X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-relayd%2Fmain.c;h=45b7de8c657f9b43a426057fc07f7a299689dc75;hp=5782175e6e7f15b57753cc29dedebd54db7a51da;hb=4f5fb4c3d8752aae822ed0066784cc77e6f0f508;hpb=6b6b9a5a667d6713eb7bcf22daf4f9bbeef00fb1 diff --git a/src/bin/lttng-relayd/main.c b/src/bin/lttng-relayd/main.c index 5782175e6..45b7de8c6 100644 --- a/src/bin/lttng-relayd/main.c +++ b/src/bin/lttng-relayd/main.c @@ -1,6 +1,8 @@ /* * Copyright (C) 2012 - Julien Desfossez * David Goulet + * 2013 - Jérémie Galarneau + * 2015 - Mathieu Desnoyers * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License, version 2 only, @@ -16,7 +18,7 @@ * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. */ -#define _GNU_SOURCE +#define _LGPL_SOURCE #include #include #include @@ -37,19 +39,24 @@ #include #include #include -#include #include #include #include #include +#include +#include #include +#include #include #include #include #include #include #include +#include +#include +#include #include "cmd.h" #include "ctf-trace.h" @@ -58,10 +65,40 @@ #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" +#include "tracefile-array.h" +#include "tcp_keep_alive.h" + +static const char *help_msg = +#ifdef LTTNG_EMBED_HELP +#include +#else +NULL +#endif +; /* 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; + +/* Size of receive buffer. */ +#define RECV_DATA_BUFFER_SIZE 65536 +#define FILE_COPY_BUFFER_SIZE 65536 + +static int recv_child_signal; /* Set to 1 when a SIGUSR1 signal is received. */ +static pid_t child_ppid; /* Internal parent PID use with daemonize. */ + static struct lttng_uri *control_uri; static struct lttng_uri *data_uri; static struct lttng_uri *live_uri; @@ -69,18 +106,21 @@ 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 * 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; @@ -90,8 +130,12 @@ static pthread_t dispatcher_thread; static pthread_t worker_thread; static pthread_t health_thread; +/* + * last_relay_stream_id_lock protects last_relay_stream_id increment + * atomicity on 32-bit architectures. + */ +static pthread_mutex_t last_relay_stream_id_lock = PTHREAD_MUTEX_INITIALIZER; static uint64_t last_relay_stream_id; -static uint64_t last_relay_session_id; /* * Relay command queue. @@ -99,131 +143,296 @@ 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; +/* Global relay sessions hash table. */ +struct lttng_ht *sessions_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' }, + { "version", 0, 0, 'V' }, + { NULL, 0, 0, 0, }, +}; + +static const char *config_ignore_options[] = { "help", "config", "version" }; + /* - * usage function on stderr + * 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 -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 URL Control port listening.\n"); - fprintf(stderr, " -D, --data-port URL Data 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"); -} - -static -int parse_args(int argc, char **argv) +static 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', }, - { "group", 1, 0, 'g', }, - { "help", 0, 0, 'h', }, - { "output", 1, 0, 'o', }, - { "verbose", 0, 0, 'v', }, - { NULL, 0, 0, 0, }, - }; + int ret; - while (1) { - int option_index = 0; - c = getopt_long(argc, argv, "dhv" "C:D:o:g:", - long_options, &option_index); - if (c == -1) { - break; + switch (opt) { + case 0: + fprintf(stderr, "option %s", optname); + if (arg) { + fprintf(stderr, " with arg %s\n", arg); } - - 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); + break; + case 'C': + if (lttng_is_setuid_setgid()) { + WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", + "-C, --control-port"); + } else { + ret = uri_parse(arg, &control_uri); if (ret < 0) { ERR("Invalid control URI specified"); - goto exit; + goto end; } if (control_uri->port == 0) { control_uri->port = DEFAULT_NETWORK_CONTROL_PORT; } - break; - case 'D': - ret = uri_parse(optarg, &data_uri); + } + break; + case 'D': + if (lttng_is_setuid_setgid()) { + WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", + "-D, -data-port"); + } else { + ret = uri_parse(arg, &data_uri); if (ret < 0) { ERR("Invalid data URI specified"); - goto exit; + goto end; } if (data_uri->port == 0) { data_uri->port = DEFAULT_NETWORK_DATA_PORT; } - break; - case 'd': - opt_daemon = 1; - break; - case 'g': - tracing_group_name = optarg; - break; - case 'h': - usage(); - exit(EXIT_FAILURE); - case 'o': - ret = asprintf(&opt_output_path, "%s", optarg); + } + break; + case 'L': + if (lttng_is_setuid_setgid()) { + WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", + "-L, -live-port"); + } else { + ret = uri_parse(arg, &live_uri); + if (ret < 0) { + ERR("Invalid live URI specified"); + goto end; + } + if (live_uri->port == 0) { + live_uri->port = DEFAULT_NETWORK_VIEWER_PORT; + } + } + break; + case 'd': + opt_daemon = 1; + break; + case 'b': + opt_background = 1; + break; + case 'g': + if (lttng_is_setuid_setgid()) { + WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", + "-g, --group"); + } else { + tracing_group_name = strdup(arg); + if (tracing_group_name == NULL) { + ret = -errno; + PERROR("strdup"); + goto end; + } + tracing_group_name_override = 1; + } + break; + case 'h': + ret = utils_show_help(8, "lttng-relayd", help_msg); + if (ret) { + ERR("Cannot show --help for `lttng-relayd`"); + perror("exec"); + } + exit(EXIT_FAILURE); + case 'V': + fprintf(stdout, "%s\n", VERSION); + exit(EXIT_SUCCESS); + case 'o': + if (lttng_is_setuid_setgid()) { + WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", + "-o, --output"); + } else { + ret = asprintf(&opt_output_path, "%s", arg); if (ret < 0) { + ret = -errno; PERROR("asprintf opt_output_path"); - goto exit; + 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/session-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, retval = 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) { + retval = -ENOMEM; + goto exit; + } + + /* Check for the --config option */ + + while ((c = getopt_long(argc, argv, optstring, long_options, + &option_index)) != -1) { + if (c == '?') { + retval = -EINVAL; + goto exit; + } else if (c != 'f') { + continue; + } + + if (lttng_is_setuid_setgid()) { + WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", + "-f, --config"); + } else { + config_path = utils_expand_path(optarg); + if (!config_path) { + ERR("Failed to resolve path: %s", optarg); + } + } + } + + 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); + } + retval = -1; + goto exit; + } + + /* Reset getopt's global state */ + optopt = orig_optopt; + optind = orig_optind; + while (1) { + c = getopt_long(argc, argv, optstring, long_options, &option_index); + if (c == -1) { 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) { + retval = -1; 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"); + retval = -1; goto exit; } @@ -231,14 +440,17 @@ int parse_args(int argc, char **argv) free(default_address); if (ret < 0) { ERR("Invalid control URI specified"); + retval = -1; goto exit; } } 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"); + retval = -1; goto exit; } @@ -246,14 +458,17 @@ int parse_args(int argc, char **argv) free(default_address); if (ret < 0) { ERR("Invalid data URI specified"); + retval = -1; goto exit; } } if (live_uri == NULL) { - ret = asprintf(&default_address, "tcp://0.0.0.0:%d", - DEFAULT_NETWORK_VIEWER_PORT); + ret = asprintf(&default_address, + "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS ":%d", + DEFAULT_NETWORK_VIEWER_PORT); if (ret < 0) { PERROR("asprintf default viewer control address"); + retval = -1; goto exit; } @@ -261,22 +476,43 @@ int parse_args(int argc, char **argv) free(default_address); if (ret < 0) { ERR("Invalid viewer control URI specified"); + retval = -1; goto exit; } } exit: - return ret; + free(optstring); + return retval; +} + +static void print_global_objects(void) +{ + rcu_register_thread(); + + print_viewer_streams(); + print_relay_streams(); + print_sessions(); + + rcu_unregister_thread(); } /* * Cleanup the daemon */ -static -void cleanup(void) +static void relayd_cleanup(void) { + print_global_objects(); + DBG("Cleaning up"); + if (viewer_streams_ht) + lttng_ht_destroy(viewer_streams_ht); + if (relay_streams_ht) + lttng_ht_destroy(relay_streams_ht); + if (sessions_ht) + lttng_ht_destroy(sessions_ht); + /* free the dynamically allocated opt_output_path */ free(opt_output_path); @@ -285,54 +521,71 @@ 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); + } } /* * Write to writable pipe used to notify a thread. */ -static -int notify_thread_pipe(int wpipe) +static int notify_thread_pipe(int wpipe) { ssize_t ret; ret = lttng_write(wpipe, "!", 1); if (ret < 1) { PERROR("write poll pipe"); + goto end; } - + ret = 0; +end: return ret; } -static void notify_health_quit_pipe(int *pipe) +static int notify_health_quit_pipe(int *pipe) { ssize_t ret; ret = lttng_write(pipe[1], "4", 1); if (ret < 1) { PERROR("write relay health quit"); + goto end; } + ret = 0; +end: + return ret; } /* - * Stop all threads by closing the thread quit pipe. + * Stop all relayd and relayd-live threads. */ -static -void stop_threads(void) +int lttng_relay_stop_threads(void) { - int ret; + int retval = 0; /* Stopping all threads */ DBG("Terminating all threads"); - ret = notify_thread_pipe(thread_quit_pipe[1]); - if (ret < 0) { + if (notify_thread_pipe(thread_quit_pipe[1])) { ERR("write error on thread quit pipe"); + retval = -1; } - notify_health_quit_pipe(health_quit_pipe); + if (notify_health_quit_pipe(health_quit_pipe)) { + ERR("write error on 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); + + if (relayd_live_stop()) { + ERR("Error stopping live threads"); + retval = -1; + } + return retval; } /* @@ -341,20 +594,23 @@ void stop_threads(void) * Simply stop all worker threads, leaving main() return gracefully after * joining all threads and calling cleanup(). */ -static -void sighandler(int sig) +static void sighandler(int sig) { switch (sig) { - case SIGPIPE: - DBG("SIGPIPE caught"); - return; case SIGINT: DBG("SIGINT caught"); - stop_threads(); + if (lttng_relay_stop_threads()) { + ERR("Error stopping threads"); + } break; case SIGTERM: DBG("SIGTERM caught"); - stop_threads(); + if (lttng_relay_stop_threads()) { + ERR("Error stopping threads"); + } + break; + case SIGUSR1: + CMM_STORE_SHARED(recv_child_signal, 1); break; default: break; @@ -365,8 +621,7 @@ void sighandler(int sig) * Setup signal handler for : * SIGINT, SIGTERM, SIGPIPE */ -static -int set_signal_handler(void) +static int set_signal_handler(void) { int ret = 0; struct sigaction sa; @@ -377,9 +632,10 @@ int set_signal_handler(void) return ret; } - sa.sa_handler = sighandler; sa.sa_mask = sigset; sa.sa_flags = 0; + + sa.sa_handler = sighandler; if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) { PERROR("sigaction"); return ret; @@ -390,23 +646,38 @@ int set_signal_handler(void) return ret; } + if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) { + PERROR("sigaction"); + return ret; + } + + sa.sa_handler = SIG_IGN; if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) { PERROR("sigaction"); return ret; } - DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT"); + 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. * * Return -1 on error or 0 if all pipes are created. */ -static -int init_thread_quit_pipe(void) +static int init_thread_quit_pipe(void) { int ret; @@ -418,8 +689,7 @@ int init_thread_quit_pipe(void) /* * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set. */ -static -int create_thread_poll_set(struct lttng_poll_event *events, int size) +static int create_thread_poll_set(struct lttng_poll_event *events, int size) { int ret; @@ -434,7 +704,7 @@ 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; } @@ -450,8 +720,7 @@ error: * * Return 1 if it was triggered else 0; */ -static -int check_thread_quit_pipe(int fd, uint32_t events) +static int check_thread_quit_pipe(int fd, uint32_t events) { if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) { return 1; @@ -463,8 +732,7 @@ int check_thread_quit_pipe(int fd, uint32_t events) /* * Create and init socket from uri. */ -static -struct lttcomm_sock *relay_init_sock(struct lttng_uri *uri) +static struct lttcomm_sock *relay_socket_create(struct lttng_uri *uri) { int ret; struct lttcomm_sock *sock = NULL; @@ -501,34 +769,12 @@ error: return NULL; } -/* - * Return nonzero if stream needs to be closed. - */ -static -int close_stream_check(struct relay_stream *stream) -{ - - if (stream->close_flag && stream->prev_seq == stream->last_net_seq_num) { - /* - * We are about to close the stream so set the data pending flag to 1 - * which will make the end data pending command skip the stream which - * is now closed and ready. Note that after proceeding to a file close, - * the written file is ready for reading. - */ - stream->data_pending_check_done = 1; - return 1; - } - return 0; -} - /* * This thread manages the listening for new connections on the network */ -static -void *relay_thread_listener(void *data) +static void *relay_thread_listener(void *data) { int i, ret, pollfd, err = -1; - int val = 1; uint32_t revents, nb_fd; struct lttng_poll_event events; struct lttcomm_sock *control_sock, *data_sock; @@ -539,18 +785,19 @@ void *relay_thread_listener(void *data) health_code_update(); - control_sock = relay_init_sock(control_uri); + control_sock = relay_socket_create(control_uri); if (!control_sock) { goto error_sock_control; } - data_sock = relay_init_sock(data_uri); + data_sock = relay_socket_create(data_uri); if (!data_sock) { goto error_sock_relay; } /* - * Pass 3 as size here for the thread quit pipe, control and data socket. + * Pass 3 as size here for the thread quit pipe, control and + * data socket. */ ret = create_thread_poll_set(&events, 3); if (ret < 0) { @@ -569,6 +816,12 @@ 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(); @@ -598,6 +851,14 @@ restart: revents = LTTNG_POLL_GETEV(&events, i); pollfd = LTTNG_POLL_GETFD(&events, i); + if (!revents) { + /* + * No activity for this FD (poll + * implementation). + */ + continue; + } + /* Thread quit pipe has been closed. Killing thread. */ ret = check_thread_quit_pipe(pollfd, revents); if (ret) { @@ -605,63 +866,73 @@ restart: goto exit; } - if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { - ERR("socket poll error"); - goto error; - } else if (revents & LPOLLIN) { + if (revents & LPOLLIN) { /* - * Get allocated in this thread, - * enqueued to a global queue, dequeued - * and freed in the worker thread. + * A new connection is requested, therefore a + * sessiond/consumerd connection is allocated in + * this thread, enqueued to a global queue and + * dequeued (and freed) in the worker thread. */ - struct relay_command *relay_cmd; + 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"); - goto error; - } + enum connection_type type; if (pollfd == data_sock->fd) { + type = RELAY_DATA; newsock = data_sock->ops->accept(data_sock); - if (!newsock) { - PERROR("accepting data sock"); - free(relay_cmd); - goto error; - } - relay_cmd->type = RELAY_DATA; - DBG("Relay data connection accepted, socket %d", newsock->fd); + DBG("Relay data connection accepted, socket %d", + newsock->fd); } else { assert(pollfd == control_sock->fd); + type = RELAY_CONTROL; newsock = control_sock->ops->accept(control_sock); - if (!newsock) { - PERROR("accepting control sock"); - free(relay_cmd); - 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); + } + if (!newsock) { + PERROR("accepting sock"); + goto error; } - ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, - &val, sizeof(int)); + + ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val, + sizeof(val)); if (ret < 0) { PERROR("setsockopt inet"); lttcomm_destroy_sock(newsock); - free(relay_cmd); goto error; } - relay_cmd->sock = newsock; - /* - * Lock free enqueue the request. - */ - cds_wfq_enqueue(&relay_cmd_queue.queue, &relay_cmd->node); + + ret = socket_apply_keep_alive_config(newsock->fd); + if (ret < 0) { + ERR("Failed to apply TCP keep-alive configuration on socket (%i)", + newsock->fd); + lttcomm_destroy_sock(newsock); + goto error; + } + + new_conn = connection_create(newsock, type); + if (!new_conn) { + lttcomm_destroy_sock(newsock); + goto error; + } + + /* Enqueue request for the dispatcher thread. */ + cds_wfcq_enqueue(&relay_conn_queue.head, &relay_conn_queue.tail, + &new_conn->qnode); /* - * Wake the dispatch queue futex. Implicit memory - * barrier with the exchange in cds_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); + } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { + ERR("socket poll error"); + goto error; + } else { + ERR("Unexpected poll events %u for sock %d", revents, pollfd); + goto error; } } } @@ -669,6 +940,7 @@ restart: exit: error: error_poll_add: +error_testpoint: lttng_poll_clean(&events); error_create_poll: if (data_sock->fd >= 0) { @@ -693,64 +965,72 @@ error_sock_control: } health_unregister(health_relayd); DBG("Relay listener thread cleanup complete"); - stop_threads(); + lttng_relay_stop_threads(); return NULL; } /* * This thread manages the dispatching of the requests to worker threads */ -static -void *relay_thread_dispatcher(void *data) +static void *relay_thread_dispatcher(void *data) { int err = -1; ssize_t ret; - struct cds_wfq_node *node; - struct relay_command *relay_cmd = NULL; + 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)) { + for (;;) { health_code_update(); /* Atomically prepare the queue futex */ - futex_nto1_prepare(&relay_cmd_queue.futex); + futex_nto1_prepare(&relay_conn_queue.futex); + + if (CMM_LOAD_SHARED(dispatch_thread_exit)) { + break; + } 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 :) + * call is blocking so we can be assured that + * the data will be read at some point in time + * or wait to the end of the world :) */ - ret = lttng_write(relay_cmd_pipe[1], relay_cmd, - sizeof(struct relay_command)); - free(relay_cmd); - if (ret < sizeof(struct relay_command)) { - PERROR("write cmd pipe"); + ret = lttng_write(relay_conn_pipe[1], &new_conn, sizeof(new_conn)); + if (ret < 0) { + PERROR("write connection pipe"); + connection_put(new_conn); goto error; } } while (node != NULL); /* Futex wait on queue. Blocking call on futex() */ health_poll_entry(); - futex_nto1_wait(&relay_cmd_queue.futex); + futex_nto1_wait(&relay_conn_queue.futex); health_poll_exit(); } @@ -758,435 +1038,392 @@ void *relay_thread_dispatcher(void *data) 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(); + lttng_relay_stop_threads(); return NULL; } /* - * Get stream from stream id. - * Need to be called with RCU read-side lock held. + * Set index data from the control port to a given index object. */ -struct relay_stream *relay_stream_find_by_id(uint64_t stream_id) -{ - struct lttng_ht_node_ulong *node; - struct lttng_ht_iter iter; - struct relay_stream *ret; - - lttng_ht_lookup(relay_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; -} - -static -void deferred_free_stream(struct rcu_head *head) +static int set_index_control_data(struct relay_index *index, + struct lttcomm_relayd_index *data, + struct relay_connection *conn) { - struct relay_stream *stream = - caa_container_of(head, struct relay_stream, rcu_node); + struct ctf_packet_index index_data; - ctf_trace_try_destroy(stream->ctf_trace); + /* + * 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_data.packet_size = data->packet_size; + index_data.content_size = data->content_size; + index_data.timestamp_begin = data->timestamp_begin; + index_data.timestamp_end = data->timestamp_end; + index_data.events_discarded = data->events_discarded; + index_data.stream_id = data->stream_id; - free(stream->path_name); - free(stream->channel_name); - free(stream); -} + if (conn->minor >= 8) { + index->index_data.stream_instance_id = data->stream_instance_id; + index->index_data.packet_seq_num = data->packet_seq_num; + } -static -void deferred_free_session(struct rcu_head *head) -{ - struct relay_session *session = - caa_container_of(head, struct relay_session, rcu_node); - free(session); + return relay_index_set_data(index, &index_data); } /* - * Close a given stream. The stream is freed using a call RCU. + * Handle the RELAYD_CREATE_SESSION command. * - * RCU read side lock MUST be acquired. If NO close_stream_check() was called - * BEFORE the stream lock MUST be acquired. + * On success, send back the session id or else return a negative value. */ -static void destroy_stream(struct relay_stream *stream) +static int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn) { - int delret; - struct relay_viewer_stream *vstream; - struct lttng_ht_iter iter; + int ret = 0, send_ret; + struct relay_session *session; + struct lttcomm_relayd_status_session reply; + char session_name[LTTNG_NAME_MAX]; + char hostname[LTTNG_HOST_NAME_MAX]; + uint32_t live_timer = 0; + bool snapshot = false; - assert(stream); + memset(session_name, 0, LTTNG_NAME_MAX); + memset(hostname, 0, LTTNG_HOST_NAME_MAX); - delret = close(stream->fd); - if (delret < 0) { - PERROR("close stream"); - } + memset(&reply, 0, sizeof(reply)); - if (stream->index_fd >= 0) { - delret = close(stream->index_fd); - if (delret < 0) { - PERROR("close stream index_fd"); - } + 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_name, + hostname, &live_timer, &snapshot); + } + if (ret < 0) { + goto send_reply; } - vstream = live_find_viewer_stream_by_id(stream->stream_handle); - if (vstream) { - /* - * Set the last good value into the viewer stream. This is done - * right before the stream gets deleted from the hash table. The - * lookup failure on the live thread side of a stream indicates - * that the viewer stream index received value should be used. - */ - vstream->total_index_received = stream->total_index_received; + session = session_create(session_name, hostname, live_timer, + snapshot, conn->major, conn->minor); + if (!session) { + ret = -1; + goto send_reply; + } + assert(!conn->session); + conn->session = session; + DBG("Created session %" PRIu64, session->id); + + reply.session_id = htobe64(session->id); + +send_reply: + if (ret < 0) { + reply.ret_code = htobe32(LTTNG_ERR_FATAL); + } else { + reply.ret_code = htobe32(LTTNG_OK); } - /* Cleanup index of that stream. */ - relay_index_destroy_by_stream_id(stream->stream_handle); + send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); + if (send_ret < 0) { + ERR("Relayd sending session id"); + ret = send_ret; + } - iter.iter.node = &stream->stream_n.node; - delret = lttng_ht_del(relay_streams_ht, &iter); - assert(!delret); - iter.iter.node = &stream->ctf_trace_node.node; - delret = lttng_ht_del(stream->ctf_traces_ht, &iter); - assert(!delret); - call_rcu(&stream->rcu_node, deferred_free_stream); - DBG("Closed tracefile %d from close stream", stream->fd); + return ret; } /* - * relay_delete_session: Free all memory associated with a session and - * close all the FDs + * When we have received all the streams and the metadata for a channel, + * we make them visible to the viewer threads. */ -static -void relay_delete_session(struct relay_command *cmd, - struct lttng_ht *sessions_ht) +static void publish_connection_local_streams(struct relay_connection *conn) { - struct lttng_ht_iter iter; - struct lttng_ht_node_ulong *node; struct relay_stream *stream; - int ret; - - if (!cmd->session) { - return; - } - - DBG("Relay deleting session %" PRIu64, cmd->session->id); + struct relay_session *session = conn->session; + /* + * We publish all streams belonging to a session atomically wrt + * session lock. + */ + pthread_mutex_lock(&session->lock); rcu_read_lock(); - cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, node, node) { - node = lttng_ht_iter_get_node_ulong(&iter); - if (!node) { - continue; - } - stream = caa_container_of(node, struct relay_stream, stream_n); - if (stream->session == cmd->session) { - destroy_stream(stream); - cmd->session->stream_count--; - assert(cmd->session->stream_count >= 0); - } + cds_list_for_each_entry_rcu(stream, &session->recv_list, + recv_node) { + stream_publish(stream); } - - /* Make this session not visible anymore. */ - iter.iter.node = &cmd->session->session_n.node; - ret = lttng_ht_del(sessions_ht, &iter); - assert(!ret); - call_rcu(&cmd->session->rcu_node, deferred_free_session); rcu_read_unlock(); -} - -/* - * Copy index data from the control port to a given index object. - */ -static void copy_index_control_data(struct relay_index *index, - struct lttcomm_relayd_index *data) -{ - 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. + * Inform the viewer that there are new streams in the session. */ - 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; + if (session->viewer_attached) { + uatomic_set(&session->new_streams, 1); + } + pthread_mutex_unlock(&session->lock); } /* - * Handle the RELAYD_CREATE_SESSION command. - * - * On success, send back the session id or else return a negative value. + * relay_add_stream: allocate a new stream for a session */ -static -int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_command *cmd, - struct lttng_ht *sessions_ht) +static int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn) { - int ret = 0, send_ret; - struct relay_session *session; - struct lttcomm_relayd_status_session reply; - - assert(recv_hdr); - assert(cmd); - - memset(&reply, 0, sizeof(reply)); + int ret; + ssize_t send_ret; + struct relay_session *session = conn->session; + struct relay_stream *stream = NULL; + struct lttcomm_relayd_status_stream reply; + struct ctf_trace *trace = NULL; + uint64_t stream_handle = -1ULL; + char *path_name = NULL, *channel_name = NULL; + uint64_t tracefile_size = 0, tracefile_count = 0; - session = zmalloc(sizeof(struct relay_session)); - if (session == NULL) { - PERROR("relay session zmalloc"); + if (!session || conn->version_check_done == 0) { + ERR("Trying to add a stream before version check"); ret = -1; - goto error; + goto end_no_session; } - session->id = ++last_relay_session_id; - session->sock = cmd->sock; - session->minor = cmd->minor; - session->major = cmd->major; - cmd->session = session; - - reply.session_id = htobe64(session->id); + switch (session->minor) { + case 1: /* LTTng sessiond 2.1. Allocates path_name and channel_name. */ + ret = cmd_recv_stream_2_1(conn, &path_name, + &channel_name); + break; + case 2: /* LTTng sessiond 2.2. Allocates path_name and channel_name. */ + default: + ret = cmd_recv_stream_2_2(conn, &path_name, + &channel_name, &tracefile_size, &tracefile_count); + break; + } + if (ret < 0) { + goto send_reply; + } - switch (cmd->minor) { - case 4: /* LTTng sessiond 2.4 */ - default: - ret = cmd_create_session_2_4(cmd, session); - break; + trace = ctf_trace_get_by_path_or_create(session, path_name); + if (!trace) { + goto send_reply; } + /* This stream here has one reference on the trace. */ - lttng_ht_node_init_ulong(&session->session_n, - (unsigned long) session->id); - lttng_ht_add_unique_ulong(sessions_ht, - &session->session_n); + pthread_mutex_lock(&last_relay_stream_id_lock); + stream_handle = ++last_relay_stream_id; + pthread_mutex_unlock(&last_relay_stream_id_lock); - DBG("Created session %" PRIu64, session->id); + /* We pass ownership of path_name and channel_name. */ + stream = stream_create(trace, stream_handle, path_name, + channel_name, tracefile_size, tracefile_count); + path_name = NULL; + channel_name = NULL; -error: - if (ret < 0) { - reply.ret_code = htobe32(LTTNG_ERR_FATAL); + /* + * Streams are the owners of their trace. Reference to trace is + * kept within stream_create(). + */ + ctf_trace_put(trace); + +send_reply: + memset(&reply, 0, sizeof(reply)); + reply.handle = htobe64(stream_handle); + if (!stream) { + reply.ret_code = htobe32(LTTNG_ERR_UNK); } else { reply.ret_code = htobe32(LTTNG_OK); } - send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); + send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, + sizeof(struct lttcomm_relayd_status_stream), 0); if (send_ret < 0) { - ERR("Relayd sending session id"); - ret = send_ret; + ERR("Relay sending stream id"); + ret = (int) send_ret; } +end_no_session: + free(path_name); + free(channel_name); return ret; } /* - * relay_add_stream: allocate a new stream for a session + * relay_close_stream: close a specific stream */ -static -int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_command *cmd, struct lttng_ht *sessions_ht) +static int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn) { - struct relay_session *session = cmd->session; - struct relay_stream *stream = NULL; - struct lttcomm_relayd_status_stream reply; 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; - if (!session || cmd->version_check_done == 0) { - ERR("Trying to add a stream before version check"); + DBG("Close stream received"); + + if (!session || conn->version_check_done == 0) { + ERR("Trying to close a stream before version check"); ret = -1; goto end_no_session; } - stream = zmalloc(sizeof(struct relay_stream)); - if (stream == NULL) { - PERROR("relay stream zmalloc"); + ret = conn->sock->ops->recvmsg(conn->sock, &stream_info, + sizeof(struct lttcomm_relayd_close_stream), 0); + if (ret < sizeof(struct lttcomm_relayd_close_stream)) { + if (ret == 0) { + /* Orderly shutdown. Not necessary to print an error. */ + DBG("Socket %d did an orderly shutdown", conn->sock->fd); + } else { + ERR("Relay didn't receive valid add_stream struct size : %d", ret); + } ret = -1; goto end_no_session; } - switch (cmd->minor) { - case 1: /* LTTng sessiond 2.1 */ - ret = cmd_recv_stream_2_1(cmd, stream); - break; - case 2: /* LTTng sessiond 2.2 */ - default: - ret = cmd_recv_stream_2_2(cmd, 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; - stream->index_fd = -1; - stream->read_index_fd = -1; - stream->ctf_trace = NULL; - 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"); + stream = stream_get_by_id(be64toh(stream_info.stream_id)); + if (!stream) { + ret = -1; goto end; } /* - * No need to use run_as API here because whatever we receives, the relayd - * uses its own credentials for the stream files. + * Set last_net_seq_num before the close flag. Required by data + * pending check. */ - ret = utils_create_stream_file(stream->path_name, stream->channel_name, - stream->tracefile_size, 0, relayd_uid, relayd_gid, NULL); - if (ret < 0) { - ERR("Create output file"); - goto end; - } - stream->fd = ret; - if (stream->tracefile_size) { - DBG("Tracefile %s/%s_0 created", stream->path_name, stream->channel_name); - } else { - DBG("Tracefile %s/%s created", stream->path_name, stream->channel_name); - } + pthread_mutex_lock(&stream->lock); + stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num); + pthread_mutex_unlock(&stream->lock); - if (!strncmp(stream->channel_name, DEFAULT_METADATA_NAME, NAME_MAX)) { - stream->metadata_flag = 1; - /* - * When we receive a new metadata stream, we create a new - * ctf_trace and we assign this ctf_trace to all streams with - * the same path. - * - * If later on we receive a new stream for the same ctf_trace, - * we copy the information from the first hit in the HT to the - * new stream. - */ - stream->ctf_trace = ctf_trace_create(); - if (!stream->ctf_trace) { - ret = -1; - goto end; + /* + * This is one of the conditions which may trigger a stream close + * with the others being: + * 1) A close command is received for a stream + * 2) The control connection owning the stream is closed + * 3) We have received all of the stream's data _after_ a close + * request. + */ + try_stream_close(stream); + if (stream->is_metadata) { + struct relay_viewer_stream *vstream; + + vstream = viewer_stream_get_by_id(stream->stream_handle); + if (vstream) { + if (vstream->metadata_sent == stream->metadata_received) { + /* + * Since all the metadata has been sent to the + * viewer and that we have a request to close + * its stream, we can safely teardown the + * corresponding metadata viewer stream. + */ + viewer_stream_put(vstream); + } + /* Put local reference. */ + viewer_stream_put(vstream); } - stream->ctf_trace->refcount++; - stream->ctf_trace->metadata_stream = stream; } - ctf_trace_assign(cmd->ctf_traces_ht, stream); - stream->ctf_traces_ht = cmd->ctf_traces_ht; - - lttng_ht_node_init_ulong(&stream->stream_n, - (unsigned long) stream->stream_handle); - lttng_ht_add_unique_ulong(relay_streams_ht, - &stream->stream_n); - - 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); + stream_put(stream); end: - reply.handle = htobe64(stream->stream_handle); - /* send the session id to the client or a negative return code on error */ + memset(&reply, 0, sizeof(reply)); if (ret < 0) { reply.ret_code = htobe32(LTTNG_ERR_UNK); - /* stream was not properly added to the ht, so free it */ - free(stream); } else { reply.ret_code = htobe32(LTTNG_OK); } - - send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, - sizeof(struct lttcomm_relayd_status_stream), 0); + 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; } - rcu_read_unlock(); end_no_session: return ret; - -err_free_stream: - free(stream->path_name); - free(stream->channel_name); - free(stream); - return ret; } /* - * relay_close_stream: close a specific stream + * relay_reset_metadata: reset a metadata stream */ static -int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_command *cmd) +int relay_reset_metadata(struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn) { int ret, send_ret; - struct relay_session *session = cmd->session; - struct lttcomm_relayd_close_stream stream_info; + struct relay_session *session = conn->session; + struct lttcomm_relayd_reset_metadata stream_info; struct lttcomm_relayd_generic_reply reply; struct relay_stream *stream; - DBG("Close stream received"); + DBG("Reset metadata received"); - if (!session || cmd->version_check_done == 0) { - ERR("Trying to close a stream before version check"); + if (!session || conn->version_check_done == 0) { + ERR("Trying to reset a metadata 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), 0); - if (ret < sizeof(struct lttcomm_relayd_close_stream)) { + ret = conn->sock->ops->recvmsg(conn->sock, &stream_info, + sizeof(struct lttcomm_relayd_reset_metadata), 0); + if (ret < sizeof(struct lttcomm_relayd_reset_metadata)) { if (ret == 0) { /* Orderly shutdown. Not necessary to print an error. */ - DBG("Socket %d did an orderly shutdown", cmd->sock->fd); + DBG("Socket %d did an orderly shutdown", conn->sock->fd); } else { - ERR("Relay didn't receive valid add_stream struct size : %d", ret); + ERR("Relay didn't receive valid reset_metadata struct " + "size : %d", ret); } ret = -1; goto end_no_session; } + DBG("Update metadata to version %" PRIu64, be64toh(stream_info.version)); - rcu_read_lock(); - stream = relay_stream_find_by_id(be64toh(stream_info.stream_id)); + /* Unsupported for live sessions for now. */ + if (session->live_timer != 0) { + ret = -1; + goto end; + } + + stream = stream_get_by_id(be64toh(stream_info.stream_id)); if (!stream) { + ret = -1; + goto end; + } + pthread_mutex_lock(&stream->lock); + if (!stream->is_metadata) { ret = -1; goto end_unlock; } - stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num); - stream->close_flag = 1; - session->stream_count--; - assert(session->stream_count >= 0); - - if (close_stream_check(stream)) { - destroy_stream(stream); + ret = utils_rotate_stream_file(stream->path_name, stream->channel_name, + 0, 0, -1, -1, stream->stream_fd->fd, NULL, + &stream->stream_fd->fd); + if (ret < 0) { + ERR("Failed to rotate metadata file %s of channel %s", + stream->path_name, stream->channel_name); + goto end_unlock; } end_unlock: - rcu_read_unlock(); + pthread_mutex_unlock(&stream->lock); + stream_put(stream); +end: + memset(&reply, 0, sizeof(reply)); if (ret < 0) { reply.ret_code = htobe32(LTTNG_ERR_UNK); } 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"); + ERR("Relay sending reset metadata reply"); ret = send_ret; } @@ -1197,14 +1434,14 @@ end_no_session: /* * relay_unknown_command: send -1 if received unknown command */ -static -void relay_unknown_command(struct relay_command *cmd) +static 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"); @@ -1215,21 +1452,21 @@ void relay_unknown_command(struct relay_command *cmd) * relay_start: send an acknowledgment to the client to tell if we are * ready to receive data. We are ready if a session is established. */ -static -int relay_start(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_command *cmd) +static int relay_start(struct lttcomm_relayd_hdr *recv_hdr, + 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"); @@ -1269,151 +1506,400 @@ end: } /* - * relay_recv_metadata: receive the metada for the session. + * Close the current index file if it is open, and create a new one. + * + * Return 0 on success, -1 on error. */ static -int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_command *cmd) +int create_rotate_index_file(struct relay_stream *stream) { - 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; - uint64_t data_size, payload_size; - - if (!session) { - ERR("Metadata sent before version check"); - ret = -1; - goto end; - } - - data_size = payload_size = be64toh(recv_hdr->data_size); - if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) { - ERR("Incorrect data size"); + int ret; + uint32_t major, minor; + + /* Put ref on previous index_file. */ + if (stream->index_file) { + lttng_index_file_put(stream->index_file); + stream->index_file = NULL; + } + major = stream->trace->session->major; + minor = stream->trace->session->minor; + stream->index_file = lttng_index_file_create(stream->path_name, + stream->channel_name, + -1, -1, stream->tracefile_size, + tracefile_array_get_file_index_head(stream->tfa), + lttng_to_index_major(major, minor), + lttng_to_index_minor(major, minor)); + if (!stream->index_file) { ret = -1; goto end; } - payload_size -= sizeof(struct lttcomm_relayd_metadata_payload); - if (data_buffer_size < data_size) { - /* In case the realloc fails, we can free the memory */ - char *tmp_data_ptr; + ret = 0; - 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, 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", cmd->sock->fd); - } else { - ERR("Relay didn't receive the whole metadata"); - } - ret = -1; - goto end; - } - metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer; +end: + return ret; +} - rcu_read_lock(); - metadata_stream = relay_stream_find_by_id( - be64toh(metadata_struct->stream_id)); - if (!metadata_stream) { - ret = -1; - goto end_unlock; - } +static +int do_rotate_stream(struct relay_stream *stream) +{ + int ret; - 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; + /* Perform the stream rotation. */ + ret = utils_rotate_stream_file(stream->path_name, + stream->channel_name, stream->tracefile_size, + stream->tracefile_count, -1, + -1, stream->stream_fd->fd, + NULL, &stream->stream_fd->fd); + if (ret < 0) { + ERR("Rotating stream output file"); + goto end; } + stream->tracefile_size_current = 0; - ret = write_padding_to_file(metadata_stream->fd, - be32toh(metadata_struct->padding_size)); - if (ret < 0) { - goto end_unlock; + /* Rotate also the index if the stream is not a metadata stream. */ + if (!stream->is_metadata) { + ret = create_rotate_index_file(stream); + if (ret < 0) { + ERR("Failed to rotate index file"); + goto end; + } } - metadata_stream->ctf_trace->metadata_received += - payload_size + be32toh(metadata_struct->padding_size); - DBG2("Relay metadata written"); + stream->rotate_at_seq_num = -1ULL; + stream->pos_after_last_complete_data_index = 0; -end_unlock: - rcu_read_unlock(); end: return ret; } /* - * relay_send_version: send relayd version number + * If too much data has been written in a tracefile before we received the + * rotation command, we have to move the excess data to the new tracefile and + * perform the rotation. This can happen because the control and data + * connections are separate, the indexes as well as the commands arrive from + * the control connection and we have no control over the order so we could be + * in a situation where too much data has been received on the data connection + * before the rotation command on the control connection arrives. We don't need + * to update the index because its order is guaranteed with the rotation + * command message. */ static -int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_command *cmd, struct lttng_ht *sessions_ht) +int rotate_truncate_stream(struct relay_stream *stream) { - int ret; - struct lttcomm_relayd_version reply, msg; - - assert(cmd); - - cmd->version_check_done = 1; - - /* Get version from the other side. */ - ret = cmd->sock->ops->recvmsg(cmd->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", cmd->sock->fd); - } else { - ERR("Relay failed to receive the version values."); - } + int ret, new_fd; + uint64_t diff, pos = 0; + char buf[FILE_COPY_BUFFER_SIZE]; + + assert(!stream->is_metadata); + + assert(stream->tracefile_size_current > + stream->pos_after_last_complete_data_index); + diff = stream->tracefile_size_current - + stream->pos_after_last_complete_data_index; + + /* Create the new tracefile. */ + new_fd = utils_create_stream_file(stream->path_name, + stream->channel_name, + stream->tracefile_size, stream->tracefile_count, + /* uid */ -1, /* gid */ -1, /* suffix */ NULL); + if (new_fd < 0) { + ERR("Failed to create new stream file at path %s for channel %s", + stream->path_name, stream->channel_name); ret = -1; goto end; } - reply.major = RELAYD_VERSION_COMM_MAJOR; - reply.minor = RELAYD_VERSION_COMM_MINOR; + /* + * Rewind the current tracefile to the position at which the rotation + * should have occured. + */ + ret = lseek(stream->stream_fd->fd, + stream->pos_after_last_complete_data_index, SEEK_SET); + if (ret < 0) { + PERROR("seek truncate stream"); + goto end; + } + + /* Move data from the old file to the new file. */ + while (pos < diff) { + uint64_t count, bytes_left; + ssize_t io_ret; + + bytes_left = diff - pos; + count = bytes_left > sizeof(buf) ? sizeof(buf) : bytes_left; + assert(count <= SIZE_MAX); + + io_ret = lttng_read(stream->stream_fd->fd, buf, count); + if (io_ret < (ssize_t) count) { + char error_string[256]; + + snprintf(error_string, sizeof(error_string), + "Failed to read %" PRIu64 " bytes from fd %i in rotate_truncate_stream(), returned %zi", + count, stream->stream_fd->fd, io_ret); + if (io_ret == -1) { + PERROR("%s", error_string); + } else { + ERR("%s", error_string); + } + ret = -1; + goto end; + } + + io_ret = lttng_write(new_fd, buf, count); + if (io_ret < (ssize_t) count) { + char error_string[256]; + + snprintf(error_string, sizeof(error_string), + "Failed to write %" PRIu64 " bytes from fd %i in rotate_truncate_stream(), returned %zi", + count, new_fd, io_ret); + if (io_ret == -1) { + PERROR("%s", error_string); + } else { + ERR("%s", error_string); + } + ret = -1; + goto end; + } + + pos += count; + } + + /* Truncate the file to get rid of the excess data. */ + ret = ftruncate(stream->stream_fd->fd, + stream->pos_after_last_complete_data_index); + if (ret) { + PERROR("ftruncate"); + goto end; + } + + ret = close(stream->stream_fd->fd); + if (ret < 0) { + PERROR("Closing tracefile"); + goto end; + } + + ret = create_rotate_index_file(stream); + if (ret < 0) { + ERR("Rotate stream index file"); + goto end; + } + + /* + * Update the offset and FD of all the eventual indexes created by the + * data connection before the rotation command arrived. + */ + ret = relay_index_switch_all_files(stream); + if (ret < 0) { + ERR("Failed to rotate index file"); + goto end; + } + + stream->stream_fd->fd = new_fd; + stream->tracefile_size_current = diff; + stream->pos_after_last_complete_data_index = 0; + stream->rotate_at_seq_num = -1ULL; + + ret = 0; + +end: + return ret; +} + +/* + * Check if a stream should perform a rotation (for session rotation). + * Must be called with the stream lock held. + * + * Return 0 on success, a negative value on error. + */ +static +int try_rotate_stream(struct relay_stream *stream) +{ + int ret = 0; + + /* No rotation expected. */ + if (stream->rotate_at_seq_num == -1ULL) { + goto end; + } + + if (stream->prev_seq < stream->rotate_at_seq_num) { + DBG("Stream %" PRIu64 " no yet ready for rotation", + stream->stream_handle); + goto end; + } else if (stream->prev_seq > stream->rotate_at_seq_num) { + DBG("Rotation after too much data has been written in tracefile " + "for stream %" PRIu64 ", need to truncate before " + "rotating", stream->stream_handle); + ret = rotate_truncate_stream(stream); + if (ret) { + ERR("Failed to truncate stream"); + goto end; + } + } else { + /* stream->prev_seq == stream->rotate_at_seq_num */ + DBG("Stream %" PRIu64 " ready for rotation", + stream->stream_handle); + ret = do_rotate_stream(stream); + } + +end: + return ret; +} + +/* + * relay_recv_metadata: receive the metadata for the session. + */ +static int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn) +{ + int ret = 0; + 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; + + if (!session) { + ERR("Metadata sent before version check"); + ret = -1; + goto end; + } + + data_size = payload_size = be64toh(recv_hdr->data_size); + if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) { + ERR("Incorrect data size"); + ret = -1; + goto end; + } + payload_size -= sizeof(struct lttcomm_relayd_metadata_payload); + + if (data_buffer_size < data_size) { + /* In case the realloc fails, we can free the memory */ + char *tmp_data_ptr; + + 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); + size_ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, data_size, 0); + if (size_ret < 0 || size_ret != data_size) { + if (size_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; + goto end; + } + metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer; + + metadata_stream = stream_get_by_id(be64toh(metadata_struct->stream_id)); + if (!metadata_stream) { + ret = -1; + goto end; + } + + pthread_mutex_lock(&metadata_stream->lock); + + size_ret = lttng_write(metadata_stream->stream_fd->fd, metadata_struct->payload, + payload_size); + if (size_ret < payload_size) { + ERR("Relay error writing metadata on file"); + ret = -1; + goto end_put; + } + + size_ret = write_padding_to_file(metadata_stream->stream_fd->fd, + be32toh(metadata_struct->padding_size)); + if (size_ret < 0) { + goto end_put; + } + + metadata_stream->metadata_received += + payload_size + be32toh(metadata_struct->padding_size); + DBG2("Relay metadata written. Updated metadata_received %" PRIu64, + metadata_stream->metadata_received); + + ret = try_rotate_stream(metadata_stream); + if (ret < 0) { + goto end_put; + } + +end_put: + pthread_mutex_unlock(&metadata_stream->lock); + stream_put(metadata_stream); +end: + return ret; +} + +/* + * relay_send_version: send relayd version number + */ +static int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn) +{ + int ret; + struct lttcomm_relayd_version reply, msg; + bool compatible = true; + + 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."); + } + 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)); - relay_delete_session(cmd, sessions_ht); - ret = 0; - goto end; + compatible = false; } - cmd->major = reply.major; + conn->major = reply.major; /* We adapt to the lowest compatible version */ if (reply.minor <= be32toh(msg.minor)) { - cmd->minor = reply.minor; + conn->minor = reply.minor; } else { - cmd->minor = be32toh(msg.minor); + conn->minor = be32toh(msg.minor); } 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 using protocol %u.%u", cmd->major, - cmd->minor); + if (!compatible) { + ret = -1; + goto end; + } + + DBG("Version check done using protocol %u.%u", conn->major, + conn->minor); end: return ret; @@ -1422,11 +1908,10 @@ end: /* * Check for data pending for a given stream id from the session daemon. */ -static -int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_command *cmd) +static int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn) { - struct relay_session *session = cmd->session; + struct relay_session *session = conn->session; struct lttcomm_relayd_data_pending msg; struct lttcomm_relayd_generic_reply reply; struct relay_stream *stream; @@ -1435,17 +1920,17 @@ int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr, DBG("Data pending command received"); - if (!session || cmd->version_check_done == 0) { + if (!session || conn->version_check_done == 0) { ERR("Trying to check for data before version check"); ret = -1; goto end_no_session; } - ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); + 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", cmd->sock->fd); + DBG("Socket %d did an orderly shutdown", conn->sock->fd); } else { ERR("Relay didn't receive valid data_pending struct size : %d", ret); @@ -1457,13 +1942,14 @@ int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr, stream_id = be64toh(msg.stream_id); last_net_seq_num = be64toh(msg.last_net_seq_num); - rcu_read_lock(); - stream = relay_stream_find_by_id(stream_id); + stream = stream_get_by_id(stream_id); if (stream == NULL) { ret = -1; - goto end_unlock; + goto end; } + pthread_mutex_lock(&stream->lock); + DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64 " and last_seq %" PRIu64, stream_id, stream->prev_seq, last_net_seq_num); @@ -1477,14 +1963,15 @@ int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr, ret = 1; } - /* Pending check is now done. */ - stream->data_pending_check_done = 1; + stream->data_pending_check_done = true; + pthread_mutex_unlock(&stream->lock); -end_unlock: - rcu_read_unlock(); + stream_put(stream); +end: + memset(&reply, 0, sizeof(reply)); reply.ret_code = htobe32(ret); - ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); + ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); if (ret < 0) { ERR("Relay data pending ret code failed"); } @@ -1496,34 +1983,33 @@ end_no_session: /* * Wait for the control socket to reach a quiescent state. * - * Note that for now, when receiving this command from the session daemon, this - * means that every subsequent commands or data received on the control socket - * has been handled. So, this is why we simply return OK here. + * Note that for now, when receiving this command from the session + * daemon, this means that every subsequent commands or data received on + * the control socket has been handled. So, this is why we simply return + * OK here. */ -static -int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_command *cmd) +static 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; DBG("Checking quiescent state on control socket"); - if (!cmd->session || cmd->version_check_done == 0) { + if (!conn->session || conn->version_check_done == 0) { ERR("Trying to check for data before version check"); ret = -1; goto end_no_session; } - ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); + 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", cmd->sock->fd); + DBG("Socket %d did an orderly shutdown", conn->sock->fd); } else { ERR("Relay didn't receive valid begin data_pending struct size: %d", ret); @@ -1533,21 +2019,19 @@ int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr, } stream_id = be64toh(msg.stream_id); - - rcu_read_lock(); - cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream, - stream_n.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(); - + stream = stream_get_by_id(stream_id); + if (!stream) { + goto reply; + } + pthread_mutex_lock(&stream->lock); + stream->data_pending_check_done = true; + pthread_mutex_unlock(&stream->lock); + DBG("Relay quiescent control pending flag set to %" PRIu64, stream_id); + stream_put(stream); +reply: + memset(&reply, 0, sizeof(reply)); reply.ret_code = htobe32(LTTNG_OK); - ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); + ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); if (ret < 0) { ERR("Relay data quiescent control ret code failed"); } @@ -1557,15 +2041,14 @@ end_no_session: } /* - * Initialize a data pending command. This means that a client is about to ask - * for data pending for each stream he/she holds. Simply iterate over all - * streams of a session and set the data_pending_check_done flag. + * Initialize a data pending command. This means that a consumer is about + * to ask for data pending for each stream it holds. Simply iterate over + * all streams of a session and set the data_pending_check_done flag. * * This command returns to the client a LTTNG_OK code. */ -static -int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_command *cmd) +static int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn) { int ret; struct lttng_ht_iter iter; @@ -1575,21 +2058,21 @@ int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr, uint64_t session_id; assert(recv_hdr); - assert(cmd); + assert(conn); DBG("Init streams for data pending"); - if (!cmd->session || cmd->version_check_done == 0) { + if (!conn->session || conn->version_check_done == 0) { ERR("Trying to check for data before version check"); ret = -1; goto end_no_session; } - ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); + 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", cmd->sock->fd); + DBG("Socket %d did an orderly shutdown", conn->sock->fd); } else { ERR("Relay didn't receive valid begin data_pending struct size: %d", ret); @@ -1601,25 +2084,33 @@ int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr, session_id = be64toh(msg.session_id); /* - * Iterate over all streams to set the begin data pending flag. For now, the - * streams are indexed by stream handle so we have to iterate over all - * streams to find the one associated with the right session_id. + * Iterate over all streams to set the begin data pending flag. + * For now, the streams are indexed by stream handle so we have + * to iterate over all streams to find the one associated with + * the right session_id. */ rcu_read_lock(); cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream, - stream_n.node) { - if (stream->session->id == session_id) { - stream->data_pending_check_done = 0; + node.node) { + if (!stream_get(stream)) { + continue; + } + if (stream->trace->session->id == session_id) { + pthread_mutex_lock(&stream->lock); + stream->data_pending_check_done = false; + pthread_mutex_unlock(&stream->lock); DBG("Set begin data pending flag to stream %" PRIu64, stream->stream_handle); } + stream_put(stream); } rcu_read_unlock(); + memset(&reply, 0, sizeof(reply)); /* All good, send back reply. */ reply.ret_code = htobe32(LTTNG_OK); - ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); + ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); if (ret < 0) { ERR("Relay begin data pending send reply failed"); } @@ -1629,17 +2120,16 @@ end_no_session: } /* - * End data pending command. This will check, for a given session id, if each - * stream associated with it has its data_pending_check_done flag set. If not, - * this means that the client lost track of the stream but the data is still - * being streamed on our side. In this case, we inform the client that data is - * inflight. + * End data pending command. This will check, for a given session id, if + * each stream associated with it has its data_pending_check_done flag + * set. If not, this means that the client lost track of the stream but + * the data is still being streamed on our side. In this case, we inform + * the client that data is in flight. * * Return to the client if there is data in flight or not with a ret_code. */ -static -int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_command *cmd) +static int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn) { int ret; struct lttng_ht_iter iter; @@ -1649,22 +2139,19 @@ int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr, uint64_t session_id; uint32_t is_data_inflight = 0; - assert(recv_hdr); - assert(cmd); - DBG("End data pending command"); - if (!cmd->session || cmd->version_check_done == 0) { + if (!conn->session || conn->version_check_done == 0) { ERR("Trying to check for data before version check"); ret = -1; goto end_no_session; } - ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); + 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", cmd->sock->fd); + DBG("Socket %d did an orderly shutdown", conn->sock->fd); } else { ERR("Relay didn't receive valid end data_pending struct size: %d", ret); @@ -1675,24 +2162,41 @@ int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr, session_id = be64toh(msg.session_id); - /* Iterate over all streams to see if the begin data pending flag is set. */ + /* + * Iterate over all streams to see if the begin data pending + * flag is set. + */ rcu_read_lock(); cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream, - stream_n.node) { - if (stream->session->id == session_id && - !stream->data_pending_check_done) { - is_data_inflight = 1; - DBG("Data is still in flight for stream %" PRIu64, - stream->stream_handle); - break; + node.node) { + if (!stream_get(stream)) { + continue; + } + if (stream->trace->session->id != session_id) { + stream_put(stream); + continue; + } + pthread_mutex_lock(&stream->lock); + if (!stream->data_pending_check_done) { + if (!stream->closed || !(((int64_t) (stream->prev_seq - stream->last_net_seq_num)) >= 0)) { + is_data_inflight = 1; + DBG("Data is still in flight for stream %" PRIu64, + stream->stream_handle); + pthread_mutex_unlock(&stream->lock); + stream_put(stream); + break; + } } + pthread_mutex_unlock(&stream->lock); + stream_put(stream); } rcu_read_unlock(); + memset(&reply, 0, sizeof(reply)); /* All good, send back reply. */ reply.ret_code = htobe32(is_data_inflight); - ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); + ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); if (ret < 0) { ERR("Relay end data pending send reply failed"); } @@ -1706,34 +2210,37 @@ end_no_session: * * Return 0 on success else a negative value. */ -static -int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_command *cmd) +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 = cmd->session; + int ret, send_ret; + struct relay_session *session = conn->session; struct lttcomm_relayd_index index_info; - struct relay_index *index, *wr_index = NULL; + struct relay_index *index; struct lttcomm_relayd_generic_reply reply; struct relay_stream *stream; uint64_t net_seq_num; + size_t msg_len; - assert(cmd); + assert(conn); DBG("Relay receiving index"); - if (!session || cmd->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, &index_info, - sizeof(index_info), 0); - if (ret < sizeof(index_info)) { + msg_len = lttcomm_relayd_index_len( + lttng_to_index_major(conn->major, conn->minor), + lttng_to_index_minor(conn->major, conn->minor)); + ret = conn->sock->ops->recvmsg(conn->sock, &index_info, + msg_len, 0); + if (ret < msg_len) { if (ret == 0) { /* Orderly shutdown. Not necessary to print an error. */ - DBG("Socket %d did an orderly shutdown", cmd->sock->fd); + DBG("Socket %d did an orderly shutdown", conn->sock->fd); } else { ERR("Relay didn't receive valid index struct size : %d", ret); } @@ -1743,142 +2250,715 @@ int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr, net_seq_num = be64toh(index_info.net_seq_num); - rcu_read_lock(); - stream = relay_stream_find_by_id(be64toh(index_info.relay_stream_id)); + stream = stream_get_by_id(be64toh(index_info.relay_stream_id)); if (!stream) { + ERR("stream_get_by_id not found"); ret = -1; - goto end_rcu_unlock; + goto end; } + pthread_mutex_lock(&stream->lock); /* Live beacon handling */ if (index_info.packet_size == 0) { - DBG("Received live beacon for stream %" PRIu64, stream->stream_handle); + DBG("Received live beacon for stream %" PRIu64, + stream->stream_handle); /* - * Only flag a stream inactive when it has already received data. + * Only flag a stream inactive when it has already + * received data and no indexes are in flight. */ - if (stream->total_index_received > 0) { - stream->beacon_ts_end = be64toh(index_info.timestamp_end); + if (stream->index_received_seqcount > 0 + && stream->indexes_in_flight == 0) { + stream->beacon_ts_end = + be64toh(index_info.timestamp_end); } ret = 0; - goto end_rcu_unlock; + goto end_stream_put; } else { stream->beacon_ts_end = -1ULL; } - index = relay_index_find(stream->stream_handle, net_seq_num); + if (stream->ctf_stream_id == -1ULL) { + stream->ctf_stream_id = be64toh(index_info.stream_id); + } + index = relay_index_get_by_id_or_create(stream, net_seq_num); if (!index) { - /* A successful creation will add the object to the HT. */ - index = relay_index_create(stream->stream_handle, net_seq_num); - if (!index) { - goto end_rcu_unlock; + ret = -1; + ERR("relay_index_get_by_id_or_create index NULL"); + goto end_stream_put; + } + if (set_index_control_data(index, &index_info, conn)) { + ERR("set_index_control_data error"); + relay_index_put(index); + ret = -1; + goto end_stream_put; + } + ret = relay_index_try_flush(index); + if (ret == 0) { + tracefile_array_commit_seq(stream->tfa); + stream->index_received_seqcount++; + stream->pos_after_last_complete_data_index += index->total_size; + } else if (ret > 0) { + /* no flush. */ + ret = 0; + } else { + ERR("relay_index_try_flush error %d", ret); + relay_index_put(index); + ret = -1; + } + +end_stream_put: + pthread_mutex_unlock(&stream->lock); + stream_put(stream); + +end: + + memset(&reply, 0, sizeof(reply)); + if (ret < 0) { + reply.ret_code = htobe32(LTTNG_ERR_UNK); + } else { + reply.ret_code = htobe32(LTTNG_OK); + } + send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); + if (send_ret < 0) { + ERR("Relay sending close index id reply"); + ret = send_ret; + } + +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; + } + + /* + * Publish every pending stream in the connection recv list which are + * now ready to be used by the viewer. + */ + publish_connection_local_streams(conn); + + memset(&reply, 0, sizeof(reply)); + reply.ret_code = htobe32(LTTNG_OK); + send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); + if (send_ret < 0) { + ERR("Relay sending sent_stream reply"); + ret = send_ret; + } else { + /* Success. */ + ret = 0; + } + +end_no_session: + return ret; +} + +/* + * relay_rotate_stream: rotate a stream to a new tracefile for the session + * rotation feature (not the tracefile rotation feature). + */ +static int relay_rotate_session_stream(struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn) +{ + int ret, send_ret; + struct relay_session *session = conn->session; + struct lttcomm_relayd_rotate_stream stream_info; + struct lttcomm_relayd_generic_reply reply; + struct relay_stream *stream; + size_t len; + char *new_pathname = NULL; + + DBG("Rotate stream received"); + + if (!session || !conn->version_check_done) { + ERR("Trying to rotate a stream before version check"); + ret = -1; + goto end_no_reply; + } + + if (session->major == 2 && session->minor < 11) { + ERR("Unsupported feature before 2.11"); + ret = -1; + goto end_no_reply; + } + + memset(&stream_info, 0, sizeof(struct lttcomm_relayd_rotate_stream)); + + /* + * Receive the struct up to the new_pathname member since we don't know + * its size yet. + */ + ret = conn->sock->ops->recvmsg(conn->sock, &stream_info, + sizeof(struct lttcomm_relayd_rotate_stream), 0); + if (ret < sizeof(struct lttcomm_relayd_rotate_stream)) { + if (ret == 0) { + /* Orderly shutdown. Not necessary to print an error. */ + DBG("Socket %d did an orderly shutdown", conn->sock->fd); + } else { + ERR("Relay didn't receive valid rotate_stream struct size : %d", ret); + } + ret = -1; + goto end_no_reply; + } + + stream = stream_get_by_id(be64toh(stream_info.stream_id)); + if (!stream) { + ret = -1; + goto end; + } + + len = be32toh(stream_info.pathname_length); + /* Ensure it fits in local filename length. */ + if (len >= LTTNG_PATH_MAX) { + ret = -ENAMETOOLONG; + ERR("Length of relay_rotate_session_stream command's path name (%zu bytes) exceeds the maximal allowed length of %i bytes", + len, LTTNG_PATH_MAX); + goto end; + } + + new_pathname = zmalloc(len); + if (!new_pathname) { + PERROR("Failed to allocation new path name of relay_rotate_session_stream command"); + ret = -1; + goto end; + } + + ret = conn->sock->ops->recvmsg(conn->sock, new_pathname, len, 0); + if (ret < len) { + 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 rotate_stream struct size : %d", ret); } - index_created = 1; + ret = -1; + goto end_no_reply; + } + + pthread_mutex_lock(&stream->lock); + + /* + * Update the trace path (just the folder, the stream name does not + * change). + */ + free(stream->path_name); + stream->path_name = create_output_path(new_pathname); + if (!stream->path_name) { + ERR("Failed to create a new output path"); + goto end_stream_unlock; + } + ret = utils_mkdir_recursive(stream->path_name, S_IRWXU | S_IRWXG, + -1, -1); + if (ret < 0) { + ERR("relay creating output directory"); + goto end_stream_unlock; + } + stream->chunk_id = be64toh(stream_info.new_chunk_id); + + if (stream->is_metadata) { + /* + * The metadata stream is sent only over the control connection + * so we know we have all the data to perform the stream + * rotation. + */ + ret = do_rotate_stream(stream); + } else { + stream->rotate_at_seq_num = be64toh(stream_info.rotate_at_seq_num); + ret = try_rotate_stream(stream); } + if (ret < 0) { + goto end_stream_unlock; + } + +end_stream_unlock: + pthread_mutex_unlock(&stream->lock); + stream_put(stream); +end: + memset(&reply, 0, sizeof(reply)); + if (ret < 0) { + reply.ret_code = htobe32(LTTNG_ERR_UNK); + } else { + reply.ret_code = htobe32(LTTNG_OK); + } + send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, + sizeof(struct lttcomm_relayd_generic_reply), 0); + if (send_ret < 0) { + ERR("Failed to send reply of rotate session stream command"); + ret = send_ret; + } + +end_no_reply: + free(new_pathname); + return ret; +} + +/* + * relay_mkdir: Create a folder on the disk. + */ +static int relay_mkdir(struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn) +{ + int ret; + ssize_t network_ret; + struct relay_session *session = conn->session; + struct lttcomm_relayd_mkdir path_info_header; + struct lttcomm_relayd_mkdir *path_info = NULL; + struct lttcomm_relayd_generic_reply reply; + char *path = NULL; - copy_index_control_data(index, &index_info); + if (!session || !conn->version_check_done) { + ERR("Trying to create a directory before version check"); + ret = -1; + goto end_no_session; + } - if (index_created) { + if (session->major == 2 && session->minor < 11) { /* - * 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. + * This client is not supposed to use this command since + * it predates its introduction. */ - relay_index_add(index, &wr_index); - if (wr_index) { - copy_index_control_data(wr_index, &index_info); - free(index); + ERR("relay_mkdir command is unsupported before LTTng 2.11"); + ret = -1; + goto end_no_session; + } + + network_ret = conn->sock->ops->recvmsg(conn->sock, &path_info_header, + sizeof(path_info_header), 0); + if (network_ret < (ssize_t) sizeof(path_info_header)) { + if (network_ret == 0) { + /* Orderly shutdown. Not necessary to print an error. */ + DBG("Socket %d did an orderly shutdown", conn->sock->fd); + } else { + ERR("Reception of mkdir command argument length failed with ret = %zi, expected %zu", + network_ret, sizeof(path_info_header)); + } + ret = -1; + goto end_no_session; + } + + path_info_header.length = be32toh(path_info_header.length); + + /* Ensure that it fits in local path length. */ + if (path_info_header.length >= LTTNG_PATH_MAX) { + ret = -ENAMETOOLONG; + ERR("Path name argument of mkdir command (%" PRIu32 " bytes) exceeds the maximal length allowed (%d bytes)", + path_info_header.length, LTTNG_PATH_MAX); + goto end; + } + + path_info = zmalloc(sizeof(path_info_header) + path_info_header.length); + if (!path_info) { + PERROR("zmalloc of mkdir command path"); + ret = -1; + goto end; + } + + network_ret = conn->sock->ops->recvmsg(conn->sock, path_info->path, + path_info_header.length, 0); + if (network_ret < (ssize_t) path_info_header.length) { + if (network_ret == 0) { + /* Orderly shutdown. Not necessary to print an error. */ + DBG("Socket %d did an orderly shutdown", conn->sock->fd); + } else { + ERR("Reception of mkdir path argument failed with ret = %zi, expected %" PRIu32, + network_ret, path_info_header.length); + } + ret = -1; + goto end_no_session; + } + + path = create_output_path(path_info->path); + if (!path) { + ERR("Failed to create output path"); + ret = -1; + goto end; + } + + ret = utils_mkdir_recursive(path, S_IRWXU | S_IRWXG, -1, -1); + if (ret < 0) { + ERR("relay creating output directory"); + goto end; + } + + ret = 0; + +end: + memset(&reply, 0, sizeof(reply)); + if (ret < 0) { + reply.ret_code = htobe32(LTTNG_ERR_UNK); + } else { + reply.ret_code = htobe32(LTTNG_OK); + } + network_ret = conn->sock->ops->sendmsg(conn->sock, &reply, + sizeof(struct lttcomm_relayd_generic_reply), 0); + if (network_ret < (ssize_t) sizeof(struct lttcomm_relayd_generic_reply)) { + ERR("Failed to send mkdir command status code with ret = %zi, expected %zu", + network_ret, + sizeof(struct lttcomm_relayd_generic_reply)); + ret = -1; + } + +end_no_session: + free(path); + free(path_info); + return ret; +} + +static int validate_rotate_rename_path_length(const char *path_type, + uint32_t path_length) +{ + int ret = 0; + + if (path_length > LTTNG_PATH_MAX) { + ret = -ENAMETOOLONG; + ERR("rotate rename \"%s\" path name length (%" PRIu32 " bytes) exceeds the allowed size of %i bytes", + path_type, path_length, LTTNG_PATH_MAX); + } else if (path_length == 0) { + ret = -EINVAL; + ERR("rotate rename \"%s\" path name has an illegal length of 0", path_type); + } + return ret; +} + +/* + * relay_rotate_rename: rename the trace folder after a rotation is + * completed. We are not closing any fd here, just moving the folder, so it + * works even if data is still in-flight. + */ +static int relay_rotate_rename(struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn) +{ + int ret; + ssize_t network_ret; + struct relay_session *session = conn->session; + struct lttcomm_relayd_generic_reply reply; + struct lttcomm_relayd_rotate_rename header; + char *received_paths = NULL; + size_t received_paths_size; + const char *received_old_path, *received_new_path; + char *complete_old_path = NULL, *complete_new_path = NULL; + + if (!session || !conn->version_check_done) { + ERR("Trying to rename a trace folder before version check"); + ret = -1; + goto end_no_reply; + } + + if (session->major == 2 && session->minor < 11) { + ERR("relay_rotate_rename command is unsupported before LTTng 2.11"); + ret = -1; + goto end_no_reply; + } + + network_ret = conn->sock->ops->recvmsg(conn->sock, &header, + sizeof(header), 0); + if (network_ret < (ssize_t) sizeof(header)) { + if (network_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 a valid rotate_rename command header: expected %zu bytes, recvmsg() returned %zi", + sizeof(header), network_ret); + } + ret = -1; + goto end_no_reply; + } + + header.old_path_length = be32toh(header.old_path_length); + header.new_path_length = be32toh(header.new_path_length); + received_paths_size = header.old_path_length + header.new_path_length; + + /* Ensure the paths don't exceed their allowed size. */ + ret = validate_rotate_rename_path_length("old", header.old_path_length); + if (ret) { + goto end; + } + ret = validate_rotate_rename_path_length("new", header.new_path_length); + if (ret) { + goto end; + } + + received_paths = zmalloc(received_paths_size); + if (!received_paths) { + PERROR("Could not allocate rotate commands paths reception buffer"); + ret = -1; + goto end; + } + + network_ret = conn->sock->ops->recvmsg(conn->sock, received_paths, + received_paths_size, 0); + if (network_ret < (ssize_t) received_paths_size) { + if (network_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 received rename command paths (%zu bytes): recvmsg() returned %zi", + received_paths_size, network_ret); } + ret = -1; + goto end_no_reply; + } + + /* Validate that both paths received are NULL terminated. */ + if (received_paths[header.old_path_length - 1] != '\0') { + ERR("relay_rotate_rename command's \"old\" path is invalid (not NULL terminated)"); + ret = -1; + goto end; + } + if (received_paths[received_paths_size - 1] != '\0') { + ERR("relay_rotate_rename command's \"new\" path is invalid (not NULL terminated)"); + ret = -1; + goto end; + } + + received_old_path = received_paths; + received_new_path = received_paths + header.old_path_length; + + complete_old_path = create_output_path(received_old_path); + if (!complete_old_path) { + ERR("Failed to build old output path in rotate_rename command"); + ret = -1; + goto end; + } + + complete_new_path = create_output_path(received_new_path); + if (!complete_new_path) { + ERR("Failed to build new output path in rotate_rename command"); + ret = -1; + goto end; + } + + ret = utils_mkdir_recursive(complete_new_path, S_IRWXU | S_IRWXG, + -1, -1); + if (ret < 0) { + ERR("Failed to mkdir() rotate_rename's \"new\" output directory at \"%s\"", + complete_new_path); + goto end; + } + + /* + * If a domain has not yet created its channel, the domain-specific + * folder might not exist, but this is not an error. + */ + ret = rename(complete_old_path, complete_new_path); + if (ret < 0 && errno != ENOENT) { + PERROR("Renaming chunk in rotate_rename command from \"%s\" to \"%s\"", + complete_old_path, complete_new_path); + goto end; + } + ret = 0; + +end: + memset(&reply, 0, sizeof(reply)); + if (ret < 0) { + reply.ret_code = htobe32(LTTNG_ERR_UNK); } else { - /* The index already exists so write it on disk. */ - wr_index = index; + reply.ret_code = htobe32(LTTNG_OK); + } + network_ret = conn->sock->ops->sendmsg(conn->sock, &reply, + sizeof(struct lttcomm_relayd_generic_reply), 0); + if (network_ret < sizeof(struct lttcomm_relayd_generic_reply)) { + ERR("Relay sending stream id"); + ret = -1; } - /* Do we have a writable ready index to write on disk. */ - if (wr_index) { - /* Starting at 2.4, create the index file if none available. */ - if (cmd->minor >= 4 && stream->index_fd < 0) { - ret = index_create_file(stream->path_name, stream->channel_name, - relayd_uid, relayd_gid, stream->tracefile_size, - stream->tracefile_count_current); - if (ret < 0) { - goto end_rcu_unlock; - } - stream->index_fd = ret; +end_no_reply: + free(received_paths); + free(complete_old_path); + free(complete_new_path); + return ret; +} + +/* + * Check if all the streams in the session have completed the last rotation. + * The chunk_id value is used to distinguish the cases where a stream was + * closed on the consumerd before the rotation started but it still active on + * the relayd, and the case where a stream appeared on the consumerd/relayd + * after the last rotation started (in that case, it is already writing in the + * new chunk folder). + */ +static +int relay_rotate_pending(struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn) +{ + struct relay_session *session = conn->session; + struct lttcomm_relayd_rotate_pending msg; + struct lttcomm_relayd_generic_reply reply; + struct lttng_ht_iter iter; + struct relay_stream *stream; + int ret = 0; + ssize_t network_ret; + uint64_t chunk_id; + bool rotate_pending = false; + + DBG("Rotate pending command received"); + + if (!session || !conn->version_check_done) { + ERR("Trying to check for data before version check"); + ret = -1; + goto end_no_reply; + } + + if (session->major == 2 && session->minor < 11) { + ERR("Unsupported feature before 2.11"); + ret = -1; + goto end_no_reply; + } + + network_ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0); + if (network_ret < (ssize_t) sizeof(msg)) { + if (network_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 rotate_pending struct size : %zi", + network_ret); } + ret = -1; + goto end_no_reply; + } - ret = relay_index_write(wr_index->fd, wr_index); - if (ret < 0) { - goto end_rcu_unlock; + chunk_id = be64toh(msg.chunk_id); + DBG("Evaluating rotate pending for chunk id %" PRIu64, chunk_id); + + /* + * Iterate over all the streams in the session and check if they are + * still waiting for data to perform their rotation. + */ + rcu_read_lock(); + cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream, + node.node) { + if (!stream_get(stream)) { + continue; + } + if (stream->trace->session != session) { + stream_put(stream); + continue; + } + pthread_mutex_lock(&stream->lock); + if (stream->rotate_at_seq_num != -1ULL) { + /* We have not yet performed the rotation. */ + rotate_pending = true; + DBG("Stream %" PRIu64 " is still rotating", + stream->stream_handle); + } else if (stream->chunk_id < chunk_id) { + /* + * Stream closed on the consumer but still active on the + * relay. + */ + rotate_pending = true; + DBG("Stream %" PRIu64 " did not exist on the consumer " + "when the last rotation started, but is" + "still waiting for data before getting" + "closed", + stream->stream_handle); + } + pthread_mutex_unlock(&stream->lock); + stream_put(stream); + if (rotate_pending) { + goto send_reply; } - stream->total_index_received++; } -end_rcu_unlock: +send_reply: rcu_read_unlock(); - - 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, sizeof(reply), 0); - if (send_ret < 0) { - ERR("Relay sending close index id reply"); - ret = send_ret; + memset(&reply, 0, sizeof(reply)); + reply.ret_code = htobe32(rotate_pending ? 1 : 0); + network_ret = conn->sock->ops->sendmsg(conn->sock, &reply, + sizeof(reply), 0); + if (network_ret < (ssize_t) sizeof(reply)) { + ERR("Relay rotate pending ret code failed"); + ret = -1; } -end_no_session: +end_no_reply: return ret; } /* * Process the commands received on the control socket */ -static -int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_command *cmd, struct relay_local_data *ctx) +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, cmd, ctx->sessions_ht); + ret = relay_create_session(recv_hdr, conn); break; case RELAYD_ADD_STREAM: - ret = relay_add_stream(recv_hdr, cmd, ctx->sessions_ht); + ret = relay_add_stream(recv_hdr, conn); break; case RELAYD_START_DATA: - ret = relay_start(recv_hdr, cmd); + ret = relay_start(recv_hdr, conn); break; case RELAYD_SEND_METADATA: - ret = relay_recv_metadata(recv_hdr, cmd); + ret = relay_recv_metadata(recv_hdr, conn); break; case RELAYD_VERSION: - ret = relay_send_version(recv_hdr, cmd, ctx->sessions_ht); + ret = relay_send_version(recv_hdr, conn); break; case RELAYD_CLOSE_STREAM: - ret = relay_close_stream(recv_hdr, cmd); + ret = relay_close_stream(recv_hdr, conn); break; case RELAYD_DATA_PENDING: - ret = relay_data_pending(recv_hdr, cmd); + ret = relay_data_pending(recv_hdr, conn); break; case RELAYD_QUIESCENT_CONTROL: - ret = relay_quiescent_control(recv_hdr, cmd); + ret = relay_quiescent_control(recv_hdr, conn); break; case RELAYD_BEGIN_DATA_PENDING: - ret = relay_begin_data_pending(recv_hdr, cmd); + ret = relay_begin_data_pending(recv_hdr, conn); break; case RELAYD_END_DATA_PENDING: - ret = relay_end_data_pending(recv_hdr, cmd); + ret = relay_end_data_pending(recv_hdr, conn); break; case RELAYD_SEND_INDEX: - ret = relay_recv_index(recv_hdr, cmd); + ret = relay_recv_index(recv_hdr, conn); + break; + case RELAYD_STREAMS_SENT: + ret = relay_streams_sent(recv_hdr, conn); + break; + case RELAYD_RESET_METADATA: + ret = relay_reset_metadata(recv_hdr, conn); + break; + case RELAYD_ROTATE_STREAM: + ret = relay_rotate_session_stream(recv_hdr, conn); + break; + case RELAYD_ROTATE_RENAME: + ret = relay_rotate_rename(recv_hdr, conn); + break; + case RELAYD_ROTATE_PENDING: + ret = relay_rotate_pending(recv_hdr, conn); + break; + case RELAYD_MKDIR: + ret = relay_mkdir(recv_hdr, conn); break; case RELAYD_UPDATE_SYNC_INFO: default: ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd)); - relay_unknown_command(cmd); + relay_unknown_command(conn); ret = -1; goto end; } @@ -1890,90 +2970,76 @@ end: /* * Handle index for a data stream. * - * RCU read side lock MUST be acquired. + * Called with the stream lock held. * * Return 0 on success else a negative value. */ static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num, - int rotate_index) + int rotate_index, bool *flushed, uint64_t total_size) { - int ret = 0, index_created = 0; - uint64_t stream_id, data_offset; - struct relay_index *index, *wr_index = NULL; - - assert(stream); + int ret = 0; + uint64_t data_offset; + struct relay_index *index; - stream_id = stream->stream_handle; /* Get data offset because we are about to update the index. */ data_offset = htobe64(stream->tracefile_size_current); + DBG("handle_index_data: stream %" PRIu64 " net_seq_num %" PRIu64 " data offset %" PRIu64, + stream->stream_handle, net_seq_num, stream->tracefile_size_current); + /* - * Lookup for an existing index for that stream id/sequence number. If on - * exists, the control thread already received the data for it thus we need - * to write it on disk. + * Lookup for an existing index for that stream id/sequence + * number. If it exists, the control thread has already received the + * data for it, thus we need to write it to disk. */ - index = relay_index_find(stream_id, net_seq_num); + index = relay_index_get_by_id_or_create(stream, net_seq_num); if (!index) { - /* A successful creation will add the object to the HT. */ - index = relay_index_create(stream_id, net_seq_num); - if (!index) { - ret = -1; - goto error; - } - index_created = 1; + ret = -1; + goto end; } - if (rotate_index || stream->index_fd < 0) { - index->to_close_fd = stream->index_fd; - ret = index_create_file(stream->path_name, stream->channel_name, - relayd_uid, relayd_gid, stream->tracefile_size, - stream->tracefile_count_current); + if (rotate_index || !stream->index_file) { + ret = create_rotate_index_file(stream); if (ret < 0) { - /* This will close the stream's index fd if one. */ - relay_index_free_safe(index); - goto error; + ERR("Failed to rotate index"); + /* Put self-ref for this index due to error. */ + relay_index_put(index); + index = NULL; + goto end; } - stream->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; + if (relay_index_set_file(index, stream->index_file, data_offset)) { + ret = -1; + /* Put self-ref for this index due to error. */ + relay_index_put(index); + index = NULL; + goto end; } - /* 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++; + ret = relay_index_try_flush(index); + if (ret == 0) { + tracefile_array_commit_seq(stream->tfa); + stream->index_received_seqcount++; + *flushed = true; + } else if (ret > 0) { + index->total_size = total_size; + /* No flush. */ + ret = 0; + } else { + /* Put self-ref for this index due to error. */ + relay_index_put(index); + index = NULL; + ret = -1; } - -error: +end: return ret; } /* * relay_process_data: Process the data received on the data socket */ -static -int relay_process_data(struct relay_command *cmd) +static int relay_process_data(struct relay_connection *conn) { int ret = 0, rotate_index = 0; ssize_t size_ret; @@ -1982,175 +3048,161 @@ int relay_process_data(struct relay_command *cmd) uint64_t stream_id; uint64_t net_seq_num; uint32_t data_size; + struct relay_session *session; + bool new_stream = false, close_requested = false; + size_t chunk_size = RECV_DATA_BUFFER_SIZE; + size_t recv_off = 0; + char data_buffer[chunk_size]; + bool index_flushed = false; - ret = cmd->sock->ops->recvmsg(cmd->sock, &data_hdr, + 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", cmd->sock->fd); + DBG("Socket %d did an orderly shutdown", conn->sock->fd); } else { - ERR("Unable to receive data header on sock %d", cmd->sock->fd); + 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 = relay_stream_find_by_id(stream_id); + stream = stream_get_by_id(stream_id); if (!stream) { + ERR("relay_process_data: Cannot find stream %" PRIu64, stream_id); ret = -1; - goto end_rcu_unlock; + goto end; } - + session = stream->trace->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 = cmd->sock->ops->recvmsg(cmd->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", cmd->sock->fd); - } - ret = -1; - goto end_rcu_unlock; - } + + pthread_mutex_lock(&stream->lock); /* Check if a rotation is needed. */ if (stream->tracefile_size > 0 && (stream->tracefile_size_current + data_size) > stream->tracefile_size) { - struct relay_viewer_stream *vstream; - uint64_t new_id; + uint64_t old_id, 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 = live_find_viewer_stream_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) { - vstream->abort_flag = 1; - vstream->close_write_flag = 1; + old_id = tracefile_array_get_file_index_head(stream->tfa); + tracefile_array_file_rotate(stream->tfa); - ret = close(vstream->read_fd); - if (ret < 0) { - PERROR("close index"); - } + /* new_id is updated by utils_rotate_stream_file. */ + new_id = old_id; - ret = close(vstream->index_read_fd); - if (ret < 0) { - PERROR("close tracefile"); - } - 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); - pthread_mutex_unlock(&stream->viewer_stream_rotation_lock); + ret = utils_rotate_stream_file(stream->path_name, + stream->channel_name, stream->tracefile_size, + stream->tracefile_count, -1, + -1, stream->stream_fd->fd, + &new_id, &stream->stream_fd->fd); if (ret < 0) { ERR("Rotating stream output file"); - goto end_rcu_unlock; + goto end_stream_unlock; } - /* Reset current size because we just perform a stream rotation. */ + /* + * Reset current size because we just performed 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. + * Index are handled in protocol version 2.4 and above. Also, + * snapshot and index are NOT supported. */ - if (stream->session->minor >= 4 && !stream->session->snapshot) { - ret = handle_index_data(stream, net_seq_num, rotate_index); + if (session->minor >= 4 && !session->snapshot) { + ret = handle_index_data(stream, net_seq_num, rotate_index, + &index_flushed, + data_size + be32toh(data_hdr.padding_size)); if (ret < 0) { - goto end_rcu_unlock; + ERR("handle_index_data: fail stream %" PRIu64 " net_seq_num %" PRIu64 " ret %d", + stream->stream_handle, net_seq_num, ret); + goto end_stream_unlock; } } - /* Write data to stream output fd. */ - size_ret = lttng_write(stream->fd, data_buffer, data_size); - if (size_ret < data_size) { - ERR("Relay error writing data to file"); - ret = -1; - goto end_rcu_unlock; - } + for (recv_off = 0; recv_off < data_size; recv_off += chunk_size) { + size_t recv_size = min(data_size - recv_off, chunk_size); + + ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, recv_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); + } else { + ERR("Socket %d error %d", conn->sock->fd, ret); + } + ret = -1; + goto end_stream_unlock; + } + + /* Write data to stream output fd. */ + size_ret = lttng_write(stream->stream_fd->fd, data_buffer, + recv_size); + if (size_ret < recv_size) { + ERR("Relay error writing data to file"); + ret = -1; + goto end_stream_unlock; + } - DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64, - ret, stream->stream_handle); + DBG2("Relay wrote %zd bytes to tracefile for stream id %" PRIu64, + size_ret, stream->stream_handle); + } - ret = write_padding_to_file(stream->fd, be32toh(data_hdr.padding_size)); + ret = write_padding_to_file(stream->stream_fd->fd, + be32toh(data_hdr.padding_size)); if (ret < 0) { - goto end_rcu_unlock; + ERR("write_padding_to_file: fail stream %" PRIu64 " net_seq_num %" PRIu64 " ret %d", + stream->stream_handle, net_seq_num, ret); + goto end_stream_unlock; + } + stream->tracefile_size_current += + data_size + be32toh(data_hdr.padding_size); + if (stream->prev_seq == -1ULL) { + new_stream = true; + } + if (index_flushed) { + stream->pos_after_last_complete_data_index = + stream->tracefile_size_current; } - stream->tracefile_size_current += data_size + be32toh(data_hdr.padding_size); stream->prev_seq = net_seq_num; - /* Check if we need to close the FD */ - if (close_stream_check(stream)) { - destroy_stream(stream); + ret = try_rotate_stream(stream); + if (ret < 0) { + goto end_stream_unlock; } -end_rcu_unlock: - rcu_read_unlock(); +end_stream_unlock: + close_requested = stream->close_requested; + pthread_mutex_unlock(&stream->lock); + if (close_requested) { + try_stream_close(stream); + } + + if (new_stream) { + pthread_mutex_lock(&session->lock); + uatomic_set(&session->new_streams, 1); + pthread_mutex_unlock(&session->lock); + } + stream_put(stream); end: return ret; } -static -void relay_cleanup_poll_connection(struct lttng_poll_event *events, int pollfd) +static void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd) { int ret; - lttng_poll_del(events, pollfd); + (void) lttng_poll_del(events, pollfd); ret = close(pollfd); if (ret < 0) { @@ -2158,97 +3210,44 @@ void relay_cleanup_poll_connection(struct lttng_poll_event *events, int pollfd) } } -static -int relay_add_connection(int fd, struct lttng_poll_event *events, - struct lttng_ht *relay_connections_ht) -{ - struct relay_command *relay_connection; - ssize_t ret; - - relay_connection = zmalloc(sizeof(struct relay_command)); - if (relay_connection == NULL) { - PERROR("Relay command zmalloc"); - goto error; - } - ret = lttng_read(fd, relay_connection, sizeof(struct relay_command)); - if (ret < sizeof(struct relay_command)) { - PERROR("read relay cmd pipe"); - 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, - (unsigned long) relay_connection->sock->fd); - rcu_read_lock(); - lttng_ht_add_unique_ulong(relay_connections_ht, - &relay_connection->sock_n); - rcu_read_unlock(); - return lttng_poll_add(events, - relay_connection->sock->fd, - LPOLLIN | LPOLLRDHUP); - -error_read: - free(relay_connection); -error: - return -1; -} - -static -void deferred_free_connection(struct rcu_head *head) -{ - struct relay_command *relay_connection = - caa_container_of(head, struct relay_command, rcu_node); - - lttcomm_destroy_sock(relay_connection->sock); - free(relay_connection); -} - -static -void relay_del_connection(struct lttng_ht *relay_connections_ht, - struct lttng_ht_iter *iter, struct relay_command *relay_connection, - struct lttng_ht *sessions_ht) +static void relay_thread_close_connection(struct lttng_poll_event *events, + int pollfd, struct relay_connection *conn) { - int ret; - - ret = lttng_ht_del(relay_connections_ht, iter); - assert(!ret); + const char *type_str; - if (relay_connection->type == RELAY_CONTROL) { - relay_delete_session(relay_connection, sessions_ht); - lttng_ht_destroy(relay_connection->ctf_traces_ht); + switch (conn->type) { + case RELAY_DATA: + type_str = "Data"; + break; + case RELAY_CONTROL: + type_str = "Control"; + break; + case RELAY_VIEWER_COMMAND: + type_str = "Viewer Command"; + break; + case RELAY_VIEWER_NOTIFICATION: + type_str = "Viewer Notification"; + break; + default: + type_str = "Unknown"; } - - call_rcu(&relay_connection->rcu_node, deferred_free_connection); + cleanup_connection_pollfd(events, pollfd); + connection_put(conn); + DBG("%s connection closed with %d", type_str, pollfd); } /* * This thread does the actual work */ -static -void *relay_thread_worker(void *data) +static void *relay_thread_worker(void *data) { int ret, err = -1, last_seen_data_fd = -1; uint32_t nb_fd; - struct relay_command *relay_connection; struct lttng_poll_event events; struct lttng_ht *relay_connections_ht; - struct lttng_ht_node_ulong *node; struct lttng_ht_iter iter; struct lttcomm_relayd_hdr recv_hdr; - struct relay_local_data *relay_ctx = (struct relay_local_data *) data; - struct lttng_ht *sessions_ht = relay_ctx->sessions_ht; + struct relay_connection *destroy_conn = NULL; DBG("[thread] Relay worker started"); @@ -2256,6 +3255,10 @@ void *relay_thread_worker(void *data) 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 */ @@ -2264,18 +3267,12 @@ void *relay_thread_worker(void *data) goto relay_connections_ht_error; } - /* Tables of received indexes indexed by index handle and net_seq_num. */ - indexes_ht = lttng_ht_new(0, LTTNG_HT_TYPE_TWO_U64); - if (!indexes_ht) { - goto indexes_ht_error; - } - ret = create_thread_poll_set(&events, 2); if (ret < 0) { goto error_poll_create; } - 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; } @@ -2304,9 +3301,9 @@ restart: nb_fd = ret; /* - * Process control. The control connection is prioritised so we don't - * starve it with high throughout put tracing data on the data - * connection. + * Process control. The control connection is + * prioritized so we don't starve it with high + * throughput tracing data on the data connection. */ for (i = 0; i < nb_fd; i++) { /* Fetch once the poll data */ @@ -2315,6 +3312,14 @@ restart: health_code_update(); + if (!revents) { + /* + * No activity for this FD (poll + * implementation). + */ + continue; + } + /* Thread quit pipe has been closed. Killing thread. */ ret = check_thread_quit_pipe(pollfd, revents); if (ret) { @@ -2322,78 +3327,35 @@ restart: goto exit; } - /* Inspect the relay cmd pipe for new connection */ - if (pollfd == relay_cmd_pipe[0]) { - if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { - ERR("Relay pipe error"); - goto error; - } else if (revents & LPOLLIN) { - DBG("Relay command received"); - ret = relay_add_connection(relay_cmd_pipe[0], - &events, relay_connections_ht); + /* Inspect the relay conn pipe for new connection */ + if (pollfd == relay_conn_pipe[0]) { + if (revents & LPOLLIN) { + struct relay_connection *conn; + + ret = lttng_read(relay_conn_pipe[0], &conn, sizeof(conn)); if (ret < 0) { goto error; } - } - } else if (revents) { - rcu_read_lock(); - lttng_ht_lookup(relay_connections_ht, - (void *)((unsigned long) pollfd), - &iter); - node = lttng_ht_iter_get_node_ulong(&iter); - if (node == NULL) { - DBG2("Relay sock %d not found", pollfd); - rcu_read_unlock(); + lttng_poll_add(&events, conn->sock->fd, + LPOLLIN | LPOLLRDHUP); + connection_ht_add(relay_connections_ht, conn); + DBG("Connection socket %d added", conn->sock->fd); + } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { + ERR("Relay connection pipe error"); + goto error; + } else { + ERR("Unexpected poll events %u for sock %d", revents, pollfd); goto error; } - 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, - &iter, relay_connection, sessions_ht); - if (last_seen_data_fd == pollfd) { - last_seen_data_fd = last_notdel_data_fd; - } - } else if (revents & (LPOLLHUP | LPOLLRDHUP)) { - DBG("Socket %d hung up", pollfd); - relay_cleanup_poll_connection(&events, pollfd); - relay_del_connection(relay_connections_ht, - &iter, relay_connection, sessions_ht); - 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), 0); - /* connection closed */ - if (ret <= 0) { - relay_cleanup_poll_connection(&events, pollfd); - relay_del_connection(relay_connections_ht, - &iter, relay_connection, sessions_ht); - 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, relay_ctx); - if (ret < 0) { - /* Clear the session on error. */ - relay_cleanup_poll_connection(&events, pollfd); - relay_del_connection(relay_connections_ht, - &iter, relay_connection, sessions_ht); - DBG("Connection closed with %d", pollfd); - } - seen_control = 1; - } - } else { + } else { + struct relay_connection *ctrl_conn; + + ctrl_conn = connection_get_by_sock(relay_connections_ht, pollfd); + /* If not found, there is a synchronization issue. */ + assert(ctrl_conn); + + if (ctrl_conn->type == RELAY_DATA) { + if (revents & LPOLLIN) { /* * Flag the last seen data fd not deleted. It will be * used as the last seen fd if any fd gets deleted in @@ -2401,8 +3363,40 @@ restart: */ last_notdel_data_fd = pollfd; } + goto put_ctrl_connection; + } + assert(ctrl_conn->type == RELAY_CONTROL); + + if (revents & LPOLLIN) { + ret = ctrl_conn->sock->ops->recvmsg(ctrl_conn->sock, + &recv_hdr, sizeof(recv_hdr), 0); + if (ret <= 0) { + /* Connection closed */ + relay_thread_close_connection(&events, pollfd, + ctrl_conn); + } else { + ret = relay_process_control(&recv_hdr, ctrl_conn); + if (ret < 0) { + /* Clear the session on error. */ + relay_thread_close_connection(&events, + pollfd, ctrl_conn); + } + seen_control = 1; + } + } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { + relay_thread_close_connection(&events, + pollfd, ctrl_conn); + if (last_seen_data_fd == pollfd) { + last_seen_data_fd = last_notdel_data_fd; + } + } else { + ERR("Unexpected poll events %u for control sock %d", + revents, pollfd); + connection_put(ctrl_conn); + goto error; } - rcu_read_unlock(); + put_ctrl_connection: + connection_put(ctrl_conn); } } @@ -2432,54 +3426,56 @@ restart: /* Fetch the poll data. */ uint32_t revents = LTTNG_POLL_GETEV(&events, i); int pollfd = LTTNG_POLL_GETFD(&events, i); + struct relay_connection *data_conn; health_code_update(); - /* Skip the command pipe. It's handled in the first loop. */ - if (pollfd == relay_cmd_pipe[0]) { + if (!revents) { + /* No activity for this FD (poll implementation). */ continue; } - if (revents) { - rcu_read_lock(); - lttng_ht_lookup(relay_connections_ht, - (void *)((unsigned long) pollfd), - &iter); - node = lttng_ht_iter_get_node_ulong(&iter); - if (node == NULL) { - /* Skip it. Might be removed before. */ - rcu_read_unlock(); - continue; - } - relay_connection = caa_container_of(node, - struct relay_command, sock_n); + /* Skip the command pipe. It's handled in the first loop. */ + if (pollfd == relay_conn_pipe[0]) { + continue; + } - if (revents & LPOLLIN) { - if (relay_connection->type != RELAY_DATA) { - continue; - } + data_conn = connection_get_by_sock(relay_connections_ht, pollfd); + if (!data_conn) { + /* Skip it. Might be removed before. */ + continue; + } + if (data_conn->type == RELAY_CONTROL) { + goto put_data_connection; + } + assert(data_conn->type == RELAY_DATA); - ret = relay_process_data(relay_connection); - /* connection closed */ - if (ret < 0) { - relay_cleanup_poll_connection(&events, pollfd); - relay_del_connection(relay_connections_ht, - &iter, relay_connection, sessions_ht); - 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; - } + if (revents & LPOLLIN) { + ret = relay_process_data(data_conn); + /* Connection closed */ + if (ret < 0) { + relay_thread_close_connection(&events, pollfd, + data_conn); + /* + * Every goto restart call sets the last seen fd where + * here we don't really care since we gracefully + * continue the loop after the connection is deleted. + */ + } else { + /* Keep last seen port. */ + last_seen_data_fd = pollfd; + connection_put(data_conn); + goto restart; } - rcu_read_unlock(); + } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { + relay_thread_close_connection(&events, pollfd, + data_conn); + } else { + ERR("Unknown poll events %u for data sock %d", + revents, pollfd); } + put_data_connection: + connection_put(data_conn); } last_seen_data_fd = -1; } @@ -2489,41 +3485,44 @@ restart: 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) { + cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, + destroy_conn, + sock_n.node) { health_code_update(); - 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, - &iter, relay_connection, sessions_ht); + if (session_abort(destroy_conn->session)) { + assert(0); } + + /* + * No need to grab another ref, because we own + * destroy_conn. + */ + relay_thread_close_connection(&events, destroy_conn->sock->fd, + destroy_conn); } rcu_read_unlock(); + + lttng_poll_clean(&events); error_poll_create: - lttng_ht_destroy(indexes_ht); -indexes_ht_error: lttng_ht_destroy(relay_connections_ht); relay_connections_ht_error: - /* Close relay cmd pipes */ - utils_close_pipe(relay_cmd_pipe); + /* 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); +error_testpoint: if (err) { health_error(); ERR("Health error occurred in %s", __func__); } health_unregister(health_relayd); rcu_unregister_thread(); - stop_threads(); + lttng_relay_stop_threads(); return NULL; } @@ -2531,11 +3530,11 @@ 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; } @@ -2545,206 +3544,224 @@ static int create_relay_cmd_pipe(void) */ int main(int argc, char **argv) { - int ret = 0; + int ret = 0, retval = 0; 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) { - goto exit; + if (set_options(argc, argv)) { + retval = -1; + goto exit_options; } - if ((ret = set_signal_handler()) < 0) { - goto exit; + if (set_signal_handler()) { + retval = -1; + goto exit_options; } /* 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; + retval = -1; + goto exit_options; } - ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG); + ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG, + -1, -1); if (ret < 0) { ERR("Unable to create %s", opt_output_path); - goto exit; + retval = -1; + goto exit_options; } } /* 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; + retval = -1; + goto exit_options; + } + + /* + * 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); } } - /* We need those values for the file/dir creation. */ - relayd_uid = getuid(); - relayd_gid = getgid(); + /* Initialize thread health monitoring */ + health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES); + if (!health_relayd) { + PERROR("health_app_create error"); + retval = -1; + goto exit_health_app_create; + } - /* Check if daemon is UID = 0 */ - if (relayd_uid == 0) { - if (control_uri->port < 1024 || data_uri->port < 1024) { - ERR("Need to be root to use ports < 1024"); - ret = -1; - goto exit; - } + /* Create thread quit pipe */ + if (init_thread_quit_pipe()) { + retval = -1; + goto exit_init_data; } /* Setup the thread apps communication pipe. */ - if ((ret = create_relay_cmd_pipe()) < 0) { - goto exit; + if (create_relay_conn_pipe()) { + retval = -1; + goto exit_init_data; } /* Init relay command queue. */ - cds_wfq_init(&relay_cmd_queue.queue); - - /* Set up max poll set size */ - lttng_poll_set_max_size(); + cds_wfcq_init(&relay_conn_queue.head, &relay_conn_queue.tail); /* Initialize communication library */ lttcomm_init(); - - relay_ctx = zmalloc(sizeof(struct relay_local_data)); - if (!relay_ctx) { - PERROR("relay_ctx"); - goto exit; - } + lttcomm_inet_init(); /* tables of sessions indexed by session ID */ - relay_ctx->sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); - if (!relay_ctx->sessions_ht) { - goto exit_relay_ctx_sessions; + sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); + if (!sessions_ht) { + retval = -1; + goto exit_init_data; } /* tables of streams indexed by stream ID */ - relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); + relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); if (!relay_streams_ht) { - goto exit_relay_ctx_streams; + retval = -1; + goto exit_init_data; } /* 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; + retval = -1; + goto exit_init_data; } ret = utils_create_pipe(health_quit_pipe); - if (ret < 0) { - goto error_health_pipe; + if (ret) { + retval = -1; + goto exit_health_quit_pipe; } /* Create thread to manage the client socket */ - ret = pthread_create(&health_thread, NULL, + ret = pthread_create(&health_thread, default_pthread_attr(), thread_manage_health, (void *) NULL); - if (ret != 0) { + if (ret) { + errno = ret; PERROR("pthread_create health"); - goto health_error; + retval = -1; + goto exit_health_thread; } /* Setup the dispatcher thread */ - ret = pthread_create(&dispatcher_thread, NULL, + ret = pthread_create(&dispatcher_thread, default_pthread_attr(), relay_thread_dispatcher, (void *) NULL); - if (ret != 0) { + if (ret) { + errno = ret; PERROR("pthread_create dispatcher"); - goto exit_dispatcher; + retval = -1; + goto exit_dispatcher_thread; } /* Setup the worker thread */ - ret = pthread_create(&worker_thread, NULL, - relay_thread_worker, (void *) relay_ctx); - if (ret != 0) { + ret = pthread_create(&worker_thread, default_pthread_attr(), + relay_thread_worker, NULL); + if (ret) { + errno = ret; PERROR("pthread_create worker"); - goto exit_worker; + retval = -1; + goto exit_worker_thread; } /* Setup the listener thread */ - ret = pthread_create(&listener_thread, NULL, + ret = pthread_create(&listener_thread, default_pthread_attr(), relay_thread_listener, (void *) NULL); - if (ret != 0) { + if (ret) { + errno = ret; PERROR("pthread_create listener"); - goto exit_listener; + retval = -1; + goto exit_listener_thread; } - ret = live_start_threads(live_uri, relay_ctx, thread_quit_pipe); - if (ret != 0) { + ret = relayd_live_create(live_uri); + if (ret) { ERR("Starting live viewer threads"); + retval = -1; goto exit_live; } + /* + * This is where we start awaiting program completion (e.g. through + * signal that asks threads to teardown). + */ + + ret = relayd_live_join(); + if (ret) { + retval = -1; + } exit_live: + ret = pthread_join(listener_thread, &status); - if (ret != 0) { - PERROR("pthread_join"); - goto error; /* join error, exit without cleanup */ + if (ret) { + errno = ret; + PERROR("pthread_join listener_thread"); + retval = -1; } -exit_listener: +exit_listener_thread: ret = pthread_join(worker_thread, &status); - if (ret != 0) { - PERROR("pthread_join"); - goto error; /* join error, exit without cleanup */ + if (ret) { + errno = ret; + PERROR("pthread_join worker_thread"); + retval = -1; } -exit_worker: +exit_worker_thread: ret = pthread_join(dispatcher_thread, &status); - if (ret != 0) { - PERROR("pthread_join"); - goto error; /* join error, exit without cleanup */ + if (ret) { + errno = ret; + PERROR("pthread_join dispatcher_thread"); + retval = -1; } +exit_dispatcher_thread: -exit_dispatcher: ret = pthread_join(health_thread, &status); - if (ret != 0) { - PERROR("pthread_join health thread"); - goto error; /* join error, exit without cleanup */ + if (ret) { + errno = ret; + PERROR("pthread_join health_thread"); + retval = -1; } +exit_health_thread: - /* - * Stop live threads only after joining other threads. - */ - live_stop_threads(); - -health_error: utils_close_pipe(health_quit_pipe); +exit_health_quit_pipe: -error_health_pipe: +exit_init_data: 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_options: + /* + * Wait for all pending call_rcu work to complete before tearing + * down data structures. call_rcu worker may be trying to + * perform lookups in those structures. + */ + rcu_barrier(); + relayd_cleanup(); -exit_relay_ctx_sessions: - free(relay_ctx); + /* Ensure all prior call_rcu are done. */ + rcu_barrier(); -exit: - cleanup(); - if (!ret) { + if (!retval) { exit(EXIT_SUCCESS); + } else { + exit(EXIT_FAILURE); } - -error: - exit(EXIT_FAILURE); }