X-Git-Url: http://git.efficios.com/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fcommon%2Fconsumer%2Fconsumer.c;h=d70a5d1d406b758e1b02cc61f22b440ee006a7b1;hp=efafeddbc3543e0d26c5cd2f66f046663c5b8572;hb=fd424d99bfddf3581bc55fdf14799a50661beddd;hpb=fcf0f774e2b567e5dcf9190a84973b7a4e06eb7d diff --git a/src/common/consumer/consumer.c b/src/common/consumer/consumer.c index efafeddbc..d70a5d1d4 100644 --- a/src/common/consumer/consumer.c +++ b/src/common/consumer/consumer.c @@ -1,20 +1,10 @@ /* - * Copyright (C) 2011 - Julien Desfossez - * Mathieu Desnoyers - * 2012 - David Goulet + * Copyright (C) 2011 Julien Desfossez + * Copyright (C) 2011 Mathieu Desnoyers + * Copyright (C) 2012 David Goulet * - * 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, - * as published by the Free Software Foundation. + * SPDX-License-Identifier: GPL-2.0-only * - * This program is distributed in the hope that it will be useful, but WITHOUT - * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or - * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for - * more details. - * - * You should have received a copy of the GNU General Public License along - * with this program; if not, write to the Free Software Foundation, Inc., - * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. */ #define _LGPL_SOURCE @@ -52,6 +42,7 @@ #include #include #include +#include struct lttng_consumer_global_data consumer_data = { .stream_count = 0, @@ -90,6 +81,20 @@ int consumer_quit; static struct lttng_ht *metadata_ht; static struct lttng_ht *data_ht; +static const char *get_consumer_domain(void) +{ + switch (consumer_data.type) { + case LTTNG_CONSUMER_KERNEL: + return DEFAULT_KERNEL_TRACE_DIR; + case LTTNG_CONSUMER64_UST: + /* Fall-through. */ + case LTTNG_CONSUMER32_UST: + return DEFAULT_UST_TRACE_DIR; + default: + abort(); + } +} + /* * Notify a thread lttng pipe to poll back again. This usually means that some * global state has changed so we just send back the thread in a poll wait @@ -410,7 +415,8 @@ void consumer_del_channel(struct lttng_consumer_channel *channel) rcu_read_unlock(); } - call_rcu(&channel->node.head, free_channel_rcu); + channel->is_deleted = true; + call_rcu(&channel->node.head, free_channel_rcu); end: pthread_mutex_unlock(&channel->lock); pthread_mutex_unlock(&consumer_data.lock); @@ -601,6 +607,7 @@ struct lttng_consumer_stream *consumer_allocate_stream(uint64_t channel_key, stream->endpoint_status = CONSUMER_ENDPOINT_ACTIVE; stream->index_file = NULL; stream->last_sequence_number = -1ULL; + stream->rotate_position = -1ULL; pthread_mutex_init(&stream->lock, NULL); pthread_mutex_init(&stream->metadata_timer_lock, NULL); @@ -707,11 +714,6 @@ void consumer_add_data_stream(struct lttng_consumer_stream *stream) pthread_mutex_unlock(&consumer_data.lock); } -void consumer_del_data_stream(struct lttng_consumer_stream *stream) -{ - consumer_del_stream(stream, data_ht); -} - /* * Add relayd socket to global consumer data hashtable. RCU read side lock MUST * be acquired before calling this. @@ -818,7 +820,7 @@ int consumer_send_relayd_stream(struct lttng_consumer_stream *stream, /* Add stream on the relayd */ pthread_mutex_lock(&relayd->ctrl_sock_mutex); ret = relayd_add_stream(&relayd->control_sock, stream->name, - path, &stream->relayd_stream_id, + get_consumer_domain(), path, &stream->relayd_stream_id, stream->chan->tracefile_size, stream->chan->tracefile_count, stream->trace_chunk); @@ -936,6 +938,7 @@ static int write_relayd_stream_header(struct lttng_consumer_stream *stream, data_hdr.stream_id = htobe64(stream->relayd_stream_id); data_hdr.data_size = htobe32(data_size); data_hdr.padding_size = htobe32(padding); + /* * Note that net_seq_num below is assigned with the *current* value of * next_net_seq_num and only after that the next_net_seq_num will be @@ -1011,32 +1014,17 @@ int lttng_consumer_channel_set_trace_chunk( struct lttng_consumer_channel *channel, struct lttng_trace_chunk *new_trace_chunk) { - int ret = 0; - const bool is_local_trace = channel->relayd_id == -1ULL; - bool update_stream_trace_chunk; - struct cds_lfht_iter iter; - struct lttng_consumer_stream *stream; - unsigned long channel_hash; - pthread_mutex_lock(&channel->lock); - /* - * A stream can transition to a state where it and its channel - * no longer belong to a trace chunk. For instance, this happens when - * a session is rotated while it is inactive. After the rotation - * of an inactive session completes, the channel and its streams no - * longer belong to a trace chunk. - * - * However, if a session is stopped, rotated, and started again, - * the session daemon will create a new chunk and send it to its peers. - * In that case, the streams' transition to a new chunk can be performed - * immediately. - * - * This trace chunk transition could also be performed lazily when - * a buffer is consumed. However, creating the files here allows the - * consumer daemon to report any creation error to the session daemon - * and cause the start of the tracing session to fail. - */ - update_stream_trace_chunk = !channel->trace_chunk && new_trace_chunk; + if (channel->is_deleted) { + /* + * The channel has been logically deleted and should no longer + * be used. It has released its reference to its current trace + * chunk and should not acquire a new one. + * + * Return success as there is nothing for the caller to do. + */ + goto end; + } /* * The acquisition of the reference cannot fail (barring @@ -1052,59 +1040,9 @@ int lttng_consumer_channel_set_trace_chunk( lttng_trace_chunk_put(channel->trace_chunk); channel->trace_chunk = new_trace_chunk; - if (!is_local_trace || !new_trace_chunk) { - /* Not an error. */ - goto end; - } - - if (!update_stream_trace_chunk) { - goto end; - } - - channel_hash = consumer_data.stream_per_chan_id_ht->hash_fct( - &channel->key, lttng_ht_seed); - rcu_read_lock(); - cds_lfht_for_each_entry_duplicate(consumer_data.stream_per_chan_id_ht->ht, - channel_hash, - consumer_data.stream_per_chan_id_ht->match_fct, - &channel->key, &iter, stream, node_channel_id.node) { - bool acquired_reference, should_regenerate_metadata = false; - - acquired_reference = lttng_trace_chunk_get(channel->trace_chunk); - assert(acquired_reference); - - pthread_mutex_lock(&stream->lock); - - /* - * On a transition from "no-chunk" to a new chunk, a metadata - * stream's content must be entirely dumped. This must occcur - * _after_ the creation of the metadata stream's output files - * as the consumption thread (not necessarily the one executing - * this) may start to consume during the call to - * consumer_metadata_stream_dump(). - */ - should_regenerate_metadata = - stream->metadata_flag && - !stream->trace_chunk && channel->trace_chunk; - stream->trace_chunk = channel->trace_chunk; - ret = consumer_stream_create_output_files(stream, true); - if (ret) { - pthread_mutex_unlock(&stream->lock); - goto end_rcu_unlock; - } - if (should_regenerate_metadata) { - ret = consumer_metadata_stream_dump(stream); - } - pthread_mutex_unlock(&stream->lock); - if (ret) { - goto end_rcu_unlock; - } - } -end_rcu_unlock: - rcu_read_unlock(); end: pthread_mutex_unlock(&channel->lock); - return ret; + return 0; } /* @@ -1395,6 +1333,7 @@ void lttng_consumer_cleanup(void) { struct lttng_ht_iter iter; struct lttng_consumer_channel *channel; + unsigned int trace_chunks_left; rcu_read_lock(); @@ -1419,6 +1358,27 @@ void lttng_consumer_cleanup(void) */ lttng_ht_destroy(consumer_data.stream_list_ht); + /* + * Trace chunks in the registry may still exist if the session + * daemon has encountered an internal error and could not + * tear down its sessions and/or trace chunks properly. + * + * Release the session daemon's implicit reference to any remaining + * trace chunk and print an error if any trace chunk was found. Note + * that there are _no_ legitimate cases for trace chunks to be left, + * it is a leak. However, it can happen following a crash of the + * session daemon and not emptying the registry would cause an assertion + * to hit. + */ + trace_chunks_left = lttng_trace_chunk_registry_put_each_chunk( + consumer_data.chunk_registry); + if (trace_chunks_left) { + ERR("%u trace chunks are leaked by lttng-consumerd. " + "This can be caused by an internal error of the session daemon.", + trace_chunks_left); + } + /* Run all callbacks freeing each chunk. */ + rcu_barrier(); lttng_trace_chunk_registry_destroy(consumer_data.chunk_registry); } @@ -1710,23 +1670,24 @@ end: */ ssize_t lttng_consumer_on_read_subbuffer_mmap( struct lttng_consumer_local_data *ctx, - struct lttng_consumer_stream *stream, unsigned long len, + struct lttng_consumer_stream *stream, + const struct lttng_buffer_view *buffer, unsigned long padding, struct ctf_packet_index *index) { - unsigned long mmap_offset; - void *mmap_base; ssize_t ret = 0; off_t orig_offset = stream->out_fd_offset; /* Default is on the disk */ int outfd = stream->out_fd; struct consumer_relayd_sock_pair *relayd = NULL; unsigned int relayd_hang_up = 0; + const size_t subbuf_content_size = buffer->size - padding; + size_t write_len; /* RCU lock for the relayd pointer */ rcu_read_lock(); - - assert(stream->chan->trace_chunk); + assert(stream->net_seq_idx != (uint64_t) -1ULL || + stream->trace_chunk); /* Flag that the current stream if set for network streaming. */ if (stream->net_seq_idx != (uint64_t) -1ULL) { @@ -1737,39 +1698,9 @@ ssize_t lttng_consumer_on_read_subbuffer_mmap( } } - /* get the offset inside the fd to mmap */ - switch (consumer_data.type) { - case LTTNG_CONSUMER_KERNEL: - mmap_base = stream->mmap_base; - ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset); - if (ret < 0) { - PERROR("tracer ctl get_mmap_read_offset"); - goto end; - } - break; - case LTTNG_CONSUMER32_UST: - case LTTNG_CONSUMER64_UST: - mmap_base = lttng_ustctl_get_mmap_base(stream); - if (!mmap_base) { - ERR("read mmap get mmap base for stream %s", stream->name); - ret = -EPERM; - goto end; - } - ret = lttng_ustctl_get_mmap_read_offset(stream, &mmap_offset); - if (ret != 0) { - PERROR("tracer ctl get_mmap_read_offset"); - ret = -EINVAL; - goto end; - } - break; - default: - ERR("Unknown consumer_data type"); - assert(0); - } - /* Handle stream on the relayd if the output is on the network */ if (relayd) { - unsigned long netlen = len; + unsigned long netlen = subbuf_content_size; /* * Lock the control socket for the complete duration of the function @@ -1807,10 +1738,10 @@ ssize_t lttng_consumer_on_read_subbuffer_mmap( goto write_error; } } - } else { - /* No streaming, we have to set the len with the full padding */ - len += padding; + write_len = subbuf_content_size; + } else { + /* No streaming; we have to write the full padding. */ if (stream->metadata_flag && stream->reset_metadata_flag) { ret = utils_truncate_stream_file(stream->out_fd, 0); if (ret < 0) { @@ -1824,7 +1755,7 @@ ssize_t lttng_consumer_on_read_subbuffer_mmap( * Check if we need to change the tracefile before writing the packet. */ if (stream->chan->tracefile_size > 0 && - (stream->tracefile_size_current + len) > + (stream->tracefile_size_current + buffer->size) > stream->chan->tracefile_size) { ret = consumer_stream_rotate_output_files(stream); if (ret) { @@ -1833,19 +1764,21 @@ ssize_t lttng_consumer_on_read_subbuffer_mmap( outfd = stream->out_fd; orig_offset = 0; } - stream->tracefile_size_current += len; + stream->tracefile_size_current += buffer->size; if (index) { index->offset = htobe64(stream->out_fd_offset); } + + write_len = buffer->size; } /* * This call guarantee that len or less is returned. It's impossible to * receive a ret value that is bigger than len. */ - ret = lttng_write(outfd, mmap_base + mmap_offset, len); - DBG("Consumer mmap write() ret %zd (len %lu)", ret, len); - if (ret < 0 || ((size_t) ret != len)) { + ret = lttng_write(outfd, buffer->data, write_len); + DBG("Consumer mmap write() ret %zd (len %lu)", ret, write_len); + if (ret < 0 || ((size_t) ret != write_len)) { /* * Report error to caller if nothing was written else at least send the * amount written. @@ -1866,7 +1799,8 @@ ssize_t lttng_consumer_on_read_subbuffer_mmap( DBG("Consumer mmap write detected relayd hang up"); } else { /* Unhandled error, print it and stop function right now. */ - PERROR("Error in write mmap (ret %zd != len %lu)", ret, len); + PERROR("Error in write mmap (ret %zd != write_len %zu)", ret, + write_len); } goto write_error; } @@ -1875,9 +1809,9 @@ ssize_t lttng_consumer_on_read_subbuffer_mmap( /* This call is useless on a socket so better save a syscall. */ if (!relayd) { /* This won't block, but will start writeout asynchronously */ - lttng_sync_file_range(outfd, stream->out_fd_offset, len, + lttng_sync_file_range(outfd, stream->out_fd_offset, write_len, SYNC_FILE_RANGE_WRITE); - stream->out_fd_offset += len; + stream->out_fd_offset += write_len; lttng_consumer_sync_trace_file(stream, orig_offset); } @@ -2219,6 +2153,7 @@ int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx, } } +static void lttng_consumer_close_all_metadata(void) { switch (consumer_data.type) { @@ -2251,7 +2186,8 @@ void lttng_consumer_close_all_metadata(void) void consumer_del_metadata_stream(struct lttng_consumer_stream *stream, struct lttng_ht *ht) { - struct lttng_consumer_channel *free_chan = NULL; + struct lttng_consumer_channel *channel = NULL; + bool free_channel = false; assert(stream); /* @@ -2263,11 +2199,17 @@ void consumer_del_metadata_stream(struct lttng_consumer_stream *stream, DBG3("Consumer delete metadata stream %d", stream->wait_fd); pthread_mutex_lock(&consumer_data.lock); - pthread_mutex_lock(&stream->chan->lock); + /* + * Note that this assumes that a stream's channel is never changed and + * that the stream's lock doesn't need to be taken to sample its + * channel. + */ + channel = stream->chan; + pthread_mutex_lock(&channel->lock); pthread_mutex_lock(&stream->lock); - if (stream->chan->metadata_cache) { + if (channel->metadata_cache) { /* Only applicable to userspace consumers. */ - pthread_mutex_lock(&stream->chan->metadata_cache->lock); + pthread_mutex_lock(&channel->metadata_cache->lock); } /* Remove any reference to that stream. */ @@ -2279,28 +2221,29 @@ void consumer_del_metadata_stream(struct lttng_consumer_stream *stream, consumer_stream_destroy_buffers(stream); /* Atomically decrement channel refcount since other threads can use it. */ - if (!uatomic_sub_return(&stream->chan->refcount, 1) - && !uatomic_read(&stream->chan->nb_init_stream_left)) { + if (!uatomic_sub_return(&channel->refcount, 1) + && !uatomic_read(&channel->nb_init_stream_left)) { /* Go for channel deletion! */ - free_chan = stream->chan; + free_channel = true; } + stream->chan = NULL; /* * Nullify the stream reference so it is not used after deletion. The * channel lock MUST be acquired before being able to check for a NULL * pointer value. */ - stream->chan->metadata_stream = NULL; + channel->metadata_stream = NULL; - if (stream->chan->metadata_cache) { - pthread_mutex_unlock(&stream->chan->metadata_cache->lock); + if (channel->metadata_cache) { + pthread_mutex_unlock(&channel->metadata_cache->lock); } pthread_mutex_unlock(&stream->lock); - pthread_mutex_unlock(&stream->chan->lock); + pthread_mutex_unlock(&channel->lock); pthread_mutex_unlock(&consumer_data.lock); - if (free_chan) { - consumer_del_channel(free_chan); + if (free_channel) { + consumer_del_channel(channel); } lttng_trace_chunk_put(stream->trace_chunk); @@ -3975,15 +3918,33 @@ int consumer_flush_buffer(struct lttng_consumer_stream *stream, int producer_act switch (consumer_data.type) { case LTTNG_CONSUMER_KERNEL: - ret = kernctl_buffer_flush(stream->wait_fd); - if (ret < 0) { - ERR("Failed to flush kernel stream"); - goto end; + if (producer_active) { + ret = kernctl_buffer_flush(stream->wait_fd); + if (ret < 0) { + ERR("Failed to flush kernel stream"); + goto end; + } + } else { + ret = kernctl_buffer_flush_empty(stream->wait_fd); + if (ret < 0) { + /* + * Doing a buffer flush which does not take into + * account empty packets. This is not perfect, + * but required as a fall-back when + * "flush_empty" is not implemented by + * lttng-modules. + */ + ret = kernctl_buffer_flush(stream->wait_fd); + if (ret < 0) { + ERR("Failed to flush kernel stream"); + goto end; + } + } } break; case LTTNG_CONSUMER32_UST: case LTTNG_CONSUMER64_UST: - lttng_ustctl_flush_buffer(stream, producer_active); + lttng_ustconsumer_flush_buffer(stream, producer_active); break; default: ERR("Unknown consumer_data type"); @@ -4011,18 +3972,34 @@ int lttng_consumer_rotate_channel(struct lttng_consumer_channel *channel, struct lttng_consumer_stream *stream; struct lttng_ht_iter iter; struct lttng_ht *ht = consumer_data.stream_per_chan_id_ht; + struct lttng_dynamic_array stream_rotation_positions; + uint64_t next_chunk_id, stream_count = 0; + enum lttng_trace_chunk_status chunk_status; + const bool is_local_trace = relayd_id == -1ULL; + struct consumer_relayd_sock_pair *relayd = NULL; + bool rotating_to_new_chunk = true; DBG("Consumer sample rotate position for channel %" PRIu64, key); + lttng_dynamic_array_init(&stream_rotation_positions, + sizeof(struct relayd_stream_rotation_position), NULL); + rcu_read_lock(); pthread_mutex_lock(&channel->lock); + assert(channel->trace_chunk); + chunk_status = lttng_trace_chunk_get_id(channel->trace_chunk, + &next_chunk_id); + if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { + ret = -1; + goto end_unlock_channel; + } cds_lfht_for_each_entry_duplicate(ht->ht, ht->hash_fct(&channel->key, lttng_ht_seed), ht->match_fct, &channel->key, &iter.iter, stream, node_channel_id.node) { - unsigned long consumed_pos; + unsigned long produced_pos = 0, consumed_pos = 0; health_code_update(); @@ -4031,101 +4008,319 @@ int lttng_consumer_rotate_channel(struct lttng_consumer_channel *channel, */ pthread_mutex_lock(&stream->lock); - ret = lttng_consumer_sample_snapshot_positions(stream); - if (ret < 0) { - ERR("Failed to sample snapshot position during channel rotation"); - goto end_unlock_stream; + if (stream->trace_chunk == stream->chan->trace_chunk) { + rotating_to_new_chunk = false; } - ret = lttng_consumer_get_produced_snapshot(stream, - &stream->rotate_position); - if (ret < 0) { - ERR("Failed to sample produced position during channel rotation"); + /* + * Do not flush an empty packet when rotating from a NULL trace + * chunk. The stream has no means to output data, and the prior + * rotation which rotated to NULL performed that side-effect already. + */ + if (stream->trace_chunk) { + /* + * For metadata stream, do an active flush, which does not + * produce empty packets. For data streams, empty-flush; + * ensures we have at least one packet in each stream per trace + * chunk, even if no data was produced. + */ + ret = consumer_flush_buffer(stream, stream->metadata_flag ? 1 : 0); + if (ret < 0) { + ERR("Failed to flush stream %" PRIu64 " during channel rotation", + stream->key); + goto end_unlock_stream; + } + } + + ret = lttng_consumer_take_snapshot(stream); + if (ret < 0 && ret != -ENODATA && ret != -EAGAIN) { + ERR("Failed to sample snapshot position during channel rotation"); goto end_unlock_stream; } + if (!ret) { + ret = lttng_consumer_get_produced_snapshot(stream, + &produced_pos); + if (ret < 0) { + ERR("Failed to sample produced position during channel rotation"); + goto end_unlock_stream; + } - lttng_consumer_get_consumed_snapshot(stream, - &consumed_pos); - if (consumed_pos == stream->rotate_position) { + ret = lttng_consumer_get_consumed_snapshot(stream, + &consumed_pos); + if (ret < 0) { + ERR("Failed to sample consumed position during channel rotation"); + goto end_unlock_stream; + } + } + /* + * Align produced position on the start-of-packet boundary of the first + * packet going into the next trace chunk. + */ + produced_pos = ALIGN_FLOOR(produced_pos, stream->max_sb_size); + if (consumed_pos == produced_pos) { + DBG("Set rotate ready for stream %" PRIu64 " produced = %lu consumed = %lu", + stream->key, produced_pos, consumed_pos); stream->rotate_ready = true; + } else { + DBG("Different consumed and produced positions " + "for stream %" PRIu64 " produced = %lu consumed = %lu", + stream->key, produced_pos, consumed_pos); } - - ret = consumer_flush_buffer(stream, 1); - if (ret < 0) { - ERR("Failed to flush stream %" PRIu64 " during channel rotation", + /* + * The rotation position is based on the packet_seq_num of the + * packet following the last packet that was consumed for this + * stream, incremented by the offset between produced and + * consumed positions. This rotation position is a lower bound + * (inclusive) at which the next trace chunk starts. Since it + * is a lower bound, it is OK if the packet_seq_num does not + * correspond exactly to the same packet identified by the + * consumed_pos, which can happen in overwrite mode. + */ + if (stream->sequence_number_unavailable) { + /* + * Rotation should never be performed on a session which + * interacts with a pre-2.8 lttng-modules, which does + * not implement packet sequence number. + */ + ERR("Failure to rotate stream %" PRIu64 ": sequence number unavailable", stream->key); + ret = -1; goto end_unlock_stream; } + stream->rotate_position = stream->last_sequence_number + 1 + + ((produced_pos - consumed_pos) / stream->max_sb_size); + DBG("Set rotation position for stream %" PRIu64 " at position %" PRIu64, + stream->key, stream->rotate_position); + if (!is_local_trace) { + /* + * The relay daemon control protocol expects a rotation + * position as "the sequence number of the first packet + * _after_ the current trace chunk". + */ + const struct relayd_stream_rotation_position position = { + .stream_id = stream->relayd_stream_id, + .rotate_at_seq_num = stream->rotate_position, + }; + + ret = lttng_dynamic_array_add_element( + &stream_rotation_positions, + &position); + if (ret) { + ERR("Failed to allocate stream rotation position"); + goto end_unlock_stream; + } + stream_count++; + } pthread_mutex_unlock(&stream->lock); } + stream = NULL; pthread_mutex_unlock(&channel->lock); + if (is_local_trace) { + ret = 0; + goto end; + } + + relayd = consumer_find_relayd(relayd_id); + if (!relayd) { + ERR("Failed to find relayd %" PRIu64, relayd_id); + ret = -1; + goto end; + } + + pthread_mutex_lock(&relayd->ctrl_sock_mutex); + ret = relayd_rotate_streams(&relayd->control_sock, stream_count, + rotating_to_new_chunk ? &next_chunk_id : NULL, + (const struct relayd_stream_rotation_position *) + stream_rotation_positions.buffer.data); + pthread_mutex_unlock(&relayd->ctrl_sock_mutex); + if (ret < 0) { + ERR("Relayd rotate stream failed. Cleaning up relayd %" PRIu64, + relayd->net_seq_idx); + lttng_consumer_cleanup_relayd(relayd); + goto end; + } + ret = 0; goto end; end_unlock_stream: pthread_mutex_unlock(&stream->lock); +end_unlock_channel: pthread_mutex_unlock(&channel->lock); end: rcu_read_unlock(); + lttng_dynamic_array_reset(&stream_rotation_positions); return ret; } -/* - * Check if a stream is ready to be rotated after extracting it. - * - * Return 1 if it is ready for rotation, 0 if it is not, a negative value on - * error. Stream lock must be held. - */ -int lttng_consumer_stream_is_rotate_ready(struct lttng_consumer_stream *stream) +static +int consumer_clear_buffer(struct lttng_consumer_stream *stream) { - int ret; - unsigned long consumed_pos; + int ret = 0; + unsigned long consumed_pos_before, consumed_pos_after; - if (!stream->rotate_position && !stream->rotate_ready) { - ret = 0; + ret = lttng_consumer_sample_snapshot_positions(stream); + if (ret < 0) { + ERR("Taking snapshot positions"); goto end; } - if (stream->rotate_ready) { - ret = 1; + ret = lttng_consumer_get_consumed_snapshot(stream, &consumed_pos_before); + if (ret < 0) { + ERR("Consumed snapshot position"); goto end; } - /* - * If we don't have the rotate_ready flag, check the consumed position - * to determine if we need to rotate. - */ + switch (consumer_data.type) { + case LTTNG_CONSUMER_KERNEL: + ret = kernctl_buffer_clear(stream->wait_fd); + if (ret < 0) { + ERR("Failed to clear kernel stream (ret = %d)", ret); + goto end; + } + break; + case LTTNG_CONSUMER32_UST: + case LTTNG_CONSUMER64_UST: + lttng_ustconsumer_clear_buffer(stream); + break; + default: + ERR("Unknown consumer_data type"); + abort(); + } + ret = lttng_consumer_sample_snapshot_positions(stream); if (ret < 0) { ERR("Taking snapshot positions"); goto end; } - - ret = lttng_consumer_get_consumed_snapshot(stream, &consumed_pos); + ret = lttng_consumer_get_consumed_snapshot(stream, &consumed_pos_after); if (ret < 0) { ERR("Consumed snapshot position"); goto end; } + DBG("clear: before: %lu after: %lu", consumed_pos_before, consumed_pos_after); +end: + return ret; +} + +static +int consumer_clear_stream(struct lttng_consumer_stream *stream) +{ + int ret; + + ret = consumer_flush_buffer(stream, 1); + if (ret < 0) { + ERR("Failed to flush stream %" PRIu64 " during channel clear", + stream->key); + ret = LTTCOMM_CONSUMERD_FATAL; + goto error; + } - /* Rotate position not reached yet (with check for overflow). */ - if ((long) (consumed_pos - stream->rotate_position) < 0) { - ret = 0; - goto end; + ret = consumer_clear_buffer(stream); + if (ret < 0) { + ERR("Failed to clear stream %" PRIu64 " during channel clear", + stream->key); + ret = LTTCOMM_CONSUMERD_FATAL; + goto error; } - ret = 1; -end: + ret = LTTCOMM_CONSUMERD_SUCCESS; +error: + return ret; +} + +static +int consumer_clear_unmonitored_channel(struct lttng_consumer_channel *channel) +{ + int ret; + struct lttng_consumer_stream *stream; + + rcu_read_lock(); + pthread_mutex_lock(&channel->lock); + cds_list_for_each_entry(stream, &channel->streams.head, send_node) { + health_code_update(); + pthread_mutex_lock(&stream->lock); + ret = consumer_clear_stream(stream); + if (ret) { + goto error_unlock; + } + pthread_mutex_unlock(&stream->lock); + } + pthread_mutex_unlock(&channel->lock); + rcu_read_unlock(); + return 0; + +error_unlock: + pthread_mutex_unlock(&stream->lock); + pthread_mutex_unlock(&channel->lock); + rcu_read_unlock(); return ret; } +/* + * Check if a stream is ready to be rotated after extracting it. + * + * Return 1 if it is ready for rotation, 0 if it is not, a negative value on + * error. Stream lock must be held. + */ +int lttng_consumer_stream_is_rotate_ready(struct lttng_consumer_stream *stream) +{ + DBG("Check is rotate ready for stream %" PRIu64 + " ready %u rotate_position %" PRIu64 + " last_sequence_number %" PRIu64, + stream->key, stream->rotate_ready, + stream->rotate_position, stream->last_sequence_number); + if (stream->rotate_ready) { + return 1; + } + + /* + * If packet seq num is unavailable, it means we are interacting + * with a pre-2.8 lttng-modules which does not implement the + * sequence number. Rotation should never be used by sessiond in this + * scenario. + */ + if (stream->sequence_number_unavailable) { + ERR("Internal error: rotation used on stream %" PRIu64 + " with unavailable sequence number", + stream->key); + return -1; + } + + if (stream->rotate_position == -1ULL || + stream->last_sequence_number == -1ULL) { + return 0; + } + + /* + * Rotate position not reached yet. The stream rotate position is + * the position of the next packet belonging to the next trace chunk, + * but consumerd considers rotation ready when reaching the last + * packet of the current chunk, hence the "rotate_position - 1". + */ + + DBG("Check is rotate ready for stream %" PRIu64 + " last_sequence_number %" PRIu64 + " rotate_position %" PRIu64, + stream->key, stream->last_sequence_number, + stream->rotate_position); + if (stream->last_sequence_number >= stream->rotate_position - 1) { + return 1; + } + + return 0; +} + /* * Reset the state for a stream after a rotation occurred. */ void lttng_consumer_reset_stream_rotate_state(struct lttng_consumer_stream *stream) { - stream->rotate_position = 0; + DBG("lttng_consumer_reset_stream_rotate_state for stream %" PRIu64, + stream->key); + stream->rotate_position = -1ULL; stream->rotate_ready = false; } @@ -4167,52 +4362,6 @@ end: return ret; } -/* - * Perform the rotation a stream file on the relay. - */ -int rotate_relay_stream(struct lttng_consumer_local_data *ctx, - struct lttng_consumer_stream *stream) -{ - int ret; - struct consumer_relayd_sock_pair *relayd; - uint64_t chunk_id; - enum lttng_trace_chunk_status chunk_status; - - DBG("Rotate relay stream"); - relayd = consumer_find_relayd(stream->net_seq_idx); - if (!relayd) { - ERR("Failed to find relayd"); - ret = -1; - goto end; - } - - chunk_status = lttng_trace_chunk_get_id(stream->chan->trace_chunk, - &chunk_id); - if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { - ERR("Failed to retrieve the id of the current trace chunk of channel \"%s\"", - stream->chan->name); - ret = -1; - goto end; - } - - pthread_mutex_lock(&relayd->ctrl_sock_mutex); - ret = relayd_rotate_stream(&relayd->control_sock, - stream->relayd_stream_id, - chunk_id, - stream->last_sequence_number); - pthread_mutex_unlock(&relayd->ctrl_sock_mutex); - if (ret < 0) { - ERR("Relayd rotate stream failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); - lttng_consumer_cleanup_relayd(relayd); - } - if (ret) { - ERR("Rotate relay stream"); - } - -end: - return ret; -} - /* * Performs the stream rotation for the rotate session feature if needed. * It must be called with the channel and stream locks held. @@ -4254,14 +4403,12 @@ int lttng_consumer_rotate_stream(struct lttng_consumer_local_data *ctx, stream->trace_chunk = stream->chan->trace_chunk; } - if (stream->net_seq_idx != (uint64_t) -1ULL) { - ret = rotate_relay_stream(ctx, stream); - } else { + if (stream->net_seq_idx == (uint64_t) -1ULL) { ret = rotate_local_stream(ctx, stream); - } - if (ret < 0) { - ERR("Failed to rotate stream, ret = %i", ret); - goto error; + if (ret < 0) { + ERR("Failed to rotate stream, ret = %i", ret); + goto error; + } } if (stream->metadata_flag && stream->trace_chunk) { @@ -4348,7 +4495,7 @@ enum lttcomm_return_code lttng_consumer_init_command( const lttng_uuid sessiond_uuid) { enum lttcomm_return_code ret; - char uuid_str[UUID_STR_LEN]; + char uuid_str[LTTNG_UUID_STR_LEN]; if (ctx->sessiond_uuid.is_set) { ret = LTTCOMM_CONSUMERD_ALREADY_SET; @@ -4374,7 +4521,7 @@ enum lttcomm_return_code lttng_consumer_create_trace_chunk( { int ret; enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS; - struct lttng_trace_chunk *created_chunk, *published_chunk; + struct lttng_trace_chunk *created_chunk = NULL, *published_chunk = NULL; enum lttng_trace_chunk_status chunk_status; char relayd_id_buffer[MAX_INT_DEC_LEN(*relayd_id)]; char creation_timestamp_buffer[ISO8601_STR_LEN]; @@ -4425,11 +4572,11 @@ enum lttcomm_return_code lttng_consumer_create_trace_chunk( * and LTTNG_CONSUMER_DESTROY_TRACE_CHUNK commands. */ created_chunk = lttng_trace_chunk_create(chunk_id, - chunk_creation_timestamp); + chunk_creation_timestamp, NULL); if (!created_chunk) { ERR("Failed to create trace chunk"); ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED; - goto end; + goto error; } if (chunk_override_name) { @@ -4437,7 +4584,7 @@ enum lttcomm_return_code lttng_consumer_create_trace_chunk( chunk_override_name); if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED; - goto end; + goto error; } } @@ -4447,7 +4594,7 @@ enum lttcomm_return_code lttng_consumer_create_trace_chunk( if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { ERR("Failed to set trace chunk credentials"); ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED; - goto end; + goto error; } /* * The consumer daemon has no ownership of the chunk output @@ -4455,10 +4602,11 @@ enum lttcomm_return_code lttng_consumer_create_trace_chunk( */ chunk_status = lttng_trace_chunk_set_as_user(created_chunk, chunk_directory_handle); + chunk_directory_handle = NULL; if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { ERR("Failed to set trace chunk's directory handle"); ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED; - goto end; + goto error; } } @@ -4470,7 +4618,7 @@ enum lttcomm_return_code lttng_consumer_create_trace_chunk( if (!published_chunk) { ERR("Failed to publish trace chunk"); ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED; - goto end; + goto error; } rcu_read_lock(); @@ -4493,11 +4641,13 @@ enum lttcomm_return_code lttng_consumer_create_trace_chunk( * channels. */ enum lttcomm_return_code close_ret; + char path[LTTNG_PATH_MAX]; DBG("Failed to set new trace chunk on existing channels, rolling back"); close_ret = lttng_consumer_close_trace_chunk(relayd_id, session_id, chunk_id, - chunk_creation_timestamp); + chunk_creation_timestamp, NULL, + path); if (close_ret != LTTCOMM_CONSUMERD_SUCCESS) { ERR("Failed to roll-back the creation of new chunk: session_id = %" PRIu64 ", chunk_id = %" PRIu64, session_id, chunk_id); @@ -4507,22 +4657,59 @@ enum lttcomm_return_code lttng_consumer_create_trace_chunk( break; } } - rcu_read_unlock(); + if (relayd_id) { + struct consumer_relayd_sock_pair *relayd; + + relayd = consumer_find_relayd(*relayd_id); + if (relayd) { + pthread_mutex_lock(&relayd->ctrl_sock_mutex); + ret = relayd_create_trace_chunk( + &relayd->control_sock, published_chunk); + pthread_mutex_unlock(&relayd->ctrl_sock_mutex); + } else { + ERR("Failed to find relay daemon socket: relayd_id = %" PRIu64, *relayd_id); + } + + if (!relayd || ret) { + enum lttcomm_return_code close_ret; + char path[LTTNG_PATH_MAX]; + + close_ret = lttng_consumer_close_trace_chunk(relayd_id, + session_id, + chunk_id, + chunk_creation_timestamp, + NULL, path); + if (close_ret != LTTCOMM_CONSUMERD_SUCCESS) { + ERR("Failed to roll-back the creation of new chunk: session_id = %" PRIu64 ", chunk_id = %" PRIu64, + session_id, + chunk_id); + } + + ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED; + goto error_unlock; + } + } +error_unlock: + rcu_read_unlock(); +error: /* Release the reference returned by the "publish" operation. */ lttng_trace_chunk_put(published_chunk); -end: + lttng_trace_chunk_put(created_chunk); return ret_code; } enum lttcomm_return_code lttng_consumer_close_trace_chunk( const uint64_t *relayd_id, uint64_t session_id, - uint64_t chunk_id, time_t chunk_close_timestamp) + uint64_t chunk_id, time_t chunk_close_timestamp, + const enum lttng_trace_chunk_command_type *close_command, + char *path) { enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS; struct lttng_trace_chunk *chunk; char relayd_id_buffer[MAX_INT_DEC_LEN(*relayd_id)]; const char *relayd_id_str = "(none)"; + const char *close_command_name = "none"; struct lttng_ht_iter iter; struct lttng_consumer_channel *channel; enum lttng_trace_chunk_status chunk_status; @@ -4538,16 +4725,21 @@ enum lttcomm_return_code lttng_consumer_close_trace_chunk( } else { relayd_id_str = "(formatting error)"; } - } + } + if (close_command) { + close_command_name = lttng_trace_chunk_command_type_get_name( + *close_command); + } DBG("Consumer close trace chunk command: relayd_id = %s" - ", session_id = %" PRIu64 - ", chunk_id = %" PRIu64, relayd_id_str, - session_id, chunk_id); + ", session_id = %" PRIu64 ", chunk_id = %" PRIu64 + ", close command = %s", + relayd_id_str, session_id, chunk_id, + close_command_name); + chunk = lttng_trace_chunk_registry_find_chunk( - consumer_data.chunk_registry, session_id, - chunk_id); - if (!chunk) { + consumer_data.chunk_registry, session_id, chunk_id); + if (!chunk) { ERR("Failed to find chunk: session_id = %" PRIu64 ", chunk_id = %" PRIu64, session_id, chunk_id); @@ -4561,12 +4753,15 @@ enum lttcomm_return_code lttng_consumer_close_trace_chunk( ret_code = LTTCOMM_CONSUMERD_CLOSE_TRACE_CHUNK_FAILED; goto end; } - /* - * Release the reference returned by the "find" operation and - * the session daemon's implicit reference to the chunk. - */ - lttng_trace_chunk_put(chunk); - lttng_trace_chunk_put(chunk); + + if (close_command) { + chunk_status = lttng_trace_chunk_set_close_command( + chunk, *close_command); + if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { + ret_code = LTTCOMM_CONSUMERD_CLOSE_TRACE_CHUNK_FAILED; + goto end; + } + } /* * chunk is now invalid to access as we no longer hold a reference to @@ -4597,8 +4792,38 @@ enum lttcomm_return_code lttng_consumer_close_trace_chunk( ret_code = LTTCOMM_CONSUMERD_CLOSE_TRACE_CHUNK_FAILED; } } + + if (relayd_id) { + int ret; + struct consumer_relayd_sock_pair *relayd; + + relayd = consumer_find_relayd(*relayd_id); + if (relayd) { + pthread_mutex_lock(&relayd->ctrl_sock_mutex); + ret = relayd_close_trace_chunk( + &relayd->control_sock, chunk, + path); + pthread_mutex_unlock(&relayd->ctrl_sock_mutex); + } else { + ERR("Failed to find relay daemon socket: relayd_id = %" PRIu64, + *relayd_id); + } + + if (!relayd || ret) { + ret_code = LTTCOMM_CONSUMERD_CLOSE_TRACE_CHUNK_FAILED; + goto error_unlock; + } + } +error_unlock: rcu_read_unlock(); end: + /* + * Release the reference returned by the "find" operation and + * the session daemon's implicit reference to the chunk. + */ + lttng_trace_chunk_put(chunk); + lttng_trace_chunk_put(chunk); + return ret_code; } @@ -4606,10 +4831,13 @@ enum lttcomm_return_code lttng_consumer_trace_chunk_exists( const uint64_t *relayd_id, uint64_t session_id, uint64_t chunk_id) { + int ret; enum lttcomm_return_code ret_code; - struct lttng_trace_chunk *chunk; char relayd_id_buffer[MAX_INT_DEC_LEN(*relayd_id)]; const char *relayd_id_str = "(none)"; + const bool is_local_trace = !relayd_id; + struct consumer_relayd_sock_pair *relayd = NULL; + bool chunk_exists_local, chunk_exists_remote; if (relayd_id) { int ret; @@ -4625,16 +4853,117 @@ enum lttcomm_return_code lttng_consumer_trace_chunk_exists( } DBG("Consumer trace chunk exists command: relayd_id = %s" - ", session_id = %" PRIu64 ", chunk_id = %" PRIu64, relayd_id_str, - session_id, chunk_id); - chunk = lttng_trace_chunk_registry_find_chunk( - consumer_data.chunk_registry, session_id, chunk_id); - DBG("Trace chunk %s locally", chunk ? "exists" : "does not exist"); - ret_code = chunk ? LTTCOMM_CONSUMERD_TRACE_CHUNK_EXISTS_LOCAL : + ret = lttng_trace_chunk_registry_chunk_exists( + consumer_data.chunk_registry, session_id, + chunk_id, &chunk_exists_local); + if (ret) { + /* Internal error. */ + ERR("Failed to query the existence of a trace chunk"); + ret_code = LTTCOMM_CONSUMERD_FATAL; + goto end; + } + DBG("Trace chunk %s locally", + chunk_exists_local ? "exists" : "does not exist"); + if (chunk_exists_local) { + ret_code = LTTCOMM_CONSUMERD_TRACE_CHUNK_EXISTS_LOCAL; + goto end; + } else if (is_local_trace) { + ret_code = LTTCOMM_CONSUMERD_UNKNOWN_TRACE_CHUNK; + goto end; + } + + rcu_read_lock(); + relayd = consumer_find_relayd(*relayd_id); + if (!relayd) { + ERR("Failed to find relayd %" PRIu64, *relayd_id); + ret_code = LTTCOMM_CONSUMERD_INVALID_PARAMETERS; + goto end_rcu_unlock; + } + DBG("Looking up existence of trace chunk on relay daemon"); + pthread_mutex_lock(&relayd->ctrl_sock_mutex); + ret = relayd_trace_chunk_exists(&relayd->control_sock, chunk_id, + &chunk_exists_remote); + pthread_mutex_unlock(&relayd->ctrl_sock_mutex); + if (ret < 0) { + ERR("Failed to look-up the existence of trace chunk on relay daemon"); + ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL; + goto end_rcu_unlock; + } + + ret_code = chunk_exists_remote ? + LTTCOMM_CONSUMERD_TRACE_CHUNK_EXISTS_REMOTE : LTTCOMM_CONSUMERD_UNKNOWN_TRACE_CHUNK; + DBG("Trace chunk %s on relay daemon", + chunk_exists_remote ? "exists" : "does not exist"); - lttng_trace_chunk_put(chunk); +end_rcu_unlock: + rcu_read_unlock(); +end: return ret_code; } + +static +int consumer_clear_monitored_channel(struct lttng_consumer_channel *channel) +{ + struct lttng_ht *ht; + struct lttng_consumer_stream *stream; + struct lttng_ht_iter iter; + int ret; + + ht = consumer_data.stream_per_chan_id_ht; + + rcu_read_lock(); + cds_lfht_for_each_entry_duplicate(ht->ht, + ht->hash_fct(&channel->key, lttng_ht_seed), + ht->match_fct, &channel->key, + &iter.iter, stream, node_channel_id.node) { + /* + * Protect against teardown with mutex. + */ + pthread_mutex_lock(&stream->lock); + if (cds_lfht_is_node_deleted(&stream->node.node)) { + goto next; + } + ret = consumer_clear_stream(stream); + if (ret) { + goto error_unlock; + } + next: + pthread_mutex_unlock(&stream->lock); + } + rcu_read_unlock(); + return LTTCOMM_CONSUMERD_SUCCESS; + +error_unlock: + pthread_mutex_unlock(&stream->lock); + rcu_read_unlock(); + return ret; +} + +int lttng_consumer_clear_channel(struct lttng_consumer_channel *channel) +{ + int ret; + + DBG("Consumer clear channel %" PRIu64, channel->key); + + if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA) { + /* + * Nothing to do for the metadata channel/stream. + * Snapshot mechanism already take care of the metadata + * handling/generation, and monitored channels only need to + * have their data stream cleared.. + */ + ret = LTTCOMM_CONSUMERD_SUCCESS; + goto end; + } + + if (!channel->monitor) { + ret = consumer_clear_unmonitored_channel(channel); + } else { + ret = consumer_clear_monitored_channel(channel); + } +end: + return ret; +}