6 +141,12 @@ void consumer_steal_stream_key(int key, struct lttng_ht *ht)
        rcu_read_unlock();
 }
 
+/*
+ * Return a channel object for the given key.
+ *
+ * RCU read side lock MUST be acquired before calling this function and
+ * protects the channel ptr.
+ */
 static struct lttng_consumer_channel *consumer_find_channel(int key)
 {
        struct lttng_ht_iter iter;
@@ -138,8 +158,6 @@ static struct lttng_consumer_channel *consumer_find_channel(int key)
                return NULL;
        }
 
-       rcu_read_lock();
-
        lttng_ht_lookup(consumer_data.channel_ht, (void *)((unsigned long) key),
                        &iter);
        node = lttng_ht_iter_get_node_ulong(&iter);
@@ -147,8 +165,6 @@ static struct lttng_consumer_channel *consumer_find_channel(int key)
                channel = caa_container_of(node, struct lttng_consumer_channel, node);
        }
 
-       rcu_read_unlock();
-
        return channel;
 }
 
@@ -214,6 +230,7 @@ static void destroy_relayd(struct consumer_relayd_sock_pair *relayd)
 {
        int ret;
        struct lttng_ht_iter iter;
+       struct lttng_ht_node_ulong *node;
 
        if (relayd == NULL) {
                return;
@@ -221,6 +238,20 @@ static void destroy_relayd(struct consumer_relayd_sock_pair *relayd)
 
        DBG("Consumer destroy and close relayd socket pair");
 
+       lttng_ht_lookup(relayd_session_id_ht,
+                       (void *)((unsigned long) relayd->session_id), &iter);
+       node = lttng_ht_iter_get_node_ulong(&iter);
+       if (node != NULL) {
+               /* We assume the relayd is being or is destroyed */
+               return;
+       }
+
+       ret = lttng_ht_del(relayd_session_id_ht, &iter);
+       if (ret != 0) {
+               /* We assume the relayd is being or is destroyed */
+               return;
+       }
+
        iter.iter.node = &relayd->node.node;
        ret = lttng_ht_del(consumer_data.relayd_ht, &iter);
        if (ret != 0) {
@@ -232,6 +263,27 @@ static void destroy_relayd(struct consumer_relayd_sock_pair *relayd)
        call_rcu(&relayd->node.head, consumer_rcu_free_relayd);
 }
 
+/*
+ * Iterate over the relayd hash table and destroy each element. Finally,
+ * destroy the whole hash table.
+ */
+static void cleanup_relayd_ht(void)
+{
+       struct lttng_ht_iter iter;
+       struct consumer_relayd_sock_pair *relayd;
+
+       rcu_read_lock();
+
+       cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd,
+                       node.node) {
+               destroy_relayd(relayd);
+       }
+
+       lttng_ht_destroy(consumer_data.relayd_ht);
+
+       rcu_read_unlock();
+}
+
 /*
  * Update the end point status of all streams having the given network sequence
  * index (relayd index).
@@ -348,8 +400,8 @@ void consumer_del_stream(struct lttng_consumer_stream *stream,
                goto free_stream;
        }
 
-       pthread_mutex_lock(&stream->lock);
        pthread_mutex_lock(&consumer_data.lock);
+       pthread_mutex_lock(&stream->lock);
 
        switch (consumer_data.type) {
        case LTTNG_CONSUMER_KERNEL:
@@ -441,8 +493,8 @@ void consumer_del_stream(struct lttng_consumer_stream *stream,
 
 end:
        consumer_data.need_update = 1;
-       pthread_mutex_unlock(&consumer_data.lock);
        pthread_mutex_unlock(&stream->lock);
+       pthread_mutex_unlock(&consumer_data.lock);
 
        if (free_chan) {
                consumer_del_channel(free_chan);
@@ -475,6 +527,8 @@ struct lttng_consumer_stream *consumer_allocate_stream(
                goto end;
        }
 
+       rcu_read_lock();
+
        /*
         * Get stream's channel reference. Needed when adding the stream to the
         * global hash table.
@@ -531,9 +585,12 @@ struct lttng_consumer_stream *consumer_allocate_stream(
                        stream->path_name, stream->key, stream->shm_fd, stream->wait_fd,
                        (unsigned long long) stream->mmap_len, stream->out_fd,
                        stream->net_seq_idx, stream->session_id);
+
+       rcu_read_unlock();
        return stream;
 
 error:
+       rcu_read_unlock();
        free(stream);
 end:
        return NULL;
@@ -757,6 +814,8 @@ void consumer_del_channel(struct lttng_consumer_channel *channel)
        int ret;
        struct lttng_ht_iter iter;
 
+       DBG("Consumer delete channel key %d", channel->key);
+
        pthread_mutex_lock(&consumer_data.lock);
 
        switch (consumer_data.type) {
@@ -910,7 +969,7 @@ static int consumer_update_poll_array(
                 * changed where this function will be called back again.
                 */
                if (stream->state != LTTNG_CONSUMER_ACTIVE_STREAM ||
-                               stream->endpoint_status) {
+                               stream->endpoint_status == CONSUMER_ENDPOINT_INACTIVE) {
                        continue;
                }
                DBG("Active FD %d", stream->wait_fd);
@@ -994,8 +1053,8 @@ int lttng_consumer_send_error(
 }
 
 /*
- * Close all the tracefiles and stream fds, should be called when all instances
- * are destroyed.
+ * Close all the tracefiles and stream fds and MUST be called when all
+ * instances are destroyed i.e. when all threads were joined and are ended.
  */
 void lttng_consumer_cleanup(void)
 {
@@ -1014,6 +1073,15 @@ void lttng_consumer_cleanup(void)
        rcu_read_unlock();
 
        lttng_ht_destroy(consumer_data.channel_ht);
+
+       cleanup_relayd_ht();
+
+       /*
+        * This HT contains streams that are freed by either the metadata thread or
+        * the data thread so we do *nothing* on the hash table and simply destroy
+        * it.
+        */
+       lttng_ht_destroy(consumer_data.stream_list_ht);
 }
 
 /*
@@ -1260,6 +1328,8 @@ end:
  * core function for writing trace buffers to either the local filesystem or
  * the network.
  *
+ * It must be called with the stream lock held.
+ *
  * Careful review MUST be put if any changes occur!
  *
  * Returns the number of bytes written
@@ -1280,8 +1350,6 @@ ssize_t lttng_consumer_on_read_subbuffer_mmap(
        /* RCU lock for the relayd pointer */
        rcu_read_lock();
 
-       pthread_mutex_lock(&stream->lock);
-
        /* Flag that the current stream if set for network streaming. */
        if (stream->net_seq_idx != -1) {
                relayd = consumer_find_relayd(stream->net_seq_idx);
@@ -1406,7 +1474,6 @@ end:
        if (relayd && stream->metadata_flag) {
                pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
        }
-       pthread_mutex_unlock(&stream->lock);
 
        rcu_read_unlock();
        return written;
@@ -1415,6 +1482,8 @@ end:
 /*
  * Splice the data from the ring buffer to the tracefile.
  *
+ * It must be called with the stream lock held.
+ *
  * Returns the number of bytes spliced.
  */
 ssize_t lttng_consumer_on_read_subbuffer_splice(
@@ -1447,8 +1516,6 @@ ssize_t lttng_consumer_on_read_subbuffer_splice(
        /* RCU lock for the relayd pointer */
        rcu_read_lock();
 
-       pthread_mutex_lock(&stream->lock);
-
        /* Flag that the current stream if set for network streaming. */
        if (stream->net_seq_idx != -1) {
                relayd = consumer_find_relayd(stream->net_seq_idx);
@@ -1616,7 +1683,6 @@ end:
        if (relayd && stream->metadata_flag) {
                pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
        }
-       pthread_mutex_unlock(&stream->lock);
 
        rcu_read_unlock();
        return written;
@@ -1690,7 +1756,6 @@ int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx,
  */
 static void destroy_data_stream_ht(struct lttng_ht *ht)
 {
-       int ret;
        struct lttng_ht_iter iter;
        struct lttng_consumer_stream *stream;
 
@@ -1700,10 +1765,11 @@ static void destroy_data_stream_ht(struct lttng_ht *ht)
 
        rcu_read_lock();
        cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
-               ret = lttng_ht_del(ht, &iter);
-               assert(!ret);
-
-               call_rcu(&stream->node.head, consumer_free_stream);
+               /*
+                * Ignore return value since we are currently cleaning up so any error
+                * can't be handled.
+                */
+               (void) consumer_del_stream(stream, ht);
        }
        rcu_read_unlock();
 
@@ -1717,7 +1783,6 @@ static void destroy_data_stream_ht(struct lttng_ht *ht)
  */
 static void destroy_stream_ht(struct lttng_ht *ht)
 {
-       int ret;
        struct lttng_ht_iter iter;
        struct lttng_consumer_stream *stream;
 
@@ -1727,10 +1792,11 @@ static void destroy_stream_ht(struct lttng_ht *ht)
 
        rcu_read_lock();
        cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
-               ret = lttng_ht_del(ht, &iter);
-               assert(!ret);
-
-               call_rcu(&stream->node.head, consumer_free_stream);
+               /*
+                * Ignore return value since we are currently cleaning up so any error
+                * can't be handled.
+                */
+               (void) consumer_del_metadata_stream(stream, ht);
        }
        rcu_read_unlock();
 
@@ -1762,9 +1828,9 @@ void consumer_del_metadata_stream(struct lttng_consumer_stream *stream,
                goto free_stream;
        }
 
+       pthread_mutex_lock(&consumer_data.lock);
        pthread_mutex_lock(&stream->lock);
 
-       pthread_mutex_lock(&consumer_data.lock);
        switch (consumer_data.type) {
        case LTTNG_CONSUMER_KERNEL:
                if (stream->mmap_base != NULL) {
@@ -1854,8 +1920,8 @@ void consumer_del_metadata_stream(struct lttng_consumer_stream *stream,
        }
 
 end:
-       pthread_mutex_unlock(&consumer_data.lock);
        pthread_mutex_unlock(&stream->lock);
+       pthread_mutex_unlock(&consumer_data.lock);
 
        if (free_chan) {
                consumer_del_channel(free_chan);
@@ -1949,7 +2015,7 @@ static void validate_endpoint_status_data_stream(void)
        rcu_read_lock();
        cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) {
                /* Validate delete flag of the stream */
-               if (stream->endpoint_status != CONSUMER_ENDPOINT_INACTIVE) {
+               if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) {
                        continue;
                }
                /* Delete it right now */
@@ -1974,7 +2040,7 @@ static void validate_endpoint_status_metadata_stream(
        rcu_read_lock();
        cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) {
                /* Validate delete flag of the stream */
-               if (!stream->endpoint_status) {
+               if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) {
                        continue;
                }
                /*
@@ -2006,6 +2072,12 @@ void *consumer_thread_metadata_poll(void *data)
 
        rcu_register_thread();
 
+       metadata_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
+       if (!metadata_ht) {
+               /* ENOMEM at this point. Better to bail out. */
+               goto error;
+       }
+
        DBG("Thread metadata poll started");
 
        /* Size is set to 1 for the consumer_metadata pipe */
@@ -2172,9 +2244,7 @@ end:
        DBG("Metadata poll thread exiting");
        lttng_poll_clean(&events);
 
-       if (metadata_ht) {
-               destroy_stream_ht(metadata_ht);
-       }
+       destroy_stream_ht(metadata_ht);
 
        rcu_unregister_thread();
        return NULL;
@@ -2199,6 +2269,7 @@ void *consumer_thread_data_poll(void *data)
 
        data_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
        if (data_ht == NULL) {
+               /* ENOMEM at this point. Better to bail out. */
                goto end;
        }
 
@@ -2435,9 +2506,7 @@ end:
                PERROR("close data pipe");
        }
 
-       if (data_ht) {
-               destroy_data_stream_ht(data_ht);
-       }
+       destroy_data_stream_ht(data_ht);
 
        rcu_unregister_thread();
        return NULL;
@@ -2449,7 +2518,7 @@ end:
  */
 void *consumer_thread_sessiond_poll(void *data)
 {
-       int sock, client_socket, ret;
+       int sock = -1, client_socket, ret;
        /*
         * structure to poll for incoming data on communication socket avoids
         * making blocking sockets.
@@ -2509,6 +2578,13 @@ void *consumer_thread_sessiond_poll(void *data)
                goto end;
        }
 
+       /* This socket is not useful anymore. */
+       ret = close(client_socket);
+       if (ret < 0) {
+               PERROR("close client_socket");
+       }
+       client_socket = -1;
+
        /* update the polling structure to poll on the established socket */
        consumer_sockpoll[1].fd = sock;
        consumer_sockpoll[1].events = POLLIN | POLLPRI;
@@ -2552,6 +2628,20 @@ end:
         */
        notify_thread_pipe(ctx->consumer_data_pipe[1]);
 
+       /* Cleaning up possibly open sockets. */
+       if (sock >= 0) {
+               ret = close(sock);
+               if (ret < 0) {
+                       PERROR("close sock sessiond poll");
+               }
+       }
+       if (client_socket >= 0) {
+               ret = close(sock);
+               if (ret < 0) {
+                       PERROR("close client_socket sessiond poll");
+               }
+       }
+
        rcu_unregister_thread();
        return NULL;
 }
@@ -2559,17 +2649,27 @@ end:
 ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream,
                struct lttng_consumer_local_data *ctx)
 {
+       ssize_t ret;
+
+       pthread_mutex_lock(&stream->lock);
+
        switch (consumer_data.type) {
        case LTTNG_CONSUMER_KERNEL:
-               return lttng_kconsumer_read_subbuffer(stream, ctx);
+               ret = lttng_kconsumer_read_subbuffer(stream, ctx);
+               break;
        case LTTNG_CONSUMER32_UST:
        case LTTNG_CONSUMER64_UST:
-               return lttng_ustconsumer_read_subbuffer(stream, ctx);
+               ret = lttng_ustconsumer_read_subbuffer(stream, ctx);
+               break;
        default:
                ERR("Unknown consumer_data type");
                assert(0);
-               return -ENOSYS;
+               ret = -ENOSYS;
+               break;
        }
+
+       pthread_mutex_unlock(&stream->lock);
+       return ret;
 }
 
 int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream)
@@ -2595,11 +2695,7 @@ void lttng_consumer_init(void)
        consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
        consumer_data.relayd_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
        consumer_data.stream_list_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
-
-       metadata_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
-       assert(metadata_ht);
-       data_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
-       assert(data_ht);
+       relayd_session_id_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
 }
 
 /*
@@ -2610,13 +2706,23 @@ void lttng_consumer_init(void)
  */
 int consumer_add_relayd_socket(int net_seq_idx, int sock_type,
                struct lttng_consumer_local_data *ctx, int sock,
-               struct pollfd *consumer_sockpoll, struct lttcomm_sock *relayd_sock)
+               struct pollfd *consumer_sockpoll, struct lttcomm_sock *relayd_sock,
+               unsigned int sessiond_id)
 {
-       int fd, ret = -1;
+       int fd = -1, ret = -1;
+       enum lttng_error_code ret_code = LTTNG_OK;
        struct consumer_relayd_sock_pair *relayd;
+       struct consumer_relayd_session_id *relayd_id_node;
 
        DBG("Consumer adding relayd socket (idx: %d)", net_seq_idx);
 
+       /* First send a status message before receiving the fds. */
+       ret = consumer_send_status_msg(sock, ret_code);
+       if (ret < 0) {
+               /* Somehow, the session daemon is not responding anymore. */
+               goto error;
+       }
+
        /* Get relayd reference if exists. */
        relayd = consumer_find_relayd(net_seq_idx);
        if (relayd == NULL) {
@@ -2639,6 +2745,14 @@ int consumer_add_relayd_socket(int net_seq_idx, int sock_type,
        if (ret != sizeof(fd)) {
                lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
                ret = -1;
+               fd = -1;        /* Just in case it gets set with an invalid value. */
+               goto error;
+       }
+
+       /* We have the fds without error. Send status back. */
+       ret = consumer_send_status_msg(sock, ret_code);
+       if (ret < 0) {
+               /* Somehow, the session daemon is not responding anymore. */
                goto error;
        }
 
@@ -2648,31 +2762,61 @@ int consumer_add_relayd_socket(int net_seq_idx, int sock_type,
                /* Copy received lttcomm socket */
                lttcomm_copy_sock(&relayd->control_sock, relayd_sock);
                ret = lttcomm_create_sock(&relayd->control_sock);
+               /* Immediately try to close the created socket if valid. */
+               if (relayd->control_sock.fd >= 0) {
+                       if (close(relayd->control_sock.fd)) {
+                               PERROR("close relayd control socket");
+                       }
+               }
+               /* Handle create_sock error. */
                if (ret < 0) {
                        goto error;
                }
 
-               /* Close the created socket fd which is useless */
-               ret = close(relayd->control_sock.fd);
+               /* Assign new file descriptor */
+               relayd->control_sock.fd = fd;
+
+               /*
+                * Create a session on the relayd and store the returned id. No need to
+                * grab the socket lock since the relayd object is not yet visible.
+                */
+               ret = relayd_create_session(&relayd->control_sock,
+                               &relayd->session_id);
                if (ret < 0) {
-                       PERROR("close relayd control socket");
+                       goto error;
                }
 
-               /* Assign new file descriptor */
-               relayd->control_sock.fd = fd;
+               /* Set up a relayd session id node. */
+               relayd_id_node = zmalloc(sizeof(struct consumer_relayd_session_id));
+               if (!relayd_id_node) {
+                       PERROR("zmalloc relayd id node");
+                       goto error;
+               }
+
+               relayd_id_node->relayd_id = relayd->session_id;
+               relayd_id_node->sessiond_id = (uint64_t) sessiond_id;
+
+               /* Indexed by session id of the session daemon. */
+               lttng_ht_node_init_ulong(&relayd_id_node->node,
+                               relayd_id_node->sessiond_id);
+               rcu_read_lock();
+               lttng_ht_add_unique_ulong(relayd_session_id_ht, &relayd_id_node->node);
+               rcu_read_unlock();
+
                break;
        case LTTNG_STREAM_DATA:
                /* Copy received lttcomm socket */
                lttcomm_copy_sock(&relayd->data_sock, relayd_sock);
                ret = lttcomm_create_sock(&relayd->data_sock);
-               if (ret < 0) {
-                       goto error;
+               /* Immediately try to close the created socket if valid. */
+               if (relayd->data_sock.fd >= 0) {
+                       if (close(relayd->data_sock.fd)) {
+                               PERROR("close relayd data socket");
+                       }
                }
-
-               /* Close the created socket fd which is useless */
-               ret = close(relayd->data_sock.fd);
+               /* Handle create_sock error. */
                if (ret < 0) {
-                       PERROR("close relayd control socket");
+                       goto error;
                }
 
                /* Assign new file descriptor */
@@ -2694,9 +2838,15 @@ int consumer_add_relayd_socket(int net_seq_idx, int sock_type,
        add_relayd(relayd);
 
        /* All good! */
-       ret = 0;
+       return 0;
 
 error:
+       /* Close received socket if valid. */
+       if (fd >= 0) {
+               if (close(fd)) {
+                       PERROR("close received socket");
+               }
+       }
        return ret;
 }
 
@@ -2838,3 +2988,17 @@ data_not_pending:
        rcu_read_unlock();
        return 1;
 }
+
+/*
+ * Send a ret code status message to the sessiond daemon.
+ *
+ * Return the sendmsg() return value.
+ */
+int consumer_send_status_msg(int sock, int ret_code)
+{
+       struct lttcomm_consumer_status_msg msg;
+
+       msg.ret_code = ret_code;
+
+       return lttcomm_send_unix_sock(sock, &msg, sizeof(msg));
+}
This page took 0.030361 seconds and 4 git commands to generate.