Fix: big relayd cleanup and refactor
[lttng-tools.git] / src / bin / lttng-relayd / live.c
index cbf32790c88f90d3f7641e05ef1467dd9c76749e..764d616a24e692c77f8e717c37ddcf3841d556fb 100644 (file)
@@ -45,6 +45,7 @@
 #include <common/compat/socket.h>
 #include <common/defaults.h>
 #include <common/futex.h>
+#include <common/index/index.h>
 #include <common/sessiond-comm/sessiond-comm.h>
 #include <common/sessiond-comm/inet.h>
 #include <common/sessiond-comm/relayd.h>
 #include "cmd.h"
 #include "live.h"
 #include "lttng-relayd.h"
-#include "lttng-viewer.h"
 #include "utils.h"
+#include "health-relayd.h"
+#include "testpoint.h"
+#include "viewer-stream.h"
+#include "stream.h"
+#include "session.h"
+#include "ctf-trace.h"
 
 static struct lttng_uri *live_uri;
 
-/*
- * Quit pipe for all threads. This permits a single cancellation point
- * for all threads when receiving an event on the pipe.
- */
-static int live_thread_quit_pipe[2] = { -1, -1 };
-
 /*
  * This pipe is used to inform the worker thread that a command is queued and
  * ready to be processed.
@@ -96,64 +96,268 @@ void cleanup(void)
 {
        DBG("Cleaning up");
 
-       /* Close thread quit pipes */
-       utils_close_pipe(live_thread_quit_pipe);
        free(live_uri);
 }
 
 /*
- * Write to writable pipe used to notify a thread.
+ * Receive a request buffer using a given socket, destination allocated buffer
+ * of length size.
+ *
+ * Return the size of the received message or else a negative value on error
+ * with errno being set by recvmsg() syscall.
  */
 static
-int notify_thread_pipe(int wpipe)
+ssize_t recv_request(struct lttcomm_sock *sock, void *buf, size_t size)
 {
-       int ret;
+       ssize_t ret;
 
-       do {
-               ret = write(wpipe, "!", 1);
-       } while (ret < 0 && errno == EINTR);
-       if (ret < 0 || ret != 1) {
-               PERROR("write poll pipe");
+       assert(sock);
+       assert(buf);
+
+       ret = sock->ops->recvmsg(sock, buf, size, 0);
+       if (ret < 0 || ret != size) {
+               if (ret == 0) {
+                       /* Orderly shutdown. Not necessary to print an error. */
+                       DBG("Socket %d did an orderly shutdown", sock->fd);
+               } else {
+                       ERR("Relay failed to receive request.");
+               }
+               ret = -1;
        }
 
        return ret;
 }
 
 /*
- * Stop all threads by closing the thread quit pipe.
+ * Send a response buffer using a given socket, source allocated buffer of
+ * length size.
+ *
+ * Return the size of the sent message or else a negative value on error with
+ * errno being set by sendmsg() syscall.
  */
 static
-void stop_threads(void)
+ssize_t send_response(struct lttcomm_sock *sock, void *buf, size_t size)
 {
-       int ret;
+       ssize_t ret;
 
-       /* Stopping all threads */
-       DBG("Terminating all live threads");
-       ret = notify_thread_pipe(live_thread_quit_pipe[1]);
+       assert(sock);
+       assert(buf);
+
+       ret = sock->ops->sendmsg(sock, buf, size, 0);
        if (ret < 0) {
-               ERR("write error on thread quit pipe");
+               ERR("Relayd failed to send response.");
        }
 
-       /* Dispatch thread */
-       CMM_STORE_SHARED(live_dispatch_thread_exit, 1);
-       futex_nto1_wake(&viewer_cmd_queue.futex);
+       return ret;
+}
+
+/*
+ * Atomically check if new streams got added in the session since the last
+ * check and reset the flag to 0.
+ *
+ * Returns 1 if new streams got added, 0 if nothing changed, a negative value
+ * on error.
+ */
+static
+int check_new_streams(uint64_t session_id, struct lttng_ht *sessions_ht)
+{
+       int ret;
+       unsigned long current_val;
+       struct relay_session *session;
+
+       assert(sessions_ht);
+
+       session = session_find_by_id(sessions_ht, session_id);
+       if (!session) {
+               DBG("Relay session %" PRIu64 " not found", session_id);
+               ret = -1;
+               goto error;
+       }
+
+       current_val = uatomic_cmpxchg(&session->new_streams, 1, 0);
+       ret = current_val;
+
+error:
+       return ret;
+}
+
+/*
+ * Send viewer streams to the given socket. The ignore_sent_flag indicates if
+ * this function should ignore the sent flag or not.
+ *
+ * Return 0 on success or else a negative value.
+ */
+static
+ssize_t send_viewer_streams(struct lttcomm_sock *sock,
+               struct relay_session *session, unsigned int ignore_sent_flag)
+{
+       ssize_t ret;
+       struct lttng_viewer_stream send_stream;
+       struct lttng_ht_iter iter;
+       struct relay_viewer_stream *vstream;
+
+       assert(session);
+
+       rcu_read_lock();
+
+       cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, vstream,
+                       stream_n.node) {
+               struct ctf_trace *ctf_trace;
+
+               health_code_update();
+
+               /* Ignore if not the same session. */
+               if (vstream->session_id != session->id ||
+                               (!ignore_sent_flag && vstream->sent_flag)) {
+                       continue;
+               }
+
+               ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
+                               vstream->path_name);
+               assert(ctf_trace);
+
+               send_stream.id = htobe64(vstream->stream_handle);
+               send_stream.ctf_trace_id = htobe64(ctf_trace->id);
+               send_stream.metadata_flag = htobe32(vstream->metadata_flag);
+               strncpy(send_stream.path_name, vstream->path_name,
+                               sizeof(send_stream.path_name));
+               strncpy(send_stream.channel_name, vstream->channel_name,
+                               sizeof(send_stream.channel_name));
+
+               DBG("Sending stream %" PRIu64 " to viewer", vstream->stream_handle);
+               ret = send_response(sock, &send_stream, sizeof(send_stream));
+               if (ret < 0) {
+                       goto end_unlock;
+               }
+               vstream->sent_flag = 1;
+       }
+
+       ret = 0;
+
+end_unlock:
+       rcu_read_unlock();
+       return ret;
 }
 
 /*
- * Init thread quit pipe.
+ * Create every viewer stream possible for the given session with the seek
+ * type. Three counters *can* be return which are in order the total amount of
+ * viewer stream of the session, the number of unsent stream and the number of
+ * stream created. Those counters can be NULL and thus will be ignored.
  *
- * Return -1 on error or 0 if all pipes are created.
+ * Return 0 on success or else a negative value.
  */
 static
-int init_thread_quit_pipe(void)
+int make_viewer_streams(struct relay_session *session,
+               enum lttng_viewer_seek seek_t, uint32_t *nb_total, uint32_t *nb_unsent,
+               uint32_t *nb_created)
 {
        int ret;
+       struct lttng_ht_iter iter;
+       struct ctf_trace *ctf_trace;
+
+       assert(session);
+
+       /*
+        * This is to make sure we create viewer streams for a full received
+        * channel. For instance, if we have 8 streams for a channel that are
+        * concurrently being flagged ready, we can end up creating just a subset
+        * of the 8 streams (the ones that are flagged). This lock avoids this
+        * limbo state.
+        */
+       pthread_mutex_lock(&session->viewer_ready_lock);
 
-       ret = utils_create_pipe_cloexec(live_thread_quit_pipe);
+       /*
+        * Create viewer streams for relay streams that are ready to be used for a
+        * the given session id only.
+        */
+       rcu_read_lock();
+       cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
+                       node.node) {
+               struct relay_stream *stream;
+
+               health_code_update();
 
+               if (ctf_trace->invalid_flag) {
+                       continue;
+               }
+
+               cds_list_for_each_entry(stream, &ctf_trace->stream_list, trace_list) {
+                       struct relay_viewer_stream *vstream;
+
+                       if (!stream->viewer_ready) {
+                               continue;
+                       }
+
+                       vstream = viewer_stream_find_by_id(stream->stream_handle);
+                       if (!vstream) {
+                               vstream = viewer_stream_create(stream, seek_t, ctf_trace);
+                               if (!vstream) {
+                                       ret = -1;
+                                       goto error_unlock;
+                               }
+                               /* Acquire reference to ctf_trace. */
+                               ctf_trace_get_ref(ctf_trace);
+
+                               if (nb_created) {
+                                       /* Update number of created stream counter. */
+                                       (*nb_created)++;
+                               }
+                       } else if (!vstream->sent_flag && nb_unsent) {
+                               /* Update number of unsent stream counter. */
+                               (*nb_unsent)++;
+                       }
+                       /* Update number of total stream counter. */
+                       if (nb_total) {
+                               (*nb_total)++;
+                       }
+               }
+       }
+
+       ret = 0;
+
+error_unlock:
+       rcu_read_unlock();
+       pthread_mutex_unlock(&session->viewer_ready_lock);
        return ret;
 }
 
+/*
+ * Write to writable pipe used to notify a thread.
+ */
+static
+int notify_thread_pipe(int wpipe)
+{
+       ssize_t ret;
+
+       ret = lttng_write(wpipe, "!", 1);
+       if (ret < 1) {
+               PERROR("write poll pipe");
+       }
+
+       return (int) ret;
+}
+
+/*
+ * Stop all threads by closing the thread quit pipe.
+ */
+static
+void stop_threads(void)
+{
+       int ret;
+
+       /* Stopping all threads */
+       DBG("Terminating all live threads");
+       ret = notify_thread_pipe(live_conn_pipe[1]);
+       if (ret < 0) {
+               ERR("write error on thread quit pipe");
+       }
+
+       /* Dispatch thread */
+       CMM_STORE_SHARED(live_dispatch_thread_exit, 1);
+       futex_nto1_wake(&viewer_cmd_queue.futex);
+}
+
 /*
  * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
  */
@@ -173,7 +377,7 @@ int create_thread_poll_set(struct lttng_poll_event *events, int size)
        }
 
        /* Add quit pipe */
-       ret = lttng_poll_add(events, live_thread_quit_pipe[0], LPOLLIN);
+       ret = lttng_poll_add(events, live_conn_pipe[0], LPOLLIN | LPOLLERR);
        if (ret < 0) {
                goto error;
        }
@@ -190,9 +394,9 @@ error:
  * Return 1 if it was triggered else 0;
  */
 static
-int check_thread_quit_pipe(int fd, uint32_t events)
+int check_live_conn_pipe(int fd, uint32_t events)
 {
-       if (fd == live_thread_quit_pipe[0] && (events & LPOLLIN)) {
+       if (fd == live_conn_pipe[0] && (events & LPOLLIN)) {
                return 1;
        }
 
@@ -254,14 +458,16 @@ void *thread_listener(void *data)
 
        DBG("[thread] Relay live listener started");
 
+       health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_LISTENER);
+
+       health_code_update();
+
        live_control_sock = init_socket(live_uri);
        if (!live_control_sock) {
                goto error_sock_control;
        }
 
-       /*
-        * Pass 3 as size here for the thread quit pipe, control and data socket.
-        */
+       /* Pass 2 as size here for the thread quit pipe and control sockets. */
        ret = create_thread_poll_set(&events, 2);
        if (ret < 0) {
                goto error_create_poll;
@@ -273,11 +479,21 @@ void *thread_listener(void *data)
                goto error_poll_add;
        }
 
+       lttng_relay_notify_ready();
+
+       if (testpoint(relayd_thread_live_listener)) {
+               goto error_testpoint;
+       }
+
        while (1) {
+               health_code_update();
+
                DBG("Listener accepting live viewers connections");
 
 restart:
+               health_poll_entry();
                ret = lttng_poll_wait(&events, -1);
+               health_poll_exit();
                if (ret < 0) {
                        /*
                         * Restart interrupted system call.
@@ -291,12 +507,14 @@ restart:
 
                DBG("Relay new viewer connection received");
                for (i = 0; i < nb_fd; i++) {
+                       health_code_update();
+
                        /* Fetch once the poll data */
                        revents = LTTNG_POLL_GETEV(&events, i);
                        pollfd = LTTNG_POLL_GETFD(&events, i);
 
                        /* Thread quit pipe has been closed. Killing thread. */
-                       ret = check_thread_quit_pipe(pollfd, revents);
+                       ret = check_live_conn_pipe(pollfd, revents);
                        if (ret) {
                                err = 0;
                                goto exit;
@@ -354,6 +572,7 @@ restart:
 exit:
 error:
 error_poll_add:
+error_testpoint:
        lttng_poll_clean(&events);
 error_create_poll:
        if (live_control_sock->fd >= 0) {
@@ -365,8 +584,10 @@ error_create_poll:
        lttcomm_destroy_sock(live_control_sock);
 error_sock_control:
        if (err) {
+               health_error();
                DBG("Live viewer listener thread exited with error");
        }
+       health_unregister(health_relayd);
        DBG("Live viewer listener thread cleanup complete");
        stop_threads();
        return NULL;
@@ -378,17 +599,30 @@ error_sock_control:
 static
 void *thread_dispatcher(void *data)
 {
-       int ret;
+       int err = -1;
+       ssize_t ret;
        struct cds_wfq_node *node;
        struct relay_command *relay_cmd = NULL;
 
        DBG("[thread] Live viewer relay dispatcher started");
 
+       health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_DISPATCHER);
+
+       if (testpoint(relayd_thread_live_dispatcher)) {
+               goto error_testpoint;
+       }
+
+       health_code_update();
+
        while (!CMM_LOAD_SHARED(live_dispatch_thread_exit)) {
+               health_code_update();
+
                /* Atomically prepare the queue futex */
                futex_nto1_prepare(&viewer_cmd_queue.futex);
 
                do {
+                       health_code_update();
+
                        /* Dequeue commands */
                        node = cds_wfq_dequeue_blocking(&viewer_cmd_queue.queue);
                        if (node == NULL) {
@@ -407,22 +641,31 @@ void *thread_dispatcher(void *data)
                         * so we can be assured that the data will be read at some point in
                         * time or wait to the end of the world :)
                         */
-                       do {
-                               ret = write(live_relay_cmd_pipe[1], relay_cmd,
-                                               sizeof(*relay_cmd));
-                       } while (ret < 0 && errno == EINTR);
+                       ret = lttng_write(live_relay_cmd_pipe[1], relay_cmd,
+                                       sizeof(*relay_cmd));
                        free(relay_cmd);
-                       if (ret < 0 || ret != sizeof(struct relay_command)) {
+                       if (ret < sizeof(struct relay_command)) {
                                PERROR("write cmd pipe");
                                goto error;
                        }
                } while (node != NULL);
 
                /* Futex wait on queue. Blocking call on futex() */
+               health_poll_entry();
                futex_nto1_wait(&viewer_cmd_queue.futex);
+               health_poll_exit();
        }
 
+       /* Normal exit, no error */
+       err = 0;
+
 error:
+error_testpoint:
+       if (err) {
+               health_error();
+               ERR("Health error occurred in %s", __func__);
+       }
+       health_unregister(health_relayd);
        DBG("Live viewer dispatch thread dying");
        stop_threads();
        return NULL;
@@ -443,27 +686,25 @@ int viewer_connect(struct relay_command *cmd)
 
        cmd->version_check_done = 1;
 
-       /* Get version from the other side. */
-       ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
-       if (ret < 0 || ret != sizeof(msg)) {
-               if (ret == 0) {
-                       /* Orderly shutdown. Not necessary to print an error. */
-                       DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
-               } else {
-                       ERR("Relay failed to receive the version values.");
-               }
-               ret = -1;
+       health_code_update();
+
+       DBG("Viewer is establishing a connection to the relayd.");
+
+       ret = recv_request(cmd->sock, &msg, sizeof(msg));
+       if (ret < 0) {
                goto end;
        }
 
+       health_code_update();
+
        reply.major = RELAYD_VERSION_COMM_MAJOR;
        reply.minor = RELAYD_VERSION_COMM_MINOR;
 
        /* Major versions must be the same */
        if (reply.major != be32toh(msg.major)) {
-               DBG("Incompatible major versions (%u vs %u)", reply.major,
-                               be32toh(msg.major));
-               ret = 0;
+               DBG("Incompatible major versions ([relayd] %u vs [client] %u)",
+                               reply.major, be32toh(msg.major));
+               ret = -1;
                goto end;
        }
 
@@ -490,12 +731,16 @@ int viewer_connect(struct relay_command *cmd)
        if (cmd->type == RELAY_VIEWER_COMMAND) {
                reply.viewer_session_id = htobe64(++last_relay_viewer_session_id);
        }
-       ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
-                       sizeof(struct lttng_viewer_connect), 0);
+
+       health_code_update();
+
+       ret = send_response(cmd->sock, &reply, sizeof(reply));
        if (ret < 0) {
-               ERR("Relay sending version");
+               goto end;
        }
 
+       health_code_update();
+
        DBG("Version check done using protocol %u.%u", cmd->major, cmd->minor);
        ret = 0;
 
@@ -516,36 +761,28 @@ int viewer_list_sessions(struct relay_command *cmd,
        struct lttng_viewer_list_sessions session_list;
        unsigned long count;
        long approx_before, approx_after;
-       struct lttng_ht_node_ulong *node;
        struct lttng_ht_iter iter;
        struct lttng_viewer_session send_session;
        struct relay_session *session;
 
        DBG("List sessions received");
 
-       if (cmd->version_check_done == 0) {
-               ERR("Trying to list sessions before version check");
-               ret = -1;
-               goto end_no_session;
-       }
-
        rcu_read_lock();
        cds_lfht_count_nodes(sessions_ht->ht, &approx_before, &count, &approx_after);
        session_list.sessions_count = htobe32(count);
 
-       ret = cmd->sock->ops->sendmsg(cmd->sock, &session_list,
-                       sizeof(session_list), 0);
+       health_code_update();
+
+       ret = send_response(cmd->sock, &session_list, sizeof(session_list));
        if (ret < 0) {
-               ERR("Relay sending sessions list");
                goto end_unlock;
        }
 
-       cds_lfht_for_each_entry(sessions_ht->ht, &iter.iter, node, node) {
-               node = lttng_ht_iter_get_node_ulong(&iter);
-               if (!node) {
-                       goto end_unlock;
-               }
-               session = caa_container_of(node, struct relay_session, session_n);
+       health_code_update();
+
+       cds_lfht_for_each_entry(sessions_ht->ht, &iter.iter, session,
+                       session_n.node) {
+               health_code_update();
 
                strncpy(send_session.session_name, session->session_name,
                                sizeof(send_session.session_name));
@@ -553,15 +790,18 @@ int viewer_list_sessions(struct relay_command *cmd,
                                sizeof(send_session.hostname));
                send_session.id = htobe64(session->id);
                send_session.live_timer = htobe32(session->live_timer);
-               send_session.clients = htobe32(session->viewer_attached);
+               send_session.clients = htobe32(session->viewer_refcount);
+               send_session.streams = htobe32(session->stream_count);
 
-               ret = cmd->sock->ops->sendmsg(cmd->sock, &send_session,
-                               sizeof(send_session), 0);
+               health_code_update();
+
+               ret = send_response(cmd->sock, &send_session, sizeof(send_session));
                if (ret < 0) {
-                       ERR("Relay sending session info");
                        goto end_unlock;
                }
        }
+       health_code_update();
+
        rcu_read_unlock();
        ret = 0;
        goto end;
@@ -570,68 +810,97 @@ end_unlock:
        rcu_read_unlock();
 
 end:
-end_no_session:
        return ret;
 }
 
 /*
- * Allocate and init a new viewer_stream.
- *
- * Copies the values from the stream passed in parameter and insert the new
- * stream in the viewer_streams_ht.
- *
- * MUST be called with rcu_read_lock held.
- *
- * Returns 0 on success or a negative value on error.
+ * Send the viewer the list of current sessions.
  */
 static
-int init_viewer_stream(struct relay_stream *stream)
+int viewer_get_new_streams(struct relay_command *cmd,
+               struct lttng_ht *sessions_ht)
 {
-       int ret;
-       struct relay_viewer_stream *viewer_stream;
+       int ret, send_streams = 0;
+       uint32_t nb_created = 0, nb_unsent = 0, nb_streams = 0;
+       struct lttng_viewer_new_streams_request request;
+       struct lttng_viewer_new_streams_response response;
+       struct relay_session *session;
+
+       assert(cmd);
+       assert(sessions_ht);
 
-       assert(stream);
+       DBG("Get new streams received");
 
-       viewer_stream = zmalloc(sizeof(*viewer_stream));
-       if (!viewer_stream) {
-               PERROR("relay viewer stream zmalloc");
-               ret = -1;
+       health_code_update();
+
+       /* Receive the request from the connected client. */
+       ret = recv_request(cmd->sock, &request, sizeof(request));
+       if (ret < 0) {
                goto error;
        }
 
-       viewer_stream->read_fd = -1;
-       viewer_stream->index_read_fd = -1;
-       viewer_stream->session_id = stream->session->id;
-       viewer_stream->stream_handle = stream->stream_handle;
-       viewer_stream->path_name = strndup(stream->path_name,
-                       LTTNG_VIEWER_PATH_MAX);
-       viewer_stream->channel_name = strndup(stream->channel_name,
-                       LTTNG_VIEWER_NAME_MAX);
-       viewer_stream->total_index_received = stream->total_index_received;
-       viewer_stream->tracefile_size = stream->tracefile_size;
-       viewer_stream->tracefile_count = stream->tracefile_count;
-       viewer_stream->metadata_flag = stream->metadata_flag;
+       health_code_update();
+
+       rcu_read_lock();
+       session = session_find_by_id(sessions_ht, be64toh(request.session_id));
+       if (!session) {
+               DBG("Relay session %" PRIu64 " not found",
+                               be64toh(request.session_id));
+               response.status = htobe32(VIEWER_NEW_STREAMS_ERR);
+               goto send_reply;
+       }
+
+       if (cmd->session_id == session->id) {
+               /* We confirmed the viewer is asking for the same session. */
+               send_streams = 1;
+               response.status = htobe32(VIEWER_NEW_STREAMS_OK);
+       } else {
+               send_streams = 0;
+               response.status = htobe32(VIEWER_NEW_STREAMS_ERR);
+               goto send_reply;
+       }
+
+       if (!send_streams) {
+               goto send_reply;
+       }
+
+       ret = make_viewer_streams(session, VIEWER_SEEK_LAST, NULL, &nb_unsent,
+                       &nb_created);
+       if (ret < 0) {
+               goto end_unlock;
+       }
+       /* Only send back the newly created streams with the unsent ones. */
+       nb_streams = nb_created + nb_unsent;
+       response.streams_count = htobe32(nb_streams);
+
+send_reply:
+       health_code_update();
+       ret = send_response(cmd->sock, &response, sizeof(response));
+       if (ret < 0) {
+               goto end_unlock;
+       }
+       health_code_update();
 
        /*
-        * This is to avoid a race between the initialization of this object and
-        * the close of the given stream. If the stream is unable to find this
-        * viewer stream when closing, this copy will at least take the latest
-        * value.
+        * Unknown or empty session, just return gracefully, the viewer knows what
+        * is happening.
         */
-       viewer_stream->total_index_received = stream->total_index_received;
+       if (!send_streams || !nb_streams) {
+               ret = 0;
+               goto end_unlock;
+       }
 
        /*
-        * The deletion of this ctf_trace object is only done in a call RCU of the
-        * relay stream making it valid as long as we have the read side lock.
+        * Send stream and *DON'T* ignore the sent flag so every viewer streams
+        * that were not sent from that point will be sent to the viewer.
         */
-       viewer_stream->ctf_trace = stream->ctf_trace;
-       uatomic_inc(&viewer_stream->ctf_trace->refcount);
-
-       lttng_ht_node_init_u64(&viewer_stream->stream_n, stream->stream_handle);
-       lttng_ht_add_unique_u64(viewer_streams_ht, &viewer_stream->stream_n);
-
-       ret = 0;
+       ret = send_viewer_streams(cmd->sock, session, 0);
+       if (ret < 0) {
+               goto end_unlock;
+       }
 
+end_unlock:
+       rcu_read_unlock();
 error:
        return ret;
 }
@@ -643,66 +912,48 @@ static
 int viewer_attach_session(struct relay_command *cmd,
                struct lttng_ht *sessions_ht)
 {
-       int ret, send_streams = 0, nb_streams = 0;
+       int send_streams = 0;
+       ssize_t ret;
+       uint32_t nb_streams = 0;
+       enum lttng_viewer_seek seek_type;
        struct lttng_viewer_attach_session_request request;
        struct lttng_viewer_attach_session_response response;
-       struct lttng_viewer_stream send_stream;
-       struct relay_stream *stream;
-       struct relay_viewer_stream *viewer_stream;
-       struct lttng_ht_node_ulong *node;
-       struct lttng_ht_node_u64 *node64;
-       struct lttng_ht_iter iter;
        struct relay_session *session;
 
        assert(cmd);
        assert(sessions_ht);
 
-       DBG("Attach session received");
+       health_code_update();
 
-       if (cmd->version_check_done == 0) {
-               ERR("Trying to attach session before version check");
-               ret = -1;
-               goto end_no_session;
-       }
-
-       ret = cmd->sock->ops->recvmsg(cmd->sock, &request, sizeof(request), 0);
-       if (ret < 0 || ret != sizeof(request)) {
-               if (ret == 0) {
-                       /* Orderly shutdown. Not necessary to print an error. */
-                       DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
-               } else {
-                       ERR("Relay failed to receive the attach parameters.");
-               }
-               ret = -1;
+       /* Receive the request from the connected client. */
+       ret = recv_request(cmd->sock, &request, sizeof(request));
+       if (ret < 0) {
                goto error;
        }
 
+       health_code_update();
+
        rcu_read_lock();
-       lttng_ht_lookup(sessions_ht,
-                       (void *)((unsigned long) be64toh(request.session_id)), &iter);
-       node = lttng_ht_iter_get_node_ulong(&iter);
-       if (node == NULL) {
+       session = session_find_by_id(sessions_ht, be64toh(request.session_id));
+       if (!session) {
                DBG("Relay session %" PRIu64 " not found",
                                be64toh(request.session_id));
                response.status = htobe32(VIEWER_ATTACH_UNK);
                goto send_reply;
        }
+       session_viewer_attach(session);
+       DBG("Attach session ID %" PRIu64 " received", be64toh(request.session_id));
 
-       session = caa_container_of(node, struct relay_session, session_n);
-       if (cmd->session_id == session->id) {
-               /* Same viewer already attached, just send the stream list. */
-               send_streams = 1;
-               response.status = htobe32(VIEWER_ATTACH_OK);
-       } else if (session->viewer_attached != 0) {
+       if (uatomic_read(&session->viewer_refcount) > 1) {
                DBG("Already a viewer attached");
                response.status = htobe32(VIEWER_ATTACH_ALREADY);
+               session_viewer_detach(session);
                goto send_reply;
        } else if (session->live_timer == 0) {
                DBG("Not live session");
                response.status = htobe32(VIEWER_ATTACH_NOT_LIVE);
                goto send_reply;
        } else {
-               session->viewer_attached++;
                send_streams = 1;
                response.status = htobe32(VIEWER_ATTACH_OK);
                cmd->session_id = session->id;
@@ -711,10 +962,8 @@ int viewer_attach_session(struct relay_command *cmd,
 
        switch (be32toh(request.seek)) {
        case VIEWER_SEEK_BEGINNING:
-               /* Default behaviour. */
-               break;
        case VIEWER_SEEK_LAST:
-               /* TODO */
+               seek_type = be32toh(request.seek);
                break;
        default:
                ERR("Wrong seek parameter");
@@ -723,188 +972,45 @@ int viewer_attach_session(struct relay_command *cmd,
                goto send_reply;
        }
 
-       if (send_streams) {
-               /* We should only be there if we have a session to attach to. */
-               assert(session);
-
-               /*
-                * Fill the viewer_streams_ht to count the number of streams
-                * ready to be sent and avoid concurrency issues on the
-                * relay_streams_ht and don't rely on a total session stream count.
-                */
-               cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, node, node) {
-                       struct relay_viewer_stream *vstream;
-
-                       node = lttng_ht_iter_get_node_ulong(&iter);
-                       if (!node) {
-                               continue;
-                       }
-                       stream = caa_container_of(node, struct relay_stream, stream_n);
-                       if (stream->session != cmd->session) {
-                               continue;
-                       }
-
-                       /*
-                        * Don't send streams with no ctf_trace, they are not ready to be
-                        * read.
-                        */
-                       if (!stream->ctf_trace) {
-                               continue;
-                       }
+       if (!send_streams) {
+               goto send_reply;
+       }
 
-                       vstream = live_find_viewer_stream_by_id(stream->stream_handle);
-                       if (!vstream) {
-                               ret = init_viewer_stream(stream);
-                               if (ret < 0) {
-                                       goto end_unlock;
-                               }
-                       }
-                       nb_streams++;
-               }
-               response.streams_count = htobe32(nb_streams);
+       ret = make_viewer_streams(session, seek_type, &nb_streams, NULL, NULL);
+       if (ret < 0) {
+               goto end_unlock;
        }
+       response.streams_count = htobe32(nb_streams);
 
 send_reply:
-       ret = cmd->sock->ops->sendmsg(cmd->sock, &response, sizeof(response), 0);
+       health_code_update();
+       ret = send_response(cmd->sock, &response, sizeof(response));
        if (ret < 0) {
-               ERR("Relay sending viewer attach response");
                goto end_unlock;
        }
+       health_code_update();
 
        /*
-        * Unknown or busy session, just return gracefully, the viewer knows what
+        * Unknown or empty session, just return gracefully, the viewer knows what
         * is happening.
         */
-       if (!send_streams) {
+       if (!send_streams || !nb_streams) {
                ret = 0;
                goto end_unlock;
        }
 
-       /* We should only be there if we have a session to attach to. */
-       assert(session);
-       cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, node, node) {
-               node64 = lttng_ht_iter_get_node_u64(&iter);
-               if (!node64) {
-                       continue;
-               }
-               viewer_stream = caa_container_of(node64, struct relay_viewer_stream,
-                               stream_n);
-               if (viewer_stream->session_id != cmd->session->id) {
-                       continue;
-               }
-
-               send_stream.id = htobe64(viewer_stream->stream_handle);
-               send_stream.ctf_trace_id = htobe64(viewer_stream->ctf_trace->id);
-               send_stream.metadata_flag = htobe32(viewer_stream->metadata_flag);
-               strncpy(send_stream.path_name, viewer_stream->path_name,
-                               sizeof(send_stream.path_name));
-               strncpy(send_stream.channel_name, viewer_stream->channel_name,
-                               sizeof(send_stream.channel_name));
-
-               ret = cmd->sock->ops->sendmsg(cmd->sock, &send_stream,
-                               sizeof(send_stream), 0);
-               if (ret < 0) {
-                       ERR("Relay sending stream %" PRIu64, viewer_stream->stream_handle);
-                       goto end_unlock;
-               }
-               DBG("Sent stream %" PRIu64 " to viewer", viewer_stream->stream_handle);
+       /* Send stream and ignore the sent flag. */
+       ret = send_viewer_streams(cmd->sock, session, 1);
+       if (ret < 0) {
+               goto end_unlock;
        }
-       ret = 0;
 
 end_unlock:
        rcu_read_unlock();
-end_no_session:
 error:
        return ret;
 }
 
-/*
- * Open index file using a given viewer stream.
- *
- * Return 0 on success or else a negative value.
- */
-static int open_index(struct relay_viewer_stream *stream)
-{
-       int ret;
-       char fullpath[PATH_MAX];
-       struct lttng_packet_index_file_hdr hdr;
-
-       if (stream->tracefile_size > 0) {
-               /* For now we don't support on-disk ring buffer. */
-               ret = -1;
-               goto end;
-       } else {
-               ret = snprintf(fullpath, sizeof(fullpath), "%s/" DEFAULT_INDEX_DIR
-                               "/%s" DEFAULT_INDEX_FILE_SUFFIX,
-                               stream->path_name, stream->channel_name);
-               if (ret < 0) {
-                       PERROR("snprintf index path");
-                       goto error;
-               }
-       }
-
-       DBG("Opening index file %s in read only", fullpath);
-       ret = open(fullpath, O_RDONLY);
-       if (ret < 0) {
-               if (errno == ENOENT) {
-                       ret = ENOENT;
-                       goto error;
-               } else {
-                       PERROR("opening index in read-only");
-               }
-               goto error;
-       }
-       stream->index_read_fd = ret;
-       DBG("Opening index file %s in read only, (fd: %d)", fullpath, ret);
-
-       do {
-               ret = read(stream->index_read_fd, &hdr, sizeof(hdr));
-       } while (ret < 0 && errno == EINTR);
-       if (ret < 0) {
-               PERROR("Reading index header");
-               goto error;
-       }
-       if (strncmp(hdr.magic, INDEX_MAGIC, sizeof(hdr.magic)) != 0) {
-               ERR("Invalid header magic");
-               ret = -1;
-               goto error;
-       }
-       if (be32toh(hdr.index_major) != INDEX_MAJOR ||
-                       be32toh(hdr.index_minor) != INDEX_MINOR) {
-               ERR("Invalid header version");
-               ret = -1;
-               goto error;
-       }
-       ret = 0;
-
-error:
-end:
-       return ret;
-}
-
-/*
- * Get viewer stream from stream id.
- *
- * RCU read side lock MUST be acquired.
- */
-struct relay_viewer_stream *live_find_viewer_stream_by_id(uint64_t stream_id)
-{
-       struct lttng_ht_node_u64 *node;
-       struct lttng_ht_iter iter;
-       struct relay_viewer_stream *stream = NULL;
-
-       lttng_ht_lookup(viewer_streams_ht, &stream_id, &iter);
-       node = lttng_ht_iter_get_node_u64(&iter);
-       if (node == NULL) {
-               DBG("Relay viewer stream %" PRIu64 " not found", stream_id);
-               goto end;
-       }
-       stream = caa_container_of(node, struct relay_viewer_stream, stream_n);
-
-end:
-       return stream;
-}
-
 /*
  * Send the next index for a stream.
  *
@@ -917,36 +1023,41 @@ int viewer_get_next_index(struct relay_command *cmd,
        int ret;
        struct lttng_viewer_get_next_index request_index;
        struct lttng_viewer_index viewer_index;
-       struct lttng_packet_index packet_index;
+       struct ctf_packet_index packet_index;
        struct relay_viewer_stream *vstream;
        struct relay_stream *rstream;
+       struct ctf_trace *ctf_trace;
+       struct relay_session *session;
 
        assert(cmd);
        assert(sessions_ht);
 
        DBG("Viewer get next index");
 
-       if (cmd->version_check_done == 0) {
-               ERR("Trying to request index before version check");
-               ret = -1;
-               goto end_no_session;
-       }
+       health_code_update();
 
-       ret = cmd->sock->ops->recvmsg(cmd->sock, &request_index,
-                       sizeof(request_index), 0);
-       if (ret < 0 || ret != sizeof(request_index)) {
-               ret = -1;
-               ERR("Relay didn't receive the whole packet");
+       ret = recv_request(cmd->sock, &request_index, sizeof(request_index));
+       if (ret < 0) {
                goto end;
        }
+       health_code_update();
 
        rcu_read_lock();
-       vstream = live_find_viewer_stream_by_id(be64toh(request_index.stream_id));
+       session = session_find_by_id(sessions_ht, cmd->session_id);
+       if (!session) {
+               ret = -1;
+               goto end_unlock;
+       }
+
+       vstream = viewer_stream_find_by_id(be64toh(request_index.stream_id));
        if (!vstream) {
                ret = -1;
                goto end_unlock;
        }
 
+       ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht, vstream->path_name);
+       assert(ctf_trace);
+
        memset(&viewer_index, 0, sizeof(viewer_index));
 
        /*
@@ -959,8 +1070,9 @@ int viewer_get_next_index(struct relay_command *cmd,
 
        /* First time, we open the index file */
        if (vstream->index_read_fd < 0) {
-               ret = open_index(vstream);
-               if (ret == ENOENT) {
+               ret = index_open(vstream->path_name, vstream->channel_name,
+                               vstream->tracefile_count, vstream->tracefile_count_current);
+               if (ret == -ENOENT) {
                        /*
                         * The index is created only when the first data packet arrives, it
                         * might not be ready at the beginning of the session
@@ -971,42 +1083,116 @@ int viewer_get_next_index(struct relay_command *cmd,
                        viewer_index.status = htobe32(VIEWER_INDEX_ERR);
                        goto send_reply;
                }
+               vstream->index_read_fd = ret;
        }
 
-       rstream = relay_stream_find_by_id(vstream->stream_handle);
-       if (rstream) {
-               if (rstream->beacon_ts_end != -1ULL &&
-                               vstream->last_sent_index == rstream->total_index_received) {
-                       viewer_index.status = htobe32(VIEWER_INDEX_INACTIVE);
-                       viewer_index.timestamp_end = htobe64(rstream->beacon_ts_end);
-                       goto send_reply;
+       rstream = stream_find_by_id(relay_streams_ht, vstream->stream_handle);
+       assert(rstream);
+
+       if (!rstream->close_flag) {
+               if (vstream->abort_flag) {
+                       /* Rotate on abort (overwrite). */
+                       DBG("Viewer rotate because of overwrite");
+                       ret = viewer_stream_rotate(vstream, rstream);
+                       if (ret < 0) {
+                               goto end_unlock;
+                       } else if (ret == 1) {
+                               viewer_index.status = htobe32(VIEWER_INDEX_HUP);
+                               viewer_stream_delete(vstream);
+                               viewer_stream_destroy(ctf_trace, vstream);
+                               goto send_reply;
+                       }
+                       /* ret == 0 means successful so we continue. */
                }
 
-               if (rstream->total_index_received <= vstream->last_sent_index) {
-                       /* No new index to send, retry later. */
-                       viewer_index.status = htobe32(VIEWER_INDEX_RETRY);
-                       goto send_reply;
+               pthread_mutex_lock(&rstream->viewer_stream_rotation_lock);
+               if (rstream->tracefile_count_current == vstream->tracefile_count_current) {
+                       if (rstream->beacon_ts_end != -1ULL &&
+                               vstream->last_sent_index == rstream->total_index_received) {
+                               viewer_index.status = htobe32(VIEWER_INDEX_INACTIVE);
+                               viewer_index.timestamp_end = htobe64(rstream->beacon_ts_end);
+                               pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
+                               goto send_reply;
+                       } else if (rstream->total_index_received <= vstream->last_sent_index
+                                       && !vstream->close_write_flag) {
+                               /*
+                                * Reader and writer are working in the same tracefile, so we care
+                                * about the number of index received and sent. Otherwise, we read
+                                * up to EOF.
+                                */
+                               pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
+                               /* No new index to send, retry later. */
+                               viewer_index.status = htobe32(VIEWER_INDEX_RETRY);
+                               goto send_reply;
+                       }
                }
-       } else if (!rstream &&
+               pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
+       } else if (rstream->close_flag && vstream->close_write_flag &&
                        vstream->total_index_received == vstream->last_sent_index) {
-               /* Last index sent and stream closed */
+               /* Last index sent and current tracefile closed in write */
                viewer_index.status = htobe32(VIEWER_INDEX_HUP);
+               viewer_stream_delete(vstream);
+               viewer_stream_destroy(ctf_trace, vstream);
                goto send_reply;
+       } else {
+               vstream->close_write_flag = 1;
        }
 
-       if (!vstream->ctf_trace->metadata_received ||
-                       vstream->ctf_trace->metadata_received >
-                       vstream->ctf_trace->metadata_sent) {
+       if (!ctf_trace->metadata_received ||
+                       ctf_trace->metadata_received > ctf_trace->metadata_sent) {
                viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
        }
 
-       do {
-               ret = read(vstream->index_read_fd, &packet_index,
-                               sizeof(packet_index));
-       } while (ret < 0 && errno == EINTR);
+       ret = check_new_streams(vstream->session_id, sessions_ht);
+       if (ret < 0) {
+               goto end_unlock;
+       } else if (ret == 1) {
+               viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
+       }
+
+       pthread_mutex_lock(&vstream->overwrite_lock);
+       if (vstream->abort_flag) {
+               /*
+                * The file is being overwritten by the writer, we cannot * use it.
+                */
+               viewer_index.status = htobe32(VIEWER_INDEX_RETRY);
+               pthread_mutex_unlock(&vstream->overwrite_lock);
+               ret = viewer_stream_rotate(vstream, rstream);
+               if (ret < 0) {
+                       goto end_unlock;
+               } else if (ret == 1) {
+                       viewer_index.status = htobe32(VIEWER_INDEX_HUP);
+                       viewer_stream_delete(vstream);
+                       viewer_stream_destroy(ctf_trace, vstream);
+                       goto send_reply;
+               }
+               goto send_reply;
+       }
+
+       ret = lttng_read(vstream->index_read_fd, &packet_index,
+                       sizeof(packet_index));
+       pthread_mutex_unlock(&vstream->overwrite_lock);
        if (ret < sizeof(packet_index)) {
-               PERROR("Relay reading index file");
-               viewer_index.status = htobe32(VIEWER_INDEX_ERR);
+               /*
+                * The tracefile is closed in write, so we read up to EOF.
+                */
+               if (vstream->close_write_flag == 1) {
+                       viewer_index.status = htobe32(VIEWER_INDEX_RETRY);
+                       /* Rotate on normal EOF */
+                       ret = viewer_stream_rotate(vstream, rstream);
+                       if (ret < 0) {
+                               goto end_unlock;
+                       } else if (ret == 1) {
+                               viewer_index.status = htobe32(VIEWER_INDEX_HUP);
+                               viewer_stream_delete(vstream);
+                               viewer_stream_destroy(ctf_trace, vstream);
+                               goto send_reply;
+                       }
+               } else {
+                       PERROR("Relay reading index file %d", vstream->index_read_fd);
+                       viewer_index.status = htobe32(VIEWER_INDEX_ERR);
+               }
+               goto send_reply;
        } else {
                viewer_index.status = htobe32(VIEWER_INDEX_OK);
                vstream->last_sent_index++;
@@ -1025,20 +1211,20 @@ int viewer_get_next_index(struct relay_command *cmd,
 
 send_reply:
        viewer_index.flags = htobe32(viewer_index.flags);
-       ret = cmd->sock->ops->sendmsg(cmd->sock, &viewer_index,
-                       sizeof(viewer_index), 0);
+       health_code_update();
+
+       ret = send_response(cmd->sock, &viewer_index, sizeof(viewer_index));
        if (ret < 0) {
-               ERR("Relay index to viewer");
                goto end_unlock;
        }
+       health_code_update();
 
-       DBG("Index %" PRIu64 "for stream %" PRIu64 "sent",
+       DBG("Index %" PRIu64 " for stream %" PRIu64 " sent",
                        vstream->last_sent_index, vstream->stream_handle);
 
 end_unlock:
        rcu_read_unlock();
 
-end_no_session:
 end:
        return ret;
 }
@@ -1049,7 +1235,8 @@ end:
  * Return 0 on success or else a negative value.
  */
 static
-int viewer_get_packet(struct relay_command *cmd)
+int viewer_get_packet(struct relay_command *cmd,
+               struct lttng_ht *sessions_ht)
 {
        int ret, send_data = 0;
        char *data = NULL;
@@ -1058,31 +1245,32 @@ int viewer_get_packet(struct relay_command *cmd)
        struct lttng_viewer_get_packet get_packet_info;
        struct lttng_viewer_trace_packet reply;
        struct relay_viewer_stream *stream;
+       struct ctf_trace *ctf_trace;
 
        assert(cmd);
 
        DBG2("Relay get data packet");
 
-       if (cmd->version_check_done == 0) {
-               ERR("Trying to get packet before version check");
-               ret = -1;
-               goto end;
-       }
+       health_code_update();
 
-       ret = cmd->sock->ops->recvmsg(cmd->sock, &get_packet_info,
-                       sizeof(get_packet_info), 0);
-       if (ret < 0 || ret != sizeof(get_packet_info)) {
-               ret = -1;
-               ERR("Relay didn't receive the whole packet");
+       ret = recv_request(cmd->sock, &get_packet_info, sizeof(get_packet_info));
+       if (ret < 0) {
                goto end;
        }
+       health_code_update();
+
+       /* From this point on, the error label can be reached. */
+       memset(&reply, 0, sizeof(reply));
 
        rcu_read_lock();
-       stream = live_find_viewer_stream_by_id(be64toh(get_packet_info.stream_id));
+       stream = viewer_stream_find_by_id(be64toh(get_packet_info.stream_id));
        if (!stream) {
                goto error;
        }
-       assert(stream->ctf_trace);
+
+       ctf_trace = ctf_trace_find_by_path(cmd->session->ctf_traces_ht,
+                       stream->path_name);
+       assert(ctf_trace);
 
        /*
         * First time we read this stream, we need open the tracefile, we should
@@ -1092,8 +1280,14 @@ int viewer_get_packet(struct relay_command *cmd)
        if (stream->read_fd < 0) {
                char fullpath[PATH_MAX];
 
-               ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name,
-                               stream->channel_name);
+               if (stream->tracefile_count > 0) {
+                       ret = snprintf(fullpath, PATH_MAX, "%s/%s_%" PRIu64, stream->path_name,
+                                       stream->channel_name,
+                                       stream->tracefile_count_current);
+               } else {
+                       ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name,
+                                       stream->channel_name);
+               }
                if (ret < 0) {
                        goto error;
                }
@@ -1105,14 +1299,19 @@ int viewer_get_packet(struct relay_command *cmd)
                stream->read_fd = ret;
        }
 
-       memset(&reply, 0, sizeof(reply));
-
-       if (!stream->ctf_trace->metadata_received ||
-                       stream->ctf_trace->metadata_received >
-                       stream->ctf_trace->metadata_sent) {
+       if (!ctf_trace->metadata_received ||
+                       ctf_trace->metadata_received > ctf_trace->metadata_sent) {
                reply.status = htobe32(VIEWER_GET_PACKET_ERR);
                reply.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
+               goto send_reply;
+       }
 
+       ret = check_new_streams(stream->session_id, sessions_ht);
+       if (ret < 0) {
+               goto end_unlock;
+       } else if (ret == 1) {
+               reply.status = htobe32(VIEWER_GET_PACKET_ERR);
+               reply.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
                goto send_reply;
        }
 
@@ -1125,14 +1324,32 @@ int viewer_get_packet(struct relay_command *cmd)
 
        ret = lseek(stream->read_fd, be64toh(get_packet_info.offset), SEEK_SET);
        if (ret < 0) {
-               PERROR("lseek");
-               goto error;
+               /*
+                * If the read fd was closed by the streaming side, the
+                * abort_flag will be set to 1, otherwise it is an error.
+                */
+               if (stream->abort_flag == 0) {
+                       PERROR("lseek");
+                       goto error;
+               }
+               reply.status = htobe32(VIEWER_GET_PACKET_EOF);
+               goto send_reply;
        }
-       read_len = read(stream->read_fd, data, len);
-       if (read_len < (ssize_t) len) {
-               PERROR("Relay reading trace file, fd: %d, offset: %" PRIu64,
-                               stream->read_fd, be64toh(get_packet_info.offset));
-               goto error;
+       read_len = lttng_read(stream->read_fd, data, len);
+       if (read_len < len) {
+               /*
+                * If the read fd was closed by the streaming side, the
+                * abort_flag will be set to 1, otherwise it is an error.
+                */
+               if (stream->abort_flag == 0) {
+                       PERROR("Relay reading trace file, fd: %d, offset: %" PRIu64,
+                                       stream->read_fd,
+                                       be64toh(get_packet_info.offset));
+                       goto error;
+               } else {
+                       reply.status = htobe32(VIEWER_GET_PACKET_EOF);
+                       goto send_reply;
+               }
        }
        reply.status = htobe32(VIEWER_GET_PACKET_OK);
        reply.len = htobe32(len);
@@ -1144,18 +1361,22 @@ error:
 
 send_reply:
        reply.flags = htobe32(reply.flags);
-       ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
+
+       health_code_update();
+
+       ret = send_response(cmd->sock, &reply, sizeof(reply));
        if (ret < 0) {
-               ERR("Relay data header to viewer");
                goto end_unlock;
        }
+       health_code_update();
 
        if (send_data) {
-               ret = cmd->sock->ops->sendmsg(cmd->sock, data, len, 0);
+               health_code_update();
+               ret = send_response(cmd->sock, data, len);
                if (ret < 0) {
-                       ERR("Relay send data to viewer");
                        goto end_unlock;
                }
+               health_code_update();
        }
 
        DBG("Sent %u bytes for stream %" PRIu64, len,
@@ -1184,37 +1405,33 @@ int viewer_get_metadata(struct relay_command *cmd)
        struct lttng_viewer_get_metadata request;
        struct lttng_viewer_metadata_packet reply;
        struct relay_viewer_stream *stream;
+       struct ctf_trace *ctf_trace;
 
        assert(cmd);
 
        DBG("Relay get metadata");
 
-       if (cmd->version_check_done == 0) {
-               ERR("Trying to get metadata before version check");
-               ret = -1;
-               goto end;
-       }
+       health_code_update();
 
-       ret = cmd->sock->ops->recvmsg(cmd->sock, &request,
-                       sizeof(request), 0);
-       if (ret < 0 || ret != sizeof(request)) {
-               ret = -1;
-               ERR("Relay didn't receive the whole packet");
+       ret = recv_request(cmd->sock, &request, sizeof(request));
+       if (ret < 0) {
                goto end;
        }
+       health_code_update();
 
        rcu_read_lock();
-       stream = live_find_viewer_stream_by_id(be64toh(request.stream_id));
+       stream = viewer_stream_find_by_id(be64toh(request.stream_id));
        if (!stream || !stream->metadata_flag) {
                ERR("Invalid metadata stream");
                goto error;
        }
-       assert(stream->ctf_trace);
-       assert(stream->ctf_trace->metadata_sent <=
-                       stream->ctf_trace->metadata_received);
 
-       len = stream->ctf_trace->metadata_received -
-               stream->ctf_trace->metadata_sent;
+       ctf_trace = ctf_trace_find_by_path(cmd->session->ctf_traces_ht,
+                       stream->path_name);
+       assert(ctf_trace);
+       assert(ctf_trace->metadata_sent <= ctf_trace->metadata_received);
+
+       len = ctf_trace->metadata_received - ctf_trace->metadata_sent;
        if (len == 0) {
                reply.status = htobe32(VIEWER_NO_NEW_METADATA);
                goto send_reply;
@@ -1244,12 +1461,12 @@ int viewer_get_metadata(struct relay_command *cmd)
                goto error;
        }
 
-       read_len = read(stream->read_fd, data, len);
-       if (read_len < (ssize_t) len) {
+       read_len = lttng_read(stream->read_fd, data, len);
+       if (read_len < len) {
                PERROR("Relay reading metadata file");
                goto error;
        }
-       stream->ctf_trace->metadata_sent += read_len;
+       ctf_trace->metadata_sent += read_len;
        reply.status = htobe32(VIEWER_METADATA_OK);
        goto send_reply;
 
@@ -1257,16 +1474,16 @@ error:
        reply.status = htobe32(VIEWER_METADATA_ERR);
 
 send_reply:
-       ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
+       health_code_update();
+       ret = send_response(cmd->sock, &reply, sizeof(reply));
        if (ret < 0) {
-               ERR("Relay data header to viewer");
                goto end_unlock;
        }
+       health_code_update();
 
        if (len > 0) {
-               ret = cmd->sock->ops->sendmsg(cmd->sock, data, len, 0);
+               ret = send_response(cmd->sock, data, len);
                if (ret < 0) {
-                       ERR("Relay send data to viewer");
                        goto end_unlock;
                }
        }
@@ -1290,14 +1507,9 @@ static
 void live_relay_unknown_command(struct relay_command *cmd)
 {
        struct lttcomm_relayd_generic_reply reply;
-       int ret;
 
        reply.ret_code = htobe32(LTTNG_ERR_UNK);
-       ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
-                       sizeof(struct lttcomm_relayd_generic_reply), 0);
-       if (ret < 0) {
-               ERR("Relay sending unknown command");
-       }
+       (void) send_response(cmd->sock, &reply, sizeof(reply));
 }
 
 /*
@@ -1308,8 +1520,25 @@ int process_control(struct lttng_viewer_cmd *recv_hdr,
                struct relay_command *cmd, struct lttng_ht *sessions_ht)
 {
        int ret = 0;
+       uint32_t msg_value;
+
+       assert(recv_hdr);
+       assert(cmd);
+       assert(sessions_ht);
 
-       switch (be32toh(recv_hdr->cmd)) {
+       msg_value = be32toh(recv_hdr->cmd);
+
+       /*
+        * Make sure we've done the version check before any command other then a
+        * new client connection.
+        */
+       if (msg_value != VIEWER_CONNECT && !cmd->version_check_done) {
+               ERR("Viewer cmd value %" PRIu32 " before version check", msg_value);
+               ret = -1;
+               goto end;
+       }
+
+       switch (msg_value) {
        case VIEWER_CONNECT:
                ret = viewer_connect(cmd);
                break;
@@ -1323,11 +1552,14 @@ int process_control(struct lttng_viewer_cmd *recv_hdr,
                ret = viewer_get_next_index(cmd, sessions_ht);
                break;
        case VIEWER_GET_PACKET:
-               ret = viewer_get_packet(cmd);
+               ret = viewer_get_packet(cmd, sessions_ht);
                break;
        case VIEWER_GET_METADATA:
                ret = viewer_get_metadata(cmd);
                break;
+       case VIEWER_GET_NEW_STREAMS:
+               ret = viewer_get_new_streams(cmd, sessions_ht);
+               break;
        default:
                ERR("Received unknown viewer command (%u)", be32toh(recv_hdr->cmd));
                live_relay_unknown_command(cmd);
@@ -1375,10 +1607,8 @@ int add_connection(int fd, struct lttng_poll_event *events,
                goto error;
        }
 
-       do {
-               ret = read(fd, relay_connection, sizeof(*relay_connection));
-       } while (ret < 0 && errno == EINTR);
-       if (ret < 0 || ret < sizeof(*relay_connection)) {
+       ret = lttng_read(fd, relay_connection, sizeof(*relay_connection));
+       if (ret < sizeof(*relay_connection)) {
                PERROR("read relay cmd pipe");
                goto error_read;
        }
@@ -1405,76 +1635,60 @@ void deferred_free_connection(struct rcu_head *head)
        struct relay_command *relay_connection =
                caa_container_of(head, struct relay_command, rcu_node);
 
-       if (relay_connection->session &&
-                       relay_connection->session->viewer_attached > 0) {
-               relay_connection->session->viewer_attached--;
-       }
        lttcomm_destroy_sock(relay_connection->sock);
        free(relay_connection);
 }
 
-static
-void deferred_free_viewer_stream(struct rcu_head *head)
-{
-       struct relay_viewer_stream *stream =
-               caa_container_of(head, struct relay_viewer_stream, rcu_node);
-
-       if (stream->ctf_trace) {
-               uatomic_dec(&stream->ctf_trace->refcount);
-               assert(uatomic_read(&stream->ctf_trace->refcount) >= 0);
-               if (uatomic_read(&stream->ctf_trace->refcount) == 0) {
-                       DBG("Freeing ctf_trace %" PRIu64, stream->ctf_trace->id);
-                       free(stream->ctf_trace);
-               }
-       }
-
-       free(stream->path_name);
-       free(stream->channel_name);
-       free(stream);
-}
-
-static
-void viewer_del_streams(uint64_t session_id)
+/*
+ * Delete all streams for a specific session ID.
+ */
+static void destroy_viewer_streams_by_session(struct relay_session *session)
 {
-       int ret;
        struct relay_viewer_stream *stream;
-       struct lttng_ht_node_u64 *node;
        struct lttng_ht_iter iter;
 
+       assert(session);
+
        rcu_read_lock();
-       cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, node, node) {
-               node = lttng_ht_iter_get_node_u64(&iter);
-               if (!node) {
-                       continue;
-               }
+       cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, stream,
+                       stream_n.node) {
+               struct ctf_trace *ctf_trace;
 
-               stream = caa_container_of(node, struct relay_viewer_stream, stream_n);
-               if (stream->session_id != session_id) {
+               health_code_update();
+               if (stream->session_id != session->id) {
                        continue;
                }
 
-               if (stream->read_fd > 0) {
-                       ret = close(stream->read_fd);
-                       if (ret < 0) {
-                               PERROR("close read_fd");
-                       }
-               }
-               if (stream->index_read_fd > 0) {
-                       ret = close(stream->index_read_fd);
-                       if (ret < 0) {
-                               PERROR("close index_read_fd");
-                       }
-               }
-               if (stream->metadata_flag && stream->ctf_trace) {
-                       stream->ctf_trace->metadata_sent = 0;
+               ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
+                               stream->path_name);
+               assert(ctf_trace);
+
+               viewer_stream_delete(stream);
+
+               if (stream->metadata_flag) {
+                       ctf_trace->metadata_sent = 0;
+                       ctf_trace->viewer_metadata_stream = NULL;
                }
-               ret = lttng_ht_del(viewer_streams_ht, &iter);
-               assert(!ret);
-               call_rcu(&stream->rcu_node, deferred_free_viewer_stream);
+
+               viewer_stream_destroy(ctf_trace, stream);
        }
        rcu_read_unlock();
 }
 
+static void try_destroy_streams(struct relay_session *session)
+{
+       struct ctf_trace *ctf_trace;
+       struct lttng_ht_iter iter;
+
+       assert(session);
+
+       cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
+                       node.node) {
+               /* Attempt to destroy the ctf trace of that session. */
+               ctf_trace_try_destroy(session, ctf_trace);
+       }
+}
+
 /*
  * Delete and free a connection.
  *
@@ -1482,18 +1696,35 @@ void viewer_del_streams(uint64_t session_id)
  */
 static
 void del_connection(struct lttng_ht *relay_connections_ht,
-               struct lttng_ht_iter *iter, struct relay_command *relay_connection)
+               struct lttng_ht_iter *iter, struct relay_command *relay_connection,
+               struct lttng_ht *sessions_ht)
 {
        int ret;
+       struct relay_session *session;
 
        assert(relay_connections_ht);
        assert(iter);
        assert(relay_connection);
+       assert(sessions_ht);
 
+       DBG("Cleaning connection of session ID %" PRIu64,
+                       relay_connection->session_id);
+
+       rcu_read_lock();
        ret = lttng_ht_del(relay_connections_ht, iter);
        assert(!ret);
 
-       viewer_del_streams(relay_connection->session_id);
+       session = session_find_by_id(sessions_ht, relay_connection->session_id);
+       if (session) {
+               /*
+                * Very important that this is done before destroying the session so we
+                * can put back every viewer stream reference from the ctf_trace.
+                */
+               destroy_viewer_streams_by_session(session);
+               try_destroy_streams(session);
+               session_viewer_try_destroy(sessions_ht, session);
+       }
+       rcu_read_unlock();
 
        call_rcu(&relay_connection->rcu_node, deferred_free_connection);
 }
@@ -1519,6 +1750,12 @@ void *thread_worker(void *data)
 
        rcu_register_thread();
 
+       health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_WORKER);
+
+       if (testpoint(relayd_thread_live_worker)) {
+               goto error_testpoint;
+       }
+
        /* table of connections indexed on socket */
        relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
        if (!relay_connections_ht) {
@@ -1539,9 +1776,13 @@ restart:
        while (1) {
                int i;
 
+               health_code_update();
+
                /* Infinite blocking call, waiting for transmission */
                DBG3("Relayd live viewer worker thread polling...");
+               health_poll_entry();
                ret = lttng_poll_wait(&events, -1);
+               health_poll_exit();
                if (ret < 0) {
                        /*
                         * Restart interrupted system call.
@@ -1564,8 +1805,10 @@ restart:
                        uint32_t revents = LTTNG_POLL_GETEV(&events, i);
                        int pollfd = LTTNG_POLL_GETFD(&events, i);
 
+                       health_code_update();
+
                        /* Thread quit pipe has been closed. Killing thread. */
-                       ret = check_thread_quit_pipe(pollfd, revents);
+                       ret = check_live_conn_pipe(pollfd, revents);
                        if (ret) {
                                err = 0;
                                goto exit;
@@ -1600,12 +1843,12 @@ restart:
                                if (revents & (LPOLLERR)) {
                                        cleanup_poll_connection(&events, pollfd);
                                        del_connection(relay_connections_ht, &iter,
-                                                       relay_connection);
+                                                       relay_connection, relay_ctx->sessions_ht);
                                } else if (revents & (LPOLLHUP | LPOLLRDHUP)) {
                                        DBG("Viewer socket %d hung up", pollfd);
                                        cleanup_poll_connection(&events, pollfd);
                                        del_connection(relay_connections_ht, &iter,
-                                                       relay_connection);
+                                                       relay_connection, relay_ctx->sessions_ht);
                                } else if (revents & LPOLLIN) {
                                        ret = relay_connection->sock->ops->recvmsg(
                                                        relay_connection->sock, &recv_hdr,
@@ -1614,8 +1857,8 @@ restart:
                                        /* connection closed */
                                        if (ret <= 0) {
                                                cleanup_poll_connection(&events, pollfd);
-                                               del_connection( relay_connections_ht, &iter,
-                                                               relay_connection);
+                                               del_connection(relay_connections_ht, &iter,
+                                                               relay_connection, relay_ctx->sessions_ht);
                                                DBG("Viewer control connection closed with %d",
                                                                pollfd);
                                        } else {
@@ -1630,7 +1873,7 @@ restart:
                                                        /* Clear the session on error. */
                                                        cleanup_poll_connection(&events, pollfd);
                                                        del_connection(relay_connections_ht, &iter,
-                                                                       relay_connection);
+                                                                       relay_connection, relay_ctx->sessions_ht);
                                                        DBG("Viewer connection closed with %d", pollfd);
                                                }
                                        }
@@ -1647,6 +1890,8 @@ error:
        /* empty the hash table and free the memory */
        rcu_read_lock();
        cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, node, node) {
+               health_code_update();
+
                node = lttng_ht_iter_get_node_ulong(&iter);
                if (!node) {
                        continue;
@@ -1654,7 +1899,8 @@ error:
 
                relay_connection = caa_container_of(node, struct relay_command,
                                sock_n);
-               del_connection(relay_connections_ht, &iter, relay_connection);
+               del_connection(relay_connections_ht, &iter, relay_connection,
+                               relay_ctx->sessions_ht);
        }
        rcu_read_unlock();
 error_poll_create:
@@ -1666,6 +1912,12 @@ relay_connections_ht_error:
                DBG("Viewer worker thread exited with error");
        }
        DBG("Viewer worker thread cleanup complete");
+error_testpoint:
+       if (err) {
+               health_error();
+               ERR("Health error occurred in %s", __func__);
+       }
+       health_unregister(health_relayd);
        stop_threads();
        rcu_unregister_thread();
        return NULL;
@@ -1684,7 +1936,7 @@ static int create_relay_cmd_pipe(void)
        return ret;
 }
 
-void live_stop_threads()
+void live_stop_threads(void)
 {
        int ret;
        void *status;
@@ -1728,11 +1980,6 @@ int live_start_threads(struct lttng_uri *uri,
        assert(uri);
        live_uri = uri;
 
-       /* Create thread quit pipe */
-       if ((ret = init_thread_quit_pipe()) < 0) {
-               goto error;
-       }
-
        /* Check if daemon is UID = 0 */
        is_root = !getuid();
 
This page took 0.044592 seconds and 4 git commands to generate.