X-Git-Url: http://git.liburcu.org/?a=blobdiff_plain;f=src%2Fcommon%2Fust-consumer%2Fust-consumer.c;h=17672fb20950e77f5d61d2738ef5109f059a161f;hb=65121c2141d0df9a1d6fc759904cea4bbc016331;hp=01d5ecf6505cc0667a41194d5cb1f07d06e3540b;hpb=83f4233dde73280d45fcfb799add96cce32098ce;p=lttng-tools.git diff --git a/src/common/ust-consumer/ust-consumer.c b/src/common/ust-consumer/ust-consumer.c index 01d5ecf65..17672fb20 100644 --- a/src/common/ust-consumer/ust-consumer.c +++ b/src/common/ust-consumer/ust-consumer.c @@ -1,6 +1,7 @@ /* * Copyright (C) 2011 - Julien Desfossez * Mathieu Desnoyers + * Copyright (C) 2017 - Jérémie Galarneau * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License, version 2 only, @@ -50,7 +51,6 @@ extern struct lttng_consumer_global_data consumer_data; extern int consumer_poll_timeout; -extern volatile int consumer_quit; /* * Free channel object and all streams associated with it. This MUST be used @@ -510,7 +510,7 @@ error_open: } /* Try to rmdir all directories under shm_path root. */ if (channel->root_shm_path[0]) { - (void) run_as_recursive_rmdir(channel->root_shm_path, + (void) run_as_rmdir_recursive(channel->root_shm_path, channel->uid, channel->gid); } free(stream_fds); @@ -768,10 +768,19 @@ static int flush_channel(uint64_t chan_key) health_code_update(); pthread_mutex_lock(&stream->lock); + + /* + * Protect against concurrent teardown of a stream. + */ + if (cds_lfht_is_node_deleted(&stream->node.node)) { + goto next; + } + if (!stream->quiescent) { ustctl_flush_buffer(stream->ustream, 0); stream->quiescent = true; } +next: pthread_mutex_unlock(&stream->lock); } error: @@ -831,6 +840,7 @@ static int close_metadata(uint64_t chan_key) { int ret = 0; struct lttng_consumer_channel *channel; + unsigned int channel_monitor; DBG("UST consumer close metadata key %" PRIu64, chan_key); @@ -849,13 +859,48 @@ static int close_metadata(uint64_t chan_key) pthread_mutex_lock(&consumer_data.lock); pthread_mutex_lock(&channel->lock); - + channel_monitor = channel->monitor; if (cds_lfht_is_node_deleted(&channel->node.node)) { goto error_unlock; } lttng_ustconsumer_close_metadata(channel); + pthread_mutex_unlock(&channel->lock); + pthread_mutex_unlock(&consumer_data.lock); + + /* + * The ownership of a metadata channel depends on the type of + * session to which it belongs. In effect, the monitor flag is checked + * to determine if this metadata channel is in "snapshot" mode or not. + * + * In the non-snapshot case, the metadata channel is created along with + * a single stream which will remain present until the metadata channel + * is destroyed (on the destruction of its session). In this case, the + * metadata stream in "monitored" by the metadata poll thread and holds + * the ownership of its channel. + * + * Closing the metadata will cause the metadata stream's "metadata poll + * pipe" to be closed. Closing this pipe will wake-up the metadata poll + * thread which will teardown the metadata stream which, in return, + * deletes the metadata channel. + * + * In the snapshot case, the metadata stream is created and destroyed + * on every snapshot record. Since the channel doesn't have an owner + * other than the session daemon, it is safe to destroy it immediately + * on reception of the CLOSE_METADATA command. + */ + if (!channel_monitor) { + /* + * The channel and consumer_data locks must be + * released before this call since consumer_del_channel + * re-acquires the channel and consumer_data locks to teardown + * the channel and queue its reclamation by the "call_rcu" + * worker thread. + */ + consumer_del_channel(channel); + } + return ret; error_unlock: pthread_mutex_unlock(&channel->lock); pthread_mutex_unlock(&consumer_data.lock); @@ -1076,9 +1121,6 @@ static int snapshot_channel(uint64_t key, char *path, uint64_t relayd_id, DBG("UST consumer snapshot channel %" PRIu64, key); cds_list_for_each_entry(stream, &channel->streams.head, send_node) { - /* Are we at a position _before_ the first available packet ? */ - bool before_first_packet = true; - health_code_update(); /* Lock stream because we are about to change its state. */ @@ -1104,12 +1146,6 @@ static int snapshot_channel(uint64_t key, char *path, uint64_t relayd_id, DBG("UST consumer snapshot stream %s/%s (%" PRIu64 ")", path, stream->name, stream->key); } - if (relayd_id != -1ULL) { - ret = consumer_send_relayd_streams_sent(relayd_id); - if (ret < 0) { - goto error_unlock; - } - } /* * If tracing is active, we want to perform a "full" buffer flush. @@ -1147,10 +1183,9 @@ static int snapshot_channel(uint64_t key, char *path, uint64_t relayd_id, produced_pos, nb_packets_per_stream, stream->max_sb_size); - while (consumed_pos < produced_pos) { + while ((long) (consumed_pos - produced_pos) < 0) { ssize_t read_len; unsigned long len, padded_len; - int lost_packet = 0; health_code_update(); @@ -1164,15 +1199,7 @@ static int snapshot_channel(uint64_t key, char *path, uint64_t relayd_id, } DBG("UST consumer get subbuf failed. Skipping it."); consumed_pos += stream->max_sb_size; - - /* - * Start accounting lost packets only when we - * already have extracted packets (to match the - * content of the final snapshot). - */ - if (!before_first_packet) { - lost_packet = 1; - } + stream->chan->lost_packets++; continue; } @@ -1208,16 +1235,6 @@ static int snapshot_channel(uint64_t key, char *path, uint64_t relayd_id, goto error_close_stream; } consumed_pos += stream->max_sb_size; - - /* - * Only account lost packets located between - * succesfully extracted packets (do not account before - * and after since they are not visible in the - * resulting snapshot). - */ - stream->chan->lost_packets += lost_packet; - lost_packet = 0; - before_first_packet = false; } /* Simply close the stream so we can use it on the next snapshot. */ @@ -1354,7 +1371,7 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, case LTTNG_CONSUMER_ADD_RELAYD_SOCKET: { /* Session daemon status message are handled in the following call. */ - ret = consumer_add_relayd_socket(msg.u.relayd_sock.net_index, + consumer_add_relayd_socket(msg.u.relayd_sock.net_index, msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll, &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id, msg.u.relayd_sock.relayd_session_id); @@ -1455,6 +1472,7 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, attr.read_timer_interval = msg.u.ask_channel.read_timer_interval; attr.chan_id = msg.u.ask_channel.chan_id; memcpy(attr.uuid, msg.u.ask_channel.uuid, sizeof(attr.uuid)); + attr.blocking_timeout= msg.u.ask_channel.blocking_timeout; /* Match channel buffer type to the UST abi. */ switch (msg.u.ask_channel.output) { @@ -1501,8 +1519,17 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, consumer_timer_switch_start(channel, attr.switch_timer_interval); attr.switch_timer_interval = 0; } else { + int monitor_start_ret; + consumer_timer_live_start(channel, msg.u.ask_channel.live_timer_interval); + monitor_start_ret = consumer_timer_monitor_start( + channel, + msg.u.ask_channel.monitor_timer_interval); + if (monitor_start_ret < 0) { + ERR("Starting channel monitoring timer failed"); + goto end_channel_error; + } } health_code_update(); @@ -1525,6 +1552,9 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, if (channel->live_timer_enabled == 1) { consumer_timer_live_stop(channel); } + if (channel->monitor_timer_enabled == 1) { + consumer_timer_monitor_stop(channel); + } goto end_channel_error; } @@ -1857,6 +1887,51 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, break; } + case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE: + { + int channel_monitor_pipe; + + ret_code = LTTCOMM_CONSUMERD_SUCCESS; + /* Successfully received the command's type. */ + ret = consumer_send_status_msg(sock, ret_code); + if (ret < 0) { + goto error_fatal; + } + + ret = lttcomm_recv_fds_unix_sock(sock, &channel_monitor_pipe, + 1); + if (ret != sizeof(channel_monitor_pipe)) { + ERR("Failed to receive channel monitor pipe"); + goto error_fatal; + } + + DBG("Received channel monitor pipe (%d)", channel_monitor_pipe); + ret = consumer_timer_thread_set_channel_monitor_pipe( + channel_monitor_pipe); + if (!ret) { + int flags; + + ret_code = LTTCOMM_CONSUMERD_SUCCESS; + /* Set the pipe as non-blocking. */ + ret = fcntl(channel_monitor_pipe, F_GETFL, 0); + if (ret == -1) { + PERROR("fcntl get flags of the channel monitoring pipe"); + goto error_fatal; + } + flags = ret; + + ret = fcntl(channel_monitor_pipe, F_SETFL, + flags | O_NONBLOCK); + if (ret == -1) { + PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe"); + goto error_fatal; + } + DBG("Channel monitor pipe set as non-blocking"); + } else { + ret_code = LTTCOMM_CONSUMERD_ALREADY_SET; + } + goto end_msg_sessiond; + } default: break; } @@ -1938,7 +2013,7 @@ void *lttng_ustctl_get_mmap_base(struct lttng_consumer_stream *stream) } /* - * Take a snapshot for a specific fd + * Take a snapshot for a specific stream. * * Returns 0 on success, < 0 on error */ @@ -1950,6 +2025,20 @@ int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream *stream) return ustctl_snapshot(stream->ustream); } +/* + * Sample consumed and produced positions for a specific stream. + * + * Returns 0 on success, < 0 on error. + */ +int lttng_ustconsumer_sample_snapshot_positions( + struct lttng_consumer_stream *stream) +{ + assert(stream); + assert(stream->ustream); + + return ustctl_snapshot_sample_positions(stream->ustream); +} + /* * Get the produced position * @@ -2067,7 +2156,7 @@ void lttng_ustconsumer_free_channel(struct lttng_consumer_channel *chan) ustctl_destroy_channel(chan->uchan); /* Try to rmdir all directories under shm_path root. */ if (chan->root_shm_path[0]) { - (void) run_as_recursive_rmdir(chan->root_shm_path, + (void) run_as_rmdir_recursive(chan->root_shm_path, chan->uid, chan->gid); } free(chan->stream_fds); @@ -2109,62 +2198,69 @@ static int get_index_values(struct ctf_packet_index *index, struct ustctl_consumer_stream *ustream) { int ret; + uint64_t packet_size, content_size, timestamp_begin, timestamp_end, + events_discarded, stream_id, stream_instance_id, + packet_seq_num; - ret = ustctl_get_timestamp_begin(ustream, &index->timestamp_begin); + ret = ustctl_get_timestamp_begin(ustream, ×tamp_begin); if (ret < 0) { PERROR("ustctl_get_timestamp_begin"); goto error; } - index->timestamp_begin = htobe64(index->timestamp_begin); - ret = ustctl_get_timestamp_end(ustream, &index->timestamp_end); + ret = ustctl_get_timestamp_end(ustream, ×tamp_end); if (ret < 0) { PERROR("ustctl_get_timestamp_end"); goto error; } - index->timestamp_end = htobe64(index->timestamp_end); - ret = ustctl_get_events_discarded(ustream, &index->events_discarded); + ret = ustctl_get_events_discarded(ustream, &events_discarded); if (ret < 0) { PERROR("ustctl_get_events_discarded"); goto error; } - index->events_discarded = htobe64(index->events_discarded); - ret = ustctl_get_content_size(ustream, &index->content_size); + ret = ustctl_get_content_size(ustream, &content_size); if (ret < 0) { PERROR("ustctl_get_content_size"); goto error; } - index->content_size = htobe64(index->content_size); - ret = ustctl_get_packet_size(ustream, &index->packet_size); + ret = ustctl_get_packet_size(ustream, &packet_size); if (ret < 0) { PERROR("ustctl_get_packet_size"); goto error; } - index->packet_size = htobe64(index->packet_size); - ret = ustctl_get_stream_id(ustream, &index->stream_id); + ret = ustctl_get_stream_id(ustream, &stream_id); if (ret < 0) { PERROR("ustctl_get_stream_id"); goto error; } - index->stream_id = htobe64(index->stream_id); - ret = ustctl_get_instance_id(ustream, &index->stream_instance_id); + ret = ustctl_get_instance_id(ustream, &stream_instance_id); if (ret < 0) { PERROR("ustctl_get_instance_id"); goto error; } - index->stream_instance_id = htobe64(index->stream_instance_id); - ret = ustctl_get_sequence_number(ustream, &index->packet_seq_num); + ret = ustctl_get_sequence_number(ustream, &packet_seq_num); if (ret < 0) { PERROR("ustctl_get_sequence_number"); goto error; } - index->packet_seq_num = htobe64(index->packet_seq_num); + + *index = (typeof(*index)) { + .offset = index->offset, + .packet_size = htobe64(packet_size), + .content_size = htobe64(content_size), + .timestamp_begin = htobe64(timestamp_begin), + .timestamp_end = htobe64(timestamp_end), + .events_discarded = htobe64(events_discarded), + .stream_id = htobe64(stream_id), + .stream_instance_id = htobe64(stream_instance_id), + .packet_seq_num = htobe64(packet_seq_num), + }; error: return ret; @@ -2241,6 +2337,13 @@ int commit_one_metadata_packet(struct lttng_consumer_stream *stream) stream->ust_metadata_pushed); ret = write_len; + /* + * Switch packet (but don't open the next one) on every commit of + * a metadata packet. Since the subbuffer is fully filled (with padding, + * if needed), the stream is "quiescent" after this commit. + */ + ustctl_flush_buffer(stream->ustream, 1); + stream->quiescent = true; end: pthread_mutex_unlock(&stream->chan->metadata_cache->lock); return ret; @@ -2273,10 +2376,10 @@ int lttng_ustconsumer_sync_metadata(struct lttng_consumer_local_data *ctx, * because we locked the metadata thread. */ ret = lttng_ustconsumer_request_metadata(ctx, metadata->chan, 0, 0); + pthread_mutex_lock(&metadata->lock); if (ret < 0) { goto end; } - pthread_mutex_lock(&metadata->lock); ret = commit_one_metadata_packet(metadata); if (ret <= 0) { @@ -2285,7 +2388,6 @@ int lttng_ustconsumer_sync_metadata(struct lttng_consumer_local_data *ctx, retry = 1; } - ustctl_flush_buffer(metadata->ustream, 1); ret = ustctl_snapshot(metadata->ustream); if (ret < 0) { if (errno != EAGAIN) { @@ -2475,7 +2577,6 @@ retry: if (ret <= 0) { goto end; } - ustctl_flush_buffer(stream->ustream, 1); goto retry; } @@ -2496,6 +2597,8 @@ retry: index.offset = htobe64(stream->out_fd_offset); ret = get_index_values(&index, ustream); if (ret < 0) { + err = ustctl_put_subbuf(ustream); + assert(err == 0); goto end; } @@ -2503,6 +2606,8 @@ retry: ret = update_stream_stats(stream); if (ret < 0) { PERROR("kernctl_get_events_discarded"); + err = ustctl_put_subbuf(ustream); + assert(err == 0); goto end; } } else { @@ -2620,14 +2725,18 @@ int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream *stream) stream->tracefile_size_current = 0; if (!stream->metadata_flag) { - ret = index_create_file(stream->chan->pathname, + struct lttng_index_file *index_file; + + index_file = lttng_index_file_create(stream->chan->pathname, stream->name, stream->uid, stream->gid, stream->chan->tracefile_size, - stream->tracefile_count_current); - if (ret < 0) { + stream->tracefile_count_current, + CTF_INDEX_MAJOR, CTF_INDEX_MINOR); + if (!index_file) { goto error; } - stream->index_fd = ret; + assert(!stream->index_file); + stream->index_file = index_file; } } ret = 0; @@ -2834,7 +2943,7 @@ int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx, request.key = channel->key; DBG("Sending metadata request to sessiond, session id %" PRIu64 - ", per-pid %" PRIu64 ", app UID %u and channek key %" PRIu64, + ", per-pid %" PRIu64 ", app UID %u and channel key %" PRIu64, request.session_id, request.session_id_per_pid, request.uid, request.key);