X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-relayd%2Fmain.c;h=02f676df8bc0e1bf56542e2c2394a5cabb59185c;hp=9a68b2b2d940fc2860fbbde22c0ebc20cb89ff59;hb=2f8f53af90479595d530f8f02e71dd0b9fb810ee;hpb=50138f511c30d4d8fe2450c9a5e8fdf98e5d2862 diff --git a/src/bin/lttng-relayd/main.c b/src/bin/lttng-relayd/main.c index 9a68b2b2d..02f676df8 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, @@ -44,12 +45,14 @@ #include #include #include +#include #include #include #include #include #include #include +#include #include "cmd.h" #include "ctf-trace.h" @@ -57,21 +60,39 @@ #include "utils.h" #include "lttng-relayd.h" #include "live.h" +#include "health-relayd.h" +#include "testpoint.h" +#include "viewer-stream.h" /* command line options */ char *opt_output_path; -static int opt_daemon; +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; +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 @@ -85,6 +106,7 @@ 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; @@ -114,6 +136,25 @@ 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 */ @@ -123,85 +164,216 @@ 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"); } +/* + * 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 { + 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; } } @@ -254,6 +426,7 @@ int parse_args(int argc, char **argv) } exit: + free(optstring); return ret; } @@ -273,6 +446,11 @@ void cleanup(void) 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); + } } /* @@ -281,18 +459,26 @@ 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 != 1) { + 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. */ @@ -308,6 +494,8 @@ 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); @@ -334,6 +522,9 @@ void sighandler(int sig) DBG("SIGTERM caught"); stop_threads(); break; + case SIGUSR1: + CMM_STORE_SHARED(recv_child_signal, 1); + break; default: break; } @@ -373,11 +564,26 @@ 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. * @@ -412,7 +618,7 @@ 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; } @@ -513,6 +719,10 @@ void *relay_thread_listener(void *data) 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; @@ -543,11 +753,21 @@ void *relay_thread_listener(void *data) goto error_poll_add; } + lttng_relay_notify_ready(); + + if (testpoint(relayd_thread_listener)) { + goto error_testpoint; + } + while (1) { + health_code_update(); + DBG("Listener accepting connections"); restart: + health_poll_entry(); ret = lttng_poll_wait(&events, -1); + health_poll_exit(); if (ret < 0) { /* * Restart interrupted system call. @@ -562,6 +782,8 @@ restart: 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); @@ -637,6 +859,7 @@ restart: exit: error: error_poll_add: +error_testpoint: lttng_poll_clean(&events); error_create_poll: if (data_sock->fd >= 0) { @@ -656,8 +879,10 @@ 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; @@ -669,17 +894,30 @@ error_sock_control: static void *relay_thread_dispatcher(void *data) { - int ret; + int err = -1; + ssize_t ret; struct cds_wfq_node *node; struct relay_command *relay_cmd = 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); do { + health_code_update(); + /* Dequeue commands */ node = cds_wfq_dequeue_blocking(&relay_cmd_queue.queue); if (node == NULL) { @@ -696,22 +934,31 @@ void *relay_thread_dispatcher(void *data) * 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); + ret = lttng_write(relay_cmd_pipe[1], relay_cmd, + sizeof(struct relay_command)); free(relay_cmd); - if (ret < 0 || ret != sizeof(struct relay_command)) { + if (ret < sizeof(struct relay_command)) { PERROR("write cmd pipe"); goto error; } } while (node != NULL); /* Futex wait on queue. Blocking call on futex() */ + health_poll_entry(); futex_nto1_wait(&relay_cmd_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; @@ -749,8 +996,6 @@ void deferred_free_stream(struct rcu_head *head) struct relay_stream *stream = caa_container_of(head, struct relay_stream, rcu_node); - ctf_trace_try_destroy(stream->ctf_trace); - free(stream->path_name); free(stream->channel_name); free(stream); @@ -770,8 +1015,7 @@ void deferred_free_session(struct rcu_head *head) * RCU read side lock MUST be acquired. If NO close_stream_check() was called * BEFORE the stream lock MUST be acquired. */ -static void destroy_stream(struct relay_stream *stream, - struct lttng_ht *ctf_traces_ht) +static void destroy_stream(struct relay_stream *stream) { int delret; struct relay_viewer_stream *vstream; @@ -791,7 +1035,7 @@ static void destroy_stream(struct relay_stream *stream, } } - vstream = live_find_viewer_stream_by_id(stream->stream_handle); + vstream = viewer_stream_find_by_id(stream->stream_handle); if (vstream) { /* * Set the last good value into the viewer stream. This is done @@ -799,7 +1043,11 @@ static void destroy_stream(struct relay_stream *stream, * lookup failure on the live thread side of a stream indicates * that the viewer stream index received value should be used. */ + pthread_mutex_lock(&stream->viewer_stream_rotation_lock); vstream->total_index_received = stream->total_index_received; + vstream->tracefile_count_last = stream->tracefile_count_current; + vstream->close_write_flag = 1; + pthread_mutex_unlock(&stream->viewer_stream_rotation_lock); } /* Cleanup index of that stream. */ @@ -809,8 +1057,13 @@ static void destroy_stream(struct relay_stream *stream, delret = lttng_ht_del(relay_streams_ht, &iter); assert(!delret); iter.iter.node = &stream->ctf_trace_node.node; - delret = lttng_ht_del(ctf_traces_ht, &iter); + delret = lttng_ht_del(stream->ctf_traces_ht, &iter); assert(!delret); + + if (stream->ctf_trace) { + ctf_trace_try_destroy(stream->ctf_trace); + } + call_rcu(&stream->rcu_node, deferred_free_stream); DBG("Closed tracefile %d from close stream", stream->fd); } @@ -842,7 +1095,9 @@ void relay_delete_session(struct relay_command *cmd, } stream = caa_container_of(node, struct relay_stream, stream_n); if (stream->session == cmd->session) { - destroy_stream(stream, cmd->ctf_traces_ht); + destroy_stream(stream); + cmd->session->stream_count--; + assert(cmd->session->stream_count >= 0); } } @@ -906,11 +1161,16 @@ int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr, session->sock = cmd->sock; session->minor = cmd->minor; session->major = cmd->major; + pthread_mutex_init(&session->viewer_ready_lock, NULL); cmd->session = session; reply.session_id = htobe64(session->id); switch (cmd->minor) { + case 1: + case 2: + case 3: + break; case 4: /* LTTng sessiond 2.4 */ default: ret = cmd_create_session_2_4(cmd, session); @@ -940,6 +1200,64 @@ error: return ret; } +/* + * 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_command *cmd) +{ + struct relay_stream_recv_handle *node, *tmp_node; + + pthread_mutex_lock(&cmd->session->viewer_ready_lock); + + cds_list_for_each_entry_safe(node, tmp_node, &cmd->recv_head, node) { + struct relay_stream *stream; + + rcu_read_lock(); + stream = relay_stream_find_by_id(node->id); + if (!stream) { + /* + * Stream is most probably being cleaned up by the data thread thus + * simply continue to the next one. + */ + rcu_read_unlock(); + continue; + } + + stream->viewer_ready = 1; + rcu_read_unlock(); + + /* Clean stream handle node. */ + cds_list_del(&node->node); + free(node); + } + + pthread_mutex_unlock(&cmd->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_handle(uint64_t handle, struct relay_command *cmd) +{ + struct relay_stream_recv_handle *node; + + assert(cmd); + + node = zmalloc(sizeof(*node)); + if (!node) { + PERROR("zmalloc queue stream handle"); + return; + } + + node->id = handle; + cds_list_add(&node->node, &cmd->recv_head); +} + /* * relay_add_stream: allocate a new stream for a session */ @@ -1030,6 +1348,14 @@ int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr, stream->ctf_trace->metadata_stream = stream; } ctf_trace_assign(cmd->ctf_traces_ht, stream); + stream->ctf_traces_ht = cmd->ctf_traces_ht; + + /* + * Add the stream handle 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_handle(stream->stream_handle, cmd); lttng_ht_node_init_ulong(&stream->stream_n, (unsigned long) stream->stream_handle); @@ -1038,6 +1364,7 @@ int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr, lttng_ht_node_init_str(&stream->ctf_trace_node, stream->path_name); lttng_ht_add_str(cmd->ctf_traces_ht, &stream->ctf_trace_node); + session->stream_count++; DBG("Relay new stream added %s with ID %" PRIu64, stream->channel_name, stream->stream_handle); @@ -1114,9 +1441,11 @@ int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr, 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)) { - destroy_stream(stream, cmd->ctf_traces_ht); + destroy_stream(stream); } end_unlock: @@ -1187,7 +1516,7 @@ int relay_start(struct lttcomm_relayd_hdr *recv_hdr, */ static int write_padding_to_file(int fd, uint32_t size) { - int ret = 0; + ssize_t ret = 0; char *zeros; if (size == 0) { @@ -1201,10 +1530,8 @@ 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 != size) { + ret = lttng_write(fd, zeros, size); + if (ret < size) { PERROR("write padding to file"); } @@ -1222,6 +1549,7 @@ int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr, struct relay_command *cmd) { int ret = htobe32(LTTNG_OK); + ssize_t size_ret; struct relay_session *session = cmd->session; struct lttcomm_relayd_metadata_payload *metadata_struct; struct relay_stream *metadata_stream; @@ -1278,11 +1606,9 @@ int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr, goto end_unlock; } - do { - ret = write(metadata_stream->fd, metadata_struct->payload, - payload_size); - } while (ret < 0 && errno == EINTR); - if (ret < 0 || 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; @@ -1779,6 +2105,53 @@ 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_command *cmd) +{ + int ret, send_ret; + struct lttcomm_relayd_generic_reply reply; + + assert(cmd); + + DBG("Relay receiving streams_sent"); + + if (!cmd->session || cmd->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(cmd); + + /* + * Inform the viewer that there are new streams in the session. + */ + uatomic_set(&cmd->session->new_streams, 1); + + reply.ret_code = htobe32(LTTNG_OK); + send_ret = cmd->sock->ops->sendmsg(cmd->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 */ @@ -1822,6 +2195,9 @@ int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr, case RELAYD_SEND_INDEX: ret = relay_recv_index(recv_hdr, cmd); break; + case RELAYD_STREAMS_SENT: + ret = relay_streams_sent(recv_hdr, cmd); + break; case RELAYD_UPDATE_SYNC_INFO: default: ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd)); @@ -1923,6 +2299,7 @@ static int relay_process_data(struct relay_command *cmd) { 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; @@ -1985,10 +2362,54 @@ int relay_process_data(struct relay_command *cmd) 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; @@ -2010,10 +2431,8 @@ int relay_process_data(struct relay_command *cmd) } /* Write data to stream output fd. */ - do { - ret = write(stream->fd, data_buffer, data_size); - } while (ret < 0 && errno == EINTR); - if (ret < 0 || ret != data_size) { + 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; @@ -2032,7 +2451,7 @@ int relay_process_data(struct relay_command *cmd) /* Check if we need to close the FD */ if (close_stream_check(stream)) { - destroy_stream(stream, cmd->ctf_traces_ht); + destroy_stream(stream); } end_rcu_unlock: @@ -2059,24 +2478,31 @@ int relay_add_connection(int fd, struct lttng_poll_event *events, struct lttng_ht *relay_connections_ht) { struct relay_command *relay_connection; - int ret; + ssize_t ret; relay_connection = zmalloc(sizeof(struct relay_command)); if (relay_connection == NULL) { PERROR("Relay command zmalloc"); goto error; } - do { - ret = read(fd, relay_connection, sizeof(struct relay_command)); - } while (ret < 0 && errno == EINTR); - if (ret < 0 || ret < sizeof(struct relay_command)) { + ret = lttng_read(fd, relay_connection, sizeof(struct relay_command)); + if (ret < sizeof(struct relay_command)) { PERROR("read relay cmd pipe"); goto error_read; } + CDS_INIT_LIST_HEAD(&relay_connection->recv_head); - relay_connection->ctf_traces_ht = lttng_ht_new(0, LTTNG_HT_TYPE_STRING); - if (!relay_connection->ctf_traces_ht) { - goto error_read; + /* + * Only used by the control side and the reference is copied inside each + * stream from that connection. Thus a destroy HT must be done after every + * stream has been destroyed. + */ + if (relay_connection->type == RELAY_CONTROL) { + relay_connection->ctf_traces_ht = lttng_ht_new(0, + LTTNG_HT_TYPE_STRING); + if (!relay_connection->ctf_traces_ht) { + goto error_read; + } } lttng_ht_node_init_ulong(&relay_connection->sock_n, @@ -2101,7 +2527,6 @@ void deferred_free_connection(struct rcu_head *head) struct relay_command *relay_connection = caa_container_of(head, struct relay_command, rcu_node); - lttng_ht_destroy(relay_connection->ctf_traces_ht); lttcomm_destroy_sock(relay_connection->sock); free(relay_connection); } @@ -2115,12 +2540,22 @@ void relay_del_connection(struct lttng_ht *relay_connections_ht, ret = lttng_ht_del(relay_connections_ht, iter); assert(!ret); + if (relay_connection->type == RELAY_CONTROL) { + struct relay_stream_recv_handle *node, *tmp_node; + relay_delete_session(relay_connection, sessions_ht); + lttng_ht_destroy(relay_connection->ctf_traces_ht); + + /* Clean up recv list. */ + cds_list_for_each_entry_safe(node, tmp_node, + &relay_connection->recv_head, node) { + cds_list_del(&node->node); + free(node); + } } - call_rcu(&relay_connection->rcu_node, - deferred_free_connection); + call_rcu(&relay_connection->rcu_node, deferred_free_connection); } /* @@ -2144,6 +2579,14 @@ void *relay_thread_worker(void *data) 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) { @@ -2170,9 +2613,13 @@ restart: while (1) { int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1; + health_code_update(); + /* Infinite blocking call, waiting for transmission */ DBG3("Relayd worker thread polling..."); + health_poll_entry(); ret = lttng_poll_wait(&events, -1); + health_poll_exit(); if (ret < 0) { /* * Restart interrupted system call. @@ -2195,6 +2642,8 @@ restart: 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); if (ret) { @@ -2297,6 +2746,9 @@ restart: 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; @@ -2310,6 +2762,8 @@ restart: 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_cmd_pipe[0]) { continue; @@ -2359,6 +2813,9 @@ restart: last_seen_data_fd = -1; } + /* Normal exit, no error */ + ret = 0; + exit: error: lttng_poll_clean(&events); @@ -2366,6 +2823,8 @@ error: /* empty the hash table and free the memory */ rcu_read_lock(); cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, node, node) { + health_code_update(); + node = lttng_ht_iter_get_node_ulong(&iter); if (node) { relay_connection = caa_container_of(node, @@ -2387,8 +2846,14 @@ relay_connections_ht_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; } @@ -2414,14 +2879,9 @@ int main(int argc, char **argv) void *status; struct relay_local_data *relay_ctx; - /* Create thread quit pipe */ - if ((ret = init_thread_quit_pipe()) < 0) { - goto error; - } - /* Parse arguments */ progname = argv[0]; - if ((ret = parse_args(argc, argv)) < 0) { + if ((ret = set_options(argc, argv)) < 0) { goto exit; } @@ -2444,12 +2904,28 @@ int main(int argc, char **argv) } /* 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); + } + } + + /* Create thread quit pipe */ + if ((ret = init_thread_quit_pipe()) < 0) { + goto error; } /* We need those values for the file/dir creation. */ @@ -2458,7 +2934,7 @@ int main(int argc, char **argv) /* Check if daemon is UID = 0 */ if (relayd_uid == 0) { - if (control_uri->port < 1024 || data_uri->port < 1024) { + 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; @@ -2478,6 +2954,7 @@ int main(int argc, char **argv) /* Initialize communication library */ lttcomm_init(); + lttcomm_inet_init(); relay_ctx = zmalloc(sizeof(struct relay_local_data)); if (!relay_ctx) { @@ -2503,6 +2980,26 @@ int main(int argc, char **argv) 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); @@ -2527,14 +3024,12 @@ int main(int argc, char **argv) goto exit_listener; } - ret = live_start_threads(live_uri, relay_ctx, thread_quit_pipe); + ret = live_start_threads(live_uri, relay_ctx); if (ret != 0) { ERR("Starting live viewer threads"); goto exit_live; } - live_stop_threads(); - exit_live: ret = pthread_join(listener_thread, &status); if (ret != 0) { @@ -2557,6 +3052,24 @@ exit_worker: } 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: