X-Git-Url: http://git.efficios.com/?a=blobdiff_plain;f=src%2Fbin%2Flttng-relayd%2Fmain.c;h=f5bb9409f29d56bcd68fea31033baed1df3b2aa8;hb=0a9eff096ef480f16445be173badc90f5b6540d2;hp=7b385b49f5264e2ace8823ffbc0f8fe4747630a5;hpb=a44ca2ca85e4b64729f7b88b1919fd6737dfff8a;p=lttng-tools.git diff --git a/src/bin/lttng-relayd/main.c b/src/bin/lttng-relayd/main.c index 7b385b49f..f5bb9409f 100644 --- a/src/bin/lttng-relayd/main.c +++ b/src/bin/lttng-relayd/main.c @@ -18,7 +18,6 @@ * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. */ -#define _GNU_SOURCE #define _LGPL_SOURCE #include #include @@ -35,12 +34,13 @@ #include #include #include +#include #include #include #include #include #include -#include +#include #include #include @@ -56,9 +56,13 @@ #include #include #include -#include +#include +#include +#include +#include #include +#include "version.h" #include "cmd.h" #include "ctf-trace.h" #include "index.h" @@ -72,10 +76,21 @@ #include "stream.h" #include "connection.h" #include "tracefile-array.h" +#include "tcp_keep_alive.h" + +enum relay_connection_status { + RELAY_CONNECTION_STATUS_OK, + /* An error occured while processing an event on the connection. */ + RELAY_CONNECTION_STATUS_ERROR, + /* Connection closed/shutdown cleanly. */ + RELAY_CONNECTION_STATUS_CLOSED, +}; /* command line options */ -char *opt_output_path; -static int opt_daemon, opt_background; +char *opt_output_path, *opt_working_directory; +static int opt_daemon, opt_background, opt_print_version; +int opt_group_output_by_session; +int opt_group_output_by_host; /* * We need to wait for listener and live listener threads, as well as @@ -83,6 +98,10 @@ static int opt_daemon, opt_background; */ #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 + 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. */ @@ -132,9 +151,8 @@ static uint64_t last_relay_stream_id; */ 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; +/* Cap of file desriptors to be in simultaneous use by the relay daemon. */ +static unsigned int lttng_opt_fd_cap; /* Global relay stream hash table. */ struct lttng_ht *relay_streams_ht; @@ -148,6 +166,9 @@ struct lttng_ht *sessions_ht; /* Relayd health monitoring */ struct health_app *health_relayd; +/* Global fd tracker. */ +struct fd_tracker *the_fd_tracker; + static struct option long_options[] = { { "control-port", 1, 0, 'C', }, { "data-port", 1, 0, 'D', }, @@ -155,31 +176,35 @@ static struct option long_options[] = { { "daemonize", 0, 0, 'd', }, { "background", 0, 0, 'b', }, { "group", 1, 0, 'g', }, + { "fd-cap", 1, 0, '\0', }, { "help", 0, 0, 'h', }, { "output", 1, 0, 'o', }, { "verbose", 0, 0, 'v', }, { "config", 1, 0, 'f' }, + { "version", 0, 0, 'V' }, + { "working-directory", 1, 0, 'w', }, + { "group-output-by-session", 0, 0, 's', }, + { "group-output-by-host", 0, 0, 'p', }, { NULL, 0, 0, 0, }, }; -static const char *config_ignore_options[] = { "help", "config" }; +static const char *config_ignore_options[] = { "help", "config", "version" }; -/* - * usage function on stderr - */ -static void usage(void) +static void print_version(void) { + fprintf(stdout, "%s\n", VERSION); +} + +static void relayd_config_log(void) { - fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname); - fprintf(stderr, " -h, --help Display this usage.\n"); - fprintf(stderr, " -d, --daemonize Start as a daemon.\n"); - fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n"); - fprintf(stderr, " -C, --control-port URL Control port listening.\n"); - fprintf(stderr, " -D, --data-port URL Data port listening.\n"); - fprintf(stderr, " -L, --live-port URL Live view port listening.\n"); - fprintf(stderr, " -o, --output PATH Output path for traces. Must use an absolute path.\n"); - fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n"); - fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n"); - fprintf(stderr, " -f --config Load daemon configuration file\n"); + DBG("LTTng-relayd " VERSION " - " VERSION_NAME "%s%s", + GIT_VERSION[0] == '\0' ? "" : " - " GIT_VERSION, + EXTRA_VERSION_NAME[0] == '\0' ? "" : " - " EXTRA_VERSION_NAME); + if (EXTRA_VERSION_DESCRIPTION[0] != '\0') { + DBG("LTTng-relayd extra version description:\n\t" EXTRA_VERSION_DESCRIPTION "\n"); + } + if (EXTRA_VERSION_PATCHES[0] != '\0') { + DBG("LTTng-relayd extra patches:\n\t" EXTRA_VERSION_PATCHES "\n"); + } } /* @@ -194,9 +219,33 @@ static int set_option(int opt, const char *arg, const char *optname) switch (opt) { case 0: - fprintf(stderr, "option %s", optname); - if (arg) { - fprintf(stderr, " with arg %s\n", arg); + if (!strcmp(optname, "fd-cap")) { + unsigned long v; + + errno = 0; + v = strtoul(arg, NULL, 0); + if (errno != 0 || !isdigit(arg[0])) { + ERR("Wrong value in --fd-cap parameter: %s", arg); + ret = -1; + goto end; + } + if (v < DEFAULT_RELAYD_MINIMAL_FD_CAP) { + ERR("File descriptor cap must be set to at least %d", + DEFAULT_RELAYD_MINIMAL_FD_CAP); + } + if (v >= UINT_MAX) { + ERR("File descriptor cap overflow in --fd-cap parameter: %s", arg); + ret = -1; + goto end; + } + lttng_opt_fd_cap = (unsigned int) v; + DBG3("File descriptor cap set to %u", lttng_opt_fd_cap); + + } else { + fprintf(stderr, "unknown option %s", optname); + if (arg) { + fprintf(stderr, " with arg %s\n", arg); + } } break; case 'C': @@ -265,8 +314,15 @@ static int set_option(int opt, const char *arg, const char *optname) } break; case 'h': - usage(); + ret = utils_show_man_page(8, "lttng-relayd"); + if (ret) { + ERR("Cannot view man page lttng-relayd(8)"); + perror("exec"); + } exit(EXIT_FAILURE); + case 'V': + opt_print_version = 1; + break; case 'o': if (lttng_is_setuid_setgid()) { WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", @@ -280,6 +336,20 @@ static int set_option(int opt, const char *arg, const char *optname) } } break; + case 'w': + if (lttng_is_setuid_setgid()) { + WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", + "-w, --working-directory"); + } else { + ret = asprintf(&opt_working_directory, "%s", arg); + if (ret < 0) { + ret = -errno; + PERROR("asprintf working_directory"); + goto end; + } + } + break; + case 'v': /* Verbose level can increase using multiple -v */ if (arg) { @@ -291,6 +361,20 @@ static int set_option(int opt, const char *arg, const char *optname) } } break; + case 's': + if (opt_group_output_by_host) { + ERR("Cannot set --group-output-by-session, --group-output-by-host already defined"); + exit(EXIT_FAILURE); + } + opt_group_output_by_session = 1; + break; + case 'p': + if (opt_group_output_by_session) { + ERR("Cannot set --group-output-by-host, --group-output-by-session already defined"); + exit(EXIT_FAILURE); + } + opt_group_output_by_host = 1; + break; default: /* Unknown option or other error. * Error is printed by getopt, just return */ @@ -307,7 +391,7 @@ end: /* * config_entry_handler_cb used to handle options read from a config file. - * See config_entry_handler_cb comment in common/config/config.h for the + * 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) @@ -360,6 +444,16 @@ end: return ret; } +static void parse_env_options(void) +{ + char *value = NULL; + + value = lttng_secure_getenv(DEFAULT_LTTNG_RELAYD_WORKING_DIRECTORY_ENV); + if (value) { + opt_working_directory = value; + } +} + static int set_options(int argc, char **argv) { int c, ret = 0, option_index = 0, retval = 0; @@ -477,6 +571,23 @@ static int set_options(int argc, char **argv) goto exit; } } + if (lttng_opt_fd_cap == 0) { + int ret; + struct rlimit rlimit; + + ret = getrlimit(RLIMIT_NOFILE, &rlimit); + if (ret) { + PERROR("Failed to get file descriptor limit"); + retval = -1; + } + + lttng_opt_fd_cap = rlimit.rlim_cur; + } + + if (!opt_group_output_by_session && !opt_group_output_by_host) { + /* Group by host by default */ + opt_group_output_by_host = 1; + } exit: free(optstring); @@ -485,13 +596,9 @@ exit: static void print_global_objects(void) { - rcu_register_thread(); - print_viewer_streams(); print_relay_streams(); print_sessions(); - - rcu_unregister_thread(); } /* @@ -514,7 +621,7 @@ static void relayd_cleanup(void) free(opt_output_path); /* Close thread quit pipes */ - utils_close_pipe(thread_quit_pipe); + (void) fd_tracker_util_pipe_close(the_fd_tracker, thread_quit_pipe); uri_free(control_uri); uri_free(data_uri); @@ -523,6 +630,7 @@ static void relayd_cleanup(void) if (tracing_group_name_override) { free((void *) tracing_group_name); } + fd_tracker_log(the_fd_tracker); } /* @@ -594,9 +702,6 @@ int lttng_relay_stop_threads(void) static void sighandler(int sig) { switch (sig) { - case SIGPIPE: - DBG("SIGPIPE caught"); - return; case SIGINT: DBG("SIGINT caught"); if (lttng_relay_stop_threads()) { @@ -632,9 +737,10 @@ static 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; @@ -645,12 +751,13 @@ static int set_signal_handler(void) return ret; } - if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) { + if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) { PERROR("sigaction"); return ret; } - if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) { + sa.sa_handler = SIG_IGN; + if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) { PERROR("sigaction"); return ret; } @@ -677,17 +784,26 @@ void lttng_relay_notify_ready(void) */ static int init_thread_quit_pipe(void) { - int ret; - - ret = utils_create_pipe_cloexec(thread_quit_pipe); + return fd_tracker_util_pipe_open_cloexec(the_fd_tracker, + "Quit pipe", thread_quit_pipe); +} - return ret; +/* + * Init health quit pipe. + * + * Return -1 on error or 0 if all pipes are created. + */ +static int init_health_quit_pipe(void) +{ + return fd_tracker_util_pipe_open_cloexec(the_fd_tracker, + "Health quit pipe", health_quit_pipe); } /* * 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_named_thread_poll_set(struct lttng_poll_event *events, + int size, const char *name) { int ret; @@ -696,10 +812,8 @@ static int create_thread_poll_set(struct lttng_poll_event *events, int size) goto error; } - ret = lttng_poll_create(events, size, LTTNG_CLOEXEC); - if (ret < 0) { - goto error; - } + ret = fd_tracker_util_poll_create(the_fd_tracker, + name, events, 1, LTTNG_CLOEXEC); /* Add quit pipe */ ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR); @@ -797,7 +911,7 @@ static void *relay_thread_listener(void *data) * Pass 3 as size here for the thread quit pipe, control and * data socket. */ - ret = create_thread_poll_set(&events, 3); + ret = create_named_thread_poll_set(&events, 3, "Listener thread epoll"); if (ret < 0) { goto error_create_poll; } @@ -864,10 +978,7 @@ restart: goto exit; } - if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { - ERR("socket poll error"); - goto error; - } else if (revents & LPOLLIN) { + if (revents & LPOLLIN) { /* * A new connection is requested, therefore a * sessiond/consumerd connection is allocated in @@ -903,6 +1014,15 @@ restart: lttcomm_destroy_sock(newsock); goto error; } + + 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); @@ -919,6 +1039,12 @@ restart: * exchange in cds_wfcq_enqueue. */ 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; } } } @@ -927,7 +1053,7 @@ exit: error: error_poll_add: error_testpoint: - lttng_poll_clean(&events); + (void) fd_tracker_util_poll_clean(the_fd_tracker, &events); error_create_poll: if (data_sock->fd >= 0) { ret = data_sock->ops->close(data_sock); @@ -975,12 +1101,16 @@ static void *relay_thread_dispatcher(void *data) health_code_update(); - while (!CMM_LOAD_SHARED(dispatch_thread_exit)) { + for (;;) { health_code_update(); /* Atomically prepare the queue futex */ futex_nto1_prepare(&relay_conn_queue.futex); + if (CMM_LOAD_SHARED(dispatch_thread_exit)) { + break; + } + do { health_code_update(); @@ -1035,22 +1165,26 @@ error_testpoint: * Set index data from the control port to a given index object. */ static int set_index_control_data(struct relay_index *index, - struct lttcomm_relayd_index *data) + struct lttcomm_relayd_index *data, + struct relay_connection *conn) { struct ctf_packet_index index_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. + * The index on disk is encoded in big endian. */ - 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; + index_data.packet_size = htobe64(data->packet_size); + index_data.content_size = htobe64(data->content_size); + index_data.timestamp_begin = htobe64(data->timestamp_begin); + index_data.timestamp_end = htobe64(data->timestamp_end); + index_data.events_discarded = htobe64(data->events_discarded); + index_data.stream_id = htobe64(data->stream_id); + + if (conn->minor >= 8) { + index->index_data.stream_instance_id = htobe64(data->stream_instance_id); + index->index_data.packet_seq_num = htobe64(data->packet_seq_num); + } + return relay_index_set_data(index, &index_data); } @@ -1059,19 +1193,21 @@ static int set_index_control_data(struct relay_index *index, * * 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_connection *conn) +static int relay_create_session(const struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn, + const struct lttng_buffer_view *payload) { - int ret = 0, send_ret; + int ret = 0; + ssize_t send_ret; struct relay_session *session; struct lttcomm_relayd_status_session reply; - char session_name[NAME_MAX]; - char hostname[HOST_NAME_MAX]; + char session_name[LTTNG_NAME_MAX]; + char hostname[LTTNG_HOST_NAME_MAX]; uint32_t live_timer = 0; bool snapshot = false; - memset(session_name, 0, NAME_MAX); - memset(hostname, 0, HOST_NAME_MAX); + memset(session_name, 0, LTTNG_NAME_MAX); + memset(hostname, 0, LTTNG_HOST_NAME_MAX); memset(&reply, 0, sizeof(reply)); @@ -1082,7 +1218,7 @@ static int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr, break; case 4: /* LTTng sessiond 2.4 */ default: - ret = cmd_create_session_2_4(conn, session_name, + ret = cmd_create_session_2_4(payload, session_name, hostname, &live_timer, &snapshot); } if (ret < 0) { @@ -1109,9 +1245,10 @@ send_reply: } send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); - if (send_ret < 0) { - ERR("Relayd sending session id"); - ret = send_ret; + if (send_ret < (ssize_t) sizeof(reply)) { + ERR("Failed to send \"create session\" command reply (ret = %zd)", + send_ret); + ret = -1; } return ret; @@ -1150,8 +1287,9 @@ static void publish_connection_local_streams(struct relay_connection *conn) /* * relay_add_stream: allocate a new stream for a session */ -static int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_connection *conn) +static int relay_add_stream(const struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn, + const struct lttng_buffer_view *payload) { int ret; ssize_t send_ret; @@ -1163,7 +1301,7 @@ static int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr, char *path_name = NULL, *channel_name = NULL; uint64_t tracefile_size = 0, tracefile_count = 0; - if (!session || conn->version_check_done == 0) { + if (!session || !conn->version_check_done) { ERR("Trying to add a stream before version check"); ret = -1; goto end_no_session; @@ -1171,13 +1309,14 @@ static int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr, 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); + ret = cmd_recv_stream_2_1(payload, &path_name, + &channel_name, session); 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); + ret = cmd_recv_stream_2_2(payload, &path_name, + &channel_name, &tracefile_size, &tracefile_count, + session); break; } if (ret < 0) { @@ -1217,9 +1356,10 @@ send_reply: send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(struct lttcomm_relayd_status_stream), 0); - if (send_ret < 0) { - ERR("Relay sending stream id"); - ret = (int) send_ret; + if (send_ret < (ssize_t) sizeof(reply)) { + ERR("Failed to send \"add stream\" command reply (ret = %zd)", + send_ret); + ret = -1; } end_no_session: @@ -1231,10 +1371,12 @@ end_no_session: /* * relay_close_stream: close a specific stream */ -static int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_connection *conn) +static int relay_close_stream(const struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn, + const struct lttng_buffer_view *payload) { - int ret, send_ret; + int ret; + ssize_t send_ret; struct relay_session *session = conn->session; struct lttcomm_relayd_close_stream stream_info; struct lttcomm_relayd_generic_reply reply; @@ -1242,33 +1384,45 @@ static int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr, DBG("Close stream received"); - if (!session || conn->version_check_done == 0) { + if (!session || !conn->version_check_done) { ERR("Trying to close a stream before version check"); ret = -1; goto end_no_session; } - 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); - } + if (payload->size < sizeof(stream_info)) { + ERR("Unexpected payload size in \"relay_close_stream\": expected >= %zu bytes, got %zu bytes", + sizeof(stream_info), payload->size); ret = -1; goto end_no_session; } + memcpy(&stream_info, payload->data, sizeof(stream_info)); + stream_info.stream_id = be64toh(stream_info.stream_id); + stream_info.last_net_seq_num = be64toh(stream_info.last_net_seq_num); - stream = stream_get_by_id(be64toh(stream_info.stream_id)); + stream = stream_get_by_id(stream_info.stream_id); if (!stream) { ret = -1; goto end; } + + /* + * Set last_net_seq_num before the close flag. Required by data + * pending check. + */ pthread_mutex_lock(&stream->lock); - stream->closed = true; - stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num); + stream->last_net_seq_num = stream_info.last_net_seq_num; + pthread_mutex_unlock(&stream->lock); + + /* + * 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; @@ -1287,6 +1441,90 @@ static int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr, viewer_stream_put(vstream); } } + stream_put(stream); + 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); + } + send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, + sizeof(struct lttcomm_relayd_generic_reply), 0); + if (send_ret < (ssize_t) sizeof(reply)) { + ERR("Failed to send \"close stream\" command reply (ret = %zd)", + send_ret); + ret = -1; + } + +end_no_session: + return ret; +} + +/* + * relay_reset_metadata: reset a metadata stream + */ +static +int relay_reset_metadata(const struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn, + const struct lttng_buffer_view *payload) +{ + int ret; + ssize_t send_ret; + struct relay_session *session = conn->session; + struct lttcomm_relayd_reset_metadata stream_info; + struct lttcomm_relayd_generic_reply reply; + struct relay_stream *stream; + + DBG("Reset metadata received"); + + if (!session || !conn->version_check_done) { + ERR("Trying to reset a metadata stream before version check"); + ret = -1; + goto end_no_session; + } + + if (payload->size < sizeof(stream_info)) { + ERR("Unexpected payload size in \"relay_reset_metadata\": expected >= %zu bytes, got %zu bytes", + sizeof(stream_info), payload->size); + ret = -1; + goto end_no_session; + } + memcpy(&stream_info, payload->data, sizeof(stream_info)); + stream_info.stream_id = be64toh(stream_info.stream_id); + stream_info.version = be64toh(stream_info.version); + + DBG("Update metadata to version %" PRIu64, stream_info.version); + + /* Unsupported for live sessions for now. */ + if (session->live_timer != 0) { + ret = -1; + goto end; + } + + stream = stream_get_by_id(stream_info.stream_id); + if (!stream) { + ret = -1; + goto end; + } + pthread_mutex_lock(&stream->lock); + if (!stream->is_metadata) { + ret = -1; + goto end_unlock; + } + + 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: pthread_mutex_unlock(&stream->lock); stream_put(stream); @@ -1299,9 +1537,10 @@ end: } 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; + if (send_ret < (ssize_t) sizeof(reply)) { + ERR("Failed to send \"reset metadata\" command reply (ret = %zd)", + send_ret); + ret = -1; } end_no_session: @@ -1314,14 +1553,13 @@ end_no_session: static void relay_unknown_command(struct relay_connection *conn) { struct lttcomm_relayd_generic_reply reply; - int ret; + ssize_t send_ret; memset(&reply, 0, sizeof(reply)); reply.ret_code = htobe32(LTTNG_ERR_UNK); - ret = conn->sock->ops->sendmsg(conn->sock, &reply, - sizeof(struct lttcomm_relayd_generic_reply), 0); - if (ret < 0) { - ERR("Relay sending unknown command"); + send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); + if (send_ret < sizeof(reply)) { + ERR("Failed to send \"unknown command\" command reply (ret = %zd)", send_ret); } } @@ -1329,10 +1567,12 @@ static void relay_unknown_command(struct relay_connection *conn) * 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_connection *conn) +static int relay_start(const struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn, + const struct lttng_buffer_view *payload) { - int ret = htobe32(LTTNG_OK); + int ret = 0; + ssize_t send_ret; struct lttcomm_relayd_generic_reply reply; struct relay_session *session = conn->session; @@ -1342,11 +1582,13 @@ static int relay_start(struct lttcomm_relayd_hdr *recv_hdr, } memset(&reply, 0, sizeof(reply)); - reply.ret_code = ret; - ret = conn->sock->ops->sendmsg(conn->sock, &reply, - sizeof(struct lttcomm_relayd_generic_reply), 0); - if (ret < 0) { - ERR("Relay sending start ack"); + reply.ret_code = htobe32(LTTNG_OK); + send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, + sizeof(reply), 0); + if (send_ret < (ssize_t) sizeof(reply)) { + ERR("Failed to send \"relay_start\" command reply (ret = %zd)", + send_ret); + ret = -1; } return ret; @@ -1385,15 +1627,16 @@ end: /* * relay_recv_metadata: receive the metadata for the session. */ -static int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_connection *conn) +static int relay_recv_metadata(const struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn, + const struct lttng_buffer_view *payload) { - int ret = htobe32(LTTNG_OK); + int ret = 0; ssize_t size_ret; struct relay_session *session = conn->session; - struct lttcomm_relayd_metadata_payload *metadata_struct; + struct lttcomm_relayd_metadata_payload metadata_payload_header; struct relay_stream *metadata_stream; - uint64_t data_size, payload_size; + uint64_t metadata_payload_size; if (!session) { ERR("Metadata sent before version check"); @@ -1401,44 +1644,22 @@ static int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr, goto end; } - data_size = payload_size = be64toh(recv_hdr->data_size); - if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) { + if (recv_hdr->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; + metadata_payload_size = recv_hdr->data_size - + sizeof(struct lttcomm_relayd_metadata_payload); - 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 = conn->sock->ops->recvmsg(conn->sock, data_buffer, data_size, 0); - if (ret < 0 || ret != data_size) { - if (ret == 0) { - /* Orderly shutdown. Not necessary to print an error. */ - DBG("Socket %d did an orderly shutdown", conn->sock->fd); - } else { - ERR("Relay didn't receive the whole metadata"); - } - ret = -1; - goto end; - } - metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer; + memcpy(&metadata_payload_header, payload->data, + sizeof(metadata_payload_header)); + metadata_payload_header.stream_id = be64toh( + metadata_payload_header.stream_id); + metadata_payload_header.padding_size = be32toh( + metadata_payload_header.padding_size); - metadata_stream = stream_get_by_id(be64toh(metadata_struct->stream_id)); + metadata_stream = stream_get_by_id(metadata_payload_header.stream_id); if (!metadata_stream) { ret = -1; goto end; @@ -1446,29 +1667,30 @@ static int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr, 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) { + size_ret = lttng_write(metadata_stream->stream_fd->fd, + payload->data + sizeof(metadata_payload_header), + metadata_payload_size); + if (size_ret < metadata_payload_size) { ERR("Relay error writing metadata on file"); ret = -1; goto end_put; } - ret = write_padding_to_file(metadata_stream->stream_fd->fd, - be32toh(metadata_struct->padding_size)); - if (ret < 0) { + size_ret = write_padding_to_file(metadata_stream->stream_fd->fd, + metadata_payload_header.padding_size); + if (size_ret < (int64_t) metadata_payload_header.padding_size) { + ret = -1; goto end_put; } metadata_stream->metadata_received += - payload_size + be32toh(metadata_struct->padding_size); + metadata_payload_size + metadata_payload_header.padding_size; DBG2("Relay metadata written. Updated metadata_received %" PRIu64, metadata_stream->metadata_received); end_put: pthread_mutex_unlock(&metadata_stream->lock); stream_put(metadata_stream); - end: return ret; } @@ -1476,54 +1698,64 @@ end: /* * relay_send_version: send relayd version number */ -static int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_connection *conn) +static int relay_send_version(const struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn, + const struct lttng_buffer_view *payload) { int ret; + ssize_t send_ret; struct lttcomm_relayd_version reply, msg; + bool compatible = true; - conn->version_check_done = 1; + conn->version_check_done = true; /* 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."); - } + if (payload->size < sizeof(msg)) { + ERR("Unexpected payload size in \"relay_send_version\": expected >= %zu bytes, got %zu bytes", + sizeof(msg), payload->size); ret = -1; goto end; } + memcpy(&msg, payload->data, sizeof(msg)); + msg.major = be32toh(msg.major); + msg.minor = be32toh(msg.minor); + 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)) { + if (reply.major != msg.major) { DBG("Incompatible major versions (%u vs %u), deleting session", - reply.major, be32toh(msg.major)); - connection_put(conn); - ret = 0; - goto end; + reply.major, msg.major); + compatible = false; } conn->major = reply.major; /* We adapt to the lowest compatible version */ - if (reply.minor <= be32toh(msg.minor)) { + if (reply.minor <= msg.minor) { conn->minor = reply.minor; } else { - conn->minor = be32toh(msg.minor); + conn->minor = msg.minor; } reply.major = htobe32(reply.major); reply.minor = htobe32(reply.minor); - ret = conn->sock->ops->sendmsg(conn->sock, &reply, - sizeof(struct lttcomm_relayd_version), 0); - if (ret < 0) { - ERR("Relay sending version"); + send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, + sizeof(reply), 0); + if (send_ret < (ssize_t) sizeof(reply)) { + ERR("Failed to send \"send version\" command reply (ret = %zd)", + send_ret); + ret = -1; + goto end; + } else { + ret = 0; + } + + if (!compatible) { + ret = -1; + goto end; } DBG("Version check done using protocol %u.%u", conn->major, @@ -1536,41 +1768,36 @@ 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_connection *conn) +static int relay_data_pending(const struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn, + const struct lttng_buffer_view *payload) { struct relay_session *session = conn->session; struct lttcomm_relayd_data_pending msg; struct lttcomm_relayd_generic_reply reply; struct relay_stream *stream; + ssize_t send_ret; int ret; - uint64_t last_net_seq_num, stream_id; DBG("Data pending command received"); - if (!session || conn->version_check_done == 0) { + if (!session || !conn->version_check_done) { ERR("Trying to check for data before version check"); ret = -1; goto end_no_session; } - ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0); - if (ret < sizeof(msg)) { - if (ret == 0) { - /* Orderly shutdown. Not necessary to print an error. */ - DBG("Socket %d did an orderly shutdown", conn->sock->fd); - } else { - ERR("Relay didn't receive valid data_pending struct size : %d", - ret); - } + if (payload->size < sizeof(msg)) { + ERR("Unexpected payload size in \"relay_data_pending\": expected >= %zu bytes, got %zu bytes", + sizeof(msg), payload->size); ret = -1; goto end_no_session; } + memcpy(&msg, payload->data, sizeof(msg)); + msg.stream_id = be64toh(msg.stream_id); + msg.last_net_seq_num = be64toh(msg.last_net_seq_num); - stream_id = be64toh(msg.stream_id); - last_net_seq_num = be64toh(msg.last_net_seq_num); - - stream = stream_get_by_id(stream_id); + stream = stream_get_by_id(msg.stream_id); if (stream == NULL) { ret = -1; goto end; @@ -1579,11 +1806,11 @@ static int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr, 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); + " and last_seq %" PRIu64, msg.stream_id, + stream->prev_seq, msg.last_net_seq_num); /* Avoid wrapping issue */ - if (((int64_t) (stream->prev_seq - last_net_seq_num)) >= 0) { + if (((int64_t) (stream->prev_seq - msg.last_net_seq_num)) >= 0) { /* Data has in fact been written and is NOT pending */ ret = 0; } else { @@ -1599,9 +1826,11 @@ end: memset(&reply, 0, sizeof(reply)); reply.ret_code = htobe32(ret); - ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); - if (ret < 0) { - ERR("Relay data pending ret code failed"); + send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); + if (send_ret < (ssize_t) sizeof(reply)) { + ERR("Failed to send \"data pending\" command reply (ret = %zd)", + send_ret); + ret = -1; } end_no_session: @@ -1616,52 +1845,53 @@ end_no_session: * 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_connection *conn) +static int relay_quiescent_control(const struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn, + const struct lttng_buffer_view *payload) { int ret; - uint64_t stream_id; + ssize_t send_ret; struct relay_stream *stream; struct lttcomm_relayd_quiescent_control msg; struct lttcomm_relayd_generic_reply reply; DBG("Checking quiescent state on control socket"); - if (!conn->session || conn->version_check_done == 0) { + if (!conn->session || !conn->version_check_done) { ERR("Trying to check for data before version check"); ret = -1; goto end_no_session; } - ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0); - if (ret < sizeof(msg)) { - if (ret == 0) { - /* Orderly shutdown. Not necessary to print an error. */ - DBG("Socket %d did an orderly shutdown", conn->sock->fd); - } else { - ERR("Relay didn't receive valid begin data_pending struct size: %d", - ret); - } + if (payload->size < sizeof(msg)) { + ERR("Unexpected payload size in \"relay_quiescent_control\": expected >= %zu bytes, got %zu bytes", + sizeof(msg), payload->size); ret = -1; goto end_no_session; } + memcpy(&msg, payload->data, sizeof(msg)); + msg.stream_id = be64toh(msg.stream_id); - stream_id = be64toh(msg.stream_id); - stream = stream_get_by_id(stream_id); + stream = stream_get_by_id(msg.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); + + DBG("Relay quiescent control pending flag set to %" PRIu64, msg.stream_id); stream_put(stream); reply: memset(&reply, 0, sizeof(reply)); reply.ret_code = htobe32(LTTNG_OK); - ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); - if (ret < 0) { - ERR("Relay data quiescent control ret code failed"); + send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); + if (send_ret < (ssize_t) sizeof(reply)) { + ERR("Failed to send \"quiescent control\" command reply (ret = %zd)", + send_ret); + ret = -1; + } else { + ret = 0; } end_no_session: @@ -1675,41 +1905,36 @@ end_no_session: * * This command returns to the client a LTTNG_OK code. */ -static int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_connection *conn) +static int relay_begin_data_pending(const struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn, + const struct lttng_buffer_view *payload) { int ret; + ssize_t send_ret; struct lttng_ht_iter iter; struct lttcomm_relayd_begin_data_pending msg; struct lttcomm_relayd_generic_reply reply; struct relay_stream *stream; - uint64_t session_id; assert(recv_hdr); assert(conn); DBG("Init streams for data pending"); - if (!conn->session || conn->version_check_done == 0) { + if (!conn->session || !conn->version_check_done) { ERR("Trying to check for data before version check"); ret = -1; goto end_no_session; } - ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0); - if (ret < sizeof(msg)) { - if (ret == 0) { - /* Orderly shutdown. Not necessary to print an error. */ - DBG("Socket %d did an orderly shutdown", conn->sock->fd); - } else { - ERR("Relay didn't receive valid begin data_pending struct size: %d", - ret); - } + if (payload->size < sizeof(msg)) { + ERR("Unexpected payload size in \"relay_begin_data_pending\": expected >= %zu bytes, got %zu bytes", + sizeof(msg), payload->size); ret = -1; goto end_no_session; } - - session_id = be64toh(msg.session_id); + memcpy(&msg, payload->data, sizeof(msg)); + msg.session_id = be64toh(msg.session_id); /* * Iterate over all streams to set the begin data pending flag. @@ -1723,7 +1948,7 @@ static int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr, if (!stream_get(stream)) { continue; } - if (stream->trace->session->id == session_id) { + if (stream->trace->session->id == msg.session_id) { pthread_mutex_lock(&stream->lock); stream->data_pending_check_done = false; pthread_mutex_unlock(&stream->lock); @@ -1738,9 +1963,13 @@ static int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr, /* All good, send back reply. */ reply.ret_code = htobe32(LTTNG_OK); - ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); - if (ret < 0) { - ERR("Relay begin data pending send reply failed"); + send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); + if (send_ret < (ssize_t) sizeof(reply)) { + ERR("Failed to send \"begin data pending\" command reply (ret = %zd)", + send_ret); + ret = -1; + } else { + ret = 0; } end_no_session: @@ -1756,39 +1985,34 @@ end_no_session: * * Return to the client if there is data in flight or not with a ret_code. */ -static int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_connection *conn) +static int relay_end_data_pending(const struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn, + const struct lttng_buffer_view *payload) { int ret; + ssize_t send_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; DBG("End data pending command"); - if (!conn->session || conn->version_check_done == 0) { + if (!conn->session || !conn->version_check_done) { ERR("Trying to check for data before version check"); ret = -1; goto end_no_session; } - ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0); - if (ret < sizeof(msg)) { - if (ret == 0) { - /* Orderly shutdown. Not necessary to print an error. */ - DBG("Socket %d did an orderly shutdown", conn->sock->fd); - } else { - ERR("Relay didn't receive valid end data_pending struct size: %d", - ret); - } + if (payload->size < sizeof(msg)) { + ERR("Unexpected payload size in \"relay_end_data_pending\": expected >= %zu bytes, got %zu bytes", + sizeof(msg), payload->size); ret = -1; goto end_no_session; } - - session_id = be64toh(msg.session_id); + memcpy(&msg, payload->data, sizeof(msg)); + msg.session_id = be64toh(msg.session_id); /* * Iterate over all streams to see if the begin data pending @@ -1800,7 +2024,7 @@ static int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr, if (!stream_get(stream)) { continue; } - if (stream->trace->session->id != session_id) { + if (stream->trace->session->id != msg.session_id) { stream_put(stream); continue; } @@ -1824,9 +2048,13 @@ static int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr, /* All good, send back reply. */ reply.ret_code = htobe32(is_data_inflight); - ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); - if (ret < 0) { - ERR("Relay end data pending send reply failed"); + send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); + if (send_ret < (ssize_t) sizeof(reply)) { + ERR("Failed to send \"end data pending\" command reply (ret = %zd)", + send_ret); + ret = -1; + } else { + ret = 0; } end_no_session: @@ -1838,43 +2066,55 @@ end_no_session: * * Return 0 on success else a negative value. */ -static int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_connection *conn) +static int relay_recv_index(const struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn, + const struct lttng_buffer_view *payload) { - int ret, send_ret; + int ret; + ssize_t send_ret; struct relay_session *session = conn->session; struct lttcomm_relayd_index index_info; struct relay_index *index; struct lttcomm_relayd_generic_reply reply; struct relay_stream *stream; - uint64_t net_seq_num; + size_t msg_len; assert(conn); DBG("Relay receiving index"); - if (!session || conn->version_check_done == 0) { + if (!session || !conn->version_check_done) { ERR("Trying to close a stream before version check"); ret = -1; goto end_no_session; } - ret = conn->sock->ops->recvmsg(conn->sock, &index_info, - sizeof(index_info), 0); - if (ret < sizeof(index_info)) { - if (ret == 0) { - /* Orderly shutdown. Not necessary to print an error. */ - DBG("Socket %d did an orderly shutdown", conn->sock->fd); - } else { - ERR("Relay didn't receive valid index struct size : %d", ret); - } + msg_len = lttcomm_relayd_index_len( + lttng_to_index_major(conn->major, conn->minor), + lttng_to_index_minor(conn->major, conn->minor)); + if (payload->size < msg_len) { + ERR("Unexpected payload size in \"relay_recv_index\": expected >= %zu bytes, got %zu bytes", + msg_len, payload->size); ret = -1; goto end_no_session; } + memcpy(&index_info, payload->data, msg_len); + index_info.relay_stream_id = be64toh(index_info.relay_stream_id); + index_info.net_seq_num = be64toh(index_info.net_seq_num); + index_info.packet_size = be64toh(index_info.packet_size); + index_info.content_size = be64toh(index_info.content_size); + index_info.timestamp_begin = be64toh(index_info.timestamp_begin); + index_info.timestamp_end = be64toh(index_info.timestamp_end); + index_info.events_discarded = be64toh(index_info.events_discarded); + index_info.stream_id = be64toh(index_info.stream_id); - net_seq_num = be64toh(index_info.net_seq_num); + if (conn->minor >= 8) { + index_info.stream_instance_id = + be64toh(index_info.stream_instance_id); + index_info.packet_seq_num = be64toh(index_info.packet_seq_num); + } - stream = stream_get_by_id(be64toh(index_info.relay_stream_id)); + stream = stream_get_by_id(index_info.relay_stream_id); if (!stream) { ERR("stream_get_by_id not found"); ret = -1; @@ -1893,8 +2133,7 @@ static int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr, */ if (stream->index_received_seqcount > 0 && stream->indexes_in_flight == 0) { - stream->beacon_ts_end = - be64toh(index_info.timestamp_end); + stream->beacon_ts_end = index_info.timestamp_end; } ret = 0; goto end_stream_put; @@ -1903,15 +2142,15 @@ static int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr, } if (stream->ctf_stream_id == -1ULL) { - stream->ctf_stream_id = be64toh(index_info.stream_id); + stream->ctf_stream_id = index_info.stream_id; } - index = relay_index_get_by_id_or_create(stream, net_seq_num); + index = relay_index_get_by_id_or_create(stream, index_info.net_seq_num); if (!index) { ret = -1; ERR("relay_index_get_by_id_or_create index NULL"); goto end_stream_put; } - if (set_index_control_data(index, &index_info)) { + if (set_index_control_data(index, &index_info, conn)) { ERR("set_index_control_data error"); relay_index_put(index); ret = -1; @@ -1943,9 +2182,9 @@ end: 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; + if (send_ret < (ssize_t) sizeof(reply)) { + ERR("Failed to send \"recv index\" command reply (ret = %zd)", send_ret); + ret = -1; } end_no_session: @@ -1957,17 +2196,19 @@ end_no_session: * * Return 0 on success else a negative value. */ -static int relay_streams_sent(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_connection *conn) +static int relay_streams_sent(const struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn, + const struct lttng_buffer_view *payload) { - int ret, send_ret; + int ret; + ssize_t send_ret; struct lttcomm_relayd_generic_reply reply; assert(conn); DBG("Relay receiving streams_sent"); - if (!conn->session || conn->version_check_done == 0) { + if (!conn->session || !conn->version_check_done) { ERR("Trying to close a stream before version check"); ret = -1; goto end_no_session; @@ -1982,9 +2223,10 @@ static int relay_streams_sent(struct lttcomm_relayd_hdr *recv_hdr, 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; + if (send_ret < (ssize_t) sizeof(reply)) { + ERR("Failed to send \"streams sent\" command reply (ret = %zd)", + send_ret); + ret = -1; } else { /* Success. */ ret = 0; @@ -1994,54 +2236,71 @@ end_no_session: return ret; } -/* - * Process the commands received on the control socket - */ -static int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr, - struct relay_connection *conn) +#define DBG_CMD(cmd_name, conn) \ + DBG3("Processing \"%s\" command for socket %i", cmd_name, conn->sock->fd); + +static int relay_process_control_command(struct relay_connection *conn, + const struct lttcomm_relayd_hdr *header, + const struct lttng_buffer_view *payload) { int ret = 0; - switch (be32toh(recv_hdr->cmd)) { + switch (header->cmd) { case RELAYD_CREATE_SESSION: - ret = relay_create_session(recv_hdr, conn); + DBG_CMD("RELAYD_CREATE_SESSION", conn); + ret = relay_create_session(header, conn, payload); break; case RELAYD_ADD_STREAM: - ret = relay_add_stream(recv_hdr, conn); + DBG_CMD("RELAYD_ADD_STREAM", conn); + ret = relay_add_stream(header, conn, payload); break; case RELAYD_START_DATA: - ret = relay_start(recv_hdr, conn); + DBG_CMD("RELAYD_START_DATA", conn); + ret = relay_start(header, conn, payload); break; case RELAYD_SEND_METADATA: - ret = relay_recv_metadata(recv_hdr, conn); + DBG_CMD("RELAYD_SEND_METADATA", conn); + ret = relay_recv_metadata(header, conn, payload); break; case RELAYD_VERSION: - ret = relay_send_version(recv_hdr, conn); + DBG_CMD("RELAYD_VERSION", conn); + ret = relay_send_version(header, conn, payload); break; case RELAYD_CLOSE_STREAM: - ret = relay_close_stream(recv_hdr, conn); + DBG_CMD("RELAYD_CLOSE_STREAM", conn); + ret = relay_close_stream(header, conn, payload); break; case RELAYD_DATA_PENDING: - ret = relay_data_pending(recv_hdr, conn); + DBG_CMD("RELAYD_DATA_PENDING", conn); + ret = relay_data_pending(header, conn, payload); break; case RELAYD_QUIESCENT_CONTROL: - ret = relay_quiescent_control(recv_hdr, conn); + DBG_CMD("RELAYD_QUIESCENT_CONTROL", conn); + ret = relay_quiescent_control(header, conn, payload); break; case RELAYD_BEGIN_DATA_PENDING: - ret = relay_begin_data_pending(recv_hdr, conn); + DBG_CMD("RELAYD_BEGIN_DATA_PENDING", conn); + ret = relay_begin_data_pending(header, conn, payload); break; case RELAYD_END_DATA_PENDING: - ret = relay_end_data_pending(recv_hdr, conn); + DBG_CMD("RELAYD_END_DATA_PENDING", conn); + ret = relay_end_data_pending(header, conn, payload); break; case RELAYD_SEND_INDEX: - ret = relay_recv_index(recv_hdr, conn); + DBG_CMD("RELAYD_SEND_INDEX", conn); + ret = relay_recv_index(header, conn, payload); break; case RELAYD_STREAMS_SENT: - ret = relay_streams_sent(recv_hdr, conn); + DBG_CMD("RELAYD_STREAMS_SENT", conn); + ret = relay_streams_sent(header, conn, payload); + break; + case RELAYD_RESET_METADATA: + DBG_CMD("RELAYD_RESET_METADATA", conn); + ret = relay_reset_metadata(header, conn, payload); break; case RELAYD_UPDATE_SYNC_INFO: default: - ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd)); + ERR("Received unknown command (%u)", header->cmd); relay_unknown_command(conn); ret = -1; goto end; @@ -2051,6 +2310,193 @@ end: return ret; } +static enum relay_connection_status relay_process_control_receive_payload( + struct relay_connection *conn) +{ + int ret = 0; + enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK; + struct lttng_dynamic_buffer *reception_buffer = + &conn->protocol.ctrl.reception_buffer; + struct ctrl_connection_state_receive_payload *state = + &conn->protocol.ctrl.state.receive_payload; + struct lttng_buffer_view payload_view; + + if (state->left_to_receive == 0) { + /* Short-circuit for payload-less commands. */ + goto reception_complete; + } + ret = conn->sock->ops->recvmsg(conn->sock, + reception_buffer->data + state->received, + state->left_to_receive, MSG_DONTWAIT); + if (ret < 0) { + if (errno != EAGAIN && errno != EWOULDBLOCK) { + PERROR("Unable to receive command payload on sock %d", + conn->sock->fd); + status = RELAY_CONNECTION_STATUS_ERROR; + } + goto end; + } else if (ret == 0) { + DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd); + status = RELAY_CONNECTION_STATUS_CLOSED; + goto end; + } + + assert(ret > 0); + assert(ret <= state->left_to_receive); + + state->left_to_receive -= ret; + state->received += ret; + + if (state->left_to_receive > 0) { + /* + * Can't transition to the protocol's next state, wait to + * receive the rest of the header. + */ + DBG3("Partial reception of control connection protocol payload (received %" PRIu64 " bytes, %" PRIu64 " bytes left to receive, fd = %i)", + state->received, state->left_to_receive, + conn->sock->fd); + goto end; + } + +reception_complete: + DBG("Done receiving control command payload: fd = %i, payload size = %" PRIu64 " bytes", + conn->sock->fd, state->received); + /* + * The payload required to process the command has been received. + * A view to the reception buffer is forwarded to the various + * commands and the state of the control is reset on success. + * + * Commands are responsible for sending their reply to the peer. + */ + payload_view = lttng_buffer_view_from_dynamic_buffer(reception_buffer, + 0, -1); + ret = relay_process_control_command(conn, + &state->header, &payload_view); + if (ret < 0) { + status = RELAY_CONNECTION_STATUS_ERROR; + goto end; + } + + ret = connection_reset_protocol_state(conn); + if (ret) { + status = RELAY_CONNECTION_STATUS_ERROR; + } +end: + return status; +} + +static enum relay_connection_status relay_process_control_receive_header( + struct relay_connection *conn) +{ + int ret = 0; + enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK; + struct lttcomm_relayd_hdr header; + struct lttng_dynamic_buffer *reception_buffer = + &conn->protocol.ctrl.reception_buffer; + struct ctrl_connection_state_receive_header *state = + &conn->protocol.ctrl.state.receive_header; + + assert(state->left_to_receive != 0); + + ret = conn->sock->ops->recvmsg(conn->sock, + reception_buffer->data + state->received, + state->left_to_receive, MSG_DONTWAIT); + if (ret < 0) { + if (errno != EAGAIN && errno != EWOULDBLOCK) { + PERROR("Unable to receive control command header on sock %d", + conn->sock->fd); + status = RELAY_CONNECTION_STATUS_ERROR; + } + goto end; + } else if (ret == 0) { + DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd); + status = RELAY_CONNECTION_STATUS_CLOSED; + goto end; + } + + assert(ret > 0); + assert(ret <= state->left_to_receive); + + state->left_to_receive -= ret; + state->received += ret; + + if (state->left_to_receive > 0) { + /* + * Can't transition to the protocol's next state, wait to + * receive the rest of the header. + */ + DBG3("Partial reception of control connection protocol header (received %" PRIu64 " bytes, %" PRIu64 " bytes left to receive, fd = %i)", + state->received, state->left_to_receive, + conn->sock->fd); + goto end; + } + + /* Transition to next state: receiving the command's payload. */ + conn->protocol.ctrl.state_id = + CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD; + memcpy(&header, reception_buffer->data, sizeof(header)); + header.circuit_id = be64toh(header.circuit_id); + header.data_size = be64toh(header.data_size); + header.cmd = be32toh(header.cmd); + header.cmd_version = be32toh(header.cmd_version); + memcpy(&conn->protocol.ctrl.state.receive_payload.header, + &header, sizeof(header)); + + DBG("Done receiving control command header: fd = %i, cmd = %" PRIu32 ", cmd_version = %" PRIu32 ", payload size = %" PRIu64 " bytes", + conn->sock->fd, header.cmd, header.cmd_version, + header.data_size); + + if (header.data_size > DEFAULT_NETWORK_RELAYD_CTRL_MAX_PAYLOAD_SIZE) { + ERR("Command header indicates a payload (%" PRIu64 " bytes) that exceeds the maximal payload size allowed on a control connection.", + header.data_size); + status = RELAY_CONNECTION_STATUS_ERROR; + goto end; + } + + conn->protocol.ctrl.state.receive_payload.left_to_receive = + header.data_size; + conn->protocol.ctrl.state.receive_payload.received = 0; + ret = lttng_dynamic_buffer_set_size(reception_buffer, + header.data_size); + if (ret) { + status = RELAY_CONNECTION_STATUS_ERROR; + goto end; + } + + if (header.data_size == 0) { + /* + * Manually invoke the next state as the poll loop + * will not wake-up to allow us to proceed further. + */ + status = relay_process_control_receive_payload(conn); + } +end: + return status; +} + +/* + * Process the commands received on the control socket + */ +static enum relay_connection_status relay_process_control( + struct relay_connection *conn) +{ + enum relay_connection_status status; + + switch (conn->protocol.ctrl.state_id) { + case CTRL_CONNECTION_STATE_RECEIVE_HEADER: + status = relay_process_control_receive_header(conn); + break; + case CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD: + status = relay_process_control_receive_payload(conn); + break; + default: + ERR("Unknown control connection protocol state encountered."); + abort(); + } + + return status; +} + /* * Handle index for a data stream. * @@ -2059,7 +2505,7 @@ end: * 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) + bool rotate_index) { int ret = 0; uint64_t data_offset; @@ -2068,8 +2514,8 @@ static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num, /* Get data offset because we are about to update the index. */ data_offset = htobe64(stream->tracefile_size_current); - DBG("handle_index_data: stream %" PRIu64 " data offset %" PRIu64, - stream->stream_handle, 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 @@ -2082,41 +2528,36 @@ static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num, goto end; } - if (rotate_index || !stream->index_fd) { - int fd; + if (rotate_index || !stream->index_file) { + uint32_t major, minor; - /* Put ref on previous index_fd. */ - if (stream->index_fd) { - stream_fd_put(stream->index_fd); - stream->index_fd = NULL; + /* Put ref on previous index_file. */ + if (stream->index_file) { + lttng_index_file_put(stream->index_file); + stream->index_file = NULL; } - - fd = index_create_file(stream->path_name, stream->channel_name, + 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)); - if (fd < 0) { + 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; /* Put self-ref for this index due to error. */ relay_index_put(index); - goto end; - } - stream->index_fd = stream_fd_create(fd); - if (!stream->index_fd) { - ret = -1; - if (close(fd)) { - PERROR("Error closing FD %d", fd); - } - /* Put self-ref for this index due to error. */ - relay_index_put(index); - /* Will put the local ref. */ + index = NULL; goto end; } } - if (relay_index_set_fd(index, stream->index_fd, data_offset)) { + 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; } @@ -2130,82 +2571,94 @@ static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num, } else { /* Put self-ref for this index due to error. */ relay_index_put(index); + index = NULL; ret = -1; } end: return ret; } -/* - * relay_process_data: Process the data received on the data socket - */ -static int relay_process_data(struct relay_connection *conn) +static enum relay_connection_status relay_process_data_receive_header( + struct relay_connection *conn) { - int ret = 0, rotate_index = 0; - ssize_t size_ret; + int ret; + enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK; + struct data_connection_state_receive_header *state = + &conn->protocol.data.state.receive_header; + struct lttcomm_relayd_data_hdr header; struct relay_stream *stream; - struct lttcomm_relayd_data_hdr data_hdr; - uint64_t stream_id; - uint64_t net_seq_num; - uint32_t data_size; - struct relay_session *session; - bool new_stream = false; - - ret = conn->sock->ops->recvmsg(conn->sock, &data_hdr, - sizeof(struct lttcomm_relayd_data_hdr), 0); - if (ret <= 0) { - if (ret == 0) { - /* Orderly shutdown. Not necessary to print an error. */ - DBG("Socket %d did an orderly shutdown", conn->sock->fd); - } else { - ERR("Unable to receive data header on sock %d", conn->sock->fd); + + assert(state->left_to_receive != 0); + + ret = conn->sock->ops->recvmsg(conn->sock, + state->header_reception_buffer + state->received, + state->left_to_receive, MSG_DONTWAIT); + if (ret < 0) { + if (errno != EAGAIN && errno != EWOULDBLOCK) { + PERROR("Unable to receive data header on sock %d", conn->sock->fd); + status = RELAY_CONNECTION_STATUS_ERROR; } - ret = -1; + goto end; + } else if (ret == 0) { + /* Orderly shutdown. Not necessary to print an error. */ + DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd); + status = RELAY_CONNECTION_STATUS_CLOSED; goto end; } - stream_id = be64toh(data_hdr.stream_id); - stream = stream_get_by_id(stream_id); - if (!stream) { - ret = -1; + assert(ret > 0); + assert(ret <= state->left_to_receive); + + state->left_to_receive -= ret; + state->received += ret; + + if (state->left_to_receive > 0) { + /* + * Can't transition to the protocol's next state, wait to + * receive the rest of the header. + */ + DBG3("Partial reception of data connection header (received %" PRIu64 " bytes, %" PRIu64 " bytes left to receive, fd = %i)", + state->received, state->left_to_receive, + conn->sock->fd); + ret = 0; 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_stream_put; - } - 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); + /* Transition to next state: receiving the payload. */ + conn->protocol.data.state_id = DATA_CONNECTION_STATE_RECEIVE_PAYLOAD; - DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64, - data_size, stream_id, net_seq_num); - ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, data_size, 0); - if (ret <= 0) { - if (ret == 0) { - /* Orderly shutdown. Not necessary to print an error. */ - DBG("Socket %d did an orderly shutdown", conn->sock->fd); - } - ret = -1; - goto end_stream_put; + memcpy(&header, state->header_reception_buffer, sizeof(header)); + header.circuit_id = be64toh(header.circuit_id); + header.stream_id = be64toh(header.stream_id); + header.data_size = be32toh(header.data_size); + header.net_seq_num = be64toh(header.net_seq_num); + header.padding_size = be32toh(header.padding_size); + memcpy(&conn->protocol.data.state.receive_payload.header, &header, sizeof(header)); + + conn->protocol.data.state.receive_payload.left_to_receive = + header.data_size; + conn->protocol.data.state.receive_payload.received = 0; + conn->protocol.data.state.receive_payload.rotate_index = false; + + DBG("Received data connection header on fd %i: circuit_id = %" PRIu64 ", stream_id = %" PRIu64 ", data_size = %" PRIu32 ", net_seq_num = %" PRIu64 ", padding_size = %" PRIu32, + conn->sock->fd, header.circuit_id, + header.stream_id, header.data_size, + header.net_seq_num, header.padding_size); + + stream = stream_get_by_id(header.stream_id); + if (!stream) { + DBG("relay_process_data_receive_payload: Cannot find stream %" PRIu64, + header.stream_id); + /* Protocol error. */ + status = RELAY_CONNECTION_STATUS_ERROR; + goto end; } 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_current + header.data_size) > stream->tracefile_size) { uint64_t old_id, new_id; @@ -2221,63 +2674,206 @@ static int relay_process_data(struct relay_connection *conn) -1, stream->stream_fd->fd, &new_id, &stream->stream_fd->fd); if (ret < 0) { - ERR("Rotating stream output file"); + ERR("Failed to rotate stream output file"); + status = RELAY_CONNECTION_STATUS_ERROR; goto end_stream_unlock; } + /* * Reset current size because we just performed a stream * rotation. */ stream->tracefile_size_current = 0; - rotate_index = 1; + conn->protocol.data.state.receive_payload.rotate_index = true; + } + + ret = 0; +end_stream_unlock: + pthread_mutex_unlock(&stream->lock); + stream_put(stream); +end: + return status; +} + +static enum relay_connection_status relay_process_data_receive_payload( + struct relay_connection *conn) +{ + int ret; + enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK; + struct relay_stream *stream; + struct data_connection_state_receive_payload *state = + &conn->protocol.data.state.receive_payload; + const size_t chunk_size = RECV_DATA_BUFFER_SIZE; + char data_buffer[chunk_size]; + bool partial_recv = false; + bool new_stream = false, close_requested = false; + uint64_t left_to_receive = state->left_to_receive; + struct relay_session *session; + + DBG3("Receiving data for stream id %" PRIu64 " seqnum %" PRIu64 ", %" PRIu64" bytes received, %" PRIu64 " bytes left to receive", + state->header.stream_id, state->header.net_seq_num, + state->received, left_to_receive); + + stream = stream_get_by_id(state->header.stream_id); + if (!stream) { + /* Protocol error. */ + ERR("relay_process_data_receive_payload: cannot find stream %" PRIu64, + state->header.stream_id); + status = RELAY_CONNECTION_STATUS_ERROR; + goto end; + } + + pthread_mutex_lock(&stream->lock); + session = stream->trace->session; + if (!conn->session) { + ret = connection_set_session(conn, session); + if (ret) { + status = RELAY_CONNECTION_STATUS_ERROR; + goto end_stream_unlock; + } } /* - * Index are handled in protocol version 2.4 and above. Also, - * snapshot and index are NOT supported. + * The size of the "chunk" received on any iteration is bounded by: + * - the data left to receive, + * - the data immediately available on the socket, + * - the on-stack data buffer */ - if (session->minor >= 4 && !session->snapshot) { - ret = handle_index_data(stream, net_seq_num, rotate_index); + while (left_to_receive > 0 && !partial_recv) { + ssize_t write_ret; + size_t recv_size = min(left_to_receive, chunk_size); + + ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, + recv_size, MSG_DONTWAIT); if (ret < 0) { + if (errno != EAGAIN && errno != EWOULDBLOCK) { + PERROR("Socket %d error", conn->sock->fd); + status = RELAY_CONNECTION_STATUS_ERROR; + } goto end_stream_unlock; + } else if (ret == 0) { + /* No more data ready to be consumed on socket. */ + DBG3("No more data ready for consumption on data socket of stream id %" PRIu64, + state->header.stream_id); + status = RELAY_CONNECTION_STATUS_CLOSED; + break; + } else if (ret < (int) recv_size) { + /* + * All the data available on the socket has been + * consumed. + */ + partial_recv = true; } + + recv_size = ret; + + /* Write data to stream output fd. */ + write_ret = lttng_write(stream->stream_fd->fd, data_buffer, + recv_size); + if (write_ret < (ssize_t) recv_size) { + ERR("Relay error writing data to file"); + status = RELAY_CONNECTION_STATUS_ERROR; + goto end_stream_unlock; + } + + left_to_receive -= recv_size; + state->received += recv_size; + state->left_to_receive = left_to_receive; + + DBG2("Relay wrote %zd bytes to tracefile for stream id %" PRIu64, + write_ret, stream->stream_handle); } - /* Write data to stream output fd. */ - size_ret = lttng_write(stream->stream_fd->fd, data_buffer, data_size); - if (size_ret < data_size) { - ERR("Relay error writing data to file"); - ret = -1; + if (state->left_to_receive > 0) { + /* + * Did not receive all the data expected, wait for more data to + * become available on the socket. + */ + DBG3("Partial receive on data connection of stream id %" PRIu64 ", %" PRIu64 " bytes received, %" PRIu64 " bytes left to receive", + state->header.stream_id, state->received, + state->left_to_receive); goto end_stream_unlock; } - DBG2("Relay wrote %zd bytes to tracefile for stream id %" PRIu64, - size_ret, stream->stream_handle); - ret = write_padding_to_file(stream->stream_fd->fd, - be32toh(data_hdr.padding_size)); - if (ret < 0) { + state->header.padding_size); + if ((int64_t) ret < (int64_t) state->header.padding_size) { + ERR("write_padding_to_file: fail stream %" PRIu64 " net_seq_num %" PRIu64 " ret %d", + stream->stream_handle, + state->header.net_seq_num, ret); + status = RELAY_CONNECTION_STATUS_ERROR; goto end_stream_unlock; } - stream->tracefile_size_current += - data_size + be32toh(data_hdr.padding_size); + + + if (session->minor >= 4 && !session->snapshot) { + ret = handle_index_data(stream, state->header.net_seq_num, + state->rotate_index); + if (ret < 0) { + ERR("handle_index_data: fail stream %" PRIu64 " net_seq_num %" PRIu64 " ret %d", + stream->stream_handle, + state->header.net_seq_num, ret); + status = RELAY_CONNECTION_STATUS_ERROR; + goto end_stream_unlock; + } + } + + stream->tracefile_size_current += state->header.data_size + + state->header.padding_size; + if (stream->prev_seq == -1ULL) { new_stream = true; } - stream->prev_seq = net_seq_num; + stream->prev_seq = state->header.net_seq_num; + + /* + * Resetting the protocol state (to RECEIVE_HEADER) will trash the + * contents of *state which are aliased (union) to the same location as + * the new state. Don't use it beyond this point. + */ + connection_reset_protocol_state(conn); + state = NULL; end_stream_unlock: + close_requested = stream->close_requested; pthread_mutex_unlock(&stream->lock); + if (close_requested && left_to_receive == 0) { + try_stream_close(stream); + } + if (new_stream) { pthread_mutex_lock(&session->lock); uatomic_set(&session->new_streams, 1); pthread_mutex_unlock(&session->lock); } -end_stream_put: + stream_put(stream); end: - return ret; + return status; +} + +/* + * relay_process_data: Process the data received on the data socket + */ +static enum relay_connection_status relay_process_data( + struct relay_connection *conn) +{ + enum relay_connection_status status; + + switch (conn->protocol.data.state_id) { + case DATA_CONNECTION_STATE_RECEIVE_HEADER: + status = relay_process_data_receive_header(conn); + break; + case DATA_CONNECTION_STATE_RECEIVE_PAYLOAD: + status = relay_process_data_receive_payload(conn); + break; + default: + ERR("Unexpected data connection communication state."); + abort(); + } + + return status; } static void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd) @@ -2328,7 +2924,6 @@ static void *relay_thread_worker(void *data) struct lttng_poll_event events; struct lttng_ht *relay_connections_ht; struct lttng_ht_iter iter; - struct lttcomm_relayd_hdr recv_hdr; struct relay_connection *destroy_conn = NULL; DBG("[thread] Relay worker started"); @@ -2349,7 +2944,7 @@ static void *relay_thread_worker(void *data) goto relay_connections_ht_error; } - ret = create_thread_poll_set(&events, 2); + ret = create_named_thread_poll_set(&events, 2, "Worker thread epoll"); if (ret < 0) { goto error_poll_create; } @@ -2411,10 +3006,7 @@ restart: /* Inspect the relay conn pipe for new connection */ if (pollfd == relay_conn_pipe[0]) { - if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { - ERR("Relay connection pipe error"); - goto error; - } else if (revents & LPOLLIN) { + if (revents & LPOLLIN) { struct relay_connection *conn; ret = lttng_read(relay_conn_pipe[0], &conn, sizeof(conn)); @@ -2425,6 +3017,12 @@ restart: 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; } } else { struct relay_connection *ctrl_conn; @@ -2433,29 +3031,8 @@ restart: /* If not found, there is a synchronization issue. */ assert(ctrl_conn); - 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 if (revents & LPOLLIN) { - if (ctrl_conn->type == RELAY_CONTROL) { - 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 (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 @@ -2463,9 +3040,54 @@ restart: */ last_notdel_data_fd = pollfd; } + goto put_ctrl_connection; + } + assert(ctrl_conn->type == RELAY_CONTROL); + + if (revents & LPOLLIN) { + enum relay_connection_status status; + + status = relay_process_control(ctrl_conn); + if (status != RELAY_CONNECTION_STATUS_OK) { + /* + * On socket error flag the session as aborted to force + * the cleanup of its stream otherwise it can leak + * during the lifetime of the relayd. + * + * This prevents situations in which streams can be + * left opened because an index was received, the + * control connection is closed, and the data + * connection is closed (uncleanly) before the packet's + * data provided. + * + * Since the control connection encountered an error, + * it is okay to be conservative and close the + * session right now as we can't rely on the protocol + * being respected anymore. + */ + if (status == RELAY_CONNECTION_STATUS_ERROR) { + session_abort(ctrl_conn->session); + } + + /* Clear the connection on error or close. */ + 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("Unknown poll events %u for sock %d", revents, pollfd); + ERR("Unexpected poll events %u for control sock %d", + revents, pollfd); + connection_put(ctrl_conn); + goto error; } + put_ctrl_connection: connection_put(ctrl_conn); } } @@ -2515,17 +3137,38 @@ restart: /* Skip it. Might be removed before. */ continue; } + if (data_conn->type == RELAY_CONTROL) { + goto put_data_connection; + } + assert(data_conn->type == RELAY_DATA); if (revents & LPOLLIN) { - if (data_conn->type != RELAY_DATA) { - goto put_connection; - } + enum relay_connection_status status; - ret = relay_process_data(data_conn); - /* Connection closed */ - if (ret < 0) { + status = relay_process_data(data_conn); + /* Connection closed or error. */ + if (status != RELAY_CONNECTION_STATUS_OK) { + /* + * On socket error flag the session as aborted to force + * the cleanup of its stream otherwise it can leak + * during the lifetime of the relayd. + * + * This prevents situations in which streams can be + * left opened because an index was received, the + * control connection is closed, and the data + * connection is closed (uncleanly) before the packet's + * data provided. + * + * Since the data connection encountered an error, + * it is okay to be conservative and close the + * session right now as we can't rely on the protocol + * being respected anymore. + */ + if (status == RELAY_CONNECTION_STATUS_ERROR) { + session_abort(data_conn->session); + } relay_thread_close_connection(&events, pollfd, - data_conn); + data_conn); /* * Every goto restart call sets the last seen fd where * here we don't really care since we gracefully @@ -2537,8 +3180,14 @@ restart: connection_put(data_conn); goto restart; } + } 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_connection: + put_data_connection: connection_put(data_conn); } last_seen_data_fd = -1; @@ -2555,6 +3204,9 @@ error: destroy_conn, sock_n.node) { health_code_update(); + + session_abort(destroy_conn->session); + /* * No need to grab another ref, because we own * destroy_conn. @@ -2564,17 +3216,17 @@ error: } rcu_read_unlock(); - lttng_poll_clean(&events); + (void) fd_tracker_util_poll_clean(the_fd_tracker, &events); error_poll_create: lttng_ht_destroy(relay_connections_ht); relay_connections_ht_error: /* Close relay conn pipes */ - utils_close_pipe(relay_conn_pipe); + (void) fd_tracker_util_pipe_close(the_fd_tracker, + relay_conn_pipe); if (err) { DBG("Thread exited with error"); } DBG("Worker thread cleanup complete"); - free(data_buffer); error_testpoint: if (err) { health_error(); @@ -2592,11 +3244,8 @@ error_testpoint: */ static int create_relay_conn_pipe(void) { - int ret; - - ret = utils_create_pipe_cloexec(relay_conn_pipe); - - return ret; + return fd_tracker_util_pipe_open_cloexec(the_fd_tracker, + "Relayd connection pipe", relay_conn_pipe); } /* @@ -2607,7 +3256,13 @@ int main(int argc, char **argv) int ret = 0, retval = 0; void *status; - /* Parse arguments */ + /* Parse environment variables */ + parse_env_options(); + + /* + * Parse arguments. + * Command line arguments overwrite environment. + */ progname = argv[0]; if (set_options(argc, argv)) { retval = -1; @@ -2619,6 +3274,19 @@ int main(int argc, char **argv) goto exit_options; } + relayd_config_log(); + + if (opt_print_version) { + print_version(); + retval = 0; + goto exit_options; + } + + ret = fclose(stdin); + if (ret) { + PERROR("Failed to close stdin"); + goto exit_options; + } /* Try to create directory if -o, --output is specified. */ if (opt_output_path) { if (*opt_output_path != '/') { @@ -2638,25 +3306,34 @@ int main(int argc, char **argv) /* Daemonize */ if (opt_daemon || opt_background) { - int i; - ret = lttng_daemonize(&child_ppid, &recv_child_signal, !opt_background); if (ret < 0) { 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); + if (opt_working_directory) { + ret = utils_change_working_dir(opt_working_directory); + if (ret) { + ERR("Changing working directory"); + goto exit_options; } } + /* + * The RCU thread registration (and use, through the fd-tracker's + * creation) is done after the daemonization to allow us to not + * deal with liburcu's fork() management as the call RCU needs to + * be restored. + */ + rcu_register_thread(); + the_fd_tracker = fd_tracker_create(lttng_opt_fd_cap); + if (!the_fd_tracker) { + retval = -1; + goto exit_options; + } /* Initialize thread health monitoring */ health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES); @@ -2672,15 +3349,6 @@ int main(int argc, char **argv) goto exit_init_data; } - /* Check if daemon is UID = 0 */ - if (!getuid()) { - if (control_uri->port < 1024 || data_uri->port < 1024 || live_uri->port < 1024) { - ERR("Need to be root to use ports < 1024"); - retval = -1; - goto exit_init_data; - } - } - /* Setup the thread apps communication pipe. */ if (create_relay_conn_pipe()) { retval = -1; @@ -2690,12 +3358,6 @@ int main(int argc, char **argv) /* Init relay command queue. */ cds_wfcq_init(&relay_conn_queue.head, &relay_conn_queue.tail); - /* Set up max poll set size */ - if (lttng_poll_set_max_size()) { - retval = -1; - goto exit_init_data; - } - /* Initialize communication library */ lttcomm_init(); lttcomm_inet_init(); @@ -2721,14 +3383,14 @@ int main(int argc, char **argv) goto exit_init_data; } - ret = utils_create_pipe(health_quit_pipe); + ret = init_health_quit_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) { errno = ret; @@ -2738,7 +3400,7 @@ int main(int argc, char **argv) } /* 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) { errno = ret; @@ -2748,7 +3410,7 @@ int main(int argc, char **argv) } /* Setup the worker thread */ - ret = pthread_create(&worker_thread, NULL, + ret = pthread_create(&worker_thread, default_pthread_attr(), relay_thread_worker, NULL); if (ret) { errno = ret; @@ -2758,7 +3420,7 @@ int main(int argc, char **argv) } /* 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) { errno = ret; @@ -2817,18 +3479,27 @@ exit_dispatcher_thread: } exit_health_thread: - utils_close_pipe(health_quit_pipe); + (void) fd_tracker_util_pipe_close(the_fd_tracker, health_quit_pipe); exit_health_quit_pipe: exit_init_data: health_app_destroy(health_relayd); exit_health_app_create: 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(); /* Ensure all prior call_rcu are done. */ rcu_barrier(); + fd_tracker_destroy(the_fd_tracker); + rcu_unregister_thread(); + if (!retval) { exit(EXIT_SUCCESS); } else {