X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-relayd%2Fmain.c;h=abefba8659d7cc429aa39588186e580de084d51b;hp=bdefd16763850b7e75a5a9f294ebfb434aca41b3;hb=6e3c5836f180eeee21271242f707f4b88a840570;hpb=25672d0260dce634e1a249dd97d3cc0b592f34a6 diff --git a/src/bin/lttng-relayd/main.c b/src/bin/lttng-relayd/main.c index bdefd1676..abefba865 100644 --- a/src/bin/lttng-relayd/main.c +++ b/src/bin/lttng-relayd/main.c @@ -49,14 +49,15 @@ #include #include #include +#include #include "lttng-relayd.h" /* command line options */ static int opt_daemon; static char *opt_output_path; -static struct lttng_uri *control_uri = NULL; -static struct lttng_uri *data_uri = NULL; +static struct lttng_uri *control_uri; +static struct lttng_uri *data_uri; const char *progname; static int is_root; /* Set to 1 if the daemon is running as root */ @@ -73,14 +74,15 @@ static int thread_quit_pipe[2] = { -1, -1 }; */ static int relay_cmd_pipe[2] = { -1, -1 }; +/* Shared between threads */ static int dispatch_thread_exit; static pthread_t listener_thread; static pthread_t dispatcher_thread; static pthread_t worker_thread; -static uint64_t last_relay_stream_id = 0; -static uint64_t last_relay_session_id = 0; +static uint64_t last_relay_stream_id; +static uint64_t last_relay_session_id; /* * Relay command queue. @@ -91,8 +93,8 @@ static uint64_t last_relay_session_id = 0; static struct relay_cmd_queue relay_cmd_queue; /* buffer allocated at startup, used to store the trace data */ -static char *data_buffer = NULL; -static unsigned int data_buffer_size = 0; +static char *data_buffer; +static unsigned int data_buffer_size; /* * usage function on stderr @@ -117,13 +119,13 @@ int parse_args(int argc, char **argv) char *default_address; static struct option long_options[] = { - { "control-port", 1, 0, 'C' }, - { "data-port", 1, 0, 'D' }, - { "daemonize", 0, 0, 'd' }, - { "help", 0, 0, 'h' }, - { "output", 1, 0, 'o' }, - { "verbose", 0, 0, 'v' }, - { NULL, 0, 0, 0 } + { "control-port", 1, 0, 'C', }, + { "data-port", 1, 0, 'D', }, + { "daemonize", 0, 0, 'd', }, + { "help", 0, 0, 'h', }, + { "output", 1, 0, 'o', }, + { "verbose", 0, 0, 'v', }, + { NULL, 0, 0, 0, }, }; while (1) { @@ -228,18 +230,16 @@ exit: static void cleanup(void) { - int i, ret; - DBG("Cleaning up"); - for (i = 0; i < 2; i++) { - if (thread_quit_pipe[i] >= 0) { - ret = close(thread_quit_pipe[i]); - if (ret) { - PERROR("close"); - } - } - } + /* free the dynamically allocated opt_output_path */ + free(opt_output_path); + + /* Close thread quit pipes */ + utils_close_pipe(thread_quit_pipe); + + /* Close relay cmd pipes */ + utils_close_pipe(relay_cmd_pipe); } /* @@ -250,7 +250,9 @@ int notify_thread_pipe(int wpipe) { int ret; - ret = write(wpipe, "!", 1); + do { + ret = write(wpipe, "!", 1); + } while (ret < 0 && errno == EINTR); if (ret < 0) { PERROR("write poll pipe"); } @@ -274,7 +276,7 @@ void stop_threads(void) } /* Dispatch thread */ - dispatch_thread_exit = 1; + CMM_STORE_SHARED(dispatch_thread_exit, 1); futex_nto1_wake(&relay_cmd_queue.futex); } @@ -351,23 +353,10 @@ int set_signal_handler(void) static int init_thread_quit_pipe(void) { - int ret, i; - - ret = pipe(thread_quit_pipe); - if (ret < 0) { - PERROR("thread quit pipe"); - goto error; - } + int ret; - for (i = 0; i < 2; i++) { - ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC); - if (ret < 0) { - PERROR("fcntl"); - goto error; - } - } + ret = utils_create_pipe_cloexec(thread_quit_pipe); -error: return ret; } @@ -463,7 +452,7 @@ error: static void *relay_thread_listener(void *data) { - int i, ret, pollfd; + int i, ret, pollfd, err = -1; int val = 1; uint32_t revents, nb_fd; struct lttng_poll_event events; @@ -479,12 +468,12 @@ void *relay_thread_listener(void *data) control_sock = relay_init_sock(control_uri); if (!control_sock) { - goto error_sock; + goto error_sock_control; } data_sock = relay_init_sock(data_uri); if (!data_sock) { - goto error_sock; + goto error_sock_relay; } /* @@ -533,7 +522,8 @@ restart: /* Thread quit pipe has been closed. Killing thread. */ ret = check_thread_quit_pipe(pollfd, revents); if (ret) { - goto error; + err = 0; + goto exit; } if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { @@ -586,28 +576,32 @@ restart: } } +exit: error: error_poll_add: lttng_poll_clean(&events); error_create_poll: - if (control_sock->fd >= 0) { - ret = control_sock->ops->close(control_sock); + if (data_sock->fd >= 0) { + ret = data_sock->ops->close(data_sock); if (ret) { PERROR("close"); } - lttcomm_destroy_sock(control_sock); } - if (data_sock->fd >= 0) { - ret = data_sock->ops->close(data_sock); + lttcomm_destroy_sock(data_sock); +error_sock_relay: + if (control_sock->fd >= 0) { + ret = control_sock->ops->close(control_sock); if (ret) { PERROR("close"); } - lttcomm_destroy_sock(data_sock); } - + lttcomm_destroy_sock(control_sock); +error_sock_control: + if (err) { + DBG("Thread exited with error"); + } DBG("Relay listener thread cleanup complete"); stop_threads(); -error_sock: return NULL; } @@ -623,7 +617,7 @@ void *relay_thread_dispatcher(void *data) DBG("[thread] Relay dispatcher started"); - while (!dispatch_thread_exit) { + while (!CMM_LOAD_SHARED(dispatch_thread_exit)) { /* Atomically prepare the queue futex */ futex_nto1_prepare(&relay_cmd_queue.futex); @@ -644,8 +638,10 @@ void *relay_thread_dispatcher(void *data) * call is blocking so we can be assured that the data will be read * at some point in time or wait to the end of the world :) */ - ret = write(relay_cmd_pipe[1], relay_cmd, - sizeof(struct relay_command)); + do { + ret = write(relay_cmd_pipe[1], relay_cmd, + sizeof(struct relay_command)); + } while (ret < 0 && errno == EINTR); free(relay_cmd); if (ret < 0) { PERROR("write cmd pipe"); @@ -673,7 +669,7 @@ static char *expand_full_path(const char *path) { const char *end_path = path; - char *next, *cut_path, *expanded_path; + char *next, *cut_path, *expanded_path, *respath; /* Find last token delimited by '/' */ while ((next = strpbrk(end_path + 1, "/"))) { @@ -685,11 +681,12 @@ char *expand_full_path(const char *path) expanded_path = malloc(PATH_MAX); if (expanded_path == NULL) { - goto error; + respath = NULL; + goto end; } - expanded_path = realpath((char *)cut_path, expanded_path); - if (expanded_path == NULL) { + respath = realpath(cut_path, expanded_path); + if (respath == NULL) { switch (errno) { case ENOENT: ERR("%s: No such file or directory", cut_path); @@ -698,18 +695,14 @@ char *expand_full_path(const char *path) PERROR("realpath"); break; } - goto error; + free(expanded_path); + } else { + /* Add end part to expanded path */ + strcat(respath, end_path); } - - /* Add end part to expanded path */ - strcat(expanded_path, end_path); - - free(cut_path); - return expanded_path; - -error: +end: free(cut_path); - return NULL; + return respath; } @@ -761,7 +754,7 @@ int mkdir_recursive(char *path, mode_t mode) if (ret < 0) { ret = mkdir(tmp, mode); if (ret < 0) { - if (!(errno == EEXIST)) { + if (errno != EEXIST) { PERROR("mkdir recursive"); ret = -errno; goto error; @@ -774,7 +767,7 @@ int mkdir_recursive(char *path, mode_t mode) ret = mkdir(tmp, mode); if (ret < 0) { - if (!(errno == EEXIST)) { + if (errno != EEXIST) { PERROR("mkdir recursive last piece"); ret = -errno; } else { @@ -786,46 +779,73 @@ error: return ret; } +static +char *create_output_path_auto(char *path_name) +{ + int ret; + char *traces_path = NULL; + char *alloc_path = NULL; + char *default_path; + + default_path = config_get_default_path(); + if (default_path == NULL) { + ERR("Home path not found.\n \ + Please specify an output path using -o, --output PATH"); + goto exit; + } + alloc_path = strdup(default_path); + if (alloc_path == NULL) { + PERROR("Path allocation"); + goto exit; + } + ret = asprintf(&traces_path, "%s/" DEFAULT_TRACE_DIR_NAME + "/%s", alloc_path, path_name); + if (ret < 0) { + PERROR("asprintf trace dir name"); + goto exit; + } +exit: + free(alloc_path); + return traces_path; +} + +static +char *create_output_path_noauto(char *path_name) +{ + int ret; + char *traces_path = NULL; + char *full_path; + + full_path = expand_full_path(opt_output_path); + ret = asprintf(&traces_path, "%s/%s", full_path, path_name); + if (ret < 0) { + PERROR("asprintf trace dir name"); + goto exit; + } +exit: + free(full_path); + return traces_path; +} + /* * create_output_path: create the output trace directory */ static char *create_output_path(char *path_name) { - int ret = 0; - char *alloc_path = NULL; - char *traces_path = NULL; - char *full_path = NULL; - - /* Auto output path */ if (opt_output_path == NULL) { - alloc_path = strdup(config_get_default_path()); - if (alloc_path == NULL) { - ERR("Home path not found.\n \ - Please specify an output path using -o, --output PATH"); - ret = -1; - goto exit; - } - - ret = asprintf(&traces_path, "%s/" DEFAULT_TRACE_DIR_NAME - "/%s", alloc_path, path_name); - if (ret < 0) { - PERROR("asprintf trace dir name"); - goto exit; - } + return create_output_path_auto(path_name); } else { - full_path = expand_full_path(opt_output_path); - ret = asprintf(&traces_path, "%s/%s", full_path, path_name); - if (ret < 0) { - PERROR("asprintf trace dir name"); - goto exit; - } + return create_output_path_noauto(path_name); } - free(alloc_path); - free(full_path); +} -exit: - return traces_path; +static +void deferred_free_stream(struct rcu_head *head) +{ + struct relay_stream *stream = + caa_container_of(head, struct relay_stream, rcu_node); + free(stream); } /* @@ -840,12 +860,14 @@ void relay_delete_session(struct relay_command *cmd, struct lttng_ht *streams_ht struct relay_stream *stream; int ret; - if (!cmd->session) + if (!cmd->session) { return; + } DBG("Relay deleting session %lu", cmd->session->id); free(cmd->session->sock); + rcu_read_lock(); cds_lfht_for_each_entry(streams_ht->ht, &iter.iter, node, node) { node = lttng_ht_iter_get_node_ulong(&iter); if (node) { @@ -855,10 +877,12 @@ void relay_delete_session(struct relay_command *cmd, struct lttng_ht *streams_ht close(stream->fd); ret = lttng_ht_del(streams_ht, &iter); assert(!ret); - free(stream); + call_rcu(&stream->rcu_node, + deferred_free_stream); } } } + rcu_read_unlock(); } /* @@ -896,6 +920,7 @@ int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr, goto end_no_session; } + rcu_read_lock(); stream->stream_handle = ++last_relay_stream_id; stream->seq = 0; stream->session = session; @@ -907,7 +932,6 @@ int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr, } ret = mkdir_recursive(root_path, S_IRWXU | S_IRWXG); if (ret < 0) { - free(root_path); ERR("relay creating output directory"); goto end; } @@ -949,6 +973,7 @@ end: if (send_ret < 0) { ERR("Relay sending stream id"); } + rcu_read_unlock(); end_no_session: return ret; @@ -1000,6 +1025,7 @@ int relay_start(struct lttcomm_relayd_hdr *recv_hdr, /* * Get stream from stream id. + * Need to be called with RCU read-side lock held. */ static struct relay_stream *relay_stream_from_stream_id(uint64_t stream_id, @@ -1064,22 +1090,28 @@ int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr, goto end; } metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer; + + rcu_read_lock(); metadata_stream = relay_stream_from_stream_id( be64toh(metadata_struct->stream_id), streams_ht); if (!metadata_stream) { ret = -1; - goto end; + goto end_unlock; } - ret = write(metadata_stream->fd, metadata_struct->payload, - payload_size); + do { + ret = write(metadata_stream->fd, metadata_struct->payload, + payload_size); + } while (ret < 0 && errno == EINTR); if (ret < (payload_size)) { ERR("Relay error writing metadata on file"); ret = -1; - goto end; + goto end_unlock; } DBG2("Relay metadata written"); +end_unlock: + rcu_read_unlock(); end: return ret; } @@ -1182,10 +1214,12 @@ int relay_process_data(struct relay_command *cmd, struct lttng_ht *streams_ht) } stream_id = be64toh(data_hdr.stream_id); + + rcu_read_lock(); stream = relay_stream_from_stream_id(stream_id, streams_ht); if (!stream) { ret = -1; - goto end; + goto end_unlock; } data_size = be32toh(data_hdr.data_size); @@ -1194,7 +1228,7 @@ int relay_process_data(struct relay_command *cmd, struct lttng_ht *streams_ht) if (!data_buffer) { ERR("Allocating data buffer"); ret = -1; - goto end; + goto end_unlock; } data_buffer_size = data_size; } @@ -1203,29 +1237,30 @@ int relay_process_data(struct relay_command *cmd, struct lttng_ht *streams_ht) ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, MSG_WAITALL); if (ret <= 0) { ret = -1; - goto end; + goto end_unlock; } - ret = write(stream->fd, data_buffer, data_size); + do { + ret = write(stream->fd, data_buffer, data_size); + } while (ret < 0 && errno == EINTR); if (ret < data_size) { ERR("Relay error writing data to file"); ret = -1; - goto end; + goto end_unlock; } DBG2("Relay wrote %d bytes to tracefile for stream id %lu", ret, stream->stream_handle); +end_unlock: + rcu_read_unlock(); end: return ret; } static -void relay_cleanup_connection(struct lttng_ht *relay_connections_ht, struct lttng_poll_event *events, - struct lttng_ht *streams_ht, int pollfd, struct lttng_ht_iter *iter) +void relay_cleanup_poll_connection(struct lttng_poll_event *events, int pollfd) { int ret; - ret = lttng_ht_del(relay_connections_ht, iter); - assert(!ret); lttng_poll_del(events, pollfd); ret = close(pollfd); @@ -1238,32 +1273,58 @@ static int relay_add_connection(int fd, struct lttng_poll_event *events, struct lttng_ht *relay_connections_ht) { - int ret; struct relay_command *relay_connection; + int ret; relay_connection = zmalloc(sizeof(struct relay_command)); if (relay_connection == NULL) { PERROR("Relay command zmalloc"); - ret = -1; - goto end; + goto error; } ret = read(fd, relay_connection, sizeof(struct relay_command)); if (ret < 0 || ret < sizeof(relay_connection)) { PERROR("read relay cmd pipe"); - ret = -1; - goto end; + 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); - ret = lttng_poll_add(events, + rcu_read_unlock(); + return lttng_poll_add(events, relay_connection->sock->fd, LPOLLIN | LPOLLRDHUP); -end: - return ret; +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); + free(relay_connection); +} + +static +void relay_del_connection(struct lttng_ht *relay_connections_ht, + struct lttng_ht *streams_ht, struct lttng_ht_iter *iter, + struct relay_command *relay_connection) +{ + int ret; + + ret = lttng_ht_del(relay_connections_ht, iter); + assert(!ret); + if (relay_connection->type == RELAY_CONTROL) { + relay_delete_session(relay_connection, streams_ht); + } + call_rcu(&relay_connection->rcu_node, + deferred_free_connection); } /* @@ -1272,7 +1333,7 @@ end: static void *relay_thread_worker(void *data) { - int i, ret, pollfd; + int i, ret, pollfd, err = -1; uint32_t revents, nb_fd; struct relay_command *relay_connection; struct lttng_poll_event events; @@ -1284,11 +1345,19 @@ void *relay_thread_worker(void *data) DBG("[thread] Relay worker started"); + rcu_register_thread(); + /* table of connections indexed on socket */ relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); + if (!relay_connections_ht) { + goto relay_connections_ht_error; + } /* tables of streams indexed by stream ID */ streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); + if (!streams_ht) { + goto streams_ht_error; + } ret = create_thread_poll_set(&events, 2); if (ret < 0) { @@ -1327,7 +1396,8 @@ void *relay_thread_worker(void *data) /* Thread quit pipe has been closed. Killing thread. */ ret = check_thread_quit_pipe(pollfd, revents); if (ret) { - goto error; + err = 0; + goto exit; } /* Inspect the relay cmd pipe for new connection */ @@ -1344,12 +1414,14 @@ void *relay_thread_worker(void *data) } } } else if (revents > 0) { + rcu_read_lock(); lttng_ht_lookup(relay_connections_ht, (void *)((unsigned long) pollfd), &iter); node = lttng_ht_iter_get_node_ulong(&iter); if (node == NULL) { DBG2("Relay sock %d not found", pollfd); + rcu_read_unlock(); goto error; } relay_connection = caa_container_of(node, @@ -1357,17 +1429,16 @@ void *relay_thread_worker(void *data) if (revents & (LPOLLERR)) { ERR("POLL ERROR"); - relay_cleanup_connection(relay_connections_ht, - &events, streams_ht, pollfd, &iter); - free(relay_connection); + relay_cleanup_poll_connection(&events, pollfd); + relay_del_connection(relay_connections_ht, + streams_ht, &iter, + relay_connection); } else if (revents & (LPOLLHUP | LPOLLRDHUP)) { DBG("Socket %d hung up", pollfd); - relay_cleanup_connection(relay_connections_ht, - &events, streams_ht, pollfd, &iter); - if (relay_connection->type == RELAY_CONTROL) { - relay_delete_session(relay_connection, streams_ht); - } - free(relay_connection); + relay_cleanup_poll_connection(&events, pollfd); + relay_del_connection(relay_connections_ht, + streams_ht, &iter, + relay_connection); } else if (revents & LPOLLIN) { /* control socket */ if (relay_connection->type == RELAY_CONTROL) { @@ -1376,10 +1447,10 @@ void *relay_thread_worker(void *data) sizeof(struct lttcomm_relayd_hdr), MSG_WAITALL); /* connection closed */ if (ret <= 0) { - relay_cleanup_connection(relay_connections_ht, - &events, streams_ht, pollfd, &iter); - relay_delete_session(relay_connection, streams_ht); - free(relay_connection); + relay_cleanup_poll_connection(&events, pollfd); + relay_del_connection(relay_connections_ht, + streams_ht, &iter, + relay_connection); DBG("Control connection closed with %d", pollfd); } else { if (relay_connection->session) { @@ -1394,9 +1465,10 @@ void *relay_thread_worker(void *data) * command: clear the session * */ if (ret < 0) { - relay_cleanup_connection(relay_connections_ht, - &events, streams_ht, pollfd, &iter); - free(relay_connection); + relay_cleanup_poll_connection(&events, pollfd); + relay_del_connection(relay_connections_ht, + streams_ht, &iter, + relay_connection); DBG("Connection closed with %d", pollfd); } } @@ -1405,36 +1477,48 @@ void *relay_thread_worker(void *data) ret = relay_process_data(relay_connection, streams_ht); /* connection closed */ if (ret < 0) { - relay_cleanup_connection(relay_connections_ht, - &events, streams_ht, pollfd, &iter); - relay_delete_session(relay_connection, streams_ht); + relay_cleanup_poll_connection(&events, pollfd); + relay_del_connection(relay_connections_ht, + streams_ht, &iter, + relay_connection); DBG("Data connection closed with %d", pollfd); } } } + rcu_read_unlock(); } } } +exit: error: lttng_poll_clean(&events); /* empty the hash table and free the memory */ + rcu_read_lock(); cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, node, node) { node = lttng_ht_iter_get_node_ulong(&iter); if (node) { relay_connection = caa_container_of(node, struct relay_command, sock_n); - free(relay_connection); + relay_del_connection(relay_connections_ht, + streams_ht, &iter, + relay_connection); } - ret = lttng_ht_del(relay_connections_ht, &iter); - assert(!ret); } + rcu_read_unlock(); error_poll_create: - free(data_buffer); + lttng_ht_destroy(streams_ht); +streams_ht_error: lttng_ht_destroy(relay_connections_ht); +relay_connections_ht_error: + if (err) { + DBG("Thread exited with error"); + } DBG("Worker thread cleanup complete"); + free(data_buffer); stop_threads(); + rcu_unregister_thread(); return NULL; } @@ -1444,23 +1528,10 @@ error_poll_create: */ static int create_relay_cmd_pipe(void) { - int ret, i; + int ret; - ret = pipe(relay_cmd_pipe); - if (ret < 0) { - PERROR("relay cmd pipe"); - goto error; - } + ret = utils_create_pipe_cloexec(relay_cmd_pipe); - for (i = 0; i < 2; i++) { - ret = fcntl(relay_cmd_pipe[i], F_SETFD, FD_CLOEXEC); - if (ret < 0) { - PERROR("fcntl relay_cmd_pipe"); - goto error; - } - } - -error: return ret; } @@ -1480,7 +1551,7 @@ int main(int argc, char **argv) /* Parse arguments */ progname = argv[0]; if ((ret = parse_args(argc, argv) < 0)) { - goto error; + goto exit; } if ((ret = set_signal_handler()) < 0) { @@ -1492,7 +1563,7 @@ int main(int argc, char **argv) ret = daemon(0, 0); if (ret < 0) { PERROR("daemon"); - goto error; + goto exit; } } @@ -1503,7 +1574,7 @@ int main(int argc, char **argv) if (control_uri->port < 1024 || data_uri->port < 1024) { ERR("Need to be root to use ports < 1024"); ret = -1; - goto error; + goto exit; } } @@ -1568,6 +1639,7 @@ exit: if (!ret) { exit(EXIT_SUCCESS); } + error: exit(EXIT_FAILURE); }