X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-relayd%2Fmain.c;h=fa5a7db86c393640c15fe9e8b3bf5f652da98111;hp=fdddcbe6efcf54309078b3ad67917b0f46b59bbf;hb=6e7241fe92e7b5db872d164c3e1a0f75ce5a1463;hpb=1d4dfdefb6efe31cd6436931f893caf7b05d0223 diff --git a/src/bin/lttng-relayd/main.c b/src/bin/lttng-relayd/main.c index fdddcbe6e..fa5a7db86 100644 --- a/src/bin/lttng-relayd/main.c +++ b/src/bin/lttng-relayd/main.c @@ -1,6 +1,7 @@ /* * Copyright (C) 2012 - Julien Desfossez * David Goulet + * 2013 - Jérémie Galarneau * * 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, @@ -43,37 +44,65 @@ #include #include #include +#include #include +#include #include #include #include -#include #include #include #include +#include +#include "cmd.h" +#include "ctf-trace.h" +#include "index.h" +#include "utils.h" #include "lttng-relayd.h" +#include "live.h" +#include "health-relayd.h" +#include "testpoint.h" +#include "viewer-stream.h" +#include "session.h" +#include "stream.h" +#include "connection.h" /* command line options */ -static int opt_daemon; -static char *opt_output_path; +char *opt_output_path; +static int opt_daemon, opt_background; + +/* + * We need to wait for listener and live listener threads, as well as + * health check thread, before being ready to signal readiness. + */ +#define NR_LTTNG_RELAY_READY 3 +static int lttng_relay_ready = NR_LTTNG_RELAY_READY; +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. */ + static struct lttng_uri *control_uri; static struct lttng_uri *data_uri; +static struct lttng_uri *live_uri; const char *progname; -static int is_root; /* Set to 1 if the daemon is running as root */ + +const char *tracing_group_name = DEFAULT_TRACING_GROUP; +static int tracing_group_name_override; + +const char * const config_section_name = "relayd"; /* * Quit pipe for all threads. This permits a single cancellation point * for all threads when receiving an event on the pipe. */ -static int thread_quit_pipe[2] = { -1, -1 }; +int thread_quit_pipe[2] = { -1, -1 }; /* * This pipe is used to inform the worker thread that a command is queued and * ready to be processed. */ -static int relay_cmd_pipe[2] = { -1, -1 }; +static int relay_conn_pipe[2] = { -1, -1 }; /* Shared between threads */ static int dispatch_thread_exit; @@ -81,9 +110,9 @@ static int dispatch_thread_exit; static pthread_t listener_thread; static pthread_t dispatcher_thread; static pthread_t worker_thread; +static pthread_t health_thread; static uint64_t last_relay_stream_id; -static uint64_t last_relay_session_id; /* * Relay command queue. @@ -91,12 +120,44 @@ static uint64_t last_relay_session_id; * The relay_thread_listener and relay_thread_dispatcher communicate with this * queue. */ -static struct relay_cmd_queue relay_cmd_queue; +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; + +/* Relayd health monitoring */ +struct health_app *health_relayd; + +static struct option long_options[] = { + { "control-port", 1, 0, 'C', }, + { "data-port", 1, 0, 'D', }, + { "live-port", 1, 0, 'L', }, + { "daemonize", 0, 0, 'd', }, + { "background", 0, 0, 'b', }, + { "group", 1, 0, 'g', }, + { "help", 0, 0, 'h', }, + { "output", 1, 0, 'o', }, + { "verbose", 0, 0, 'v', }, + { "config", 1, 0, 'f' }, + { NULL, 0, 0, 0, }, +}; + +static const char *config_ignore_options[] = { "help", "config" }; + /* * usage function on stderr */ @@ -104,95 +165,230 @@ 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, " -C, --control-port Control port listening (URI)\n"); - fprintf(stderr, " -D, --data-port Data port listening (URI)\n"); - fprintf(stderr, " -o, --output Output path for traces (PATH)\n"); - fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n"); + 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"); } +/* + * Take an option from the getopt output and set it in the right variable to be + * used later. + * + * Return 0 on success else a negative value. + */ static -int parse_args(int argc, char **argv) +int set_option(int opt, const char *arg, const char *optname) { - int c; - int ret = 0; - char *default_address; - - static struct option long_options[] = { - { "control-port", 1, 0, 'C', }, - { "data-port", 1, 0, 'D', }, - { "daemonize", 0, 0, 'd', }, - { "help", 0, 0, 'h', }, - { "output", 1, 0, 'o', }, - { "verbose", 0, 0, 'v', }, - { NULL, 0, 0, 0, }, - }; + int ret; + + switch (opt) { + case 0: + fprintf(stderr, "option %s", optname); + if (arg) { + fprintf(stderr, " with arg %s\n", arg); + } + 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; + } + 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; + } + 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; + } + break; + case 'd': + opt_daemon = 1; + break; + case 'b': + opt_background = 1; + break; + case 'g': + tracing_group_name = strdup(arg); + tracing_group_name_override = 1; + break; + case 'h': + usage(); + exit(EXIT_FAILURE); + case 'o': + ret = asprintf(&opt_output_path, "%s", arg); + if (ret < 0) { + ret = -errno; + PERROR("asprintf opt_output_path"); + goto end; + } + break; + case 'v': + /* Verbose level can increase using multiple -v */ + if (arg) { + lttng_opt_verbose = config_parse_value(arg); + } else { + /* Only 3 level of verbosity (-vvv). */ + if (lttng_opt_verbose < 3) { + lttng_opt_verbose += 1; + } + } + break; + default: + /* Unknown option or other error. + * Error is printed by getopt, just return */ + ret = -1; + goto end; + } + + /* All good. */ + ret = 0; + +end: + return ret; +} + +/* + * 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 + * return value conventions. + */ +static +int config_entry_handler(const struct config_entry *entry, void *unused) +{ + int ret = 0, i; + if (!entry || !entry->name || !entry->value) { + ret = -EINVAL; + goto end; + } + + /* Check if the option is to be ignored */ + for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) { + if (!strcmp(entry->name, config_ignore_options[i])) { + goto end; + } + } + + for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1; i++) { + /* Ignore if entry name is not fully matched. */ + if (strcmp(entry->name, long_options[i].name)) { + continue; + } + + /* + * 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); + if (ret <= 0) { + if (ret) { + WARN("Invalid configuration value \"%s\" for option %s", + entry->value, entry->name); + } + /* False, skip boolean config option. */ + goto end; + } + } + + ret = set_option(long_options[i].val, entry->value, entry->name); + goto end; + } + + WARN("Unrecognized option \"%s\" in daemon configuration file.", + entry->name); + +end: + return ret; +} + +static +int set_options(int argc, char **argv) +{ + int c, ret = 0, option_index = 0; + int orig_optopt = optopt, orig_optind = optind; + char *default_address, *optstring; + const char *config_path = NULL; + + optstring = utils_generate_optstring(long_options, + sizeof(long_options) / sizeof(struct option)); + if (!optstring) { + ret = -ENOMEM; + goto exit; + } + + /* Check for the --config option */ + + while ((c = getopt_long(argc, argv, optstring, long_options, + &option_index)) != -1) { + if (c == '?') { + ret = -EINVAL; + goto exit; + } else if (c != 'f') { + continue; + } + + config_path = utils_expand_path(optarg); + if (!config_path) { + ERR("Failed to resolve path: %s", optarg); + } + } + + ret = config_get_section_entries(config_path, config_section_name, + config_entry_handler, NULL); + if (ret) { + if (ret > 0) { + ERR("Invalid configuration option at line %i", ret); + ret = -1; + } + goto exit; + } + + /* Reset getopt's global state */ + optopt = orig_optopt; + optind = orig_optind; while (1) { - int option_index = 0; - c = getopt_long(argc, argv, "dhv" "C:D:o:", - long_options, &option_index); + c = getopt_long(argc, argv, optstring, long_options, &option_index); if (c == -1) { break; } - switch (c) { - case 0: - fprintf(stderr, "option %s", long_options[option_index].name); - if (optarg) { - fprintf(stderr, " with arg %s\n", optarg); - } - break; - case 'C': - ret = uri_parse(optarg, &control_uri); - if (ret < 0) { - ERR("Invalid control URI specified"); - goto exit; - } - if (control_uri->port == 0) { - control_uri->port = DEFAULT_NETWORK_CONTROL_PORT; - } - break; - case 'D': - ret = uri_parse(optarg, &data_uri); - if (ret < 0) { - ERR("Invalid data URI specified"); - goto exit; - } - if (data_uri->port == 0) { - data_uri->port = DEFAULT_NETWORK_DATA_PORT; - } - break; - case 'd': - opt_daemon = 1; - break; - case 'h': - usage(); - exit(EXIT_FAILURE); - case 'o': - ret = asprintf(&opt_output_path, "%s", optarg); - if (ret < 0) { - PERROR("asprintf opt_output_path"); - goto exit; - } - break; - case 'v': - /* Verbose level can increase using multiple -v */ - lttng_opt_verbose += 1; - break; - default: - /* Unknown option or other error. - * Error is printed by getopt, just return */ - ret = -1; + ret = set_option(c, optarg, long_options[option_index].name); + if (ret < 0) { goto exit; } } /* assign default values */ if (control_uri == NULL) { - ret = asprintf(&default_address, "tcp://0.0.0.0:%d", - DEFAULT_NETWORK_CONTROL_PORT); + ret = asprintf(&default_address, + "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS ":%d", + DEFAULT_NETWORK_CONTROL_PORT); if (ret < 0) { PERROR("asprintf default data address"); goto exit; @@ -206,8 +402,9 @@ int parse_args(int argc, char **argv) } } if (data_uri == NULL) { - ret = asprintf(&default_address, "tcp://0.0.0.0:%d", - DEFAULT_NETWORK_DATA_PORT); + ret = asprintf(&default_address, + "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS ":%d", + DEFAULT_NETWORK_DATA_PORT); if (ret < 0) { PERROR("asprintf default data address"); goto exit; @@ -220,8 +417,25 @@ int parse_args(int argc, char **argv) goto exit; } } + if (live_uri == NULL) { + ret = asprintf(&default_address, + "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS ":%d", + DEFAULT_NETWORK_VIEWER_PORT); + if (ret < 0) { + PERROR("asprintf default viewer control address"); + goto exit; + } + + ret = uri_parse(default_address, &live_uri); + free(default_address); + if (ret < 0) { + ERR("Invalid viewer control URI specified"); + goto exit; + } + } exit: + free(optstring); return ret; } @@ -239,8 +453,13 @@ void cleanup(void) /* Close thread quit pipes */ utils_close_pipe(thread_quit_pipe); - /* Close relay cmd pipes */ - utils_close_pipe(relay_cmd_pipe); + uri_free(control_uri); + uri_free(data_uri); + /* Live URI is freed in the live thread. */ + + if (tracing_group_name_override) { + free((void *) tracing_group_name); + } } /* @@ -249,18 +468,26 @@ void cleanup(void) static int notify_thread_pipe(int wpipe) { - int ret; + ssize_t ret; - do { - ret = write(wpipe, "!", 1); - } while (ret < 0 && errno == EINTR); - if (ret < 0) { + ret = lttng_write(wpipe, "!", 1); + if (ret < 1) { PERROR("write poll pipe"); } return ret; } +static void notify_health_quit_pipe(int *pipe) +{ + ssize_t ret; + + ret = lttng_write(pipe[1], "4", 1); + if (ret < 1) { + PERROR("write relay health quit"); + } +} + /* * Stop all threads by closing the thread quit pipe. */ @@ -276,9 +503,11 @@ void stop_threads(void) ERR("write error on thread quit pipe"); } + notify_health_quit_pipe(health_quit_pipe); + /* Dispatch thread */ CMM_STORE_SHARED(dispatch_thread_exit, 1); - futex_nto1_wake(&relay_cmd_queue.futex); + futex_nto1_wake(&relay_conn_queue.futex); } /* @@ -302,6 +531,9 @@ void sighandler(int sig) DBG("SIGTERM caught"); stop_threads(); break; + case SIGUSR1: + CMM_STORE_SHARED(recv_child_signal, 1); + break; default: break; } @@ -341,11 +573,26 @@ int set_signal_handler(void) return ret; } - DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT"); + if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) { + PERROR("sigaction"); + return ret; + } + + DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT"); return ret; } +void lttng_relay_notify_ready(void) +{ + /* Notify the parent of the fork() process that we are ready. */ + if (opt_daemon || opt_background) { + if (uatomic_sub_return(<tng_relay_ready, 1) == 0) { + kill(child_ppid, SIGUSR1); + } + } +} + /* * Init thread quit pipe. * @@ -380,7 +627,7 @@ int create_thread_poll_set(struct lttng_poll_event *events, int size) } /* Add quit pipe */ - ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN); + ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR); if (ret < 0) { goto error; } @@ -453,13 +700,53 @@ error: 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 */ @@ -467,19 +754,16 @@ static void *relay_thread_listener(void *data) { int i, ret, pollfd, err = -1; - int val = 1; uint32_t revents, nb_fd; struct lttng_poll_event events; struct lttcomm_sock *control_sock, *data_sock; - /* - * Get allocated in this thread, enqueued to a global queue, dequeued and - * freed in the worker thread. - */ - struct relay_command *relay_cmd = NULL; - DBG("[thread] Relay listener started"); + health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER); + + health_code_update(); + control_sock = relay_init_sock(control_uri); if (!control_sock) { goto error_sock_control; @@ -510,13 +794,21 @@ void *relay_thread_listener(void *data) goto error_poll_add; } + lttng_relay_notify_ready(); + + if (testpoint(relayd_thread_listener)) { + goto error_testpoint; + } + while (1) { - DBG("Listener accepting connections"); + health_code_update(); - nb_fd = LTTNG_POLL_GETNB(&events); + DBG("Listener accepting connections"); restart: + health_poll_entry(); ret = lttng_poll_wait(&events, -1); + health_poll_exit(); if (ret < 0) { /* * Restart interrupted system call. @@ -527,8 +819,12 @@ restart: goto error; } + nb_fd = ret; + DBG("Relay new connection received"); for (i = 0; i < nb_fd; i++) { + health_code_update(); + /* Fetch once the poll data */ revents = LTTNG_POLL_GETEV(&events, i); pollfd = LTTNG_POLL_GETFD(&events, i); @@ -544,48 +840,56 @@ restart: ERR("socket poll error"); goto error; } else if (revents & LPOLLIN) { - struct lttcomm_sock *newsock = NULL; + /* + * Get allocated in this thread, enqueued to a global queue, + * dequeued and freed in the worker thread. + */ + int val = 1; + struct relay_connection *new_conn; + struct lttcomm_sock *newsock; - relay_cmd = zmalloc(sizeof(struct relay_command)); - if (relay_cmd == NULL) { - PERROR("relay command zmalloc"); + new_conn = connection_create(); + if (!new_conn) { goto error; } if (pollfd == data_sock->fd) { + new_conn->type = RELAY_DATA; newsock = data_sock->ops->accept(data_sock); - if (newsock < 0) { - PERROR("accepting data sock"); - goto error; - } - relay_cmd->type = RELAY_DATA; - DBG("Relay data connection accepted, socket %d", newsock->fd); - } else if (pollfd == control_sock->fd) { + DBG("Relay data connection accepted, socket %d", + newsock->fd); + } else { + assert(pollfd == control_sock->fd); + new_conn->type = RELAY_CONTROL; newsock = control_sock->ops->accept(control_sock); - if (newsock < 0) { - PERROR("accepting control sock"); - goto error; - } - relay_cmd->type = RELAY_CONTROL; - DBG("Relay control connection accepted, socket %d", newsock->fd); + DBG("Relay control connection accepted, socket %d", + newsock->fd); } - ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, - &val, sizeof(int)); + if (!newsock) { + PERROR("accepting sock"); + connection_free(new_conn); + goto error; + } + + ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val, + sizeof(val)); if (ret < 0) { PERROR("setsockopt inet"); + lttcomm_destroy_sock(newsock); + connection_free(new_conn); goto error; } - relay_cmd->sock = newsock; - /* - * Lock free enqueue the request. - */ - cds_wfq_enqueue(&relay_cmd_queue.queue, &relay_cmd->node); + new_conn->sock = newsock; + + /* 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_wfq_enqueue. + * Wake the dispatch queue futex. Implicit memory barrier with + * the exchange in cds_wfcq_enqueue. */ - futex_nto1_wake(&relay_cmd_queue.futex); + futex_nto1_wake(&relay_conn_queue.futex); } } } @@ -593,6 +897,7 @@ restart: exit: error: error_poll_add: +error_testpoint: lttng_poll_clean(&events); error_create_poll: if (data_sock->fd >= 0) { @@ -612,8 +917,10 @@ error_sock_relay: lttcomm_destroy_sock(control_sock); error_sock_control: if (err) { - DBG("Thread exited with error"); + health_error(); + ERR("Health error occurred in %s", __func__); } + health_unregister(health_relayd); DBG("Relay listener thread cleanup complete"); stop_threads(); return NULL; @@ -625,278 +932,234 @@ error_sock_control: static void *relay_thread_dispatcher(void *data) { - int ret; - struct cds_wfq_node *node; - struct relay_command *relay_cmd = NULL; + int err = -1; + ssize_t ret; + struct cds_wfcq_node *node; + struct relay_connection *new_conn = NULL; DBG("[thread] Relay dispatcher started"); + health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER); + + if (testpoint(relayd_thread_dispatcher)) { + goto error_testpoint; + } + + health_code_update(); + while (!CMM_LOAD_SHARED(dispatch_thread_exit)) { + health_code_update(); + /* Atomically prepare the queue futex */ - futex_nto1_prepare(&relay_cmd_queue.futex); + futex_nto1_prepare(&relay_conn_queue.futex); do { + health_code_update(); + /* Dequeue commands */ - node = cds_wfq_dequeue_blocking(&relay_cmd_queue.queue); + node = cds_wfcq_dequeue_blocking(&relay_conn_queue.head, + &relay_conn_queue.tail); if (node == NULL) { DBG("Woken up but nothing in the relay command queue"); /* Continue thread execution */ break; } + new_conn = caa_container_of(node, struct relay_connection, qnode); - relay_cmd = caa_container_of(node, struct relay_command, node); - DBG("Dispatching request waiting on sock %d", relay_cmd->sock->fd); + 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 :) */ - do { - ret = write(relay_cmd_pipe[1], relay_cmd, - sizeof(struct relay_command)); - } while (ret < 0 && errno == EINTR); - free(relay_cmd); + ret = lttng_write(relay_conn_pipe[1], &new_conn, sizeof(new_conn)); if (ret < 0) { - PERROR("write cmd pipe"); + PERROR("write connection pipe"); + connection_destroy(new_conn); goto error; } } while (node != NULL); /* Futex wait on queue. Blocking call on futex() */ - futex_nto1_wait(&relay_cmd_queue.futex); + health_poll_entry(); + futex_nto1_wait(&relay_conn_queue.futex); + health_poll_exit(); } + /* Normal exit, no error */ + err = 0; + error: +error_testpoint: + if (err) { + health_error(); + ERR("Health error occurred in %s", __func__); + } + health_unregister(health_relayd); DBG("Dispatch thread dying"); stop_threads(); return NULL; } -/* - * Return the realpath(3) of the path even if the last directory token does not - * exist. For example, with /tmp/test1/test2, if test2/ does not exist but the - * /tmp/test1 does, the real path is returned. In normal time, realpath(3) - * fails if the end point directory does not exist. - */ -static -char *expand_full_path(const char *path) +static void try_close_streams(struct relay_session *session) { - const char *end_path = path; - char *next, *cut_path, *expanded_path, *respath; - - /* Find last token delimited by '/' */ - while ((next = strpbrk(end_path + 1, "/"))) { - end_path = next; - } + struct ctf_trace *ctf_trace; + struct lttng_ht_iter iter; - /* Cut last token from original path */ - cut_path = strndup(path, end_path - path); + assert(session); - expanded_path = malloc(PATH_MAX); - if (expanded_path == NULL) { - respath = NULL; - goto end; - } + 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; - respath = realpath(cut_path, expanded_path); - if (respath == NULL) { - switch (errno) { - case ENOENT: - ERR("%s: No such file or directory", cut_path); - break; - default: - PERROR("realpath"); - break; + /* Close streams. */ + cds_list_for_each_entry(stream, &ctf_trace->stream_list, trace_list) { + stream_close(session, stream); } - free(expanded_path); - } else { - /* Add end part to expanded path */ - strcat(respath, end_path); + + ctf_trace->invalid_flag = 1; + ctf_trace_try_destroy(session, ctf_trace); } -end: - free(cut_path); - return respath; + rcu_read_unlock(); + pthread_mutex_unlock(&session->viewer_ready_lock); } +/* + * Try to destroy a session within a connection. + */ +static void destroy_session(struct relay_session *session, + struct lttng_ht *sessions_ht) +{ + assert(session); + assert(sessions_ht); + + /* Indicate that this session can be destroyed from now on. */ + session->close_flag = 1; + + try_close_streams(session); + + /* + * This will try to delete and destroy the session if no viewer is attached + * to it meaning the refcount is down to zero. + */ + session_try_destroy(sessions_ht, session); +} /* - * config_get_default_path - * - * Returns the HOME directory path. Caller MUST NOT free(3) the return pointer. + * Copy index data from the control port to a given index object. */ -static -char *config_get_default_path(void) +static void copy_index_control_data(struct relay_index *index, + struct lttcomm_relayd_index *data) { - return getenv("HOME"); + 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; } /* - * Create recursively directory using the FULL path. + * Handle the RELAYD_CREATE_SESSION command. + * + * On success, send back the session id or else return a negative value. */ static -int mkdir_recursive(char *path, mode_t mode) +int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn) { - char *p, tmp[PATH_MAX]; - struct stat statbuf; - size_t len; - int ret; + int ret = 0, send_ret; + struct relay_session *session; + struct lttcomm_relayd_status_session reply; - ret = snprintf(tmp, sizeof(tmp), "%s", path); - if (ret < 0) { - PERROR("snprintf mkdir"); + assert(recv_hdr); + assert(conn); + + memset(&reply, 0, sizeof(reply)); + + session = session_create(); + if (!session) { + ret = -1; goto error; } + session->minor = conn->minor; + session->major = conn->major; + conn->session_id = session->id; + conn->session = session; - len = ret; - if (tmp[len - 1] == '/') { - tmp[len - 1] = 0; - } + reply.session_id = htobe64(session->id); - for (p = tmp + 1; *p; p++) { - if (*p == '/') { - *p = 0; - if (tmp[strlen(tmp) - 1] == '.' && - tmp[strlen(tmp) - 2] == '.' && - tmp[strlen(tmp) - 3] == '/') { - ERR("Using '/../' is not permitted in the trace path (%s)", - tmp); - ret = -1; - goto error; - } - ret = stat(tmp, &statbuf); - if (ret < 0) { - ret = mkdir(tmp, mode); - if (ret < 0) { - if (errno != EEXIST) { - PERROR("mkdir recursive"); - ret = -errno; - goto error; - } - } - } - *p = '/'; - } + 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); } - ret = mkdir(tmp, mode); + lttng_ht_add_unique_u64(conn->sessions_ht, &session->session_n); + DBG("Created session %" PRIu64, session->id); + +error: if (ret < 0) { - if (errno != EEXIST) { - PERROR("mkdir recursive last piece"); - ret = -errno; - } else { - ret = 0; - } + reply.ret_code = htobe32(LTTNG_ERR_FATAL); + } else { + reply.ret_code = htobe32(LTTNG_OK); } -error: - return ret; -} - -static -char *create_output_path_auto(char *path_name) -{ - int ret; - char *traces_path = NULL; - char *alloc_path = NULL; - char *default_path; - - default_path = config_get_default_path(); - if (default_path == NULL) { - ERR("Home path not found.\n \ - Please specify an output path using -o, --output PATH"); - goto exit; - } - alloc_path = strdup(default_path); - if (alloc_path == NULL) { - PERROR("Path allocation"); - goto exit; - } - ret = asprintf(&traces_path, "%s/" DEFAULT_TRACE_DIR_NAME - "/%s", alloc_path, path_name); - if (ret < 0) { - PERROR("asprintf trace dir name"); - goto exit; + send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); + if (send_ret < 0) { + ERR("Relayd sending session id"); + ret = send_ret; } -exit: - free(alloc_path); - return traces_path; -} - -static -char *create_output_path_noauto(char *path_name) -{ - int ret; - char *traces_path = NULL; - char *full_path; - full_path = expand_full_path(opt_output_path); - ret = asprintf(&traces_path, "%s/%s", full_path, path_name); - if (ret < 0) { - PERROR("asprintf trace dir name"); - goto exit; - } -exit: - free(full_path); - return traces_path; + return ret; } /* - * create_output_path: create the output trace directory + * When we have received all the streams and the metadata for a channel, + * we make them visible to the viewer threads. */ static -char *create_output_path(char *path_name) +void set_viewer_ready_flag(struct relay_connection *conn) { - if (opt_output_path == NULL) { - return create_output_path_auto(path_name); - } else { - return create_output_path_noauto(path_name); - } -} + struct relay_stream *stream, *tmp_stream; -static -void deferred_free_stream(struct rcu_head *head) -{ - struct relay_stream *stream = - caa_container_of(head, struct relay_stream, rcu_node); - free(stream); + 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); + } + pthread_mutex_unlock(&conn->session->viewer_ready_lock); + return; } /* - * relay_delete_session: Free all memory associated with a session and - * close all the FDs + * 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 relay_delete_session(struct relay_command *cmd, struct lttng_ht *streams_ht) +static void queue_stream(struct relay_stream *stream, + struct relay_connection *conn) { - struct lttng_ht_iter iter; - struct lttng_ht_node_ulong *node; - struct relay_stream *stream; - int ret; + assert(conn); + assert(stream); - if (!cmd->session) { - return; - } - - DBG("Relay deleting session %" PRIu64, cmd->session->id); - free(cmd->session->sock); - - rcu_read_lock(); - cds_lfht_for_each_entry(streams_ht->ht, &iter.iter, node, node) { - node = lttng_ht_iter_get_node_ulong(&iter); - if (node) { - stream = caa_container_of(node, - struct relay_stream, stream_n); - if (stream->session == cmd->session) { - close(stream->fd); - ret = lttng_ht_del(streams_ht, &iter); - assert(!ret); - call_rcu(&stream->rcu_node, - deferred_free_stream); - } - } - } - rcu_read_unlock(); + cds_list_add(&stream->recv_list, &conn->recv_head); } /* @@ -904,29 +1167,20 @@ void relay_delete_session(struct relay_command *cmd, struct lttng_ht *streams_ht */ static int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_command *cmd, struct lttng_ht *streams_ht) + struct relay_connection *conn) { - struct relay_session *session = cmd->session; - struct lttcomm_relayd_add_stream stream_info; + int ret, send_ret; + struct relay_session *session = conn->session; struct relay_stream *stream = NULL; struct lttcomm_relayd_status_stream reply; - char *path = NULL, *root_path = NULL; - int ret, send_ret; + struct ctf_trace *trace; - if (!session || session->version_check_done == 0) { + if (!session || conn->version_check_done == 0) { ERR("Trying to add a stream before version check"); ret = -1; goto end_no_session; } - /* FIXME : use data_size for something ? */ - ret = cmd->sock->ops->recvmsg(cmd->sock, &stream_info, - sizeof(struct lttcomm_relayd_add_stream), MSG_WAITALL); - if (ret < sizeof(struct lttcomm_relayd_add_stream)) { - ERR("Relay didn't receive valid add_stream struct size : %d", ret); - ret = -1; - goto end_no_session; - } stream = zmalloc(sizeof(struct relay_stream)); if (stream == NULL) { PERROR("relay stream zmalloc"); @@ -934,63 +1188,116 @@ int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr, 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 = session; - - root_path = create_output_path(stream_info.pathname); - if (!root_path) { - ret = -1; - goto end; - } - ret = mkdir_recursive(root_path, S_IRWXU | S_IRWXG); + 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; } - ret = asprintf(&path, "%s/%s", root_path, stream_info.channel_name); + /* + * 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) { - PERROR("asprintf stream path"); + 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); + } - ret = open(path, O_WRONLY|O_CREAT|O_TRUNC, S_IRWXU|S_IRWXG|S_IRWXO); - if (ret < 0) { - PERROR("Relay creating trace file"); - goto end; + trace = ctf_trace_find_by_path(session->ctf_traces_ht, stream->path_name); + if (!trace) { + trace = ctf_trace_create(stream->path_name); + if (!trace) { + ret = -1; + goto end; + } + ctf_trace_add(session->ctf_traces_ht, trace); } + ctf_trace_get_ref(trace); - stream->fd = ret; - DBG("Tracefile %s created", path); + 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; + } + + /* + * 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); - lttng_ht_node_init_ulong(&stream->stream_n, - (unsigned long) stream->stream_handle); - lttng_ht_add_unique_ulong(streams_ht, - &stream->stream_n); + /* + * 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. + */ + lttng_ht_add_unique_u64(relay_streams_ht, &stream->node); + cds_list_add_tail(&stream->trace_list, &trace->stream_list); - DBG("Relay new stream added %s", stream_info.channel_name); + session->stream_count++; + + DBG("Relay new stream added %s with ID %" PRIu64, stream->channel_name, + stream->stream_handle); end: - free(path); - free(root_path); + memset(&reply, 0, sizeof(reply)); + reply.handle = htobe64(stream->stream_handle); /* send the session id to the client or a negative return code on error */ if (ret < 0) { reply.ret_code = htobe32(LTTNG_ERR_UNK); + /* stream was not properly added to the ht, so free it */ + free(stream); } else { reply.ret_code = htobe32(LTTNG_OK); } - reply.handle = htobe64(stream->stream_handle); - send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, + + 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; } rcu_read_unlock(); end_no_session: return ret; + +err_free_stream: + free(stream->path_name); + free(stream->channel_name); + free(stream); + return ret; } /* @@ -998,74 +1305,65 @@ end_no_session: */ static int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_command *cmd, struct lttng_ht *streams_ht) + struct relay_connection *conn) { - struct relay_session *session = cmd->session; + int ret, send_ret; + struct relay_session *session = conn->session; struct lttcomm_relayd_close_stream stream_info; struct lttcomm_relayd_generic_reply reply; struct relay_stream *stream; - int ret, send_ret; - struct lttng_ht_node_ulong *node; - struct lttng_ht_iter iter; DBG("Close stream received"); - if (!session || session->version_check_done == 0) { + if (!session || conn->version_check_done == 0) { ERR("Trying to close a stream before version check"); ret = -1; goto end_no_session; } - ret = cmd->sock->ops->recvmsg(cmd->sock, &stream_info, - sizeof(struct lttcomm_relayd_close_stream), MSG_WAITALL); + ret = conn->sock->ops->recvmsg(conn->sock, &stream_info, + sizeof(struct lttcomm_relayd_close_stream), 0); if (ret < sizeof(struct lttcomm_relayd_close_stream)) { - ERR("Relay didn't receive valid add_stream struct size : %d", ret); + 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); + } ret = -1; goto end_no_session; } rcu_read_lock(); - lttng_ht_lookup(streams_ht, - (void *)((unsigned long) be64toh(stream_info.stream_id)), - &iter); - node = lttng_ht_iter_get_node_ulong(&iter); - if (node == NULL) { - DBG("Relay stream %" PRIu64 " not found", be64toh(stream_info.stream_id)); - ret = -1; - goto end_unlock; - } - - stream = caa_container_of(node, struct relay_stream, stream_n); + stream = stream_find_by_id(relay_streams_ht, + be64toh(stream_info.stream_id)); if (!stream) { ret = -1; goto end_unlock; } + stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num); stream->close_flag = 1; + session->stream_count--; + assert(session->stream_count >= 0); - if (close_stream_check(stream)) { - int delret; - - close(stream->fd); - delret = lttng_ht_del(streams_ht, &iter); - assert(!delret); - call_rcu(&stream->rcu_node, - deferred_free_stream); - DBG("Closed tracefile %d from close stream", stream->fd); - } + /* Check if we can close it or else the data will do it. */ + try_close_stream(session, stream); end_unlock: rcu_read_unlock(); + memset(&reply, 0, sizeof(reply)); if (ret < 0) { reply.ret_code = htobe32(LTTNG_ERR_UNK); } else { reply.ret_code = htobe32(LTTNG_OK); } - send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, + 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; } end_no_session: @@ -1076,13 +1374,14 @@ end_no_session: * relay_unknown_command: send -1 if received unknown command */ static -void relay_unknown_command(struct relay_command *cmd) +void relay_unknown_command(struct relay_connection *conn) { struct lttcomm_relayd_generic_reply reply; int ret; + memset(&reply, 0, sizeof(reply)); reply.ret_code = htobe32(LTTNG_ERR_UNK); - ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, + ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(struct lttcomm_relayd_generic_reply), 0); if (ret < 0) { ERR("Relay sending unknown command"); @@ -1095,19 +1394,20 @@ void relay_unknown_command(struct relay_command *cmd) */ static int relay_start(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_command *cmd) + struct relay_connection *conn) { int ret = htobe32(LTTNG_OK); struct lttcomm_relayd_generic_reply reply; - struct relay_session *session = cmd->session; + struct relay_session *session = conn->session; if (!session) { DBG("Trying to start the streaming without a session established"); ret = htobe32(LTTNG_ERR_UNK); } + memset(&reply, 0, sizeof(reply)); reply.ret_code = ret; - ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, + ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(struct lttcomm_relayd_generic_reply), 0); if (ret < 0) { ERR("Relay sending start ack"); @@ -1116,40 +1416,12 @@ int relay_start(struct lttcomm_relayd_hdr *recv_hdr, return ret; } -/* - * Get stream from stream id. - * Need to be called with RCU read-side lock held. - */ -static -struct relay_stream *relay_stream_from_stream_id(uint64_t stream_id, - struct lttng_ht *streams_ht) -{ - struct lttng_ht_node_ulong *node; - struct lttng_ht_iter iter; - struct relay_stream *ret; - - lttng_ht_lookup(streams_ht, - (void *)((unsigned long) stream_id), - &iter); - node = lttng_ht_iter_get_node_ulong(&iter); - if (node == NULL) { - DBG("Relay stream %" PRIu64 " not found", stream_id); - ret = NULL; - goto end; - } - - ret = caa_container_of(node, struct relay_stream, stream_n); - -end: - return ret; -} - /* * Append padding to the file pointed by the file descriptor fd. */ static int write_padding_to_file(int fd, uint32_t size) { - int ret = 0; + ssize_t ret = 0; char *zeros; if (size == 0) { @@ -1163,13 +1435,13 @@ static int write_padding_to_file(int fd, uint32_t size) goto end; } - do { - ret = write(fd, zeros, size); - } while (ret < 0 && errno == EINTR); - if (ret < 0) { + ret = lttng_write(fd, zeros, size); + if (ret < size) { PERROR("write padding to file"); } + free(zeros); + end: return ret; } @@ -1179,13 +1451,15 @@ end: */ static int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_command *cmd, struct lttng_ht *streams_ht) + struct relay_connection *conn) { int ret = htobe32(LTTNG_OK); - struct relay_session *session = cmd->session; + 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; if (!session) { ERR("Metadata sent before version check"); @@ -1202,38 +1476,45 @@ int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr, payload_size -= sizeof(struct lttcomm_relayd_metadata_payload); if (data_buffer_size < data_size) { - data_buffer = realloc(data_buffer, data_size); - if (!data_buffer) { + /* In case the realloc fails, we can free the memory */ + char *tmp_data_ptr; + + tmp_data_ptr = realloc(data_buffer, data_size); + if (!tmp_data_ptr) { ERR("Allocating data buffer"); + free(data_buffer); ret = -1; goto end; } + data_buffer = tmp_data_ptr; data_buffer_size = data_size; } memset(data_buffer, 0, data_size); DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size); - ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, - MSG_WAITALL); + 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"); + } ret = -1; - ERR("Relay didn't receive the whole metadata"); goto end; } metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer; rcu_read_lock(); - metadata_stream = relay_stream_from_stream_id( - be64toh(metadata_struct->stream_id), streams_ht); + metadata_stream = stream_find_by_id(relay_streams_ht, + be64toh(metadata_struct->stream_id)); if (!metadata_stream) { ret = -1; goto end_unlock; } - do { - ret = write(metadata_stream->fd, metadata_struct->payload, - payload_size); - } while (ret < 0 && errno == EINTR); - if (ret < payload_size) { + size_ret = lttng_write(metadata_stream->fd, metadata_struct->payload, + payload_size); + if (size_ret < payload_size) { ERR("Relay error writing metadata on file"); ret = -1; goto end_unlock; @@ -1245,6 +1526,12 @@ int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr, goto end_unlock; } + ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht, + metadata_stream->path_name); + assert(ctf_trace); + ctf_trace->metadata_received += + payload_size + be32toh(metadata_struct->padding_size); + DBG2("Relay metadata written"); end_unlock: @@ -1258,241 +1545,872 @@ end: */ static int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_command *cmd) + struct relay_connection *conn) { - int ret = htobe32(LTTNG_OK); - struct lttcomm_relayd_version reply; - struct relay_session *session = NULL; + int ret; + struct lttcomm_relayd_version reply, msg; - if (cmd->session == NULL) { - session = zmalloc(sizeof(struct relay_session)); - if (session == NULL) { - PERROR("relay session zmalloc"); - ret = -1; - goto end; + assert(conn); + + conn->version_check_done = 1; + + /* 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."); } - session->id = ++last_relay_session_id; - DBG("Created session %" PRIu64, session->id); - cmd->session = session; + ret = -1; + goto end; + } + + memset(&reply, 0, sizeof(reply)); + reply.major = RELAYD_VERSION_COMM_MAJOR; + reply.minor = RELAYD_VERSION_COMM_MINOR; + + /* Major versions must be the same */ + if (reply.major != be32toh(msg.major)) { + DBG("Incompatible major versions (%u vs %u), deleting session", + reply.major, be32toh(msg.major)); + destroy_session(conn->session, conn->sessions_ht); + ret = 0; + goto end; + } + + conn->major = reply.major; + /* We adapt to the lowest compatible version */ + if (reply.minor <= be32toh(msg.minor)) { + conn->minor = reply.minor; + } else { + conn->minor = be32toh(msg.minor); } - session->version_check_done = 1; - sscanf(VERSION, "%u.%u", &reply.major, &reply.minor); reply.major = htobe32(reply.major); reply.minor = htobe32(reply.minor); - ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, + ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(struct lttcomm_relayd_version), 0); if (ret < 0) { ERR("Relay sending version"); } - DBG("Version check done (%u.%u)", be32toh(reply.major), - be32toh(reply.minor)); - -end: - return ret; -} - -/* - * relay_process_control: Process the commands received on the control socket - */ -static -int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_command *cmd, struct lttng_ht *streams_ht) -{ - int ret = 0; - switch (be32toh(recv_hdr->cmd)) { - /* - case RELAYD_CREATE_SESSION: - ret = relay_create_session(recv_hdr, cmd); - break; - */ - case RELAYD_ADD_STREAM: - ret = relay_add_stream(recv_hdr, cmd, streams_ht); - break; - case RELAYD_START_DATA: - ret = relay_start(recv_hdr, cmd); - break; - case RELAYD_SEND_METADATA: - ret = relay_recv_metadata(recv_hdr, cmd, streams_ht); - break; - case RELAYD_VERSION: - ret = relay_send_version(recv_hdr, cmd); - break; - case RELAYD_CLOSE_STREAM: - ret = relay_close_stream(recv_hdr, cmd, streams_ht); - break; - case RELAYD_UPDATE_SYNC_INFO: - default: - ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd)); - relay_unknown_command(cmd); - ret = -1; - goto end; - } + DBG("Version check done using protocol %u.%u", conn->major, + conn->minor); end: return ret; } /* - * relay_process_data: Process the data received on the data socket + * Check for data pending for a given stream id from the session daemon. */ static -int relay_process_data(struct relay_command *cmd, struct lttng_ht *streams_ht) +int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn) { - int ret = 0; + struct relay_session *session = conn->session; + struct lttcomm_relayd_data_pending msg; + struct lttcomm_relayd_generic_reply reply; struct relay_stream *stream; - struct lttcomm_relayd_data_hdr data_hdr; - uint64_t stream_id; - uint64_t net_seq_num; - uint32_t data_size; - - ret = cmd->sock->ops->recvmsg(cmd->sock, &data_hdr, - sizeof(struct lttcomm_relayd_data_hdr), MSG_WAITALL); - if (ret <= 0) { - ERR("Connections seems to be closed"); - ret = -1; - goto end; - } + int ret; + uint64_t last_net_seq_num, stream_id; - stream_id = be64toh(data_hdr.stream_id); + DBG("Data pending command received"); - rcu_read_lock(); - stream = relay_stream_from_stream_id(stream_id, streams_ht); - if (!stream) { + if (!session || conn->version_check_done == 0) { + ERR("Trying to check for data before version check"); ret = -1; - goto end_unlock; + goto end_no_session; } - data_size = be32toh(data_hdr.data_size); - if (data_buffer_size < data_size) { - data_buffer = realloc(data_buffer, data_size); - if (!data_buffer) { - ERR("Allocating data buffer"); - ret = -1; - goto end_unlock; + 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); } - data_buffer_size = data_size; + ret = -1; + goto end_no_session; } - memset(data_buffer, 0, data_size); - net_seq_num = be64toh(data_hdr.net_seq_num); + stream_id = be64toh(msg.stream_id); + last_net_seq_num = be64toh(msg.last_net_seq_num); - DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64, - data_size, stream_id, net_seq_num); - ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, MSG_WAITALL); - if (ret <= 0) { + rcu_read_lock(); + stream = stream_find_by_id(relay_streams_ht, stream_id); + if (stream == NULL) { ret = -1; goto end_unlock; } - do { - ret = write(stream->fd, data_buffer, data_size); - } while (ret < 0 && errno == EINTR); - if (ret < data_size) { - ERR("Relay error writing data to file"); - ret = -1; - goto end_unlock; - } + DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64 + " and last_seq %" PRIu64, stream_id, stream->prev_seq, + last_net_seq_num); - ret = write_padding_to_file(stream->fd, be32toh(data_hdr.padding_size)); - if (ret < 0) { - goto end_unlock; + /* Avoid wrapping issue */ + if (((int64_t) (stream->prev_seq - last_net_seq_num)) >= 0) { + /* Data has in fact been written and is NOT pending */ + ret = 0; + } else { + /* Data still being streamed thus pending */ + ret = 1; } - DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64, - ret, stream->stream_handle); - - stream->prev_seq = net_seq_num; + /* Pending check is now done. */ + stream->data_pending_check_done = 1; - /* Check if we need to close the FD */ - if (close_stream_check(stream)) { - struct lttng_ht_iter iter; +end_unlock: + rcu_read_unlock(); - close(stream->fd); - iter.iter.node = &stream->stream_n.node; - ret = lttng_ht_del(streams_ht, &iter); - assert(!ret); - call_rcu(&stream->rcu_node, - deferred_free_stream); - DBG("Closed tracefile %d after recv data", stream->fd); + 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"); } -end_unlock: - rcu_read_unlock(); -end: +end_no_session: return ret; } +/* + * 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. + */ static -void relay_cleanup_poll_connection(struct lttng_poll_event *events, int pollfd) +int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn) { int ret; + uint64_t stream_id; + struct relay_stream *stream; + struct lttng_ht_iter iter; + struct lttcomm_relayd_quiescent_control msg; + struct lttcomm_relayd_generic_reply reply; - lttng_poll_del(events, pollfd); + DBG("Checking quiescent state on control socket"); - ret = close(pollfd); - if (ret < 0) { - ERR("Closing pollfd %d", pollfd); + if (!conn->session || conn->version_check_done == 0) { + ERR("Trying to check for data before version check"); + ret = -1; + goto end_no_session; } -} - -static -int relay_add_connection(int fd, struct lttng_poll_event *events, - struct lttng_ht *relay_connections_ht) -{ - struct relay_command *relay_connection; - int ret; - relay_connection = zmalloc(sizeof(struct relay_command)); - if (relay_connection == NULL) { - PERROR("Relay command zmalloc"); - goto error; - } - ret = read(fd, relay_connection, sizeof(struct relay_command)); - if (ret < 0 || ret < sizeof(struct relay_command)) { - PERROR("read relay cmd pipe"); - goto error_read; + 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); + } + ret = -1; + goto end_no_session; } - lttng_ht_node_init_ulong(&relay_connection->sock_n, - (unsigned long) relay_connection->sock->fd); + stream_id = be64toh(msg.stream_id); + rcu_read_lock(); - lttng_ht_add_unique_ulong(relay_connections_ht, - &relay_connection->sock_n); + 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; + } + } rcu_read_unlock(); - return lttng_poll_add(events, - relay_connection->sock->fd, - LPOLLIN | LPOLLRDHUP); -error_read: - free(relay_connection); -error: - return -1; -} + 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"); + } -static -void deferred_free_connection(struct rcu_head *head) -{ - struct relay_command *relay_connection = - caa_container_of(head, struct relay_command, rcu_node); - free(relay_connection); +end_no_session: + return ret; } +/* + * 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. + * + * This command returns to the client a LTTNG_OK code. + */ static -void relay_del_connection(struct lttng_ht *relay_connections_ht, - struct lttng_ht *streams_ht, struct lttng_ht_iter *iter, - struct relay_command *relay_connection) +int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn) { int 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) { + 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); + } + ret = -1; + goto end_no_session; + } + + 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. + */ + 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; + DBG("Set begin data pending flag to stream %" PRIu64, + stream->stream_handle); + } + } + rcu_read_unlock(); + + memset(&reply, 0, sizeof(reply)); + /* 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"); + } + +end_no_session: + return ret; +} + +/* + * 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. + * + * 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) +{ + int 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) { + 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); + } + ret = -1; + goto end_no_session; + } + + session_id = be64toh(msg.session_id); + + /* 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; + } + } + rcu_read_unlock(); + + memset(&reply, 0, sizeof(reply)); + /* 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"); + } + +end_no_session: + return ret; +} + +/* + * Receive an index for a specific stream. + * + * Return 0 on success else a negative value. + */ +static +int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn) +{ + int ret, send_ret, index_created = 0; + struct relay_session *session = conn->session; + struct lttcomm_relayd_index index_info; + struct relay_index *index, *wr_index = NULL; + struct lttcomm_relayd_generic_reply reply; + struct relay_stream *stream; + uint64_t net_seq_num; + + assert(conn); + + DBG("Relay receiving index"); + + if (!session || conn->version_check_done == 0) { + 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); + } + ret = -1; + goto end_no_session; + } + + net_seq_num = be64toh(index_info.net_seq_num); + + rcu_read_lock(); + stream = stream_find_by_id(relay_streams_ht, + be64toh(index_info.relay_stream_id)); + if (!stream) { + ret = -1; + goto end_rcu_unlock; + } + + /* Live beacon handling */ + if (index_info.packet_size == 0) { + DBG("Received live beacon for stream %" PRIu64, stream->stream_handle); + + /* + * Only flag a stream inactive when it has already received data + * and no indexes are in flight. + */ + if (stream->total_index_received > 0 && stream->indexes_in_flight == 0) { + stream->beacon_ts_end = be64toh(index_info.timestamp_end); + } + ret = 0; + goto end_rcu_unlock; + } else { + stream->beacon_ts_end = -1ULL; + } + + index = relay_index_find(stream->stream_handle, net_seq_num); + if (!index) { + /* A successful creation will add the object to the HT. */ + index = relay_index_create(stream->stream_handle, net_seq_num); + if (!index) { + goto end_rcu_unlock; + } + index_created = 1; + stream->indexes_in_flight++; + } + + copy_index_control_data(index, &index_info); + if (stream->ctf_stream_id == -1ULL) { + stream->ctf_stream_id = be64toh(index_info.stream_id); + } + + if (index_created) { + /* + * Try to add the relay index object to the hash table. If an object + * already exist, destroy back the index created, set the data in this + * object and write it on disk. + */ + relay_index_add(index, &wr_index); + if (wr_index) { + copy_index_control_data(wr_index, &index_info); + free(index); + } + } else { + /* The index already exists so write it on disk. */ + wr_index = index; + } + + /* 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_rcu_unlock: + rcu_read_unlock(); - ret = lttng_ht_del(relay_connections_ht, iter); - assert(!ret); - if (relay_connection->type == RELAY_CONTROL) { - relay_delete_session(relay_connection, streams_ht); + 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 < 0) { + ERR("Relay sending close index id reply"); + ret = send_ret; } - call_rcu(&relay_connection->rcu_node, - deferred_free_connection); + +end_no_session: + return ret; +} + +/* + * Receive the streams_sent message. + * + * Return 0 on success else a negative value. + */ +static +int relay_streams_sent(struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn) +{ + int ret, send_ret; + struct lttcomm_relayd_generic_reply reply; + + assert(conn); + + DBG("Relay receiving streams_sent"); + + if (!conn->session || conn->version_check_done == 0) { + 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. + */ + set_viewer_ready_flag(conn); + + /* + * Inform the viewer that there are new streams in the session. + */ + if (conn->session->viewer_refcount) { + uatomic_set(&conn->session->new_streams, 1); + } + + 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; + } + +end_no_session: + return ret; +} + +/* + * Process the commands received on the control socket + */ +static +int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn) +{ + int ret = 0; + + 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); + 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); + break; + case RELAYD_UPDATE_SYNC_INFO: + default: + ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd)); + relay_unknown_command(conn); + ret = -1; + goto end; + } + +end: + return ret; +} + +/* + * Handle index for a data stream. + * + * RCU read side lock MUST be acquired. + * + * Return 0 on success else a negative value. + */ +static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num, + int rotate_index) +{ + int ret = 0, index_created = 0; + uint64_t stream_id, data_offset; + struct relay_index *index, *wr_index = NULL; + + assert(stream); + + stream_id = stream->stream_handle; + /* Get data offset because we are about to update the index. */ + data_offset = htobe64(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. + */ + index = relay_index_find(stream_id, net_seq_num); + if (!index) { + /* A successful creation will add the object to the HT. */ + index = relay_index_create(stream_id, net_seq_num); + if (!index) { + ret = -1; + goto error; + } + index_created = 1; + stream->indexes_in_flight++; + } + + 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 (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. + */ + 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; + } + + /* Do we have a writable ready index to write on disk. */ + if (wr_index) { + ret = relay_index_write(wr_index->fd, wr_index); + if (ret < 0) { + goto error; + } + stream->total_index_received++; + stream->indexes_in_flight--; + assert(stream->indexes_in_flight >= 0); + } + +error: + return ret; +} + +/* + * relay_process_data: Process the data received on the data socket + */ +static +int relay_process_data(struct relay_connection *conn) +{ + int ret = 0, rotate_index = 0; + ssize_t size_ret; + 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); + + 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 = -1; + goto end; + } + + stream_id = be64toh(data_hdr.stream_id); + + rcu_read_lock(); + stream = stream_find_by_id(relay_streams_ht, stream_id); + if (!stream) { + ret = -1; + goto end_rcu_unlock; + } + + session = session_find_by_id(conn->sessions_ht, stream->session_id); + assert(session); + + data_size = be32toh(data_hdr.data_size); + if (data_buffer_size < data_size) { + char *tmp_data_ptr; + + 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); + + net_seq_num = be64toh(data_hdr.net_seq_num); + + 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; + } + + /* Check if a rotation is needed. */ + if (stream->tracefile_size > 0 && + (stream->tracefile_size_current + data_size) > + stream->tracefile_size) { + struct relay_viewer_stream *vstream; + uint64_t new_id; + + new_id = (stream->tracefile_count_current + 1) % + stream->tracefile_count; + /* + * When we wrap-around back to 0, we start overwriting old + * trace data. + */ + if (!stream->tracefile_overwrite && new_id == 0) { + stream->tracefile_overwrite = 1; + } + pthread_mutex_lock(&stream->viewer_stream_rotation_lock); + if (stream->tracefile_overwrite) { + stream->oldest_tracefile_id = + (stream->oldest_tracefile_id + 1) % + stream->tracefile_count; + } + vstream = viewer_stream_find_by_id(stream->stream_handle); + if (vstream) { + /* + * The viewer is reading a file about to be + * overwritten. Close the FDs it is + * currently using and let it handle the fault. + */ + if (vstream->tracefile_count_current == new_id) { + pthread_mutex_lock(&vstream->overwrite_lock); + vstream->abort_flag = 1; + pthread_mutex_unlock(&vstream->overwrite_lock); + DBG("Streaming side setting abort_flag on stream %s_%lu\n", + stream->channel_name, new_id); + } else if (vstream->tracefile_count_current == + stream->tracefile_count_current) { + /* + * The reader and writer were in the + * same trace file, inform the viewer + * that no new index will ever be added + * to this file. + */ + vstream->close_write_flag = 1; + } + } + 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); + stream->total_index_received = 0; + pthread_mutex_unlock(&stream->viewer_stream_rotation_lock); + if (ret < 0) { + ERR("Rotating stream output file"); + goto end_rcu_unlock; + } + /* Reset current size because we just perform a stream rotation. */ + stream->tracefile_size_current = 0; + rotate_index = 1; + } + + /* + * 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 (ret < 0) { + goto end_rcu_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; + } + + DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64, + ret, stream->stream_handle); + + ret = write_padding_to_file(stream->fd, be32toh(data_hdr.padding_size)); + if (ret < 0) { + goto end_rcu_unlock; + } + stream->tracefile_size_current += data_size + be32toh(data_hdr.padding_size); + + stream->prev_seq = net_seq_num; + + try_close_stream(session, stream); + +end_rcu_unlock: + rcu_read_unlock(); +end: + return ret; +} + +static +void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd) +{ + int ret; + + assert(events); + + (void) lttng_poll_del(events, pollfd); + + ret = close(pollfd); + if (ret < 0) { + ERR("Closing pollfd %d", pollfd); + } +} + +static void destroy_connection(struct lttng_ht *relay_connections_ht, + struct relay_connection *conn) +{ + assert(relay_connections_ht); + assert(conn); + + 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); + } + + connection_destroy(conn); } /* @@ -1501,30 +2419,38 @@ void relay_del_connection(struct lttng_ht *relay_connections_ht, static void *relay_thread_worker(void *data) { - int i, ret, pollfd, err = -1; - uint32_t revents, nb_fd; - struct relay_command *relay_connection; + 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_node_ulong *node; struct lttng_ht_iter iter; - struct lttng_ht *streams_ht; 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; DBG("[thread] Relay worker started"); rcu_register_thread(); + health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER); + + if (testpoint(relayd_thread_worker)) { + goto error_testpoint; + } + + health_code_update(); + /* table of connections indexed on socket */ relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); if (!relay_connections_ht) { goto relay_connections_ht_error; } - /* tables of streams indexed by stream ID */ - streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); - if (!streams_ht) { - goto streams_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); @@ -1532,21 +2458,22 @@ void *relay_thread_worker(void *data) goto error_poll_create; } - ret = lttng_poll_add(&events, relay_cmd_pipe[0], LPOLLIN | LPOLLRDHUP); + ret = lttng_poll_add(&events, relay_conn_pipe[0], LPOLLIN | LPOLLRDHUP); if (ret < 0) { goto error; } +restart: while (1) { - /* Zeroed the events structure */ - lttng_poll_reset(&events); + int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1; - nb_fd = LTTNG_POLL_GETNB(&events); + health_code_update(); /* Infinite blocking call, waiting for transmission */ - restart: DBG3("Relayd worker thread polling..."); + health_poll_entry(); ret = lttng_poll_wait(&events, -1); + health_poll_exit(); if (ret < 0) { /* * Restart interrupted system call. @@ -1557,10 +2484,19 @@ void *relay_thread_worker(void *data) goto error; } + 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. + */ for (i = 0; i < nb_fd; i++) { /* Fetch once the poll data */ - revents = LTTNG_POLL_GETEV(&events, i); - pollfd = LTTNG_POLL_GETFD(&events, i); + uint32_t revents = LTTNG_POLL_GETEV(&events, i); + int pollfd = LTTNG_POLL_GETFD(&events, i); + + health_code_update(); /* Thread quit pipe has been closed. Killing thread. */ ret = check_thread_quit_pipe(pollfd, revents); @@ -1569,125 +2505,179 @@ void *relay_thread_worker(void *data) goto exit; } - /* Inspect the relay cmd pipe for new connection */ - if (pollfd == relay_cmd_pipe[0]) { + /* Inspect the relay conn pipe for new connection */ + if (pollfd == relay_conn_pipe[0]) { if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { - ERR("Relay pipe error"); + ERR("Relay connection pipe error"); goto error; } else if (revents & LPOLLIN) { - DBG("Relay command received"); - ret = relay_add_connection(relay_cmd_pipe[0], - &events, relay_connections_ht); + ret = lttng_read(relay_conn_pipe[0], &conn, sizeof(conn)); if (ret < 0) { goto error; } - } - } else if (revents > 0) { - rcu_read_lock(); - lttng_ht_lookup(relay_connections_ht, - (void *)((unsigned long) pollfd), - &iter); - node = lttng_ht_iter_get_node_ulong(&iter); - if (node == NULL) { - DBG2("Relay sock %d not found", pollfd); + 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(); - goto error; + DBG("Connection socket %d added", conn->sock->fd); } - relay_connection = caa_container_of(node, - struct relay_command, sock_n); - - if (revents & (LPOLLERR)) { - ERR("POLL ERROR"); - relay_cleanup_poll_connection(&events, pollfd); - relay_del_connection(relay_connections_ht, - streams_ht, &iter, - relay_connection); - } else if (revents & (LPOLLHUP | LPOLLRDHUP)) { - DBG("Socket %d hung up", pollfd); - relay_cleanup_poll_connection(&events, pollfd); - relay_del_connection(relay_connections_ht, - streams_ht, &iter, - relay_connection); + } else { + rcu_read_lock(); + conn = connection_find_by_sock(relay_connections_ht, pollfd); + /* If not found, there is a synchronization issue. */ + assert(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) { - /* control socket */ - if (relay_connection->type == RELAY_CONTROL) { - ret = relay_connection->sock->ops->recvmsg( - relay_connection->sock, &recv_hdr, - sizeof(struct lttcomm_relayd_hdr), MSG_WAITALL); - /* connection closed */ + if (conn->type == RELAY_CONTROL) { + ret = conn->sock->ops->recvmsg(conn->sock, &recv_hdr, + sizeof(recv_hdr), 0); if (ret <= 0) { - relay_cleanup_poll_connection(&events, pollfd); - relay_del_connection(relay_connections_ht, - streams_ht, &iter, - relay_connection); + /* Connection closed */ + cleanup_connection_pollfd(&events, pollfd); + destroy_connection(relay_connections_ht, conn); DBG("Control connection closed with %d", pollfd); } else { - if (relay_connection->session) { - DBG2("Relay worker receiving data for session : %" PRIu64, - relay_connection->session->id); - } - ret = relay_process_control(&recv_hdr, - relay_connection, - streams_ht); - /* - * there was an error in processing a control - * command: clear the session - * */ + ret = relay_process_control(&recv_hdr, conn); if (ret < 0) { - relay_cleanup_poll_connection(&events, pollfd); - relay_del_connection(relay_connections_ht, - streams_ht, &iter, - relay_connection); + /* 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; } - /* data socket */ - } else if (relay_connection->type == RELAY_DATA) { - ret = relay_process_data(relay_connection, streams_ht); - /* connection closed */ - if (ret < 0) { - relay_cleanup_poll_connection(&events, pollfd); - relay_del_connection(relay_connections_ht, - streams_ht, &iter, - relay_connection); - DBG("Data connection closed with %d", pollfd); - } + } else { + /* + * Flag the last seen data fd not deleted. It will be + * used as the last seen fd if any fd gets deleted in + * this first loop. + */ + last_notdel_data_fd = pollfd; + } + } else { + ERR("Unknown poll events %u for sock %d", revents, pollfd); + } + rcu_read_unlock(); + } + } + + /* + * The last loop handled a control request, go back to poll to make + * sure we prioritise the control socket. + */ + if (seen_control) { + continue; + } + + if (last_seen_data_fd >= 0) { + for (i = 0; i < nb_fd; i++) { + int pollfd = LTTNG_POLL_GETFD(&events, i); + + health_code_update(); + + if (last_seen_data_fd == pollfd) { + idx = i; + break; + } + } + } + + /* Process data connection. */ + for (i = idx + 1; i < nb_fd; i++) { + /* Fetch the poll data. */ + uint32_t revents = LTTNG_POLL_GETEV(&events, i); + int pollfd = LTTNG_POLL_GETFD(&events, i); + + health_code_update(); + + /* Skip the command pipe. It's handled in the first loop. */ + if (pollfd == relay_conn_pipe[0]) { + continue; + } + + if (revents) { + rcu_read_lock(); + conn = connection_find_by_sock(relay_connections_ht, pollfd); + if (!conn) { + /* Skip it. Might be removed before. */ + rcu_read_unlock(); + continue; + } + + if (revents & LPOLLIN) { + if (conn->type != RELAY_DATA) { + continue; + } + + ret = relay_process_data(conn); + /* Connection closed */ + if (ret < 0) { + cleanup_connection_pollfd(&events, pollfd); + destroy_connection(relay_connections_ht, conn); + DBG("Data connection closed with %d", pollfd); + /* + * Every goto restart call sets the last seen fd where + * here we don't really care since we gracefully + * continue the loop after the connection is deleted. + */ + } else { + /* Keep last seen port. */ + last_seen_data_fd = pollfd; + rcu_read_unlock(); + goto restart; } } rcu_read_unlock(); } } + last_seen_data_fd = -1; } + /* Normal exit, no error */ + ret = 0; + exit: error: lttng_poll_clean(&events); - /* empty the hash table and free the memory */ + /* Cleanup reamaining connection object. */ rcu_read_lock(); - cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, node, node) { - node = lttng_ht_iter_get_node_ulong(&iter); - if (node) { - relay_connection = caa_container_of(node, - struct relay_command, sock_n); - relay_del_connection(relay_connections_ht, - streams_ht, &iter, - relay_connection); - } + cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, conn, + sock_n.node) { + health_code_update(); + destroy_connection(relay_connections_ht, conn); } rcu_read_unlock(); error_poll_create: - lttng_ht_destroy(streams_ht); -streams_ht_error: + lttng_ht_destroy(indexes_ht); +indexes_ht_error: lttng_ht_destroy(relay_connections_ht); relay_connections_ht_error: + /* Close relay conn pipes */ + utils_close_pipe(relay_conn_pipe); if (err) { DBG("Thread exited with error"); } DBG("Worker thread cleanup complete"); free(data_buffer); - stop_threads(); +error_testpoint: + if (err) { + health_error(); + ERR("Health error occurred in %s", __func__); + } + health_unregister(health_relayd); rcu_unregister_thread(); + stop_threads(); return NULL; } @@ -1695,11 +2685,11 @@ relay_connections_ht_error: * Create the relay command pipe to wake thread_manage_apps. * Closed in cleanup(). */ -static int create_relay_cmd_pipe(void) +static int create_relay_conn_pipe(void) { int ret; - ret = utils_create_pipe_cloexec(relay_cmd_pipe); + ret = utils_create_pipe_cloexec(relay_conn_pipe); return ret; } @@ -1711,15 +2701,11 @@ int main(int argc, char **argv) { int ret = 0; void *status; - - /* Create thread quit pipe */ - if ((ret = init_thread_quit_pipe()) < 0) { - goto error; - } + struct relay_local_data *relay_ctx; /* Parse arguments */ progname = argv[0]; - if ((ret = parse_args(argc, argv) < 0)) { + if ((ret = set_options(argc, argv)) < 0) { goto exit; } @@ -1727,20 +2713,52 @@ int main(int argc, char **argv) goto exit; } + /* Try to create directory if -o, --output is specified. */ + if (opt_output_path) { + if (*opt_output_path != '/') { + ERR("Please specify an absolute path for -o, --output PATH"); + goto exit; + } + + ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG); + if (ret < 0) { + ERR("Unable to create %s", opt_output_path); + goto exit; + } + } + /* Daemonize */ - if (opt_daemon) { - ret = daemon(0, 0); + if (opt_daemon || opt_background) { + int i; + + ret = lttng_daemonize(&child_ppid, &recv_child_signal, + !opt_background); if (ret < 0) { - PERROR("daemon"); goto exit; } + + /* + * We are in the child. Make sure all other file + * descriptors are closed, in case we are called with + * more opened file descriptors than the standard ones. + */ + for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) { + (void) close(i); + } } - /* Check if daemon is UID = 0 */ - is_root = !getuid(); + /* Create thread quit pipe */ + if ((ret = init_thread_quit_pipe()) < 0) { + goto error; + } + + /* We need those values for the file/dir creation. */ + relayd_uid = getuid(); + relayd_gid = getgid(); - if (!is_root) { - if (control_uri->port < 1024 || data_uri->port < 1024) { + /* 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"); ret = -1; goto exit; @@ -1748,16 +2766,64 @@ int main(int argc, char **argv) } /* Setup the thread apps communication pipe. */ - if ((ret = create_relay_cmd_pipe()) < 0) { + if ((ret = create_relay_conn_pipe()) < 0) { goto exit; } /* Init relay command queue. */ - cds_wfq_init(&relay_cmd_queue.queue); + cds_wfcq_init(&relay_conn_queue.head, &relay_conn_queue.tail); /* Set up max poll set size */ lttng_poll_set_max_size(); + /* Initialize communication library */ + lttcomm_init(); + lttcomm_inet_init(); + + relay_ctx = zmalloc(sizeof(struct relay_local_data)); + if (!relay_ctx) { + PERROR("relay_ctx"); + goto exit; + } + + /* tables of sessions indexed by session ID */ + relay_ctx->sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); + if (!relay_ctx->sessions_ht) { + goto exit_relay_ctx_sessions; + } + + /* tables of streams indexed by stream ID */ + relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); + if (!relay_streams_ht) { + goto exit_relay_ctx_streams; + } + + /* tables of streams indexed by stream ID */ + viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); + if (!viewer_streams_ht) { + goto exit_relay_ctx_viewer_streams; + } + + /* Initialize thread health monitoring */ + health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES); + if (!health_relayd) { + PERROR("health_app_create error"); + goto exit_health_app_create; + } + + ret = utils_create_pipe(health_quit_pipe); + if (ret < 0) { + goto error_health_pipe; + } + + /* Create thread to manage the client socket */ + ret = pthread_create(&health_thread, NULL, + thread_manage_health, (void *) NULL); + if (ret != 0) { + PERROR("pthread_create health"); + goto health_error; + } + /* Setup the dispatcher thread */ ret = pthread_create(&dispatcher_thread, NULL, relay_thread_dispatcher, (void *) NULL); @@ -1768,7 +2834,7 @@ int main(int argc, char **argv) /* Setup the worker thread */ ret = pthread_create(&worker_thread, NULL, - relay_thread_worker, (void *) NULL); + relay_thread_worker, (void *) relay_ctx); if (ret != 0) { PERROR("pthread_create worker"); goto exit_worker; @@ -1782,27 +2848,63 @@ int main(int argc, char **argv) goto exit_listener; } -exit_listener: + ret = live_start_threads(live_uri, relay_ctx); + if (ret != 0) { + ERR("Starting live viewer threads"); + goto exit_live; + } + +exit_live: ret = pthread_join(listener_thread, &status); if (ret != 0) { PERROR("pthread_join"); goto error; /* join error, exit without cleanup */ } -exit_worker: +exit_listener: ret = pthread_join(worker_thread, &status); if (ret != 0) { PERROR("pthread_join"); goto error; /* join error, exit without cleanup */ } -exit_dispatcher: +exit_worker: ret = pthread_join(dispatcher_thread, &status); if (ret != 0) { PERROR("pthread_join"); goto error; /* join error, exit without cleanup */ } +exit_dispatcher: + ret = pthread_join(health_thread, &status); + if (ret != 0) { + PERROR("pthread_join health thread"); + goto error; /* join error, exit without cleanup */ + } + + /* + * Stop live threads only after joining other threads. + */ + live_stop_threads(); + +health_error: + utils_close_pipe(health_quit_pipe); + +error_health_pipe: + health_app_destroy(health_relayd); + +exit_health_app_create: + lttng_ht_destroy(viewer_streams_ht); + +exit_relay_ctx_viewer_streams: + lttng_ht_destroy(relay_streams_ht); + +exit_relay_ctx_streams: + lttng_ht_destroy(relay_ctx->sessions_ht); + +exit_relay_ctx_sessions: + free(relay_ctx); + exit: cleanup(); if (!ret) {