X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-relayd%2Fmain.c;h=1e2e9050fef2d6d77ca78e9767a2fea759ba99be;hp=fb290bacc1a4345e12850d475fa10c720115f77f;hb=c6db3843828a8fbf08444a2bc4191291a4807936;hpb=f48c25b760239f20a6e82f3839e04f82d07bdeea diff --git a/src/bin/lttng-relayd/main.c b/src/bin/lttng-relayd/main.c index fb290bacc..1e2e9050f 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 @@ -39,13 +39,13 @@ #include #include #include -#include #include #include #include #include #include +#include #include #include #include @@ -54,7 +54,11 @@ #include #include #include -#include +#include +#include +#include +#include +#include #include "cmd.h" #include "ctf-trace.h" @@ -68,6 +72,24 @@ #include "session.h" #include "stream.h" #include "connection.h" +#include "tracefile-array.h" +#include "tcp_keep_alive.h" + +static const char *help_msg = +#ifdef LTTNG_EMBED_HELP +#include +#else +NULL +#endif +; + +enum relay_connection_status { + RELAY_CONNECTION_STATUS_OK, + /* An error occured 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 +101,11 @@ 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 +#define FILE_COPY_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 +140,11 @@ 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; /* @@ -123,22 +155,14 @@ 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; @@ -154,29 +178,11 @@ 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 +190,7 @@ 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 +202,48 @@ 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 +253,40 @@ 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 +316,10 @@ 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 +342,9 @@ 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 +369,7 @@ 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 +394,14 @@ 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 +492,32 @@ 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 +537,7 @@ 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 +551,7 @@ end: return ret; } -static -int notify_health_quit_pipe(int *pipe) +static int notify_health_quit_pipe(int *pipe) { ssize_t ret; @@ -551,13 +600,9 @@ 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 +627,7 @@ 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 +638,10 @@ 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 +652,13 @@ 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 +683,7 @@ 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 +695,7 @@ 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 +726,7 @@ 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 +738,7 @@ 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 +757,7 @@ 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 +776,10 @@ 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 +792,19 @@ 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) { @@ -869,7 +859,10 @@ restart: pollfd = LTTNG_POLL_GETFD(&events, i); if (!revents) { - /* No activity for this FD (poll implementation). */ + /* + * No activity for this FD (poll + * implementation). + */ continue; } @@ -880,38 +873,32 @@ 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; } @@ -920,20 +907,39 @@ 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; } } } @@ -973,8 +979,7 @@ 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; @@ -991,12 +996,16 @@ 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(); @@ -1013,14 +1022,15 @@ 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); @@ -1046,72 +1056,36 @@ 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. + * Set index data from the control port to a given index object. */ -static void destroy_session(struct relay_session *session, - struct lttng_ht *sessions_ht) +static int set_index_control_data(struct relay_index *index, + struct lttcomm_relayd_index *data, + struct relay_connection *conn) { - assert(session); - assert(sessions_ht); - - /* Indicate that this session can be destroyed from now on. */ - session->close_flag = 1; - - try_close_streams(session); + struct ctf_packet_index index_data; /* - * This will try to delete and destroy the session if no viewer is attached - * to it meaning the refcount is down to zero. + * The index on disk is encoded in big endian. */ - session_try_destroy(sessions_ht, session); + index_data.packet_size = htobe64(data->packet_size); + index_data.content_size = htobe64(data->content_size); + index_data.timestamp_begin = htobe64(data->timestamp_begin); + index_data.timestamp_end = htobe64(data->timestamp_end); + index_data.events_discarded = htobe64(data->events_discarded); + index_data.stream_id = htobe64(data->stream_id); + + if (conn->minor >= 8) { + index->index_data.stream_instance_id = htobe64(data->stream_instance_id); + index->index_data.packet_seq_num = htobe64(data->packet_seq_num); + } + + return relay_index_set_data(index, &index_data); } -/* - * 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; } /* @@ -1119,45 +1093,54 @@ 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; + int ret = 0; + ssize_t send_ret; struct relay_session *session; struct lttcomm_relayd_status_session reply; + char session_name[LTTNG_NAME_MAX]; + char hostname[LTTNG_HOST_NAME_MAX]; + uint32_t live_timer = 0; + bool snapshot = false; - assert(recv_hdr); - assert(conn); + memset(session_name, 0, LTTNG_NAME_MAX); + memset(hostname, 0, LTTNG_HOST_NAME_MAX); memset(&reply, 0, sizeof(reply)); - session = session_create(); + 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 { + /* From 2.11 to ... */ + ret = cmd_create_session_2_11(payload, session_name, + hostname, &live_timer, &snapshot); + } + + if (ret < 0) { + goto send_reply; + } + + session = session_create(session_name, hostname, live_timer, + snapshot, conn->major, conn->minor); 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; + DBG("Created session %" PRIu64, session->id); 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: +send_reply: if (ret < 0) { reply.ret_code = htobe32(LTTNG_ERR_FATAL); } else { @@ -1165,9 +1148,10 @@ error: } 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 (send_ret < (ssize_t) sizeof(reply)) { + ERR("Failed to send \"create session\" command reply (ret = %zd)", + send_ret); + ret = -1; } return ret; @@ -1177,181 +1161,131 @@ 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; - 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); + /* + * 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); } - pthread_mutex_unlock(&conn->session->viewer_ready_lock); - return; -} - -/* - * 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) -{ - assert(conn); - assert(stream); + rcu_read_unlock(); - cds_list_add(&stream->recv_list, &conn->recv_head); + /* + * Inform the viewer that there are new streams in the session. + */ + if (session->viewer_attached) { + uatomic_set(&session->new_streams, 1); + } + pthread_mutex_unlock(&session->lock); } /* * 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; + struct relay_stream_chunk_id stream_chunk_id = { 0 }; - 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; - } - - 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; - } - - 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 (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; } - /* - * 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); + goto send_reply; } - trace = ctf_trace_find_by_path(session->ctf_traces_ht, stream->path_name); + trace = ctf_trace_get_by_path_or_create(session, 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); + goto send_reply; } - ctf_trace_get_ref(trace); + /* This stream here has one reference on the 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; - } + pthread_mutex_lock(&last_relay_stream_id_lock); + stream_handle = ++last_relay_stream_id; + pthread_mutex_unlock(&last_relay_stream_id_lock); - /* - * 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. - */ - queue_stream(stream, conn); + /* We pass ownership of path_name and channel_name. */ + stream = stream_create(trace, stream_handle, path_name, + channel_name, tracefile_size, tracefile_count, + &stream_chunk_id); + path_name = NULL; + channel_name = NULL; /* - * 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. + * Streams are the owners of their trace. Reference to trace is + * kept within stream_create(). */ - lttng_ht_add_unique_u64(relay_streams_ht, &stream->node); - cds_list_add_tail(&stream->trace_list, &trace->stream_list); - - session->stream_count++; - - DBG("Relay new stream added %s with ID %" PRIu64, stream->channel_name, - stream->stream_handle); + ctf_trace_put(trace); -end: +send_reply: 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.handle = htobe64(stream_handle); + if (!stream) { 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; + if (send_ret < (ssize_t) sizeof(reply)) { + ERR("Failed to send \"add 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); + free(path_name); + free(channel_name); return ret; } /* * relay_close_stream: close a specific stream */ -static -int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_connection *conn) +static int relay_close_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 lttcomm_relayd_close_stream stream_info; struct lttcomm_relayd_generic_reply reply; @@ -1359,43 +1293,151 @@ int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr, DBG("Close stream received"); - 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, &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_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); - rcu_read_lock(); - stream = stream_find_by_id(relay_streams_ht, - be64toh(stream_info.stream_id)); + stream = stream_get_by_id(stream_info.stream_id); if (!stream) { ret = -1; - goto end_unlock; + goto end; } - stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num); - stream->close_flag = 1; - session->stream_count--; + /* + * Set last_net_seq_num before the close flag. Required by data + * pending check. + */ + pthread_mutex_lock(&stream->lock); + stream->last_net_seq_num = stream_info.last_net_seq_num; + pthread_mutex_unlock(&stream->lock); - /* Check if we can close it or else the data will do it. */ - try_close_stream(session, stream); + /* + * 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. + */ + try_stream_close(stream); + if (stream->is_metadata) { + struct relay_viewer_stream *vstream; + + 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)); + if (ret < 0) { + 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_generic_reply), 0); + if (send_ret < (ssize_t) sizeof(reply)) { + ERR("Failed to send \"close stream\" command reply (ret = %zd)", + send_ret); + ret = -1; + } + +end_no_session: + return ret; +} + +/* + * relay_reset_metadata: reset a metadata stream + */ +static +int relay_reset_metadata(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_reset_metadata stream_info; + struct lttcomm_relayd_generic_reply reply; + struct relay_stream *stream; + + DBG("Reset metadata received"); + + if (!session || !conn->version_check_done) { + ERR("Trying to reset a metadata stream before version check"); + ret = -1; + goto end_no_session; + } + + 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); + + 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; + } + + ret = utils_rotate_stream_file(stream->path_name, stream->channel_name, + 0, 0, -1, -1, stream->stream_fd->fd, NULL, + &stream->stream_fd->fd); + if (ret < 0) { + ERR("Failed to rotate metadata file %s of channel %s", + 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); @@ -1404,9 +1446,10 @@ 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: @@ -1416,18 +1459,16 @@ 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); } } @@ -1435,11 +1476,12 @@ 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; @@ -1449,11 +1491,13 @@ 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"); + 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; } return ret; @@ -1490,95 +1534,396 @@ end: } /* - * relay_recv_metadata: receive the metada for the session. + * Close the current index file if it is open, and create a new one. + * + * Return 0 on success, -1 on error. */ static -int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_connection *conn) +int create_rotate_index_file(struct relay_stream *stream, + const char *stream_path) { - int ret = htobe32(LTTNG_OK); - ssize_t size_ret; - struct relay_session *session = conn->session; - struct lttcomm_relayd_metadata_payload *metadata_struct; - struct relay_stream *metadata_stream; - uint64_t data_size, payload_size; - struct ctf_trace *ctf_trace; + int ret; + uint32_t major, minor; + + /* Put ref on previous index_file. */ + if (stream->index_file) { + lttng_index_file_put(stream->index_file); + stream->index_file = NULL; + } + major = stream->trace->session->major; + minor = stream->trace->session->minor; + stream->index_file = lttng_index_file_create(stream_path, + stream->channel_name, + -1, -1, stream->tracefile_size, + tracefile_array_get_file_index_head(stream->tfa), + lttng_to_index_major(major, minor), + lttng_to_index_minor(major, minor)); + if (!stream->index_file) { + ret = -1; + goto end; + } - if (!session) { - ERR("Metadata sent before version check"); + ret = 0; + +end: + return ret; +} + +static +int do_rotate_stream_data(struct relay_stream *stream) +{ + int ret; + + DBG("Rotating stream %" PRIu64 " data file", + stream->stream_handle); + /* Perform the stream rotation. */ + ret = utils_rotate_stream_file(stream->path_name, + stream->channel_name, stream->tracefile_size, + stream->tracefile_count, -1, + -1, stream->stream_fd->fd, + NULL, &stream->stream_fd->fd); + if (ret < 0) { + ERR("Rotating stream output file"); + goto end; + } + stream->tracefile_size_current = 0; + stream->pos_after_last_complete_data_index = 0; + stream->data_rotated = true; + + if (stream->data_rotated && stream->index_rotated) { + /* Rotation completed; reset its state. */ + DBG("Rotation completed for stream %" PRIu64, + stream->stream_handle); + stream->rotate_at_seq_num = -1ULL; + stream->data_rotated = false; + stream->index_rotated = false; + } +end: + return ret; +} + +/* + * If too much data has been written in a tracefile before we received the + * rotation command, we have to move the excess data to the new tracefile and + * perform the rotation. This can happen because the control and data + * connections are separate, the indexes as well as the commands arrive from + * the control connection and we have no control over the order so we could be + * in a situation where too much data has been received on the data connection + * before the rotation command on the control connection arrives. + */ +static +int rotate_truncate_stream(struct relay_stream *stream) +{ + int ret, new_fd; + off_t lseek_ret; + uint64_t diff, pos = 0; + char buf[FILE_COPY_BUFFER_SIZE]; + + assert(!stream->is_metadata); + + assert(stream->tracefile_size_current > + stream->pos_after_last_complete_data_index); + diff = stream->tracefile_size_current - + stream->pos_after_last_complete_data_index; + + /* Create the new tracefile. */ + new_fd = utils_create_stream_file(stream->path_name, + stream->channel_name, + stream->tracefile_size, stream->tracefile_count, + /* uid */ -1, /* gid */ -1, /* suffix */ NULL); + if (new_fd < 0) { + ERR("Failed to create new stream file at path %s for channel %s", + stream->path_name, stream->channel_name); ret = -1; goto end; } - data_size = payload_size = be64toh(recv_hdr->data_size); - if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) { - ERR("Incorrect data size"); + /* + * Rewind the current tracefile to the position at which the rotation + * should have occured. + */ + lseek_ret = lseek(stream->stream_fd->fd, + stream->pos_after_last_complete_data_index, SEEK_SET); + if (lseek_ret < 0) { + PERROR("seek truncate stream"); ret = -1; goto end; } - payload_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; + /* Move data from the old file to the new file. */ + while (pos < diff) { + uint64_t count, bytes_left; + ssize_t io_ret; + + bytes_left = diff - pos; + count = bytes_left > sizeof(buf) ? sizeof(buf) : bytes_left; + assert(count <= SIZE_MAX); + + io_ret = lttng_read(stream->stream_fd->fd, buf, count); + if (io_ret < (ssize_t) count) { + char error_string[256]; + + snprintf(error_string, sizeof(error_string), + "Failed to read %" PRIu64 " bytes from fd %i in rotate_truncate_stream(), returned %zi", + count, stream->stream_fd->fd, io_ret); + if (io_ret == -1) { + PERROR("%s", error_string); + } else { + ERR("%s", error_string); + } + ret = -1; + goto end; + } + + io_ret = lttng_write(new_fd, buf, count); + if (io_ret < (ssize_t) count) { + char error_string[256]; - tmp_data_ptr = realloc(data_buffer, data_size); - if (!tmp_data_ptr) { - ERR("Allocating data buffer"); - free(data_buffer); + snprintf(error_string, sizeof(error_string), + "Failed to write %" PRIu64 " bytes from fd %i in rotate_truncate_stream(), returned %zi", + count, new_fd, io_ret); + if (io_ret == -1) { + PERROR("%s", error_string); + } else { + ERR("%s", error_string); + } ret = -1; goto end; } - data_buffer = tmp_data_ptr; - data_buffer_size = data_size; + + pos += count; } - 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"); + + /* Truncate the file to get rid of the excess data. */ + ret = ftruncate(stream->stream_fd->fd, + stream->pos_after_last_complete_data_index); + if (ret) { + PERROR("ftruncate"); + goto end; + } + + ret = close(stream->stream_fd->fd); + if (ret < 0) { + PERROR("Closing tracefile"); + goto end; + } + + /* + * Update the offset and FD of all the eventual indexes created by the + * data connection before the rotation command arrived. + */ + ret = relay_index_switch_all_files(stream); + if (ret < 0) { + ERR("Failed to rotate index file"); + goto end; + } + + stream->stream_fd->fd = new_fd; + stream->tracefile_size_current = diff; + stream->pos_after_last_complete_data_index = 0; + stream->rotate_at_seq_num = -1ULL; + + ret = 0; + +end: + return ret; +} + +/* + * Check if a stream's index file should be rotated (for session rotation). + * Must be called with the stream lock held. + * + * Return 0 on success, a negative value on error. + */ +static +int try_rotate_stream_index(struct relay_stream *stream) +{ + int ret = 0; + + if (stream->rotate_at_seq_num == -1ULL) { + /* No rotation expected. */ + goto end; + } + + if (stream->index_rotated) { + /* Rotation of the index has already occurred. */ + goto end; + } + + if (stream->prev_index_seq == -1ULL || + stream->prev_index_seq < stream->rotate_at_seq_num) { + DBG("Stream %" PRIu64 " index not yet ready for rotation (rotate_at_seq_num = %" PRIu64 ", prev_index_seq = %" PRIu64 ")", + stream->stream_handle, + stream->rotate_at_seq_num, + stream->prev_index_seq); + goto end; + } else if (stream->prev_index_seq != stream->rotate_at_seq_num) { + /* + * Unexpected, protocol error/bug. + * It could mean that we received a rotation position + * that is in the past. + */ + ERR("Stream %" PRIu64 " index is in an inconsistent state (rotate_at_seq_num = %" PRIu64 ", prev_data_seq = %" PRIu64 ", prev_index_seq = %" PRIu64 ")", + stream->stream_handle, + stream->rotate_at_seq_num, + stream->prev_data_seq, + stream->prev_index_seq); + ret = -1; + goto end; + } else { + DBG("Rotating stream %" PRIu64 " index file", + stream->stream_handle); + ret = create_rotate_index_file(stream, stream->path_name); + stream->index_rotated = true; + + if (stream->data_rotated && stream->index_rotated) { + /* Rotation completed; reset its state. */ + DBG("Rotation completed for stream %" PRIu64, + stream->stream_handle); + stream->rotate_at_seq_num = -1ULL; + stream->data_rotated = false; + stream->index_rotated = false; + } + } + +end: + return ret; +} + +/* + * Check if a stream's data file (as opposed to index) should be rotated + * (for session rotation). + * Must be called with the stream lock held. + * + * Return 0 on success, a negative value on error. + */ +static +int try_rotate_stream_data(struct relay_stream *stream) +{ + int ret = 0; + + if (stream->rotate_at_seq_num == -1ULL) { + /* No rotation expected. */ + goto end; + } + + if (stream->data_rotated) { + /* Rotation of the data file has already occurred. */ + goto end; + } + + if (stream->prev_data_seq == -1ULL || + stream->prev_data_seq < stream->rotate_at_seq_num) { + DBG("Stream %" PRIu64 " not yet ready for rotation (rotate_at_seq_num = %" PRIu64 ", prev_data_seq = %" PRIu64 ")", + stream->stream_handle, + stream->rotate_at_seq_num, + stream->prev_data_seq); + goto end; + } else if (stream->prev_data_seq > stream->rotate_at_seq_num) { + /* + * prev_data_seq is checked here since indexes and rotation + * commands are serialized with respect to each other. + */ + DBG("Rotation after too much data has been written in tracefile " + "for stream %" PRIu64 ", need to truncate before " + "rotating", stream->stream_handle); + ret = rotate_truncate_stream(stream); + if (ret) { + ERR("Failed to truncate stream"); + goto end; } + } else if (stream->prev_data_seq != stream->rotate_at_seq_num) { + /* + * Unexpected, protocol error/bug. + * It could mean that we received a rotation position + * that is in the past. + */ + ERR("Stream %" PRIu64 " data is in an inconsistent state (rotate_at_seq_num = %" PRIu64 ", prev_data_seq = %" PRIu64 ")", + stream->stream_handle, + stream->rotate_at_seq_num, + stream->prev_data_seq); ret = -1; goto end; + } else { + ret = do_rotate_stream_data(stream); } - 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)); +end: + return ret; +} + +/* + * relay_recv_metadata: receive the metadata for the session. + */ +static int relay_recv_metadata(const struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn, + const struct lttng_buffer_view *payload) +{ + int ret = 0; + ssize_t size_ret; + struct relay_session *session = conn->session; + struct lttcomm_relayd_metadata_payload metadata_payload_header; + struct relay_stream *metadata_stream; + uint64_t metadata_payload_size; + + if (!session) { + ERR("Metadata sent before version check"); + ret = -1; + goto end; + } + + if (recv_hdr->data_size < sizeof(struct lttcomm_relayd_metadata_payload)) { + ERR("Incorrect data size"); + ret = -1; + goto end; + } + metadata_payload_size = recv_hdr->data_size - + sizeof(struct lttcomm_relayd_metadata_payload); + + 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); + + metadata_stream = stream_get_by_id(metadata_payload_header.stream_id); if (!metadata_stream) { ret = -1; - goto end_unlock; + goto end; } - size_ret = lttng_write(metadata_stream->fd, metadata_struct->payload, - payload_size); - if (size_ret < payload_size) { + pthread_mutex_lock(&metadata_stream->lock); + + size_ret = lttng_write(metadata_stream->stream_fd->fd, + payload->data + sizeof(metadata_payload_header), + metadata_payload_size); + if (size_ret < metadata_payload_size) { ERR("Relay error writing metadata on file"); 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; + size_ret = write_padding_to_file(metadata_stream->stream_fd->fd, + metadata_payload_header.padding_size); + if (size_ret < (int64_t) metadata_payload_header.padding_size) { + ret = -1; + goto end_put; } - 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); + metadata_stream->metadata_received += + metadata_payload_size + metadata_payload_header.padding_size; + DBG2("Relay metadata written. Updated metadata_received %" PRIu64, + metadata_stream->metadata_received); - DBG2("Relay metadata written"); + ret = try_rotate_stream_data(metadata_stream); + if (ret < 0) { + goto end_put; + } -end_unlock: - rcu_read_unlock(); +end_put: + pthread_mutex_unlock(&metadata_stream->lock); + stream_put(metadata_stream); end: return ret; } @@ -1586,57 +1931,64 @@ 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, @@ -1649,54 +2001,61 @@ 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 { @@ -1704,17 +2063,19 @@ 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: @@ -1724,61 +2085,58 @@ 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: @@ -1786,62 +2144,63 @@ 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(); @@ -1849,9 +2208,13 @@ 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: @@ -1859,63 +2222,81 @@ 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(); @@ -1923,9 +2304,13 @@ 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: @@ -1937,113 +2322,126 @@ 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 relay_index *index; 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; + goto end; } + pthread_mutex_lock(&stream->lock); /* Live beacon handling */ if (index_info.packet_size == 0) { - DBG("Received live beacon for stream %" PRIu64, stream->stream_handle); + 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. + * 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); + if (stream->index_received_seqcount > 0 + && stream->indexes_in_flight == 0) { + stream->beacon_ts_end = index_info.timestamp_end; } ret = 0; - goto end_rcu_unlock; + goto end_stream_put; } else { stream->beacon_ts_end = -1ULL; } - index = relay_index_find(stream->stream_handle, net_seq_num); + if (stream->ctf_stream_id == -1ULL) { + stream->ctf_stream_id = index_info.stream_id; + } + index = relay_index_get_by_id_or_create(stream, index_info.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++; + ret = -1; + ERR("relay_index_get_by_id_or_create index NULL"); + goto end_stream_put; } - - copy_index_control_data(index, &index_info); - if (stream->ctf_stream_id == -1ULL) { - stream->ctf_stream_id = be64toh(index_info.stream_id); + if (set_index_control_data(index, &index_info, conn)) { + ERR("set_index_control_data error"); + relay_index_put(index); + ret = -1; + goto end_stream_put; } + ret = relay_index_try_flush(index); + if (ret == 0) { + tracefile_array_commit_seq(stream->tfa); + stream->index_received_seqcount++; + stream->pos_after_last_complete_data_index += index->total_size; + stream->prev_index_seq = index_info.net_seq_num; - 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); + ret = try_rotate_stream_index(stream); + if (ret < 0) { + goto end_stream_put; } + } else if (ret > 0) { + /* no flush. */ + ret = 0; } else { - /* The index already exists so write it on disk. */ - wr_index = index; + /* + * ret < 0 + * + * relay_index_try_flush is responsible for the self-reference + * put of the index object on error. + */ + ERR("relay_index_try_flush error %d", ret); + ret = -1; } - /* 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); - } +end_stream_put: + pthread_mutex_unlock(&stream->lock); + stream_put(stream); -end_rcu_unlock: - rcu_read_unlock(); +end: memset(&reply, 0, sizeof(reply)); if (ret < 0) { @@ -2052,9 +2450,9 @@ 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: @@ -2066,370 +2464,1243 @@ 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); + 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 < (ssize_t) sizeof(reply)) { + ERR("Failed to send \"streams sent\" command reply (ret = %zd)", + send_ret); + ret = -1; + } else { + /* Success. */ + ret = 0; + } + +end_no_session: + return ret; +} + +/* + * 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_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_rotate_stream stream_info; + struct lttcomm_relayd_generic_reply reply; + struct relay_stream *stream; + size_t header_len; + size_t path_len; + struct lttng_buffer_view new_path_view; + + DBG("Rotate stream received"); + + 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; + } + + header_len = sizeof(struct lttcomm_relayd_rotate_stream); + + 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(&stream_info, payload->data, header_len); + + /* Convert to host */ + stream_info.pathname_length = be32toh(stream_info.pathname_length); + stream_info.stream_id = be64toh(stream_info.stream_id); + stream_info.new_chunk_id = be64toh(stream_info.new_chunk_id); + stream_info.rotate_at_seq_num = be64toh(stream_info.rotate_at_seq_num); + + path_len = stream_info.pathname_length; + if (payload->size < header_len + path_len) { + ERR("Unexpected payload size in \"relay_rotate_session_stream\" including path: expected >= %zu bytes, got %zu bytes", + header_len + path_len, payload->size); + ret = -1; + goto end_no_reply; + } + + /* Ensure it fits in local filename length. */ + if (path_len >= LTTNG_PATH_MAX) { + ret = -ENAMETOOLONG; + ERR("Length of relay_rotate_session_stream command's path name (%zu bytes) exceeds the maximal allowed length of %i bytes", + path_len, LTTNG_PATH_MAX); + goto end; + } + + new_path_view = lttng_buffer_view_from_view(payload, header_len, + stream_info.pathname_length); + + stream = stream_get_by_id(stream_info.stream_id); + if (!stream) { + ret = -1; + goto end; + } + + pthread_mutex_lock(&stream->lock); /* - * Inform the viewer that there are new streams in the session. + * Update the trace path (just the folder, the stream name does not + * change). + */ + free(stream->prev_path_name); + stream->prev_path_name = stream->path_name; + stream->path_name = create_output_path(new_path_view.data); + if (!stream->path_name) { + ERR("Failed to create a new output path"); + ret = -1; + goto end_stream_unlock; + } + ret = utils_mkdir_recursive(stream->path_name, S_IRWXU | S_IRWXG, + -1, -1); + if (ret < 0) { + ERR("relay creating output directory"); + ret = -1; + goto end_stream_unlock; + } + + assert(stream->current_chunk_id.is_set); + stream->current_chunk_id.value = stream_info.new_chunk_id; + + if (stream->is_metadata) { + /* + * Metadata streams have no index; consider its rotation + * complete. + */ + stream->index_rotated = true; + /* + * The metadata stream is sent only over the control connection + * so we know we have all the data to perform the stream + * rotation. + */ + ret = do_rotate_stream_data(stream); + } else { + stream->rotate_at_seq_num = stream_info.rotate_at_seq_num; + ret = try_rotate_stream_data(stream); + if (ret < 0) { + goto end_stream_unlock; + } + + ret = try_rotate_stream_index(stream); + if (ret < 0) { + goto end_stream_unlock; + } + } + +end_stream_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); + } else { + reply.ret_code = htobe32(LTTNG_OK); + } + 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; + } + +end_no_reply: + return ret; +} + +/* + * relay_mkdir: Create a folder on the disk. + */ +static int relay_mkdir(const struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn, + const struct lttng_buffer_view *payload) +{ + int ret; + struct relay_session *session = conn->session; + struct lttcomm_relayd_mkdir path_info_header; + struct lttcomm_relayd_generic_reply reply; + char *path = NULL; + size_t header_len; + ssize_t send_ret; + struct lttng_buffer_view path_view; + + if (!session || !conn->version_check_done) { + ERR("Trying to create a directory before version check"); + ret = -1; + goto end_no_session; + } + + if (session->major == 2 && session->minor < 11) { + /* + * This client is not supposed to use this command since + * it predates its introduction. + */ + ERR("relay_mkdir command is unsupported before LTTng 2.11"); + ret = -1; + goto end_no_session; + } + + header_len = sizeof(path_info_header); + if (payload->size < header_len) { + ERR("Unexpected payload size in \"relay_mkdir\": expected >= %zu bytes, got %zu bytes", + header_len, payload->size); + ret = -1; + goto end_no_session; + } + + memcpy(&path_info_header, payload->data, header_len); + + path_info_header.length = be32toh(path_info_header.length); + + if (payload->size < header_len + path_info_header.length) { + ERR("Unexpected payload size in \"relay_mkdir\" including path: expected >= %zu bytes, got %zu bytes", + header_len + path_info_header.length, payload->size); + ret = -1; + goto end_no_session; + } + + /* Ensure that it fits in local path length. */ + if (path_info_header.length >= LTTNG_PATH_MAX) { + ret = -ENAMETOOLONG; + ERR("Path name argument of mkdir command (%" PRIu32 " bytes) exceeds the maximal length allowed (%d bytes)", + path_info_header.length, LTTNG_PATH_MAX); + goto end; + } + + path_view = lttng_buffer_view_from_view(payload, header_len, + path_info_header.length); + + path = create_output_path(path_view.data); + if (!path) { + ERR("Failed to create output path"); + ret = -1; + goto end; + } + + DBG("MKDIR command has path \"%s\", changed to \"%s\"", path_view.data, path); + ret = utils_mkdir_recursive(path, S_IRWXU | S_IRWXG, -1, -1); + if (ret < 0) { + ERR("relay creating output directory"); + goto end; + } + + ret = 0; + +end: + memset(&reply, 0, sizeof(reply)); + if (ret < 0) { + reply.ret_code = htobe32(LTTNG_ERR_UNK); + } else { + 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 \"mkdir\" command reply (ret = %zd)", send_ret); + ret = -1; + } + +end_no_session: + free(path); + return ret; +} + +static int validate_rotate_rename_path_length(const char *path_type, + uint32_t path_length) +{ + int ret = 0; + + if (path_length > LTTNG_PATH_MAX) { + ret = -ENAMETOOLONG; + ERR("rotate rename \"%s\" path name length (%" PRIu32 " bytes) exceeds the allowed size of %i bytes", + path_type, path_length, LTTNG_PATH_MAX); + } else if (path_length == 0) { + ret = -EINVAL; + ERR("rotate rename \"%s\" path name has an illegal length of 0", path_type); + } + return ret; +} + +/* + * relay_rotate_rename: rename the trace folder after a rotation is + * completed. We are not closing any fd here, just moving the folder, so it + * works even if data is still in-flight. + */ +static int relay_rotate_rename(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_generic_reply reply; + struct lttcomm_relayd_rotate_rename header; + size_t header_len; + size_t received_paths_size; + char *complete_old_path = NULL, *complete_new_path = NULL; + struct lttng_buffer_view old_path_view; + struct lttng_buffer_view new_path_view; + + if (!session || !conn->version_check_done) { + ERR("Trying to rename a trace folder before version check"); + ret = -1; + goto end_no_reply; + } + + if (session->major == 2 && session->minor < 11) { + ERR("relay_rotate_rename command is unsupported before LTTng 2.11"); + ret = -1; + goto end_no_reply; + } + + header_len = sizeof(header); + if (payload->size < header_len) { + ERR("Unexpected payload size in \"relay_rotate_rename\": expected >= %zu bytes, got %zu bytes", + header_len, payload->size); + ret = -1; + goto end_no_reply; + } + + memcpy(&header, payload->data, header_len); + + header.old_path_length = be32toh(header.old_path_length); + header.new_path_length = be32toh(header.new_path_length); + received_paths_size = header.old_path_length + header.new_path_length; + + if (payload->size < header_len + received_paths_size) { + ERR("Unexpected payload size in \"relay_rotate_rename\" including paths: expected >= %zu bytes, got %zu bytes", + header_len, payload->size); + ret = -1; + goto end_no_reply; + } + + /* Ensure the paths don't exceed their allowed size. */ + ret = validate_rotate_rename_path_length("old", header.old_path_length); + if (ret) { + goto end; + } + ret = validate_rotate_rename_path_length("new", header.new_path_length); + if (ret) { + goto end; + } + + old_path_view = lttng_buffer_view_from_view(payload, header_len, + header.old_path_length); + new_path_view = lttng_buffer_view_from_view(payload, + header_len + header.old_path_length, + header.new_path_length); + + /* Validate that both paths received are NULL terminated. */ + if (old_path_view.data[old_path_view.size - 1] != '\0') { + ERR("relay_rotate_rename command's \"old\" path is invalid (not NULL terminated)"); + ret = -1; + goto end; + } + if (new_path_view.data[new_path_view.size - 1] != '\0') { + ERR("relay_rotate_rename command's \"new\" path is invalid (not NULL terminated)"); + ret = -1; + goto end; + } + + DBG("ROTATE_RENAME command has argument old path = \"%s\", new_path = \"%s\"", + old_path_view.data, new_path_view.data); + complete_old_path = create_output_path(old_path_view.data); + if (!complete_old_path) { + ERR("Failed to build old output path in rotate_rename command"); + ret = -1; + goto end; + } + + complete_new_path = create_output_path(new_path_view.data); + if (!complete_new_path) { + ERR("Failed to build new output path in rotate_rename command"); + ret = -1; + goto end; + } + DBG("Expanded ROTATE_RENAME arguments to old path = \"%s\", new_path = \"%s\"", + complete_old_path, complete_new_path); + + ret = utils_mkdir_recursive(complete_new_path, S_IRWXU | S_IRWXG, + -1, -1); + if (ret < 0) { + ERR("Failed to mkdir() rotate_rename's \"new\" output directory at \"%s\"", + complete_new_path); + goto end; + } + + /* + * If a domain has not yet created its channel, the domain-specific + * folder might not exist, but this is not an error. + */ + ret = rename(complete_old_path, complete_new_path); + if (ret < 0 && errno != ENOENT) { + PERROR("Renaming chunk in rotate_rename command from \"%s\" to \"%s\"", + complete_old_path, complete_new_path); + goto end; + } + ret = 0; + +end: + memset(&reply, 0, sizeof(reply)); + if (ret < 0) { + reply.ret_code = htobe32(LTTNG_ERR_UNK); + } else { + reply.ret_code = htobe32(LTTNG_OK); + } + send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, + sizeof(reply), 0); + if (send_ret < sizeof(reply)) { + ERR("Failed to send \"rotate rename\" command reply (ret = %zd)", + send_ret); + ret = -1; + } + +end_no_reply: + free(complete_old_path); + free(complete_new_path); + return ret; +} + +/* + * Check if all the streams in the session have completed the last rotation. + * The chunk_id value is used to distinguish the cases where a stream was + * closed on the consumerd before the rotation started but it still active on + * the relayd, and the case where a stream appeared on the consumerd/relayd + * after the last rotation started (in that case, it is already writing in the + * new chunk folder). + */ +static +int relay_rotate_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_rotate_pending msg; + struct lttcomm_relayd_rotate_pending_reply reply; + struct lttng_ht_iter iter; + struct relay_stream *stream; + int ret = 0; + ssize_t send_ret; + uint64_t chunk_id; + bool rotate_pending = false; + + DBG("Rotate pending command received"); + + if (!session || !conn->version_check_done) { + ERR("Trying to check for data 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 < sizeof(msg)) { + ERR("Unexpected payload size in \"relay_rotate_pending\": expected >= %zu bytes, got %zu bytes", + sizeof(msg), payload->size); + ret = -1; + goto end_no_reply; + } + + memcpy(&msg, payload->data, sizeof(msg)); + + chunk_id = be64toh(msg.chunk_id); + + DBG("Evaluating rotate pending for session \"%s\" and chunk id %" PRIu64, + session->session_name, chunk_id); + + /* + * Iterate over all the streams in the session and check if they are + * still waiting for data to perform their rotation. + */ + rcu_read_lock(); + cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream, + node.node) { + if (!stream_get(stream)) { + continue; + } + if (stream->trace->session != session) { + stream_put(stream); + continue; + } + pthread_mutex_lock(&stream->lock); + if (stream->rotate_at_seq_num != -1ULL) { + /* We have not yet performed the rotation. */ + rotate_pending = true; + DBG("Stream %" PRIu64 " is still rotating", + stream->stream_handle); + } else if (stream->current_chunk_id.value < chunk_id) { + /* + * Stream closed on the consumer but still active on the + * relay. + */ + rotate_pending = true; + DBG("Stream %" PRIu64 " did not exist on the consumer " + "when the last rotation started, but is" + "still waiting for data before getting" + "closed", + stream->stream_handle); + } + pthread_mutex_unlock(&stream->lock); + stream_put(stream); + if (rotate_pending) { + goto send_reply; + } + } + +send_reply: + rcu_read_unlock(); + memset(&reply, 0, sizeof(reply)); + reply.generic.ret_code = htobe32((uint32_t) LTTNG_OK); + reply.is_pending = (uint8_t) !!rotate_pending; + send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, + sizeof(reply), 0); + if (send_ret < (ssize_t) sizeof(reply)) { + ERR("Failed to send \"rotate pending\" command reply (ret = %zd)", + send_ret); + ret = -1; + } + +end_no_reply: + 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 (header->cmd) { + case RELAYD_CREATE_SESSION: + DBG_CMD("RELAYD_CREATE_SESSION", conn); + ret = relay_create_session(header, conn, payload); + break; + case RELAYD_ADD_STREAM: + DBG_CMD("RELAYD_ADD_STREAM", conn); + ret = relay_add_stream(header, conn, payload); + break; + case RELAYD_START_DATA: + DBG_CMD("RELAYD_START_DATA", conn); + ret = relay_start(header, conn, payload); + break; + case RELAYD_SEND_METADATA: + DBG_CMD("RELAYD_SEND_METADATA", conn); + ret = relay_recv_metadata(header, conn, payload); + break; + case RELAYD_VERSION: + DBG_CMD("RELAYD_VERSION", conn); + ret = relay_send_version(header, conn, payload); + break; + case RELAYD_CLOSE_STREAM: + DBG_CMD("RELAYD_CLOSE_STREAM", conn); + ret = relay_close_stream(header, conn, payload); + break; + case RELAYD_DATA_PENDING: + DBG_CMD("RELAYD_DATA_PENDING", conn); + ret = relay_data_pending(header, conn, payload); + break; + case RELAYD_QUIESCENT_CONTROL: + DBG_CMD("RELAYD_QUIESCENT_CONTROL", conn); + ret = relay_quiescent_control(header, conn, payload); + break; + case RELAYD_BEGIN_DATA_PENDING: + DBG_CMD("RELAYD_BEGIN_DATA_PENDING", conn); + ret = relay_begin_data_pending(header, conn, payload); + break; + case RELAYD_END_DATA_PENDING: + DBG_CMD("RELAYD_END_DATA_PENDING", conn); + ret = relay_end_data_pending(header, conn, payload); + break; + case RELAYD_SEND_INDEX: + DBG_CMD("RELAYD_SEND_INDEX", conn); + ret = relay_recv_index(header, conn, payload); + break; + case RELAYD_STREAMS_SENT: + 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_STREAM: + DBG_CMD("RELAYD_ROTATE_STREAM", conn); + ret = relay_rotate_session_stream(header, conn, payload); + break; + case RELAYD_ROTATE_RENAME: + DBG_CMD("RELAYD_ROTATE_RENAME", conn); + ret = relay_rotate_rename(header, conn, payload); + break; + case RELAYD_ROTATE_PENDING: + DBG_CMD("RELAYD_ROTATE_PENDING", conn); + ret = relay_rotate_pending(header, conn, payload); + break; + case RELAYD_MKDIR: + DBG_CMD("RELAYD_MKDIR", conn); + ret = relay_mkdir(header, conn, payload); + break; + case RELAYD_UPDATE_SYNC_INFO: + default: + ERR("Received unknown command (%u)", header->cmd); + relay_unknown_command(conn); + ret = -1; + goto end; + } + +end: + return ret; +} + +static enum relay_connection_status relay_process_control_receive_payload( + struct relay_connection *conn) +{ + 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); + + 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 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); + /* + * 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. */ - if (conn->session->viewer_refcount) { - uatomic_set(&conn->session->new_streams, 1); + 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; + } + + 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; + } + 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); + + 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 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; + } + + /* 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; + } + + 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; } - 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; - } else { - /* Success. */ - ret = 0; + 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_no_session: - return ret; +end: + return status; } /* * Process the commands received on the control socket */ -static -int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr, +static enum relay_connection_status relay_process_control( struct relay_connection *conn) { - int ret = 0; + enum relay_connection_status status; - switch (be32toh(recv_hdr->cmd)) { - case RELAYD_CREATE_SESSION: - ret = relay_create_session(recv_hdr, conn); - break; - case RELAYD_ADD_STREAM: - ret = relay_add_stream(recv_hdr, conn); - break; - case RELAYD_START_DATA: - ret = relay_start(recv_hdr, conn); - break; - case RELAYD_SEND_METADATA: - ret = relay_recv_metadata(recv_hdr, conn); - break; - case RELAYD_VERSION: - ret = relay_send_version(recv_hdr, conn); - break; - case RELAYD_CLOSE_STREAM: - ret = relay_close_stream(recv_hdr, conn); - break; - case RELAYD_DATA_PENDING: - ret = relay_data_pending(recv_hdr, conn); - break; - case RELAYD_QUIESCENT_CONTROL: - ret = relay_quiescent_control(recv_hdr, conn); - break; - case RELAYD_BEGIN_DATA_PENDING: - ret = relay_begin_data_pending(recv_hdr, conn); + switch (conn->protocol.ctrl.state_id) { + case CTRL_CONNECTION_STATE_RECEIVE_HEADER: + status = relay_process_control_receive_header(conn); break; - case RELAYD_END_DATA_PENDING: - ret = relay_end_data_pending(recv_hdr, conn); - break; - case RELAYD_SEND_INDEX: - ret = relay_recv_index(recv_hdr, conn); - break; - case RELAYD_STREAMS_SENT: - ret = relay_streams_sent(recv_hdr, conn); + case CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD: + status = relay_process_control_receive_payload(conn); break; - case RELAYD_UPDATE_SYNC_INFO: default: - ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd)); - relay_unknown_command(conn); - ret = -1; - goto end; + ERR("Unknown control connection protocol state encountered."); + abort(); } -end: - return ret; + return status; } /* * Handle index for a data stream. * - * RCU read side lock MUST be acquired. + * Called with the stream lock held. * * 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) + bool rotate_index, bool *flushed, uint64_t total_size) { - int ret = 0, index_created = 0; - uint64_t stream_id, data_offset; - struct relay_index *index, *wr_index = NULL; - - assert(stream); + int ret = 0; + uint64_t data_offset; + struct relay_index *index; - stream_id = stream->stream_handle; /* Get data offset because we are about to update the index. */ data_offset = htobe64(stream->tracefile_size_current); + DBG("handle_index_data: stream %" PRIu64 " net_seq_num %" PRIu64 " data offset %" PRIu64, + stream->stream_handle, net_seq_num, stream->tracefile_size_current); + /* - * 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. + * Lookup for an existing index for that stream id/sequence + * number. If it exists, the control thread has already received the + * data for it, thus we need to write it to disk. */ - index = relay_index_find(stream_id, net_seq_num); + index = relay_index_get_by_id_or_create(stream, 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++; + ret = -1; + 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; - } - stream->index_fd = ret; - } - index->fd = stream->index_fd; - index->index_data.offset = data_offset; + if (rotate_index || !stream->index_file) { + const char *stream_path; - if (index_created) { /* - * 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. + * The data connection creates the stream's first index file. + * + * This can happen _after_ a ROTATE_STREAM command. In + * other words, the data of the first packet of this stream + * can be received after a ROTATE_STREAM command. + * + * The ROTATE_STREAM command changes the stream's path_name + * to point to the "next" chunk. If a rotation is pending for + * this stream, as indicated by "rotate_at_seq_num != -1ULL", + * it means that we are still receiving data that belongs in the + * stream's former path. + * + * In this very specific case, we must ensure that the index + * file is created in the streams's former path, + * "prev_path_name". + * + * All other rotations beyond the first one are not affected + * by this problem since the actual rotation operation creates + * the new chunk's index file. */ - 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; - } + stream_path = stream->rotate_at_seq_num == -1ULL ? + stream->path_name: + stream->prev_path_name; - /* Do we have a writable ready index to write on disk. */ - if (wr_index) { - ret = relay_index_write(wr_index->fd, wr_index); + ret = create_rotate_index_file(stream, stream_path); if (ret < 0) { - goto error; + ERR("Failed to rotate index"); + /* Put self-ref for this index due to error. */ + relay_index_put(index); + index = NULL; + goto end; } - stream->total_index_received++; - stream->indexes_in_flight--; - assert(stream->indexes_in_flight >= 0); } -error: + if (relay_index_set_file(index, stream->index_file, data_offset)) { + ret = -1; + /* Put self-ref for this index due to error. */ + relay_index_put(index); + index = NULL; + goto end; + } + + ret = relay_index_try_flush(index); + if (ret == 0) { + tracefile_array_commit_seq(stream->tfa); + stream->index_received_seqcount++; + *flushed = true; + } else if (ret > 0) { + index->total_size = total_size; + /* No flush. */ + ret = 0; + } else { + /* + * ret < 0 + * + * relay_index_try_flush is responsible for the self-reference + * put of the index object on error. + */ + ERR("relay_index_try_flush error %d", ret); + ret = -1; + } +end: return ret; } -/* - * relay_process_data: Process the data received on the data socket - */ -static -int relay_process_data(struct relay_connection *conn) +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; - } - memset(data_buffer, 0, 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; - net_seq_num = be64toh(data_hdr.net_seq_num); + 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); - 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); - } - ret = -1; - goto end_rcu_unlock; + 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); + /* Check if a rotation is needed. */ if (stream->tracefile_size > 0 && - (stream->tracefile_size_current + data_size) > + (stream->tracefile_size_current + header.data_size) > stream->tracefile_size) { - struct relay_viewer_stream *vstream; - uint64_t new_id; + uint64_t old_id, new_id; + + old_id = tracefile_array_get_file_index_head(stream->tfa); + tracefile_array_file_rotate(stream->tfa); + + /* new_id is updated by utils_rotate_stream_file. */ + new_id = old_id; + + ret = utils_rotate_stream_file(stream->path_name, + stream->channel_name, stream->tracefile_size, + stream->tracefile_count, -1, + -1, stream->stream_fd->fd, + &new_id, &stream->stream_fd->fd); + if (ret < 0) { + ERR("Failed to rotate stream output file"); + status = RELAY_CONNECTION_STATUS_ERROR; + goto end_stream_unlock; + } - new_id = (stream->tracefile_count_current + 1) % - stream->tracefile_count; /* - * When we wrap-around back to 0, we start overwriting old - * trace data. + * Reset current size because we just performed a stream + * rotation. */ - 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; + stream->tracefile_size_current = 0; + conn->protocol.data.state.receive_payload.rotate_index = true; + } + +end_stream_unlock: + pthread_mutex_unlock(&stream->lock); + 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; + } + + 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; } - vstream = viewer_stream_find_by_id(stream->stream_handle); - if (vstream) { + } + + /* + * 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) { + ssize_t write_ret; + size_t recv_size = min(left_to_receive, chunk_size); + + 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_%" PRIu64 "\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; } - 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; + + recv_size = ret; + + /* Write data to stream output fd. */ + write_ret = lttng_write(stream->stream_fd->fd, data_buffer, + recv_size); + if (write_ret < (ssize_t) recv_size) { + 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; + + DBG2("Relay wrote %zd bytes to tracefile for stream id %" PRIu64, + write_ret, stream->stream_handle); } - /* - * 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 = write_padding_to_file(stream->stream_fd->fd, + state->header.padding_size); + if ((int64_t) ret < (int64_t) state->header.padding_size) { + ERR("write_padding_to_file: fail 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; + } + + + if (session_streams_have_index(session)) { + ret = handle_index_data(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("handle_index_data: fail 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; + stream->tracefile_size_current += state->header.data_size + + state->header.padding_size; + + if (stream->prev_data_seq == -1ULL) { + new_stream = true; + } + if (index_flushed) { + stream->pos_after_last_complete_data_index = + stream->tracefile_size_current; + stream->prev_index_seq = state->header.net_seq_num; + ret = try_rotate_stream_index(stream); + if (ret < 0) { + goto end_stream_unlock; + } } - DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64, - ret, stream->stream_handle); + stream->prev_data_seq = state->header.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; - ret = write_padding_to_file(stream->fd, be32toh(data_hdr.padding_size)); + ret = try_rotate_stream_data(stream); if (ret < 0) { - goto end_rcu_unlock; + 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; +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(); + } - assert(events); + return status; +} + +static void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd) +{ + int ret; (void) lttng_poll_del(events, pollfd); @@ -2439,38 +3710,43 @@ 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); + const char *type_str; - connection_delete(relay_connections_ht, conn); - - /* 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"); @@ -2490,12 +3766,6 @@ 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; @@ -2530,9 +3800,9 @@ 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 */ @@ -2542,7 +3812,10 @@ restart: health_code_update(); if (!revents) { - /* No activity for this FD (poll implementation). */ + /* + * No activity for this FD (poll + * implementation). + */ continue; } @@ -2555,56 +3828,33 @@ 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 @@ -2612,10 +3862,55 @@ 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); } } @@ -2645,6 +3940,7 @@ 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(); @@ -2658,26 +3954,43 @@ restart: continue; } - rcu_read_lock(); - conn = connection_find_by_sock(relay_connections_ht, pollfd); - if (!conn) { + data_conn = connection_get_by_sock(relay_connections_ht, pollfd); + if (!data_conn) { /* Skip it. Might be removed before. */ - rcu_read_unlock(); continue; } + if (data_conn->type == RELAY_CONTROL) { + goto put_data_connection; + } + assert(data_conn->type == RELAY_DATA); if (revents & LPOLLIN) { - if (conn->type != RELAY_DATA) { - rcu_read_unlock(); - continue; - } + enum relay_connection_status status; - 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); + 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 @@ -2686,11 +3999,18 @@ restart: } else { /* Keep last seen port. */ last_seen_data_fd = pollfd; - rcu_read_unlock(); + connection_put(data_conn); goto restart; } + } 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); } - rcu_read_unlock(); + put_data_connection: + connection_put(data_conn); } last_seen_data_fd = -1; } @@ -2700,27 +4020,26 @@ 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 */ @@ -2729,7 +4048,6 @@ relay_connections_ht_error: DBG("Thread exited with error"); } DBG("Worker thread cleanup complete"); - free(data_buffer); error_testpoint: if (err) { health_error(); @@ -2761,7 +4079,6 @@ 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]; @@ -2783,7 +4100,8 @@ 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; @@ -2812,7 +4130,6 @@ int main(int argc, char **argv) } } - /* Initialize thread health monitoring */ health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES); if (!health_relayd) { @@ -2827,19 +4144,6 @@ 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; @@ -2849,26 +4153,13 @@ 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 */ - if (lttng_poll_set_max_size()) { - retval = -1; - goto exit_init_data; - } - /* 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; } @@ -2894,7 +4185,7 @@ 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; @@ -2904,7 +4195,7 @@ 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; @@ -2914,8 +4205,8 @@ 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"); @@ -2924,7 +4215,7 @@ 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; @@ -2933,7 +4224,7 @@ 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; @@ -2990,7 +4281,16 @@ exit_init_data: health_app_destroy(health_relayd); exit_health_app_create: 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);