X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-relayd%2Fmain.c;h=07e7a012245d5fa0d36fd919f40d00c380070be1;hp=1313091b1df48bb84f7c7144b81be1636957188c;hb=794e2e5f064718252b249e17914c9b2b089d8d0c;hpb=f86f6389a47137d2f5e8fc707916b6650fd1de00 diff --git a/src/bin/lttng-relayd/main.c b/src/bin/lttng-relayd/main.c index 1313091b1..07e7a0122 100644 --- a/src/bin/lttng-relayd/main.c +++ b/src/bin/lttng-relayd/main.c @@ -34,11 +34,15 @@ #include #include #include +#include #include #include #include +#include #include #include +#include +#include #include #include @@ -58,22 +62,25 @@ #include #include #include -#include +#include +#include "backward-compatibility-group-by.h" #include "cmd.h" +#include "connection.h" #include "ctf-trace.h" +#include "health-relayd.h" #include "index.h" -#include "utils.h" -#include "lttng-relayd.h" #include "live.h" -#include "health-relayd.h" -#include "testpoint.h" -#include "viewer-stream.h" +#include "lttng-relayd.h" #include "session.h" +#include "sessiond-trace-chunks.h" #include "stream.h" -#include "connection.h" -#include "tracefile-array.h" #include "tcp_keep_alive.h" +#include "testpoint.h" +#include "tracefile-array.h" +#include "utils.h" +#include "version.h" +#include "viewer-stream.h" static const char *help_msg = #ifdef LTTNG_EMBED_HELP @@ -85,15 +92,16 @@ NULL enum relay_connection_status { RELAY_CONNECTION_STATUS_OK, - /* An error occured while processing an event on the connection. */ + /* An error occurred 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, opt_allow_clear = 1; +enum relay_group_output_by opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_UNKNOWN; /* * We need to wait for listener and live listener threads, as well as @@ -104,7 +112,6 @@ static int lttng_relay_ready = NR_LTTNG_RELAY_READY; /* Size of receive buffer. */ #define RECV_DATA_BUFFER_SIZE 65536 -#define FILE_COPY_BUFFER_SIZE 65536 static int recv_child_signal; /* Set to 1 when a SIGUSR1 signal is received. */ static pid_t child_ppid; /* Internal parent PID use with daemonize. */ @@ -155,6 +162,9 @@ static uint64_t last_relay_stream_id; */ static struct relay_conn_queue relay_conn_queue; +/* 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; @@ -167,6 +177,8 @@ struct lttng_ht *sessions_ht; /* Relayd health monitoring */ struct health_app *health_relayd; +struct sessiond_trace_chunk_registry *sessiond_trace_chunk_registry; + static struct option long_options[] = { { "control-port", 1, 0, 'C', }, { "data-port", 1, 0, 'D', }, @@ -174,16 +186,38 @@ 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', }, + { "disallow-clear", 0, 0, 'x' }, { NULL, 0, 0, 0, }, }; static const char *config_ignore_options[] = { "help", "config", "version" }; +static void print_version(void) { + fprintf(stdout, "%s\n", VERSION); +} + +static void relayd_config_log(void) +{ + 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"); + } +} + /* * Take an option from the getopt output and set it in the right variable to be * used later. @@ -196,9 +230,34 @@ 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': @@ -274,8 +333,8 @@ static int set_option(int opt, const char *arg, const char *optname) } exit(EXIT_FAILURE); case 'V': - fprintf(stdout, "%s\n", VERSION); - exit(EXIT_SUCCESS); + opt_print_version = 1; + break; case 'o': if (lttng_is_setuid_setgid()) { WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", @@ -289,6 +348,20 @@ 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 opt_working_directory"); + goto end; + } + } + break; + case 'v': /* Verbose level can increase using multiple -v */ if (arg) { @@ -300,6 +373,24 @@ static int set_option(int opt, const char *arg, const char *optname) } } break; + case 's': + if (opt_group_output_by != RELAYD_GROUP_OUTPUT_BY_UNKNOWN) { + ERR("Cannot set --group-output-by-session, another --group-output-by argument is present"); + exit(EXIT_FAILURE); + } + opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_SESSION; + break; + case 'p': + if (opt_group_output_by != RELAYD_GROUP_OUTPUT_BY_UNKNOWN) { + ERR("Cannot set --group-output-by-host, another --group-output-by argument is present"); + exit(EXIT_FAILURE); + } + opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_HOST; + break; + case 'x': + /* Disallow clear */ + opt_allow_clear = 0; + break; default: /* Unknown option or other error. * Error is printed by getopt, just return */ @@ -369,6 +460,23 @@ end: return ret; } +static int parse_env_options(void) +{ + int ret = 0; + char *value = NULL; + + value = lttng_secure_getenv(DEFAULT_LTTNG_RELAYD_WORKING_DIRECTORY_ENV); + if (value) { + opt_working_directory = strdup(value); + if (!opt_working_directory) { + ERR("Failed to allocate working directory string (\"%s\")", + value); + ret = -1; + } + } + return ret; +} + static int set_options(int argc, char **argv) { int c, ret = 0, option_index = 0, retval = 0; @@ -486,6 +594,35 @@ 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 == RELAYD_GROUP_OUTPUT_BY_UNKNOWN) { + opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_HOST; + } + if (opt_allow_clear) { + /* Check if env variable exists. */ + const char *value = lttng_secure_getenv(DEFAULT_LTTNG_RELAYD_DISALLOW_CLEAR_ENV); + if (value) { + ret = config_parse_value(value); + if (ret < 0) { + ERR("Invalid value for %s specified", DEFAULT_LTTNG_RELAYD_DISALLOW_CLEAR_ENV); + retval = -1; + goto exit; + } + opt_allow_clear = !ret; + } + } exit: free(optstring); @@ -519,12 +656,21 @@ static void relayd_cleanup(void) if (sessions_ht) lttng_ht_destroy(sessions_ht); - /* free the dynamically allocated opt_output_path */ free(opt_output_path); + free(opt_working_directory); + if (health_relayd) { + health_app_destroy(health_relayd); + } /* Close thread quit pipes */ + utils_close_pipe(health_quit_pipe); utils_close_pipe(thread_quit_pipe); + if (sessiond_trace_chunk_registry) { + sessiond_trace_chunk_registry_destroy( + sessiond_trace_chunk_registry); + } + uri_free(control_uri); uri_free(data_uri); /* Live URI is freed in the live thread. */ @@ -858,14 +1004,6 @@ restart: revents = LTTNG_POLL_GETEV(&events, i); pollfd = LTTNG_POLL_GETFD(&events, i); - if (!revents) { - /* - * No activity for this FD (poll - * implementation). - */ - continue; - } - /* Thread quit pipe has been closed. Killing thread. */ ret = check_thread_quit_pipe(pollfd, revents); if (ret) { @@ -1056,31 +1194,9 @@ error_testpoint: return NULL; } -/* - * 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 relay_connection *conn) +static bool session_streams_have_index(const struct relay_session *session) { - struct ctf_packet_index index_data; - - /* - * The index on disk is encoded in big endian. - */ - 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); + return session->minor >= 4 && !session->snapshot; } /* @@ -1094,17 +1210,22 @@ static int relay_create_session(const struct lttcomm_relayd_hdr *recv_hdr, { int ret = 0; ssize_t send_ret; - struct relay_session *session; - struct lttcomm_relayd_status_session reply; - char session_name[LTTNG_NAME_MAX]; - char hostname[LTTNG_HOST_NAME_MAX]; + struct relay_session *session = NULL; + struct lttcomm_relayd_create_session_reply_2_11 reply = {}; + char session_name[LTTNG_NAME_MAX] = {}; + char hostname[LTTNG_HOST_NAME_MAX] = {}; uint32_t live_timer = 0; bool snapshot = false; + bool session_name_contains_creation_timestamp = false; + /* Left nil for peers < 2.11. */ + char base_path[LTTNG_PATH_MAX] = {}; + lttng_uuid sessiond_uuid = {}; + LTTNG_OPTIONAL(uint64_t) id_sessiond = {}; + LTTNG_OPTIONAL(uint64_t) current_chunk_id = {}; + LTTNG_OPTIONAL(time_t) creation_time = {}; + struct lttng_dynamic_buffer reply_payload; - memset(session_name, 0, LTTNG_NAME_MAX); - memset(hostname, 0, LTTNG_HOST_NAME_MAX); - - memset(&reply, 0, sizeof(reply)); + lttng_dynamic_buffer_init(&reply_payload); if (conn->minor < 4) { /* From 2.1 to 2.3 */ @@ -1114,17 +1235,42 @@ static int relay_create_session(const struct lttcomm_relayd_hdr *recv_hdr, ret = cmd_create_session_2_4(payload, session_name, hostname, &live_timer, &snapshot); } else { + bool has_current_chunk; + uint64_t current_chunk_id_value; + time_t creation_time_value; + uint64_t id_sessiond_value; + /* From 2.11 to ... */ - ret = cmd_create_session_2_11(payload, session_name, - hostname, &live_timer, &snapshot); + ret = cmd_create_session_2_11(payload, session_name, hostname, + base_path, &live_timer, &snapshot, &id_sessiond_value, + sessiond_uuid, &has_current_chunk, + ¤t_chunk_id_value, &creation_time_value, + &session_name_contains_creation_timestamp); + if (lttng_uuid_is_nil(sessiond_uuid)) { + /* The nil UUID is reserved for pre-2.11 clients. */ + ERR("Illegal nil UUID announced by peer in create session command"); + ret = -1; + goto send_reply; + } + LTTNG_OPTIONAL_SET(&id_sessiond, id_sessiond_value); + LTTNG_OPTIONAL_SET(&creation_time, creation_time_value); + if (has_current_chunk) { + LTTNG_OPTIONAL_SET(¤t_chunk_id, + current_chunk_id_value); + } } if (ret < 0) { goto send_reply; } - session = session_create(session_name, hostname, live_timer, - snapshot, conn->major, conn->minor); + session = session_create(session_name, hostname, base_path, live_timer, + snapshot, sessiond_uuid, + id_sessiond.is_set ? &id_sessiond.value : NULL, + current_chunk_id.is_set ? ¤t_chunk_id.value : NULL, + creation_time.is_set ? &creation_time.value : NULL, + conn->major, conn->minor, + session_name_contains_creation_timestamp); if (!session) { ret = -1; goto send_reply; @@ -1133,22 +1279,59 @@ static int relay_create_session(const struct lttcomm_relayd_hdr *recv_hdr, conn->session = session; DBG("Created session %" PRIu64, session->id); - reply.session_id = htobe64(session->id); + reply.generic.session_id = htobe64(session->id); send_reply: if (ret < 0) { - reply.ret_code = htobe32(LTTNG_ERR_FATAL); + reply.generic.ret_code = htobe32(LTTNG_ERR_FATAL); } else { - reply.ret_code = htobe32(LTTNG_OK); + reply.generic.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 \"create session\" command reply (ret = %zd)", - send_ret); - ret = -1; + if (conn->minor < 11) { + /* From 2.1 to 2.10 */ + ret = lttng_dynamic_buffer_append(&reply_payload, + &reply.generic, sizeof(reply.generic)); + if (ret) { + ERR("Failed to append \"create session\" command reply header to payload buffer"); + ret = -1; + goto end; + } + } else { + const uint32_t output_path_length = + session ? strlen(session->output_path) + 1 : 0; + + reply.output_path_length = htobe32(output_path_length); + ret = lttng_dynamic_buffer_append( + &reply_payload, &reply, sizeof(reply)); + if (ret) { + ERR("Failed to append \"create session\" command reply header to payload buffer"); + goto end; + } + + if (output_path_length) { + ret = lttng_dynamic_buffer_append(&reply_payload, + session->output_path, + output_path_length); + if (ret) { + ERR("Failed to append \"create session\" command reply path to payload buffer"); + goto end; + } + } } + send_ret = conn->sock->ops->sendmsg(conn->sock, reply_payload.data, + reply_payload.size, 0); + if (send_ret < (ssize_t) reply_payload.size) { + ERR("Failed to send \"create session\" command reply of %zu bytes (ret = %zd)", + reply_payload.size, send_ret); + ret = -1; + } +end: + if (ret < 0 && session) { + session_put(session); + } + lttng_dynamic_buffer_reset(&reply_payload); return ret; } @@ -1182,6 +1365,34 @@ static void publish_connection_local_streams(struct relay_connection *conn) pthread_mutex_unlock(&session->lock); } +static int conform_channel_path(char *channel_path) +{ + int ret = 0; + + if (strstr("../", channel_path)) { + ERR("Refusing channel path as it walks up the path hierarchy: \"%s\"", + channel_path); + ret = -1; + goto end; + } + + if (*channel_path == '/') { + const size_t len = strlen(channel_path); + + /* + * Channel paths from peers prior to 2.11 are expressed as an + * absolute path that is, in reality, relative to the relay + * daemon's output directory. Remove the leading slash so it + * is correctly interpreted as a relative path later on. + * + * len (and not len - 1) is used to copy the trailing NULL. + */ + bcopy(channel_path + 1, channel_path, len); + } +end: + return ret; +} + /* * relay_add_stream: allocate a new stream for a session */ @@ -1198,6 +1409,7 @@ static int relay_add_stream(const struct lttcomm_relayd_hdr *recv_hdr, uint64_t stream_handle = -1ULL; char *path_name = NULL, *channel_name = NULL; uint64_t tracefile_size = 0, tracefile_count = 0; + LTTNG_OPTIONAL(uint64_t) stream_chunk_id = {}; if (!session || !conn->version_check_done) { ERR("Trying to add a stream before version check"); @@ -1205,34 +1417,88 @@ static int relay_add_stream(const struct lttcomm_relayd_hdr *recv_hdr, goto end_no_session; } - switch (session->minor) { - case 1: /* LTTng sessiond 2.1. Allocates path_name and channel_name. */ + if (session->minor == 1) { + /* For 2.1 */ ret = cmd_recv_stream_2_1(payload, &path_name, &channel_name); - break; - case 2: /* LTTng sessiond 2.2. Allocates path_name and channel_name. */ - default: + } else if (session->minor > 1 && session->minor < 11) { + /* From 2.2 to 2.10 */ ret = cmd_recv_stream_2_2(payload, &path_name, &channel_name, &tracefile_size, &tracefile_count); - break; + } else { + /* From 2.11 to ... */ + ret = cmd_recv_stream_2_11(payload, &path_name, + &channel_name, &tracefile_size, &tracefile_count, + &stream_chunk_id.value); + stream_chunk_id.is_set = true; } + if (ret < 0) { goto send_reply; } + if (conform_channel_path(path_name)) { + goto send_reply; + } + + /* + * Backward compatibility for --group-output-by-session. + * Prior to lttng 2.11, the complete path is passed by the stream. + * Starting at 2.11, lttng-relayd uses chunk. When dealing with producer + * >=2.11 the chunk is responsible for the output path. When dealing + * with producer < 2.11 the chunk output_path is the root output path + * and the stream carries the complete path (path_name). + * To support --group-output-by-session with older producer (<2.11), we + * need to craft the path based on the stream path. + */ + if (opt_group_output_by == RELAYD_GROUP_OUTPUT_BY_SESSION) { + if (conn->minor < 4) { + /* + * From 2.1 to 2.3, the session_name is not passed on + * the RELAYD_CREATE_SESSION command. The session name + * is necessary to detect the presence of a base_path + * inside the stream path. Without it we cannot perform + * a valid group-output-by-session transformation. + */ + WARN("Unable to perform a --group-by-session transformation for session %" PRIu64 + " for stream with path \"%s\" as it is produced by a peer using a protocol older than v2.4", + session->id, path_name); + } else if (conn->minor >= 4 && conn->minor < 11) { + char *group_by_session_path_name; + + assert(session->session_name[0] != '\0'); + + group_by_session_path_name = + backward_compat_group_by_session( + path_name, + session->session_name); + if (!group_by_session_path_name) { + ERR("Failed to apply group by session to stream of session %" PRIu64, + session->id); + goto send_reply; + } + + DBG("Transformed session path from \"%s\" to \"%s\" to honor per-session name grouping", + path_name, group_by_session_path_name); + + free(path_name); + path_name = group_by_session_path_name; + } + } + trace = ctf_trace_get_by_path_or_create(session, path_name); if (!trace) { goto send_reply; } - /* This stream here has one reference on the trace. */ + /* This stream here has one reference on the trace. */ pthread_mutex_lock(&last_relay_stream_id_lock); stream_handle = ++last_relay_stream_id; pthread_mutex_unlock(&last_relay_stream_id_lock); /* We pass ownership of path_name and channel_name. */ stream = stream_create(trace, stream_handle, path_name, - channel_name, tracefile_size, tracefile_count); + channel_name, tracefile_size, tracefile_count); path_name = NULL; channel_name = NULL; @@ -1325,7 +1591,7 @@ static int relay_close_stream(const struct lttcomm_relayd_hdr *recv_hdr, vstream = viewer_stream_get_by_id(stream->stream_handle); if (vstream) { - if (vstream->metadata_sent == stream->metadata_received) { + if (stream->no_new_metadata_notified) { /* * Since all the metadata has been sent to the * viewer and that we have a request to close @@ -1412,15 +1678,14 @@ int relay_reset_metadata(const struct lttcomm_relayd_hdr *recv_hdr, 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); + ret = stream_reset_file(stream); if (ret < 0) { - ERR("Failed to rotate metadata file %s of channel %s", - stream->path_name, stream->channel_name); + ERR("Failed to reset metadata stream %" PRIu64 + ": stream_path = %s, channel = %s", + stream->stream_handle, stream->path_name, + stream->channel_name); goto end_unlock; } - end_unlock: pthread_mutex_unlock(&stream->lock); stream_put(stream); @@ -1492,391 +1757,104 @@ static int relay_start(const struct lttcomm_relayd_hdr *recv_hdr, } /* - * Append padding to the file pointed by the file descriptor fd. + * relay_recv_metadata: receive the metadata for the session. */ -static int write_padding_to_file(int fd, uint32_t size) +static int relay_recv_metadata(const struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn, + const struct lttng_buffer_view *payload) { - ssize_t ret = 0; - char *zeros; + int ret = 0; + struct relay_session *session = conn->session; + struct lttcomm_relayd_metadata_payload metadata_payload_header; + struct relay_stream *metadata_stream; + uint64_t metadata_payload_size; + struct lttng_buffer_view packet_view; - if (size == 0) { + if (!session) { + ERR("Metadata sent before version check"); + ret = -1; goto end; } - zeros = zmalloc(size); - if (zeros == NULL) { - PERROR("zmalloc zeros for padding"); + if (recv_hdr->data_size < sizeof(struct lttcomm_relayd_metadata_payload)) { + ERR("Incorrect data size"); ret = -1; goto end; } + metadata_payload_size = recv_hdr->data_size - + sizeof(struct lttcomm_relayd_metadata_payload); - ret = lttng_write(fd, zeros, size); - if (ret < size) { - PERROR("write padding to file"); - } - - free(zeros); - -end: - return ret; -} + 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); -/* - * Close the current index file if it is open, and create a new one. - * - * Return 0 on success, -1 on error. - */ -static -int create_rotate_index_file(struct relay_stream *stream) -{ - int ret; - uint32_t major, minor; - - /* Put ref on previous index_file. */ - if (stream->index_file) { - lttng_index_file_put(stream->index_file); - stream->index_file = NULL; - } - major = stream->trace->session->major; - minor = stream->trace->session->minor; - stream->index_file = lttng_index_file_create(stream->path_name, - stream->channel_name, - -1, -1, stream->tracefile_size, - tracefile_array_get_file_index_head(stream->tfa), - lttng_to_index_major(major, minor), - lttng_to_index_minor(major, minor)); - if (!stream->index_file) { + metadata_stream = stream_get_by_id(metadata_payload_header.stream_id); + if (!metadata_stream) { ret = -1; goto end; } - ret = 0; - -end: - return ret; -} - -static -int do_rotate_stream(struct relay_stream *stream) -{ - int ret; - - /* Perform the stream rotation. */ - ret = utils_rotate_stream_file(stream->path_name, - stream->channel_name, stream->tracefile_size, - stream->tracefile_count, -1, - -1, stream->stream_fd->fd, - NULL, &stream->stream_fd->fd); - if (ret < 0) { - ERR("Rotating stream output file"); - goto end; + packet_view = lttng_buffer_view_from_view(payload, + sizeof(metadata_payload_header), metadata_payload_size); + if (!packet_view.data) { + ERR("Invalid metadata packet length announced by header"); + ret = -1; + goto end_put; } - stream->tracefile_size_current = 0; - /* Rotate also the index if the stream is not a metadata stream. */ - if (!stream->is_metadata) { - ret = create_rotate_index_file(stream); - if (ret < 0) { - ERR("Failed to rotate index file"); - goto end; - } + pthread_mutex_lock(&metadata_stream->lock); + ret = stream_write(metadata_stream, &packet_view, + metadata_payload_header.padding_size); + pthread_mutex_unlock(&metadata_stream->lock); + if (ret){ + ret = -1; + goto end_put; } - - stream->rotate_at_seq_num = -1ULL; - stream->pos_after_last_complete_data_index = 0; - +end_put: + stream_put(metadata_stream); end: return ret; } /* - * If too much data has been written in a tracefile before we received the - * rotation command, we have to move the excess data to the new tracefile and - * perform the rotation. This can happen because the control and data - * connections are separate, the indexes as well as the commands arrive from - * the control connection and we have no control over the order so we could be - * in a situation where too much data has been received on the data connection - * before the rotation command on the control connection arrives. We don't need - * to update the index because its order is guaranteed with the rotation - * command message. + * relay_send_version: send relayd version number */ -static -int rotate_truncate_stream(struct relay_stream *stream) +static int relay_send_version(const struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn, + const struct lttng_buffer_view *payload) { - int ret, new_fd; - off_t lseek_ret; - uint64_t diff, pos = 0; - char buf[FILE_COPY_BUFFER_SIZE]; - - assert(!stream->is_metadata); - - assert(stream->tracefile_size_current > - stream->pos_after_last_complete_data_index); - diff = stream->tracefile_size_current - - stream->pos_after_last_complete_data_index; - - /* Create the new tracefile. */ - new_fd = utils_create_stream_file(stream->path_name, - stream->channel_name, - stream->tracefile_size, stream->tracefile_count, - /* uid */ -1, /* gid */ -1, /* suffix */ NULL); - if (new_fd < 0) { - ERR("Failed to create new stream file at path %s for channel %s", - stream->path_name, stream->channel_name); - ret = -1; - goto end; - } + int ret; + ssize_t send_ret; + struct lttcomm_relayd_version reply, msg; + bool compatible = true; - /* - * Rewind the current tracefile to the position at which the rotation - * should have occured. - */ - lseek_ret = lseek(stream->stream_fd->fd, - stream->pos_after_last_complete_data_index, SEEK_SET); - if (lseek_ret < 0) { - PERROR("seek truncate stream"); + conn->version_check_done = true; + + /* Get version from the other side. */ + 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; } - /* Move data from the old file to the new file. */ - while (pos < diff) { - uint64_t count, bytes_left; - ssize_t io_ret; + memcpy(&msg, payload->data, sizeof(msg)); + msg.major = be32toh(msg.major); + msg.minor = be32toh(msg.minor); - bytes_left = diff - pos; - count = bytes_left > sizeof(buf) ? sizeof(buf) : bytes_left; - assert(count <= SIZE_MAX); + memset(&reply, 0, sizeof(reply)); + reply.major = RELAYD_VERSION_COMM_MAJOR; + reply.minor = RELAYD_VERSION_COMM_MINOR; - io_ret = lttng_read(stream->stream_fd->fd, buf, count); - if (io_ret < (ssize_t) count) { - char error_string[256]; - - snprintf(error_string, sizeof(error_string), - "Failed to read %" PRIu64 " bytes from fd %i in rotate_truncate_stream(), returned %zi", - count, stream->stream_fd->fd, io_ret); - if (io_ret == -1) { - PERROR("%s", error_string); - } else { - ERR("%s", error_string); - } - ret = -1; - goto end; - } - - io_ret = lttng_write(new_fd, buf, count); - if (io_ret < (ssize_t) count) { - char error_string[256]; - - snprintf(error_string, sizeof(error_string), - "Failed to write %" PRIu64 " bytes from fd %i in rotate_truncate_stream(), returned %zi", - count, new_fd, io_ret); - if (io_ret == -1) { - PERROR("%s", error_string); - } else { - ERR("%s", error_string); - } - ret = -1; - goto end; - } - - pos += count; - } - - /* Truncate the file to get rid of the excess data. */ - ret = ftruncate(stream->stream_fd->fd, - stream->pos_after_last_complete_data_index); - if (ret) { - PERROR("ftruncate"); - goto end; - } - - ret = close(stream->stream_fd->fd); - if (ret < 0) { - PERROR("Closing tracefile"); - goto end; - } - - ret = create_rotate_index_file(stream); - if (ret < 0) { - ERR("Rotate stream index file"); - goto end; - } - - /* - * Update the offset and FD of all the eventual indexes created by the - * data connection before the rotation command arrived. - */ - ret = relay_index_switch_all_files(stream); - if (ret < 0) { - ERR("Failed to rotate index file"); - goto end; - } - - stream->stream_fd->fd = new_fd; - stream->tracefile_size_current = diff; - stream->pos_after_last_complete_data_index = 0; - stream->rotate_at_seq_num = -1ULL; - - ret = 0; - -end: - return ret; -} - -/* - * Check if a stream should perform a rotation (for session rotation). - * Must be called with the stream lock held. - * - * Return 0 on success, a negative value on error. - */ -static -int try_rotate_stream(struct relay_stream *stream) -{ - int ret = 0; - - /* No rotation expected. */ - if (stream->rotate_at_seq_num == -1ULL) { - goto end; - } - - if (stream->prev_seq < stream->rotate_at_seq_num || - stream->prev_seq == -1ULL) { - DBG("Stream %" PRIu64 " no yet ready for rotation", - stream->stream_handle); - goto end; - } else if (stream->prev_seq > stream->rotate_at_seq_num) { - DBG("Rotation after too much data has been written in tracefile " - "for stream %" PRIu64 ", need to truncate before " - "rotating", stream->stream_handle); - ret = rotate_truncate_stream(stream); - if (ret) { - ERR("Failed to truncate stream"); - goto end; - } - } else { - /* stream->prev_seq == stream->rotate_at_seq_num */ - DBG("Stream %" PRIu64 " ready for rotation", - stream->stream_handle); - ret = do_rotate_stream(stream); - } - -end: - return ret; -} - -/* - * relay_recv_metadata: receive the metadata for the session. - */ -static int relay_recv_metadata(const struct lttcomm_relayd_hdr *recv_hdr, - struct relay_connection *conn, - const struct lttng_buffer_view *payload) -{ - int ret = 0; - ssize_t size_ret; - struct relay_session *session = conn->session; - struct lttcomm_relayd_metadata_payload metadata_payload_header; - struct relay_stream *metadata_stream; - uint64_t metadata_payload_size; - - if (!session) { - ERR("Metadata sent before version check"); - ret = -1; - goto end; - } - - if (recv_hdr->data_size < sizeof(struct lttcomm_relayd_metadata_payload)) { - ERR("Incorrect data size"); - ret = -1; - goto end; - } - metadata_payload_size = recv_hdr->data_size - - sizeof(struct lttcomm_relayd_metadata_payload); - - 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(metadata_payload_header.stream_id); - if (!metadata_stream) { - ret = -1; - goto end; - } - - pthread_mutex_lock(&metadata_stream->lock); - - 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; - } - - 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 += - metadata_payload_size + metadata_payload_header.padding_size; - DBG2("Relay metadata written. Updated metadata_received %" PRIu64, - metadata_stream->metadata_received); - - ret = try_rotate_stream(metadata_stream); - if (ret < 0) { - goto end_put; - } - -end_put: - pthread_mutex_unlock(&metadata_stream->lock); - stream_put(metadata_stream); -end: - return ret; -} - -/* - * relay_send_version: send relayd version number - */ -static int relay_send_version(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 = true; - - /* Get version from the other side. */ - 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 != msg.major) { - DBG("Incompatible major versions (%u vs %u), deleting session", - reply.major, msg.major); - compatible = false; - } + /* Major versions must be the same */ + if (reply.major != msg.major) { + DBG("Incompatible major versions (%u vs %u), deleting session", + reply.major, msg.major); + compatible = false; + } conn->major = reply.major; /* We adapt to the lowest compatible version */ @@ -1924,6 +1902,7 @@ static int relay_data_pending(const struct lttcomm_relayd_hdr *recv_hdr, struct relay_stream *stream; ssize_t send_ret; int ret; + uint64_t stream_seq; DBG("Data pending command received"); @@ -1951,12 +1930,23 @@ static int relay_data_pending(const 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, msg.stream_id, - stream->prev_seq, msg.last_net_seq_num); + if (session_streams_have_index(session)) { + /* + * Ensure that both the index and stream data have been + * flushed up to the requested point. + */ + stream_seq = min(stream->prev_data_seq, stream->prev_index_seq); + } else { + stream_seq = stream->prev_data_seq; + } + DBG("Data pending for stream id %" PRIu64 ": prev_data_seq %" PRIu64 + ", prev_index_seq %" PRIu64 + ", and last_seq %" PRIu64, msg.stream_id, + stream->prev_data_seq, stream->prev_index_seq, + msg.last_net_seq_num); /* Avoid wrapping issue */ - if (((int64_t) (stream->prev_seq - msg.last_net_seq_num)) >= 0) { + if (((int64_t) (stream_seq - msg.last_net_seq_num)) >= 0) { /* Data has in fact been written and is NOT pending */ ret = 0; } else { @@ -2176,7 +2166,18 @@ static int relay_end_data_pending(const struct lttcomm_relayd_hdr *recv_hdr, } pthread_mutex_lock(&stream->lock); if (!stream->data_pending_check_done) { - if (!stream->closed || !(((int64_t) (stream->prev_seq - stream->last_net_seq_num)) >= 0)) { + uint64_t stream_seq; + + if (session_streams_have_index(conn->session)) { + /* + * Ensure that both the index and stream data have been + * flushed up to the requested point. + */ + stream_seq = min(stream->prev_data_seq, stream->prev_index_seq); + } else { + stream_seq = stream->prev_data_seq; + } + if (!stream->closed || !(((int64_t) (stream_seq - stream->last_net_seq_num)) >= 0)) { is_data_inflight = 1; DBG("Data is still in flight for stream %" PRIu64, stream->stream_handle); @@ -2220,7 +2221,6 @@ static int relay_recv_index(const struct lttcomm_relayd_hdr *recv_hdr, 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; size_t msg_len; @@ -2258,6 +2258,9 @@ static int relay_recv_index(const struct lttcomm_relayd_hdr *recv_hdr, index_info.stream_instance_id = be64toh(index_info.stream_instance_id); index_info.packet_seq_num = be64toh(index_info.packet_seq_num); + } else { + index_info.stream_instance_id = -1ULL; + index_info.packet_seq_num = -1ULL; } stream = stream_get_by_id(index_info.relay_stream_id); @@ -2266,62 +2269,17 @@ static int relay_recv_index(const struct lttcomm_relayd_hdr *recv_hdr, ret = -1; goto end; } - pthread_mutex_lock(&stream->lock); - - /* Live beacon handling */ - if (index_info.packet_size == 0) { - DBG("Received live beacon for stream %" PRIu64, - stream->stream_handle); - - /* - * Only flag a stream inactive when it has already - * received data and no indexes are in flight. - */ - if (stream->index_received_seqcount > 0 - && stream->indexes_in_flight == 0) { - stream->beacon_ts_end = index_info.timestamp_end; - } - ret = 0; - goto end_stream_put; - } else { - stream->beacon_ts_end = -1ULL; - } - if (stream->ctf_stream_id == -1ULL) { - stream->ctf_stream_id = index_info.stream_id; - } - 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, conn)) { - ERR("set_index_control_data error"); - relay_index_put(index); - ret = -1; + pthread_mutex_lock(&stream->lock); + ret = stream_add_index(stream, &index_info); + pthread_mutex_unlock(&stream->lock); + if (ret) { goto end_stream_put; } - ret = relay_index_try_flush(index); - if (ret == 0) { - tracefile_array_commit_seq(stream->tfa); - stream->index_received_seqcount++; - stream->pos_after_last_complete_data_index += index->total_size; - } else if (ret > 0) { - /* no flush. */ - ret = 0; - } else { - ERR("relay_index_try_flush error %d", ret); - relay_index_put(index); - ret = -1; - } end_stream_put: - pthread_mutex_unlock(&stream->lock); stream_put(stream); - end: - memset(&reply, 0, sizeof(reply)); if (ret < 0) { reply.ret_code = htobe32(LTTNG_ERR_UNK); @@ -2384,24 +2342,27 @@ end_no_session: } /* - * relay_rotate_session_stream: rotate a stream to a new tracefile for the session - * rotation feature (not the tracefile rotation feature). + * relay_rotate_session_stream: rotate a stream to a new tracefile for the + * session rotation feature (not the tracefile rotation feature). */ -static int relay_rotate_session_stream(const struct lttcomm_relayd_hdr *recv_hdr, +static int relay_rotate_session_streams( + const struct lttcomm_relayd_hdr *recv_hdr, struct relay_connection *conn, const struct lttng_buffer_view *payload) { - int ret; + int ret = 0; + uint32_t i; ssize_t send_ret; + enum lttng_error_code reply_code = LTTNG_ERR_UNK; struct relay_session *session = conn->session; - struct lttcomm_relayd_rotate_stream stream_info; - struct lttcomm_relayd_generic_reply reply; - struct relay_stream *stream; - size_t header_len; - size_t path_len; - struct lttng_buffer_view new_path_view; - - DBG("Rotate stream received"); + struct lttcomm_relayd_rotate_streams rotate_streams; + struct lttcomm_relayd_generic_reply reply = {}; + struct relay_stream *stream = NULL; + const size_t header_len = sizeof(struct lttcomm_relayd_rotate_streams); + struct lttng_trace_chunk *next_trace_chunk = NULL; + struct lttng_buffer_view stream_positions; + char chunk_id_buf[MAX_INT_DEC_LEN(uint64_t)]; + const char *chunk_id_str = "none"; if (!session || !conn->version_check_done) { ERR("Trying to rotate a stream before version check"); @@ -2415,8 +2376,6 @@ static int relay_rotate_session_stream(const struct lttcomm_relayd_hdr *recv_hdr goto end_no_reply; } - header_len = sizeof(struct lttcomm_relayd_rotate_stream); - if (payload->size < header_len) { ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected >= %zu bytes, got %zu bytes", header_len, payload->size); @@ -2424,87 +2383,101 @@ static int relay_rotate_session_stream(const struct lttcomm_relayd_hdr *recv_hdr goto end_no_reply; } - memcpy(&stream_info, payload->data, header_len); + memcpy(&rotate_streams, payload->data, header_len); - /* Convert to host */ - stream_info.pathname_length = be32toh(stream_info.pathname_length); - stream_info.stream_id = be64toh(stream_info.stream_id); - stream_info.new_chunk_id = be64toh(stream_info.new_chunk_id); - stream_info.rotate_at_seq_num = be64toh(stream_info.rotate_at_seq_num); + /* Convert header to host endianness. */ + rotate_streams = (typeof(rotate_streams)) { + .stream_count = be32toh(rotate_streams.stream_count), + .new_chunk_id = (typeof(rotate_streams.new_chunk_id)) { + .is_set = !!rotate_streams.new_chunk_id.is_set, + .value = be64toh(rotate_streams.new_chunk_id.value), + } + }; - path_len = stream_info.pathname_length; - if (payload->size < header_len + path_len) { - ERR("Unexpected payload size in \"relay_rotate_session_stream\" including path: expected >= %zu bytes, got %zu bytes", - header_len + path_len, payload->size); - ret = -1; - goto end_no_reply; - } - - /* Ensure it fits in local filename length. */ - if (path_len >= LTTNG_PATH_MAX) { - ret = -ENAMETOOLONG; - ERR("Length of relay_rotate_session_stream command's path name (%zu bytes) exceeds the maximal allowed length of %i bytes", - path_len, LTTNG_PATH_MAX); - goto end; + if (rotate_streams.new_chunk_id.is_set) { + /* + * Retrieve the trace chunk the stream must transition to. As + * per the protocol, this chunk should have been created + * before this command is received. + */ + next_trace_chunk = sessiond_trace_chunk_registry_get_chunk( + sessiond_trace_chunk_registry, + session->sessiond_uuid, session->id, + rotate_streams.new_chunk_id.value); + if (!next_trace_chunk) { + char uuid_str[LTTNG_UUID_STR_LEN]; + + lttng_uuid_to_str(session->sessiond_uuid, uuid_str); + ERR("Unknown next trace chunk in ROTATE_STREAMS command: sessiond_uuid = {%s}, session_id = %" PRIu64 + ", trace_chunk_id = %" PRIu64, + uuid_str, session->id, + rotate_streams.new_chunk_id.value); + reply_code = LTTNG_ERR_INVALID_PROTOCOL; + ret = -1; + goto end; + } + + ret = snprintf(chunk_id_buf, sizeof(chunk_id_buf), "%" PRIu64, + rotate_streams.new_chunk_id.value); + if (ret < 0 || ret >= sizeof(chunk_id_buf)) { + chunk_id_str = "formatting error"; + } else { + chunk_id_str = chunk_id_buf; + } } - new_path_view = lttng_buffer_view_from_view(payload, header_len, - stream_info.pathname_length); + DBG("Rotate %" PRIu32 " streams of session \"%s\" to chunk \"%s\"", + rotate_streams.stream_count, session->session_name, + chunk_id_str); - stream = stream_get_by_id(stream_info.stream_id); - if (!stream) { + stream_positions = lttng_buffer_view_from_view(payload, + sizeof(rotate_streams), -1); + if (!stream_positions.data || + stream_positions.size < + (rotate_streams.stream_count * + sizeof(struct lttcomm_relayd_stream_rotation_position))) { + reply_code = LTTNG_ERR_INVALID_PROTOCOL; ret = -1; goto end; } - pthread_mutex_lock(&stream->lock); + for (i = 0; i < rotate_streams.stream_count; i++) { + struct lttcomm_relayd_stream_rotation_position *position_comm = + &((typeof(position_comm)) stream_positions.data)[i]; + const struct lttcomm_relayd_stream_rotation_position pos = { + .stream_id = be64toh(position_comm->stream_id), + .rotate_at_seq_num = be64toh( + position_comm->rotate_at_seq_num), + }; - /* - * Update the trace path (just the folder, the stream name does not - * change). - */ - free(stream->path_name); - stream->path_name = create_output_path(new_path_view.data); - if (!stream->path_name) { - ERR("Failed to create a new output path"); - ret = -1; - goto end_stream_unlock; - } - ret = utils_mkdir_recursive(stream->path_name, S_IRWXU | S_IRWXG, - -1, -1); - if (ret < 0) { - ERR("relay creating output directory"); - ret = -1; - goto end_stream_unlock; - } + stream = stream_get_by_id(pos.stream_id); + if (!stream) { + reply_code = LTTNG_ERR_INVALID; + ret = -1; + goto end; + } - stream->chunk_id = stream_info.new_chunk_id; + pthread_mutex_lock(&stream->lock); + ret = stream_set_pending_rotation(stream, next_trace_chunk, + pos.rotate_at_seq_num); + pthread_mutex_unlock(&stream->lock); + if (ret) { + reply_code = LTTNG_ERR_FILE_CREATION_ERROR; + goto end; + } - if (stream->is_metadata) { - /* - * The metadata stream is sent only over the control connection - * so we know we have all the data to perform the stream - * rotation. - */ - ret = do_rotate_stream(stream); - } else { - stream->rotate_at_seq_num = stream_info.rotate_at_seq_num; - ret = try_rotate_stream(stream); - } - if (ret < 0) { - goto end_stream_unlock; + stream_put(stream); + stream = NULL; } -end_stream_unlock: - pthread_mutex_unlock(&stream->lock); - stream_put(stream); + reply_code = LTTNG_OK; + 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); + if (stream) { + stream_put(stream); } + + reply.ret_code = htobe32((uint32_t) reply_code); send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(struct lttcomm_relayd_generic_reply), 0); if (send_ret < (ssize_t) sizeof(reply)) { @@ -2512,360 +2485,566 @@ end: send_ret); ret = -1; } - end_no_reply: + lttng_trace_chunk_put(next_trace_chunk); return ret; } + + /* - * relay_mkdir: Create a folder on the disk. + * relay_create_trace_chunk: create a new trace chunk */ -static int relay_mkdir(const struct lttcomm_relayd_hdr *recv_hdr, +static int relay_create_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr, struct relay_connection *conn, const struct lttng_buffer_view *payload) { - int ret; - struct relay_session *session = conn->session; - struct lttcomm_relayd_mkdir path_info_header; - struct lttcomm_relayd_generic_reply reply; - char *path = NULL; - size_t header_len; + int ret = 0; ssize_t send_ret; - struct lttng_buffer_view path_view; + struct relay_session *session = conn->session; + struct lttcomm_relayd_create_trace_chunk *msg; + struct lttcomm_relayd_generic_reply reply = {}; + struct lttng_buffer_view header_view; + struct lttng_buffer_view chunk_name_view; + struct lttng_trace_chunk *chunk = NULL, *published_chunk = NULL; + enum lttng_error_code reply_code = LTTNG_OK; + enum lttng_trace_chunk_status chunk_status; + struct lttng_directory_handle *session_output = NULL; + const char *new_path; if (!session || !conn->version_check_done) { - ERR("Trying to create a directory before version check"); + ERR("Trying to create a trace chunk before version check"); ret = -1; - goto end_no_session; + goto end_no_reply; } if (session->major == 2 && session->minor < 11) { - /* - * This client is not supposed to use this command since - * it predates its introduction. - */ - ERR("relay_mkdir command is unsupported before LTTng 2.11"); + ERR("Chunk creation command is unsupported before 2.11"); ret = -1; - goto end_no_session; + goto end_no_reply; } - header_len = sizeof(path_info_header); - if (payload->size < header_len) { - ERR("Unexpected payload size in \"relay_mkdir\": expected >= %zu bytes, got %zu bytes", - header_len, payload->size); + header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg)); + if (!header_view.data) { + ERR("Failed to receive payload of chunk creation command"); ret = -1; - goto end_no_session; + goto end_no_reply; } - memcpy(&path_info_header, payload->data, header_len); + /* Convert to host endianness. */ + msg = (typeof(msg)) header_view.data; + msg->chunk_id = be64toh(msg->chunk_id); + msg->creation_timestamp = be64toh(msg->creation_timestamp); + msg->override_name_length = be32toh(msg->override_name_length); - path_info_header.length = be32toh(path_info_header.length); - - if (payload->size < header_len + path_info_header.length) { - ERR("Unexpected payload size in \"relay_mkdir\" including path: expected >= %zu bytes, got %zu bytes", - header_len + path_info_header.length, payload->size); - ret = -1; - goto end_no_session; + if (session->current_trace_chunk && + !lttng_trace_chunk_get_name_overridden(session->current_trace_chunk)) { + chunk_status = lttng_trace_chunk_rename_path(session->current_trace_chunk, + DEFAULT_CHUNK_TMP_OLD_DIRECTORY); + if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { + ERR("Failed to rename old chunk"); + ret = -1; + reply_code = LTTNG_ERR_UNK; + goto end; + } } - - /* Ensure that it fits in local path length. */ - if (path_info_header.length >= LTTNG_PATH_MAX) { - ret = -ENAMETOOLONG; - ERR("Path name argument of mkdir command (%" PRIu32 " bytes) exceeds the maximal length allowed (%d bytes)", - path_info_header.length, LTTNG_PATH_MAX); + session->ongoing_rotation = true; + if (!session->current_trace_chunk) { + if (!session->has_rotated) { + new_path = ""; + } else { + new_path = NULL; + } + } else { + new_path = DEFAULT_CHUNK_TMP_NEW_DIRECTORY; + } + chunk = lttng_trace_chunk_create( + msg->chunk_id, msg->creation_timestamp, new_path); + if (!chunk) { + ERR("Failed to create trace chunk in trace chunk creation command"); + ret = -1; + reply_code = LTTNG_ERR_NOMEM; goto end; } - path_view = lttng_buffer_view_from_view(payload, header_len, - path_info_header.length); + if (msg->override_name_length) { + const char *name; - path = create_output_path(path_view.data); - if (!path) { - ERR("Failed to create output path"); - ret = -1; - goto end; + chunk_name_view = lttng_buffer_view_from_view(payload, + sizeof(*msg), + msg->override_name_length); + name = chunk_name_view.data; + if (!name || name[msg->override_name_length - 1]) { + ERR("Failed to receive payload of chunk creation command"); + ret = -1; + reply_code = LTTNG_ERR_INVALID; + goto end; + } + + chunk_status = lttng_trace_chunk_override_name( + chunk, chunk_name_view.data); + switch (chunk_status) { + case LTTNG_TRACE_CHUNK_STATUS_OK: + break; + case LTTNG_TRACE_CHUNK_STATUS_INVALID_ARGUMENT: + ERR("Failed to set the name of new trace chunk in trace chunk creation command (invalid name)"); + reply_code = LTTNG_ERR_INVALID; + ret = -1; + goto end; + default: + ERR("Failed to set the name of new trace chunk in trace chunk creation command (unknown error)"); + reply_code = LTTNG_ERR_UNK; + ret = -1; + goto end; + } } - ret = utils_mkdir_recursive(path, S_IRWXU | S_IRWXG, -1, -1); - if (ret < 0) { - ERR("relay creating output directory"); + chunk_status = lttng_trace_chunk_set_credentials_current_user(chunk); + if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { + reply_code = LTTNG_ERR_UNK; + ret = -1; goto end; } - ret = 0; - -end: - memset(&reply, 0, sizeof(reply)); - if (ret < 0) { - reply.ret_code = htobe32(LTTNG_ERR_UNK); - } else { - reply.ret_code = htobe32(LTTNG_OK); + session_output = session_create_output_directory_handle( + conn->session); + if (!session_output) { + reply_code = LTTNG_ERR_CREATE_DIR_FAIL; + goto end; } - send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); - if (send_ret < (ssize_t) sizeof(reply)) { - ERR("Failed to send \"mkdir\" command reply (ret = %zd)", send_ret); + chunk_status = lttng_trace_chunk_set_as_owner(chunk, session_output); + lttng_directory_handle_put(session_output); + session_output = NULL; + if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { + reply_code = LTTNG_ERR_UNK; ret = -1; + goto end; } -end_no_session: - free(path); - return ret; -} + published_chunk = sessiond_trace_chunk_registry_publish_chunk( + sessiond_trace_chunk_registry, + conn->session->sessiond_uuid, + conn->session->id, + chunk); + if (!published_chunk) { + char uuid_str[LTTNG_UUID_STR_LEN]; -static int validate_rotate_rename_path_length(const char *path_type, - uint32_t path_length) -{ - int ret = 0; + lttng_uuid_to_str(conn->session->sessiond_uuid, uuid_str); + ERR("Failed to publish chunk: sessiond_uuid = %s, session_id = %" PRIu64 ", chunk_id = %" PRIu64, + uuid_str, + conn->session->id, + msg->chunk_id); + ret = -1; + reply_code = LTTNG_ERR_NOMEM; + goto end; + } - if (path_length > LTTNG_PATH_MAX) { - ret = -ENAMETOOLONG; - ERR("rotate rename \"%s\" path name length (%" PRIu32 " bytes) exceeds the allowed size of %i bytes", - path_type, path_length, LTTNG_PATH_MAX); - } else if (path_length == 0) { - ret = -EINVAL; - ERR("rotate rename \"%s\" path name has an illegal length of 0", path_type); + pthread_mutex_lock(&conn->session->lock); + if (conn->session->pending_closure_trace_chunk) { + /* + * Invalid; this means a second create_trace_chunk command was + * received before a close_trace_chunk. + */ + ERR("Invalid trace chunk close command received; a trace chunk is already waiting for a trace chunk close command"); + reply_code = LTTNG_ERR_INVALID_PROTOCOL; + ret = -1; + goto end_unlock_session; + } + conn->session->pending_closure_trace_chunk = + conn->session->current_trace_chunk; + conn->session->current_trace_chunk = published_chunk; + published_chunk = NULL; + if (!conn->session->pending_closure_trace_chunk) { + session->ongoing_rotation = false; + } +end_unlock_session: + pthread_mutex_unlock(&conn->session->lock); +end: + reply.ret_code = htobe32((uint32_t) reply_code); + 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 \"create trace chunk\" command reply (ret = %zd)", + send_ret); + ret = -1; } +end_no_reply: + lttng_trace_chunk_put(chunk); + lttng_trace_chunk_put(published_chunk); + lttng_directory_handle_put(session_output); return ret; } /* - * relay_rotate_rename: rename the trace folder after a rotation is - * completed. We are not closing any fd here, just moving the folder, so it - * works even if data is still in-flight. + * relay_close_trace_chunk: close a trace chunk */ -static int relay_rotate_rename(const struct lttcomm_relayd_hdr *recv_hdr, +static int relay_close_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr, struct relay_connection *conn, const struct lttng_buffer_view *payload) { - int ret; + int ret = 0, buf_ret; ssize_t send_ret; struct relay_session *session = conn->session; - struct lttcomm_relayd_generic_reply reply; - struct lttcomm_relayd_rotate_rename header; - size_t header_len; - size_t received_paths_size; - char *complete_old_path = NULL, *complete_new_path = NULL; - struct lttng_buffer_view old_path_view; - struct lttng_buffer_view new_path_view; + struct lttcomm_relayd_close_trace_chunk *msg; + struct lttcomm_relayd_close_trace_chunk_reply reply = {}; + struct lttng_buffer_view header_view; + struct lttng_trace_chunk *chunk = NULL; + enum lttng_error_code reply_code = LTTNG_OK; + enum lttng_trace_chunk_status chunk_status; + uint64_t chunk_id; + LTTNG_OPTIONAL(enum lttng_trace_chunk_command_type) close_command = {}; + time_t close_timestamp; + char closed_trace_chunk_path[LTTNG_PATH_MAX]; + size_t path_length = 0; + const char *chunk_name = NULL; + struct lttng_dynamic_buffer reply_payload; + const char *new_path; + + lttng_dynamic_buffer_init(&reply_payload); if (!session || !conn->version_check_done) { - ERR("Trying to rename a trace folder before version check"); + ERR("Trying to close a trace chunk before version check"); ret = -1; goto end_no_reply; } if (session->major == 2 && session->minor < 11) { - ERR("relay_rotate_rename command is unsupported before LTTng 2.11"); + ERR("Chunk close command is unsupported before 2.11"); ret = -1; goto end_no_reply; } - header_len = sizeof(header); - if (payload->size < header_len) { - ERR("Unexpected payload size in \"relay_rotate_rename\": expected >= %zu bytes, got %zu bytes", - header_len, payload->size); + header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg)); + if (!header_view.data) { + ERR("Failed to receive payload of chunk close command"); ret = -1; goto end_no_reply; } - memcpy(&header, payload->data, header_len); - - header.old_path_length = be32toh(header.old_path_length); - header.new_path_length = be32toh(header.new_path_length); - received_paths_size = header.old_path_length + header.new_path_length; - - if (payload->size < header_len + received_paths_size) { - ERR("Unexpected payload size in \"relay_rotate_rename\" including paths: expected >= %zu bytes, got %zu bytes", - header_len, payload->size); + /* Convert to host endianness. */ + msg = (typeof(msg)) header_view.data; + chunk_id = be64toh(msg->chunk_id); + close_timestamp = (time_t) be64toh(msg->close_timestamp); + close_command = (typeof(close_command)){ + .value = be32toh(msg->close_command.value), + .is_set = msg->close_command.is_set, + }; + + chunk = sessiond_trace_chunk_registry_get_chunk( + sessiond_trace_chunk_registry, + conn->session->sessiond_uuid, + conn->session->id, + chunk_id); + if (!chunk) { + char uuid_str[LTTNG_UUID_STR_LEN]; + + lttng_uuid_to_str(conn->session->sessiond_uuid, uuid_str); + ERR("Failed to find chunk to close: sessiond_uuid = %s, session_id = %" PRIu64 ", chunk_id = %" PRIu64, + uuid_str, + conn->session->id, + msg->chunk_id); ret = -1; - goto end_no_reply; + reply_code = LTTNG_ERR_NOMEM; + goto end; } - /* Ensure the paths don't exceed their allowed size. */ - ret = validate_rotate_rename_path_length("old", header.old_path_length); - if (ret) { - goto end; + pthread_mutex_lock(&session->lock); + if (close_command.is_set && + close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_DELETE) { + /* + * Clear command. It is a protocol error to ask for a + * clear on a relay which does not allow it. Querying + * the configuration allows figuring out whether + * clearing is allowed before doing the clear. + */ + if (!opt_allow_clear) { + ret = -1; + reply_code = LTTNG_ERR_INVALID_PROTOCOL; + goto end_unlock_session; + } } - ret = validate_rotate_rename_path_length("new", header.new_path_length); - if (ret) { - goto end; + if (session->pending_closure_trace_chunk && + session->pending_closure_trace_chunk != chunk) { + ERR("Trace chunk close command for session \"%s\" does not target the trace chunk pending closure", + session->session_name); + reply_code = LTTNG_ERR_INVALID_PROTOCOL; + ret = -1; + goto end_unlock_session; } - old_path_view = lttng_buffer_view_from_view(payload, header_len, - header.old_path_length); - new_path_view = lttng_buffer_view_from_view(payload, - header_len + header.old_path_length, - header.new_path_length); + if (session->current_trace_chunk && session->current_trace_chunk != chunk && + !lttng_trace_chunk_get_name_overridden(session->current_trace_chunk)) { + if (close_command.is_set && + close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_DELETE && + !session->has_rotated) { + /* New chunk stays in session output directory. */ + new_path = ""; + } else { + /* Use chunk name for new chunk. */ + new_path = NULL; + } + /* Rename new chunk path. */ + chunk_status = lttng_trace_chunk_rename_path(session->current_trace_chunk, + new_path); + if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { + ret = -1; + goto end; + } + session->ongoing_rotation = false; + } + if ((!close_command.is_set || + close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_NO_OPERATION) && + !lttng_trace_chunk_get_name_overridden(chunk)) { + const char *old_path; - /* Validate that both paths received are NULL terminated. */ - if (old_path_view.data[old_path_view.size - 1] != '\0') { - ERR("relay_rotate_rename command's \"old\" path is invalid (not NULL terminated)"); - ret = -1; - goto end; + if (!session->has_rotated) { + old_path = ""; + } else { + old_path = NULL; + } + /* We need to move back the .tmp_old_chunk to its rightful place. */ + chunk_status = lttng_trace_chunk_rename_path(chunk, old_path); + if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { + ret = -1; + goto end; + } } - if (new_path_view.data[new_path_view.size - 1] != '\0') { - ERR("relay_rotate_rename command's \"new\" path is invalid (not NULL terminated)"); + chunk_status = lttng_trace_chunk_set_close_timestamp( + chunk, close_timestamp); + if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { + ERR("Failed to set trace chunk close timestamp"); ret = -1; - goto end; + reply_code = LTTNG_ERR_UNK; + goto end_unlock_session; } - complete_old_path = create_output_path(old_path_view.data); - if (!complete_old_path) { - ERR("Failed to build old output path in rotate_rename command"); + if (close_command.is_set) { + chunk_status = lttng_trace_chunk_set_close_command( + chunk, close_command.value); + if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { + ret = -1; + reply_code = LTTNG_ERR_INVALID; + goto end_unlock_session; + } + } + chunk_status = lttng_trace_chunk_get_name(chunk, &chunk_name, NULL); + if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { + ERR("Failed to get chunk name"); ret = -1; - goto end; + reply_code = LTTNG_ERR_UNK; + goto end_unlock_session; } - - complete_new_path = create_output_path(new_path_view.data); - if (!complete_new_path) { - ERR("Failed to build new output path in rotate_rename command"); + if (!session->has_rotated && !session->snapshot) { + ret = lttng_strncpy(closed_trace_chunk_path, + session->output_path, + sizeof(closed_trace_chunk_path)); + if (ret) { + ERR("Failed to send trace chunk path: path length of %zu bytes exceeds the maximal allowed length of %zu bytes", + strlen(session->output_path), + sizeof(closed_trace_chunk_path)); + reply_code = LTTNG_ERR_NOMEM; + ret = -1; + goto end_unlock_session; + } + } else { + if (session->snapshot) { + ret = snprintf(closed_trace_chunk_path, + sizeof(closed_trace_chunk_path), + "%s/%s", session->output_path, + chunk_name); + } else { + ret = snprintf(closed_trace_chunk_path, + sizeof(closed_trace_chunk_path), + "%s/" DEFAULT_ARCHIVED_TRACE_CHUNKS_DIRECTORY + "/%s", + session->output_path, chunk_name); + } + if (ret < 0 || ret == sizeof(closed_trace_chunk_path)) { + ERR("Failed to format closed trace chunk resulting path"); + reply_code = ret < 0 ? LTTNG_ERR_UNK : LTTNG_ERR_NOMEM; + ret = -1; + goto end_unlock_session; + } + } + if (close_command.is_set && + close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED) { + session->has_rotated = true; + } + DBG("Reply chunk path on close: %s", closed_trace_chunk_path); + path_length = strlen(closed_trace_chunk_path) + 1; + if (path_length > UINT32_MAX) { + ERR("Closed trace chunk path exceeds the maximal length allowed by the protocol"); ret = -1; - goto end; + reply_code = LTTNG_ERR_INVALID_PROTOCOL; + goto end_unlock_session; } - ret = utils_mkdir_recursive(complete_new_path, S_IRWXU | S_IRWXG, - -1, -1); - if (ret < 0) { - ERR("Failed to mkdir() rotate_rename's \"new\" output directory at \"%s\"", - complete_new_path); - goto end; + if (session->current_trace_chunk == chunk) { + /* + * After a trace chunk close command, no new streams + * referencing the chunk may be created. Hence, on the + * event that no new trace chunk have been created for + * the session, the reference to the current trace chunk + * is released in order to allow it to be reclaimed when + * the last stream releases its reference to it. + */ + lttng_trace_chunk_put(session->current_trace_chunk); + session->current_trace_chunk = NULL; } + lttng_trace_chunk_put(session->pending_closure_trace_chunk); + session->pending_closure_trace_chunk = NULL; +end_unlock_session: + pthread_mutex_unlock(&session->lock); - /* - * If a domain has not yet created its channel, the domain-specific - * folder might not exist, but this is not an error. - */ - ret = rename(complete_old_path, complete_new_path); - if (ret < 0 && errno != ENOENT) { - PERROR("Renaming chunk in rotate_rename command from \"%s\" to \"%s\"", - complete_old_path, complete_new_path); - goto end; +end: + reply.generic.ret_code = htobe32((uint32_t) reply_code); + reply.path_length = htobe32((uint32_t) path_length); + buf_ret = lttng_dynamic_buffer_append( + &reply_payload, &reply, sizeof(reply)); + if (buf_ret) { + ERR("Failed to append \"close trace chunk\" command reply header to payload buffer"); + goto end_no_reply; } - 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); + if (reply_code == LTTNG_OK) { + buf_ret = lttng_dynamic_buffer_append(&reply_payload, + closed_trace_chunk_path, path_length); + if (buf_ret) { + ERR("Failed to append \"close trace chunk\" command reply path to payload buffer"); + goto end_no_reply; + } } - send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, - sizeof(reply), 0); - if (send_ret < sizeof(reply)) { - ERR("Failed to send \"rotate rename\" command reply (ret = %zd)", - send_ret); + + send_ret = conn->sock->ops->sendmsg(conn->sock, + reply_payload.data, + reply_payload.size, + 0); + if (send_ret < reply_payload.size) { + ERR("Failed to send \"close trace chunk\" command reply of %zu bytes (ret = %zd)", + reply_payload.size, send_ret); ret = -1; + goto end_no_reply; } - end_no_reply: - free(complete_old_path); - free(complete_new_path); + lttng_trace_chunk_put(chunk); + lttng_dynamic_buffer_reset(&reply_payload); return ret; } /* - * Check if all the streams in the session have completed the last rotation. - * The chunk_id value is used to distinguish the cases where a stream was - * closed on the consumerd before the rotation started but it still active on - * the relayd, and the case where a stream appeared on the consumerd/relayd - * after the last rotation started (in that case, it is already writing in the - * new chunk folder). + * relay_trace_chunk_exists: check if a trace chunk exists */ -static -int relay_rotate_pending(const struct lttcomm_relayd_hdr *recv_hdr, +static int relay_trace_chunk_exists(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_rotate_pending msg; - struct lttcomm_relayd_rotate_pending_reply reply; - struct lttng_ht_iter iter; - struct relay_stream *stream; int ret = 0; ssize_t send_ret; + struct relay_session *session = conn->session; + struct lttcomm_relayd_trace_chunk_exists *msg; + struct lttcomm_relayd_trace_chunk_exists_reply reply = {}; + struct lttng_buffer_view header_view; uint64_t chunk_id; - bool rotate_pending = false; - - DBG("Rotate pending command received"); + bool chunk_exists; if (!session || !conn->version_check_done) { - ERR("Trying to check for data before version check"); + ERR("Trying to close a trace chunk before version check"); ret = -1; goto end_no_reply; } if (session->major == 2 && session->minor < 11) { - ERR("Unsupported feature before 2.11"); + ERR("Chunk close command is unsupported before 2.11"); ret = -1; goto end_no_reply; } - if (payload->size < sizeof(msg)) { - ERR("Unexpected payload size in \"relay_rotate_pending\": expected >= %zu bytes, got %zu bytes", - sizeof(msg), payload->size); + header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg)); + if (!header_view.data) { + ERR("Failed to receive payload of chunk close command"); ret = -1; goto end_no_reply; } - memcpy(&msg, payload->data, sizeof(msg)); - - chunk_id = be64toh(msg.chunk_id); - - DBG("Evaluating rotate pending for chunk id %" PRIu64, chunk_id); + /* Convert to host endianness. */ + msg = (typeof(msg)) header_view.data; + chunk_id = be64toh(msg->chunk_id); + ret = sessiond_trace_chunk_registry_chunk_exists( + sessiond_trace_chunk_registry, + conn->session->sessiond_uuid, + conn->session->id, + chunk_id, &chunk_exists); /* - * Iterate over all the streams in the session and check if they are - * still waiting for data to perform their rotation. + * If ret is not 0, send the reply and report the error to the caller. + * It is a protocol (or internal) error and the session/connection + * should be torn down. */ - rcu_read_lock(); - cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream, - node.node) { - if (!stream_get(stream)) { - continue; - } - if (stream->trace->session != session) { - stream_put(stream); - continue; - } - pthread_mutex_lock(&stream->lock); - if (stream->rotate_at_seq_num != -1ULL) { - /* We have not yet performed the rotation. */ - rotate_pending = true; - DBG("Stream %" PRIu64 " is still rotating", - stream->stream_handle); - } else if (stream->chunk_id < chunk_id) { - /* - * Stream closed on the consumer but still active on the - * relay. - */ - rotate_pending = true; - DBG("Stream %" PRIu64 " did not exist on the consumer " - "when the last rotation started, but is" - "still waiting for data before getting" - "closed", - stream->stream_handle); - } - pthread_mutex_unlock(&stream->lock); - stream_put(stream); - if (rotate_pending) { - goto send_reply; - } + reply = (typeof(reply)){ + .generic.ret_code = htobe32((uint32_t) + (ret == 0 ? LTTNG_OK : LTTNG_ERR_INVALID_PROTOCOL)), + .trace_chunk_exists = ret == 0 ? chunk_exists : 0, + }; + send_ret = conn->sock->ops->sendmsg( + conn->sock, &reply, sizeof(reply), 0); + if (send_ret < (ssize_t) sizeof(reply)) { + ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)", + send_ret); + ret = -1; } +end_no_reply: + return ret; +} -send_reply: - rcu_read_unlock(); - memset(&reply, 0, sizeof(reply)); - reply.generic.ret_code = htobe32((uint32_t) LTTNG_OK); - reply.is_pending = (uint8_t) !!rotate_pending; - send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, - sizeof(reply), 0); +/* + * relay_get_configuration: query whether feature is available + */ +static int relay_get_configuration(const struct lttcomm_relayd_hdr *recv_hdr, + struct relay_connection *conn, + const struct lttng_buffer_view *payload) +{ + int ret = 0; + ssize_t send_ret; + struct lttcomm_relayd_get_configuration *msg; + struct lttcomm_relayd_get_configuration_reply reply = {}; + struct lttng_buffer_view header_view; + uint64_t query_flags = 0; + uint64_t result_flags = 0; + + header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg)); + if (!header_view.data) { + ERR("Failed to receive payload of chunk close command"); + ret = -1; + goto end_no_reply; + } + + /* Convert to host endianness. */ + msg = (typeof(msg)) header_view.data; + query_flags = be64toh(msg->query_flags); + + if (query_flags) { + ret = LTTNG_ERR_INVALID_PROTOCOL; + goto reply; + } + if (opt_allow_clear) { + result_flags |= LTTCOMM_RELAYD_CONFIGURATION_FLAG_CLEAR_ALLOWED; + } + ret = 0; +reply: + reply = (typeof(reply)){ + .generic.ret_code = htobe32((uint32_t) + (ret == 0 ? LTTNG_OK : LTTNG_ERR_INVALID_PROTOCOL)), + .relayd_configuration_flags = htobe64(result_flags), + }; + send_ret = conn->sock->ops->sendmsg( + conn->sock, &reply, sizeof(reply), 0); if (send_ret < (ssize_t) sizeof(reply)) { - ERR("Failed to send \"rotate pending\" command reply (ret = %zd)", + ERR("Failed to send \"get configuration\" command reply (ret = %zd)", send_ret); ret = -1; } - end_no_reply: return ret; } @@ -2932,21 +3111,25 @@ static int relay_process_control_command(struct relay_connection *conn, DBG_CMD("RELAYD_RESET_METADATA", conn); ret = relay_reset_metadata(header, conn, payload); break; - case RELAYD_ROTATE_STREAM: - DBG_CMD("RELAYD_ROTATE_STREAM", conn); - ret = relay_rotate_session_stream(header, conn, payload); + case RELAYD_ROTATE_STREAMS: + DBG_CMD("RELAYD_ROTATE_STREAMS", conn); + ret = relay_rotate_session_streams(header, conn, payload); + break; + case RELAYD_CREATE_TRACE_CHUNK: + DBG_CMD("RELAYD_CREATE_TRACE_CHUNK", conn); + ret = relay_create_trace_chunk(header, conn, payload); break; - case RELAYD_ROTATE_RENAME: - DBG_CMD("RELAYD_ROTATE_RENAME", conn); - ret = relay_rotate_rename(header, conn, payload); + case RELAYD_CLOSE_TRACE_CHUNK: + DBG_CMD("RELAYD_CLOSE_TRACE_CHUNK", conn); + ret = relay_close_trace_chunk(header, conn, payload); break; - case RELAYD_ROTATE_PENDING: - DBG_CMD("RELAYD_ROTATE_PENDING", conn); - ret = relay_rotate_pending(header, conn, payload); + case RELAYD_TRACE_CHUNK_EXISTS: + DBG_CMD("RELAYD_TRACE_CHUNK_EXISTS", conn); + ret = relay_trace_chunk_exists(header, conn, payload); break; - case RELAYD_MKDIR: - DBG_CMD("RELAYD_MKDIR", conn); - ret = relay_mkdir(header, conn, payload); + case RELAYD_GET_CONFIGURATION: + DBG_CMD("RELAYD_GET_CONFIGURATION", conn); + ret = relay_get_configuration(header, conn, payload); break; case RELAYD_UPDATE_SYNC_INFO: default: @@ -3148,75 +3331,6 @@ static enum relay_connection_status relay_process_control( return status; } -/* - * Handle index for a data stream. - * - * Called with the stream lock held. - * - * Return 0 on success else a negative value. - */ -static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num, - bool rotate_index, bool *flushed, uint64_t total_size) -{ - int ret = 0; - uint64_t data_offset; - struct relay_index *index; - - /* Get data offset because we are about to update the index. */ - data_offset = htobe64(stream->tracefile_size_current); - - DBG("handle_index_data: stream %" PRIu64 " net_seq_num %" PRIu64 " data offset %" PRIu64, - stream->stream_handle, net_seq_num, stream->tracefile_size_current); - - /* - * Lookup for an existing index for that stream id/sequence - * number. If it exists, the control thread has already received the - * data for it, thus we need to write it to disk. - */ - index = relay_index_get_by_id_or_create(stream, net_seq_num); - if (!index) { - ret = -1; - goto end; - } - - if (rotate_index || !stream->index_file) { - ret = create_rotate_index_file(stream); - if (ret < 0) { - ERR("Failed to rotate index"); - /* Put self-ref for this index due to error. */ - relay_index_put(index); - index = NULL; - goto end; - } - } - - 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; - } - - ret = relay_index_try_flush(index); - if (ret == 0) { - tracefile_array_commit_seq(stream->tfa); - stream->index_received_seqcount++; - *flushed = true; - } else if (ret > 0) { - index->total_size = total_size; - /* No flush. */ - ret = 0; - } else { - /* Put self-ref for this index due to error. */ - relay_index_put(index); - index = NULL; - ret = -1; - } -end: - return ret; -} - static enum relay_connection_status relay_process_data_receive_header( struct relay_connection *conn) { @@ -3293,40 +3407,17 @@ static enum relay_connection_status relay_process_data_receive_header( } pthread_mutex_lock(&stream->lock); - - /* Check if a rotation is needed. */ - if (stream->tracefile_size > 0 && - (stream->tracefile_size_current + header.data_size) > - stream->tracefile_size) { - uint64_t old_id, new_id; - - old_id = tracefile_array_get_file_index_head(stream->tfa); - tracefile_array_file_rotate(stream->tfa); - - /* new_id is updated by utils_rotate_stream_file. */ - new_id = old_id; - - ret = utils_rotate_stream_file(stream->path_name, - stream->channel_name, stream->tracefile_size, - stream->tracefile_count, -1, - -1, stream->stream_fd->fd, - &new_id, &stream->stream_fd->fd); - if (ret < 0) { - ERR("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; - conn->protocol.data.state.receive_payload.rotate_index = true; + /* Prepare stream for the reception of a new packet. */ + ret = stream_init_packet(stream, header.data_size, + &conn->protocol.data.state.receive_payload.rotate_index); + pthread_mutex_unlock(&stream->lock); + if (ret) { + ERR("Failed to rotate stream output file"); + status = RELAY_CONNECTION_STATUS_ERROR; + goto end_stream_unlock; } end_stream_unlock: - pthread_mutex_unlock(&stream->lock); stream_put(stream); end: return status; @@ -3377,8 +3468,8 @@ static enum relay_connection_status relay_process_data_receive_payload( * - the on-stack data buffer */ while (left_to_receive > 0 && !partial_recv) { - ssize_t write_ret; size_t recv_size = min(left_to_receive, chunk_size); + struct lttng_buffer_view packet_chunk; ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, recv_size, MSG_DONTWAIT); @@ -3400,14 +3491,15 @@ static enum relay_connection_status relay_process_data_receive_payload( * consumed. */ partial_recv = true; + recv_size = ret; } - recv_size = ret; + packet_chunk = lttng_buffer_view_init(data_buffer, + 0, recv_size); + assert(packet_chunk.data); - /* 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) { + ret = stream_write(stream, &packet_chunk, 0); + if (ret) { ERR("Relay error writing data to file"); status = RELAY_CONNECTION_STATUS_ERROR; goto end_stream_unlock; @@ -3416,9 +3508,6 @@ static enum relay_connection_status relay_process_data_receive_payload( 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); } if (state->left_to_receive > 0) { @@ -3432,22 +3521,18 @@ static enum relay_connection_status relay_process_data_receive_payload( goto end_stream_unlock; } - ret = write_padding_to_file(stream->stream_fd->fd, - 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); + ret = stream_write(stream, NULL, state->header.padding_size); + if (ret) { status = RELAY_CONNECTION_STATUS_ERROR; goto end_stream_unlock; } - - if (session->minor >= 4 && !session->snapshot) { - ret = handle_index_data(stream, state->header.net_seq_num, - state->rotate_index, &index_flushed, state->header.data_size + state->header.padding_size); + if (session_streams_have_index(session)) { + ret = stream_update_index(stream, state->header.net_seq_num, + state->rotate_index, &index_flushed, + state->header.data_size + state->header.padding_size); if (ret < 0) { - ERR("handle_index_data: fail stream %" PRIu64 " net_seq_num %" PRIu64 " ret %d", + ERR("Failed to update index: stream %" PRIu64 " net_seq_num %" PRIu64 " ret %d", stream->stream_handle, state->header.net_seq_num, ret); status = RELAY_CONNECTION_STATUS_ERROR; @@ -3455,18 +3540,17 @@ static enum relay_connection_status relay_process_data_receive_payload( } } - stream->tracefile_size_current += state->header.data_size + - state->header.padding_size; - - if (stream->prev_seq == -1ULL) { + if (stream->prev_data_seq == -1ULL) { new_stream = true; } - if (index_flushed) { - stream->pos_after_last_complete_data_index = - stream->tracefile_size_current; - } - stream->prev_seq = state->header.net_seq_num; + ret = stream_complete_packet(stream, state->header.data_size + + state->header.padding_size, state->header.net_seq_num, + index_flushed); + if (ret) { + status = RELAY_CONNECTION_STATUS_ERROR; + goto end_stream_unlock; + } /* * Resetting the protocol state (to RECEIVE_HEADER) will trash the @@ -3476,12 +3560,6 @@ static enum relay_connection_status relay_process_data_receive_payload( connection_reset_protocol_state(conn); state = NULL; - ret = try_rotate_stream(stream); - if (ret < 0) { - status = RELAY_CONNECTION_STATUS_ERROR; - goto end_stream_unlock; - } - end_stream_unlock: close_requested = stream->close_requested; pthread_mutex_unlock(&stream->lock); @@ -3636,14 +3714,6 @@ restart: health_code_update(); - if (!revents) { - /* - * No activity for this FD (poll - * implementation). - */ - continue; - } - /* Thread quit pipe has been closed. Killing thread. */ ret = check_thread_quit_pipe(pollfd, revents); if (ret) { @@ -3660,8 +3730,13 @@ restart: if (ret < 0) { goto error; } - lttng_poll_add(&events, conn->sock->fd, + ret = lttng_poll_add(&events, + conn->sock->fd, LPOLLIN | LPOLLRDHUP); + if (ret) { + ERR("Failed to add new connection file descriptor to poll set"); + goto error; + } connection_ht_add(relay_connections_ht, conn); DBG("Connection socket %d added", conn->sock->fd); } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { @@ -3905,7 +3980,17 @@ int main(int argc, char **argv) int ret = 0, retval = 0; void *status; - /* Parse arguments */ + /* Parse environment variables */ + ret = parse_env_options(); + if (ret) { + retval = -1; + goto exit_options; + } + + /* + * Parse arguments. + * Command line arguments overwrite environment. + */ progname = argv[0]; if (set_options(argc, argv)) { retval = -1; @@ -3917,6 +4002,22 @@ 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; + } + + DBG("Clear command %s", opt_allow_clear ? "allowed" : "disallowed"); + /* Try to create directory if -o, --output is specified. */ if (opt_output_path) { if (*opt_output_path != '/') { @@ -3955,24 +4056,39 @@ int main(int argc, char **argv) } } + if (opt_working_directory) { + ret = utils_change_working_directory(opt_working_directory); + if (ret) { + /* All errors are already logged. */ + goto exit_options; + } + } + + sessiond_trace_chunk_registry = sessiond_trace_chunk_registry_create(); + if (!sessiond_trace_chunk_registry) { + ERR("Failed to initialize session daemon trace chunk registry"); + retval = -1; + goto exit_options; + } + /* Initialize thread health monitoring */ health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES); if (!health_relayd) { PERROR("health_app_create error"); retval = -1; - goto exit_health_app_create; + goto exit_options; } /* Create thread quit pipe */ if (init_thread_quit_pipe()) { retval = -1; - goto exit_init_data; + goto exit_options; } /* Setup the thread apps communication pipe. */ if (create_relay_conn_pipe()) { retval = -1; - goto exit_init_data; + goto exit_options; } /* Init relay command queue. */ @@ -3986,27 +4102,27 @@ int main(int argc, char **argv) sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); if (!sessions_ht) { retval = -1; - goto exit_init_data; + goto exit_options; } /* tables of streams indexed by stream ID */ relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); if (!relay_streams_ht) { retval = -1; - goto exit_init_data; + goto exit_options; } /* tables of streams indexed by stream ID */ viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); if (!viewer_streams_ht) { retval = -1; - goto exit_init_data; + goto exit_options; } ret = utils_create_pipe(health_quit_pipe); if (ret) { retval = -1; - goto exit_health_quit_pipe; + goto exit_options; } /* Create thread to manage the client socket */ @@ -4016,7 +4132,7 @@ int main(int argc, char **argv) errno = ret; PERROR("pthread_create health"); retval = -1; - goto exit_health_thread; + goto exit_options; } /* Setup the dispatcher thread */ @@ -4097,14 +4213,6 @@ exit_dispatcher_thread: PERROR("pthread_join health_thread"); retval = -1; } -exit_health_thread: - - utils_close_pipe(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