X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-relayd%2Fmain.c;h=bb038a670b0a763afd41b1343cbbb1e919fbe1f7;hp=a00fb3c48ab258459de880d3a4720ef9dfd9b0ed;hb=309167d2a6f59d0c8cbf64eb23ba912cdea76a34;hpb=f66c074ce00abc984ec7cb57a4fa31b1d9cffd72 diff --git a/src/bin/lttng-relayd/main.c b/src/bin/lttng-relayd/main.c index a00fb3c48..bb038a670 100644 --- a/src/bin/lttng-relayd/main.c +++ b/src/bin/lttng-relayd/main.c @@ -47,16 +47,17 @@ #include #include #include -#include #include #include #include +#include "cmd.h" +#include "utils.h" #include "lttng-relayd.h" /* command line options */ +char *opt_output_path; static int opt_daemon; -static char *opt_output_path; static struct lttng_uri *control_uri; static struct lttng_uri *data_uri; @@ -104,12 +105,12 @@ static void usage(void) { fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname); - fprintf(stderr, " -h, --help Display this usage.\n"); - fprintf(stderr, " -d, --daemonize Start as a daemon.\n"); - fprintf(stderr, " -C, --control-port Control port listening (URI)\n"); - fprintf(stderr, " -D, --data-port Data port listening (URI)\n"); - fprintf(stderr, " -o, --output Output path for traces (PATH)\n"); - fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n"); + fprintf(stderr, " -h, --help Display this usage.\n"); + fprintf(stderr, " -d, --daemonize Start as a daemon.\n"); + fprintf(stderr, " -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"); } static @@ -254,7 +255,7 @@ int notify_thread_pipe(int wpipe) do { ret = write(wpipe, "!", 1); } while (ret < 0 && errno == EINTR); - if (ret < 0) { + if (ret < 0 || ret != 1) { PERROR("write poll pipe"); } @@ -455,6 +456,13 @@ 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; @@ -507,8 +515,6 @@ void *relay_thread_listener(void *data) while (1) { DBG("Listener accepting connections"); - nb_fd = LTTNG_POLL_GETNB(&events); - restart: ret = lttng_poll_wait(&events, -1); if (ret < 0) { @@ -521,6 +527,8 @@ restart: goto error; } + nb_fd = ret; + DBG("Relay new connection received"); for (i = 0; i < nb_fd; i++) { /* Fetch once the poll data */ @@ -662,7 +670,7 @@ void *relay_thread_dispatcher(void *data) sizeof(struct relay_command)); } while (ret < 0 && errno == EINTR); free(relay_cmd); - if (ret < 0) { + if (ret < 0 || ret != sizeof(struct relay_command)) { PERROR("write cmd pipe"); goto error; } @@ -679,191 +687,40 @@ error: } /* - * Return the realpath(3) of the path even if the last directory token does not - * exist. For example, with /tmp/test1/test2, if test2/ does not exist but the - * /tmp/test1 does, the real path is returned. In normal time, realpath(3) - * fails if the end point directory does not exist. + * Get stream from stream id. + * Need to be called with RCU read-side lock held. */ static -char *expand_full_path(const char *path) +struct relay_stream *relay_stream_from_stream_id(uint64_t stream_id, + struct lttng_ht *streams_ht) { - const char *end_path = path; - char *next, *cut_path, *expanded_path, *respath; - - /* Find last token delimited by '/' */ - while ((next = strpbrk(end_path + 1, "/"))) { - end_path = next; - } - - /* Cut last token from original path */ - cut_path = strndup(path, end_path - path); + struct lttng_ht_node_ulong *node; + struct lttng_ht_iter iter; + struct relay_stream *ret; - expanded_path = malloc(PATH_MAX); - if (expanded_path == NULL) { - respath = NULL; + lttng_ht_lookup(streams_ht, + (void *)((unsigned long) stream_id), + &iter); + node = lttng_ht_iter_get_node_ulong(&iter); + if (node == NULL) { + DBG("Relay stream %" PRIu64 " not found", stream_id); + ret = NULL; goto end; } - respath = realpath(cut_path, expanded_path); - if (respath == NULL) { - switch (errno) { - case ENOENT: - ERR("%s: No such file or directory", cut_path); - break; - default: - PERROR("realpath"); - break; - } - free(expanded_path); - } else { - /* Add end part to expanded path */ - strcat(respath, end_path); - } -end: - free(cut_path); - return respath; -} - - -/* - * config_get_default_path - * - * Returns the HOME directory path. Caller MUST NOT free(3) the return pointer. - */ -static -char *config_get_default_path(void) -{ - return getenv("HOME"); -} - -/* - * Create recursively directory using the FULL path. - */ -static -int mkdir_recursive(char *path, mode_t mode) -{ - char *p, tmp[PATH_MAX]; - struct stat statbuf; - size_t len; - int ret; - - ret = snprintf(tmp, sizeof(tmp), "%s", path); - if (ret < 0) { - PERROR("snprintf mkdir"); - goto error; - } - - len = ret; - if (tmp[len - 1] == '/') { - tmp[len - 1] = 0; - } - - for (p = tmp + 1; *p; p++) { - if (*p == '/') { - *p = 0; - if (tmp[strlen(tmp) - 1] == '.' && - tmp[strlen(tmp) - 2] == '.' && - tmp[strlen(tmp) - 3] == '/') { - ERR("Using '/../' is not permitted in the trace path (%s)", - tmp); - ret = -1; - goto error; - } - ret = stat(tmp, &statbuf); - if (ret < 0) { - ret = mkdir(tmp, mode); - if (ret < 0) { - if (errno != EEXIST) { - PERROR("mkdir recursive"); - ret = -errno; - goto error; - } - } - } - *p = '/'; - } - } - - ret = mkdir(tmp, mode); - if (ret < 0) { - if (errno != EEXIST) { - PERROR("mkdir recursive last piece"); - ret = -errno; - } else { - ret = 0; - } - } + ret = caa_container_of(node, struct relay_stream, stream_n); -error: +end: 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) -{ - if (opt_output_path == NULL) { - return create_output_path_auto(path_name); - } else { - return create_output_path_noauto(path_name); - } -} - static void deferred_free_stream(struct rcu_head *head) { struct relay_stream *stream = caa_container_of(head, struct relay_stream, rcu_node); + free(stream->path_name); + free(stream->channel_name); free(stream); } @@ -885,8 +742,6 @@ void relay_delete_session(struct relay_command *cmd, struct lttng_ht *streams_ht DBG("Relay deleting session %" PRIu64, cmd->session->id); - lttcomm_destroy_sock(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); @@ -910,6 +765,55 @@ void relay_delete_session(struct relay_command *cmd, struct lttng_ht *streams_ht free(cmd->session); } +/* + * Handle the RELAYD_CREATE_SESSION command. + * + * On success, send back the session id or else return a negative value. + */ +static +int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr, + struct relay_command *cmd) +{ + 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)); + + session = zmalloc(sizeof(struct relay_session)); + if (session == NULL) { + PERROR("relay session zmalloc"); + ret = -1; + goto error; + } + + session->id = ++last_relay_session_id; + session->sock = cmd->sock; + cmd->session = session; + + reply.session_id = htobe64(session->id); + + DBG("Created session %" PRIu64, session->id); + +error: + if (ret < 0) { + reply.ret_code = htobe32(LTTNG_ERR_FATAL); + } else { + reply.ret_code = htobe32(LTTNG_OK); + } + + send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); + if (send_ret < 0) { + ERR("Relayd sending session id"); + ret = send_ret; + } + + return ret; +} + /* * relay_add_stream: allocate a new stream for a session */ @@ -918,26 +822,16 @@ int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr, struct relay_command *cmd, struct lttng_ht *streams_ht) { struct relay_session *session = cmd->session; - struct lttcomm_relayd_add_stream stream_info; struct relay_stream *stream = NULL; struct lttcomm_relayd_status_stream reply; - char *path = NULL, *root_path = NULL; int ret, send_ret; - if (!session || session->version_check_done == 0) { + if (!session || cmd->version_check_done == 0) { ERR("Trying to add a stream before version check"); ret = -1; goto end_no_session; } - /* FIXME : use data_size for something ? */ - ret = cmd->sock->ops->recvmsg(cmd->sock, &stream_info, - sizeof(struct lttcomm_relayd_add_stream), MSG_WAITALL); - if (ret < sizeof(struct lttcomm_relayd_add_stream)) { - ERR("Relay didn't receive valid add_stream struct size : %d", ret); - ret = -1; - goto end_no_session; - } stream = zmalloc(sizeof(struct relay_stream)); if (stream == NULL) { PERROR("relay stream zmalloc"); @@ -945,63 +839,81 @@ int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr, 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; - root_path = create_output_path(stream_info.pathname); - if (!root_path) { - ret = -1; - goto end; - } - ret = mkdir_recursive(root_path, S_IRWXU | S_IRWXG); + ret = utils_mkdir_recursive(stream->path_name, S_IRWXU | S_IRWXG); if (ret < 0) { ERR("relay creating output directory"); goto end; } - ret = asprintf(&path, "%s/%s", root_path, stream_info.channel_name); - if (ret < 0) { - PERROR("asprintf stream path"); - goto end; - } - - ret = open(path, O_WRONLY|O_CREAT|O_TRUNC, S_IRWXU|S_IRWXG|S_IRWXO); + /* + * No need to use run_as API here because whatever we receives, the relayd + * uses its own credentials for the stream files. + */ + ret = utils_create_stream_file(stream->path_name, stream->channel_name, + stream->tracefile_size, 0, -1, -1, NULL); if (ret < 0) { - PERROR("Relay creating trace file"); + ERR("Create output file"); goto end; } - stream->fd = ret; - DBG("Tracefile %s created", path); + 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); + } lttng_ht_node_init_ulong(&stream->stream_n, (unsigned long) stream->stream_handle); lttng_ht_add_unique_ulong(streams_ht, &stream->stream_n); - DBG("Relay new stream added %s", stream_info.channel_name); + DBG("Relay new stream added %s", stream->channel_name); end: - free(path); - free(root_path); + reply.handle = htobe64(stream->stream_handle); /* send the session id to the client or a negative return code on error */ if (ret < 0) { reply.ret_code = htobe32(LTTNG_ERR_UNK); + /* stream was not properly added to the ht, so free it */ + free(stream); } else { reply.ret_code = htobe32(LTTNG_OK); } - reply.handle = htobe64(stream->stream_handle); + send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(struct lttcomm_relayd_status_stream), 0); if (send_ret < 0) { ERR("Relay sending stream id"); + ret = send_ret; } rcu_read_unlock(); end_no_session: return ret; + +err_free_stream: + free(stream->path_name); + free(stream->channel_name); + free(stream); + return ret; } /* @@ -1016,37 +928,32 @@ int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr, struct lttcomm_relayd_generic_reply reply; struct relay_stream *stream; int ret, send_ret; - struct lttng_ht_node_ulong *node; struct lttng_ht_iter iter; DBG("Close stream received"); - if (!session || session->version_check_done == 0) { + if (!session || cmd->version_check_done == 0) { ERR("Trying to close a stream before version check"); ret = -1; goto end_no_session; } ret = cmd->sock->ops->recvmsg(cmd->sock, &stream_info, - sizeof(struct lttcomm_relayd_close_stream), MSG_WAITALL); + sizeof(struct lttcomm_relayd_close_stream), 0); if (ret < sizeof(struct lttcomm_relayd_close_stream)) { - ERR("Relay didn't receive valid add_stream struct size : %d", ret); + if (ret == 0) { + /* Orderly shutdown. Not necessary to print an error. */ + DBG("Socket %d did an orderly shutdown", cmd->sock->fd); + } else { + ERR("Relay didn't receive valid add_stream struct size : %d", ret); + } ret = -1; goto end_no_session; } rcu_read_lock(); - lttng_ht_lookup(streams_ht, - (void *)((unsigned long) be64toh(stream_info.stream_id)), - &iter); - node = lttng_ht_iter_get_node_ulong(&iter); - if (node == NULL) { - DBG("Relay stream %" PRIu64 " not found", be64toh(stream_info.stream_id)); - ret = -1; - goto end_unlock; - } - - stream = caa_container_of(node, struct relay_stream, stream_n); + stream = relay_stream_from_stream_id(be64toh(stream_info.stream_id), + streams_ht); if (!stream) { ret = -1; goto end_unlock; @@ -1062,6 +969,7 @@ int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr, if (delret < 0) { PERROR("close stream"); } + iter.iter.node = &stream->stream_n.node; delret = lttng_ht_del(streams_ht, &iter); assert(!delret); call_rcu(&stream->rcu_node, @@ -1081,6 +989,7 @@ end_unlock: sizeof(struct lttcomm_relayd_generic_reply), 0); if (send_ret < 0) { ERR("Relay sending stream id"); + ret = send_ret; } end_no_session: @@ -1131,34 +1040,6 @@ int relay_start(struct lttcomm_relayd_hdr *recv_hdr, return ret; } -/* - * Get stream from stream id. - * Need to be called with RCU read-side lock held. - */ -static -struct relay_stream *relay_stream_from_stream_id(uint64_t stream_id, - struct lttng_ht *streams_ht) -{ - struct lttng_ht_node_ulong *node; - struct lttng_ht_iter iter; - struct relay_stream *ret; - - lttng_ht_lookup(streams_ht, - (void *)((unsigned long) stream_id), - &iter); - node = lttng_ht_iter_get_node_ulong(&iter); - if (node == NULL) { - DBG("Relay stream %" PRIu64 " not found", stream_id); - ret = NULL; - goto end; - } - - ret = caa_container_of(node, struct relay_stream, stream_n); - -end: - return ret; -} - /* * Append padding to the file pointed by the file descriptor fd. */ @@ -1181,7 +1062,7 @@ static int write_padding_to_file(int fd, uint32_t size) do { ret = write(fd, zeros, size); } while (ret < 0 && errno == EINTR); - if (ret < 0) { + if (ret < 0 || ret != size) { PERROR("write padding to file"); } @@ -1220,23 +1101,29 @@ int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr, if (data_buffer_size < data_size) { /* In case the realloc fails, we can free the memory */ - char *tmp_data_ptr = data_buffer; - data_buffer = realloc(data_buffer, data_size); - if (!data_buffer) { + char *tmp_data_ptr; + + tmp_data_ptr = realloc(data_buffer, data_size); + if (!tmp_data_ptr) { ERR("Allocating data buffer"); - free(tmp_data_ptr); + free(data_buffer); ret = -1; goto end; } + data_buffer = tmp_data_ptr; data_buffer_size = data_size; } memset(data_buffer, 0, data_size); DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size); - ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, - MSG_WAITALL); + ret = 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; - ERR("Relay didn't receive the whole metadata"); goto end; } metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer; @@ -1253,7 +1140,7 @@ int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr, ret = write(metadata_stream->fd, metadata_struct->payload, payload_size); } while (ret < 0 && errno == EINTR); - if (ret < payload_size) { + if (ret < 0 || ret != payload_size) { ERR("Relay error writing metadata on file"); ret = -1; goto end_unlock; @@ -1278,33 +1165,48 @@ end: */ static int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_command *cmd) + struct relay_command *cmd, struct lttng_ht *streams_ht) { int ret; - struct lttcomm_relayd_version reply; - struct relay_session *session; + struct lttcomm_relayd_version reply, msg; - if (cmd->session == NULL) { - session = zmalloc(sizeof(struct relay_session)); - if (session == NULL) { - PERROR("relay session zmalloc"); - ret = -1; - goto end; + assert(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."); } - session->id = ++last_relay_session_id; - DBG("Created session %" PRIu64, session->id); - cmd->session = session; - } else { - session = cmd->session; + ret = -1; + goto end; } - session->version_check_done = 1; - ret = sscanf(VERSION, "%u.%u", &reply.major, &reply.minor); - if (ret < 2) { - ERR("Error in scanning version"); - ret = -1; + 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, streams_ht); + ret = 0; goto end; } + + cmd->major = reply.major; + /* We adapt to the lowest compatible version */ + if (reply.minor <= be32toh(msg.minor)) { + cmd->minor = reply.minor; + } else { + cmd->minor = be32toh(msg.minor); + } + reply.major = htobe32(reply.major); reply.minor = htobe32(reply.minor); ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, @@ -1312,8 +1214,9 @@ int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr, if (ret < 0) { ERR("Relay sending version"); } - DBG("Version check done (%u.%u)", be32toh(reply.major), - be32toh(reply.minor)); + + DBG("Version check done using protocol %u.%u", cmd->major, + cmd->minor); end: return ret; @@ -1331,21 +1234,25 @@ int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr, struct lttcomm_relayd_generic_reply reply; struct relay_stream *stream; int ret; - struct lttng_ht_node_ulong *node; - struct lttng_ht_iter iter; uint64_t last_net_seq_num, stream_id; DBG("Data pending command received"); - if (!session || session->version_check_done == 0) { + if (!session || cmd->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), MSG_WAITALL); + ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); if (ret < sizeof(msg)) { - ERR("Relay didn't receive valid data_pending struct size : %d", ret); + 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 valid data_pending struct size : %d", + ret); + } ret = -1; goto end_no_session; } @@ -1354,23 +1261,18 @@ int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr, last_net_seq_num = be64toh(msg.last_net_seq_num); rcu_read_lock(); - lttng_ht_lookup(streams_ht, (void *)((unsigned long) stream_id), &iter); - node = lttng_ht_iter_get_node_ulong(&iter); - if (node == NULL) { - DBG("Relay stream %" PRIu64 " not found", stream_id); + stream = relay_stream_from_stream_id(stream_id, streams_ht); + if (stream == NULL) { ret = -1; goto end_unlock; } - stream = caa_container_of(node, struct relay_stream, stream_n); - assert(stream); - DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64 " and last_seq %" PRIu64, stream_id, stream->prev_seq, last_net_seq_num); /* Avoid wrapping issue */ - if (((int64_t) (stream->prev_seq - last_net_seq_num)) <= 0) { + if (((int64_t) (stream->prev_seq - last_net_seq_num)) >= 0) { /* Data has in fact been written and is NOT pending */ ret = 0; } else { @@ -1378,6 +1280,9 @@ int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr, ret = 1; } + /* Pending check is now done. */ + stream->data_pending_check_done = 1; + end_unlock: rcu_read_unlock(); @@ -1400,19 +1305,201 @@ end_no_session: */ static int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_command *cmd) + struct relay_command *cmd, struct lttng_ht *streams_ht) { 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) { + 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); + 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); + } else { + ERR("Relay didn't receive valid begin data_pending struct size: %d", + ret); + } + ret = -1; + goto end_no_session; + } + + stream_id = be64toh(msg.stream_id); + + rcu_read_lock(); + cds_lfht_for_each_entry(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(); + reply.ret_code = htobe32(LTTNG_OK); ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); if (ret < 0) { ERR("Relay data quiescent control ret code failed"); } +end_no_session: + return ret; +} + +/* + * Initialize a data pending command. This means that a client is about to ask + * for data pending for each stream he/she holds. Simply iterate over all + * streams of a session and set the data_pending_check_done flag. + * + * This command returns to the client a LTTNG_OK code. + */ +static +int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr, + struct relay_command *cmd, struct lttng_ht *streams_ht) +{ + int ret; + struct lttng_ht_iter iter; + struct lttcomm_relayd_begin_data_pending msg; + struct lttcomm_relayd_generic_reply reply; + struct relay_stream *stream; + uint64_t session_id; + + assert(recv_hdr); + assert(cmd); + assert(streams_ht); + + DBG("Init streams for data pending"); + + if (!cmd->session || cmd->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); + 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); + } else { + ERR("Relay didn't receive valid begin data_pending struct size: %d", + ret); + } + ret = -1; + goto end_no_session; + } + + session_id = be64toh(msg.session_id); + + /* + * Iterate over all streams to set the begin data pending flag. For now, the + * streams are indexed by stream handle so we have to iterate over all + * streams to find the one associated with the right session_id. + */ + rcu_read_lock(); + cds_lfht_for_each_entry(streams_ht->ht, &iter.iter, stream, stream_n.node) { + if (stream->session->id == session_id) { + stream->data_pending_check_done = 0; + DBG("Set begin data pending flag to stream %" PRIu64, + stream->stream_handle); + } + } + rcu_read_unlock(); + + /* All good, send back reply. */ + reply.ret_code = htobe32(LTTNG_OK); + + ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); + if (ret < 0) { + ERR("Relay begin data pending send reply failed"); + } + +end_no_session: + return ret; +} + +/* + * End data pending command. This will check, for a given session id, if each + * stream associated with it has its data_pending_check_done flag set. If not, + * this means that the client lost track of the stream but the data is still + * being streamed on our side. In this case, we inform the client that data is + * inflight. + * + * Return to the client if there is data in flight or not with a ret_code. + */ +static +int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr, + struct relay_command *cmd, struct lttng_ht *streams_ht) +{ + int ret; + struct lttng_ht_iter iter; + struct lttcomm_relayd_end_data_pending msg; + struct lttcomm_relayd_generic_reply reply; + struct relay_stream *stream; + uint64_t session_id; + uint32_t is_data_inflight = 0; + + assert(recv_hdr); + assert(cmd); + assert(streams_ht); + + DBG("End data pending command"); + + if (!cmd->session || cmd->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); + 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); + } else { + ERR("Relay didn't receive valid end data_pending struct size: %d", + ret); + } + ret = -1; + goto end_no_session; + } + + session_id = be64toh(msg.session_id); + + /* Iterate over all streams to see if the begin data pending flag is set. */ + rcu_read_lock(); + cds_lfht_for_each_entry(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; + } + } + rcu_read_unlock(); + + /* All good, send back reply. */ + reply.ret_code = htobe32(is_data_inflight); + + ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); + if (ret < 0) { + ERR("Relay end data pending send reply failed"); + } + +end_no_session: return ret; } @@ -1426,11 +1513,9 @@ int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr, int ret = 0; switch (be32toh(recv_hdr->cmd)) { - /* case RELAYD_CREATE_SESSION: ret = relay_create_session(recv_hdr, cmd); break; - */ case RELAYD_ADD_STREAM: ret = relay_add_stream(recv_hdr, cmd, streams_ht); break; @@ -1441,7 +1526,7 @@ int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr, ret = relay_recv_metadata(recv_hdr, cmd, streams_ht); break; case RELAYD_VERSION: - ret = relay_send_version(recv_hdr, cmd); + ret = relay_send_version(recv_hdr, cmd, streams_ht); break; case RELAYD_CLOSE_STREAM: ret = relay_close_stream(recv_hdr, cmd, streams_ht); @@ -1450,7 +1535,13 @@ int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr, ret = relay_data_pending(recv_hdr, cmd, streams_ht); break; case RELAYD_QUIESCENT_CONTROL: - ret = relay_quiescent_control(recv_hdr, cmd); + ret = relay_quiescent_control(recv_hdr, cmd, streams_ht); + break; + case RELAYD_BEGIN_DATA_PENDING: + ret = relay_begin_data_pending(recv_hdr, cmd, streams_ht); + break; + case RELAYD_END_DATA_PENDING: + ret = relay_end_data_pending(recv_hdr, cmd, streams_ht); break; case RELAYD_UPDATE_SYNC_INFO: default: @@ -1478,9 +1569,14 @@ int relay_process_data(struct relay_command *cmd, struct lttng_ht *streams_ht) uint32_t data_size; ret = cmd->sock->ops->recvmsg(cmd->sock, &data_hdr, - sizeof(struct lttcomm_relayd_data_hdr), MSG_WAITALL); + sizeof(struct lttcomm_relayd_data_hdr), 0); if (ret <= 0) { - ERR("Connections seems to be closed"); + if (ret == 0) { + /* Orderly shutdown. Not necessary to print an error. */ + DBG("Socket %d did an orderly shutdown", cmd->sock->fd); + } else { + ERR("Unable to receive data header on sock %d", cmd->sock->fd); + } ret = -1; goto end; } @@ -1496,14 +1592,16 @@ int relay_process_data(struct relay_command *cmd, struct lttng_ht *streams_ht) data_size = be32toh(data_hdr.data_size); if (data_buffer_size < data_size) { - char *tmp_data_ptr = data_buffer; - data_buffer = realloc(data_buffer, data_size); - if (!data_buffer) { + char *tmp_data_ptr; + + tmp_data_ptr = realloc(data_buffer, data_size); + if (!tmp_data_ptr) { ERR("Allocating data buffer"); - free(tmp_data_ptr); + free(data_buffer); ret = -1; goto end_unlock; } + data_buffer = tmp_data_ptr; data_buffer_size = data_size; } memset(data_buffer, 0, data_size); @@ -1512,16 +1610,37 @@ int relay_process_data(struct relay_command *cmd, struct lttng_ht *streams_ht) DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64, data_size, stream_id, net_seq_num); - ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, MSG_WAITALL); + 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_unlock; } + if (stream->tracefile_size > 0 && + (stream->tracefile_size_current + data_size) > + stream->tracefile_size) { + ret = utils_rotate_stream_file(stream->path_name, + stream->channel_name, stream->tracefile_size, + stream->tracefile_count, -1, -1, + stream->fd, &(stream->tracefile_count_current), + &stream->fd); + if (ret < 0) { + ERR("Rotating output file"); + goto end; + } + stream->fd = ret; + /* Reset current size because we just perform a stream rotation. */ + stream->tracefile_size_current = 0; + } + stream->tracefile_size_current += data_size; do { ret = write(stream->fd, data_buffer, data_size); } while (ret < 0 && errno == EINTR); - if (ret < data_size) { + if (ret < 0 || ret != data_size) { ERR("Relay error writing data to file"); ret = -1; goto end_unlock; @@ -1585,7 +1704,9 @@ int relay_add_connection(int fd, struct lttng_poll_event *events, PERROR("Relay command zmalloc"); goto error; } - ret = read(fd, relay_connection, sizeof(struct relay_command)); + do { + ret = read(fd, relay_connection, sizeof(struct relay_command)); + } while (ret < 0 && errno == EINTR); if (ret < 0 || ret < sizeof(struct relay_command)) { PERROR("read relay cmd pipe"); goto error_read; @@ -1640,8 +1761,8 @@ void relay_del_connection(struct lttng_ht *relay_connections_ht, static void *relay_thread_worker(void *data) { - int i, ret, pollfd, err = -1; - uint32_t revents, nb_fd; + 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; @@ -1676,14 +1797,11 @@ void *relay_thread_worker(void *data) goto error; } +restart: while (1) { - /* Zeroed the events structure */ - lttng_poll_reset(&events); - - nb_fd = LTTNG_POLL_GETNB(&events); + int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1; /* Infinite blocking call, waiting for transmission */ - restart: DBG3("Relayd worker thread polling..."); ret = lttng_poll_wait(&events, -1); if (ret < 0) { @@ -1696,10 +1814,17 @@ void *relay_thread_worker(void *data) goto error; } + nb_fd = ret; + + /* + * Process control. The control connection is prioritised so we don't + * starve it with high throughout put tracing data on the data + * connection. + */ for (i = 0; i < nb_fd; i++) { /* Fetch once the poll data */ - revents = LTTNG_POLL_GETEV(&events, i); - pollfd = LTTNG_POLL_GETFD(&events, i); + uint32_t revents = LTTNG_POLL_GETEV(&events, i); + int pollfd = LTTNG_POLL_GETFD(&events, i); /* Thread quit pipe has been closed. Killing thread. */ ret = check_thread_quit_pipe(pollfd, revents); @@ -1721,7 +1846,7 @@ void *relay_thread_worker(void *data) goto error; } } - } else if (revents > 0) { + } else if (revents) { rcu_read_lock(); lttng_ht_lookup(relay_connections_ht, (void *)((unsigned long) pollfd), @@ -1741,18 +1866,24 @@ void *relay_thread_worker(void *data) relay_del_connection(relay_connections_ht, streams_ht, &iter, relay_connection); + 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, streams_ht, &iter, relay_connection); + if (last_seen_data_fd == pollfd) { + last_seen_data_fd = last_notdel_data_fd; + } } else if (revents & LPOLLIN) { /* control socket */ if (relay_connection->type == RELAY_CONTROL) { ret = relay_connection->sock->ops->recvmsg( relay_connection->sock, &recv_hdr, - sizeof(struct lttcomm_relayd_hdr), MSG_WAITALL); + sizeof(struct lttcomm_relayd_hdr), 0); /* connection closed */ if (ret <= 0) { relay_cleanup_poll_connection(&events, pollfd); @@ -1768,34 +1899,101 @@ void *relay_thread_worker(void *data) ret = relay_process_control(&recv_hdr, relay_connection, streams_ht); - /* - * there was an error in processing a control - * command: clear the session - * */ if (ret < 0) { + /* Clear the session on error. */ relay_cleanup_poll_connection(&events, pollfd); relay_del_connection(relay_connections_ht, streams_ht, &iter, relay_connection); DBG("Connection closed with %d", pollfd); } + seen_control = 1; } - /* data socket */ - } else if (relay_connection->type == RELAY_DATA) { - ret = relay_process_data(relay_connection, streams_ht); - /* connection closed */ - if (ret < 0) { - relay_cleanup_poll_connection(&events, pollfd); - relay_del_connection(relay_connections_ht, - streams_ht, &iter, - relay_connection); - DBG("Data connection closed with %d", pollfd); - } + } else { + /* + * Flag the last seen data fd not deleted. It will be + * used as the last seen fd if any fd gets deleted in + * this first loop. + */ + last_notdel_data_fd = pollfd; } } rcu_read_unlock(); } } + + /* + * The last loop handled a control request, go back to poll to make + * sure we prioritise the control socket. + */ + if (seen_control) { + continue; + } + + if (last_seen_data_fd >= 0) { + for (i = 0; i < nb_fd; i++) { + int pollfd = LTTNG_POLL_GETFD(&events, i); + if (last_seen_data_fd == pollfd) { + idx = i; + break; + } + } + } + + /* Process data connection. */ + for (i = idx + 1; i < nb_fd; i++) { + /* Fetch the poll data. */ + uint32_t revents = LTTNG_POLL_GETEV(&events, i); + int pollfd = LTTNG_POLL_GETFD(&events, i); + + /* Skip the command pipe. It's handled in the first loop. */ + if (pollfd == relay_cmd_pipe[0]) { + 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); + + if (revents & LPOLLIN) { + if (relay_connection->type != RELAY_DATA) { + continue; + } + + ret = relay_process_data(relay_connection, streams_ht); + /* connection closed */ + if (ret < 0) { + relay_cleanup_poll_connection(&events, pollfd); + relay_del_connection(relay_connections_ht, + streams_ht, &iter, + relay_connection); + DBG("Data connection closed with %d", pollfd); + /* + * Every goto restart call sets the last seen fd where + * here we don't really care since we gracefully + * continue the loop after the connection is deleted. + */ + } else { + /* Keep last seen port. */ + last_seen_data_fd = pollfd; + rcu_read_unlock(); + goto restart; + } + } + rcu_read_unlock(); + } + } + last_seen_data_fd = -1; } exit: @@ -1860,7 +2058,7 @@ int main(int argc, char **argv) /* Parse arguments */ progname = argv[0]; - if ((ret = parse_args(argc, argv) < 0)) { + if ((ret = parse_args(argc, argv)) < 0) { goto exit; } @@ -1868,6 +2066,20 @@ int main(int argc, char **argv) goto exit; } + /* Try to create directory if -o, --output is specified. */ + if (opt_output_path) { + if (*opt_output_path != '/') { + ERR("Please specify an absolute path for -o, --output PATH"); + goto exit; + } + + ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG); + if (ret < 0) { + ERR("Unable to create %s", opt_output_path); + goto exit; + } + } + /* Daemonize */ if (opt_daemon) { ret = daemon(0, 0); @@ -1899,6 +2111,9 @@ int main(int argc, char **argv) /* Set up max poll set size */ lttng_poll_set_max_size(); + /* Initialize communication library */ + lttcomm_init(); + /* Setup the dispatcher thread */ ret = pthread_create(&dispatcher_thread, NULL, relay_thread_dispatcher, (void *) NULL);