X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-relayd%2Flive.c;h=8c4f501aa3446afd5b5cb5d7e72e42bf82dd2d11;hp=1a691a849c11e972fee17059df4d5c7068846f01;hb=d995f382e348c509c15673aba195c3ecb624391c;hpb=4a9daf1745ccbd2aab029206a664f39fcbd640ce diff --git a/src/bin/lttng-relayd/live.c b/src/bin/lttng-relayd/live.c index 1a691a849..8c4f501aa 100644 --- a/src/bin/lttng-relayd/live.c +++ b/src/bin/lttng-relayd/live.c @@ -1,6 +1,7 @@ /* * Copyright (C) 2013 - Julien Desfossez * David Goulet + * 2015 - Mathieu Desnoyers * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License, version 2 only, @@ -16,7 +17,7 @@ * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. */ -#define _GNU_SOURCE +#define _LGPL_SOURCE #include #include #include @@ -35,16 +36,18 @@ #include #include #include +#include #include #include -#include #include #include #include #include +#include #include #include +#include #include #include #include @@ -54,10 +57,17 @@ #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" +#include "connection.h" +#include "viewer-session.h" + +#define SESSION_BUF_DEFAULT_COUNT 16 static struct lttng_uri *live_uri; @@ -65,7 +75,7 @@ static struct lttng_uri *live_uri; * This pipe is used to inform the worker thread that a command is queued and * ready to be processed. */ -static int live_relay_cmd_pipe[2] = { -1, -1 }; +static int live_conn_pipe[2] = { -1, -1 }; /* Shared between threads */ static int live_dispatch_thread_exit; @@ -80,15 +90,17 @@ static pthread_t live_worker_thread; * The live_thread_listener and live_thread_dispatcher communicate with this * queue. */ -static struct relay_cmd_queue viewer_cmd_queue; +static struct relay_conn_queue viewer_conn_queue; static uint64_t last_relay_viewer_session_id; +static pthread_mutex_t last_relay_viewer_session_id_lock = + PTHREAD_MUTEX_INITIALIZER; /* * Cleanup the daemon */ static -void cleanup(void) +void cleanup_relayd_live(void) { DBG("Cleaning up"); @@ -96,39 +108,285 @@ void cleanup(void) } /* - * 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) { ssize_t ret; - ret = lttng_write(wpipe, "!", 1); - if (ret < 1) { - PERROR("write poll pipe"); + 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 (int) ret; + 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(thread_quit_pipe[1]); + ret = sock->ops->sendmsg(sock, buf, size, 0); if (ret < 0) { - ERR("write error on thread quit pipe"); + ERR("Relayd failed to send response."); + } + + return ret; +} + +/* + * Atomically check if new streams got added in one of the sessions attached + * 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(struct relay_connection *conn) +{ + struct relay_session *session; + unsigned long current_val; + int ret = 0; + + if (!conn->viewer_session) { + goto end; + } + rcu_read_lock(); + cds_list_for_each_entry_rcu(session, + &conn->viewer_session->session_list, + viewer_session_node) { + if (!session_get(session)) { + continue; + } + current_val = uatomic_cmpxchg(&session->new_streams, 1, 0); + ret = current_val; + session_put(session); + if (ret == 1) { + goto end; + } + } +end: + rcu_read_unlock(); + 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, + uint64_t session_id, unsigned int ignore_sent_flag) +{ + ssize_t ret; + struct lttng_viewer_stream send_stream; + struct lttng_ht_iter iter; + struct relay_viewer_stream *vstream; + + 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(); + + if (!viewer_stream_get(vstream)) { + continue; + } + + pthread_mutex_lock(&vstream->stream->lock); + /* Ignore if not the same session. */ + if (vstream->stream->trace->session->id != session_id || + (!ignore_sent_flag && vstream->sent_flag)) { + pthread_mutex_unlock(&vstream->stream->lock); + viewer_stream_put(vstream); + continue; + } + + ctf_trace = vstream->stream->trace; + send_stream.id = htobe64(vstream->stream->stream_handle); + send_stream.ctf_trace_id = htobe64(ctf_trace->id); + send_stream.metadata_flag = htobe32( + vstream->stream->is_metadata); + if (lttng_strncpy(send_stream.path_name, vstream->path_name, + sizeof(send_stream.path_name))) { + pthread_mutex_unlock(&vstream->stream->lock); + viewer_stream_put(vstream); + ret = -1; /* Error. */ + goto end_unlock; + } + if (lttng_strncpy(send_stream.channel_name, + vstream->channel_name, + sizeof(send_stream.channel_name))) { + pthread_mutex_unlock(&vstream->stream->lock); + viewer_stream_put(vstream); + ret = -1; /* Error. */ + goto end_unlock; + } + + DBG("Sending stream %" PRIu64 " to viewer", + vstream->stream->stream_handle); + vstream->sent_flag = 1; + pthread_mutex_unlock(&vstream->stream->lock); + + ret = send_response(sock, &send_stream, sizeof(send_stream)); + viewer_stream_put(vstream); + if (ret < 0) { + goto end_unlock; + } + } + + ret = 0; + +end_unlock: + rcu_read_unlock(); + return ret; +} + +/* + * 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. + * + * session must be locked to ensure that we see either none or all initial + * streams for a session, but no intermediate state.. + * + * Return 0 on success or else a negative value. + */ +static int make_viewer_streams(struct relay_session *session, + struct lttng_trace_chunk *viewer_trace_chunk, + enum lttng_viewer_seek seek_t, + uint32_t *nb_total, + uint32_t *nb_unsent, + uint32_t *nb_created, + bool *closed) +{ + int ret; + struct lttng_ht_iter iter; + struct ctf_trace *ctf_trace; + + assert(session); + ASSERT_LOCKED(session->lock); + + if (session->connection_closed) { + *closed = true; + } + + /* + * 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_get(ctf_trace)) { + continue; + } + + cds_list_for_each_entry_rcu(stream, &ctf_trace->stream_list, stream_node) { + struct relay_viewer_stream *vstream; + + if (!stream_get(stream)) { + continue; + } + /* + * stream published is protected by the session lock. + */ + if (!stream->published) { + goto next; + } + vstream = viewer_stream_get_by_id(stream->stream_handle); + if (!vstream) { + vstream = viewer_stream_create(stream, + viewer_trace_chunk, seek_t); + if (!vstream) { + ret = -1; + ctf_trace_put(ctf_trace); + stream_put(stream); + goto error_unlock; + } + + if (nb_created) { + /* Update number of created stream counter. */ + (*nb_created)++; + } + /* + * Ensure a self-reference is preserved even + * after we have put our local reference. + */ + if (!viewer_stream_get(vstream)) { + ERR("Unable to get self-reference on viewer stream, logic error."); + abort(); + } + } else { + if (!vstream->sent_flag && nb_unsent) { + /* Update number of unsent stream counter. */ + (*nb_unsent)++; + } + } + /* Update number of total stream counter. */ + if (nb_total) { + if (stream->is_metadata) { + if (!stream->closed || + stream->metadata_received > vstream->metadata_sent) { + (*nb_total)++; + } + } else { + if (!stream->closed || + !(((int64_t) (stream->prev_data_seq - stream->last_net_seq_num)) >= 0)) { + + (*nb_total)++; + } + } + } + /* Put local reference. */ + viewer_stream_put(vstream); + next: + stream_put(stream); + } + ctf_trace_put(ctf_trace); } - /* Dispatch thread */ + ret = 0; + +error_unlock: + rcu_read_unlock(); + return ret; +} + +int relayd_live_stop(void) +{ + /* Stop dispatch thread */ CMM_STORE_SHARED(live_dispatch_thread_exit, 1); - futex_nto1_wake(&viewer_cmd_queue.futex); + futex_nto1_wake(&viewer_conn_queue.futex); + return 0; } /* @@ -195,10 +453,11 @@ struct lttcomm_sock *init_socket(struct lttng_uri *uri) if (ret < 0) { goto error; } - DBG("Listening on sock %d for live", sock->fd); + DBG("Listening on sock %d for lttng-live", sock->fd); ret = sock->ops->bind(sock); if (ret < 0) { + PERROR("Failed to bind lttng-live socket"); goto error; } @@ -224,7 +483,6 @@ static void *thread_listener(void *data) { int i, ret, pollfd, err = -1; - int val = 1; uint32_t revents, nb_fd; struct lttng_poll_event events; struct lttcomm_sock *live_control_sock; @@ -293,51 +551,55 @@ restart: goto exit; } - if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { - ERR("socket poll error"); - goto error; - } else if (revents & LPOLLIN) { + if (revents & LPOLLIN) { /* - * Get allocated in this thread, enqueued to a global queue, - * dequeued and freed in the worker thread. + * A new connection is requested, therefore a + * viewer connection is allocated in this + * thread, enqueued to a global queue and + * dequeued (and freed) in the worker thread. */ - struct relay_command *relay_cmd; + int val = 1; + struct relay_connection *new_conn; struct lttcomm_sock *newsock; - relay_cmd = zmalloc(sizeof(*relay_cmd)); - if (!relay_cmd) { - PERROR("relay command zmalloc"); - goto error; - } - - assert(pollfd == live_control_sock->fd); newsock = live_control_sock->ops->accept(live_control_sock); if (!newsock) { PERROR("accepting control sock"); - free(relay_cmd); goto error; } DBG("Relay viewer connection accepted socket %d", newsock->fd); + ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val, - sizeof(int)); + sizeof(val)); if (ret < 0) { PERROR("setsockopt inet"); lttcomm_destroy_sock(newsock); - free(relay_cmd); goto error; } - relay_cmd->sock = newsock; + new_conn = connection_create(newsock, RELAY_CONNECTION_UNKNOWN); + if (!new_conn) { + lttcomm_destroy_sock(newsock); + goto error; + } + /* Ownership assumed by the connection. */ + newsock = NULL; - /* - * Lock free enqueue the request. - */ - cds_wfq_enqueue(&viewer_cmd_queue.queue, &relay_cmd->node); + /* Enqueue request for the dispatcher thread. */ + cds_wfcq_enqueue(&viewer_conn_queue.head, &viewer_conn_queue.tail, + &new_conn->qnode); /* - * Wake the dispatch queue futex. Implicit memory - * barrier with the exchange in cds_wfq_enqueue. + * Wake the dispatch queue futex. + * Implicit memory barrier with the + * exchange in cds_wfcq_enqueue. */ - futex_nto1_wake(&viewer_cmd_queue.futex); + futex_nto1_wake(&viewer_conn_queue.futex); + } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { + ERR("socket poll error"); + goto error; + } else { + ERR("Unexpected poll events %u for sock %d", revents, pollfd); + goto error; } } } @@ -362,7 +624,9 @@ error_sock_control: } health_unregister(health_relayd); DBG("Live viewer listener thread cleanup complete"); - stop_threads(); + if (lttng_relay_stop_threads()) { + ERR("Error stopping threads"); + } return NULL; } @@ -374,8 +638,8 @@ void *thread_dispatcher(void *data) { int err = -1; ssize_t ret; - struct cds_wfq_node *node; - struct relay_command *relay_cmd = NULL; + struct cds_wfcq_node *node; + struct relay_connection *conn = NULL; DBG("[thread] Live viewer relay dispatcher started"); @@ -387,45 +651,49 @@ void *thread_dispatcher(void *data) health_code_update(); - while (!CMM_LOAD_SHARED(live_dispatch_thread_exit)) { + for (;;) { health_code_update(); /* Atomically prepare the queue futex */ - futex_nto1_prepare(&viewer_cmd_queue.futex); + futex_nto1_prepare(&viewer_conn_queue.futex); + + if (CMM_LOAD_SHARED(live_dispatch_thread_exit)) { + break; + } do { health_code_update(); /* Dequeue commands */ - node = cds_wfq_dequeue_blocking(&viewer_cmd_queue.queue); + node = cds_wfcq_dequeue_blocking(&viewer_conn_queue.head, + &viewer_conn_queue.tail); if (node == NULL) { DBG("Woken up but nothing in the live-viewer " "relay command queue"); /* Continue thread execution */ break; } - - relay_cmd = caa_container_of(node, struct relay_command, node); + conn = caa_container_of(node, struct relay_connection, qnode); DBG("Dispatching viewer request waiting on sock %d", - relay_cmd->sock->fd); + conn->sock->fd); /* - * Inform worker thread of the new request. This call is blocking - * so we can be assured that the data will be read at some point in - * time or wait to the end of the world :) + * Inform worker thread of the new request. This + * call is blocking so we can be assured that + * the data will be read at some point in time + * or wait to the end of the world :) */ - ret = lttng_write(live_relay_cmd_pipe[1], relay_cmd, - sizeof(*relay_cmd)); - free(relay_cmd); - if (ret < sizeof(struct relay_command)) { - PERROR("write cmd pipe"); + ret = lttng_write(live_conn_pipe[1], &conn, sizeof(conn)); + if (ret < 0) { + PERROR("write conn pipe"); + connection_put(conn); goto error; } } while (node != NULL); /* Futex wait on queue. Blocking call on futex() */ health_poll_entry(); - futex_nto1_wait(&viewer_cmd_queue.futex); + futex_nto1_wait(&viewer_conn_queue.futex); health_poll_exit(); } @@ -440,7 +708,9 @@ error_testpoint: } health_unregister(health_relayd); DBG("Live viewer dispatch thread dying"); - stop_threads(); + if (lttng_relay_stop_threads()) { + ERR("Error stopping threads"); + } return NULL; } @@ -450,55 +720,48 @@ error_testpoint: * Return 0 on success or else negative value. */ static -int viewer_connect(struct relay_command *cmd) +int viewer_connect(struct relay_connection *conn) { int ret; struct lttng_viewer_connect reply, msg; - assert(cmd); - - cmd->version_check_done = 1; + conn->version_check_done = 1; health_code_update(); - /* 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; + DBG("Viewer is establishing a connection to the relayd."); + + ret = recv_request(conn->sock, &msg, sizeof(msg)); + if (ret < 0) { goto end; } health_code_update(); + memset(&reply, 0, sizeof(reply)); 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)); + DBG("Incompatible major versions ([relayd] %u vs [client] %u)", + reply.major, be32toh(msg.major)); ret = -1; goto end; } - cmd->major = reply.major; + conn->major = reply.major; /* We adapt to the lowest compatible version */ if (reply.minor <= be32toh(msg.minor)) { - cmd->minor = reply.minor; + conn->minor = reply.minor; } else { - cmd->minor = be32toh(msg.minor); + conn->minor = be32toh(msg.minor); } - if (be32toh(msg.type) == VIEWER_CLIENT_COMMAND) { - cmd->type = RELAY_VIEWER_COMMAND; - } else if (be32toh(msg.type) == VIEWER_CLIENT_NOTIFICATION) { - cmd->type = RELAY_VIEWER_NOTIFICATION; + if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_COMMAND) { + conn->type = RELAY_VIEWER_COMMAND; + } else if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_NOTIFICATION) { + conn->type = RELAY_VIEWER_NOTIFICATION; } else { ERR("Unknown connection type : %u", be32toh(msg.type)); ret = -1; @@ -507,21 +770,28 @@ int viewer_connect(struct relay_command *cmd) reply.major = htobe32(reply.major); reply.minor = htobe32(reply.minor); - if (cmd->type == RELAY_VIEWER_COMMAND) { - reply.viewer_session_id = htobe64(++last_relay_viewer_session_id); + if (conn->type == RELAY_VIEWER_COMMAND) { + /* + * Increment outside of htobe64 macro, because the argument can + * be used more than once within the macro, and thus the + * operation may be undefined. + */ + pthread_mutex_lock(&last_relay_viewer_session_id_lock); + last_relay_viewer_session_id++; + pthread_mutex_unlock(&last_relay_viewer_session_id_lock); + reply.viewer_session_id = htobe64(last_relay_viewer_session_id); } health_code_update(); - ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, - sizeof(struct lttng_viewer_connect), 0); + ret = send_response(conn->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); + DBG("Version check done using protocol %u.%u", conn->major, conn->minor); ret = 0; end: @@ -530,650 +800,523 @@ end: /* * Send the viewer the list of current sessions. + * We need to create a copy of the hash table content because otherwise + * we cannot assume the number of entries stays the same between getting + * the number of HT elements and iteration over the HT. * * Return 0 on success or else a negative value. */ static -int viewer_list_sessions(struct relay_command *cmd, - struct lttng_ht *sessions_ht) +int viewer_list_sessions(struct relay_connection *conn) { - int ret; + int ret = 0; 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; + struct lttng_viewer_session *send_session_buf = NULL; + uint32_t buf_count = SESSION_BUF_DEFAULT_COUNT; + uint32_t count = 0; DBG("List sessions received"); - if (cmd->version_check_done == 0) { - ERR("Trying to list sessions before version check"); - ret = -1; - goto end_no_session; + send_session_buf = zmalloc(SESSION_BUF_DEFAULT_COUNT * sizeof(*send_session_buf)); + if (!send_session_buf) { + return -1; } rcu_read_lock(); - cds_lfht_count_nodes(sessions_ht->ht, &approx_before, &count, &approx_after); - session_list.sessions_count = htobe32(count); + cds_lfht_for_each_entry(sessions_ht->ht, &iter.iter, session, + session_n.node) { + struct lttng_viewer_session *send_session; - health_code_update(); - - ret = cmd->sock->ops->sendmsg(cmd->sock, &session_list, - sizeof(session_list), 0); - if (ret < 0) { - ERR("Relay sending sessions list"); - goto end_unlock; - } - - health_code_update(); - - cds_lfht_for_each_entry(sessions_ht->ht, &iter.iter, node, node) { health_code_update(); - node = lttng_ht_iter_get_node_ulong(&iter); - if (!node) { - goto end_unlock; + pthread_mutex_lock(&session->lock); + if (session->connection_closed) { + /* Skip closed session */ + goto next_session; + } + if (!session->current_trace_chunk) { + /* + * Skip un-attachable session. It is either + * being destroyed or has not had a trace + * chunk created against it yet. + */ + goto next_session; } - session = caa_container_of(node, struct relay_session, session_n); - - strncpy(send_session.session_name, session->session_name, - sizeof(send_session.session_name)); - strncpy(send_session.hostname, session->hostname, - 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.streams = htobe32(session->stream_count); - health_code_update(); + if (count >= buf_count) { + struct lttng_viewer_session *newbuf; + uint32_t new_buf_count = buf_count << 1; - ret = cmd->sock->ops->sendmsg(cmd->sock, &send_session, - sizeof(send_session), 0); - if (ret < 0) { - ERR("Relay sending session info"); - goto end_unlock; + newbuf = realloc(send_session_buf, + new_buf_count * sizeof(*send_session_buf)); + if (!newbuf) { + ret = -1; + goto break_loop; + } + send_session_buf = newbuf; + buf_count = new_buf_count; } + send_session = &send_session_buf[count]; + if (lttng_strncpy(send_session->session_name, + session->session_name, + sizeof(send_session->session_name))) { + ret = -1; + goto break_loop; + } + if (lttng_strncpy(send_session->hostname, session->hostname, + sizeof(send_session->hostname))) { + ret = -1; + goto break_loop; + } + send_session->id = htobe64(session->id); + send_session->live_timer = htobe32(session->live_timer); + if (session->viewer_attached) { + send_session->clients = htobe32(1); + } else { + send_session->clients = htobe32(0); + } + send_session->streams = htobe32(session->stream_count); + count++; + next_session: + pthread_mutex_unlock(&session->lock); + continue; + break_loop: + pthread_mutex_unlock(&session->lock); + break; } - health_code_update(); - - rcu_read_unlock(); - ret = 0; - goto end; - -end_unlock: rcu_read_unlock(); + if (ret < 0) { + goto end_free; + } -end: -end_no_session: - return ret; -} + session_list.sessions_count = htobe32(count); -/* - * 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 ctf_packet_index_file_hdr hdr; + health_code_update(); - if (stream->tracefile_count > 0) { - ret = snprintf(fullpath, sizeof(fullpath), "%s/" DEFAULT_INDEX_DIR "/%s_%" - PRIu64 DEFAULT_INDEX_FILE_SUFFIX, stream->path_name, - stream->channel_name, stream->tracefile_count_current); - } else { - ret = snprintf(fullpath, sizeof(fullpath), "%s/" DEFAULT_INDEX_DIR "/%s" - DEFAULT_INDEX_FILE_SUFFIX, stream->path_name, - stream->channel_name); - } + ret = send_response(conn->sock, &session_list, sizeof(session_list)); if (ret < 0) { - PERROR("snprintf index path"); - goto error; + goto end_free; } - DBG("Opening index file %s in read only", fullpath); - ret = open(fullpath, O_RDONLY); + health_code_update(); + + ret = send_response(conn->sock, send_session_buf, + count * sizeof(*send_session_buf)); if (ret < 0) { - if (errno == ENOENT) { - ret = -ENOENT; - goto error; - } else { - PERROR("opening index in read-only"); - } - goto error; + goto end_free; } - stream->index_read_fd = ret; - DBG("Opening index file %s in read only, (fd: %d)", fullpath, ret); + health_code_update(); - ret = lttng_read(stream->index_read_fd, &hdr, sizeof(hdr)); - if (ret < sizeof(hdr)) { - PERROR("Reading index header"); - goto error; - } - if (be32toh(hdr.magic) != CTF_INDEX_MAGIC) { - ERR("Invalid header magic"); - ret = -1; - goto error; - } - if (be32toh(hdr.index_major) != CTF_INDEX_MAJOR || - be32toh(hdr.index_minor) != CTF_INDEX_MINOR) { - ERR("Invalid header version"); - ret = -1; - goto error; - } ret = 0; - -error: +end_free: + free(send_session_buf); 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 streams. */ static -int init_viewer_stream(struct relay_stream *stream, int seek_last) +int viewer_get_new_streams(struct relay_connection *conn) { - int ret; - struct relay_viewer_stream *viewer_stream; + int ret, send_streams = 0; + uint32_t nb_created = 0, nb_unsent = 0, nb_streams = 0, nb_total = 0; + struct lttng_viewer_new_streams_request request; + struct lttng_viewer_new_streams_response response; + struct relay_session *session = NULL; + uint64_t session_id; + bool closed = false; - assert(stream); + assert(conn); - viewer_stream = zmalloc(sizeof(*viewer_stream)); - if (!viewer_stream) { - PERROR("relay viewer stream zmalloc"); - ret = -1; + DBG("Get new streams received"); + + health_code_update(); + + /* Receive the request from the connected client. */ + ret = recv_request(conn->sock, &request, sizeof(request)); + if (ret < 0) { goto error; } - 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->tracefile_count = stream->tracefile_count; - viewer_stream->metadata_flag = stream->metadata_flag; - viewer_stream->tracefile_count_last = -1ULL; - if (seek_last) { - viewer_stream->tracefile_count_current = - stream->tracefile_count_current; - } else { - viewer_stream->tracefile_count_current = - stream->oldest_tracefile_id; + session_id = be64toh(request.session_id); + + health_code_update(); + + memset(&response, 0, sizeof(response)); + + session = session_get_by_id(session_id); + if (!session) { + DBG("Relay session %" PRIu64 " not found", session_id); + response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR); + goto send_reply; } - viewer_stream->ctf_trace = stream->ctf_trace; - if (viewer_stream->metadata_flag) { - viewer_stream->ctf_trace->viewer_metadata_stream = - viewer_stream; + if (!viewer_session_is_attached(conn->viewer_session, session)) { + send_streams = 0; + response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR); + goto send_reply; } - 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); + send_streams = 1; + response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_OK); - viewer_stream->index_read_fd = -1; - viewer_stream->read_fd = -1; + pthread_mutex_lock(&session->lock); + if (!conn->viewer_session->current_trace_chunk && + session->current_trace_chunk) { + ret = viewer_session_set_trace_chunk(conn->viewer_session, + session->current_trace_chunk); + if (ret) { + goto error_unlock_session; + } + } + ret = make_viewer_streams(session, + conn->viewer_session->current_trace_chunk, + LTTNG_VIEWER_SEEK_LAST, &nb_total, &nb_unsent, + &nb_created, &closed); + if (ret < 0) { + goto error_unlock_session; + } + pthread_mutex_unlock(&session->lock); - /* - * 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. - * We also need that for the seek_last. - */ - viewer_stream->total_index_received = stream->total_index_received; + /* Only send back the newly created streams with the unsent ones. */ + nb_streams = nb_created + nb_unsent; + response.streams_count = htobe32(nb_streams); /* - * If we never received an index for the current stream, delay - * the opening of the index, otherwise open it right now. + * If the session is closed, HUP when there are no more streams + * with data. */ - if (viewer_stream->tracefile_count_current == - stream->tracefile_count_current && - viewer_stream->total_index_received == 0) { - viewer_stream->index_read_fd = -1; - } else { - ret = open_index(viewer_stream); - if (ret < 0) { - goto error; - } + if (closed && nb_total == 0) { + send_streams = 0; + response.streams_count = 0; + response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP); + goto send_reply; } - if (seek_last && viewer_stream->index_read_fd > 0) { - ret = lseek(viewer_stream->index_read_fd, - viewer_stream->total_index_received * - sizeof(struct ctf_packet_index), - SEEK_CUR); - if (ret < 0) { - goto error; - } - viewer_stream->last_sent_index = - viewer_stream->total_index_received; +send_reply: + health_code_update(); + ret = send_response(conn->sock, &response, sizeof(response)); + if (ret < 0) { + goto end_put_session; } + health_code_update(); - ret = 0; - -error: - return ret; -} - -/* - * Rotate a stream to the next tracefile. - * - * Returns 0 on success, 1 on EOF, a negative value on error. - */ -static -int rotate_viewer_stream(struct relay_viewer_stream *viewer_stream, - struct relay_stream *stream) -{ - int ret; - uint64_t tracefile_id; - - assert(viewer_stream); - - tracefile_id = (viewer_stream->tracefile_count_current + 1) % - viewer_stream->tracefile_count; /* - * Detect the last tracefile to open. + * Unknown or empty session, just return gracefully, the viewer + * knows what is happening. */ - if (viewer_stream->tracefile_count_last != -1ULL && - viewer_stream->tracefile_count_last == - viewer_stream->tracefile_count_current) { - ret = 1; - goto end; + if (!send_streams || !nb_streams) { + ret = 0; + goto end_put_session; } - if (stream) { - pthread_mutex_lock(&stream->viewer_stream_rotation_lock); - } /* - * The writer and the reader are not working in the same - * tracefile, we can read up to EOF, we don't care about the - * total_index_received. + * 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. */ - if (!stream || (stream->tracefile_count_current != tracefile_id)) { - viewer_stream->close_write_flag = 1; - } else { - /* - * We are opening a file that is still open in write, make - * sure we limit our reading to the number of indexes - * received. - */ - viewer_stream->close_write_flag = 0; - if (stream) { - viewer_stream->total_index_received = - stream->total_index_received; - } - } - viewer_stream->tracefile_count_current = tracefile_id; - - ret = close(viewer_stream->index_read_fd); - if (ret < 0) { - PERROR("close index file %d", - viewer_stream->index_read_fd); - } - viewer_stream->index_read_fd = -1; - ret = close(viewer_stream->read_fd); + ret = send_viewer_streams(conn->sock, session_id, 0); if (ret < 0) { - PERROR("close tracefile %d", - viewer_stream->read_fd); + goto end_put_session; } - viewer_stream->read_fd = -1; - pthread_mutex_lock(&viewer_stream->overwrite_lock); - viewer_stream->abort_flag = 0; - pthread_mutex_unlock(&viewer_stream->overwrite_lock); - - viewer_stream->index_read_fd = -1; - viewer_stream->read_fd = -1; - - if (stream) { - pthread_mutex_unlock(&stream->viewer_stream_rotation_lock); +end_put_session: + if (session) { + session_put(session); } - ret = open_index(viewer_stream); - if (ret < 0) { - goto error; - } - - ret = 0; - -end: error: return ret; +error_unlock_session: + pthread_mutex_unlock(&session->lock); + session_put(session); + return ret; } /* * Send the viewer the list of current sessions. */ static -int viewer_attach_session(struct relay_command *cmd, - struct lttng_ht *sessions_ht) +int viewer_attach_session(struct relay_connection *conn) { - int ret, send_streams = 0; - uint32_t nb_streams = 0, nb_streams_ready = 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; - int seek_last = 0; + struct relay_session *session = NULL; + bool closed = false; + uint64_t session_id; - assert(cmd); - assert(sessions_ht); - - DBG("Attach session received"); - - if (cmd->version_check_done == 0) { - ERR("Trying to attach session before version check"); - ret = -1; - goto end_no_session; - } + assert(conn); health_code_update(); - 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(conn->sock, &request, sizeof(request)); + if (ret < 0) { goto error; } + session_id = be64toh(request.session_id); 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) { - DBG("Relay session %" PRIu64 " not found", - be64toh(request.session_id)); - response.status = htobe32(VIEWER_ATTACH_UNK); + memset(&response, 0, sizeof(response)); + + if (!conn->viewer_session) { + DBG("Client trying to attach before creating a live viewer session"); + response.status = htobe32(LTTNG_VIEWER_ATTACH_NO_SESSION); goto send_reply; } - 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) { - DBG("Already a viewer attached"); - response.status = htobe32(VIEWER_ATTACH_ALREADY); + session = session_get_by_id(session_id); + if (!session) { + DBG("Relay session %" PRIu64 " not found", session_id); + response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK); goto send_reply; - } else if (session->live_timer == 0) { + } + DBG("Attach session ID %" PRIu64 " received", session_id); + + pthread_mutex_lock(&session->lock); + if (session->live_timer == 0) { DBG("Not live session"); - response.status = htobe32(VIEWER_ATTACH_NOT_LIVE); + response.status = htobe32(LTTNG_VIEWER_ATTACH_NOT_LIVE); + goto send_reply; + } + + send_streams = 1; + ret = viewer_session_attach(conn->viewer_session, session); + if (ret) { + DBG("Already a viewer attached"); + response.status = htobe32(LTTNG_VIEWER_ATTACH_ALREADY); goto send_reply; - } else { - session->viewer_attached++; - send_streams = 1; - response.status = htobe32(VIEWER_ATTACH_OK); - cmd->session_id = session->id; - cmd->session = session; } switch (be32toh(request.seek)) { - case VIEWER_SEEK_BEGINNING: - /* Default behaviour. */ - break; - case VIEWER_SEEK_LAST: - seek_last = 1; + case LTTNG_VIEWER_SEEK_BEGINNING: + case LTTNG_VIEWER_SEEK_LAST: + response.status = htobe32(LTTNG_VIEWER_ATTACH_OK); + seek_type = be32toh(request.seek); break; default: ERR("Wrong seek parameter"); - response.status = htobe32(VIEWER_ATTACH_SEEK_ERR); + response.status = htobe32(LTTNG_VIEWER_ATTACH_SEEK_ERR); send_streams = 0; 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; - - health_code_update(); - - 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; - } - nb_streams++; - - /* - * Don't send streams with no ctf_trace, they are not - * ready to be read. - */ - if (!stream->ctf_trace || !stream->viewer_ready) { - continue; - } - nb_streams_ready++; - - vstream = live_find_viewer_stream_by_id(stream->stream_handle); - if (!vstream) { - ret = init_viewer_stream(stream, seek_last); - if (ret < 0) { - goto end_unlock; - } - } + if (!conn->viewer_session->current_trace_chunk && + session->current_trace_chunk) { + ret = viewer_session_set_trace_chunk(conn->viewer_session, + session->current_trace_chunk); + if (ret) { + goto end_put_session; } + } + ret = make_viewer_streams(session, + conn->viewer_session->current_trace_chunk, seek_type, + &nb_streams, NULL, NULL, &closed); + if (ret < 0) { + goto end_put_session; + } + pthread_mutex_unlock(&session->lock); + session_put(session); + session = NULL; - /* We must have the same amount of existing stream and ready stream. */ - if (nb_streams != nb_streams_ready) { - nb_streams = 0; - } - response.streams_count = htobe32(nb_streams); + response.streams_count = htobe32(nb_streams); + /* + * If the session is closed when the viewer is attaching, it + * means some of the streams may have been concurrently removed, + * so we don't allow the viewer to attach, even if there are + * streams available. + */ + if (closed) { + send_streams = 0; + response.streams_count = 0; + response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK); + goto send_reply; } send_reply: health_code_update(); - ret = cmd->sock->ops->sendmsg(cmd->sock, &response, sizeof(response), 0); + ret = send_response(conn->sock, &response, sizeof(response)); if (ret < 0) { - ERR("Relay sending viewer attach response"); - goto end_unlock; + goto end_put_session; } health_code_update(); /* - * Unknown or empty session, just return gracefully, the viewer knows what - * is happening. + * Unknown or empty session, just return gracefully, the viewer + * knows what is happening. */ if (!send_streams || !nb_streams) { ret = 0; - goto end_unlock; + goto end_put_session; } - /* 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) { - health_code_update(); - - 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(conn->sock, session_id, 1); + if (ret < 0) { + goto end_put_session; } - ret = 0; -end_unlock: - rcu_read_unlock(); -end_no_session: +end_put_session: + if (session) { + pthread_mutex_unlock(&session->lock); + session_put(session); + } error: return ret; } /* - * Get viewer stream from stream id. + * Open the index file if needed for the given vstream. * - * RCU read side lock MUST be acquired. + * If an index file is successfully opened, the vstream will set it as its + * current index file. + * + * Return 0 on success, a negative value on error (-ENOENT if not ready yet). + * + * Called with rstream lock held. */ -struct relay_viewer_stream *live_find_viewer_stream_by_id(uint64_t stream_id) +static int try_open_index(struct relay_viewer_stream *vstream, + struct relay_stream *rstream) { - struct lttng_ht_node_u64 *node; - struct lttng_ht_iter iter; - struct relay_viewer_stream *stream = NULL; + int ret = 0; + const uint32_t connection_major = rstream->trace->session->major; + const uint32_t connection_minor = rstream->trace->session->minor; - 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); + if (vstream->index_file) { goto end; } - stream = caa_container_of(node, struct relay_viewer_stream, stream_n); - -end: - return stream; -} - -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); - - free(stream->path_name); - free(stream->channel_name); - free(stream); -} - -static -void delete_viewer_stream(struct relay_viewer_stream *vstream) -{ - int delret; - struct lttng_ht_iter iter; - - iter.iter.node = &vstream->stream_n.node; - delret = lttng_ht_del(viewer_streams_ht, &iter); - assert(!delret); -} - -static -void destroy_viewer_stream(struct relay_viewer_stream *vstream) -{ - unsigned long ret_ref; - int ret; - - assert(vstream); - ret_ref = uatomic_add_return(&vstream->ctf_trace->refcount, -1); - assert(ret_ref >= 0); - - if (vstream->read_fd >= 0) { - ret = close(vstream->read_fd); - if (ret < 0) { - PERROR("close read_fd"); - } - } - if (vstream->index_read_fd >= 0) { - ret = close(vstream->index_read_fd); - if (ret < 0) { - PERROR("close index_read_fd"); - } - } /* - * If the only stream left in the HT is the metadata stream, - * we need to remove it because we won't detect a EOF for this - * stream. + * First time, we open the index file and at least one index is ready. */ - if (ret_ref == 1 && vstream->ctf_trace->viewer_metadata_stream) { - delete_viewer_stream(vstream->ctf_trace->viewer_metadata_stream); - destroy_viewer_stream(vstream->ctf_trace->viewer_metadata_stream); - vstream->ctf_trace->metadata_stream = NULL; - DBG("Freeing ctf_trace %" PRIu64, vstream->ctf_trace->id); - /* - * The streaming-side is already closed and we can't receive a new - * stream concurrently at this point (since the session is being - * destroyed), so when we detect the refcount equals 0, we are the - * only owners of the ctf_trace and we can free it ourself. - */ - free(vstream->ctf_trace); + if (rstream->index_received_seqcount == 0) { + ret = -ENOENT; + goto end; + } + vstream->index_file = lttng_index_file_create_from_trace_chunk_read_only( + vstream->stream_file.trace_chunk, rstream->path_name, + rstream->channel_name, rstream->tracefile_size, + vstream->current_tracefile_id, + lttng_to_index_major(connection_major, connection_minor), + lttng_to_index_minor(connection_major, connection_minor)); + if (!vstream->index_file) { + ret = -1; } - call_rcu(&vstream->rcu_node, deferred_free_viewer_stream); +end: + return ret; } /* - * Atomically check if new streams got added in the session since the last - * check and reset the flag to 0. + * Check the status of the index for the given stream. This function + * updates the index structure if needed and can put (close) the vstream + * in the HUP situation. * - * Returns 1 if new streams got added, 0 if nothing changed, a negative value - * on error. + * Return 0 means that we can proceed with the index. A value of 1 means + * that the index has been updated and is ready to be sent to the + * client. A negative value indicates an error that can't be handled. + * + * Called with rstream lock held. */ -static -int check_new_streams(uint64_t session_id, struct lttng_ht *sessions_ht) +static int check_index_status(struct relay_viewer_stream *vstream, + struct relay_stream *rstream, struct ctf_trace *trace, + struct lttng_viewer_index *index) { - struct lttng_ht_node_ulong *node; - struct lttng_ht_iter iter; - struct relay_session *session; - unsigned long current_val; int ret; - lttng_ht_lookup(sessions_ht, - (void *)((unsigned long) session_id), &iter); - node = lttng_ht_iter_get_node_ulong(&iter); - if (node == NULL) { - DBG("Relay session %" PRIu64 " not found", session_id); - ret = -1; - goto error; + if ((trace->session->connection_closed || rstream->closed) + && rstream->index_received_seqcount + == vstream->index_sent_seqcount) { + /* + * Last index sent and session connection or relay + * stream are closed. + */ + index->status = htobe32(LTTNG_VIEWER_INDEX_HUP); + goto hup; + } else if (rstream->beacon_ts_end != -1ULL && + rstream->index_received_seqcount + == vstream->index_sent_seqcount) { + /* + * We've received a synchronization beacon and the last index + * available has been sent, the index for now is inactive. + * + * In this case, we have received a beacon which allows us to + * inform the client of a time interval during which we can + * guarantee that there are no events to read (and never will + * be). + */ + index->status = htobe32(LTTNG_VIEWER_INDEX_INACTIVE); + index->timestamp_end = htobe64(rstream->beacon_ts_end); + index->stream_id = htobe64(rstream->ctf_stream_id); + goto index_ready; + } else if (rstream->index_received_seqcount + == vstream->index_sent_seqcount) { + /* + * This checks whether received == sent seqcount. In + * this case, we have not received a beacon. Therefore, + * we can only ask the client to retry later. + */ + index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY); + goto index_ready; + } else if (!tracefile_array_seq_in_file(rstream->tfa, + vstream->current_tracefile_id, + vstream->index_sent_seqcount)) { + /* + * The next index we want to send cannot be read either + * because we need to perform a rotation, or due to + * the producer having overwritten its trace file. + */ + DBG("Viewer stream %" PRIu64 " rotation", + vstream->stream->stream_handle); + ret = viewer_stream_rotate(vstream); + if (ret < 0) { + goto end; + } else if (ret == 1) { + /* EOF across entire stream. */ + index->status = htobe32(LTTNG_VIEWER_INDEX_HUP); + goto hup; + } + /* + * If we have been pushed due to overwrite, it + * necessarily means there is data that can be read in + * the stream. If we rotated because we reached the end + * of a tracefile, it means the following tracefile + * needs to contain at least one index, else we would + * have already returned LTTNG_VIEWER_INDEX_RETRY to the + * viewer. The updated index_sent_seqcount needs to + * point to a readable index entry now. + * + * In the case where we "rotate" on a single file, we + * can end up in a case where the requested index is + * still unavailable. + */ + if (rstream->tracefile_count == 1 && + !tracefile_array_seq_in_file( + rstream->tfa, + vstream->current_tracefile_id, + vstream->index_sent_seqcount)) { + index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY); + goto index_ready; + } + assert(tracefile_array_seq_in_file(rstream->tfa, + vstream->current_tracefile_id, + vstream->index_sent_seqcount)); } - - session = caa_container_of(node, struct relay_session, session_n); - - current_val = uatomic_cmpxchg(&session->new_streams, 1, 0); - ret = current_val; - -error: + /* ret == 0 means successful so we continue. */ + ret = 0; +end: return ret; + +hup: + viewer_stream_put(vstream); +index_ready: + return 1; } /* @@ -1182,183 +1325,147 @@ error: * Return 0 on success or else a negative value. */ static -int viewer_get_next_index(struct relay_command *cmd, - struct lttng_ht *sessions_ht) +int viewer_get_next_index(struct relay_connection *conn) { int ret; struct lttng_viewer_get_next_index request_index; struct lttng_viewer_index viewer_index; struct ctf_packet_index packet_index; - struct relay_viewer_stream *vstream; - struct relay_stream *rstream; + struct relay_viewer_stream *vstream = NULL; + struct relay_stream *rstream = NULL; + struct ctf_trace *ctf_trace = NULL; + struct relay_viewer_stream *metadata_viewer_stream = NULL; - assert(cmd); - assert(sessions_ht); + assert(conn); 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; - } - + memset(&viewer_index, 0, sizeof(viewer_index)); 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(conn->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)); + vstream = viewer_stream_get_by_id(be64toh(request_index.stream_id)); if (!vstream) { - ret = -1; - goto end_unlock; + DBG("Client requested index of unknown stream id %" PRIu64, + (uint64_t) be64toh(request_index.stream_id)); + viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR); + goto send_reply; } - memset(&viewer_index, 0, sizeof(viewer_index)); + /* Use back. ref. Protected by refcounts. */ + rstream = vstream->stream; + ctf_trace = rstream->trace; + + /* metadata_viewer_stream may be NULL. */ + metadata_viewer_stream = + ctf_trace_get_viewer_metadata_stream(ctf_trace); + + pthread_mutex_lock(&rstream->lock); /* * The viewer should not ask for index on metadata stream. */ - if (vstream->metadata_flag) { - viewer_index.status = htobe32(VIEWER_INDEX_HUP); + if (rstream->is_metadata) { + viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP); goto send_reply; } - /* First time, we open the index file */ - if (vstream->index_read_fd < 0) { - ret = open_index(vstream); + /* Try to open an index if one is needed for that stream. */ + ret = try_open_index(vstream, rstream); + if (ret < 0) { 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 + * The index is created only when the first data + * packet arrives, it might not be ready at the + * beginning of the session */ - viewer_index.status = htobe32(VIEWER_INDEX_RETRY); - goto send_reply; - } else if (ret < 0) { - viewer_index.status = htobe32(VIEWER_INDEX_ERR); - goto send_reply; + viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY); + } else { + /* Unhandled error. */ + viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR); } + goto send_reply; } - rstream = relay_stream_find_by_id(vstream->stream_handle); - if (rstream) { - if (vstream->abort_flag) { - /* Rotate on abort (overwrite). */ - DBG("Viewer rotate because of overwrite"); - ret = rotate_viewer_stream(vstream, rstream); - if (ret < 0) { - goto end_unlock; - } else if (ret == 1) { - viewer_index.status = htobe32(VIEWER_INDEX_HUP); - delete_viewer_stream(vstream); - destroy_viewer_stream(vstream); - 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; - /* - * 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. - */ - } else if (rstream->total_index_received <= vstream->last_sent_index - && !vstream->close_write_flag) { - 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; - } - } - pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock); - } else if (!rstream && vstream->close_write_flag && - vstream->total_index_received == vstream->last_sent_index) { - /* Last index sent and current tracefile closed in write */ - viewer_index.status = htobe32(VIEWER_INDEX_HUP); - delete_viewer_stream(vstream); - destroy_viewer_stream(vstream); + ret = check_index_status(vstream, rstream, ctf_trace, &viewer_index); + if (ret < 0) { + goto error_put; + } else if (ret == 1) { + /* + * We have no index to send and check_index_status has populated + * viewer_index's status. + */ goto send_reply; - } else { - vstream->close_write_flag = 1; } + /* At this point, ret is 0 thus we will be able to read the index. */ + assert(!ret); - if (!vstream->ctf_trace->metadata_received || - vstream->ctf_trace->metadata_received > - vstream->ctf_trace->metadata_sent) { - viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA; + /* + * vstream->stream_fd may be NULL if it has been closed by + * tracefile rotation, or if we are at the beginning of the + * stream. We open the data stream file here to protect against + * overwrite caused by tracefile rotation (in association with + * unlink performed before overwrite). + */ + if (!vstream->stream_file.fd) { + int fd; + char file_path[LTTNG_PATH_MAX]; + enum lttng_trace_chunk_status status; + + ret = utils_stream_file_path(rstream->path_name, + rstream->channel_name, rstream->tracefile_size, + vstream->current_tracefile_id, NULL, file_path, + sizeof(file_path)); + if (ret < 0) { + goto error_put; + } + + status = lttng_trace_chunk_open_file( + vstream->stream_file.trace_chunk, + file_path, O_RDONLY, 0, &fd); + if (status != LTTNG_TRACE_CHUNK_STATUS_OK) { + PERROR("Failed to open trace file for viewer stream"); + goto error_put; + } + vstream->stream_file.fd = stream_fd_create(fd); + if (!vstream->stream_file.fd) { + if (close(fd)) { + PERROR("Failed to close viewer stream file"); + } + goto error_put; + } } - ret = check_new_streams(vstream->session_id, sessions_ht); + ret = check_new_streams(conn); if (ret < 0) { - goto end_unlock; + viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR); + goto send_reply; } 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 = rotate_viewer_stream(vstream, rstream); - if (ret < 0) { - goto end_unlock; - } else if (ret == 1) { - viewer_index.status = htobe32(VIEWER_INDEX_HUP); - delete_viewer_stream(vstream); - destroy_viewer_stream(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)) { - /* - * 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 = rotate_viewer_stream(vstream, rstream); - if (ret < 0) { - goto end_unlock; - } else if (ret == 1) { - viewer_index.status = htobe32(VIEWER_INDEX_HUP); - delete_viewer_stream(vstream); - destroy_viewer_stream(vstream); - goto send_reply; - } - } else { - PERROR("Relay reading index file %d", - vstream->index_read_fd); - viewer_index.status = htobe32(VIEWER_INDEX_ERR); - } + ret = lttng_index_file_read(vstream->index_file, &packet_index); + if (ret) { + ERR("Relay error reading index file %d", + vstream->index_file->fd); + viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR); goto send_reply; } else { - viewer_index.status = htobe32(VIEWER_INDEX_OK); - vstream->last_sent_index++; + viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_OK); + vstream->index_sent_seqcount++; } /* * Indexes are stored in big endian, no need to switch before sending. */ + DBG("Sending viewer index for stream %" PRIu64 " offset %" PRIu64, + rstream->stream_handle, + (uint64_t) be64toh(packet_index.offset)); viewer_index.offset = packet_index.offset; viewer_index.packet_size = packet_index.packet_size; viewer_index.content_size = packet_index.content_size; @@ -1368,24 +1475,53 @@ int viewer_get_next_index(struct relay_command *cmd, viewer_index.stream_id = packet_index.stream_id; send_reply: + if (rstream) { + pthread_mutex_unlock(&rstream->lock); + } + + if (metadata_viewer_stream) { + pthread_mutex_lock(&metadata_viewer_stream->stream->lock); + DBG("get next index metadata check: recv %" PRIu64 + " sent %" PRIu64, + metadata_viewer_stream->stream->metadata_received, + metadata_viewer_stream->metadata_sent); + if (!metadata_viewer_stream->stream->metadata_received || + metadata_viewer_stream->stream->metadata_received > + metadata_viewer_stream->metadata_sent) { + viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA; + } + pthread_mutex_unlock(&metadata_viewer_stream->stream->lock); + } + viewer_index.flags = htobe32(viewer_index.flags); health_code_update(); - ret = cmd->sock->ops->sendmsg(cmd->sock, &viewer_index, - sizeof(viewer_index), 0); + + ret = send_response(conn->sock, &viewer_index, sizeof(viewer_index)); if (ret < 0) { - ERR("Relay index to viewer"); - goto end_unlock; + goto end; } health_code_update(); - DBG("Index %" PRIu64 "for stream %" PRIu64 "sent", - vstream->last_sent_index, vstream->stream_handle); - -end_unlock: - rcu_read_unlock(); - -end_no_session: + if (vstream) { + DBG("Index %" PRIu64 " for stream %" PRIu64 " sent", + vstream->index_sent_seqcount, + vstream->stream->stream_handle); + } end: + if (metadata_viewer_stream) { + viewer_stream_put(metadata_viewer_stream); + } + if (vstream) { + viewer_stream_put(vstream); + } + return ret; + +error_put: + pthread_mutex_unlock(&rstream->lock); + if (metadata_viewer_stream) { + viewer_stream_put(metadata_viewer_stream); + } + viewer_stream_put(vstream); return ret; } @@ -1395,164 +1531,106 @@ end: * Return 0 on success or else a negative value. */ static -int viewer_get_packet(struct relay_command *cmd, - struct lttng_ht *sessions_ht) +int viewer_get_packet(struct relay_connection *conn) { - int ret, send_data = 0; - char *data = NULL; - uint32_t len = 0; - ssize_t read_len; + int ret; + off_t lseek_ret; + char *reply = NULL; struct lttng_viewer_get_packet get_packet_info; - struct lttng_viewer_trace_packet reply; - struct relay_viewer_stream *stream; - - 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; - } + struct lttng_viewer_trace_packet reply_header; + struct relay_viewer_stream *vstream = NULL; + uint32_t reply_size = sizeof(reply_header); + uint32_t packet_data_len = 0; + ssize_t read_len; + + DBG2("Relay get data packet"); 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(conn->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)); - if (!stream) { - goto error; - } - assert(stream->ctf_trace); - - /* - * First time we read this stream, we need open the tracefile, we should - * only arrive here if an index has already been sent to the viewer, so the - * tracefile must exist, if it does not it is a fatal error. - */ - if (stream->read_fd < 0) { - char fullpath[PATH_MAX]; - - 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; - } - ret = open(fullpath, O_RDONLY); - if (ret < 0) { - PERROR("Relay opening trace file"); - goto error; - } - stream->read_fd = ret; - } + memset(&reply_header, 0, sizeof(reply_header)); - if (!stream->ctf_trace->metadata_received || - stream->ctf_trace->metadata_received > - stream->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; + vstream = viewer_stream_get_by_id(be64toh(get_packet_info.stream_id)); + if (!vstream) { + DBG("Client requested packet of unknown stream id %" PRIu64, + (uint64_t) be64toh(get_packet_info.stream_id)); + reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR); + goto send_reply_nolock; + } else { + packet_data_len = be32toh(get_packet_info.len); + reply_size += packet_data_len; } - len = be32toh(get_packet_info.len); - data = zmalloc(len); - if (!data) { - PERROR("relay data zmalloc"); + reply = zmalloc(reply_size); + if (!reply) { + PERROR("packet reply zmalloc"); + reply_size = sizeof(reply_header); goto error; } - ret = lseek(stream->read_fd, be64toh(get_packet_info.offset), SEEK_SET); - if (ret < 0) { - /* - * 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; + pthread_mutex_lock(&vstream->stream->lock); + lseek_ret = lseek(vstream->stream_file.fd->fd, + be64toh(get_packet_info.offset), SEEK_SET); + if (lseek_ret < 0) { + PERROR("lseek fd %d to offset %" PRIu64, + vstream->stream_file.fd->fd, + (uint64_t) 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; - } + read_len = lttng_read(vstream->stream_file.fd->fd, + reply + sizeof(reply_header), packet_data_len); + if (read_len < packet_data_len) { + PERROR("Relay reading trace file, fd: %d, offset: %" PRIu64, + vstream->stream_file.fd->fd, + (uint64_t) be64toh(get_packet_info.offset)); + goto error; } - reply.status = htobe32(VIEWER_GET_PACKET_OK); - reply.len = htobe32(len); - send_data = 1; + reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_OK); + reply_header.len = htobe32(packet_data_len); goto send_reply; error: - reply.status = htobe32(VIEWER_GET_PACKET_ERR); + reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR); send_reply: - reply.flags = htobe32(reply.flags); - - health_code_update(); - ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); - if (ret < 0) { - ERR("Relay data header to viewer"); - goto end_unlock; + if (vstream) { + pthread_mutex_unlock(&vstream->stream->lock); } +send_reply_nolock: + health_code_update(); - if (send_data) { - health_code_update(); - ret = cmd->sock->ops->sendmsg(cmd->sock, data, len, 0); - if (ret < 0) { - ERR("Relay send data to viewer"); - goto end_unlock; - } - health_code_update(); + if (reply) { + memcpy(reply, &reply_header, sizeof(reply_header)); + ret = send_response(conn->sock, reply, reply_size); + } else { + /* No reply to send. */ + ret = send_response(conn->sock, &reply_header, + reply_size); } - DBG("Sent %u bytes for stream %" PRIu64, len, - be64toh(get_packet_info.stream_id)); + health_code_update(); + if (ret < 0) { + PERROR("sendmsg of packet data failed"); + goto end_free; + } -end_unlock: - free(data); - rcu_read_unlock(); + DBG("Sent %u bytes for stream %" PRIu64, reply_size, + (uint64_t) be64toh(get_packet_info.stream_id)); +end_free: + free(reply); end: + if (vstream) { + viewer_stream_put(vstream); + } return ret; } @@ -1562,7 +1640,7 @@ end: * Return 0 on success else a negative value. */ static -int viewer_get_metadata(struct relay_command *cmd) +int viewer_get_metadata(struct relay_connection *conn) { int ret = 0; ssize_t read_len; @@ -1570,60 +1648,80 @@ int viewer_get_metadata(struct relay_command *cmd) char *data = NULL; struct lttng_viewer_get_metadata request; struct lttng_viewer_metadata_packet reply; - struct relay_viewer_stream *stream; + struct relay_viewer_stream *vstream = NULL; - assert(cmd); + assert(conn); 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(conn->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)); - if (!stream || !stream->metadata_flag) { + memset(&reply, 0, sizeof(reply)); + + vstream = viewer_stream_get_by_id(be64toh(request.stream_id)); + if (!vstream) { + /* + * The metadata stream can be closed by a CLOSE command + * just before we attach. It can also be closed by + * per-pid tracing during tracing. Therefore, it is + * possible that we cannot find this viewer stream. + * Reply back to the client with an error if we cannot + * find it. + */ + DBG("Client requested metadata of unknown stream id %" PRIu64, + (uint64_t) be64toh(request.stream_id)); + reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR); + goto send_reply; + } + pthread_mutex_lock(&vstream->stream->lock); + if (!vstream->stream->is_metadata) { 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; + assert(vstream->metadata_sent <= vstream->stream->metadata_received); + + len = vstream->stream->metadata_received - vstream->metadata_sent; if (len == 0) { - reply.status = htobe32(VIEWER_NO_NEW_METADATA); + reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA); goto send_reply; } /* first time, we open the metadata file */ - if (stream->read_fd < 0) { - char fullpath[PATH_MAX]; - - ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name, - stream->channel_name); + if (!vstream->stream_file.fd) { + int fd; + char file_path[LTTNG_PATH_MAX]; + enum lttng_trace_chunk_status status; + struct relay_stream *rstream = vstream->stream; + + ret = utils_stream_file_path(rstream->path_name, + rstream->channel_name, rstream->tracefile_size, + vstream->current_tracefile_id, NULL, file_path, + sizeof(file_path)); if (ret < 0) { goto error; } - ret = open(fullpath, O_RDONLY); - if (ret < 0) { - PERROR("Relay opening metadata file"); + + status = lttng_trace_chunk_open_file( + vstream->stream_file.trace_chunk, + file_path, O_RDONLY, 0, &fd); + if (status != LTTNG_TRACE_CHUNK_STATUS_OK) { + PERROR("Failed to open metadata file for viewer stream"); + goto error; + } + vstream->stream_file.fd = stream_fd_create(fd); + if (!vstream->stream_file.fd) { + if (close(fd)) { + PERROR("Failed to close viewer metadata file"); + } goto error; } - stream->read_fd = ret; } reply.len = htobe64(len); @@ -1633,235 +1731,251 @@ int viewer_get_metadata(struct relay_command *cmd) goto error; } - read_len = lttng_read(stream->read_fd, data, len); + read_len = lttng_read(vstream->stream_file.fd->fd, data, len); if (read_len < len) { PERROR("Relay reading metadata file"); goto error; } - stream->ctf_trace->metadata_sent += read_len; - reply.status = htobe32(VIEWER_METADATA_OK); + vstream->metadata_sent += read_len; + if (vstream->metadata_sent == vstream->stream->metadata_received + && vstream->stream->closed) { + /* Release ownership for the viewer metadata stream. */ + viewer_stream_put(vstream); + } + + reply.status = htobe32(LTTNG_VIEWER_METADATA_OK); + goto send_reply; error: - reply.status = htobe32(VIEWER_METADATA_ERR); + reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR); send_reply: health_code_update(); - ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); + if (vstream) { + pthread_mutex_unlock(&vstream->stream->lock); + } + ret = send_response(conn->sock, &reply, sizeof(reply)); if (ret < 0) { - ERR("Relay data header to viewer"); - goto end_unlock; + goto end_free; } health_code_update(); if (len > 0) { - ret = cmd->sock->ops->sendmsg(cmd->sock, data, len, 0); + ret = send_response(conn->sock, data, len); if (ret < 0) { - ERR("Relay send data to viewer"); - goto end_unlock; + goto end_free; } } DBG("Sent %" PRIu64 " bytes of metadata for stream %" PRIu64, len, - be64toh(request.stream_id)); + (uint64_t) be64toh(request.stream_id)); DBG("Metadata sent"); -end_unlock: +end_free: free(data); - rcu_read_unlock(); end: + if (vstream) { + viewer_stream_put(vstream); + } return ret; } /* - * live_relay_unknown_command: send -1 if received unknown command + * Create a viewer session. + * + * Return 0 on success or else a negative value. */ static -void live_relay_unknown_command(struct relay_command *cmd) +int viewer_create_session(struct relay_connection *conn) { - struct lttcomm_relayd_generic_reply reply; int ret; + struct lttng_viewer_create_session_response resp; - 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"); - } -} + DBG("Viewer create session received"); -/* - * Process the commands received on the control socket - */ -static -int process_control(struct lttng_viewer_cmd *recv_hdr, - struct relay_command *cmd, struct lttng_ht *sessions_ht) -{ - int ret = 0; + memset(&resp, 0, sizeof(resp)); + resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_OK); + conn->viewer_session = viewer_session_create(); + if (!conn->viewer_session) { + ERR("Allocation viewer session"); + resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_ERR); + goto send_reply; + } - switch (be32toh(recv_hdr->cmd)) { - case VIEWER_CONNECT: - ret = viewer_connect(cmd); - break; - case VIEWER_LIST_SESSIONS: - ret = viewer_list_sessions(cmd, sessions_ht); - break; - case VIEWER_ATTACH_SESSION: - ret = viewer_attach_session(cmd, sessions_ht); - break; - case VIEWER_GET_NEXT_INDEX: - ret = viewer_get_next_index(cmd, sessions_ht); - break; - case VIEWER_GET_PACKET: - ret = viewer_get_packet(cmd, sessions_ht); - break; - case VIEWER_GET_METADATA: - ret = viewer_get_metadata(cmd); - break; - default: - ERR("Received unknown viewer command (%u)", be32toh(recv_hdr->cmd)); - live_relay_unknown_command(cmd); - ret = -1; +send_reply: + health_code_update(); + ret = send_response(conn->sock, &resp, sizeof(resp)); + if (ret < 0) { goto end; } + health_code_update(); + ret = 0; end: return ret; } +/* + * Detach a viewer session. + * + * Return 0 on success or else a negative value. + */ static -void cleanup_poll_connection(struct lttng_poll_event *events, int pollfd) +int viewer_detach_session(struct relay_connection *conn) { int ret; + struct lttng_viewer_detach_session_response response; + struct lttng_viewer_detach_session_request request; + struct relay_session *session = NULL; + uint64_t viewer_session_to_close; - assert(events); + DBG("Viewer detach session received"); - lttng_poll_del(events, pollfd); + assert(conn); - ret = close(pollfd); + health_code_update(); + + /* Receive the request from the connected client. */ + ret = recv_request(conn->sock, &request, sizeof(request)); if (ret < 0) { - ERR("Closing pollfd %d", pollfd); + goto end; } -} + viewer_session_to_close = be64toh(request.session_id); -/* - * Create and add connection to the given hash table. - * - * Return poll add value or else -1 on error. - */ -static -int add_connection(int fd, struct lttng_poll_event *events, - struct lttng_ht *relay_connections_ht) -{ - int ret; - struct relay_command *relay_connection; + if (!conn->viewer_session) { + DBG("Client trying to detach before creating a live viewer session"); + response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_ERR); + goto send_reply; + } + + health_code_update(); - assert(events); - assert(relay_connections_ht); + memset(&response, 0, sizeof(response)); + DBG("Detaching from session ID %" PRIu64, viewer_session_to_close); - relay_connection = zmalloc(sizeof(struct relay_command)); - if (relay_connection == NULL) { - PERROR("Relay command zmalloc"); - goto error; + session = session_get_by_id(be64toh(request.session_id)); + if (!session) { + DBG("Relay session %" PRIu64 " not found", + (uint64_t) be64toh(request.session_id)); + response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_UNK); + goto send_reply; } - ret = lttng_read(fd, relay_connection, sizeof(*relay_connection)); - if (ret < sizeof(*relay_connection)) { - PERROR("read relay cmd pipe"); - goto error_read; + ret = viewer_session_is_attached(conn->viewer_session, session); + if (ret != 1) { + DBG("Not attached to this session"); + response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_ERR); + goto send_reply_put; } - lttng_ht_node_init_ulong(&relay_connection->sock_n, - (unsigned long) relay_connection->sock->fd); - rcu_read_lock(); - lttng_ht_add_unique_ulong(relay_connections_ht, - &relay_connection->sock_n); - rcu_read_unlock(); + viewer_session_close_one_session(conn->viewer_session, session); + response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_OK); + DBG("Session %" PRIu64 " detached.", viewer_session_to_close); - return lttng_poll_add(events, relay_connection->sock->fd, - LPOLLIN | LPOLLRDHUP); +send_reply_put: + session_put(session); -error_read: - free(relay_connection); -error: - return -1; +send_reply: + health_code_update(); + ret = send_response(conn->sock, &response, sizeof(response)); + if (ret < 0) { + goto end; + } + health_code_update(); + ret = 0; + +end: + return ret; } +/* + * live_relay_unknown_command: send -1 if received unknown command + */ static -void deferred_free_connection(struct rcu_head *head) +void live_relay_unknown_command(struct relay_connection *conn) { - struct relay_command *relay_connection = - caa_container_of(head, struct relay_command, rcu_node); + struct lttcomm_relayd_generic_reply reply; - if (relay_connection->session && - relay_connection->session->viewer_attached > 0) { - relay_connection->session->viewer_attached--; - } - lttcomm_destroy_sock(relay_connection->sock); - free(relay_connection); + memset(&reply, 0, sizeof(reply)); + reply.ret_code = htobe32(LTTNG_ERR_UNK); + (void) send_response(conn->sock, &reply, sizeof(reply)); } /* - * Delete all streams for a specific session ID. + * Process the commands received on the control socket */ static -void viewer_del_streams(uint64_t session_id) +int process_control(struct lttng_viewer_cmd *recv_hdr, + struct relay_connection *conn) { - struct relay_viewer_stream *stream; - struct lttng_ht_iter iter; - - rcu_read_lock(); - cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, stream, - stream_n.node) { - health_code_update(); + int ret = 0; + uint32_t msg_value; - if (stream->session_id != session_id) { - continue; - } + msg_value = be32toh(recv_hdr->cmd); - delete_viewer_stream(stream); - assert(stream->ctf_trace); + /* + * Make sure we've done the version check before any command other then a + * new client connection. + */ + if (msg_value != LTTNG_VIEWER_CONNECT && !conn->version_check_done) { + ERR("Viewer conn value %" PRIu32 " before version check", msg_value); + ret = -1; + goto end; + } - if (stream->metadata_flag) { - /* - * The metadata viewer stream is destroyed once the refcount on the - * ctf trace goes to 0 in the destroy stream function thus there is - * no explicit call to that function here. - */ - stream->ctf_trace->metadata_sent = 0; - stream->ctf_trace->viewer_metadata_stream = NULL; - } else { - destroy_viewer_stream(stream); - } + switch (msg_value) { + case LTTNG_VIEWER_CONNECT: + ret = viewer_connect(conn); + break; + case LTTNG_VIEWER_LIST_SESSIONS: + ret = viewer_list_sessions(conn); + break; + case LTTNG_VIEWER_ATTACH_SESSION: + ret = viewer_attach_session(conn); + break; + case LTTNG_VIEWER_GET_NEXT_INDEX: + ret = viewer_get_next_index(conn); + break; + case LTTNG_VIEWER_GET_PACKET: + ret = viewer_get_packet(conn); + break; + case LTTNG_VIEWER_GET_METADATA: + ret = viewer_get_metadata(conn); + break; + case LTTNG_VIEWER_GET_NEW_STREAMS: + ret = viewer_get_new_streams(conn); + break; + case LTTNG_VIEWER_CREATE_SESSION: + ret = viewer_create_session(conn); + break; + case LTTNG_VIEWER_DETACH_SESSION: + ret = viewer_detach_session(conn); + break; + default: + ERR("Received unknown viewer command (%u)", + be32toh(recv_hdr->cmd)); + live_relay_unknown_command(conn); + ret = -1; + goto end; } - rcu_read_unlock(); + +end: + return ret; } -/* - * Delete and free a connection. - * - * RCU read side lock MUST be acquired. - */ static -void del_connection(struct lttng_ht *relay_connections_ht, - struct lttng_ht_iter *iter, struct relay_command *relay_connection) +void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd) { int ret; - assert(relay_connections_ht); - assert(iter); - assert(relay_connection); - - DBG("Cleaning connection of session ID %" PRIu64, - relay_connection->session_id); - - ret = lttng_ht_del(relay_connections_ht, iter); - assert(!ret); - - viewer_del_streams(relay_connection->session_id); + (void) lttng_poll_del(events, pollfd); - call_rcu(&relay_connection->rcu_node, deferred_free_connection); + ret = close(pollfd); + if (ret < 0) { + ERR("Closing pollfd %d", pollfd); + } } /* @@ -1872,14 +1986,11 @@ void *thread_worker(void *data) { int ret, err = -1; uint32_t nb_fd; - struct relay_command *relay_connection; struct lttng_poll_event events; - struct lttng_ht *relay_connections_ht; - struct lttng_ht_node_ulong *node; + struct lttng_ht *viewer_connections_ht; struct lttng_ht_iter iter; struct lttng_viewer_cmd recv_hdr; - struct relay_local_data *relay_ctx = (struct relay_local_data *) data; - struct lttng_ht *sessions_ht = relay_ctx->sessions_ht; + struct relay_connection *destroy_conn; DBG("[thread] Live viewer relay worker started"); @@ -1892,9 +2003,9 @@ void *thread_worker(void *data) } /* table of connections indexed on socket */ - relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); - if (!relay_connections_ht) { - goto relay_connections_ht_error; + viewer_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); + if (!viewer_connections_ht) { + goto viewer_connections_ht_error; } ret = create_thread_poll_set(&events, 2); @@ -1902,7 +2013,7 @@ void *thread_worker(void *data) goto error_poll_create; } - ret = lttng_poll_add(&events, live_relay_cmd_pipe[0], LPOLLIN | LPOLLRDHUP); + ret = lttng_poll_add(&events, live_conn_pipe[0], LPOLLIN | LPOLLRDHUP); if (ret < 0) { goto error; } @@ -1949,71 +2060,71 @@ restart: goto exit; } - /* Inspect the relay cmd pipe for new connection */ - if (pollfd == live_relay_cmd_pipe[0]) { - if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { - ERR("Relay live pipe error"); - goto error; - } else if (revents & LPOLLIN) { - DBG("Relay live viewer command received"); - ret = add_connection(live_relay_cmd_pipe[0], - &events, relay_connections_ht); + /* Inspect the relay conn pipe for new connection. */ + if (pollfd == live_conn_pipe[0]) { + if (revents & LPOLLIN) { + struct relay_connection *conn; + + ret = lttng_read(live_conn_pipe[0], + &conn, sizeof(conn)); if (ret < 0) { goto error; } - } - } else if (revents) { - rcu_read_lock(); - lttng_ht_lookup(relay_connections_ht, - (void *)((unsigned long) pollfd), &iter); - node = lttng_ht_iter_get_node_ulong(&iter); - if (node == NULL) { - DBG2("Relay viewer sock %d not found", pollfd); - rcu_read_unlock(); + ret = lttng_poll_add(&events, + conn->sock->fd, + LPOLLIN | LPOLLRDHUP); + if (ret) { + ERR("Failed to add new live connection file descriptor to poll set"); + goto error; + } + connection_ht_add(viewer_connections_ht, conn); + DBG("Connection socket %d added to poll", conn->sock->fd); + } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { + ERR("Relay live pipe error"); + goto error; + } else { + ERR("Unexpected poll events %u for sock %d", revents, pollfd); goto error; } - relay_connection = caa_container_of(node, struct relay_command, - sock_n); - - if (revents & (LPOLLERR)) { - cleanup_poll_connection(&events, pollfd); - del_connection(relay_connections_ht, &iter, - relay_connection); - } 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); - } else if (revents & LPOLLIN) { - ret = relay_connection->sock->ops->recvmsg( - relay_connection->sock, &recv_hdr, - sizeof(struct lttng_viewer_cmd), - 0); - /* connection closed */ + } else { + /* Connection activity. */ + struct relay_connection *conn; + + conn = connection_get_by_sock(viewer_connections_ht, pollfd); + if (!conn) { + continue; + } + + if (revents & LPOLLIN) { + ret = conn->sock->ops->recvmsg(conn->sock, &recv_hdr, + sizeof(recv_hdr), 0); if (ret <= 0) { - cleanup_poll_connection(&events, pollfd); - del_connection(relay_connections_ht, &iter, - relay_connection); - DBG("Viewer control connection closed with %d", - pollfd); + /* Connection closed. */ + cleanup_connection_pollfd(&events, pollfd); + /* Put "create" ownership reference. */ + connection_put(conn); + DBG("Viewer control conn closed with %d", pollfd); } else { - if (relay_connection->session) { - DBG2("Relay viewer worker receiving data for " - "session: %" PRIu64, - relay_connection->session->id); - } - ret = process_control(&recv_hdr, relay_connection, - sessions_ht); + ret = process_control(&recv_hdr, conn); if (ret < 0) { /* Clear the session on error. */ - cleanup_poll_connection(&events, pollfd); - del_connection(relay_connections_ht, &iter, - relay_connection); + cleanup_connection_pollfd(&events, pollfd); + /* Put "create" ownership reference. */ + connection_put(conn); DBG("Viewer connection closed with %d", pollfd); } } + } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { + cleanup_connection_pollfd(&events, pollfd); + /* Put "create" ownership reference. */ + connection_put(conn); + } else { + ERR("Unexpected poll events %u for sock %d", revents, pollfd); + connection_put(conn); + goto error; } - rcu_read_unlock(); + /* Put local "get_by_sock" reference. */ + connection_put(conn); } } } @@ -2022,26 +2133,20 @@ exit: error: lttng_poll_clean(&events); - /* empty the hash table and free the memory */ + /* Cleanup remaining connection object. */ rcu_read_lock(); - cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, node, node) { + cds_lfht_for_each_entry(viewer_connections_ht->ht, &iter.iter, + destroy_conn, + sock_n.node) { health_code_update(); - - node = lttng_ht_iter_get_node_ulong(&iter); - if (!node) { - continue; - } - - relay_connection = caa_container_of(node, struct relay_command, - sock_n); - del_connection(relay_connections_ht, &iter, relay_connection); + connection_put(destroy_conn); } rcu_read_unlock(); error_poll_create: - lttng_ht_destroy(relay_connections_ht); -relay_connections_ht_error: - /* Close relay cmd pipes */ - utils_close_pipe(live_relay_cmd_pipe); + lttng_ht_destroy(viewer_connections_ht); +viewer_connections_ht_error: + /* Close relay conn pipes */ + utils_close_pipe(live_conn_pipe); if (err) { DBG("Viewer worker thread exited with error"); } @@ -2052,7 +2157,9 @@ error_testpoint: ERR("Health error occurred in %s", __func__); } health_unregister(health_relayd); - stop_threads(); + if (lttng_relay_stop_threads()) { + ERR("Error stopping threads"); + } rcu_unregister_thread(); return NULL; } @@ -2061,57 +2168,55 @@ error_testpoint: * Create the relay command pipe to wake thread_manage_apps. * Closed in cleanup(). */ -static int create_relay_cmd_pipe(void) +static int create_conn_pipe(void) { - int ret; - - ret = utils_create_pipe_cloexec(live_relay_cmd_pipe); - - return ret; + return utils_create_pipe_cloexec(live_conn_pipe); } -void live_stop_threads(void) +int relayd_live_join(void) { - int ret; + int ret, retval = 0; void *status; - stop_threads(); - ret = pthread_join(live_listener_thread, &status); - if (ret != 0) { + if (ret) { + errno = ret; PERROR("pthread_join live listener"); - goto error; /* join error, exit without cleanup */ + retval = -1; } ret = pthread_join(live_worker_thread, &status); - if (ret != 0) { + if (ret) { + errno = ret; PERROR("pthread_join live worker"); - goto error; /* join error, exit without cleanup */ + retval = -1; } ret = pthread_join(live_dispatcher_thread, &status); - if (ret != 0) { + if (ret) { + errno = ret; PERROR("pthread_join live dispatcher"); - goto error; /* join error, exit without cleanup */ + retval = -1; } - cleanup(); + cleanup_relayd_live(); -error: - return; + return retval; } /* * main */ -int live_start_threads(struct lttng_uri *uri, - struct relay_local_data *relay_ctx) +int relayd_live_create(struct lttng_uri *uri) { - int ret = 0; + int ret = 0, retval = 0; void *status; int is_root; - assert(uri); + if (!uri) { + retval = -1; + goto exit_init_data; + } live_uri = uri; /* Check if daemon is UID = 0 */ @@ -2120,74 +2225,86 @@ int live_start_threads(struct lttng_uri *uri, if (!is_root) { if (live_uri->port < 1024) { ERR("Need to be root to use ports < 1024"); - ret = -1; - goto exit; + retval = -1; + goto exit_init_data; } } /* Setup the thread apps communication pipe. */ - if ((ret = create_relay_cmd_pipe()) < 0) { - goto exit; + if (create_conn_pipe()) { + retval = -1; + goto exit_init_data; } /* Init relay command queue. */ - cds_wfq_init(&viewer_cmd_queue.queue); + cds_wfcq_init(&viewer_conn_queue.head, &viewer_conn_queue.tail); /* Set up max poll set size */ - lttng_poll_set_max_size(); + if (lttng_poll_set_max_size()) { + retval = -1; + goto exit_init_data; + } /* Setup the dispatcher thread */ - ret = pthread_create(&live_dispatcher_thread, NULL, + ret = pthread_create(&live_dispatcher_thread, default_pthread_attr(), thread_dispatcher, (void *) NULL); - if (ret != 0) { + if (ret) { + errno = ret; PERROR("pthread_create viewer dispatcher"); - goto exit_dispatcher; + retval = -1; + goto exit_dispatcher_thread; } /* Setup the worker thread */ - ret = pthread_create(&live_worker_thread, NULL, - thread_worker, relay_ctx); - if (ret != 0) { + ret = pthread_create(&live_worker_thread, default_pthread_attr(), + thread_worker, NULL); + if (ret) { + errno = ret; PERROR("pthread_create viewer worker"); - goto exit_worker; + retval = -1; + goto exit_worker_thread; } /* Setup the listener thread */ - ret = pthread_create(&live_listener_thread, NULL, + ret = pthread_create(&live_listener_thread, default_pthread_attr(), thread_listener, (void *) NULL); - if (ret != 0) { + if (ret) { + errno = ret; PERROR("pthread_create viewer listener"); - goto exit_listener; + retval = -1; + goto exit_listener_thread; } - ret = 0; - goto end; + /* + * All OK, started all threads. + */ + return retval; -exit_listener: - ret = pthread_join(live_listener_thread, &status); - if (ret != 0) { - PERROR("pthread_join live listener"); - goto error; /* join error, exit without cleanup */ - } + /* + * Join on the live_listener_thread should anything be added after + * the live_listener thread's creation. + */ + +exit_listener_thread: -exit_worker: ret = pthread_join(live_worker_thread, &status); - if (ret != 0) { + if (ret) { + errno = ret; PERROR("pthread_join live worker"); - goto error; /* join error, exit without cleanup */ + retval = -1; } +exit_worker_thread: -exit_dispatcher: ret = pthread_join(live_dispatcher_thread, &status); - if (ret != 0) { + if (ret) { + errno = ret; PERROR("pthread_join live dispatcher"); - goto error; /* join error, exit without cleanup */ + retval = -1; } +exit_dispatcher_thread: -exit: - cleanup(); +exit_init_data: + cleanup_relayd_live(); -end: -error: - return ret; + return retval; }