X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fcommon%2Fconsumer%2Fconsumer.c;h=cbd99e7ec3ab08146142706383085b2933e314ba;hp=8483146c65ab6ecf3d1b026d1f33c62be4442c8a;hb=e66d26f5110d65c1f9549904a33986797a3759a4;hpb=948411cd7aa5e2327177c2e4ed8be8dd65c8504c diff --git a/src/common/consumer/consumer.c b/src/common/consumer/consumer.c index 8483146c6..cbd99e7ec 100644 --- a/src/common/consumer/consumer.c +++ b/src/common/consumer/consumer.c @@ -1,22 +1,13 @@ /* - * 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. */ +#include "common/index/ctf-index.h" #define _LGPL_SOURCE #include #include @@ -91,6 +82,20 @@ 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 @@ -411,7 +416,8 @@ 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); @@ -563,94 +569,6 @@ void consumer_stream_update_channel_attributes( channel->tracefile_size; } -struct lttng_consumer_stream *consumer_allocate_stream(uint64_t channel_key, - uint64_t stream_key, - const char *channel_name, - uint64_t relayd_id, - uint64_t session_id, - struct lttng_trace_chunk *trace_chunk, - int cpu, - int *alloc_ret, - enum consumer_channel_type type, - unsigned int monitor) -{ - int ret; - struct lttng_consumer_stream *stream; - - stream = zmalloc(sizeof(*stream)); - if (stream == NULL) { - PERROR("malloc struct lttng_consumer_stream"); - ret = -ENOMEM; - goto end; - } - - if (trace_chunk && !lttng_trace_chunk_get(trace_chunk)) { - ERR("Failed to acquire trace chunk reference during the creation of a stream"); - ret = -1; - goto error; - } - - rcu_read_lock(); - stream->key = stream_key; - stream->trace_chunk = trace_chunk; - stream->out_fd = -1; - stream->out_fd_offset = 0; - stream->output_written = 0; - stream->net_seq_idx = relayd_id; - stream->session_id = session_id; - stream->monitor = monitor; - stream->endpoint_status = CONSUMER_ENDPOINT_ACTIVE; - stream->index_file = NULL; - stream->last_sequence_number = -1ULL; - pthread_mutex_init(&stream->lock, NULL); - pthread_mutex_init(&stream->metadata_timer_lock, NULL); - - /* If channel is the metadata, flag this stream as metadata. */ - if (type == CONSUMER_CHANNEL_TYPE_METADATA) { - stream->metadata_flag = 1; - /* Metadata is flat out. */ - strncpy(stream->name, DEFAULT_METADATA_NAME, sizeof(stream->name)); - /* Live rendez-vous point. */ - pthread_cond_init(&stream->metadata_rdv, NULL); - pthread_mutex_init(&stream->metadata_rdv_lock, NULL); - } else { - /* Format stream name to _ */ - ret = snprintf(stream->name, sizeof(stream->name), "%s_%d", - channel_name, cpu); - if (ret < 0) { - PERROR("snprintf stream name"); - goto error; - } - } - - /* Key is always the wait_fd for streams. */ - lttng_ht_node_init_u64(&stream->node, stream->key); - - /* Init node per channel id key */ - lttng_ht_node_init_u64(&stream->node_channel_id, channel_key); - - /* Init session id node with the stream session id */ - lttng_ht_node_init_u64(&stream->node_session_id, stream->session_id); - - DBG3("Allocated stream %s (key %" PRIu64 ", chan_key %" PRIu64 - " relayd_id %" PRIu64 ", session_id %" PRIu64, - stream->name, stream->key, channel_key, - stream->net_seq_idx, stream->session_id); - - rcu_read_unlock(); - return stream; - -error: - rcu_read_unlock(); - lttng_trace_chunk_put(stream->trace_chunk); - free(stream); -end: - if (alloc_ret) { - *alloc_ret = ret; - } - return NULL; -} - /* * Add a stream to the global list protected by a mutex. */ @@ -708,11 +626,6 @@ 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. @@ -819,7 +732,7 @@ 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); @@ -1013,32 +926,17 @@ 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 @@ -1054,59 +952,9 @@ 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; } /* @@ -1127,6 +975,7 @@ struct lttng_consumer_channel *consumer_allocate_channel(uint64_t key, uint64_t session_id_per_pid, unsigned int monitor, unsigned int live_timer_interval, + bool is_in_live_session, const char *root_shm_path, const char *shm_path) { @@ -1158,6 +1007,7 @@ struct lttng_consumer_channel *consumer_allocate_channel(uint64_t key, channel->tracefile_count = tracefile_count; channel->monitor = monitor; channel->live_timer_interval = live_timer_interval; + channel->is_live = is_in_live_session; pthread_mutex_init(&channel->lock, NULL); pthread_mutex_init(&channel->timer_lock, NULL); @@ -1397,6 +1247,7 @@ void lttng_consumer_cleanup(void) { struct lttng_ht_iter iter; struct lttng_consumer_channel *channel; + unsigned int trace_chunks_left; rcu_read_lock(); @@ -1421,6 +1272,27 @@ 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); } @@ -1504,7 +1376,7 @@ void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream, struct lttng_consumer_local_data *lttng_consumer_create( enum lttng_consumer_type type, ssize_t (*buffer_ready)(struct lttng_consumer_stream *stream, - struct lttng_consumer_local_data *ctx), + struct lttng_consumer_local_data *ctx, bool locked_by_caller), int (*recv_channel)(struct lttng_consumer_channel *channel), int (*recv_stream)(struct lttng_consumer_stream *stream), int (*update_stream)(uint64_t stream_key, uint32_t state)) @@ -1711,19 +1583,18 @@ end: * Returns the number of bytes written */ ssize_t lttng_consumer_on_read_subbuffer_mmap( - struct lttng_consumer_local_data *ctx, - struct lttng_consumer_stream *stream, unsigned long len, - unsigned long padding, - struct ctf_packet_index *index) + struct lttng_consumer_stream *stream, + const struct lttng_buffer_view *buffer, + unsigned long padding) { - 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(); @@ -1739,39 +1610,9 @@ 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 @@ -1809,10 +1650,10 @@ 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) { @@ -1826,7 +1667,7 @@ 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) { @@ -1835,19 +1676,17 @@ ssize_t lttng_consumer_on_read_subbuffer_mmap( outfd = stream->out_fd; orig_offset = 0; } - stream->tracefile_size_current += len; - if (index) { - index->offset = htobe64(stream->out_fd_offset); - } + stream->tracefile_size_current += buffer->size; + 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 %zu)", 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. @@ -1868,7 +1707,8 @@ 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; } @@ -1877,9 +1717,9 @@ 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); } @@ -1913,8 +1753,7 @@ end: ssize_t lttng_consumer_on_read_subbuffer_splice( struct lttng_consumer_local_data *ctx, struct lttng_consumer_stream *stream, unsigned long len, - unsigned long padding, - struct ctf_packet_index *index) + unsigned long padding) { ssize_t ret = 0, written = 0, ret_splice = 0; loff_t offset = 0; @@ -2018,7 +1857,6 @@ ssize_t lttng_consumer_on_read_subbuffer_splice( orig_offset = 0; } stream->tracefile_size_current += len; - index->offset = htobe64(stream->out_fd_offset); } while (len > 0) { @@ -2221,6 +2059,7 @@ int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx, } } +static void lttng_consumer_close_all_metadata(void) { switch (consumer_data.type) { @@ -2253,7 +2092,8 @@ 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); /* @@ -2265,11 +2105,17 @@ 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. */ @@ -2281,28 +2127,29 @@ 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); @@ -2576,7 +2423,7 @@ restart: do { health_code_update(); - len = ctx->on_buffer_ready(stream, ctx); + len = ctx->on_buffer_ready(stream, ctx, false); /* * We don't check the return value here since if we get * a negative len, it means an error occurred thus we @@ -2603,7 +2450,7 @@ restart: do { health_code_update(); - len = ctx->on_buffer_ready(stream, ctx); + len = ctx->on_buffer_ready(stream, ctx, false); /* * We don't check the return value here since if we get * a negative len, it means an error occurred thus we @@ -2819,7 +2666,7 @@ void *consumer_thread_data_poll(void *data) if (pollfd[i].revents & POLLPRI) { DBG("Urgent read on fd %d", pollfd[i].fd); high_prio = 1; - len = ctx->on_buffer_ready(local_stream[i], ctx); + len = ctx->on_buffer_ready(local_stream[i], ctx, false); /* it's ok to have an unavailable sub-buffer */ if (len < 0 && len != -EAGAIN && len != -ENODATA) { /* Clean the stream and free it. */ @@ -2850,7 +2697,7 @@ void *consumer_thread_data_poll(void *data) local_stream[i]->hangup_flush_done || local_stream[i]->has_data) { DBG("Normal read on fd %d", pollfd[i].fd); - len = ctx->on_buffer_ready(local_stream[i], ctx); + len = ctx->on_buffer_ready(local_stream[i], ctx, false); /* it's ok to have an unavailable sub-buffer */ if (len < 0 && len != -EAGAIN && len != -ENODATA) { /* Clean the stream and free it. */ @@ -3464,39 +3311,123 @@ error_testpoint: } ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream, - struct lttng_consumer_local_data *ctx) + struct lttng_consumer_local_data *ctx, + bool locked_by_caller) { - ssize_t ret; + ssize_t ret, written_bytes; + int rotation_ret; + struct stream_subbuffer subbuffer = {}; - pthread_mutex_lock(&stream->chan->lock); - pthread_mutex_lock(&stream->lock); - if (stream->metadata_flag) { - pthread_mutex_lock(&stream->metadata_rdv_lock); + if (!locked_by_caller) { + stream->read_subbuffer_ops.lock(stream); } - switch (consumer_data.type) { - case LTTNG_CONSUMER_KERNEL: - ret = lttng_kconsumer_read_subbuffer(stream, ctx); - break; - case LTTNG_CONSUMER32_UST: - case LTTNG_CONSUMER64_UST: - ret = lttng_ustconsumer_read_subbuffer(stream, ctx); - break; - default: - ERR("Unknown consumer_data type"); - assert(0); - ret = -ENOSYS; - break; + if (stream->read_subbuffer_ops.on_wake_up) { + ret = stream->read_subbuffer_ops.on_wake_up(stream); + if (ret) { + goto end; + } } - if (stream->metadata_flag) { - pthread_cond_broadcast(&stream->metadata_rdv); - pthread_mutex_unlock(&stream->metadata_rdv_lock); + /* + * If the stream was flagged to be ready for rotation before we extract + * the next packet, rotate it now. + */ + if (stream->rotate_ready) { + DBG("Rotate stream before consuming data"); + ret = lttng_consumer_rotate_stream(ctx, stream); + if (ret < 0) { + ERR("Stream rotation error before consuming data"); + goto end; + } + } + + ret = stream->read_subbuffer_ops.get_next_subbuffer(stream, &subbuffer); + if (ret) { + if (ret == -ENODATA) { + /* Not an error. */ + ret = 0; + goto sleep_stream; + } + goto end; + } + + ret = stream->read_subbuffer_ops.pre_consume_subbuffer( + stream, &subbuffer); + if (ret) { + goto error_put_subbuf; + } + + written_bytes = stream->read_subbuffer_ops.consume_subbuffer( + ctx, stream, &subbuffer); + /* + * Should write subbuf_size amount of data when network streaming or + * the full padded size when we are not streaming. + */ + if ((written_bytes != subbuffer.info.data.subbuf_size && + stream->net_seq_idx != (uint64_t) -1ULL) || + (written_bytes != subbuffer.info.data.padded_subbuf_size && + stream->net_seq_idx == + (uint64_t) -1ULL)) { + /* + * Display the error but continue processing to try to + * release the subbuffer. This is a DBG statement + * since this can happen without being a critical + * error. + */ + DBG("Failed to write to tracefile (written_bytes: %zd != padded subbuffer size: %lu, subbuffer size: %lu)", + written_bytes, subbuffer.info.data.subbuf_size, + subbuffer.info.data.padded_subbuf_size); + } + + ret = stream->read_subbuffer_ops.put_next_subbuffer(stream, &subbuffer); + if (ret) { + goto end; + } + + if (stream->read_subbuffer_ops.post_consume) { + ret = stream->read_subbuffer_ops.post_consume(stream, &subbuffer, ctx); + if (ret) { + goto end; + } + } + + /* + * After extracting the packet, we check if the stream is now ready to + * be rotated and perform the action immediately. + * + * Don't overwrite `ret` as callers expect the number of bytes + * consumed to be returned on success. + */ + rotation_ret = lttng_consumer_stream_is_rotate_ready(stream); + if (rotation_ret == 1) { + rotation_ret = lttng_consumer_rotate_stream(ctx, stream); + if (rotation_ret < 0) { + ret = rotation_ret; + ERR("Stream rotation error after consuming data"); + goto end; + } + } else if (rotation_ret < 0) { + ret = rotation_ret; + ERR("Failed to check if stream was ready to rotate after consuming data"); + goto end; + } + +stream_sleep: + if (stream->read_subbuffer_ops.on_sleep) { + stream->read_subbuffer_ops.on_sleep(stream, ctx); + } + + ret = written_bytes; +end: + if (!locked_by_caller) { + stream->read_subbuffer_ops.unlock(stream); } - pthread_mutex_unlock(&stream->lock); - pthread_mutex_unlock(&stream->chan->lock); return ret; +error_put_subbuf: + (void) stream->read_subbuffer_ops.put_next_subbuffer(stream, &subbuffer); + goto end; } int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream) @@ -3977,15 +3908,33 @@ 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"); @@ -4040,7 +3989,7 @@ int lttng_consumer_rotate_channel(struct lttng_consumer_channel *channel, 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(); @@ -4053,65 +4002,95 @@ int lttng_consumer_rotate_channel(struct lttng_consumer_channel *channel, rotating_to_new_chunk = false; } - ret = lttng_consumer_sample_snapshot_positions(stream); - if (ret < 0) { - ERR("Failed to sample snapshot position during channel rotation"); - goto end_unlock_stream; + /* + * 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_get_produced_snapshot(stream, - &stream->rotate_position); - if (ret < 0) { - ERR("Failed to sample produced position during channel rotation"); + 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); } - /* - * Active flush; has no effect if the production position - * is at a packet boundary. + * 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. */ - ret = consumer_flush_buffer(stream, 1); - if (ret < 0) { - ERR("Failed to flush stream %" PRIu64 " during channel rotation", + 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. - * - * At the moment when the positions of the buffers are - * sampled, the production position does not necessarily - * sit at a packet boundary. The 'active' flush - * operation above will push the production position to - * the next packet boundary _if_ it is not already - * sitting at such a boundary. - * - * Assuming a current production position that is not - * on the bound of a packet, the 'target' sequence - * number is - * (consumed_pos / subbuffer_size) + 1 - * Note the '+ 1' to ensure the current packet is - * part of the current trace chunk. - * - * However, if the production position is already at - * a packet boundary, the '+ 1' is not necessary as the - * last packet of the current chunk is already - * 'complete'. + * _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 / stream->max_sb_size) + - !!(stream->rotate_position % stream->max_sb_size), + .rotate_at_seq_num = stream->rotate_position, }; ret = lttng_dynamic_array_add_element( @@ -4166,60 +4145,172 @@ end: 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; +} - /* Rotate position not reached yet (with check for overflow). */ - if ((long) (consumed_pos - stream->rotate_position) < 0) { - ret = 0; - goto end; +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; } - ret = 1; -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 = 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; } @@ -4394,7 +4485,7 @@ 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; @@ -4420,7 +4511,7 @@ 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]; @@ -4471,11 +4562,11 @@ 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) { @@ -4483,7 +4574,7 @@ 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; } } @@ -4493,7 +4584,7 @@ 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 @@ -4501,10 +4592,11 @@ 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; } } @@ -4516,7 +4608,7 @@ 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(); @@ -4539,11 +4631,13 @@ 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, NULL); + 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); @@ -4569,12 +4663,13 @@ enum lttcomm_return_code lttng_consumer_create_trace_chunk( 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); + 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, @@ -4582,21 +4677,23 @@ enum lttcomm_return_code lttng_consumer_create_trace_chunk( } ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED; - goto error; + goto error_unlock; } } -error: +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, - const enum lttng_trace_chunk_command_type *close_command) + 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; @@ -4694,7 +4791,8 @@ enum lttcomm_return_code lttng_consumer_close_trace_chunk( if (relayd) { pthread_mutex_lock(&relayd->ctrl_sock_mutex); ret = relayd_close_trace_chunk( - &relayd->control_sock, chunk); + &relayd->control_sock, chunk, + path); pthread_mutex_unlock(&relayd->ctrl_sock_mutex); } else { ERR("Failed to find relay daemon socket: relayd_id = %" PRIu64, @@ -4725,12 +4823,11 @@ enum lttcomm_return_code lttng_consumer_trace_chunk_exists( { 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_remote; + bool chunk_exists_local, chunk_exists_remote; if (relayd_id) { int ret; @@ -4748,13 +4845,19 @@ enum lttcomm_return_code lttng_consumer_trace_chunk_exists( DBG("Consumer trace chunk exists command: relayd_id = %s" ", chunk_id = %" PRIu64, relayd_id_str, chunk_id); - chunk = lttng_trace_chunk_registry_find_chunk( + ret = lttng_trace_chunk_registry_chunk_exists( consumer_data.chunk_registry, session_id, - chunk_id); - DBG("Trace chunk %s locally", chunk ? "exists" : "does not exist"); - if (chunk) { + 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; - lttng_trace_chunk_put(chunk); goto end; } else if (is_local_trace) { ret_code = LTTCOMM_CONSUMERD_UNKNOWN_TRACE_CHUNK; @@ -4790,3 +4893,67 @@ end_rcu_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; +}