X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-relayd%2Flive.c;h=3ddafe2a16fb0075c33cbf921b9fb91981cfa646;hp=764d616a24e692c77f8e717c37ddcf3841d556fb;hb=d546eeec57f3d23255e55d6d7384ab825bae8218;hpb=2a174661a1e0ab551b41ff1cae7191688525fc1f diff --git a/src/bin/lttng-relayd/live.c b/src/bin/lttng-relayd/live.c index 764d616a2..3ddafe2a1 100644 --- a/src/bin/lttng-relayd/live.c +++ b/src/bin/lttng-relayd/live.c @@ -1,24 +1,17 @@ /* - * Copyright (C) 2013 - Julien Desfossez - * David Goulet + * Copyright (C) 2013 Julien Desfossez + * Copyright (C) 2013 David Goulet + * Copyright (C) 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, - * as published by the Free Software Foundation. + * SPDX-License-Identifier: GPL-2.0-only * - * This program is distributed in the hope that it will be useful, but WITHOUT - * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or - * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for - * more details. - * - * You should have received a copy of the GNU General Public License along - * with this program; if not, write to the Free Software Foundation, Inc., - * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. */ -#define _GNU_SOURCE +#define _LGPL_SOURCE +#include #include #include +#include #include #include #include @@ -32,36 +25,41 @@ #include #include #include -#include +#include #include +#include #include -#include -#include -#include -#include #include +#include #include #include #include +#include +#include #include #include -#include #include #include +#include #include #include +#include #include "cmd.h" +#include "connection.h" +#include "ctf-trace.h" +#include "health-relayd.h" #include "live.h" #include "lttng-relayd.h" -#include "utils.h" -#include "health-relayd.h" +#include "session.h" +#include "stream.h" #include "testpoint.h" +#include "utils.h" +#include "viewer-session.h" #include "viewer-stream.h" -#include "stream.h" -#include "session.h" -#include "ctf-trace.h" + +#define SESSION_BUF_DEFAULT_COUNT 16 static struct lttng_uri *live_uri; @@ -69,7 +67,7 @@ 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; @@ -84,15 +82,17 @@ 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"); @@ -111,9 +111,6 @@ ssize_t recv_request(struct lttcomm_sock *sock, void *buf, size_t size) { ssize_t ret; - assert(sock); - assert(buf); - ret = sock->ops->recvmsg(sock, buf, size, 0); if (ret < 0 || ret != size) { if (ret == 0) { @@ -140,9 +137,6 @@ ssize_t send_response(struct lttcomm_sock *sock, void *buf, size_t size) { ssize_t ret; - assert(sock); - assert(buf); - ret = sock->ops->sendmsg(sock, buf, size, 0); if (ret < 0) { ERR("Relayd failed to send response."); @@ -152,32 +146,38 @@ ssize_t send_response(struct lttcomm_sock *sock, void *buf, size_t size) } /* - * Atomically check if new streams got added in the session since the last - * check and reset the flag to 0. + * 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(uint64_t session_id, struct lttng_ht *sessions_ht) +int check_new_streams(struct relay_connection *conn) { - int ret; - unsigned long current_val; struct relay_session *session; + unsigned long current_val; + int ret = 0; - 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; + if (!conn->viewer_session) { + goto end; } - - current_val = uatomic_cmpxchg(&session->new_streams, 1, 0); - ret = current_val; - -error: + 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; } @@ -189,15 +189,13 @@ error: */ static ssize_t send_viewer_streams(struct lttcomm_sock *sock, - struct relay_session *session, unsigned int ignore_sent_flag) + 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; - assert(session); - rcu_read_lock(); cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, vstream, @@ -206,30 +204,50 @@ ssize_t send_viewer_streams(struct lttcomm_sock *sock, health_code_update(); + if (!viewer_stream_get(vstream)) { + continue; + } + + pthread_mutex_lock(&vstream->stream->lock); /* Ignore if not the same session. */ - if (vstream->session_id != session->id || + 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 = ctf_trace_find_by_path(session->ctf_traces_ht, - vstream->path_name); - assert(ctf_trace); - - send_stream.id = htobe64(vstream->stream_handle); + 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->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)); + 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); - DBG("Sending stream %" PRIu64 " to viewer", vstream->stream_handle); ret = send_response(sock, &send_stream, sizeof(send_stream)); + viewer_stream_put(vstream); if (ret < 0) { goto end_unlock; } - vstream->sent_flag = 1; } ret = 0; @@ -245,124 +263,172 @@ end_unlock: * 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, - enum lttng_viewer_seek seek_t, uint32_t *nb_total, uint32_t *nb_unsent, - uint32_t *nb_created) +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); - /* - * 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); + if (!viewer_trace_chunk) { + ERR("Internal error: viewer session associated with session \"%s\" has a NULL trace chunk", + session->session_name); + ret = -1; + goto error; + } + + 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. + * 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) { + bool trace_has_metadata_stream = false; struct relay_stream *stream; health_code_update(); - if (ctf_trace->invalid_flag) { + if (!ctf_trace_get(ctf_trace)) { + continue; + } + + /* + * Iterate over all the streams of the trace to see if we have a + * metadata stream. + */ + cds_list_for_each_entry_rcu( + stream, &ctf_trace->stream_list, stream_node) + { + if (stream->is_metadata) { + trace_has_metadata_stream = true; + break; + } + } + + /* + * If there is no metadata stream in this trace at the moment + * and we never sent one to the viewer, skip the trace. We + * accept that the viewer will not see this trace at all. + */ + if (!trace_has_metadata_stream && + !ctf_trace->metadata_stream_sent_to_viewer) { + ctf_trace_put(ctf_trace); continue; } - cds_list_for_each_entry(stream, &ctf_trace->stream_list, trace_list) { + cds_list_for_each_entry_rcu(stream, &ctf_trace->stream_list, stream_node) { struct relay_viewer_stream *vstream; - if (!stream->viewer_ready) { + if (!stream_get(stream)) { continue; } - - vstream = viewer_stream_find_by_id(stream->stream_handle); + /* + * 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, seek_t, ctf_trace); + /* + * Save that we sent the metadata stream to the + * viewer. So that we know what trace the viewer + * is aware of. + */ + if (stream->is_metadata) { + ctf_trace->metadata_stream_sent_to_viewer = + true; + } + 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; } - /* 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)++; + /* + * 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) { - (*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); } ret = 0; error_unlock: rcu_read_unlock(); - pthread_mutex_unlock(&session->viewer_ready_lock); +error: 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 relayd_live_stop(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 */ + /* 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; } /* * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set. */ static -int create_thread_poll_set(struct lttng_poll_event *events, int size) +int create_named_thread_poll_set(struct lttng_poll_event *events, + int size, const char *name) { int ret; @@ -371,13 +437,11 @@ int create_thread_poll_set(struct lttng_poll_event *events, int size) goto error; } - ret = lttng_poll_create(events, size, LTTNG_CLOEXEC); - if (ret < 0) { - goto error; - } + ret = fd_tracker_util_poll_create(the_fd_tracker, + name, events, 1, LTTNG_CLOEXEC); /* Add quit pipe */ - ret = lttng_poll_add(events, live_conn_pipe[0], LPOLLIN | LPOLLERR); + ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR); if (ret < 0) { goto error; } @@ -394,23 +458,85 @@ error: * Return 1 if it was triggered else 0; */ static -int check_live_conn_pipe(int fd, uint32_t events) +int check_thread_quit_pipe(int fd, uint32_t events) { - if (fd == live_conn_pipe[0] && (events & LPOLLIN)) { + if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) { return 1; } return 0; } +static +int create_sock(void *data, int *out_fd) +{ + int ret; + struct lttcomm_sock *sock = data; + + ret = lttcomm_create_sock(sock); + if (ret < 0) { + goto end; + } + + *out_fd = sock->fd; +end: + return ret; +} + +static +int close_sock(void *data, int *in_fd) +{ + struct lttcomm_sock *sock = data; + + return sock->ops->close(sock); +} + +static int accept_sock(void *data, int *out_fd) +{ + int ret = 0; + /* Socks is an array of in_sock, out_sock. */ + struct lttcomm_sock **socks = data; + struct lttcomm_sock *in_sock = socks[0]; + + socks[1] = in_sock->ops->accept(in_sock); + if (!socks[1]) { + ret = -1; + goto end; + } + *out_fd = socks[1]->fd; +end: + return ret; +} + +static +struct lttcomm_sock *accept_live_sock(struct lttcomm_sock *listening_sock, + const char *name) +{ + int out_fd, ret; + struct lttcomm_sock *socks[2] = { listening_sock, NULL }; + struct lttcomm_sock *new_sock = NULL; + + ret = fd_tracker_open_unsuspendable_fd(the_fd_tracker, &out_fd, + (const char **) &name, 1, accept_sock, &socks); + if (ret) { + goto end; + } + new_sock = socks[1]; + DBG("%s accepted, socket %d", name, new_sock->fd); +end: + return new_sock; +} + /* * Create and init socket from uri. */ static -struct lttcomm_sock *init_socket(struct lttng_uri *uri) +struct lttcomm_sock *init_socket(struct lttng_uri *uri, const char *name) { - int ret; + int ret, sock_fd; struct lttcomm_sock *sock = NULL; + char uri_str[LTTNG_PATH_MAX]; + char *formated_name = NULL; sock = lttcomm_alloc_sock_from_uri(uri); if (sock == NULL) { @@ -418,14 +544,33 @@ struct lttcomm_sock *init_socket(struct lttng_uri *uri) goto error; } - ret = lttcomm_create_sock(sock); - if (ret < 0) { + /* + * Don't fail to create the socket if the name can't be built as it is + * only used for debugging purposes. + */ + ret = uri_to_str_url(uri, uri_str, sizeof(uri_str)); + uri_str[sizeof(uri_str) - 1] = '\0'; + if (ret >= 0) { + ret = asprintf(&formated_name, "%s socket @ %s", name, + uri_str); + if (ret < 0) { + formated_name = NULL; + } + } + + ret = fd_tracker_open_unsuspendable_fd(the_fd_tracker, &sock_fd, + (const char **) (formated_name ? &formated_name : NULL), + 1, create_sock, sock); + if (ret) { + PERROR("Failed to create \"%s\" socket", + formated_name ?: "Unknown"); goto error; } - DBG("Listening on sock %d for live", sock->fd); + DBG("Listening on %s socket %d", name, sock->fd); ret = sock->ops->bind(sock); if (ret < 0) { + PERROR("Failed to bind lttng-live socket"); goto error; } @@ -435,12 +580,14 @@ struct lttcomm_sock *init_socket(struct lttng_uri *uri) } + free(formated_name); return sock; error: if (sock) { lttcomm_destroy_sock(sock); } + free(formated_name); return NULL; } @@ -451,24 +598,25 @@ 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; DBG("[thread] Relay live listener started"); + rcu_register_thread(); health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_LISTENER); health_code_update(); - live_control_sock = init_socket(live_uri); + live_control_sock = init_socket(live_uri, "Live listener"); if (!live_control_sock) { goto error_sock_control; } /* Pass 2 as size here for the thread quit pipe and control sockets. */ - ret = create_thread_poll_set(&events, 2); + ret = create_named_thread_poll_set(&events, 2, + "Live listener thread epoll"); if (ret < 0) { goto error_create_poll; } @@ -514,57 +662,62 @@ restart: pollfd = LTTNG_POLL_GETFD(&events, i); /* Thread quit pipe has been closed. Killing thread. */ - ret = check_live_conn_pipe(pollfd, revents); + ret = check_thread_quit_pipe(pollfd, revents); if (ret) { err = 0; 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); + newsock = accept_live_sock(live_control_sock, + "Live socket to client"); 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; } } } @@ -573,13 +726,18 @@ exit: error: error_poll_add: error_testpoint: - lttng_poll_clean(&events); + (void) fd_tracker_util_poll_clean(the_fd_tracker, &events); error_create_poll: if (live_control_sock->fd >= 0) { - ret = live_control_sock->ops->close(live_control_sock); + int sock_fd = live_control_sock->fd; + + ret = fd_tracker_close_unsuspendable_fd(the_fd_tracker, + &sock_fd, 1, close_sock, + live_control_sock); if (ret) { PERROR("close"); } + live_control_sock->fd = -1; } lttcomm_destroy_sock(live_control_sock); error_sock_control: @@ -588,8 +746,11 @@ error_sock_control: DBG("Live viewer listener thread exited with error"); } health_unregister(health_relayd); + rcu_unregister_thread(); DBG("Live viewer listener thread cleanup complete"); - stop_threads(); + if (lttng_relay_stop_threads()) { + ERR("Error stopping threads"); + } return NULL; } @@ -601,8 +762,8 @@ 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"); @@ -614,45 +775,49 @@ 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(); } @@ -667,7 +832,9 @@ 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; } @@ -677,26 +844,25 @@ 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(); DBG("Viewer is establishing a connection to the relayd."); - ret = recv_request(cmd->sock, &msg, sizeof(msg)); + 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; @@ -708,18 +874,18 @@ int viewer_connect(struct relay_command *cmd) 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; @@ -728,20 +894,28 @@ 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 = send_response(cmd->sock, &reply, sizeof(reply)); + ret = send_response(conn->sock, &reply, sizeof(reply)); if (ret < 0) { 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: @@ -750,167 +924,236 @@ 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_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"); - rcu_read_lock(); - cds_lfht_count_nodes(sessions_ht->ht, &approx_before, &count, &approx_after); - session_list.sessions_count = htobe32(count); - - health_code_update(); - - ret = send_response(cmd->sock, &session_list, sizeof(session_list)); - if (ret < 0) { - goto end_unlock; + send_session_buf = zmalloc(SESSION_BUF_DEFAULT_COUNT * sizeof(*send_session_buf)); + if (!send_session_buf) { + return -1; } - health_code_update(); - + rcu_read_lock(); cds_lfht_for_each_entry(sessions_ht->ht, &iter.iter, session, session_n.node) { + struct lttng_viewer_session *send_session; + health_code_update(); - 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_refcount); - send_session.streams = htobe32(session->stream_count); + 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; + } - health_code_update(); + if (count >= buf_count) { + struct lttng_viewer_session *newbuf; + uint32_t new_buf_count = buf_count << 1; - ret = send_response(cmd->sock, &send_session, sizeof(send_session)); - if (ret < 0) { - 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; + } + rcu_read_unlock(); + if (ret < 0) { + goto end_free; } + + session_list.sessions_count = htobe32(count); + health_code_update(); - rcu_read_unlock(); - ret = 0; - goto end; + ret = send_response(conn->sock, &session_list, sizeof(session_list)); + if (ret < 0) { + goto end_free; + } -end_unlock: - rcu_read_unlock(); + health_code_update(); -end: + ret = send_response(conn->sock, send_session_buf, + count * sizeof(*send_session_buf)); + if (ret < 0) { + goto end_free; + } + health_code_update(); + + ret = 0; +end_free: + free(send_session_buf); return ret; } /* - * Send the viewer the list of current sessions. + * Send the viewer the list of current streams. */ static -int viewer_get_new_streams(struct relay_command *cmd, - struct lttng_ht *sessions_ht) +int viewer_get_new_streams(struct relay_connection *conn) { int ret, send_streams = 0; - uint32_t nb_created = 0, nb_unsent = 0, nb_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; + struct relay_session *session = NULL; + uint64_t session_id; + bool closed = false; - assert(cmd); - assert(sessions_ht); + assert(conn); DBG("Get new streams received"); health_code_update(); /* Receive the request from the connected client. */ - ret = recv_request(cmd->sock, &request, sizeof(request)); + 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(); - 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; - } + memset(&response, 0, sizeof(response)); - 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); + 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; } - if (!send_streams) { + if (!viewer_session_is_attached(conn->viewer_session, session)) { + response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR); goto send_reply; } - ret = make_viewer_streams(session, VIEWER_SEEK_LAST, NULL, &nb_unsent, - &nb_created); + pthread_mutex_lock(&session->lock); + 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 end_unlock; + goto error_unlock_session; } + send_streams = 1; + response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_OK); + /* Only send back the newly created streams with the unsent ones. */ nb_streams = nb_created + nb_unsent; response.streams_count = htobe32(nb_streams); + /* + * If the session is closed, HUP when there are no more streams + * with data. + */ + if (closed && nb_total == 0) { + send_streams = 0; + response.streams_count = 0; + response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP); + goto send_reply_unlock; + } +send_reply_unlock: + pthread_mutex_unlock(&session->lock); + send_reply: health_code_update(); - ret = send_response(cmd->sock, &response, sizeof(response)); + ret = send_response(conn->sock, &response, sizeof(response)); if (ret < 0) { - 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; } /* - * 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. + * 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. */ - ret = send_viewer_streams(cmd->sock, session, 0); + ret = send_viewer_streams(conn->sock, session_id, 0); if (ret < 0) { - goto end_unlock; + goto end_put_session; } -end_unlock: - rcu_read_unlock(); +end_put_session: + if (session) { + session_put(session); + } 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 send_streams = 0; ssize_t ret; @@ -918,289 +1161,530 @@ int viewer_attach_session(struct relay_command *cmd, enum lttng_viewer_seek seek_type; struct lttng_viewer_attach_session_request request; struct lttng_viewer_attach_session_response response; - struct relay_session *session; + struct relay_session *session = NULL; + enum lttng_viewer_attach_return_code viewer_attach_status; + bool closed = false; + uint64_t session_id; - assert(cmd); - assert(sessions_ht); + assert(conn); health_code_update(); /* Receive the request from the connected client. */ - ret = recv_request(cmd->sock, &request, sizeof(request)); + 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(); - session = session_find_by_id(sessions_ht, be64toh(request.session_id)); + 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 = session_get_by_id(session_id); if (!session) { - DBG("Relay session %" PRIu64 " not found", - be64toh(request.session_id)); - response.status = htobe32(VIEWER_ATTACH_UNK); + DBG("Relay session %" PRIu64 " not found", session_id); + response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK); goto send_reply; } - session_viewer_attach(session); - DBG("Attach session ID %" PRIu64 " received", be64toh(request.session_id)); + DBG("Attach session ID %" PRIu64 " received", session_id); - if (uatomic_read(&session->viewer_refcount) > 1) { - DBG("Already a viewer attached"); - response.status = htobe32(VIEWER_ATTACH_ALREADY); - session_viewer_detach(session); + pthread_mutex_lock(&session->lock); + if (!session->current_trace_chunk) { + /* + * Session is either being destroyed or it never had a trace + * chunk created against it. + */ + DBG("Session requested by live client has no current trace chunk, returning unknown session"); + response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK); goto send_reply; - } else if (session->live_timer == 0) { + } + 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; + viewer_attach_status = viewer_session_attach(conn->viewer_session, + session); + if (viewer_attach_status != LTTNG_VIEWER_ATTACH_OK) { + response.status = htobe32(viewer_attach_status); goto send_reply; - } else { - 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: - case VIEWER_SEEK_LAST: + 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) { - goto send_reply; - } - - ret = make_viewer_streams(session, seek_type, &nb_streams, NULL, NULL); + ret = make_viewer_streams(session, + conn->viewer_session->current_trace_chunk, seek_type, + &nb_streams, NULL, NULL, &closed); if (ret < 0) { - goto end_unlock; + goto end_put_session; } + pthread_mutex_unlock(&session->lock); + session_put(session); + session = NULL; + 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 = send_response(cmd->sock, &response, sizeof(response)); + ret = send_response(conn->sock, &response, sizeof(response)); if (ret < 0) { - 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; } /* Send stream and ignore the sent flag. */ - ret = send_viewer_streams(cmd->sock, session, 1); + ret = send_viewer_streams(conn->sock, session_id, 1); if (ret < 0) { - goto end_unlock; + goto end_put_session; } -end_unlock: - rcu_read_unlock(); +end_put_session: + if (session) { + pthread_mutex_unlock(&session->lock); + session_put(session); + } error: return ret; } +/* + * Open the index file if needed for the given vstream. + * + * 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. + */ +static int try_open_index(struct relay_viewer_stream *vstream, + struct relay_stream *rstream) +{ + int ret = 0; + const uint32_t connection_major = rstream->trace->session->major; + const uint32_t connection_minor = rstream->trace->session->minor; + enum lttng_trace_chunk_status chunk_status; + + if (vstream->index_file) { + goto end; + } + + /* + * First time, we open the index file and at least one index is ready. + */ + if (rstream->index_received_seqcount == 0) { + ret = -ENOENT; + goto end; + } + chunk_status = 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), + true, &vstream->index_file); + if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { + if (chunk_status == LTTNG_TRACE_CHUNK_STATUS_NO_FILE) { + ret = -ENOENT; + } else { + ret = -1; + } + } + +end: + return ret; +} + +/* + * 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. + * + * 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_index_status(struct relay_viewer_stream *vstream, + struct relay_stream *rstream, struct ctf_trace *trace, + struct lttng_viewer_index *index) +{ + int ret; + + DBG("Check index status: index_received_seqcount %" PRIu64 " " + "index_sent_seqcount %" PRIu64 " " + "for stream %" PRIu64, + rstream->index_received_seqcount, + vstream->index_sent_seqcount, + vstream->stream->stream_handle); + 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 == 0 || + (vstream->index_sent_seqcount != 0 && + 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). + * + * The sent seqcount can grow higher than receive seqcount on + * clear because the rotation performed by clear will push + * the index_sent_seqcount ahead (see + * viewer_stream_sync_tracefile_array_tail) and skip over + * packet sequence numbers. + */ + index->status = htobe32(LTTNG_VIEWER_INDEX_INACTIVE); + index->timestamp_end = htobe64(rstream->beacon_ts_end); + index->stream_id = htobe64(rstream->ctf_stream_id); + DBG("Check index status: inactive with beacon, for stream %" PRIu64, + vstream->stream->stream_handle); + goto index_ready; + } else if (rstream->index_received_seqcount == 0 || + (vstream->index_sent_seqcount != 0 && + 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. + * + * The sent seqcount can grow higher than receive seqcount on + * clear because the rotation performed by clear will push + * the index_sent_seqcount ahead (see + * viewer_stream_sync_tracefile_array_tail) and skip over + * packet sequence numbers. + */ + index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY); + DBG("Check index status: retry for stream %" PRIu64, + vstream->stream->stream_handle); + 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 == 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); + DBG("Check index status: retry: " + "tracefile array sequence number %" PRIu64 + " not in file for stream %" PRIu64, + vstream->index_sent_seqcount, + vstream->stream->stream_handle); + goto index_ready; + } + assert(tracefile_array_seq_in_file(rstream->tfa, + vstream->current_tracefile_id, + vstream->index_sent_seqcount)); + } + /* ret == 0 means successful so we continue. */ + ret = 0; + return ret; + +hup: + viewer_stream_put(vstream); +index_ready: + return 1; +} + /* * Send the next index for a stream. * * 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 ctf_trace *ctf_trace; - struct relay_session *session; + 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"); + memset(&viewer_index, 0, sizeof(viewer_index)); health_code_update(); - ret = recv_request(cmd->sock, &request_index, sizeof(request_index)); + ret = recv_request(conn->sock, &request_index, sizeof(request_index)); if (ret < 0) { goto end; } health_code_update(); - rcu_read_lock(); - 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)); + 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; } - ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht, vstream->path_name); - assert(ctf_trace); + /* Use back. ref. Protected by refcounts. */ + rstream = vstream->stream; + ctf_trace = rstream->trace; - memset(&viewer_index, 0, sizeof(viewer_index)); + /* 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 = 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 - */ - viewer_index.status = htobe32(VIEWER_INDEX_RETRY); - goto send_reply; - } else if (ret < 0) { - viewer_index.status = htobe32(VIEWER_INDEX_ERR); - goto send_reply; - } - vstream->index_read_fd = ret; + if (rstream->ongoing_rotation.is_set) { + /* Rotation is ongoing, try again later. */ + viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY); + goto send_reply; } - rstream = stream_find_by_id(relay_streams_ht, vstream->stream_handle); - assert(rstream); + if (rstream->trace->session->ongoing_rotation) { + /* Rotation is ongoing, try again later. */ + viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY); + goto send_reply; + } - 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->trace_chunk) { + uint64_t rchunk_id, vchunk_id; + + /* + * If the relay stream is not yet closed, ensure the viewer + * chunk matches the relay chunk after clear. + */ + if (lttng_trace_chunk_get_id(rstream->trace_chunk, + &rchunk_id) != LTTNG_TRACE_CHUNK_STATUS_OK) { + viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR); + goto send_reply; + } + if (lttng_trace_chunk_get_id( + conn->viewer_session->current_trace_chunk, + &vchunk_id) != LTTNG_TRACE_CHUNK_STATUS_OK) { + viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR); + 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); + if (rchunk_id != vchunk_id) { + DBG("Relay and viewer chunk ids differ: " + "rchunk_id %" PRIu64 " vchunk_id %" PRIu64, + rchunk_id, vchunk_id); + + lttng_trace_chunk_put( + conn->viewer_session->current_trace_chunk); + conn->viewer_session->current_trace_chunk = NULL; + ret = viewer_session_set_trace_chunk_copy( + conn->viewer_session, + rstream->trace_chunk); + if (ret) { + viewer_index.status = + htobe32(LTTNG_VIEWER_INDEX_ERR); goto send_reply; } } - 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 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 (conn->viewer_session->current_trace_chunk != + vstream->stream_file.trace_chunk) { + bool acquired_reference; - if (!ctf_trace->metadata_received || - ctf_trace->metadata_received > ctf_trace->metadata_sent) { - viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA; + DBG("Viewer session and viewer stream chunk differ: " + "vsession chunk %p vstream chunk %p", + conn->viewer_session->current_trace_chunk, + vstream->stream_file.trace_chunk); + lttng_trace_chunk_put(vstream->stream_file.trace_chunk); + acquired_reference = lttng_trace_chunk_get(conn->viewer_session->current_trace_chunk); + assert(acquired_reference); + vstream->stream_file.trace_chunk = + conn->viewer_session->current_trace_chunk; + viewer_stream_sync_tracefile_array_tail(vstream); + viewer_stream_close_files(vstream); } - ret = check_new_streams(vstream->session_id, sessions_ht); + ret = check_index_status(vstream, rstream, ctf_trace, &viewer_index); if (ret < 0) { - goto end_unlock; + goto error_put; } 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. + * We have no index to send and check_index_status has populated + * viewer_index's status. */ - 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; + } + /* At this point, ret is 0 thus we will be able to read the index. */ + assert(!ret); + + /* Try to open an index if one is needed for that stream. */ + ret = try_open_index(vstream, rstream); + if (ret == -ENOENT) { + if (rstream->closed) { + viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP); goto send_reply; - } + } else { + viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY); + goto send_reply; + } + } + if (ret < 0) { + viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR); 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)) { + /* + * 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.handle) { + char file_path[LTTNG_PATH_MAX]; + enum lttng_trace_chunk_status status; + struct fs_handle *fs_handle; + + 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; + } + /* - * The tracefile is closed in write, so we read up to EOF. + * It is possible the the file we are trying to open is + * missing if the stream has been closed (application exits with + * per-pid buffers) and a clear command has been performed. */ - 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); + status = lttng_trace_chunk_open_fs_handle( + vstream->stream_file.trace_chunk, + file_path, O_RDONLY, 0, &fs_handle, true); + if (status != LTTNG_TRACE_CHUNK_STATUS_OK) { + if (status == LTTNG_TRACE_CHUNK_STATUS_NO_FILE && + rstream->closed) { + viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP); goto send_reply; } - } else { - PERROR("Relay reading index file %d", vstream->index_read_fd); - viewer_index.status = htobe32(VIEWER_INDEX_ERR); + PERROR("Failed to open trace file for viewer stream"); + goto error_put; } + vstream->stream_file.handle = fs_handle; + } + + ret = check_new_streams(conn); + if (ret < 0) { + viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR); + goto send_reply; + } else if (ret == 1) { + viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM; + } + + ret = lttng_index_file_read(vstream->index_file, &packet_index); + if (ret) { + ERR("Relay error reading index file"); + 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; @@ -1210,22 +1694,53 @@ 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 = send_response(cmd->sock, &viewer_index, sizeof(viewer_index)); + ret = send_response(conn->sock, &viewer_index, sizeof(viewer_index)); if (ret < 0) { - 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(); - + 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; } @@ -1235,158 +1750,109 @@ 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; - struct ctf_trace *ctf_trace; - - assert(cmd); + 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; + uint64_t stream_id; DBG2("Relay get data packet"); health_code_update(); - ret = recv_request(cmd->sock, &get_packet_info, sizeof(get_packet_info)); + 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)); + memset(&reply_header, 0, sizeof(reply_header)); + stream_id = (uint64_t) be64toh(get_packet_info.stream_id); - rcu_read_lock(); - stream = viewer_stream_find_by_id(be64toh(get_packet_info.stream_id)); - if (!stream) { - goto error; - } - - 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 - * 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; - } - - 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; + vstream = viewer_stream_get_by_id(stream_id); + if (!vstream) { + DBG("Client requested packet of unknown stream id %" PRIu64, + 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 = fs_handle_seek(vstream->stream_file.handle, + be64toh(get_packet_info.offset), SEEK_SET); + if (lseek_ret < 0) { + PERROR("Failed to seek file system handle of viewer stream %" PRIu64 + " to offset %" PRIu64, + stream_id, + (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 = fs_handle_read(vstream->stream_file.handle, + reply + sizeof(reply_header), packet_data_len); + if (read_len < packet_data_len) { + PERROR("Failed to read from file system handle of viewer stream id %" PRIu64 + ", offset: %" PRIu64, + stream_id, + (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); + if (vstream) { + pthread_mutex_unlock(&vstream->stream->lock); + } +send_reply_nolock: health_code_update(); - ret = send_response(cmd->sock, &reply, sizeof(reply)); - if (ret < 0) { - goto end_unlock; + 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); } - health_code_update(); - if (send_data) { - health_code_update(); - ret = send_response(cmd->sock, data, len); - if (ret < 0) { - goto end_unlock; - } - health_code_update(); + health_code_update(); + if (ret < 0) { + PERROR("sendmsg of packet data failed"); + goto end_free; } - DBG("Sent %u bytes for stream %" PRIu64, len, - be64toh(get_packet_info.stream_id)); - -end_unlock: - free(data); - rcu_read_unlock(); + DBG("Sent %u bytes for stream %" PRIu64, reply_size, stream_id); +end_free: + free(reply); end: + if (vstream) { + viewer_stream_put(vstream); + } return ret; } @@ -1396,62 +1862,114 @@ 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; + int fd = -1; ssize_t read_len; uint64_t len = 0; char *data = NULL; struct lttng_viewer_get_metadata request; struct lttng_viewer_metadata_packet reply; - struct relay_viewer_stream *stream; - struct ctf_trace *ctf_trace; + struct relay_viewer_stream *vstream = NULL; - assert(cmd); + assert(conn); DBG("Relay get metadata"); health_code_update(); - ret = recv_request(cmd->sock, &request, sizeof(request)); + ret = recv_request(conn->sock, &request, sizeof(request)); if (ret < 0) { goto end; } health_code_update(); - rcu_read_lock(); - stream = viewer_stream_find_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; } - 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); + if (vstream->metadata_sent >= vstream->stream->metadata_received) { + /* + * The live viewers expect to receive a NO_NEW_METADATA + * status before a stream disappears, otherwise they abort the + * entire live connection when receiving an error status. + * + * Clear feature resets the metadata_sent to 0 until the + * same metadata is received again. + */ + reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA); + /* + * The live viewer considers a closed 0 byte metadata stream as + * an error. + */ + if (vstream->metadata_sent > 0) { + vstream->stream->no_new_metadata_notified = true; + if (vstream->stream->closed) { + /* Release ownership for the viewer metadata stream. */ + viewer_stream_put(vstream); + } + } goto send_reply; } - /* first time, we open the metadata file */ - if (stream->read_fd < 0) { - char fullpath[PATH_MAX]; + len = vstream->stream->metadata_received - vstream->metadata_sent; - ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name, - stream->channel_name); + /* first time, we open the metadata file */ + if (!vstream->stream_file.handle) { + struct fs_handle *fs_handle; + 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"); + + /* + * It is possible the the metadata file we are trying to open is + * missing if the stream has been closed (application exits with + * per-pid buffers) and a clear command has been performed. + */ + status = lttng_trace_chunk_open_fs_handle( + vstream->stream_file.trace_chunk, + file_path, O_RDONLY, 0, &fs_handle, true); + if (status != LTTNG_TRACE_CHUNK_STATUS_OK) { + if (status == LTTNG_TRACE_CHUNK_STATUS_NO_FILE) { + reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA); + len = 0; + if (vstream->stream->closed) { + viewer_stream_put(vstream); + } + goto send_reply; + } + PERROR("Failed to open metadata file for viewer stream"); goto error; } - stream->read_fd = ret; + vstream->stream_file.handle = fs_handle; } reply.len = htobe64(len); @@ -1461,272 +1979,253 @@ int viewer_get_metadata(struct relay_command *cmd) goto error; } - read_len = lttng_read(stream->read_fd, data, len); + fd = fs_handle_get_fd(vstream->stream_file.handle); + if (fd < 0) { + ERR("Failed to restore viewer stream file system handle"); + goto error; + } + read_len = lttng_read(fd, data, len); + fs_handle_put_fd(vstream->stream_file.handle); + fd = -1; if (read_len < len) { PERROR("Relay reading metadata file"); goto error; } - ctf_trace->metadata_sent += read_len; - reply.status = htobe32(VIEWER_METADATA_OK); + vstream->metadata_sent += read_len; + 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 = send_response(cmd->sock, &reply, sizeof(reply)); + if (vstream) { + pthread_mutex_unlock(&vstream->stream->lock); + } + ret = send_response(conn->sock, &reply, sizeof(reply)); if (ret < 0) { - goto end_unlock; + goto end_free; } health_code_update(); if (len > 0) { - ret = send_response(cmd->sock, data, len); + ret = send_response(conn->sock, data, len); if (ret < 0) { - 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 - */ -static -void live_relay_unknown_command(struct relay_command *cmd) -{ - struct lttcomm_relayd_generic_reply reply; - - reply.ret_code = htobe32(LTTNG_ERR_UNK); - (void) send_response(cmd->sock, &reply, sizeof(reply)); -} - -/* - * Process the commands received on the control socket + * Create a viewer session. + * + * Return 0 on success or else a negative value. */ static -int process_control(struct lttng_viewer_cmd *recv_hdr, - struct relay_command *cmd, struct lttng_ht *sessions_ht) +int viewer_create_session(struct relay_connection *conn) { - int ret = 0; - uint32_t msg_value; - - assert(recv_hdr); - assert(cmd); - assert(sessions_ht); + int ret; + struct lttng_viewer_create_session_response resp; - msg_value = be32toh(recv_hdr->cmd); + DBG("Viewer create session received"); - /* - * 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; + 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 (msg_value) { - 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; - 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); - 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; + + DBG("Viewer detach session received"); - assert(events); + assert(conn); - lttng_poll_del(events, pollfd); + health_code_update(); - ret = close(pollfd); + /* 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; + } - assert(events); - assert(relay_connections_ht); + health_code_update(); - relay_connection = zmalloc(sizeof(struct relay_command)); - if (relay_connection == NULL) { - PERROR("Relay command zmalloc"); - goto error; + memset(&response, 0, sizeof(response)); + DBG("Detaching from session ID %" PRIu64, viewer_session_to_close); + + 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; - 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 destroy_viewer_streams_by_session(struct relay_session *session) +static +int process_control(struct lttng_viewer_cmd *recv_hdr, + struct relay_connection *conn) { - struct relay_viewer_stream *stream; - struct lttng_ht_iter iter; - - assert(session); - - rcu_read_lock(); - cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, stream, - stream_n.node) { - struct ctf_trace *ctf_trace; - - health_code_update(); - if (stream->session_id != session->id) { - continue; - } - - ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht, - stream->path_name); - assert(ctf_trace); - - viewer_stream_delete(stream); + int ret = 0; + uint32_t msg_value; - if (stream->metadata_flag) { - ctf_trace->metadata_sent = 0; - ctf_trace->viewer_metadata_stream = NULL; - } + msg_value = be32toh(recv_hdr->cmd); - viewer_stream_destroy(ctf_trace, stream); + /* + * 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; } - 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); + 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; } + +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, - struct lttng_ht *sessions_ht) +void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd) { int ret; - struct relay_session *session; - assert(relay_connections_ht); - assert(iter); - assert(relay_connection); - assert(sessions_ht); + (void) lttng_poll_del(events, pollfd); - DBG("Cleaning connection of session ID %" PRIu64, - relay_connection->session_id); - - rcu_read_lock(); - ret = lttng_ht_del(relay_connections_ht, iter); - assert(!ret); - - 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); + ret = fd_tracker_close_unsuspendable_fd(the_fd_tracker, &pollfd, 1, + fd_tracker_util_close_fd, NULL); + if (ret < 0) { + ERR("Closing pollfd %d", pollfd); } - rcu_read_unlock(); - - call_rcu(&relay_connection->rcu_node, deferred_free_connection); } /* @@ -1737,14 +2236,11 @@ 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"); @@ -1757,17 +2253,18 @@ 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); + ret = create_named_thread_poll_set(&events, 2, + "Live viewer worker thread epoll"); if (ret < 0) { 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; } @@ -1808,106 +2305,99 @@ restart: health_code_update(); /* Thread quit pipe has been closed. Killing thread. */ - ret = check_live_conn_pipe(pollfd, revents); + ret = check_thread_quit_pipe(pollfd, revents); if (ret) { err = 0; 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, 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_ctx->sessions_ht); - } 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, relay_ctx->sessions_ht); - 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, relay_ctx->sessions_ht); + 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); } } } exit: error: - lttng_poll_clean(&events); + (void) fd_tracker_util_poll_clean(the_fd_tracker, &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, - relay_ctx->sessions_ht); + 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 */ + (void) fd_tracker_util_pipe_close(the_fd_tracker, live_conn_pipe); if (err) { DBG("Viewer worker thread exited with error"); } @@ -1918,7 +2408,9 @@ 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; } @@ -1927,57 +2419,56 @@ 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 fd_tracker_util_pipe_open_cloexec(the_fd_tracker, + "Live connection pipe", 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 */ @@ -1986,74 +2477,86 @@ 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; }