X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-relayd%2Fmain.c;h=62a5bd7cefc16026ff05f3e3741e4b4258837fd1;hp=1ec1eeaa4b14be22fa0a8cde2d708279a50bcf4b;hb=9898f786493c5e60dcf99fa3102bb95752ad73ed;hpb=b4aacfdce9fe9fddc01b738a8cc807d764245cef diff --git a/src/bin/lttng-relayd/main.c b/src/bin/lttng-relayd/main.c index 1ec1eeaa4..62a5bd7ce 100644 --- a/src/bin/lttng-relayd/main.c +++ b/src/bin/lttng-relayd/main.c @@ -2,6 +2,7 @@ * Copyright (C) 2012 - Julien Desfossez * David Goulet * 2013 - Jérémie Galarneau + * 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, @@ -17,7 +18,6 @@ * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. */ -#define _GNU_SOURCE #define _LGPL_SOURCE #include #include @@ -37,15 +37,17 @@ #include #include #include +#include #include #include -#include +#include #include #include #include #include #include +#include #include #include #include @@ -54,7 +56,11 @@ #include #include #include -#include +#include +#include +#include +#include +#include #include "cmd.h" #include "ctf-trace.h" @@ -68,6 +74,25 @@ #include "session.h" #include "stream.h" #include "connection.h" +#include "tracefile-array.h" +#include "tcp_keep_alive.h" +#include "sessiond-trace-chunks.h" + +static const char *help_msg = +#ifdef LTTNG_EMBED_HELP +#include +#else +NULL +#endif +; + +enum relay_connection_status { + RELAY_CONNECTION_STATUS_OK, + /* An error occurred while processing an event on the connection. */ + RELAY_CONNECTION_STATUS_ERROR, + /* Connection closed/shutdown cleanly. */ + RELAY_CONNECTION_STATUS_CLOSED, +}; /* command line options */ char *opt_output_path; @@ -79,6 +104,10 @@ static int opt_daemon, opt_background; */ #define NR_LTTNG_RELAY_READY 3 static int lttng_relay_ready = NR_LTTNG_RELAY_READY; + +/* Size of receive buffer. */ +#define RECV_DATA_BUFFER_SIZE 65536 + static int recv_child_signal; /* Set to 1 when a SIGUSR1 signal is received. */ static pid_t child_ppid; /* Internal parent PID use with daemonize. */ @@ -113,6 +142,11 @@ static pthread_t dispatcher_thread; static pthread_t worker_thread; static pthread_t health_thread; +/* + * last_relay_stream_id_lock protects last_relay_stream_id increment + * atomicity on 32-bit architectures. + */ +static pthread_mutex_t last_relay_stream_id_lock = PTHREAD_MUTEX_INITIALIZER; static uint64_t last_relay_stream_id; /* @@ -123,26 +157,20 @@ static uint64_t last_relay_stream_id; */ static struct relay_conn_queue relay_conn_queue; -/* buffer allocated at startup, used to store the trace data */ -static char *data_buffer; -static unsigned int data_buffer_size; - -/* We need those values for the file/dir creation. */ -static uid_t relayd_uid; -static gid_t relayd_gid; - /* Global relay stream hash table. */ struct lttng_ht *relay_streams_ht; /* Global relay viewer stream hash table. */ struct lttng_ht *viewer_streams_ht; -/* Global hash table that stores relay index object. */ -struct lttng_ht *indexes_ht; +/* Global relay sessions hash table. */ +struct lttng_ht *sessions_ht; /* Relayd health monitoring */ struct health_app *health_relayd; +struct sessiond_trace_chunk_registry *sessiond_trace_chunk_registry; + static struct option long_options[] = { { "control-port", 1, 0, 'C', }, { "data-port", 1, 0, 'D', }, @@ -154,29 +182,11 @@ static struct option long_options[] = { { "output", 1, 0, 'o', }, { "verbose", 0, 0, 'v', }, { "config", 1, 0, 'f' }, + { "version", 0, 0, 'V' }, { NULL, 0, 0, 0, }, }; -static const char *config_ignore_options[] = { "help", "config" }; - -/* - * usage function on stderr - */ -static -void usage(void) -{ - fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname); - fprintf(stderr, " -h, --help Display this usage.\n"); - fprintf(stderr, " -d, --daemonize Start as a daemon.\n"); - fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n"); - fprintf(stderr, " -C, --control-port URL Control port listening.\n"); - fprintf(stderr, " -D, --data-port URL Data port listening.\n"); - fprintf(stderr, " -L, --live-port URL Live view port listening.\n"); - fprintf(stderr, " -o, --output PATH Output path for traces. Must use an absolute path.\n"); - fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n"); - fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n"); - fprintf(stderr, " -f --config Load daemon configuration file\n"); -} +static const char *config_ignore_options[] = { "help", "config", "version" }; /* * Take an option from the getopt output and set it in the right variable to be @@ -184,8 +194,7 @@ void usage(void) * * Return 0 on success else a negative value. */ -static -int set_option(int opt, const char *arg, const char *optname) +static int set_option(int opt, const char *arg, const char *optname) { int ret; @@ -197,33 +206,48 @@ int set_option(int opt, const char *arg, const char *optname) } break; case 'C': - ret = uri_parse(arg, &control_uri); - if (ret < 0) { - ERR("Invalid control URI specified"); - goto end; - } - if (control_uri->port == 0) { - control_uri->port = DEFAULT_NETWORK_CONTROL_PORT; + if (lttng_is_setuid_setgid()) { + WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", + "-C, --control-port"); + } else { + ret = uri_parse(arg, &control_uri); + if (ret < 0) { + ERR("Invalid control URI specified"); + goto end; + } + if (control_uri->port == 0) { + control_uri->port = DEFAULT_NETWORK_CONTROL_PORT; + } } break; case 'D': - ret = uri_parse(arg, &data_uri); - if (ret < 0) { - ERR("Invalid data URI specified"); - goto end; - } - if (data_uri->port == 0) { - data_uri->port = DEFAULT_NETWORK_DATA_PORT; + if (lttng_is_setuid_setgid()) { + WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", + "-D, -data-port"); + } else { + ret = uri_parse(arg, &data_uri); + if (ret < 0) { + ERR("Invalid data URI specified"); + goto end; + } + if (data_uri->port == 0) { + data_uri->port = DEFAULT_NETWORK_DATA_PORT; + } } break; case 'L': - ret = uri_parse(arg, &live_uri); - if (ret < 0) { - ERR("Invalid live URI specified"); - goto end; - } - if (live_uri->port == 0) { - live_uri->port = DEFAULT_NETWORK_VIEWER_PORT; + if (lttng_is_setuid_setgid()) { + WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", + "-L, -live-port"); + } else { + ret = uri_parse(arg, &live_uri); + if (ret < 0) { + ERR("Invalid live URI specified"); + goto end; + } + if (live_uri->port == 0) { + live_uri->port = DEFAULT_NETWORK_VIEWER_PORT; + } } break; case 'd': @@ -233,23 +257,40 @@ int set_option(int opt, const char *arg, const char *optname) opt_background = 1; break; case 'g': - tracing_group_name = strdup(arg); - if (tracing_group_name == NULL) { - ret = -errno; - PERROR("strdup"); - goto end; + if (lttng_is_setuid_setgid()) { + WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", + "-g, --group"); + } else { + tracing_group_name = strdup(arg); + if (tracing_group_name == NULL) { + ret = -errno; + PERROR("strdup"); + goto end; + } + tracing_group_name_override = 1; } - tracing_group_name_override = 1; break; case 'h': - usage(); + ret = utils_show_help(8, "lttng-relayd", help_msg); + if (ret) { + ERR("Cannot show --help for `lttng-relayd`"); + perror("exec"); + } exit(EXIT_FAILURE); + case 'V': + fprintf(stdout, "%s\n", VERSION); + exit(EXIT_SUCCESS); case 'o': - ret = asprintf(&opt_output_path, "%s", arg); - if (ret < 0) { - ret = -errno; - PERROR("asprintf opt_output_path"); - goto end; + if (lttng_is_setuid_setgid()) { + WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", + "-o, --output"); + } else { + ret = asprintf(&opt_output_path, "%s", arg); + if (ret < 0) { + ret = -errno; + PERROR("asprintf opt_output_path"); + goto end; + } } break; case 'v': @@ -279,11 +320,10 @@ end: /* * config_entry_handler_cb used to handle options read from a config file. - * See config_entry_handler_cb comment in common/config/config.h for the + * See config_entry_handler_cb comment in common/config/session-config.h for the * return value conventions. */ -static -int config_entry_handler(const struct config_entry *entry, void *unused) +static int config_entry_handler(const struct config_entry *entry, void *unused) { int ret = 0, i; @@ -306,9 +346,9 @@ int config_entry_handler(const struct config_entry *entry, void *unused) } /* - * If the option takes no argument on the command line, we have to - * check if the value is "true". We support non-zero numeric values, - * true, on and yes. + * If the option takes no argument on the command line, + * we have to check if the value is "true". We support + * non-zero numeric values, true, on and yes. */ if (!long_options[i].has_arg) { ret = config_parse_value(entry->value); @@ -333,8 +373,7 @@ end: return ret; } -static -int set_options(int argc, char **argv) +static int set_options(int argc, char **argv) { int c, ret = 0, option_index = 0, retval = 0; int orig_optopt = optopt, orig_optind = optind; @@ -359,9 +398,14 @@ int set_options(int argc, char **argv) continue; } - config_path = utils_expand_path(optarg); - if (!config_path) { - ERR("Failed to resolve path: %s", optarg); + if (lttng_is_setuid_setgid()) { + WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", + "-f, --config"); + } else { + config_path = utils_expand_path(optarg); + if (!config_path) { + ERR("Failed to resolve path: %s", optarg); + } } } @@ -452,21 +496,32 @@ exit: return retval; } +static void print_global_objects(void) +{ + rcu_register_thread(); + + print_viewer_streams(); + print_relay_streams(); + print_sessions(); + + rcu_unregister_thread(); +} + /* * Cleanup the daemon */ -static -void relayd_cleanup(struct relay_local_data *relay_ctx) +static void relayd_cleanup(void) { + print_global_objects(); + DBG("Cleaning up"); if (viewer_streams_ht) lttng_ht_destroy(viewer_streams_ht); if (relay_streams_ht) lttng_ht_destroy(relay_streams_ht); - if (relay_ctx && relay_ctx->sessions_ht) - lttng_ht_destroy(relay_ctx->sessions_ht); - free(relay_ctx); + if (sessions_ht) + lttng_ht_destroy(sessions_ht); /* free the dynamically allocated opt_output_path */ free(opt_output_path); @@ -486,8 +541,7 @@ void relayd_cleanup(struct relay_local_data *relay_ctx) /* * Write to writable pipe used to notify a thread. */ -static -int notify_thread_pipe(int wpipe) +static int notify_thread_pipe(int wpipe) { ssize_t ret; @@ -501,8 +555,7 @@ end: return ret; } -static -int notify_health_quit_pipe(int *pipe) +static int notify_health_quit_pipe(int *pipe) { ssize_t ret; @@ -551,13 +604,9 @@ int lttng_relay_stop_threads(void) * Simply stop all worker threads, leaving main() return gracefully after * joining all threads and calling cleanup(). */ -static -void sighandler(int sig) +static void sighandler(int sig) { switch (sig) { - case SIGPIPE: - DBG("SIGPIPE caught"); - return; case SIGINT: DBG("SIGINT caught"); if (lttng_relay_stop_threads()) { @@ -582,8 +631,7 @@ void sighandler(int sig) * Setup signal handler for : * SIGINT, SIGTERM, SIGPIPE */ -static -int set_signal_handler(void) +static int set_signal_handler(void) { int ret = 0; struct sigaction sa; @@ -594,9 +642,10 @@ int set_signal_handler(void) return ret; } - sa.sa_handler = sighandler; sa.sa_mask = sigset; sa.sa_flags = 0; + + sa.sa_handler = sighandler; if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) { PERROR("sigaction"); return ret; @@ -607,12 +656,13 @@ int set_signal_handler(void) return ret; } - if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) { + if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) { PERROR("sigaction"); return ret; } - if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) { + sa.sa_handler = SIG_IGN; + if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) { PERROR("sigaction"); return ret; } @@ -637,8 +687,7 @@ void lttng_relay_notify_ready(void) * * Return -1 on error or 0 if all pipes are created. */ -static -int init_thread_quit_pipe(void) +static int init_thread_quit_pipe(void) { int ret; @@ -650,8 +699,7 @@ int init_thread_quit_pipe(void) /* * 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) +static int create_thread_poll_set(struct lttng_poll_event *events, int size) { int ret; @@ -682,8 +730,7 @@ error: * * Return 1 if it was triggered else 0; */ -static -int check_thread_quit_pipe(int fd, uint32_t events) +static int check_thread_quit_pipe(int fd, uint32_t events) { if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) { return 1; @@ -695,8 +742,7 @@ int check_thread_quit_pipe(int fd, uint32_t events) /* * Create and init socket from uri. */ -static -struct lttcomm_sock *relay_init_sock(struct lttng_uri *uri) +static struct lttcomm_sock *relay_socket_create(struct lttng_uri *uri) { int ret; struct lttcomm_sock *sock = NULL; @@ -715,6 +761,7 @@ struct lttcomm_sock *relay_init_sock(struct lttng_uri *uri) ret = sock->ops->bind(sock); if (ret < 0) { + PERROR("Failed to bind socket"); goto error; } @@ -733,64 +780,10 @@ error: return NULL; } -/* - * Return nonzero if stream needs to be closed. - */ -static -int close_stream_check(struct relay_stream *stream) -{ - if (stream->close_flag && stream->prev_seq == stream->last_net_seq_num) { - /* - * We are about to close the stream so set the data pending flag to 1 - * which will make the end data pending command skip the stream which - * is now closed and ready. Note that after proceeding to a file close, - * the written file is ready for reading. - */ - stream->data_pending_check_done = 1; - return 1; - } - return 0; -} - -static void try_close_stream(struct relay_session *session, - struct relay_stream *stream) -{ - int ret; - struct ctf_trace *ctf_trace; - - assert(session); - assert(stream); - - if (!close_stream_check(stream)) { - /* Can't close it, not ready for that. */ - goto end; - } - - ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht, - stream->path_name); - assert(ctf_trace); - - pthread_mutex_lock(&session->viewer_ready_lock); - ctf_trace->invalid_flag = 1; - pthread_mutex_unlock(&session->viewer_ready_lock); - - ret = stream_close(session, stream); - if (ret || session->snapshot) { - /* Already close thus the ctf trace is being or has been destroyed. */ - goto end; - } - - ctf_trace_try_destroy(session, ctf_trace); - -end: - return; -} - /* * This thread manages the listening for new connections on the network */ -static -void *relay_thread_listener(void *data) +static void *relay_thread_listener(void *data) { int i, ret, pollfd, err = -1; uint32_t revents, nb_fd; @@ -803,18 +796,19 @@ void *relay_thread_listener(void *data) health_code_update(); - control_sock = relay_init_sock(control_uri); + control_sock = relay_socket_create(control_uri); if (!control_sock) { goto error_sock_control; } - data_sock = relay_init_sock(data_uri); + data_sock = relay_socket_create(data_uri); if (!data_sock) { goto error_sock_relay; } /* - * Pass 3 as size here for the thread quit pipe, control and data socket. + * Pass 3 as size here for the thread quit pipe, control and + * data socket. */ ret = create_thread_poll_set(&events, 3); if (ret < 0) { @@ -875,38 +869,32 @@ 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 + * sessiond/consumerd connection is allocated in + * this thread, enqueued to a global queue and + * dequeued (and freed) in the worker thread. */ int val = 1; struct relay_connection *new_conn; struct lttcomm_sock *newsock; - - new_conn = connection_create(); - if (!new_conn) { - goto error; - } + enum connection_type type; if (pollfd == data_sock->fd) { - new_conn->type = RELAY_DATA; + type = RELAY_DATA; newsock = data_sock->ops->accept(data_sock); DBG("Relay data connection accepted, socket %d", newsock->fd); } else { assert(pollfd == control_sock->fd); - new_conn->type = RELAY_CONTROL; + type = RELAY_CONTROL; newsock = control_sock->ops->accept(control_sock); DBG("Relay control connection accepted, socket %d", newsock->fd); } if (!newsock) { PERROR("accepting sock"); - connection_free(new_conn); goto error; } @@ -915,20 +903,39 @@ restart: if (ret < 0) { PERROR("setsockopt inet"); lttcomm_destroy_sock(newsock); - connection_free(new_conn); goto error; } - new_conn->sock = newsock; + + ret = socket_apply_keep_alive_config(newsock->fd); + if (ret < 0) { + ERR("Failed to apply TCP keep-alive configuration on socket (%i)", + newsock->fd); + lttcomm_destroy_sock(newsock); + goto error; + } + + new_conn = connection_create(newsock, type); + if (!new_conn) { + lttcomm_destroy_sock(newsock); + goto error; + } /* Enqueue request for the dispatcher thread. */ cds_wfcq_enqueue(&relay_conn_queue.head, &relay_conn_queue.tail, &new_conn->qnode); /* - * Wake the dispatch queue futex. Implicit memory barrier with - * the exchange in cds_wfcq_enqueue. + * Wake the dispatch queue futex. + * Implicit memory barrier with the + * exchange in cds_wfcq_enqueue. */ futex_nto1_wake(&relay_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; } } } @@ -968,8 +975,7 @@ error_sock_control: /* * This thread manages the dispatching of the requests to worker threads */ -static -void *relay_thread_dispatcher(void *data) +static void *relay_thread_dispatcher(void *data) { int err = -1; ssize_t ret; @@ -986,12 +992,16 @@ void *relay_thread_dispatcher(void *data) health_code_update(); - while (!CMM_LOAD_SHARED(dispatch_thread_exit)) { + for (;;) { health_code_update(); /* Atomically prepare the queue futex */ futex_nto1_prepare(&relay_conn_queue.futex); + if (CMM_LOAD_SHARED(dispatch_thread_exit)) { + break; + } + do { health_code_update(); @@ -1008,14 +1018,15 @@ void *relay_thread_dispatcher(void *data) DBG("Dispatching request waiting on sock %d", new_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(relay_conn_pipe[1], &new_conn, sizeof(new_conn)); if (ret < 0) { PERROR("write connection pipe"); - connection_destroy(new_conn); + connection_put(new_conn); goto error; } } while (node != NULL); @@ -1041,72 +1052,9 @@ error_testpoint: return NULL; } -static void try_close_streams(struct relay_session *session) -{ - struct ctf_trace *ctf_trace; - struct lttng_ht_iter iter; - - assert(session); - - pthread_mutex_lock(&session->viewer_ready_lock); - rcu_read_lock(); - cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace, - node.node) { - struct relay_stream *stream; - - /* Close streams. */ - cds_list_for_each_entry(stream, &ctf_trace->stream_list, trace_list) { - stream_close(session, stream); - } - - ctf_trace->invalid_flag = 1; - ctf_trace_try_destroy(session, ctf_trace); - } - rcu_read_unlock(); - pthread_mutex_unlock(&session->viewer_ready_lock); -} - -/* - * Try to destroy a session within a connection. - */ -static void destroy_session(struct relay_session *session, - struct lttng_ht *sessions_ht) -{ - assert(session); - assert(sessions_ht); - - /* Indicate that this session can be destroyed from now on. */ - session->close_flag = 1; - - try_close_streams(session); - - /* - * This will try to delete and destroy the session if no viewer is attached - * to it meaning the refcount is down to zero. - */ - session_try_destroy(sessions_ht, session); -} - -/* - * Copy index data from the control port to a given index object. - */ -static void copy_index_control_data(struct relay_index *index, - struct lttcomm_relayd_index *data) +static bool session_streams_have_index(const struct relay_session *session) { - assert(index); - assert(data); - - /* - * The index on disk is encoded in big endian, so we don't need to convert - * the data received on the network. The data_offset value is NEVER - * modified here and is updated by the data thread. - */ - index->index_data.packet_size = data->packet_size; - index->index_data.content_size = data->content_size; - index->index_data.timestamp_begin = data->timestamp_begin; - index->index_data.timestamp_end = data->timestamp_end; - index->index_data.events_discarded = data->events_discarded; - index->index_data.stream_id = data->stream_id; + return session->minor >= 4 && !session->snapshot; } /* @@ -1114,57 +1062,134 @@ static void copy_index_control_data(struct relay_index *index, * * On success, send back the session id or else return a negative value. */ -static -int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_connection *conn) +static int relay_create_session(const struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn, + const struct lttng_buffer_view *payload) { - int ret = 0, send_ret; - struct relay_session *session; - struct lttcomm_relayd_status_session reply; - - assert(recv_hdr); - assert(conn); + int ret = 0; + ssize_t send_ret; + struct relay_session *session = NULL; + struct lttcomm_relayd_create_session_reply_2_11 reply = {}; + char session_name[LTTNG_NAME_MAX] = {}; + char hostname[LTTNG_HOST_NAME_MAX] = {}; + uint32_t live_timer = 0; + bool snapshot = false; + bool session_name_contains_creation_timestamp = false; + /* Left nil for peers < 2.11. */ + char base_path[LTTNG_PATH_MAX] = {}; + lttng_uuid sessiond_uuid = {}; + LTTNG_OPTIONAL(uint64_t) id_sessiond = {}; + LTTNG_OPTIONAL(uint64_t) current_chunk_id = {}; + LTTNG_OPTIONAL(time_t) creation_time = {}; + struct lttng_dynamic_buffer reply_payload; + + lttng_dynamic_buffer_init(&reply_payload); + + if (conn->minor < 4) { + /* From 2.1 to 2.3 */ + ret = 0; + } else if (conn->minor >= 4 && conn->minor < 11) { + /* From 2.4 to 2.10 */ + ret = cmd_create_session_2_4(payload, session_name, + hostname, &live_timer, &snapshot); + } else { + bool has_current_chunk; + uint64_t current_chunk_id_value; + time_t creation_time_value; + uint64_t id_sessiond_value; + + /* From 2.11 to ... */ + ret = cmd_create_session_2_11(payload, session_name, hostname, + base_path, &live_timer, &snapshot, &id_sessiond_value, + sessiond_uuid, &has_current_chunk, + ¤t_chunk_id_value, &creation_time_value, + &session_name_contains_creation_timestamp); + if (lttng_uuid_is_nil(sessiond_uuid)) { + /* The nil UUID is reserved for pre-2.11 clients. */ + ERR("Illegal nil UUID announced by peer in create session command"); + ret = -1; + goto send_reply; + } + LTTNG_OPTIONAL_SET(&id_sessiond, id_sessiond_value); + LTTNG_OPTIONAL_SET(&creation_time, creation_time_value); + if (has_current_chunk) { + LTTNG_OPTIONAL_SET(¤t_chunk_id, + current_chunk_id_value); + } + } - memset(&reply, 0, sizeof(reply)); + if (ret < 0) { + goto send_reply; + } - session = session_create(); + session = session_create(session_name, hostname, base_path, live_timer, + snapshot, sessiond_uuid, + id_sessiond.is_set ? &id_sessiond.value : NULL, + current_chunk_id.is_set ? ¤t_chunk_id.value : NULL, + creation_time.is_set ? &creation_time.value : NULL, + conn->major, conn->minor, + session_name_contains_creation_timestamp); if (!session) { ret = -1; - goto error; + goto send_reply; } - session->minor = conn->minor; - session->major = conn->major; - conn->session_id = session->id; + assert(!conn->session); conn->session = session; - - reply.session_id = htobe64(session->id); - - switch (conn->minor) { - case 1: - case 2: - case 3: - break; - case 4: /* LTTng sessiond 2.4 */ - default: - ret = cmd_create_session_2_4(conn, session); - } - - lttng_ht_add_unique_u64(conn->sessions_ht, &session->session_n); DBG("Created session %" PRIu64, session->id); -error: + reply.generic.session_id = htobe64(session->id); + +send_reply: if (ret < 0) { - reply.ret_code = htobe32(LTTNG_ERR_FATAL); + reply.generic.ret_code = htobe32(LTTNG_ERR_FATAL); } else { - reply.ret_code = htobe32(LTTNG_OK); + reply.generic.ret_code = htobe32(LTTNG_OK); } - send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); - if (send_ret < 0) { - ERR("Relayd sending session id"); - ret = send_ret; + if (conn->minor < 11) { + /* From 2.1 to 2.10 */ + ret = lttng_dynamic_buffer_append(&reply_payload, + &reply.generic, sizeof(reply.generic)); + if (ret) { + ERR("Failed to append \"create session\" command reply header to payload buffer"); + ret = -1; + goto end; + } + } else { + const uint32_t output_path_length = + session ? strlen(session->output_path) + 1 : 0; + + reply.output_path_length = htobe32(output_path_length); + ret = lttng_dynamic_buffer_append( + &reply_payload, &reply, sizeof(reply)); + if (ret) { + ERR("Failed to append \"create session\" command reply header to payload buffer"); + goto end; + } + + if (output_path_length) { + ret = lttng_dynamic_buffer_append(&reply_payload, + session->output_path, + output_path_length); + if (ret) { + ERR("Failed to append \"create session\" command reply path to payload buffer"); + goto end; + } + } } + send_ret = conn->sock->ops->sendmsg(conn->sock, reply_payload.data, + reply_payload.size, 0); + if (send_ret < (ssize_t) reply_payload.size) { + ERR("Failed to send \"create session\" command reply of %zu bytes (ret = %zd)", + reply_payload.size, send_ret); + ret = -1; + } +end: + if (ret < 0 && session) { + session_put(session); + } + lttng_dynamic_buffer_reset(&reply_payload); return ret; } @@ -1172,225 +1197,313 @@ error: * When we have received all the streams and the metadata for a channel, * we make them visible to the viewer threads. */ -static -void set_viewer_ready_flag(struct relay_connection *conn) +static void publish_connection_local_streams(struct relay_connection *conn) { - struct relay_stream *stream, *tmp_stream; + struct relay_stream *stream; + struct relay_session *session = conn->session; + + /* + * We publish all streams belonging to a session atomically wrt + * session lock. + */ + pthread_mutex_lock(&session->lock); + rcu_read_lock(); + cds_list_for_each_entry_rcu(stream, &session->recv_list, + recv_node) { + stream_publish(stream); + } + rcu_read_unlock(); - pthread_mutex_lock(&conn->session->viewer_ready_lock); - cds_list_for_each_entry_safe(stream, tmp_stream, &conn->recv_head, - recv_list) { - stream->viewer_ready = 1; - cds_list_del(&stream->recv_list); + /* + * Inform the viewer that there are new streams in the session. + */ + if (session->viewer_attached) { + uatomic_set(&session->new_streams, 1); } - pthread_mutex_unlock(&conn->session->viewer_ready_lock); - return; + pthread_mutex_unlock(&session->lock); } -/* - * Add a recv handle node to the connection recv list with the given stream - * handle. A new node is allocated thus must be freed when the node is deleted - * from the list. - */ -static void queue_stream(struct relay_stream *stream, - struct relay_connection *conn) +static int conform_channel_path(char *channel_path) { - assert(conn); - assert(stream); + int ret = 0; + + if (strstr("../", channel_path)) { + ERR("Refusing channel path as it walks up the path hierarchy: \"%s\"", + channel_path); + ret = -1; + goto end; + } + + if (*channel_path == '/') { + const size_t len = strlen(channel_path); - cds_list_add(&stream->recv_list, &conn->recv_head); + /* + * Channel paths from peers prior to 2.11 are expressed as an + * absolute path that is, in reality, relative to the relay + * daemon's output directory. Remove the leading slash so it + * is correctly interpreted as a relative path later on. + * + * len (and not len - 1) is used to copy the trailing NULL. + */ + bcopy(channel_path + 1, channel_path, len); + } +end: + return ret; } /* * relay_add_stream: allocate a new stream for a session */ -static -int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_connection *conn) +static int relay_add_stream(const struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn, + const struct lttng_buffer_view *payload) { - int ret, send_ret; + int ret; + ssize_t send_ret; struct relay_session *session = conn->session; struct relay_stream *stream = NULL; struct lttcomm_relayd_status_stream reply; - struct ctf_trace *trace; + struct ctf_trace *trace = NULL; + uint64_t stream_handle = -1ULL; + char *path_name = NULL, *channel_name = NULL; + uint64_t tracefile_size = 0, tracefile_count = 0; + LTTNG_OPTIONAL(uint64_t) stream_chunk_id = {}; - if (!session || conn->version_check_done == 0) { + if (!session || !conn->version_check_done) { ERR("Trying to add a stream before version check"); ret = -1; goto end_no_session; } - stream = zmalloc(sizeof(struct relay_stream)); - if (stream == NULL) { - PERROR("relay stream zmalloc"); - ret = -1; - goto end_no_session; + if (session->minor == 1) { + /* For 2.1 */ + ret = cmd_recv_stream_2_1(payload, &path_name, + &channel_name); + } else if (session->minor > 1 && session->minor < 11) { + /* From 2.2 to 2.10 */ + ret = cmd_recv_stream_2_2(payload, &path_name, + &channel_name, &tracefile_size, &tracefile_count); + } else { + /* From 2.11 to ... */ + ret = cmd_recv_stream_2_11(payload, &path_name, + &channel_name, &tracefile_size, &tracefile_count, + &stream_chunk_id.value); + stream_chunk_id.is_set = true; } - switch (conn->minor) { - case 1: /* LTTng sessiond 2.1 */ - ret = cmd_recv_stream_2_1(conn, stream); - break; - case 2: /* LTTng sessiond 2.2 */ - default: - ret = cmd_recv_stream_2_2(conn, stream); - break; - } if (ret < 0) { - goto err_free_stream; + goto send_reply; } - rcu_read_lock(); - stream->stream_handle = ++last_relay_stream_id; - stream->prev_seq = -1ULL; - stream->session_id = session->id; - stream->index_fd = -1; - stream->read_index_fd = -1; - stream->ctf_stream_id = -1ULL; - lttng_ht_node_init_u64(&stream->node, stream->stream_handle); - pthread_mutex_init(&stream->lock, NULL); - - ret = utils_mkdir_recursive(stream->path_name, S_IRWXU | S_IRWXG); - if (ret < 0) { - ERR("relay creating output directory"); - goto end; + if (conform_channel_path(path_name)) { + goto send_reply; } - /* - * No need to use run_as API here because whatever we receives, the relayd - * uses its own credentials for the stream files. - */ - ret = utils_create_stream_file(stream->path_name, stream->channel_name, - stream->tracefile_size, 0, relayd_uid, relayd_gid, NULL); - if (ret < 0) { - ERR("Create output file"); - goto end; - } - stream->fd = ret; - if (stream->tracefile_size) { - DBG("Tracefile %s/%s_0 created", stream->path_name, stream->channel_name); - } else { - DBG("Tracefile %s/%s created", stream->path_name, stream->channel_name); + trace = ctf_trace_get_by_path_or_create(session, path_name); + if (!trace) { + goto send_reply; } + /* This stream here has one reference on the trace. */ - trace = ctf_trace_find_by_path(session->ctf_traces_ht, stream->path_name); - if (!trace) { - trace = ctf_trace_create(stream->path_name); - if (!trace) { - ret = -1; - goto end; - } - ctf_trace_add(session->ctf_traces_ht, trace); + pthread_mutex_lock(&last_relay_stream_id_lock); + stream_handle = ++last_relay_stream_id; + pthread_mutex_unlock(&last_relay_stream_id_lock); + + /* We pass ownership of path_name and channel_name. */ + stream = stream_create(trace, stream_handle, path_name, + channel_name, tracefile_size, tracefile_count); + path_name = NULL; + channel_name = NULL; + + /* + * Streams are the owners of their trace. Reference to trace is + * kept within stream_create(). + */ + ctf_trace_put(trace); + +send_reply: + memset(&reply, 0, sizeof(reply)); + reply.handle = htobe64(stream_handle); + if (!stream) { + reply.ret_code = htobe32(LTTNG_ERR_UNK); + } else { + reply.ret_code = htobe32(LTTNG_OK); + } + + send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, + sizeof(struct lttcomm_relayd_status_stream), 0); + if (send_ret < (ssize_t) sizeof(reply)) { + ERR("Failed to send \"add stream\" command reply (ret = %zd)", + send_ret); + ret = -1; + } + +end_no_session: + free(path_name); + free(channel_name); + return ret; +} + +/* + * relay_close_stream: close a specific stream + */ +static int relay_close_stream(const struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn, + const struct lttng_buffer_view *payload) +{ + int ret; + ssize_t send_ret; + struct relay_session *session = conn->session; + struct lttcomm_relayd_close_stream stream_info; + struct lttcomm_relayd_generic_reply reply; + struct relay_stream *stream; + + DBG("Close stream received"); + + if (!session || !conn->version_check_done) { + ERR("Trying to close a stream before version check"); + ret = -1; + goto end_no_session; } - ctf_trace_get_ref(trace); - if (!strncmp(stream->channel_name, DEFAULT_METADATA_NAME, NAME_MAX)) { - stream->metadata_flag = 1; - /* Assign quick reference to the metadata stream in the trace. */ - trace->metadata_stream = stream; + if (payload->size < sizeof(stream_info)) { + ERR("Unexpected payload size in \"relay_close_stream\": expected >= %zu bytes, got %zu bytes", + sizeof(stream_info), payload->size); + ret = -1; + goto end_no_session; + } + memcpy(&stream_info, payload->data, sizeof(stream_info)); + stream_info.stream_id = be64toh(stream_info.stream_id); + stream_info.last_net_seq_num = be64toh(stream_info.last_net_seq_num); + + stream = stream_get_by_id(stream_info.stream_id); + if (!stream) { + ret = -1; + goto end; } /* - * Add the stream in the recv list of the connection. Once the end stream - * message is received, this list is emptied and streams are set with the - * viewer ready flag. + * Set last_net_seq_num before the close flag. Required by data + * pending check. */ - queue_stream(stream, conn); + pthread_mutex_lock(&stream->lock); + stream->last_net_seq_num = stream_info.last_net_seq_num; + pthread_mutex_unlock(&stream->lock); /* - * Both in the ctf_trace object and the global stream ht since the data - * side of the relayd does not have the concept of session. + * This is one of the conditions which may trigger a stream close + * with the others being: + * 1) A close command is received for a stream + * 2) The control connection owning the stream is closed + * 3) We have received all of the stream's data _after_ a close + * request. */ - lttng_ht_add_unique_u64(relay_streams_ht, &stream->node); - cds_list_add_tail(&stream->trace_list, &trace->stream_list); - - session->stream_count++; + try_stream_close(stream); + if (stream->is_metadata) { + struct relay_viewer_stream *vstream; - DBG("Relay new stream added %s with ID %" PRIu64, stream->channel_name, - stream->stream_handle); + vstream = viewer_stream_get_by_id(stream->stream_handle); + if (vstream) { + if (vstream->metadata_sent == stream->metadata_received) { + /* + * Since all the metadata has been sent to the + * viewer and that we have a request to close + * its stream, we can safely teardown the + * corresponding metadata viewer stream. + */ + viewer_stream_put(vstream); + } + /* Put local reference. */ + viewer_stream_put(vstream); + } + } + stream_put(stream); + ret = 0; end: memset(&reply, 0, sizeof(reply)); - reply.handle = htobe64(stream->stream_handle); - /* send the session id to the client or a negative return code on error */ if (ret < 0) { reply.ret_code = htobe32(LTTNG_ERR_UNK); - /* stream was not properly added to the ht, so free it */ - free(stream); } else { reply.ret_code = htobe32(LTTNG_OK); } - send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, - sizeof(struct lttcomm_relayd_status_stream), 0); - if (send_ret < 0) { - ERR("Relay sending stream id"); - ret = send_ret; + sizeof(struct lttcomm_relayd_generic_reply), 0); + if (send_ret < (ssize_t) sizeof(reply)) { + ERR("Failed to send \"close stream\" command reply (ret = %zd)", + send_ret); + ret = -1; } - rcu_read_unlock(); end_no_session: return ret; - -err_free_stream: - free(stream->path_name); - free(stream->channel_name); - free(stream); - return ret; } /* - * relay_close_stream: close a specific stream + * relay_reset_metadata: reset a metadata stream */ static -int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_connection *conn) +int relay_reset_metadata(const struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn, + const struct lttng_buffer_view *payload) { - int ret, send_ret; + int ret; + ssize_t send_ret; struct relay_session *session = conn->session; - struct lttcomm_relayd_close_stream stream_info; + struct lttcomm_relayd_reset_metadata stream_info; struct lttcomm_relayd_generic_reply reply; struct relay_stream *stream; - DBG("Close stream received"); + DBG("Reset metadata received"); - if (!session || conn->version_check_done == 0) { - ERR("Trying to close a stream before version check"); + if (!session || !conn->version_check_done) { + ERR("Trying to reset a metadata stream before version check"); ret = -1; goto end_no_session; } - ret = conn->sock->ops->recvmsg(conn->sock, &stream_info, - sizeof(struct lttcomm_relayd_close_stream), 0); - if (ret < sizeof(struct lttcomm_relayd_close_stream)) { - if (ret == 0) { - /* Orderly shutdown. Not necessary to print an error. */ - DBG("Socket %d did an orderly shutdown", conn->sock->fd); - } else { - ERR("Relay didn't receive valid add_stream struct size : %d", ret); - } + if (payload->size < sizeof(stream_info)) { + ERR("Unexpected payload size in \"relay_reset_metadata\": expected >= %zu bytes, got %zu bytes", + sizeof(stream_info), payload->size); ret = -1; goto end_no_session; } + memcpy(&stream_info, payload->data, sizeof(stream_info)); + stream_info.stream_id = be64toh(stream_info.stream_id); + stream_info.version = be64toh(stream_info.version); - rcu_read_lock(); - stream = stream_find_by_id(relay_streams_ht, - be64toh(stream_info.stream_id)); + DBG("Update metadata to version %" PRIu64, stream_info.version); + + /* Unsupported for live sessions for now. */ + if (session->live_timer != 0) { + ret = -1; + goto end; + } + + stream = stream_get_by_id(stream_info.stream_id); if (!stream) { + ret = -1; + goto end; + } + pthread_mutex_lock(&stream->lock); + if (!stream->is_metadata) { ret = -1; goto end_unlock; } - stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num); - stream->close_flag = 1; - session->stream_count--; - - /* Check if we can close it or else the data will do it. */ - try_close_stream(session, stream); - + ret = stream_reset_file(stream); + if (ret < 0) { + ERR("Failed to reset metadata stream %" PRIu64 + ": stream_path = %s, channel = %s", + stream->stream_handle, stream->path_name, + stream->channel_name); + goto end_unlock; + } end_unlock: - rcu_read_unlock(); + pthread_mutex_unlock(&stream->lock); + stream_put(stream); +end: memset(&reply, 0, sizeof(reply)); if (ret < 0) { reply.ret_code = htobe32(LTTNG_ERR_UNK); @@ -1399,9 +1512,10 @@ end_unlock: } send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(struct lttcomm_relayd_generic_reply), 0); - if (send_ret < 0) { - ERR("Relay sending stream id"); - ret = send_ret; + if (send_ret < (ssize_t) sizeof(reply)) { + ERR("Failed to send \"reset metadata\" command reply (ret = %zd)", + send_ret); + ret = -1; } end_no_session: @@ -1411,18 +1525,16 @@ end_no_session: /* * relay_unknown_command: send -1 if received unknown command */ -static -void relay_unknown_command(struct relay_connection *conn) +static void relay_unknown_command(struct relay_connection *conn) { struct lttcomm_relayd_generic_reply reply; - int ret; + ssize_t send_ret; memset(&reply, 0, sizeof(reply)); reply.ret_code = htobe32(LTTNG_ERR_UNK); - ret = conn->sock->ops->sendmsg(conn->sock, &reply, - sizeof(struct lttcomm_relayd_generic_reply), 0); - if (ret < 0) { - ERR("Relay sending unknown command"); + send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); + if (send_ret < sizeof(reply)) { + ERR("Failed to send \"unknown command\" command reply (ret = %zd)", send_ret); } } @@ -1430,11 +1542,12 @@ void relay_unknown_command(struct relay_connection *conn) * relay_start: send an acknowledgment to the client to tell if we are * ready to receive data. We are ready if a session is established. */ -static -int relay_start(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_connection *conn) +static int relay_start(const struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn, + const struct lttng_buffer_view *payload) { - int ret = htobe32(LTTNG_OK); + int ret = 0; + ssize_t send_ret; struct lttcomm_relayd_generic_reply reply; struct relay_session *session = conn->session; @@ -1444,60 +1557,31 @@ int relay_start(struct lttcomm_relayd_hdr *recv_hdr, } memset(&reply, 0, sizeof(reply)); - reply.ret_code = ret; - ret = conn->sock->ops->sendmsg(conn->sock, &reply, - sizeof(struct lttcomm_relayd_generic_reply), 0); - if (ret < 0) { - ERR("Relay sending start ack"); - } - - return ret; -} - -/* - * Append padding to the file pointed by the file descriptor fd. - */ -static int write_padding_to_file(int fd, uint32_t size) -{ - ssize_t ret = 0; - char *zeros; - - if (size == 0) { - goto end; - } - - zeros = zmalloc(size); - if (zeros == NULL) { - PERROR("zmalloc zeros for padding"); + reply.ret_code = htobe32(LTTNG_OK); + send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, + sizeof(reply), 0); + if (send_ret < (ssize_t) sizeof(reply)) { + ERR("Failed to send \"relay_start\" command reply (ret = %zd)", + send_ret); ret = -1; - goto end; - } - - ret = lttng_write(fd, zeros, size); - if (ret < size) { - PERROR("write padding to file"); } - free(zeros); - -end: return ret; } /* - * relay_recv_metadata: receive the metada for the session. + * relay_recv_metadata: receive the metadata for the session. */ -static -int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_connection *conn) +static int relay_recv_metadata(const struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn, + const struct lttng_buffer_view *payload) { - int ret = htobe32(LTTNG_OK); - ssize_t size_ret; + int ret = 0; struct relay_session *session = conn->session; - struct lttcomm_relayd_metadata_payload *metadata_struct; + struct lttcomm_relayd_metadata_payload metadata_payload_header; struct relay_stream *metadata_stream; - uint64_t data_size, payload_size; - struct ctf_trace *ctf_trace; + uint64_t metadata_payload_size; + struct lttng_buffer_view packet_view; if (!session) { ERR("Metadata sent before version check"); @@ -1505,75 +1589,45 @@ int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr, goto end; } - data_size = payload_size = be64toh(recv_hdr->data_size); - if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) { + if (recv_hdr->data_size < sizeof(struct lttcomm_relayd_metadata_payload)) { ERR("Incorrect data size"); ret = -1; goto end; } - payload_size -= sizeof(struct lttcomm_relayd_metadata_payload); + metadata_payload_size = recv_hdr->data_size - + sizeof(struct lttcomm_relayd_metadata_payload); - if (data_buffer_size < data_size) { - /* In case the realloc fails, we can free the memory */ - char *tmp_data_ptr; + memcpy(&metadata_payload_header, payload->data, + sizeof(metadata_payload_header)); + metadata_payload_header.stream_id = be64toh( + metadata_payload_header.stream_id); + metadata_payload_header.padding_size = be32toh( + metadata_payload_header.padding_size); - tmp_data_ptr = realloc(data_buffer, data_size); - if (!tmp_data_ptr) { - ERR("Allocating data buffer"); - free(data_buffer); - ret = -1; - goto end; - } - data_buffer = tmp_data_ptr; - data_buffer_size = data_size; - } - memset(data_buffer, 0, data_size); - DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size); - ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, data_size, 0); - if (ret < 0 || ret != data_size) { - if (ret == 0) { - /* Orderly shutdown. Not necessary to print an error. */ - DBG("Socket %d did an orderly shutdown", conn->sock->fd); - } else { - ERR("Relay didn't receive the whole metadata"); - } + metadata_stream = stream_get_by_id(metadata_payload_header.stream_id); + if (!metadata_stream) { ret = -1; goto end; } - metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer; - rcu_read_lock(); - metadata_stream = stream_find_by_id(relay_streams_ht, - be64toh(metadata_struct->stream_id)); - if (!metadata_stream) { + packet_view = lttng_buffer_view_from_view(payload, + sizeof(metadata_payload_header), metadata_payload_size); + if (!packet_view.data) { + ERR("Invalid metadata packet length announced by header"); ret = -1; - goto end_unlock; + goto end_put; } - size_ret = lttng_write(metadata_stream->fd, metadata_struct->payload, - payload_size); - if (size_ret < payload_size) { - ERR("Relay error writing metadata on file"); + pthread_mutex_lock(&metadata_stream->lock); + ret = stream_write(metadata_stream, &packet_view, + metadata_payload_header.padding_size); + pthread_mutex_unlock(&metadata_stream->lock); + if (ret){ ret = -1; - goto end_unlock; + goto end_put; } - - ret = write_padding_to_file(metadata_stream->fd, - be32toh(metadata_struct->padding_size)); - if (ret < 0) { - goto end_unlock; - } - - ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht, - metadata_stream->path_name); - assert(ctf_trace); - ctf_trace->metadata_received += - payload_size + be32toh(metadata_struct->padding_size); - - DBG2("Relay metadata written"); - -end_unlock: - rcu_read_unlock(); +end_put: + stream_put(metadata_stream); end: return ret; } @@ -1581,57 +1635,64 @@ end: /* * relay_send_version: send relayd version number */ -static -int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_connection *conn) +static int relay_send_version(const struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn, + const struct lttng_buffer_view *payload) { int ret; + ssize_t send_ret; struct lttcomm_relayd_version reply, msg; + bool compatible = true; - assert(conn); - - conn->version_check_done = 1; + conn->version_check_done = true; /* Get version from the other side. */ - ret = conn->sock->ops->recvmsg(conn->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", conn->sock->fd); - } else { - ERR("Relay failed to receive the version values."); - } + if (payload->size < sizeof(msg)) { + ERR("Unexpected payload size in \"relay_send_version\": expected >= %zu bytes, got %zu bytes", + sizeof(msg), payload->size); ret = -1; goto end; } + memcpy(&msg, payload->data, sizeof(msg)); + msg.major = be32toh(msg.major); + msg.minor = be32toh(msg.minor); + 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)) { + if (reply.major != msg.major) { DBG("Incompatible major versions (%u vs %u), deleting session", - reply.major, be32toh(msg.major)); - destroy_session(conn->session, conn->sessions_ht); - ret = 0; - goto end; + reply.major, msg.major); + compatible = false; } conn->major = reply.major; /* We adapt to the lowest compatible version */ - if (reply.minor <= be32toh(msg.minor)) { + if (reply.minor <= msg.minor) { conn->minor = reply.minor; } else { - conn->minor = be32toh(msg.minor); + conn->minor = msg.minor; } reply.major = htobe32(reply.major); reply.minor = htobe32(reply.minor); - ret = conn->sock->ops->sendmsg(conn->sock, &reply, - sizeof(struct lttcomm_relayd_version), 0); - if (ret < 0) { - ERR("Relay sending version"); + send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, + sizeof(reply), 0); + if (send_ret < (ssize_t) sizeof(reply)) { + ERR("Failed to send \"send version\" command reply (ret = %zd)", + send_ret); + ret = -1; + goto end; + } else { + ret = 0; + } + + if (!compatible) { + ret = -1; + goto end; } DBG("Version check done using protocol %u.%u", conn->major, @@ -1644,54 +1705,61 @@ end: /* * Check for data pending for a given stream id from the session daemon. */ -static -int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_connection *conn) +static int relay_data_pending(const struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn, + const struct lttng_buffer_view *payload) { struct relay_session *session = conn->session; struct lttcomm_relayd_data_pending msg; struct lttcomm_relayd_generic_reply reply; struct relay_stream *stream; + ssize_t send_ret; int ret; - uint64_t last_net_seq_num, stream_id; + uint64_t stream_seq; DBG("Data pending command received"); - if (!session || conn->version_check_done == 0) { + if (!session || !conn->version_check_done) { ERR("Trying to check for data before version check"); ret = -1; goto end_no_session; } - ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0); - if (ret < sizeof(msg)) { - if (ret == 0) { - /* Orderly shutdown. Not necessary to print an error. */ - DBG("Socket %d did an orderly shutdown", conn->sock->fd); - } else { - ERR("Relay didn't receive valid data_pending struct size : %d", - ret); - } + if (payload->size < sizeof(msg)) { + ERR("Unexpected payload size in \"relay_data_pending\": expected >= %zu bytes, got %zu bytes", + sizeof(msg), payload->size); ret = -1; goto end_no_session; } + memcpy(&msg, payload->data, sizeof(msg)); + msg.stream_id = be64toh(msg.stream_id); + msg.last_net_seq_num = be64toh(msg.last_net_seq_num); - stream_id = be64toh(msg.stream_id); - last_net_seq_num = be64toh(msg.last_net_seq_num); - - rcu_read_lock(); - stream = stream_find_by_id(relay_streams_ht, stream_id); + stream = stream_get_by_id(msg.stream_id); if (stream == NULL) { ret = -1; - goto end_unlock; + goto end; } - DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64 - " and last_seq %" PRIu64, stream_id, stream->prev_seq, - last_net_seq_num); + pthread_mutex_lock(&stream->lock); + + if (session_streams_have_index(session)) { + /* + * Ensure that both the index and stream data have been + * flushed up to the requested point. + */ + stream_seq = min(stream->prev_data_seq, stream->prev_index_seq); + } else { + stream_seq = stream->prev_data_seq; + } + DBG("Data pending for stream id %" PRIu64 ": prev_data_seq %" PRIu64 + ", prev_index_seq %" PRIu64 + ", and last_seq %" PRIu64, msg.stream_id, + stream->prev_data_seq, stream->prev_index_seq, + msg.last_net_seq_num); /* Avoid wrapping issue */ - if (((int64_t) (stream->prev_seq - last_net_seq_num)) >= 0) { + if (((int64_t) (stream_seq - msg.last_net_seq_num)) >= 0) { /* Data has in fact been written and is NOT pending */ ret = 0; } else { @@ -1699,17 +1767,19 @@ int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr, ret = 1; } - /* Pending check is now done. */ - stream->data_pending_check_done = 1; + stream->data_pending_check_done = true; + pthread_mutex_unlock(&stream->lock); -end_unlock: - rcu_read_unlock(); + stream_put(stream); +end: memset(&reply, 0, sizeof(reply)); reply.ret_code = htobe32(ret); - ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); - if (ret < 0) { - ERR("Relay data pending ret code failed"); + send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); + if (send_ret < (ssize_t) sizeof(reply)) { + ERR("Failed to send \"data pending\" command reply (ret = %zd)", + send_ret); + ret = -1; } end_no_session: @@ -1719,61 +1789,58 @@ end_no_session: /* * Wait for the control socket to reach a quiescent state. * - * Note that for now, when receiving this command from the session daemon, this - * means that every subsequent commands or data received on the control socket - * has been handled. So, this is why we simply return OK here. + * Note that for now, when receiving this command from the session + * daemon, this means that every subsequent commands or data received on + * the control socket has been handled. So, this is why we simply return + * OK here. */ -static -int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_connection *conn) +static int relay_quiescent_control(const struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn, + const struct lttng_buffer_view *payload) { int ret; - uint64_t stream_id; + ssize_t send_ret; struct relay_stream *stream; - struct lttng_ht_iter iter; struct lttcomm_relayd_quiescent_control msg; struct lttcomm_relayd_generic_reply reply; DBG("Checking quiescent state on control socket"); - if (!conn->session || conn->version_check_done == 0) { + if (!conn->session || !conn->version_check_done) { ERR("Trying to check for data before version check"); ret = -1; goto end_no_session; } - ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0); - if (ret < sizeof(msg)) { - if (ret == 0) { - /* Orderly shutdown. Not necessary to print an error. */ - DBG("Socket %d did an orderly shutdown", conn->sock->fd); - } else { - ERR("Relay didn't receive valid begin data_pending struct size: %d", - ret); - } + if (payload->size < sizeof(msg)) { + ERR("Unexpected payload size in \"relay_quiescent_control\": expected >= %zu bytes, got %zu bytes", + sizeof(msg), payload->size); ret = -1; goto end_no_session; } + memcpy(&msg, payload->data, sizeof(msg)); + msg.stream_id = be64toh(msg.stream_id); - stream_id = be64toh(msg.stream_id); - - rcu_read_lock(); - cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream, - node.node) { - if (stream->stream_handle == stream_id) { - stream->data_pending_check_done = 1; - DBG("Relay quiescent control pending flag set to %" PRIu64, - stream_id); - break; - } + stream = stream_get_by_id(msg.stream_id); + if (!stream) { + goto reply; } - rcu_read_unlock(); + pthread_mutex_lock(&stream->lock); + stream->data_pending_check_done = true; + pthread_mutex_unlock(&stream->lock); + DBG("Relay quiescent control pending flag set to %" PRIu64, msg.stream_id); + stream_put(stream); +reply: memset(&reply, 0, sizeof(reply)); reply.ret_code = htobe32(LTTNG_OK); - ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); - if (ret < 0) { - ERR("Relay data quiescent control ret code failed"); + send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); + if (send_ret < (ssize_t) sizeof(reply)) { + ERR("Failed to send \"quiescent control\" command reply (ret = %zd)", + send_ret); + ret = -1; + } else { + ret = 0; } end_no_session: @@ -1781,62 +1848,63 @@ end_no_session: } /* - * Initialize a data pending command. This means that a client is about to ask - * for data pending for each stream he/she holds. Simply iterate over all - * streams of a session and set the data_pending_check_done flag. + * Initialize a data pending command. This means that a consumer is about + * to ask for data pending for each stream it holds. Simply iterate over + * all streams of a session and set the data_pending_check_done flag. * * This command returns to the client a LTTNG_OK code. */ -static -int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_connection *conn) +static int relay_begin_data_pending(const struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn, + const struct lttng_buffer_view *payload) { int ret; + ssize_t send_ret; struct lttng_ht_iter iter; struct lttcomm_relayd_begin_data_pending msg; struct lttcomm_relayd_generic_reply reply; struct relay_stream *stream; - uint64_t session_id; assert(recv_hdr); assert(conn); DBG("Init streams for data pending"); - if (!conn->session || conn->version_check_done == 0) { + if (!conn->session || !conn->version_check_done) { ERR("Trying to check for data before version check"); ret = -1; goto end_no_session; } - ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0); - if (ret < sizeof(msg)) { - if (ret == 0) { - /* Orderly shutdown. Not necessary to print an error. */ - DBG("Socket %d did an orderly shutdown", conn->sock->fd); - } else { - ERR("Relay didn't receive valid begin data_pending struct size: %d", - ret); - } + if (payload->size < sizeof(msg)) { + ERR("Unexpected payload size in \"relay_begin_data_pending\": expected >= %zu bytes, got %zu bytes", + sizeof(msg), payload->size); ret = -1; goto end_no_session; } - - session_id = be64toh(msg.session_id); + memcpy(&msg, payload->data, sizeof(msg)); + msg.session_id = be64toh(msg.session_id); /* - * Iterate over all streams to set the begin data pending flag. For now, the - * streams are indexed by stream handle so we have to iterate over all - * streams to find the one associated with the right session_id. + * Iterate over all streams to set the begin data pending flag. + * For now, the streams are indexed by stream handle so we have + * to iterate over all streams to find the one associated with + * the right session_id. */ rcu_read_lock(); cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream, node.node) { - if (stream->session_id == session_id) { - stream->data_pending_check_done = 0; + if (!stream_get(stream)) { + continue; + } + if (stream->trace->session->id == msg.session_id) { + pthread_mutex_lock(&stream->lock); + stream->data_pending_check_done = false; + pthread_mutex_unlock(&stream->lock); DBG("Set begin data pending flag to stream %" PRIu64, stream->stream_handle); } + stream_put(stream); } rcu_read_unlock(); @@ -1844,9 +1912,13 @@ int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr, /* All good, send back reply. */ reply.ret_code = htobe32(LTTNG_OK); - ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); - if (ret < 0) { - ERR("Relay begin data pending send reply failed"); + send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); + if (send_ret < (ssize_t) sizeof(reply)) { + ERR("Failed to send \"begin data pending\" command reply (ret = %zd)", + send_ret); + ret = -1; + } else { + ret = 0; } end_no_session: @@ -1854,63 +1926,81 @@ end_no_session: } /* - * End data pending command. This will check, for a given session id, if each - * stream associated with it has its data_pending_check_done flag set. If not, - * this means that the client lost track of the stream but the data is still - * being streamed on our side. In this case, we inform the client that data is - * inflight. + * End data pending command. This will check, for a given session id, if + * each stream associated with it has its data_pending_check_done flag + * set. If not, this means that the client lost track of the stream but + * the data is still being streamed on our side. In this case, we inform + * the client that data is in flight. * * Return to the client if there is data in flight or not with a ret_code. */ -static -int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_connection *conn) +static int relay_end_data_pending(const struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn, + const struct lttng_buffer_view *payload) { int ret; + ssize_t send_ret; struct lttng_ht_iter iter; struct lttcomm_relayd_end_data_pending msg; struct lttcomm_relayd_generic_reply reply; struct relay_stream *stream; - uint64_t session_id; uint32_t is_data_inflight = 0; - assert(recv_hdr); - assert(conn); - DBG("End data pending command"); - if (!conn->session || conn->version_check_done == 0) { + if (!conn->session || !conn->version_check_done) { ERR("Trying to check for data before version check"); ret = -1; goto end_no_session; } - ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0); - if (ret < sizeof(msg)) { - if (ret == 0) { - /* Orderly shutdown. Not necessary to print an error. */ - DBG("Socket %d did an orderly shutdown", conn->sock->fd); - } else { - ERR("Relay didn't receive valid end data_pending struct size: %d", - ret); - } + if (payload->size < sizeof(msg)) { + ERR("Unexpected payload size in \"relay_end_data_pending\": expected >= %zu bytes, got %zu bytes", + sizeof(msg), payload->size); ret = -1; goto end_no_session; } + memcpy(&msg, payload->data, sizeof(msg)); + msg.session_id = be64toh(msg.session_id); - session_id = be64toh(msg.session_id); - - /* Iterate over all streams to see if the begin data pending flag is set. */ + /* + * Iterate over all streams to see if the begin data pending + * flag is set. + */ rcu_read_lock(); cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream, node.node) { - if (stream->session_id == session_id && - !stream->data_pending_check_done && !stream->terminated_flag) { - is_data_inflight = 1; - DBG("Data is still in flight for stream %" PRIu64, - stream->stream_handle); - break; + if (!stream_get(stream)) { + continue; + } + if (stream->trace->session->id != msg.session_id) { + stream_put(stream); + continue; } + pthread_mutex_lock(&stream->lock); + if (!stream->data_pending_check_done) { + uint64_t stream_seq; + + if (session_streams_have_index(conn->session)) { + /* + * Ensure that both the index and stream data have been + * flushed up to the requested point. + */ + stream_seq = min(stream->prev_data_seq, stream->prev_index_seq); + } else { + stream_seq = stream->prev_data_seq; + } + if (!stream->closed || !(((int64_t) (stream_seq - stream->last_net_seq_num)) >= 0)) { + is_data_inflight = 1; + DBG("Data is still in flight for stream %" PRIu64, + stream->stream_handle); + pthread_mutex_unlock(&stream->lock); + stream_put(stream); + break; + } + } + pthread_mutex_unlock(&stream->lock); + stream_put(stream); } rcu_read_unlock(); @@ -1918,9 +2008,13 @@ int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr, /* All good, send back reply. */ reply.ret_code = htobe32(is_data_inflight); - ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); - if (ret < 0) { - ERR("Relay end data pending send reply failed"); + send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); + if (send_ret < (ssize_t) sizeof(reply)) { + ERR("Failed to send \"end data pending\" command reply (ret = %zd)", + send_ret); + ret = -1; + } else { + ret = 0; } end_no_session: @@ -1932,114 +2026,70 @@ end_no_session: * * Return 0 on success else a negative value. */ -static -int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_connection *conn) +static int relay_recv_index(const struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn, + const struct lttng_buffer_view *payload) { - int ret, send_ret, index_created = 0; + int ret; + ssize_t send_ret; struct relay_session *session = conn->session; struct lttcomm_relayd_index index_info; - struct relay_index *index, *wr_index = NULL; struct lttcomm_relayd_generic_reply reply; struct relay_stream *stream; - uint64_t net_seq_num; + size_t msg_len; assert(conn); DBG("Relay receiving index"); - if (!session || conn->version_check_done == 0) { + if (!session || !conn->version_check_done) { ERR("Trying to close a stream before version check"); ret = -1; goto end_no_session; } - ret = conn->sock->ops->recvmsg(conn->sock, &index_info, - sizeof(index_info), 0); - if (ret < sizeof(index_info)) { - if (ret == 0) { - /* Orderly shutdown. Not necessary to print an error. */ - DBG("Socket %d did an orderly shutdown", conn->sock->fd); - } else { - ERR("Relay didn't receive valid index struct size : %d", ret); - } + msg_len = lttcomm_relayd_index_len( + lttng_to_index_major(conn->major, conn->minor), + lttng_to_index_minor(conn->major, conn->minor)); + if (payload->size < msg_len) { + ERR("Unexpected payload size in \"relay_recv_index\": expected >= %zu bytes, got %zu bytes", + msg_len, payload->size); ret = -1; goto end_no_session; } + memcpy(&index_info, payload->data, msg_len); + index_info.relay_stream_id = be64toh(index_info.relay_stream_id); + index_info.net_seq_num = be64toh(index_info.net_seq_num); + index_info.packet_size = be64toh(index_info.packet_size); + index_info.content_size = be64toh(index_info.content_size); + index_info.timestamp_begin = be64toh(index_info.timestamp_begin); + index_info.timestamp_end = be64toh(index_info.timestamp_end); + index_info.events_discarded = be64toh(index_info.events_discarded); + index_info.stream_id = be64toh(index_info.stream_id); - net_seq_num = be64toh(index_info.net_seq_num); + if (conn->minor >= 8) { + index_info.stream_instance_id = + be64toh(index_info.stream_instance_id); + index_info.packet_seq_num = be64toh(index_info.packet_seq_num); + } - rcu_read_lock(); - stream = stream_find_by_id(relay_streams_ht, - be64toh(index_info.relay_stream_id)); + stream = stream_get_by_id(index_info.relay_stream_id); if (!stream) { + ERR("stream_get_by_id not found"); ret = -1; - goto end_rcu_unlock; - } - - /* Live beacon handling */ - if (index_info.packet_size == 0) { - DBG("Received live beacon for stream %" PRIu64, stream->stream_handle); - - /* - * Only flag a stream inactive when it has already received data - * and no indexes are in flight. - */ - if (stream->total_index_received > 0 && stream->indexes_in_flight == 0) { - stream->beacon_ts_end = be64toh(index_info.timestamp_end); - } - ret = 0; - goto end_rcu_unlock; - } else { - stream->beacon_ts_end = -1ULL; - } - - index = relay_index_find(stream->stream_handle, net_seq_num); - if (!index) { - /* A successful creation will add the object to the HT. */ - index = relay_index_create(stream->stream_handle, net_seq_num); - if (!index) { - goto end_rcu_unlock; - } - index_created = 1; - stream->indexes_in_flight++; - } - - copy_index_control_data(index, &index_info); - if (stream->ctf_stream_id == -1ULL) { - stream->ctf_stream_id = be64toh(index_info.stream_id); - } - - if (index_created) { - /* - * Try to add the relay index object to the hash table. If an object - * already exist, destroy back the index created, set the data in this - * object and write it on disk. - */ - relay_index_add(index, &wr_index); - if (wr_index) { - copy_index_control_data(wr_index, &index_info); - free(index); - } - } else { - /* The index already exists so write it on disk. */ - wr_index = index; + goto end; } - /* Do we have a writable ready index to write on disk. */ - if (wr_index) { - ret = relay_index_write(wr_index->fd, wr_index); - if (ret < 0) { - goto end_rcu_unlock; - } - stream->total_index_received++; - stream->indexes_in_flight--; - assert(stream->indexes_in_flight >= 0); + pthread_mutex_lock(&stream->lock); + ret = stream_add_index(stream, &index_info); + pthread_mutex_unlock(&stream->lock); + if (ret) { + goto end_stream_put; } -end_rcu_unlock: - rcu_read_unlock(); - +end_stream_put: + stream_put(stream); +end: memset(&reply, 0, sizeof(reply)); if (ret < 0) { reply.ret_code = htobe32(LTTNG_ERR_UNK); @@ -2047,9 +2097,9 @@ end_rcu_unlock: reply.ret_code = htobe32(LTTNG_OK); } send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); - if (send_ret < 0) { - ERR("Relay sending close index id reply"); - ret = send_ret; + if (send_ret < (ssize_t) sizeof(reply)) { + ERR("Failed to send \"recv index\" command reply (ret = %zd)", send_ret); + ret = -1; } end_no_session: @@ -2061,42 +2111,37 @@ end_no_session: * * Return 0 on success else a negative value. */ -static -int relay_streams_sent(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_connection *conn) +static int relay_streams_sent(const struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn, + const struct lttng_buffer_view *payload) { - int ret, send_ret; + int ret; + ssize_t send_ret; struct lttcomm_relayd_generic_reply reply; assert(conn); DBG("Relay receiving streams_sent"); - if (!conn->session || conn->version_check_done == 0) { + if (!conn->session || !conn->version_check_done) { ERR("Trying to close a stream before version check"); ret = -1; goto end_no_session; } /* - * Flag every pending stream in the connection recv list that they are - * ready to be used by the viewer. + * Publish every pending stream in the connection recv list which are + * now ready to be used by the viewer. */ - set_viewer_ready_flag(conn); - - /* - * Inform the viewer that there are new streams in the session. - */ - if (conn->session->viewer_refcount) { - uatomic_set(&conn->session->new_streams, 1); - } + publish_connection_local_streams(conn); memset(&reply, 0, sizeof(reply)); reply.ret_code = htobe32(LTTNG_OK); send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); - if (send_ret < 0) { - ERR("Relay sending sent_stream reply"); - ret = send_ret; + if (send_ret < (ssize_t) sizeof(reply)) { + ERR("Failed to send \"streams sent\" command reply (ret = %zd)", + send_ret); + ret = -1; } else { /* Success. */ ret = 0; @@ -2106,55 +2151,659 @@ end_no_session: return ret; } -/* - * Process the commands received on the control socket - */ -static -int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_connection *conn) +/* + * relay_rotate_session_stream: rotate a stream to a new tracefile for the + * session rotation feature (not the tracefile rotation feature). + */ +static int relay_rotate_session_streams( + const struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn, + const struct lttng_buffer_view *payload) +{ + int ret = 0; + uint32_t i; + ssize_t send_ret; + enum lttng_error_code reply_code = LTTNG_ERR_UNK; + struct relay_session *session = conn->session; + struct lttcomm_relayd_rotate_streams rotate_streams; + struct lttcomm_relayd_generic_reply reply = {}; + struct relay_stream *stream = NULL; + const size_t header_len = sizeof(struct lttcomm_relayd_rotate_streams); + struct lttng_trace_chunk *next_trace_chunk = NULL; + struct lttng_buffer_view stream_positions; + char chunk_id_buf[MAX_INT_DEC_LEN(uint64_t)]; + const char *chunk_id_str = "none"; + + if (!session || !conn->version_check_done) { + ERR("Trying to rotate a stream before version check"); + ret = -1; + goto end_no_reply; + } + + if (session->major == 2 && session->minor < 11) { + ERR("Unsupported feature before 2.11"); + ret = -1; + goto end_no_reply; + } + + if (payload->size < header_len) { + ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected >= %zu bytes, got %zu bytes", + header_len, payload->size); + ret = -1; + goto end_no_reply; + } + + memcpy(&rotate_streams, payload->data, header_len); + + /* Convert header to host endianness. */ + rotate_streams = (typeof(rotate_streams)) { + .stream_count = be32toh(rotate_streams.stream_count), + .new_chunk_id = (typeof(rotate_streams.new_chunk_id)) { + .is_set = !!rotate_streams.new_chunk_id.is_set, + .value = be64toh(rotate_streams.new_chunk_id.value), + } + }; + + if (rotate_streams.new_chunk_id.is_set) { + /* + * Retrieve the trace chunk the stream must transition to. As + * per the protocol, this chunk should have been created + * before this command is received. + */ + next_trace_chunk = sessiond_trace_chunk_registry_get_chunk( + sessiond_trace_chunk_registry, + session->sessiond_uuid, session->id, + rotate_streams.new_chunk_id.value); + if (!next_trace_chunk) { + char uuid_str[UUID_STR_LEN]; + + lttng_uuid_to_str(session->sessiond_uuid, uuid_str); + ERR("Unknown next trace chunk in ROTATE_STREAMS command: sessiond_uuid = {%s}, session_id = %" PRIu64 + ", trace_chunk_id = %" PRIu64, + uuid_str, session->id, + rotate_streams.new_chunk_id.value); + reply_code = LTTNG_ERR_INVALID_PROTOCOL; + ret = -1; + goto end; + } + + ret = snprintf(chunk_id_buf, sizeof(chunk_id_buf), "%" PRIu64, + rotate_streams.new_chunk_id.value); + if (ret < 0 || ret >= sizeof(chunk_id_buf)) { + chunk_id_str = "formatting error"; + } else { + chunk_id_str = chunk_id_buf; + } + session->has_rotated = true; + } + + DBG("Rotate %" PRIu32 " streams of session \"%s\" to chunk \"%s\"", + rotate_streams.stream_count, session->session_name, + chunk_id_str); + + stream_positions = lttng_buffer_view_from_view(payload, + sizeof(rotate_streams), -1); + if (!stream_positions.data || + stream_positions.size < + (rotate_streams.stream_count * + sizeof(struct lttcomm_relayd_stream_rotation_position))) { + reply_code = LTTNG_ERR_INVALID_PROTOCOL; + ret = -1; + goto end; + } + + for (i = 0; i < rotate_streams.stream_count; i++) { + struct lttcomm_relayd_stream_rotation_position *position_comm = + &((typeof(position_comm)) stream_positions.data)[i]; + const struct lttcomm_relayd_stream_rotation_position pos = { + .stream_id = be64toh(position_comm->stream_id), + .rotate_at_seq_num = be64toh( + position_comm->rotate_at_seq_num), + }; + + stream = stream_get_by_id(pos.stream_id); + if (!stream) { + reply_code = LTTNG_ERR_INVALID; + ret = -1; + goto end; + } + + pthread_mutex_lock(&stream->lock); + ret = stream_set_pending_rotation(stream, next_trace_chunk, + pos.rotate_at_seq_num); + pthread_mutex_unlock(&stream->lock); + if (ret) { + reply_code = LTTNG_ERR_FILE_CREATION_ERROR; + goto end; + } + + stream_put(stream); + stream = NULL; + } + + reply_code = LTTNG_OK; +end: + if (stream) { + stream_put(stream); + } + + reply.ret_code = htobe32((uint32_t) reply_code); + send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, + sizeof(struct lttcomm_relayd_generic_reply), 0); + if (send_ret < (ssize_t) sizeof(reply)) { + ERR("Failed to send \"rotate session stream\" command reply (ret = %zd)", + send_ret); + ret = -1; + } + + ret = 0; +end_no_reply: + lttng_trace_chunk_put(next_trace_chunk); + return ret; +} + + + +/* + * relay_create_trace_chunk: create a new trace chunk + */ +static int relay_create_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn, + const struct lttng_buffer_view *payload) +{ + int ret = 0; + ssize_t send_ret; + struct relay_session *session = conn->session; + struct lttcomm_relayd_create_trace_chunk *msg; + struct lttcomm_relayd_generic_reply reply = {}; + struct lttng_buffer_view header_view; + struct lttng_buffer_view chunk_name_view; + struct lttng_trace_chunk *chunk = NULL, *published_chunk = NULL; + enum lttng_error_code reply_code = LTTNG_OK; + enum lttng_trace_chunk_status chunk_status; + struct lttng_directory_handle session_output; + + if (!session || !conn->version_check_done) { + ERR("Trying to create a trace chunk before version check"); + ret = -1; + goto end_no_reply; + } + + if (session->major == 2 && session->minor < 11) { + ERR("Chunk creation command is unsupported before 2.11"); + ret = -1; + goto end_no_reply; + } + + header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg)); + if (!header_view.data) { + ERR("Failed to receive payload of chunk creation command"); + ret = -1; + goto end_no_reply; + } + + /* Convert to host endianness. */ + msg = (typeof(msg)) header_view.data; + msg->chunk_id = be64toh(msg->chunk_id); + msg->creation_timestamp = be64toh(msg->creation_timestamp); + msg->override_name_length = be32toh(msg->override_name_length); + + chunk = lttng_trace_chunk_create( + msg->chunk_id, msg->creation_timestamp); + if (!chunk) { + ERR("Failed to create trace chunk in trace chunk creation command"); + ret = -1; + reply_code = LTTNG_ERR_NOMEM; + goto end; + } + + if (msg->override_name_length) { + const char *name; + + chunk_name_view = lttng_buffer_view_from_view(payload, + sizeof(*msg), + msg->override_name_length); + name = chunk_name_view.data; + if (!name || name[msg->override_name_length - 1]) { + ERR("Failed to receive payload of chunk creation command"); + ret = -1; + reply_code = LTTNG_ERR_INVALID; + goto end; + } + + chunk_status = lttng_trace_chunk_override_name( + chunk, chunk_name_view.data); + switch (chunk_status) { + case LTTNG_TRACE_CHUNK_STATUS_OK: + break; + case LTTNG_TRACE_CHUNK_STATUS_INVALID_ARGUMENT: + ERR("Failed to set the name of new trace chunk in trace chunk creation command (invalid name)"); + reply_code = LTTNG_ERR_INVALID; + ret = -1; + goto end; + default: + ERR("Failed to set the name of new trace chunk in trace chunk creation command (unknown error)"); + reply_code = LTTNG_ERR_UNK; + ret = -1; + goto end; + } + } + + chunk_status = lttng_trace_chunk_set_credentials_current_user(chunk); + if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { + reply_code = LTTNG_ERR_UNK; + ret = -1; + goto end; + } + + ret = session_init_output_directory_handle( + conn->session, &session_output); + if (ret) { + reply_code = LTTNG_ERR_CREATE_DIR_FAIL; + goto end; + } + chunk_status = lttng_trace_chunk_set_as_owner(chunk, &session_output); + lttng_directory_handle_fini(&session_output); + if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { + reply_code = LTTNG_ERR_UNK; + ret = -1; + goto end; + } + + published_chunk = sessiond_trace_chunk_registry_publish_chunk( + sessiond_trace_chunk_registry, + conn->session->sessiond_uuid, + conn->session->id, + chunk); + if (!published_chunk) { + char uuid_str[UUID_STR_LEN]; + + lttng_uuid_to_str(conn->session->sessiond_uuid, uuid_str); + ERR("Failed to publish chunk: sessiond_uuid = %s, session_id = %" PRIu64 ", chunk_id = %" PRIu64, + uuid_str, + conn->session->id, + msg->chunk_id); + ret = -1; + reply_code = LTTNG_ERR_NOMEM; + goto end; + } + + pthread_mutex_lock(&conn->session->lock); + if (conn->session->pending_closure_trace_chunk) { + /* + * Invalid; this means a second create_trace_chunk command was + * received before a close_trace_chunk. + */ + ERR("Invalid trace chunk close command received; a trace chunk is already waiting for a trace chunk close command"); + reply_code = LTTNG_ERR_INVALID_PROTOCOL; + ret = -1; + goto end_unlock_session; + } + conn->session->pending_closure_trace_chunk = + conn->session->current_trace_chunk; + conn->session->current_trace_chunk = published_chunk; + published_chunk = NULL; +end_unlock_session: + pthread_mutex_unlock(&conn->session->lock); +end: + reply.ret_code = htobe32((uint32_t) reply_code); + send_ret = conn->sock->ops->sendmsg(conn->sock, + &reply, + sizeof(struct lttcomm_relayd_generic_reply), + 0); + if (send_ret < (ssize_t) sizeof(reply)) { + ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)", + send_ret); + ret = -1; + } +end_no_reply: + lttng_trace_chunk_put(chunk); + lttng_trace_chunk_put(published_chunk); + return ret; +} + +/* + * relay_close_trace_chunk: close a trace chunk + */ +static int relay_close_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn, + const struct lttng_buffer_view *payload) +{ + int ret = 0, buf_ret; + ssize_t send_ret; + struct relay_session *session = conn->session; + struct lttcomm_relayd_close_trace_chunk *msg; + struct lttcomm_relayd_close_trace_chunk_reply reply = {}; + struct lttng_buffer_view header_view; + struct lttng_trace_chunk *chunk = NULL; + enum lttng_error_code reply_code = LTTNG_OK; + enum lttng_trace_chunk_status chunk_status; + uint64_t chunk_id; + LTTNG_OPTIONAL(enum lttng_trace_chunk_command_type) close_command = {}; + time_t close_timestamp; + char closed_trace_chunk_path[LTTNG_PATH_MAX]; + size_t path_length = 0; + const char *chunk_name = NULL; + struct lttng_dynamic_buffer reply_payload; + + lttng_dynamic_buffer_init(&reply_payload); + + if (!session || !conn->version_check_done) { + ERR("Trying to close a trace chunk before version check"); + ret = -1; + goto end_no_reply; + } + + if (session->major == 2 && session->minor < 11) { + ERR("Chunk close command is unsupported before 2.11"); + ret = -1; + goto end_no_reply; + } + + header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg)); + if (!header_view.data) { + ERR("Failed to receive payload of chunk close command"); + ret = -1; + goto end_no_reply; + } + + /* Convert to host endianness. */ + msg = (typeof(msg)) header_view.data; + chunk_id = be64toh(msg->chunk_id); + close_timestamp = (time_t) be64toh(msg->close_timestamp); + close_command = (typeof(close_command)){ + .value = be32toh(msg->close_command.value), + .is_set = msg->close_command.is_set, + }; + + chunk = sessiond_trace_chunk_registry_get_chunk( + sessiond_trace_chunk_registry, + conn->session->sessiond_uuid, + conn->session->id, + chunk_id); + if (!chunk) { + char uuid_str[UUID_STR_LEN]; + + lttng_uuid_to_str(conn->session->sessiond_uuid, uuid_str); + ERR("Failed to find chunk to close: sessiond_uuid = %s, session_id = %" PRIu64 ", chunk_id = %" PRIu64, + uuid_str, + conn->session->id, + msg->chunk_id); + ret = -1; + reply_code = LTTNG_ERR_NOMEM; + goto end; + } + + pthread_mutex_lock(&session->lock); + if (session->pending_closure_trace_chunk && + session->pending_closure_trace_chunk != chunk) { + ERR("Trace chunk close command for session \"%s\" does not target the trace chunk pending closure", + session->session_name); + reply_code = LTTNG_ERR_INVALID_PROTOCOL; + ret = -1; + goto end_unlock_session; + } + + chunk_status = lttng_trace_chunk_set_close_timestamp( + chunk, close_timestamp); + if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { + ERR("Failed to set trace chunk close timestamp"); + ret = -1; + reply_code = LTTNG_ERR_UNK; + goto end_unlock_session; + } + + if (close_command.is_set) { + chunk_status = lttng_trace_chunk_set_close_command( + chunk, close_command.value); + if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { + ret = -1; + reply_code = LTTNG_ERR_INVALID; + goto end_unlock_session; + } + } + chunk_status = lttng_trace_chunk_get_name(chunk, &chunk_name, NULL); + if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { + ERR("Failed to get chunk name"); + ret = -1; + reply_code = LTTNG_ERR_UNK; + goto end_unlock_session; + } + if (!session->has_rotated && !session->snapshot) { + ret = lttng_strncpy(closed_trace_chunk_path, + session->output_path, + sizeof(closed_trace_chunk_path)); + if (ret) { + ERR("Failed to send trace chunk path: path length of %zu bytes exceeds the maximal allowed length of %zu bytes", + strlen(session->output_path), + sizeof(closed_trace_chunk_path)); + reply_code = LTTNG_ERR_NOMEM; + ret = -1; + goto end_unlock_session; + } + } else { + if (session->snapshot) { + ret = snprintf(closed_trace_chunk_path, + sizeof(closed_trace_chunk_path), + "%s/%s", session->output_path, + chunk_name); + } else { + ret = snprintf(closed_trace_chunk_path, + sizeof(closed_trace_chunk_path), + "%s/" DEFAULT_ARCHIVED_TRACE_CHUNKS_DIRECTORY + "/%s", + session->output_path, chunk_name); + } + if (ret < 0 || ret == sizeof(closed_trace_chunk_path)) { + ERR("Failed to format closed trace chunk resulting path"); + reply_code = ret < 0 ? LTTNG_ERR_UNK : LTTNG_ERR_NOMEM; + ret = -1; + goto end_unlock_session; + } + } + DBG("Reply chunk path on close: %s", closed_trace_chunk_path); + path_length = strlen(closed_trace_chunk_path) + 1; + if (path_length > UINT32_MAX) { + ERR("Closed trace chunk path exceeds the maximal length allowed by the protocol"); + ret = -1; + reply_code = LTTNG_ERR_INVALID_PROTOCOL; + goto end_unlock_session; + } + + if (session->current_trace_chunk == chunk) { + /* + * After a trace chunk close command, no new streams + * referencing the chunk may be created. Hence, on the + * event that no new trace chunk have been created for + * the session, the reference to the current trace chunk + * is released in order to allow it to be reclaimed when + * the last stream releases its reference to it. + */ + lttng_trace_chunk_put(session->current_trace_chunk); + session->current_trace_chunk = NULL; + } + lttng_trace_chunk_put(session->pending_closure_trace_chunk); + session->pending_closure_trace_chunk = NULL; +end_unlock_session: + pthread_mutex_unlock(&session->lock); + +end: + reply.generic.ret_code = htobe32((uint32_t) reply_code); + reply.path_length = htobe32((uint32_t) path_length); + buf_ret = lttng_dynamic_buffer_append( + &reply_payload, &reply, sizeof(reply)); + if (buf_ret) { + ERR("Failed to append \"close trace chunk\" command reply header to payload buffer"); + goto end_no_reply; + } + + if (reply_code == LTTNG_OK) { + buf_ret = lttng_dynamic_buffer_append(&reply_payload, + closed_trace_chunk_path, path_length); + if (buf_ret) { + ERR("Failed to append \"close trace chunk\" command reply path to payload buffer"); + goto end_no_reply; + } + } + + send_ret = conn->sock->ops->sendmsg(conn->sock, + reply_payload.data, + reply_payload.size, + 0); + if (send_ret < reply_payload.size) { + ERR("Failed to send \"close trace chunk\" command reply of %zu bytes (ret = %zd)", + reply_payload.size, send_ret); + ret = -1; + goto end_no_reply; + } +end_no_reply: + lttng_trace_chunk_put(chunk); + lttng_dynamic_buffer_reset(&reply_payload); + return ret; +} + +/* + * relay_trace_chunk_exists: check if a trace chunk exists + */ +static int relay_trace_chunk_exists(const struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn, + const struct lttng_buffer_view *payload) +{ + int ret = 0; + ssize_t send_ret; + struct relay_session *session = conn->session; + struct lttcomm_relayd_trace_chunk_exists *msg; + struct lttcomm_relayd_trace_chunk_exists_reply reply = {}; + struct lttng_buffer_view header_view; + struct lttng_trace_chunk *chunk = NULL; + uint64_t chunk_id; + + if (!session || !conn->version_check_done) { + ERR("Trying to close a trace chunk before version check"); + ret = -1; + goto end_no_reply; + } + + if (session->major == 2 && session->minor < 11) { + ERR("Chunk close command is unsupported before 2.11"); + ret = -1; + goto end_no_reply; + } + + header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg)); + if (!header_view.data) { + ERR("Failed to receive payload of chunk close command"); + ret = -1; + goto end_no_reply; + } + + /* Convert to host endianness. */ + msg = (typeof(msg)) header_view.data; + chunk_id = be64toh(msg->chunk_id); + + chunk = sessiond_trace_chunk_registry_get_chunk( + sessiond_trace_chunk_registry, + conn->session->sessiond_uuid, + conn->session->id, + chunk_id); + + reply = (typeof(reply)) { + .generic.ret_code = htobe32((uint32_t) LTTNG_OK), + .trace_chunk_exists = !!chunk, + }; + send_ret = conn->sock->ops->sendmsg(conn->sock, + &reply, sizeof(reply), 0); + if (send_ret < (ssize_t) sizeof(reply)) { + ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)", + send_ret); + ret = -1; + } +end_no_reply: + lttng_trace_chunk_put(chunk); + return ret; +} + +#define DBG_CMD(cmd_name, conn) \ + DBG3("Processing \"%s\" command for socket %i", cmd_name, conn->sock->fd); + +static int relay_process_control_command(struct relay_connection *conn, + const struct lttcomm_relayd_hdr *header, + const struct lttng_buffer_view *payload) { int ret = 0; - switch (be32toh(recv_hdr->cmd)) { + switch (header->cmd) { case RELAYD_CREATE_SESSION: - ret = relay_create_session(recv_hdr, conn); + DBG_CMD("RELAYD_CREATE_SESSION", conn); + ret = relay_create_session(header, conn, payload); break; case RELAYD_ADD_STREAM: - ret = relay_add_stream(recv_hdr, conn); + DBG_CMD("RELAYD_ADD_STREAM", conn); + ret = relay_add_stream(header, conn, payload); break; case RELAYD_START_DATA: - ret = relay_start(recv_hdr, conn); + DBG_CMD("RELAYD_START_DATA", conn); + ret = relay_start(header, conn, payload); break; case RELAYD_SEND_METADATA: - ret = relay_recv_metadata(recv_hdr, conn); + DBG_CMD("RELAYD_SEND_METADATA", conn); + ret = relay_recv_metadata(header, conn, payload); break; case RELAYD_VERSION: - ret = relay_send_version(recv_hdr, conn); + DBG_CMD("RELAYD_VERSION", conn); + ret = relay_send_version(header, conn, payload); break; case RELAYD_CLOSE_STREAM: - ret = relay_close_stream(recv_hdr, conn); + DBG_CMD("RELAYD_CLOSE_STREAM", conn); + ret = relay_close_stream(header, conn, payload); break; case RELAYD_DATA_PENDING: - ret = relay_data_pending(recv_hdr, conn); + DBG_CMD("RELAYD_DATA_PENDING", conn); + ret = relay_data_pending(header, conn, payload); break; case RELAYD_QUIESCENT_CONTROL: - ret = relay_quiescent_control(recv_hdr, conn); + DBG_CMD("RELAYD_QUIESCENT_CONTROL", conn); + ret = relay_quiescent_control(header, conn, payload); break; case RELAYD_BEGIN_DATA_PENDING: - ret = relay_begin_data_pending(recv_hdr, conn); + DBG_CMD("RELAYD_BEGIN_DATA_PENDING", conn); + ret = relay_begin_data_pending(header, conn, payload); break; case RELAYD_END_DATA_PENDING: - ret = relay_end_data_pending(recv_hdr, conn); + DBG_CMD("RELAYD_END_DATA_PENDING", conn); + ret = relay_end_data_pending(header, conn, payload); break; case RELAYD_SEND_INDEX: - ret = relay_recv_index(recv_hdr, conn); + DBG_CMD("RELAYD_SEND_INDEX", conn); + ret = relay_recv_index(header, conn, payload); break; case RELAYD_STREAMS_SENT: - ret = relay_streams_sent(recv_hdr, conn); + DBG_CMD("RELAYD_STREAMS_SENT", conn); + ret = relay_streams_sent(header, conn, payload); + break; + case RELAYD_RESET_METADATA: + DBG_CMD("RELAYD_RESET_METADATA", conn); + ret = relay_reset_metadata(header, conn, payload); + break; + case RELAYD_ROTATE_STREAMS: + DBG_CMD("RELAYD_ROTATE_STREAMS", conn); + ret = relay_rotate_session_streams(header, conn, payload); + break; + case RELAYD_CREATE_TRACE_CHUNK: + DBG_CMD("RELAYD_CREATE_TRACE_CHUNK", conn); + ret = relay_create_trace_chunk(header, conn, payload); + break; + case RELAYD_CLOSE_TRACE_CHUNK: + DBG_CMD("RELAYD_CLOSE_TRACE_CHUNK", conn); + ret = relay_close_trace_chunk(header, conn, payload); + break; + case RELAYD_TRACE_CHUNK_EXISTS: + DBG_CMD("RELAYD_TRACE_CHUNK_EXISTS", conn); + ret = relay_trace_chunk_exists(header, conn, payload); break; case RELAYD_UPDATE_SYNC_INFO: default: - ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd)); + ERR("Received unknown command (%u)", header->cmd); relay_unknown_command(conn); ret = -1; goto end; @@ -2164,267 +2813,467 @@ end: return ret; } -/* - * Handle index for a data stream. - * - * RCU read side lock MUST be acquired. - * - * Return 0 on success else a negative value. - */ -static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num, - int rotate_index) +static enum relay_connection_status relay_process_control_receive_payload( + struct relay_connection *conn) { - int ret = 0, index_created = 0; - uint64_t stream_id, data_offset; - struct relay_index *index, *wr_index = NULL; + int ret = 0; + enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK; + struct lttng_dynamic_buffer *reception_buffer = + &conn->protocol.ctrl.reception_buffer; + struct ctrl_connection_state_receive_payload *state = + &conn->protocol.ctrl.state.receive_payload; + struct lttng_buffer_view payload_view; + + if (state->left_to_receive == 0) { + /* Short-circuit for payload-less commands. */ + goto reception_complete; + } + + ret = conn->sock->ops->recvmsg(conn->sock, + reception_buffer->data + state->received, + state->left_to_receive, MSG_DONTWAIT); + if (ret < 0) { + if (errno != EAGAIN && errno != EWOULDBLOCK) { + PERROR("Unable to receive command payload on sock %d", + conn->sock->fd); + status = RELAY_CONNECTION_STATUS_ERROR; + } + goto end; + } else if (ret == 0) { + DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd); + status = RELAY_CONNECTION_STATUS_CLOSED; + goto end; + } + + assert(ret > 0); + assert(ret <= state->left_to_receive); - assert(stream); + state->left_to_receive -= ret; + state->received += ret; - stream_id = stream->stream_handle; - /* Get data offset because we are about to update the index. */ - data_offset = htobe64(stream->tracefile_size_current); + if (state->left_to_receive > 0) { + /* + * Can't transition to the protocol's next state, wait to + * receive the rest of the header. + */ + DBG3("Partial reception of control connection protocol payload (received %" PRIu64 " bytes, %" PRIu64 " bytes left to receive, fd = %i)", + state->received, state->left_to_receive, + conn->sock->fd); + goto end; + } +reception_complete: + DBG("Done receiving control command payload: fd = %i, payload size = %" PRIu64 " bytes", + conn->sock->fd, state->received); /* - * Lookup for an existing index for that stream id/sequence number. If on - * exists, the control thread already received the data for it thus we need - * to write it on disk. + * The payload required to process the command has been received. + * A view to the reception buffer is forwarded to the various + * commands and the state of the control is reset on success. + * + * Commands are responsible for sending their reply to the peer. */ - index = relay_index_find(stream_id, net_seq_num); - if (!index) { - /* A successful creation will add the object to the HT. */ - index = relay_index_create(stream_id, net_seq_num); - if (!index) { - ret = -1; - goto error; - } - index_created = 1; - stream->indexes_in_flight++; + payload_view = lttng_buffer_view_from_dynamic_buffer(reception_buffer, + 0, -1); + ret = relay_process_control_command(conn, + &state->header, &payload_view); + if (ret < 0) { + status = RELAY_CONNECTION_STATUS_ERROR; + goto end; } - if (rotate_index || stream->index_fd < 0) { - index->to_close_fd = stream->index_fd; - ret = index_create_file(stream->path_name, stream->channel_name, - relayd_uid, relayd_gid, stream->tracefile_size, - stream->tracefile_count_current); - if (ret < 0) { - /* This will close the stream's index fd if one. */ - relay_index_free_safe(index); - goto error; + ret = connection_reset_protocol_state(conn); + if (ret) { + status = RELAY_CONNECTION_STATUS_ERROR; + } +end: + return status; +} + +static enum relay_connection_status relay_process_control_receive_header( + struct relay_connection *conn) +{ + int ret = 0; + enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK; + struct lttcomm_relayd_hdr header; + struct lttng_dynamic_buffer *reception_buffer = + &conn->protocol.ctrl.reception_buffer; + struct ctrl_connection_state_receive_header *state = + &conn->protocol.ctrl.state.receive_header; + + assert(state->left_to_receive != 0); + + ret = conn->sock->ops->recvmsg(conn->sock, + reception_buffer->data + state->received, + state->left_to_receive, MSG_DONTWAIT); + if (ret < 0) { + if (errno != EAGAIN && errno != EWOULDBLOCK) { + PERROR("Unable to receive control command header on sock %d", + conn->sock->fd); + status = RELAY_CONNECTION_STATUS_ERROR; } - stream->index_fd = ret; + goto end; + } else if (ret == 0) { + DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd); + status = RELAY_CONNECTION_STATUS_CLOSED; + goto end; } - index->fd = stream->index_fd; - index->index_data.offset = data_offset; - if (index_created) { + assert(ret > 0); + assert(ret <= state->left_to_receive); + + state->left_to_receive -= ret; + state->received += ret; + + if (state->left_to_receive > 0) { /* - * Try to add the relay index object to the hash table. If an object - * already exist, destroy back the index created and set the data. + * Can't transition to the protocol's next state, wait to + * receive the rest of the header. */ - relay_index_add(index, &wr_index); - if (wr_index) { - /* Copy back data from the created index. */ - wr_index->fd = index->fd; - wr_index->to_close_fd = index->to_close_fd; - wr_index->index_data.offset = data_offset; - free(index); - } - } else { - /* The index already exists so write it on disk. */ - wr_index = index; + DBG3("Partial reception of control connection protocol header (received %" PRIu64 " bytes, %" PRIu64 " bytes left to receive, fd = %i)", + state->received, state->left_to_receive, + conn->sock->fd); + goto end; } - /* Do we have a writable ready index to write on disk. */ - if (wr_index) { - ret = relay_index_write(wr_index->fd, wr_index); - if (ret < 0) { - goto error; - } - stream->total_index_received++; - stream->indexes_in_flight--; - assert(stream->indexes_in_flight >= 0); + /* Transition to next state: receiving the command's payload. */ + conn->protocol.ctrl.state_id = + CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD; + memcpy(&header, reception_buffer->data, sizeof(header)); + header.circuit_id = be64toh(header.circuit_id); + header.data_size = be64toh(header.data_size); + header.cmd = be32toh(header.cmd); + header.cmd_version = be32toh(header.cmd_version); + memcpy(&conn->protocol.ctrl.state.receive_payload.header, + &header, sizeof(header)); + + DBG("Done receiving control command header: fd = %i, cmd = %" PRIu32 ", cmd_version = %" PRIu32 ", payload size = %" PRIu64 " bytes", + conn->sock->fd, header.cmd, header.cmd_version, + header.data_size); + + if (header.data_size > DEFAULT_NETWORK_RELAYD_CTRL_MAX_PAYLOAD_SIZE) { + ERR("Command header indicates a payload (%" PRIu64 " bytes) that exceeds the maximal payload size allowed on a control connection.", + header.data_size); + status = RELAY_CONNECTION_STATUS_ERROR; + goto end; } -error: - return ret; + conn->protocol.ctrl.state.receive_payload.left_to_receive = + header.data_size; + conn->protocol.ctrl.state.receive_payload.received = 0; + ret = lttng_dynamic_buffer_set_size(reception_buffer, + header.data_size); + if (ret) { + status = RELAY_CONNECTION_STATUS_ERROR; + goto end; + } + + if (header.data_size == 0) { + /* + * Manually invoke the next state as the poll loop + * will not wake-up to allow us to proceed further. + */ + status = relay_process_control_receive_payload(conn); + } +end: + return status; } /* - * relay_process_data: Process the data received on the data socket + * Process the commands received on the control socket */ -static -int relay_process_data(struct relay_connection *conn) +static enum relay_connection_status relay_process_control( + struct relay_connection *conn) +{ + enum relay_connection_status status; + + switch (conn->protocol.ctrl.state_id) { + case CTRL_CONNECTION_STATE_RECEIVE_HEADER: + status = relay_process_control_receive_header(conn); + break; + case CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD: + status = relay_process_control_receive_payload(conn); + break; + default: + ERR("Unknown control connection protocol state encountered."); + abort(); + } + + return status; +} + +static enum relay_connection_status relay_process_data_receive_header( + struct relay_connection *conn) { - int ret = 0, rotate_index = 0; - ssize_t size_ret; + int ret; + enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK; + struct data_connection_state_receive_header *state = + &conn->protocol.data.state.receive_header; + struct lttcomm_relayd_data_hdr header; struct relay_stream *stream; - struct lttcomm_relayd_data_hdr data_hdr; - uint64_t stream_id; - uint64_t net_seq_num; - uint32_t data_size; - struct relay_session *session; - assert(conn); + assert(state->left_to_receive != 0); - ret = conn->sock->ops->recvmsg(conn->sock, &data_hdr, - sizeof(struct lttcomm_relayd_data_hdr), 0); - if (ret <= 0) { - if (ret == 0) { - /* Orderly shutdown. Not necessary to print an error. */ - DBG("Socket %d did an orderly shutdown", conn->sock->fd); - } else { - ERR("Unable to receive data header on sock %d", conn->sock->fd); + ret = conn->sock->ops->recvmsg(conn->sock, + state->header_reception_buffer + state->received, + state->left_to_receive, MSG_DONTWAIT); + if (ret < 0) { + if (errno != EAGAIN && errno != EWOULDBLOCK) { + PERROR("Unable to receive data header on sock %d", conn->sock->fd); + status = RELAY_CONNECTION_STATUS_ERROR; } - ret = -1; + goto end; + } else if (ret == 0) { + /* Orderly shutdown. Not necessary to print an error. */ + DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd); + status = RELAY_CONNECTION_STATUS_CLOSED; goto end; } - stream_id = be64toh(data_hdr.stream_id); + assert(ret > 0); + assert(ret <= state->left_to_receive); - rcu_read_lock(); - stream = stream_find_by_id(relay_streams_ht, stream_id); - if (!stream) { - ret = -1; - goto end_rcu_unlock; + state->left_to_receive -= ret; + state->received += ret; + + if (state->left_to_receive > 0) { + /* + * Can't transition to the protocol's next state, wait to + * receive the rest of the header. + */ + DBG3("Partial reception of data connection header (received %" PRIu64 " bytes, %" PRIu64 " bytes left to receive, fd = %i)", + state->received, state->left_to_receive, + conn->sock->fd); + goto end; } - session = session_find_by_id(conn->sessions_ht, stream->session_id); - assert(session); + /* Transition to next state: receiving the payload. */ + conn->protocol.data.state_id = DATA_CONNECTION_STATE_RECEIVE_PAYLOAD; - data_size = be32toh(data_hdr.data_size); - if (data_buffer_size < data_size) { - char *tmp_data_ptr; + memcpy(&header, state->header_reception_buffer, sizeof(header)); + header.circuit_id = be64toh(header.circuit_id); + header.stream_id = be64toh(header.stream_id); + header.data_size = be32toh(header.data_size); + header.net_seq_num = be64toh(header.net_seq_num); + header.padding_size = be32toh(header.padding_size); + memcpy(&conn->protocol.data.state.receive_payload.header, &header, sizeof(header)); - tmp_data_ptr = realloc(data_buffer, data_size); - if (!tmp_data_ptr) { - ERR("Allocating data buffer"); - free(data_buffer); - ret = -1; - goto end_rcu_unlock; - } - data_buffer = tmp_data_ptr; - data_buffer_size = data_size; + conn->protocol.data.state.receive_payload.left_to_receive = + header.data_size; + conn->protocol.data.state.receive_payload.received = 0; + conn->protocol.data.state.receive_payload.rotate_index = false; + + DBG("Received data connection header on fd %i: circuit_id = %" PRIu64 ", stream_id = %" PRIu64 ", data_size = %" PRIu32 ", net_seq_num = %" PRIu64 ", padding_size = %" PRIu32, + conn->sock->fd, header.circuit_id, + header.stream_id, header.data_size, + header.net_seq_num, header.padding_size); + + stream = stream_get_by_id(header.stream_id); + if (!stream) { + DBG("relay_process_data_receive_payload: Cannot find stream %" PRIu64, + header.stream_id); + /* Protocol error. */ + status = RELAY_CONNECTION_STATUS_ERROR; + goto end; + } + + pthread_mutex_lock(&stream->lock); + /* Prepare stream for the reception of a new packet. */ + ret = stream_init_packet(stream, header.data_size, + &conn->protocol.data.state.receive_payload.rotate_index); + pthread_mutex_unlock(&stream->lock); + if (ret) { + ERR("Failed to rotate stream output file"); + status = RELAY_CONNECTION_STATUS_ERROR; + goto end_stream_unlock; } - memset(data_buffer, 0, data_size); - net_seq_num = be64toh(data_hdr.net_seq_num); +end_stream_unlock: + stream_put(stream); +end: + return status; +} + +static enum relay_connection_status relay_process_data_receive_payload( + struct relay_connection *conn) +{ + int ret; + enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK; + struct relay_stream *stream; + struct data_connection_state_receive_payload *state = + &conn->protocol.data.state.receive_payload; + const size_t chunk_size = RECV_DATA_BUFFER_SIZE; + char data_buffer[chunk_size]; + bool partial_recv = false; + bool new_stream = false, close_requested = false, index_flushed = false; + uint64_t left_to_receive = state->left_to_receive; + struct relay_session *session; + + DBG3("Receiving data for stream id %" PRIu64 " seqnum %" PRIu64 ", %" PRIu64" bytes received, %" PRIu64 " bytes left to receive", + state->header.stream_id, state->header.net_seq_num, + state->received, left_to_receive); + + stream = stream_get_by_id(state->header.stream_id); + if (!stream) { + /* Protocol error. */ + ERR("relay_process_data_receive_payload: cannot find stream %" PRIu64, + state->header.stream_id); + status = RELAY_CONNECTION_STATUS_ERROR; + goto end; + } - DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64, - data_size, stream_id, net_seq_num); - ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, data_size, 0); - if (ret <= 0) { - if (ret == 0) { - /* Orderly shutdown. Not necessary to print an error. */ - DBG("Socket %d did an orderly shutdown", conn->sock->fd); + pthread_mutex_lock(&stream->lock); + session = stream->trace->session; + if (!conn->session) { + ret = connection_set_session(conn, session); + if (ret) { + status = RELAY_CONNECTION_STATUS_ERROR; + goto end_stream_unlock; } - ret = -1; - goto end_rcu_unlock; } - /* Check if a rotation is needed. */ - if (stream->tracefile_size > 0 && - (stream->tracefile_size_current + data_size) > - stream->tracefile_size) { - struct relay_viewer_stream *vstream; - uint64_t new_id; + /* + * The size of the "chunk" received on any iteration is bounded by: + * - the data left to receive, + * - the data immediately available on the socket, + * - the on-stack data buffer + */ + while (left_to_receive > 0 && !partial_recv) { + size_t recv_size = min(left_to_receive, chunk_size); + struct lttng_buffer_view packet_chunk; - new_id = (stream->tracefile_count_current + 1) % - stream->tracefile_count; - /* - * When we wrap-around back to 0, we start overwriting old - * trace data. - */ - if (!stream->tracefile_overwrite && new_id == 0) { - stream->tracefile_overwrite = 1; - } - pthread_mutex_lock(&stream->viewer_stream_rotation_lock); - if (stream->tracefile_overwrite) { - stream->oldest_tracefile_id = - (stream->oldest_tracefile_id + 1) % - stream->tracefile_count; - } - vstream = viewer_stream_find_by_id(stream->stream_handle); - if (vstream) { + ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, + recv_size, MSG_DONTWAIT); + if (ret < 0) { + if (errno != EAGAIN && errno != EWOULDBLOCK) { + PERROR("Socket %d error", conn->sock->fd); + status = RELAY_CONNECTION_STATUS_ERROR; + } + goto end_stream_unlock; + } else if (ret == 0) { + /* No more data ready to be consumed on socket. */ + DBG3("No more data ready for consumption on data socket of stream id %" PRIu64, + state->header.stream_id); + status = RELAY_CONNECTION_STATUS_CLOSED; + break; + } else if (ret < (int) recv_size) { /* - * The viewer is reading a file about to be - * overwritten. Close the FDs it is - * currently using and let it handle the fault. + * All the data available on the socket has been + * consumed. */ - if (vstream->tracefile_count_current == new_id) { - pthread_mutex_lock(&vstream->overwrite_lock); - vstream->abort_flag = 1; - pthread_mutex_unlock(&vstream->overwrite_lock); - DBG("Streaming side setting abort_flag on stream %s_%lu\n", - stream->channel_name, new_id); - } else if (vstream->tracefile_count_current == - stream->tracefile_count_current) { - /* - * The reader and writer were in the - * same trace file, inform the viewer - * that no new index will ever be added - * to this file. - */ - vstream->close_write_flag = 1; - } + partial_recv = true; + recv_size = ret; } - ret = utils_rotate_stream_file(stream->path_name, stream->channel_name, - stream->tracefile_size, stream->tracefile_count, - relayd_uid, relayd_gid, stream->fd, - &(stream->tracefile_count_current), &stream->fd); - pthread_mutex_unlock(&stream->viewer_stream_rotation_lock); - if (ret < 0) { - ERR("Rotating stream output file"); - goto end_rcu_unlock; + + packet_chunk = lttng_buffer_view_init(data_buffer, + 0, recv_size); + assert(packet_chunk.data); + + ret = stream_write(stream, &packet_chunk, 0); + if (ret) { + ERR("Relay error writing data to file"); + status = RELAY_CONNECTION_STATUS_ERROR; + goto end_stream_unlock; } - /* Reset current size because we just perform a stream rotation. */ - stream->tracefile_size_current = 0; - rotate_index = 1; + + left_to_receive -= recv_size; + state->received += recv_size; + state->left_to_receive = left_to_receive; } - /* - * Index are handled in protocol version 2.4 and above. Also, snapshot and - * index are NOT supported. - */ - if (session->minor >= 4 && !session->snapshot) { - ret = handle_index_data(stream, net_seq_num, rotate_index); + if (state->left_to_receive > 0) { + /* + * Did not receive all the data expected, wait for more data to + * become available on the socket. + */ + DBG3("Partial receive on data connection of stream id %" PRIu64 ", %" PRIu64 " bytes received, %" PRIu64 " bytes left to receive", + state->header.stream_id, state->received, + state->left_to_receive); + goto end_stream_unlock; + } + + ret = stream_write(stream, NULL, state->header.padding_size); + if (ret) { + status = RELAY_CONNECTION_STATUS_ERROR; + goto end_stream_unlock; + } + + if (session_streams_have_index(session)) { + ret = stream_update_index(stream, state->header.net_seq_num, + state->rotate_index, &index_flushed, + state->header.data_size + state->header.padding_size); if (ret < 0) { - goto end_rcu_unlock; + ERR("Failed to update index: stream %" PRIu64 " net_seq_num %" PRIu64 " ret %d", + stream->stream_handle, + state->header.net_seq_num, ret); + status = RELAY_CONNECTION_STATUS_ERROR; + goto end_stream_unlock; } } - /* Write data to stream output fd. */ - size_ret = lttng_write(stream->fd, data_buffer, data_size); - if (size_ret < data_size) { - ERR("Relay error writing data to file"); - ret = -1; - goto end_rcu_unlock; + if (stream->prev_data_seq == -1ULL) { + new_stream = true; } - DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64, - ret, stream->stream_handle); - - ret = write_padding_to_file(stream->fd, be32toh(data_hdr.padding_size)); - if (ret < 0) { - goto end_rcu_unlock; + ret = stream_complete_packet(stream, state->header.data_size + + state->header.padding_size, state->header.net_seq_num, + index_flushed); + if (ret) { + status = RELAY_CONNECTION_STATUS_ERROR; + goto end_stream_unlock; } - stream->tracefile_size_current += data_size + be32toh(data_hdr.padding_size); - stream->prev_seq = net_seq_num; + /* + * Resetting the protocol state (to RECEIVE_HEADER) will trash the + * contents of *state which are aliased (union) to the same location as + * the new state. Don't use it beyond this point. + */ + connection_reset_protocol_state(conn); + state = NULL; + +end_stream_unlock: + close_requested = stream->close_requested; + pthread_mutex_unlock(&stream->lock); + if (close_requested && left_to_receive == 0) { + try_stream_close(stream); + } - try_close_stream(session, stream); + if (new_stream) { + pthread_mutex_lock(&session->lock); + uatomic_set(&session->new_streams, 1); + pthread_mutex_unlock(&session->lock); + } -end_rcu_unlock: - rcu_read_unlock(); + stream_put(stream); end: - return ret; + return status; } -static -void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd) +/* + * relay_process_data: Process the data received on the data socket + */ +static enum relay_connection_status relay_process_data( + struct relay_connection *conn) { - int ret; + enum relay_connection_status status; + + switch (conn->protocol.data.state_id) { + case DATA_CONNECTION_STATE_RECEIVE_HEADER: + status = relay_process_data_receive_header(conn); + break; + case DATA_CONNECTION_STATE_RECEIVE_PAYLOAD: + status = relay_process_data_receive_payload(conn); + break; + default: + ERR("Unexpected data connection communication state."); + abort(); + } + + return status; +} - assert(events); +static void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd) +{ + int ret; (void) lttng_poll_del(events, pollfd); @@ -2434,38 +3283,43 @@ void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd) } } -static void destroy_connection(struct lttng_ht *relay_connections_ht, - struct relay_connection *conn) +static void relay_thread_close_connection(struct lttng_poll_event *events, + int pollfd, struct relay_connection *conn) { - assert(relay_connections_ht); - assert(conn); - - connection_delete(relay_connections_ht, conn); + const char *type_str; - /* For the control socket, we try to destroy the session. */ - if (conn->type == RELAY_CONTROL && conn->session) { - destroy_session(conn->session, conn->sessions_ht); + switch (conn->type) { + case RELAY_DATA: + type_str = "Data"; + break; + case RELAY_CONTROL: + type_str = "Control"; + break; + case RELAY_VIEWER_COMMAND: + type_str = "Viewer Command"; + break; + case RELAY_VIEWER_NOTIFICATION: + type_str = "Viewer Notification"; + break; + default: + type_str = "Unknown"; } - - connection_destroy(conn); + cleanup_connection_pollfd(events, pollfd); + connection_put(conn); + DBG("%s connection closed with %d", type_str, pollfd); } /* * This thread does the actual work */ -static -void *relay_thread_worker(void *data) +static void *relay_thread_worker(void *data) { int ret, err = -1, last_seen_data_fd = -1; uint32_t nb_fd; - struct relay_connection *conn; struct lttng_poll_event events; struct lttng_ht *relay_connections_ht; struct lttng_ht_iter iter; - struct lttcomm_relayd_hdr recv_hdr; - struct relay_local_data *relay_ctx = (struct relay_local_data *) data; - struct lttng_ht *sessions_ht = relay_ctx->sessions_ht; - struct relay_index *index; + struct relay_connection *destroy_conn = NULL; DBG("[thread] Relay worker started"); @@ -2485,12 +3339,6 @@ void *relay_thread_worker(void *data) goto relay_connections_ht_error; } - /* Tables of received indexes indexed by index handle and net_seq_num. */ - indexes_ht = lttng_ht_new(0, LTTNG_HT_TYPE_TWO_U64); - if (!indexes_ht) { - goto indexes_ht_error; - } - ret = create_thread_poll_set(&events, 2); if (ret < 0) { goto error_poll_create; @@ -2525,9 +3373,9 @@ restart: nb_fd = ret; /* - * Process control. The control connection is prioritised so we don't - * starve it with high throughout put tracing data on the data - * connection. + * Process control. The control connection is + * prioritized so we don't starve it with high + * throughput tracing data on the data connection. */ for (i = 0; i < nb_fd; i++) { /* Fetch once the poll data */ @@ -2545,56 +3393,33 @@ restart: /* Inspect the relay conn pipe for new connection */ if (pollfd == relay_conn_pipe[0]) { - if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { - ERR("Relay connection pipe error"); - goto error; - } else if (revents & LPOLLIN) { + if (revents & LPOLLIN) { + struct relay_connection *conn; + ret = lttng_read(relay_conn_pipe[0], &conn, sizeof(conn)); if (ret < 0) { goto error; } - conn->sessions_ht = sessions_ht; - connection_init(conn); lttng_poll_add(&events, conn->sock->fd, LPOLLIN | LPOLLRDHUP); - rcu_read_lock(); - lttng_ht_add_unique_ulong(relay_connections_ht, - &conn->sock_n); - rcu_read_unlock(); + connection_ht_add(relay_connections_ht, conn); DBG("Connection socket %d added", conn->sock->fd); + } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { + ERR("Relay connection pipe error"); + goto error; + } else { + ERR("Unexpected poll events %u for sock %d", revents, pollfd); + goto error; } } else { - rcu_read_lock(); - conn = connection_find_by_sock(relay_connections_ht, pollfd); + struct relay_connection *ctrl_conn; + + ctrl_conn = connection_get_by_sock(relay_connections_ht, pollfd); /* If not found, there is a synchronization issue. */ - assert(conn); + assert(ctrl_conn); - if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { - cleanup_connection_pollfd(&events, pollfd); - destroy_connection(relay_connections_ht, conn); - if (last_seen_data_fd == pollfd) { - last_seen_data_fd = last_notdel_data_fd; - } - } else if (revents & LPOLLIN) { - if (conn->type == RELAY_CONTROL) { - ret = conn->sock->ops->recvmsg(conn->sock, &recv_hdr, - sizeof(recv_hdr), 0); - if (ret <= 0) { - /* Connection closed */ - cleanup_connection_pollfd(&events, pollfd); - destroy_connection(relay_connections_ht, conn); - DBG("Control connection closed with %d", pollfd); - } else { - ret = relay_process_control(&recv_hdr, conn); - if (ret < 0) { - /* Clear the session on error. */ - cleanup_connection_pollfd(&events, pollfd); - destroy_connection(relay_connections_ht, conn); - DBG("Connection closed with %d", pollfd); - } - seen_control = 1; - } - } else { + if (ctrl_conn->type == RELAY_DATA) { + if (revents & LPOLLIN) { /* * Flag the last seen data fd not deleted. It will be * used as the last seen fd if any fd gets deleted in @@ -2602,10 +3427,55 @@ restart: */ last_notdel_data_fd = pollfd; } + goto put_ctrl_connection; + } + assert(ctrl_conn->type == RELAY_CONTROL); + + if (revents & LPOLLIN) { + enum relay_connection_status status; + + status = relay_process_control(ctrl_conn); + if (status != RELAY_CONNECTION_STATUS_OK) { + /* + * On socket error flag the session as aborted to force + * the cleanup of its stream otherwise it can leak + * during the lifetime of the relayd. + * + * This prevents situations in which streams can be + * left opened because an index was received, the + * control connection is closed, and the data + * connection is closed (uncleanly) before the packet's + * data provided. + * + * Since the control connection encountered an error, + * it is okay to be conservative and close the + * session right now as we can't rely on the protocol + * being respected anymore. + */ + if (status == RELAY_CONNECTION_STATUS_ERROR) { + session_abort(ctrl_conn->session); + } + + /* Clear the connection on error or close. */ + relay_thread_close_connection(&events, + pollfd, + ctrl_conn); + } + seen_control = 1; + } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { + relay_thread_close_connection(&events, + pollfd, ctrl_conn); + if (last_seen_data_fd == pollfd) { + last_seen_data_fd = last_notdel_data_fd; + } } else { - ERR("Unknown poll events %u for sock %d", revents, pollfd); + ERR("Unexpected poll events %u for control sock %d", + revents, pollfd); + connection_put(ctrl_conn); + goto error; } - rcu_read_unlock(); + put_ctrl_connection: + connection_put(ctrl_conn); } } @@ -2635,49 +3505,77 @@ restart: /* Fetch the poll data. */ uint32_t revents = LTTNG_POLL_GETEV(&events, i); int pollfd = LTTNG_POLL_GETFD(&events, i); + struct relay_connection *data_conn; health_code_update(); + if (!revents) { + /* No activity for this FD (poll implementation). */ + continue; + } + /* Skip the command pipe. It's handled in the first loop. */ if (pollfd == relay_conn_pipe[0]) { continue; } - if (revents) { - rcu_read_lock(); - conn = connection_find_by_sock(relay_connections_ht, pollfd); - if (!conn) { - /* Skip it. Might be removed before. */ - rcu_read_unlock(); - continue; - } - - if (revents & LPOLLIN) { - if (conn->type != RELAY_DATA) { - rcu_read_unlock(); - continue; - } - - ret = relay_process_data(conn); - /* Connection closed */ - if (ret < 0) { - cleanup_connection_pollfd(&events, pollfd); - destroy_connection(relay_connections_ht, conn); - DBG("Data connection closed with %d", pollfd); - /* - * Every goto restart call sets the last seen fd where - * here we don't really care since we gracefully - * continue the loop after the connection is deleted. - */ - } else { - /* Keep last seen port. */ - last_seen_data_fd = pollfd; - rcu_read_unlock(); - goto restart; + data_conn = connection_get_by_sock(relay_connections_ht, pollfd); + if (!data_conn) { + /* Skip it. Might be removed before. */ + continue; + } + if (data_conn->type == RELAY_CONTROL) { + goto put_data_connection; + } + assert(data_conn->type == RELAY_DATA); + + if (revents & LPOLLIN) { + enum relay_connection_status status; + + status = relay_process_data(data_conn); + /* Connection closed or error. */ + if (status != RELAY_CONNECTION_STATUS_OK) { + /* + * On socket error flag the session as aborted to force + * the cleanup of its stream otherwise it can leak + * during the lifetime of the relayd. + * + * This prevents situations in which streams can be + * left opened because an index was received, the + * control connection is closed, and the data + * connection is closed (uncleanly) before the packet's + * data provided. + * + * Since the data connection encountered an error, + * it is okay to be conservative and close the + * session right now as we can't rely on the protocol + * being respected anymore. + */ + if (status == RELAY_CONNECTION_STATUS_ERROR) { + session_abort(data_conn->session); } + relay_thread_close_connection(&events, pollfd, + data_conn); + /* + * Every goto restart call sets the last seen fd where + * here we don't really care since we gracefully + * continue the loop after the connection is deleted. + */ + } else { + /* Keep last seen port. */ + last_seen_data_fd = pollfd; + connection_put(data_conn); + goto restart; } - rcu_read_unlock(); + } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { + relay_thread_close_connection(&events, pollfd, + data_conn); + } else { + ERR("Unknown poll events %u for data sock %d", + revents, pollfd); } + put_data_connection: + connection_put(data_conn); } last_seen_data_fd = -1; } @@ -2687,27 +3585,26 @@ restart: exit: error: - lttng_poll_clean(&events); - - /* Cleanup reamaining connection object. */ + /* Cleanup remaining connection object. */ rcu_read_lock(); - cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, conn, + cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, + destroy_conn, sock_n.node) { health_code_update(); - destroy_connection(relay_connections_ht, conn); + + session_abort(destroy_conn->session); + + /* + * No need to grab another ref, because we own + * destroy_conn. + */ + relay_thread_close_connection(&events, destroy_conn->sock->fd, + destroy_conn); } rcu_read_unlock(); + + lttng_poll_clean(&events); error_poll_create: - rcu_read_lock(); - cds_lfht_for_each_entry(indexes_ht->ht, &iter.iter, index, - index_n.node) { - health_code_update(); - relay_index_delete(index); - relay_index_free_safe(index); - } - rcu_read_unlock(); - lttng_ht_destroy(indexes_ht); -indexes_ht_error: lttng_ht_destroy(relay_connections_ht); relay_connections_ht_error: /* Close relay conn pipes */ @@ -2716,7 +3613,6 @@ relay_connections_ht_error: DBG("Thread exited with error"); } DBG("Worker thread cleanup complete"); - free(data_buffer); error_testpoint: if (err) { health_error(); @@ -2748,7 +3644,6 @@ int main(int argc, char **argv) { int ret = 0, retval = 0; void *status; - struct relay_local_data *relay_ctx = NULL; /* Parse arguments */ progname = argv[0]; @@ -2770,7 +3665,8 @@ int main(int argc, char **argv) goto exit_options; } - ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG); + ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG, + -1, -1); if (ret < 0) { ERR("Unable to create %s", opt_output_path); retval = -1; @@ -2799,6 +3695,12 @@ int main(int argc, char **argv) } } + sessiond_trace_chunk_registry = sessiond_trace_chunk_registry_create(); + if (!sessiond_trace_chunk_registry) { + ERR("Failed to initialize session daemon trace chunk registry"); + retval = -1; + goto exit_sessiond_trace_chunk_registry; + } /* Initialize thread health monitoring */ health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES); @@ -2814,19 +3716,6 @@ int main(int argc, char **argv) goto exit_init_data; } - /* We need those values for the file/dir creation. */ - relayd_uid = getuid(); - relayd_gid = getgid(); - - /* Check if daemon is UID = 0 */ - if (relayd_uid == 0) { - if (control_uri->port < 1024 || data_uri->port < 1024 || live_uri->port < 1024) { - ERR("Need to be root to use ports < 1024"); - retval = -1; - goto exit_init_data; - } - } - /* Setup the thread apps communication pipe. */ if (create_relay_conn_pipe()) { retval = -1; @@ -2836,23 +3725,13 @@ int main(int argc, char **argv) /* Init relay command queue. */ cds_wfcq_init(&relay_conn_queue.head, &relay_conn_queue.tail); - /* Set up max poll set size */ - lttng_poll_set_max_size(); - /* Initialize communication library */ lttcomm_init(); lttcomm_inet_init(); - relay_ctx = zmalloc(sizeof(struct relay_local_data)); - if (!relay_ctx) { - PERROR("relay_ctx"); - retval = -1; - goto exit_init_data; - } - /* tables of sessions indexed by session ID */ - relay_ctx->sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); - if (!relay_ctx->sessions_ht) { + sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); + if (!sessions_ht) { retval = -1; goto exit_init_data; } @@ -2878,7 +3757,7 @@ int main(int argc, char **argv) } /* Create thread to manage the client socket */ - ret = pthread_create(&health_thread, NULL, + ret = pthread_create(&health_thread, default_pthread_attr(), thread_manage_health, (void *) NULL); if (ret) { errno = ret; @@ -2888,7 +3767,7 @@ int main(int argc, char **argv) } /* Setup the dispatcher thread */ - ret = pthread_create(&dispatcher_thread, NULL, + ret = pthread_create(&dispatcher_thread, default_pthread_attr(), relay_thread_dispatcher, (void *) NULL); if (ret) { errno = ret; @@ -2898,8 +3777,8 @@ int main(int argc, char **argv) } /* Setup the worker thread */ - ret = pthread_create(&worker_thread, NULL, - relay_thread_worker, (void *) relay_ctx); + ret = pthread_create(&worker_thread, default_pthread_attr(), + relay_thread_worker, NULL); if (ret) { errno = ret; PERROR("pthread_create worker"); @@ -2908,7 +3787,7 @@ int main(int argc, char **argv) } /* Setup the listener thread */ - ret = pthread_create(&listener_thread, NULL, + ret = pthread_create(&listener_thread, default_pthread_attr(), relay_thread_listener, (void *) NULL); if (ret) { errno = ret; @@ -2917,7 +3796,7 @@ int main(int argc, char **argv) goto exit_listener_thread; } - ret = relayd_live_create(live_uri, relay_ctx); + ret = relayd_live_create(live_uri); if (ret) { ERR("Starting live viewer threads"); retval = -1; @@ -2972,9 +3851,20 @@ exit_health_quit_pipe: exit_init_data: health_app_destroy(health_relayd); + sessiond_trace_chunk_registry_destroy(sessiond_trace_chunk_registry); exit_health_app_create: +exit_sessiond_trace_chunk_registry: exit_options: - relayd_cleanup(relay_ctx); + /* + * Wait for all pending call_rcu work to complete before tearing + * down data structures. call_rcu worker may be trying to + * perform lookups in those structures. + */ + rcu_barrier(); + relayd_cleanup(); + + /* Ensure all prior call_rcu are done. */ + rcu_barrier(); if (!retval) { exit(EXIT_SUCCESS);