X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fcommon%2Fconsumer%2Fconsumer.c;h=84fdbcaaa6902e076476489c3ab4e6b1aa7a1c9a;hp=46b961e8980d8f30e68062a73dd2979c6ca6bf92;hb=1cbd136b2479ef142bfb339b13d3d25aa772dda5;hpb=ec135c61effc83fff0695e0e36a4182a8be83c5c diff --git a/src/common/consumer/consumer.c b/src/common/consumer/consumer.c index 46b961e89..84fdbcaaa 100644 --- a/src/common/consumer/consumer.c +++ b/src/common/consumer/consumer.c @@ -1,24 +1,14 @@ /* - * 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 #include #include @@ -33,6 +23,7 @@ #include #include #include +#include #include #include #include @@ -48,11 +39,15 @@ #include #include #include - -struct lttng_consumer_global_data consumer_data = { - .stream_count = 0, - .need_update = 1, - .type = LTTNG_CONSUMER_UNKNOWN, +#include +#include +#include +#include + +struct lttng_consumer_global_data the_consumer_data = { + .stream_count = 0, + .need_update = 1, + .type = LTTNG_CONSUMER_UNKNOWN, }; enum consumer_channel_action { @@ -86,6 +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 (the_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 @@ -95,7 +104,7 @@ static void notify_thread_lttng_pipe(struct lttng_pipe *pipe) { struct lttng_consumer_stream *null_stream = NULL; - assert(pipe); + LTTNG_ASSERT(pipe); (void) lttng_pipe_write(pipe, &null_stream, sizeof(null_stream)); } @@ -163,7 +172,7 @@ static void clean_channel_stream_list(struct lttng_consumer_channel *channel) { struct lttng_consumer_stream *stream, *stmp; - assert(channel); + LTTNG_ASSERT(channel); /* Delete streams that might have been left in the stream list. */ cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head, @@ -191,7 +200,7 @@ static struct lttng_consumer_stream *find_stream(uint64_t key, struct lttng_ht_node_u64 *node; struct lttng_consumer_stream *stream = NULL; - assert(ht); + LTTNG_ASSERT(ht); /* -1ULL keys are lookup failures */ if (key == (uint64_t) -1ULL) { @@ -246,7 +255,7 @@ struct lttng_consumer_channel *consumer_find_channel(uint64_t key) return NULL; } - lttng_ht_lookup(consumer_data.channel_ht, &key, &iter); + lttng_ht_lookup(the_consumer_data.channel_ht, &key, &iter); node = lttng_ht_iter_get_node_u64(&iter); if (node != NULL) { channel = caa_container_of(node, struct lttng_consumer_channel, node); @@ -288,7 +297,7 @@ static void free_channel_rcu(struct rcu_head *head) struct lttng_consumer_channel *channel = caa_container_of(node, struct lttng_consumer_channel, node); - switch (consumer_data.type) { + switch (the_consumer_data.type) { case LTTNG_CONSUMER_KERNEL: break; case LTTNG_CONSUMER32_UST: @@ -323,6 +332,7 @@ static void free_relayd_rcu(struct rcu_head *head) (void) relayd_close(&relayd->control_sock); (void) relayd_close(&relayd->data_sock); + pthread_mutex_destroy(&relayd->ctrl_sock_mutex); free(relayd); } @@ -341,7 +351,7 @@ void consumer_destroy_relayd(struct consumer_relayd_sock_pair *relayd) DBG("Consumer destroy and close relayd socket pair"); iter.iter.node = &relayd->node.node; - ret = lttng_ht_del(consumer_data.relayd_ht, &iter); + ret = lttng_ht_del(the_consumer_data.relayd_ht, &iter); if (ret != 0) { /* We assume the relayd is being or is destroyed */ return; @@ -357,12 +367,11 @@ void consumer_destroy_relayd(struct consumer_relayd_sock_pair *relayd) */ void consumer_del_channel(struct lttng_consumer_channel *channel) { - int ret; struct lttng_ht_iter iter; DBG("Consumer delete channel key %" PRIu64, channel->key); - pthread_mutex_lock(&consumer_data.lock); + pthread_mutex_lock(&the_consumer_data.lock); pthread_mutex_lock(&channel->lock); /* Destroy streams that might have been left in the stream list. */ @@ -375,7 +384,7 @@ void consumer_del_channel(struct lttng_consumer_channel *channel) consumer_timer_monitor_stop(channel); } - switch (consumer_data.type) { + switch (the_consumer_data.type) { case LTTNG_CONSUMER_KERNEL: break; case LTTNG_CONSUMER32_UST: @@ -384,20 +393,33 @@ void consumer_del_channel(struct lttng_consumer_channel *channel) break; default: ERR("Unknown consumer_data type"); - assert(0); + abort(); goto end; } - rcu_read_lock(); - iter.iter.node = &channel->node.node; - ret = lttng_ht_del(consumer_data.channel_ht, &iter); - assert(!ret); - rcu_read_unlock(); + lttng_trace_chunk_put(channel->trace_chunk); + channel->trace_chunk = NULL; + + if (channel->is_published) { + int ret; + rcu_read_lock(); + iter.iter.node = &channel->node.node; + ret = lttng_ht_del(the_consumer_data.channel_ht, &iter); + LTTNG_ASSERT(!ret); + + iter.iter.node = &channel->channels_by_session_id_ht_node.node; + ret = lttng_ht_del(the_consumer_data.channels_by_session_id_ht, + &iter); + LTTNG_ASSERT(!ret); + rcu_read_unlock(); + } + + channel->is_deleted = true; call_rcu(&channel->node.head, free_channel_rcu); end: pthread_mutex_unlock(&channel->lock); - pthread_mutex_unlock(&consumer_data.lock); + pthread_mutex_unlock(&the_consumer_data.lock); } /* @@ -411,14 +433,14 @@ static void cleanup_relayd_ht(void) rcu_read_lock(); - cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd, - node.node) { + cds_lfht_for_each_entry(the_consumer_data.relayd_ht->ht, &iter.iter, + relayd, node.node) { consumer_destroy_relayd(relayd); } rcu_read_unlock(); - lttng_ht_destroy(consumer_data.relayd_ht); + lttng_ht_destroy(the_consumer_data.relayd_ht); } /* @@ -464,14 +486,13 @@ static void update_endpoint_status_by_netidx(uint64_t net_seq_idx, * If a local data context is available, notify the threads that the streams' * state have changed. */ -static void cleanup_relayd(struct consumer_relayd_sock_pair *relayd, - struct lttng_consumer_local_data *ctx) +void lttng_consumer_cleanup_relayd(struct consumer_relayd_sock_pair *relayd) { uint64_t netidx; - assert(relayd); + LTTNG_ASSERT(relayd); - DBG("Cleaning up relayd sockets"); + DBG("Cleaning up relayd object ID %"PRIu64, relayd->net_seq_idx); /* Save the net sequence index before destroying the object */ netidx = relayd->net_seq_idx; @@ -491,10 +512,8 @@ static void cleanup_relayd(struct consumer_relayd_sock_pair *relayd, * memory barrier ordering the updates of the end point status from the * read of this status which happens AFTER receiving this notify. */ - if (ctx) { - notify_thread_lttng_pipe(ctx->consumer_data_pipe); - notify_thread_lttng_pipe(ctx->consumer_metadata_pipe); - } + notify_thread_lttng_pipe(relayd->ctx->consumer_data_pipe); + notify_thread_lttng_pipe(relayd->ctx->consumer_metadata_pipe); } /* @@ -505,7 +524,7 @@ static void cleanup_relayd(struct consumer_relayd_sock_pair *relayd, */ void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair *relayd) { - assert(relayd); + LTTNG_ASSERT(relayd); /* Set destroy flag for this object */ uatomic_set(&relayd->destroy_flag, 1); @@ -547,96 +566,6 @@ void consumer_stream_update_channel_attributes( { stream->channel_read_only_attributes.tracefile_size = channel->tracefile_size; - memcpy(stream->channel_read_only_attributes.path, channel->pathname, - sizeof(stream->channel_read_only_attributes.path)); -} - -struct lttng_consumer_stream *consumer_allocate_stream(uint64_t channel_key, - uint64_t stream_key, - enum lttng_consumer_stream_state state, - const char *channel_name, - uid_t uid, - gid_t gid, - uint64_t relayd_id, - uint64_t session_id, - int cpu, - int *alloc_ret, - enum consumer_channel_type type, - unsigned int monitor, - uint64_t trace_archive_id) -{ - int ret; - struct lttng_consumer_stream *stream; - - stream = zmalloc(sizeof(*stream)); - if (stream == NULL) { - PERROR("malloc struct lttng_consumer_stream"); - ret = -ENOMEM; - goto end; - } - - rcu_read_lock(); - - stream->key = stream_key; - stream->out_fd = -1; - stream->out_fd_offset = 0; - stream->output_written = 0; - stream->state = state; - stream->uid = uid; - stream->gid = gid; - 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; - stream->trace_archive_id = trace_archive_id; - 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(); - free(stream); -end: - if (alloc_ret) { - *alloc_ret = ret; - } - return NULL; } /* @@ -646,12 +575,12 @@ void consumer_add_data_stream(struct lttng_consumer_stream *stream) { struct lttng_ht *ht = data_ht; - assert(stream); - assert(ht); + LTTNG_ASSERT(stream); + LTTNG_ASSERT(ht); DBG3("Adding consumer stream %" PRIu64, stream->key); - pthread_mutex_lock(&consumer_data.lock); + pthread_mutex_lock(&the_consumer_data.lock); pthread_mutex_lock(&stream->chan->lock); pthread_mutex_lock(&stream->chan->timer_lock); pthread_mutex_lock(&stream->lock); @@ -662,7 +591,7 @@ void consumer_add_data_stream(struct lttng_consumer_stream *stream) lttng_ht_add_unique_u64(ht, &stream->node); - lttng_ht_add_u64(consumer_data.stream_per_chan_id_ht, + lttng_ht_add_u64(the_consumer_data.stream_per_chan_id_ht, &stream->node_channel_id); /* @@ -670,7 +599,8 @@ void consumer_add_data_stream(struct lttng_consumer_stream *stream) * the key since the HT does not use it and we allow to add redundant keys * into this table. */ - lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id); + lttng_ht_add_u64(the_consumer_data.stream_list_ht, + &stream->node_session_id); /* * When nb_init_stream_left reaches 0, we don't need to trigger any action @@ -686,19 +616,14 @@ void consumer_add_data_stream(struct lttng_consumer_stream *stream) } /* Update consumer data once the node is inserted. */ - consumer_data.stream_count++; - consumer_data.need_update = 1; + the_consumer_data.stream_count++; + the_consumer_data.need_update = 1; rcu_read_unlock(); pthread_mutex_unlock(&stream->lock); pthread_mutex_unlock(&stream->chan->timer_lock); pthread_mutex_unlock(&stream->chan->lock); - pthread_mutex_unlock(&consumer_data.lock); -} - -void consumer_del_data_stream(struct lttng_consumer_stream *stream) -{ - consumer_del_stream(stream, data_ht); + pthread_mutex_unlock(&the_consumer_data.lock); } /* @@ -711,15 +636,15 @@ static int add_relayd(struct consumer_relayd_sock_pair *relayd) struct lttng_ht_node_u64 *node; struct lttng_ht_iter iter; - assert(relayd); + LTTNG_ASSERT(relayd); - lttng_ht_lookup(consumer_data.relayd_ht, - &relayd->net_seq_idx, &iter); + lttng_ht_lookup(the_consumer_data.relayd_ht, &relayd->net_seq_idx, + &iter); node = lttng_ht_iter_get_node_u64(&iter); if (node != NULL) { goto end; } - lttng_ht_add_unique_u64(consumer_data.relayd_ht, &relayd->node); + lttng_ht_add_unique_u64(the_consumer_data.relayd_ht, &relayd->node); end: return ret; @@ -774,8 +699,7 @@ struct consumer_relayd_sock_pair *consumer_find_relayd(uint64_t key) goto error; } - lttng_ht_lookup(consumer_data.relayd_ht, &key, - &iter); + lttng_ht_lookup(the_consumer_data.relayd_ht, &key, &iter); node = lttng_ht_iter_get_node_u64(&iter); if (node != NULL) { relayd = caa_container_of(node, struct consumer_relayd_sock_pair, node); @@ -796,9 +720,9 @@ int consumer_send_relayd_stream(struct lttng_consumer_stream *stream, int ret = 0; struct consumer_relayd_sock_pair *relayd; - assert(stream); - assert(stream->net_seq_idx != -1ULL); - assert(path); + LTTNG_ASSERT(stream); + LTTNG_ASSERT(stream->net_seq_idx != -1ULL); + LTTNG_ASSERT(path); /* The stream is not metadata. Get relayd reference if exists. */ rcu_read_lock(); @@ -807,11 +731,14 @@ 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, - stream->chan->tracefile_size, stream->chan->tracefile_count, - stream->trace_archive_id); + get_consumer_domain(), path, &stream->relayd_stream_id, + stream->chan->tracefile_size, + stream->chan->tracefile_count, + stream->trace_chunk); pthread_mutex_unlock(&relayd->ctrl_sock_mutex); if (ret < 0) { + ERR("Relayd add stream failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); + lttng_consumer_cleanup_relayd(relayd); goto end; } @@ -842,7 +769,7 @@ int consumer_send_relayd_streams_sent(uint64_t net_seq_idx) int ret = 0; struct consumer_relayd_sock_pair *relayd; - assert(net_seq_idx != -1ULL); + LTTNG_ASSERT(net_seq_idx != -1ULL); /* The stream is not metadata. Get relayd reference if exists. */ rcu_read_lock(); @@ -853,6 +780,8 @@ int consumer_send_relayd_streams_sent(uint64_t net_seq_idx) ret = relayd_streams_sent(&relayd->control_sock); pthread_mutex_unlock(&relayd->ctrl_sock_mutex); if (ret < 0) { + ERR("Relayd streams sent failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); + lttng_consumer_cleanup_relayd(relayd); goto end; } } else { @@ -900,8 +829,8 @@ static int write_relayd_stream_header(struct lttng_consumer_stream *stream, struct lttcomm_relayd_data_hdr data_hdr; /* Safety net */ - assert(stream); - assert(relayd); + LTTNG_ASSERT(stream); + LTTNG_ASSERT(relayd); /* Reset data header */ memset(&data_hdr, 0, sizeof(data_hdr)); @@ -920,6 +849,7 @@ 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 @@ -946,6 +876,123 @@ error: return outfd; } +/* + * Write a character on the metadata poll pipe to wake the metadata thread. + * Returns 0 on success, -1 on error. + */ +int consumer_metadata_wakeup_pipe(const struct lttng_consumer_channel *channel) +{ + int ret = 0; + + DBG("Waking up metadata poll thread (writing to pipe): channel name = '%s'", + channel->name); + if (channel->monitor && channel->metadata_stream) { + const char dummy = 'c'; + const ssize_t write_ret = lttng_write( + channel->metadata_stream->ust_metadata_poll_pipe[1], + &dummy, 1); + + if (write_ret < 1) { + if (errno == EWOULDBLOCK) { + /* + * This is fine, the metadata poll thread + * is having a hard time keeping-up, but + * it will eventually wake-up and consume + * the available data. + */ + ret = 0; + } else { + PERROR("Failed to write to UST metadata pipe while attempting to wake-up the metadata poll thread"); + ret = -1; + goto end; + } + } + } + +end: + return ret; +} + +/* + * Trigger a dump of the metadata content. Following/during the succesful + * completion of this call, the metadata poll thread will start receiving + * metadata packets to consume. + * + * The caller must hold the channel and stream locks. + */ +static +int consumer_metadata_stream_dump(struct lttng_consumer_stream *stream) +{ + int ret; + + ASSERT_LOCKED(stream->chan->lock); + ASSERT_LOCKED(stream->lock); + LTTNG_ASSERT(stream->metadata_flag); + LTTNG_ASSERT(stream->chan->trace_chunk); + + switch (the_consumer_data.type) { + case LTTNG_CONSUMER_KERNEL: + /* + * Reset the position of what has been read from the + * metadata cache to 0 so we can dump it again. + */ + ret = kernctl_metadata_cache_dump(stream->wait_fd); + break; + case LTTNG_CONSUMER32_UST: + case LTTNG_CONSUMER64_UST: + /* + * Reset the position pushed from the metadata cache so it + * will write from the beginning on the next push. + */ + stream->ust_metadata_pushed = 0; + ret = consumer_metadata_wakeup_pipe(stream->chan); + break; + default: + ERR("Unknown consumer_data type"); + abort(); + } + if (ret < 0) { + ERR("Failed to dump the metadata cache"); + } + return ret; +} + +static +int lttng_consumer_channel_set_trace_chunk( + struct lttng_consumer_channel *channel, + struct lttng_trace_chunk *new_trace_chunk) +{ + pthread_mutex_lock(&channel->lock); + 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 + * a severe internal error) since a reference to the published + * chunk is already held by the caller. + */ + if (new_trace_chunk) { + const bool acquired_reference = lttng_trace_chunk_get( + new_trace_chunk); + + LTTNG_ASSERT(acquired_reference); + } + + lttng_trace_chunk_put(channel->trace_chunk); + channel->trace_chunk = new_trace_chunk; +end: + pthread_mutex_unlock(&channel->lock); + return 0; +} + /* * Allocate and return a new lttng_consumer_channel object using the given key * to initialize the hash table node. @@ -954,10 +1001,9 @@ error: */ struct lttng_consumer_channel *consumer_allocate_channel(uint64_t key, uint64_t session_id, + const uint64_t *chunk_id, const char *pathname, const char *name, - uid_t uid, - gid_t gid, uint64_t relayd_id, enum lttng_event_output output, uint64_t tracefile_size, @@ -965,10 +1011,22 @@ 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) { - struct lttng_consumer_channel *channel; + struct lttng_consumer_channel *channel = NULL; + struct lttng_trace_chunk *trace_chunk = NULL; + + if (chunk_id) { + trace_chunk = lttng_trace_chunk_registry_find_chunk( + the_consumer_data.chunk_registry, session_id, + *chunk_id); + if (!trace_chunk) { + ERR("Failed to find trace chunk reference during creation of channel"); + goto end; + } + } channel = zmalloc(sizeof(*channel)); if (channel == NULL) { @@ -980,13 +1038,12 @@ struct lttng_consumer_channel *consumer_allocate_channel(uint64_t key, channel->refcount = 0; channel->session_id = session_id; channel->session_id_per_pid = session_id_per_pid; - channel->uid = uid; - channel->gid = gid; channel->relayd_id = relayd_id; channel->tracefile_size = tracefile_size; 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); @@ -998,7 +1055,7 @@ struct lttng_consumer_channel *consumer_allocate_channel(uint64_t key, channel->output = CONSUMER_CHANNEL_MMAP; break; default: - assert(0); + abort(); free(channel); channel = NULL; goto end; @@ -1033,15 +1090,29 @@ struct lttng_consumer_channel *consumer_allocate_channel(uint64_t key, } lttng_ht_node_init_u64(&channel->node, channel->key); + lttng_ht_node_init_u64(&channel->channels_by_session_id_ht_node, + channel->session_id); channel->wait_fd = -1; - CDS_INIT_LIST_HEAD(&channel->streams.head); + if (trace_chunk) { + int ret = lttng_consumer_channel_set_trace_chunk(channel, + trace_chunk); + if (ret) { + goto error; + } + } + DBG("Allocated channel (key %" PRIu64 ")", channel->key); end: + lttng_trace_chunk_put(trace_chunk); return channel; +error: + consumer_del_channel(channel); + channel = NULL; + goto end; } /* @@ -1052,7 +1123,7 @@ end: int consumer_add_channel(struct lttng_consumer_channel *channel, struct lttng_consumer_local_data *ctx) { - pthread_mutex_lock(&consumer_data.lock); + pthread_mutex_lock(&the_consumer_data.lock); pthread_mutex_lock(&channel->lock); pthread_mutex_lock(&channel->timer_lock); @@ -1064,12 +1135,15 @@ int consumer_add_channel(struct lttng_consumer_channel *channel, steal_channel_key(channel->key); rcu_read_lock(); - lttng_ht_add_unique_u64(consumer_data.channel_ht, &channel->node); + lttng_ht_add_unique_u64(the_consumer_data.channel_ht, &channel->node); + lttng_ht_add_u64(the_consumer_data.channels_by_session_id_ht, + &channel->channels_by_session_id_ht_node); rcu_read_unlock(); + channel->is_published = true; pthread_mutex_unlock(&channel->timer_lock); pthread_mutex_unlock(&channel->lock); - pthread_mutex_unlock(&consumer_data.lock); + pthread_mutex_unlock(&the_consumer_data.lock); if (channel->wait_fd != -1 && channel->type == CONSUMER_CHANNEL_TYPE_DATA) { notify_channel_pipe(ctx, channel, -1, CONSUMER_CHANNEL_ADD); @@ -1093,10 +1167,10 @@ static int update_poll_array(struct lttng_consumer_local_data *ctx, struct lttng_ht_iter iter; struct lttng_consumer_stream *stream; - assert(ctx); - assert(ht); - assert(pollfd); - assert(local_stream); + LTTNG_ASSERT(ctx); + LTTNG_ASSERT(ht); + LTTNG_ASSERT(pollfd); + LTTNG_ASSERT(local_stream); DBG("Updating poll fd array"); *nb_inactive_fd = 0; @@ -1115,16 +1189,13 @@ static int update_poll_array(struct lttng_consumer_local_data *ctx, * closed by the polling thread after a wakeup on the data_pipe or * metadata_pipe. */ - if (stream->state != LTTNG_CONSUMER_ACTIVE_STREAM || - stream->endpoint_status == CONSUMER_ENDPOINT_INACTIVE) { + if (stream->endpoint_status == CONSUMER_ENDPOINT_INACTIVE) { (*nb_inactive_fd)++; continue; } /* * This clobbers way too much the debug output. Uncomment that if you * need it for debugging purposes. - * - * DBG("Active FD %d", stream->wait_fd); */ (*pollfd)[i].fd = stream->wait_fd; (*pollfd)[i].events = POLLIN | POLLPRI; @@ -1212,28 +1283,53 @@ void lttng_consumer_cleanup(void) { struct lttng_ht_iter iter; struct lttng_consumer_channel *channel; + unsigned int trace_chunks_left; rcu_read_lock(); - cds_lfht_for_each_entry(consumer_data.channel_ht->ht, &iter.iter, channel, - node.node) { + cds_lfht_for_each_entry(the_consumer_data.channel_ht->ht, &iter.iter, + channel, node.node) { consumer_del_channel(channel); } rcu_read_unlock(); - lttng_ht_destroy(consumer_data.channel_ht); + lttng_ht_destroy(the_consumer_data.channel_ht); + lttng_ht_destroy(the_consumer_data.channels_by_session_id_ht); cleanup_relayd_ht(); - lttng_ht_destroy(consumer_data.stream_per_chan_id_ht); + lttng_ht_destroy(the_consumer_data.stream_per_chan_id_ht); /* * This HT contains streams that are freed by either the metadata thread or * the data thread so we do *nothing* on the hash table and simply destroy * it. */ - lttng_ht_destroy(consumer_data.stream_list_ht); + lttng_ht_destroy(the_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( + the_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(the_consumer_data.chunk_registry); } /* @@ -1316,7 +1412,7 @@ 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)) @@ -1324,9 +1420,9 @@ struct lttng_consumer_local_data *lttng_consumer_create( int ret; struct lttng_consumer_local_data *ctx; - assert(consumer_data.type == LTTNG_CONSUMER_UNKNOWN || - consumer_data.type == type); - consumer_data.type = type; + LTTNG_ASSERT(the_consumer_data.type == LTTNG_CONSUMER_UNKNOWN || + the_consumer_data.type == type); + the_consumer_data.type = type; ctx = zmalloc(sizeof(struct lttng_consumer_local_data)); if (ctx == NULL) { @@ -1516,29 +1612,30 @@ end: * core function for writing trace buffers to either the local filesystem or * the network. * - * It must be called with the stream lock held. + * It must be called with the stream and the channel lock held. * * Careful review MUST be put if any changes occur! * * 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(); + LTTNG_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) { @@ -1549,39 +1646,9 @@ 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 @@ -1619,10 +1686,10 @@ 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) { @@ -1636,49 +1703,26 @@ 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 = utils_rotate_stream_file(stream->chan->pathname, - stream->name, stream->chan->tracefile_size, - stream->chan->tracefile_count, stream->uid, stream->gid, - stream->out_fd, &(stream->tracefile_count_current), - &stream->out_fd); - if (ret < 0) { - ERR("Rotating output file"); + ret = consumer_stream_rotate_output_files(stream); + if (ret) { goto end; } outfd = stream->out_fd; - - if (stream->index_file) { - lttng_index_file_put(stream->index_file); - stream->index_file = lttng_index_file_create(stream->chan->pathname, - stream->name, stream->uid, stream->gid, - stream->chan->tracefile_size, - stream->tracefile_count_current, - CTF_INDEX_MAJOR, CTF_INDEX_MINOR); - if (!stream->index_file) { - goto end; - } - } - - /* Reset current size because we just perform a rotation. */ - stream->tracefile_size_current = 0; - stream->out_fd_offset = 0; 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. @@ -1689,7 +1733,7 @@ ssize_t lttng_consumer_on_read_subbuffer_mmap( relayd_hang_up = 1; /* Socket operation failed. We consider the relayd dead */ - if (errno == EPIPE || errno == EINVAL || errno == EBADF) { + if (errno == EPIPE) { /* * This is possible if the fd is closed on the other side * (outfd) or any write problem. It can be verbose a bit for a @@ -1699,7 +1743,8 @@ 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; } @@ -1708,9 +1753,9 @@ 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); } @@ -1720,7 +1765,8 @@ write_error: * cleanup the relayd object and all associated streams. */ if (relayd && relayd_hang_up) { - cleanup_relayd(relayd, ctx); + ERR("Relayd hangup. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); + lttng_consumer_cleanup_relayd(relayd); } end: @@ -1743,8 +1789,7 @@ 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; @@ -1756,7 +1801,7 @@ ssize_t lttng_consumer_on_read_subbuffer_splice( int *splice_pipe; unsigned int relayd_hang_up = 0; - switch (consumer_data.type) { + switch (the_consumer_data.type) { case LTTNG_CONSUMER_KERNEL: break; case LTTNG_CONSUMER32_UST: @@ -1765,7 +1810,7 @@ ssize_t lttng_consumer_on_read_subbuffer_splice( return -ENOSYS; default: ERR("Unknown consumer_data type"); - assert(0); + abort(); } /* RCU lock for the relayd pointer */ @@ -1839,37 +1884,15 @@ ssize_t lttng_consumer_on_read_subbuffer_splice( if (stream->chan->tracefile_size > 0 && (stream->tracefile_size_current + len) > stream->chan->tracefile_size) { - ret = utils_rotate_stream_file(stream->chan->pathname, - stream->name, stream->chan->tracefile_size, - stream->chan->tracefile_count, stream->uid, stream->gid, - stream->out_fd, &(stream->tracefile_count_current), - &stream->out_fd); + ret = consumer_stream_rotate_output_files(stream); if (ret < 0) { written = ret; - ERR("Rotating output file"); goto end; } outfd = stream->out_fd; - - if (stream->index_file) { - lttng_index_file_put(stream->index_file); - stream->index_file = lttng_index_file_create(stream->chan->pathname, - stream->name, stream->uid, stream->gid, - stream->chan->tracefile_size, - stream->tracefile_count_current, - CTF_INDEX_MAJOR, CTF_INDEX_MINOR); - if (!stream->index_file) { - goto end; - } - } - - /* Reset current size because we just perform a rotation. */ - stream->tracefile_size_current = 0; - stream->out_fd_offset = 0; orig_offset = 0; } stream->tracefile_size_current += len; - index->offset = htobe64(stream->out_fd_offset); } while (len > 0) { @@ -1946,7 +1969,8 @@ write_error: * cleanup the relayd object and all associated streams. */ if (relayd && relayd_hang_up) { - cleanup_relayd(relayd, ctx); + ERR("Relayd hangup. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); + lttng_consumer_cleanup_relayd(relayd); /* Skip splice error so the consumer does not fail */ goto end; } @@ -1981,7 +2005,7 @@ end: */ int lttng_consumer_sample_snapshot_positions(struct lttng_consumer_stream *stream) { - switch (consumer_data.type) { + switch (the_consumer_data.type) { case LTTNG_CONSUMER_KERNEL: return lttng_kconsumer_sample_snapshot_positions(stream); case LTTNG_CONSUMER32_UST: @@ -1989,7 +2013,7 @@ int lttng_consumer_sample_snapshot_positions(struct lttng_consumer_stream *strea return lttng_ustconsumer_sample_snapshot_positions(stream); default: ERR("Unknown consumer_data type"); - assert(0); + abort(); return -ENOSYS; } } @@ -2000,7 +2024,7 @@ int lttng_consumer_sample_snapshot_positions(struct lttng_consumer_stream *strea */ int lttng_consumer_take_snapshot(struct lttng_consumer_stream *stream) { - switch (consumer_data.type) { + switch (the_consumer_data.type) { case LTTNG_CONSUMER_KERNEL: return lttng_kconsumer_take_snapshot(stream); case LTTNG_CONSUMER32_UST: @@ -2008,7 +2032,7 @@ int lttng_consumer_take_snapshot(struct lttng_consumer_stream *stream) return lttng_ustconsumer_take_snapshot(stream); default: ERR("Unknown consumer_data type"); - assert(0); + abort(); return -ENOSYS; } } @@ -2021,7 +2045,7 @@ int lttng_consumer_take_snapshot(struct lttng_consumer_stream *stream) int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream *stream, unsigned long *pos) { - switch (consumer_data.type) { + switch (the_consumer_data.type) { case LTTNG_CONSUMER_KERNEL: return lttng_kconsumer_get_produced_snapshot(stream, pos); case LTTNG_CONSUMER32_UST: @@ -2029,7 +2053,7 @@ int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream *stream, return lttng_ustconsumer_get_produced_snapshot(stream, pos); default: ERR("Unknown consumer_data type"); - assert(0); + abort(); return -ENOSYS; } } @@ -2042,7 +2066,7 @@ int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream *stream, int lttng_consumer_get_consumed_snapshot(struct lttng_consumer_stream *stream, unsigned long *pos) { - switch (consumer_data.type) { + switch (the_consumer_data.type) { case LTTNG_CONSUMER_KERNEL: return lttng_kconsumer_get_consumed_snapshot(stream, pos); case LTTNG_CONSUMER32_UST: @@ -2050,7 +2074,7 @@ int lttng_consumer_get_consumed_snapshot(struct lttng_consumer_stream *stream, return lttng_ustconsumer_get_consumed_snapshot(stream, pos); default: ERR("Unknown consumer_data type"); - assert(0); + abort(); return -ENOSYS; } } @@ -2058,7 +2082,7 @@ int lttng_consumer_get_consumed_snapshot(struct lttng_consumer_stream *stream, int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx, int sock, struct pollfd *consumer_sockpoll) { - switch (consumer_data.type) { + switch (the_consumer_data.type) { case LTTNG_CONSUMER_KERNEL: return lttng_kconsumer_recv_cmd(ctx, sock, consumer_sockpoll); case LTTNG_CONSUMER32_UST: @@ -2066,14 +2090,15 @@ int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx, return lttng_ustconsumer_recv_cmd(ctx, sock, consumer_sockpoll); default: ERR("Unknown consumer_data type"); - assert(0); + abort(); return -ENOSYS; } } +static void lttng_consumer_close_all_metadata(void) { - switch (consumer_data.type) { + switch (the_consumer_data.type) { case LTTNG_CONSUMER_KERNEL: /* * The Kernel consumer has a different metadata scheme so we don't @@ -2093,7 +2118,7 @@ void lttng_consumer_close_all_metadata(void) break; default: ERR("Unknown consumer_data type"); - assert(0); + abort(); } } @@ -2103,23 +2128,30 @@ 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); + LTTNG_ASSERT(stream); /* * This call should NEVER receive regular stream. It must always be * metadata stream and this is crucial for data structure synchronization. */ - assert(stream->metadata_flag); + LTTNG_ASSERT(stream->metadata_flag); DBG3("Consumer delete metadata stream %d", stream->wait_fd); - pthread_mutex_lock(&consumer_data.lock); - pthread_mutex_lock(&stream->chan->lock); + pthread_mutex_lock(&the_consumer_data.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. */ @@ -2131,30 +2163,33 @@ 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(&consumer_data.lock); + pthread_mutex_unlock(&channel->lock); + pthread_mutex_unlock(&the_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); + stream->trace_chunk = NULL; consumer_stream_free(stream); } @@ -2168,12 +2203,12 @@ void consumer_add_metadata_stream(struct lttng_consumer_stream *stream) struct lttng_ht_iter iter; struct lttng_ht_node_u64 *node; - assert(stream); - assert(ht); + LTTNG_ASSERT(stream); + LTTNG_ASSERT(ht); DBG3("Adding metadata stream %" PRIu64 " to hash table", stream->key); - pthread_mutex_lock(&consumer_data.lock); + pthread_mutex_lock(&the_consumer_data.lock); pthread_mutex_lock(&stream->chan->lock); pthread_mutex_lock(&stream->chan->timer_lock); pthread_mutex_lock(&stream->lock); @@ -2191,7 +2226,7 @@ void consumer_add_metadata_stream(struct lttng_consumer_stream *stream) */ lttng_ht_lookup(ht, &stream->key, &iter); node = lttng_ht_iter_get_node_u64(&iter); - assert(!node); + LTTNG_ASSERT(!node); /* * When nb_init_stream_left reaches 0, we don't need to trigger any action @@ -2208,22 +2243,23 @@ void consumer_add_metadata_stream(struct lttng_consumer_stream *stream) lttng_ht_add_unique_u64(ht, &stream->node); - lttng_ht_add_u64(consumer_data.stream_per_chan_id_ht, - &stream->node_channel_id); + lttng_ht_add_u64(the_consumer_data.stream_per_chan_id_ht, + &stream->node_channel_id); /* * Add stream to the stream_list_ht of the consumer data. No need to steal * the key since the HT does not use it and we allow to add redundant keys * into this table. */ - lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id); + lttng_ht_add_u64(the_consumer_data.stream_list_ht, + &stream->node_session_id); rcu_read_unlock(); pthread_mutex_unlock(&stream->lock); pthread_mutex_unlock(&stream->chan->lock); pthread_mutex_unlock(&stream->chan->timer_lock); - pthread_mutex_unlock(&consumer_data.lock); + pthread_mutex_unlock(&the_consumer_data.lock); } /* @@ -2259,7 +2295,7 @@ static void validate_endpoint_status_metadata_stream( DBG("Consumer delete flagged metadata stream"); - assert(pollset); + LTTNG_ASSERT(pollset); rcu_read_lock(); cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) { @@ -2279,72 +2315,6 @@ static void validate_endpoint_status_metadata_stream( rcu_read_unlock(); } -static -int rotate_notify_sessiond(struct lttng_consumer_local_data *ctx, - uint64_t key) -{ - ssize_t ret; - - do { - ret = write(ctx->channel_rotate_pipe, &key, sizeof(key)); - } while (ret == -1 && errno == EINTR); - if (ret == -1) { - PERROR("Failed to write to the channel rotation pipe"); - } else { - DBG("Sent channel rotation notification for channel key %" - PRIu64, key); - ret = 0; - } - - return (int) ret; -} - -/* - * Perform operations that need to be done after a stream has - * rotated and released the stream lock. - * - * Multiple rotations cannot occur simultaneously, so we know the state of the - * "rotated" stream flag cannot change. - * - * This MUST be called WITHOUT the stream lock held. - */ -static -int consumer_post_rotation(struct lttng_consumer_stream *stream, - struct lttng_consumer_local_data *ctx) -{ - int ret = 0; - - pthread_mutex_lock(&stream->chan->lock); - - switch (consumer_data.type) { - case LTTNG_CONSUMER_KERNEL: - break; - case LTTNG_CONSUMER32_UST: - case LTTNG_CONSUMER64_UST: - /* - * The ust_metadata_pushed counter has been reset to 0, so now - * we can wakeup the metadata thread so it dumps the metadata - * cache to the new file. - */ - if (stream->metadata_flag) { - consumer_metadata_wakeup_pipe(stream->chan); - } - break; - default: - ERR("Unknown consumer_data type"); - abort(); - } - - if (--stream->chan->nr_stream_rotate_pending == 0) { - DBG("Rotation of channel \"%s\" completed, notifying the session daemon", - stream->chan->name); - ret = rotate_notify_sessiond(ctx, stream->chan->key); - } - pthread_mutex_unlock(&stream->chan->lock); - - return ret; -} - /* * Thread polls on metadata file descriptor and write them on disk or on the * network. @@ -2418,11 +2388,6 @@ restart: revents = LTTNG_POLL_GETEV(&events, i); pollfd = LTTNG_POLL_GETFD(&events, i); - if (!revents) { - /* No activity for this FD (poll implementation). */ - continue; - } - if (pollfd == lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)) { if (revents & LPOLLIN) { ssize_t pipe_len; @@ -2482,7 +2447,7 @@ restart: lttng_ht_lookup(metadata_ht, &tmp_id, &iter); } node = lttng_ht_iter_get_node_u64(&iter); - assert(node); + LTTNG_ASSERT(node); stream = caa_container_of(node, struct lttng_consumer_stream, node); @@ -2490,12 +2455,12 @@ restart: if (revents & (LPOLLIN | LPOLLPRI)) { /* Get the data out of the metadata file descriptor */ DBG("Metadata available on fd %d", pollfd); - assert(stream->wait_fd == pollfd); + LTTNG_ASSERT(stream->wait_fd == pollfd); 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 @@ -2512,9 +2477,10 @@ restart: } } else if (revents & (LPOLLERR | LPOLLHUP)) { DBG("Metadata fd %d is hup|err.", pollfd); - if (!stream->hangup_flush_done - && (consumer_data.type == LTTNG_CONSUMER32_UST - || consumer_data.type == LTTNG_CONSUMER64_UST)) { + if (!stream->hangup_flush_done && + (the_consumer_data.type == LTTNG_CONSUMER32_UST || + the_consumer_data.type == + LTTNG_CONSUMER64_UST)) { DBG("Attempting to flush and consume the UST buffers"); lttng_ustconsumer_on_stream_hangup(stream); @@ -2522,7 +2488,7 @@ 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 @@ -2576,7 +2542,9 @@ void *consumer_thread_data_poll(void *data) /* local view of the streams */ struct lttng_consumer_stream **local_stream = NULL, *new_stream = NULL; /* local view of consumer_data.fds_count */ - int nb_fd = 0, nb_pipes_fd; + int nb_fd = 0; + /* 2 for the consumer_data_pipe and wake up pipe */ + const int nb_pipes_fd = 2; /* Number of FDs with CONSUMER_ENDPOINT_INACTIVE but still open. */ int nb_inactive_fd = 0; struct lttng_consumer_local_data *ctx = data; @@ -2608,32 +2576,30 @@ void *consumer_thread_data_poll(void *data) * the fds set has been updated, we need to update our * local array as well */ - pthread_mutex_lock(&consumer_data.lock); - if (consumer_data.need_update) { + pthread_mutex_lock(&the_consumer_data.lock); + if (the_consumer_data.need_update) { free(pollfd); pollfd = NULL; free(local_stream); local_stream = NULL; - /* - * Allocate for all fds + 2: - * +1 for the consumer_data_pipe - * +1 for wake up pipe - */ - nb_pipes_fd = 2; - pollfd = zmalloc((consumer_data.stream_count + nb_pipes_fd) * sizeof(struct pollfd)); + /* Allocate for all fds */ + pollfd = zmalloc((the_consumer_data.stream_count + + nb_pipes_fd) * + sizeof(struct pollfd)); if (pollfd == NULL) { PERROR("pollfd malloc"); - pthread_mutex_unlock(&consumer_data.lock); + pthread_mutex_unlock(&the_consumer_data.lock); goto end; } - local_stream = zmalloc((consumer_data.stream_count + nb_pipes_fd) * + local_stream = zmalloc((the_consumer_data.stream_count + + nb_pipes_fd) * sizeof(struct lttng_consumer_stream *)); if (local_stream == NULL) { PERROR("local_stream malloc"); - pthread_mutex_unlock(&consumer_data.lock); + pthread_mutex_unlock(&the_consumer_data.lock); goto end; } ret = update_poll_array(ctx, &pollfd, local_stream, @@ -2641,13 +2607,13 @@ void *consumer_thread_data_poll(void *data) if (ret < 0) { ERR("Error in allocating pollfd or local_outfds"); lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); - pthread_mutex_unlock(&consumer_data.lock); + pthread_mutex_unlock(&the_consumer_data.lock); goto end; } nb_fd = ret; - consumer_data.need_update = 0; + the_consumer_data.need_update = 0; } - pthread_mutex_unlock(&consumer_data.lock); + pthread_mutex_unlock(&the_consumer_data.lock); /* No FDs and consumer_quit, consumer_cleanup the thread */ if (nb_fd == 0 && nb_inactive_fd == 0 && @@ -2741,7 +2707,7 @@ 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. */ @@ -2772,7 +2738,7 @@ 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. */ @@ -2793,8 +2759,8 @@ void *consumer_thread_data_poll(void *data) } if (!local_stream[i]->hangup_flush_done && (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL)) - && (consumer_data.type == LTTNG_CONSUMER32_UST - || consumer_data.type == LTTNG_CONSUMER64_UST)) { + && (the_consumer_data.type == LTTNG_CONSUMER32_UST + || the_consumer_data.type == LTTNG_CONSUMER64_UST)) { DBG("fd %d is hup|err|nval. Attempting flush and read.", pollfd[i].fd); lttng_ustconsumer_on_stream_hangup(local_stream[i]); @@ -2873,7 +2839,7 @@ void consumer_close_channel_streams(struct lttng_consumer_channel *channel) struct lttng_consumer_stream *stream; struct lttng_ht_iter iter; - ht = consumer_data.stream_per_chan_id_ht; + ht = the_consumer_data.stream_per_chan_id_ht; rcu_read_lock(); cds_lfht_for_each_entry_duplicate(ht->ht, @@ -2887,7 +2853,7 @@ void consumer_close_channel_streams(struct lttng_consumer_channel *channel) if (cds_lfht_is_node_deleted(&stream->node.node)) { goto next; } - switch (consumer_data.type) { + switch (the_consumer_data.type) { case LTTNG_CONSUMER_KERNEL: break; case LTTNG_CONSUMER32_UST: @@ -2906,7 +2872,7 @@ void consumer_close_channel_streams(struct lttng_consumer_channel *channel) break; default: ERR("Unknown consumer_data type"); - assert(0); + abort(); } next: pthread_mutex_unlock(&stream->lock); @@ -2927,7 +2893,7 @@ static void destroy_channel_ht(struct lttng_ht *ht) rcu_read_lock(); cds_lfht_for_each_entry(ht->ht, &iter.iter, channel, wait_fd_node.node) { ret = lttng_ht_del(ht, &iter); - assert(ret != 0); + LTTNG_ASSERT(ret != 0); } rcu_read_unlock(); @@ -3015,11 +2981,6 @@ restart: revents = LTTNG_POLL_GETEV(&events, i); pollfd = LTTNG_POLL_GETFD(&events, i); - if (!revents) { - /* No activity for this FD (poll implementation). */ - continue; - } - if (pollfd == ctx->consumer_channel_pipe[0]) { if (revents & LPOLLIN) { enum consumer_channel_action action; @@ -3070,9 +3031,9 @@ restart: lttng_poll_del(&events, chan->wait_fd); iter.iter.node = &chan->wait_fd_node.node; ret = lttng_ht_del(channel_ht, &iter); - assert(ret == 0); + LTTNG_ASSERT(ret == 0); - switch (consumer_data.type) { + switch (the_consumer_data.type) { case LTTNG_CONSUMER_KERNEL: break; case LTTNG_CONSUMER32_UST: @@ -3083,7 +3044,7 @@ restart: break; default: ERR("Unknown consumer_data type"); - assert(0); + abort(); } /* @@ -3131,7 +3092,7 @@ restart: lttng_ht_lookup(channel_ht, &tmp_id, &iter); } node = lttng_ht_iter_get_node_u64(&iter); - assert(node); + LTTNG_ASSERT(node); chan = caa_container_of(node, struct lttng_consumer_channel, wait_fd_node); @@ -3142,7 +3103,7 @@ restart: lttng_poll_del(&events, chan->wait_fd); ret = lttng_ht_del(channel_ht, &iter); - assert(ret == 0); + LTTNG_ASSERT(ret == 0); /* * This will close the wait fd for each stream associated to @@ -3190,8 +3151,8 @@ static int set_metadata_socket(struct lttng_consumer_local_data *ctx, { int ret; - assert(ctx); - assert(sockpoll); + LTTNG_ASSERT(ctx); + LTTNG_ASSERT(sockpoll); ret = lttng_consumer_poll_socket(sockpoll); if (ret) { @@ -3333,7 +3294,7 @@ void *consumer_thread_sessiond_poll(void *data) err = 0; /* All is OK */ goto end; } - DBG("received command on sock"); + DBG("Received command on sock"); } /* All is OK */ err = 0; @@ -3390,52 +3351,144 @@ error_testpoint: return NULL; } -ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream, +static int post_consume(struct lttng_consumer_stream *stream, + const struct stream_subbuffer *subbuffer, struct lttng_consumer_local_data *ctx) { - ssize_t ret; - int rotate_ret; - bool rotated = false; + size_t i; + int ret = 0; + const size_t count = lttng_dynamic_array_get_count( + &stream->read_subbuffer_ops.post_consume_cbs); - pthread_mutex_lock(&stream->lock); - if (stream->metadata_flag) { - pthread_mutex_lock(&stream->metadata_rdv_lock); + for (i = 0; i < count; i++) { + const post_consume_cb op = *(post_consume_cb *) lttng_dynamic_array_get_element( + &stream->read_subbuffer_ops.post_consume_cbs, + i); + + ret = op(stream, subbuffer, ctx); + if (ret) { + goto end; + } } +end: + return ret; +} - switch (consumer_data.type) { - case LTTNG_CONSUMER_KERNEL: - ret = lttng_kconsumer_read_subbuffer(stream, ctx, &rotated); - break; - case LTTNG_CONSUMER32_UST: - case LTTNG_CONSUMER64_UST: - ret = lttng_ustconsumer_read_subbuffer(stream, ctx, &rotated); +ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream, + struct lttng_consumer_local_data *ctx, + bool locked_by_caller) +{ + ssize_t ret, written_bytes = 0; + int rotation_ret; + struct stream_subbuffer subbuffer = {}; + enum get_next_subbuffer_status get_next_status; + + if (!locked_by_caller) { + stream->read_subbuffer_ops.lock(stream); + } + + if (stream->read_subbuffer_ops.on_wake_up) { + ret = stream->read_subbuffer_ops.on_wake_up(stream); + if (ret) { + goto end; + } + } + + /* + * 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; + } + } + + get_next_status = stream->read_subbuffer_ops.get_next_subbuffer( + stream, &subbuffer); + switch (get_next_status) { + case GET_NEXT_SUBBUFFER_STATUS_OK: break; + case GET_NEXT_SUBBUFFER_STATUS_NO_DATA: + /* Not an error. */ + ret = 0; + goto sleep_stream; + case GET_NEXT_SUBBUFFER_STATUS_ERROR: + ret = -1; + goto end; default: - ERR("Unknown consumer_data type"); - assert(0); - ret = -ENOSYS; - break; + abort(); } - if (stream->metadata_flag) { - pthread_cond_broadcast(&stream->metadata_rdv); - pthread_mutex_unlock(&stream->metadata_rdv_lock); + ret = stream->read_subbuffer_ops.pre_consume_subbuffer( + stream, &subbuffer); + if (ret) { + goto error_put_subbuf; } - pthread_mutex_unlock(&stream->lock); - if (rotated) { - rotate_ret = consumer_post_rotation(stream, ctx); - if (rotate_ret < 0) { - ERR("Failed after a rotation"); - ret = -1; + + written_bytes = stream->read_subbuffer_ops.consume_subbuffer( + ctx, stream, &subbuffer); + if (written_bytes <= 0) { + ERR("Error consuming subbuffer: (%zd)", written_bytes); + ret = (int) written_bytes; + goto error_put_subbuf; + } + + ret = stream->read_subbuffer_ops.put_next_subbuffer(stream, &subbuffer); + if (ret) { + goto end; + } + + ret = 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; + } + +sleep_stream: + 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); } 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) { - switch (consumer_data.type) { + switch (the_consumer_data.type) { case LTTNG_CONSUMER_KERNEL: return lttng_kconsumer_on_recv_stream(stream); case LTTNG_CONSUMER32_UST: @@ -3443,7 +3496,7 @@ int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream) return lttng_ustconsumer_on_recv_stream(stream); default: ERR("Unknown consumer_data type"); - assert(0); + abort(); return -ENOSYS; } } @@ -3453,23 +3506,30 @@ int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream) */ int lttng_consumer_init(void) { - consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); - if (!consumer_data.channel_ht) { + the_consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); + if (!the_consumer_data.channel_ht) { goto error; } - consumer_data.relayd_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); - if (!consumer_data.relayd_ht) { + the_consumer_data.channels_by_session_id_ht = + lttng_ht_new(0, LTTNG_HT_TYPE_U64); + if (!the_consumer_data.channels_by_session_id_ht) { goto error; } - consumer_data.stream_list_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); - if (!consumer_data.stream_list_ht) { + the_consumer_data.relayd_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); + if (!the_consumer_data.relayd_ht) { goto error; } - consumer_data.stream_per_chan_id_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); - if (!consumer_data.stream_per_chan_id_ht) { + the_consumer_data.stream_list_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); + if (!the_consumer_data.stream_list_ht) { + goto error; + } + + the_consumer_data.stream_per_chan_id_ht = + lttng_ht_new(0, LTTNG_HT_TYPE_U64); + if (!the_consumer_data.stream_per_chan_id_ht) { goto error; } @@ -3483,6 +3543,11 @@ int lttng_consumer_init(void) goto error; } + the_consumer_data.chunk_registry = lttng_trace_chunk_registry_create(); + if (!the_consumer_data.chunk_registry) { + goto error; + } + return 0; error: @@ -3505,15 +3570,15 @@ error: enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS; struct consumer_relayd_sock_pair *relayd = NULL; - assert(ctx); - assert(relayd_sock); + LTTNG_ASSERT(ctx); + LTTNG_ASSERT(relayd_sock); DBG("Consumer adding relayd socket (idx: %" PRIu64 ")", net_seq_idx); /* Get relayd reference if exists. */ relayd = consumer_find_relayd(net_seq_idx); if (relayd == NULL) { - assert(sock_type == LTTNG_STREAM_CONTROL); + LTTNG_ASSERT(sock_type == LTTNG_STREAM_CONTROL); /* Not found. Allocate one. */ relayd = consumer_allocate_relayd_sock_pair(net_seq_idx); if (relayd == NULL) { @@ -3533,7 +3598,7 @@ error: /* * relayd key should never be found for control socket. */ - assert(sock_type != LTTNG_STREAM_CONTROL); + LTTNG_ASSERT(sock_type != LTTNG_STREAM_CONTROL); } /* First send a status message before receiving the fds. */ @@ -3593,7 +3658,6 @@ error: /* Assign new file descriptor */ relayd->control_sock.sock.fd = fd; - fd = -1; /* For error path */ /* Assign version values. */ relayd->control_sock.major = relayd_sock->major; relayd->control_sock.minor = relayd_sock->minor; @@ -3621,7 +3685,6 @@ error: /* Assign new file descriptor */ relayd->data_sock.sock.fd = fd; - fd = -1; /* for eventual error paths */ /* Assign version values. */ relayd->data_sock.major = relayd_sock->major; relayd->data_sock.minor = relayd_sock->minor; @@ -3635,6 +3698,11 @@ error: DBG("Consumer %s socket created successfully with net idx %" PRIu64 " (fd: %d)", sock_type == LTTNG_STREAM_CONTROL ? "control" : "data", relayd->net_seq_idx, fd); + /* + * We gave the ownership of the fd to the relayd structure. Set the + * fd to -1 so we don't call close() on it in the error path below. + */ + fd = -1; /* We successfully added the socket. Send status back. */ ret = consumer_send_status_msg(sock, ret_code); @@ -3648,6 +3716,7 @@ error: * Add relayd socket pair to consumer data hashtable. If object already * exists or on error, the function gracefully returns. */ + relayd->ctx = ctx; add_relayd(relayd); /* All good! */ @@ -3671,34 +3740,6 @@ error_nosignal: } } -/* - * Try to lock the stream mutex. - * - * On success, 1 is returned else 0 indicating that the mutex is NOT lock. - */ -static int stream_try_lock(struct lttng_consumer_stream *stream) -{ - int ret; - - assert(stream); - - /* - * Try to lock the stream mutex. On failure, we know that the stream is - * being used else where hence there is data still being extracted. - */ - ret = pthread_mutex_trylock(&stream->lock); - if (ret) { - /* For both EBUSY and EINVAL error, the mutex is NOT locked. */ - ret = 0; - goto end; - } - - ret = 1; - -end: - return ret; -} - /* * Search for a relayd associated to the session id and return the reference. * @@ -3711,8 +3752,8 @@ static struct consumer_relayd_sock_pair *find_relayd_by_session_id(uint64_t id) struct consumer_relayd_sock_pair *relayd = NULL; /* Iterate over all relayd since they are indexed by net_seq_idx. */ - cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd, - node.node) { + cds_lfht_for_each_entry(the_consumer_data.relayd_ht->ht, &iter.iter, + relayd, node.node) { /* * Check by sessiond id which is unique here where the relayd session * id might not be when having multiple relayd. @@ -3747,9 +3788,9 @@ int consumer_data_pending(uint64_t id) DBG("Consumer data pending command on session id %" PRIu64, id); rcu_read_lock(); - pthread_mutex_lock(&consumer_data.lock); + pthread_mutex_lock(&the_consumer_data.lock); - switch (consumer_data.type) { + switch (the_consumer_data.type) { case LTTNG_CONSUMER_KERNEL: data_pending = lttng_kconsumer_data_pending; break; @@ -3759,34 +3800,17 @@ int consumer_data_pending(uint64_t id) break; default: ERR("Unknown consumer data type"); - assert(0); + abort(); } /* Ease our life a bit */ - ht = consumer_data.stream_list_ht; - - relayd = find_relayd_by_session_id(id); - if (relayd) { - /* Send init command for data pending. */ - pthread_mutex_lock(&relayd->ctrl_sock_mutex); - ret = relayd_begin_data_pending(&relayd->control_sock, - relayd->relayd_session_id); - pthread_mutex_unlock(&relayd->ctrl_sock_mutex); - if (ret < 0) { - /* Communication error thus the relayd so no data pending. */ - goto data_not_pending; - } - } + ht = the_consumer_data.stream_list_ht; cds_lfht_for_each_entry_duplicate(ht->ht, ht->hash_fct(&id, lttng_ht_seed), ht->match_fct, &id, &iter.iter, stream, node_session_id.node) { - /* If this call fails, the stream is being used hence data pending. */ - ret = stream_try_lock(stream); - if (!ret) { - goto data_pending; - } + pthread_mutex_lock(&stream->lock); /* * A removed node from the hash table indicates that the stream has @@ -3804,9 +3828,27 @@ int consumer_data_pending(uint64_t id) } } - /* Relayd check */ - if (relayd) { - pthread_mutex_lock(&relayd->ctrl_sock_mutex); + pthread_mutex_unlock(&stream->lock); + } + + relayd = find_relayd_by_session_id(id); + if (relayd) { + unsigned int is_data_inflight = 0; + + /* Send init command for data pending. */ + pthread_mutex_lock(&relayd->ctrl_sock_mutex); + ret = relayd_begin_data_pending(&relayd->control_sock, + relayd->relayd_session_id); + if (ret < 0) { + pthread_mutex_unlock(&relayd->ctrl_sock_mutex); + /* Communication error thus the relayd so no data pending. */ + goto data_not_pending; + } + + cds_lfht_for_each_entry_duplicate(ht->ht, + ht->hash_fct(&id, lttng_ht_seed), + ht->match_fct, &id, + &iter.iter, stream, node_session_id.node) { if (stream->metadata_flag) { ret = relayd_quiescent_control(&relayd->control_sock, stream->relayd_stream_id); @@ -3815,24 +3857,25 @@ int consumer_data_pending(uint64_t id) stream->relayd_stream_id, stream->next_net_seq_num - 1); } - pthread_mutex_unlock(&relayd->ctrl_sock_mutex); + if (ret == 1) { - pthread_mutex_unlock(&stream->lock); + pthread_mutex_unlock(&relayd->ctrl_sock_mutex); goto data_pending; + } else if (ret < 0) { + ERR("Relayd data pending failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); + lttng_consumer_cleanup_relayd(relayd); + pthread_mutex_unlock(&relayd->ctrl_sock_mutex); + goto data_not_pending; } } - pthread_mutex_unlock(&stream->lock); - } - - if (relayd) { - unsigned int is_data_inflight = 0; - /* Send init command for data pending. */ - pthread_mutex_lock(&relayd->ctrl_sock_mutex); + /* Send end command for data pending. */ ret = relayd_end_data_pending(&relayd->control_sock, relayd->relayd_session_id, &is_data_inflight); pthread_mutex_unlock(&relayd->ctrl_sock_mutex); if (ret < 0) { + ERR("Relayd end data pending failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); + lttng_consumer_cleanup_relayd(relayd); goto data_not_pending; } if (is_data_inflight) { @@ -3848,13 +3891,13 @@ int consumer_data_pending(uint64_t id) data_not_pending: /* Data is available to be read by a viewer. */ - pthread_mutex_unlock(&consumer_data.lock); + pthread_mutex_unlock(&the_consumer_data.lock); rcu_read_unlock(); return 0; data_pending: /* Data is still being extracted from buffers. */ - pthread_mutex_unlock(&consumer_data.lock); + pthread_mutex_unlock(&the_consumer_data.lock); rcu_read_unlock(); return 1; } @@ -3884,7 +3927,7 @@ int consumer_send_status_channel(int sock, { struct lttcomm_consumer_status_channel msg; - assert(sock >= 0); + LTTNG_ASSERT(sock >= 0); memset(&msg, 0, sizeof(msg)); if (!channel) { @@ -3907,7 +3950,7 @@ unsigned long consumer_get_consume_start_pos(unsigned long consumed_pos, if (!nb_packets_per_stream) { return consumed_pos; /* Grab everything */ } - start_pos = produced_pos - offset_align_floor(produced_pos, max_sb_size); + start_pos = produced_pos - lttng_offset_align_floor(produced_pos, max_sb_size); start_pos -= max_sb_size * nb_packets_per_stream; if ((long) (start_pos - consumed_pos) < 0) { return consumed_pos; /* Grab everything */ @@ -3915,26 +3958,30 @@ unsigned long consumer_get_consume_start_pos(unsigned long consumed_pos, return start_pos; } -static -int consumer_flush_buffer(struct lttng_consumer_stream *stream, int producer_active) +/* Stream lock must be held by the caller. */ +static int sample_stream_positions(struct lttng_consumer_stream *stream, + unsigned long *produced, unsigned long *consumed) { - int ret = 0; + int ret; - 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; - } - break; - case LTTNG_CONSUMER32_UST: - case LTTNG_CONSUMER64_UST: - lttng_ustctl_flush_buffer(stream, producer_active); - break; - default: - ERR("Unknown consumer_data type"); - abort(); + ASSERT_LOCKED(stream->lock); + + ret = lttng_consumer_sample_snapshot_positions(stream); + if (ret < 0) { + ERR("Failed to sample snapshot positions"); + goto end; + } + + ret = lttng_consumer_get_produced_snapshot(stream, produced); + if (ret < 0) { + ERR("Failed to sample produced position"); + goto end; + } + + ret = lttng_consumer_get_consumed_snapshot(stream, consumed); + if (ret < 0) { + ERR("Failed to sample consumed position"); + goto end; } end: @@ -3946,61 +3993,50 @@ end: * is already at the rotate position (produced == consumed), we flag it as * ready for rotation. The rotation of ready streams occurs after we have * replied to the session daemon that we have finished sampling the positions. + * Must be called with RCU read-side lock held to ensure existence of channel. * * Returns 0 on success, < 0 on error */ -int lttng_consumer_rotate_channel(uint64_t key, const char *path, - uint64_t relayd_id, uint32_t metadata, uint64_t new_chunk_id, +int lttng_consumer_rotate_channel(struct lttng_consumer_channel *channel, + uint64_t key, uint64_t relayd_id, uint32_t metadata, struct lttng_consumer_local_data *ctx) { int ret; - 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_ht *ht = the_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; + /* Array of `struct lttng_consumer_stream *` */ + struct lttng_dynamic_pointer_array streams_packet_to_open; + size_t stream_idx; DBG("Consumer sample rotate position for channel %" PRIu64, key); - rcu_read_lock(); + lttng_dynamic_array_init(&stream_rotation_positions, + sizeof(struct relayd_stream_rotation_position), NULL); + lttng_dynamic_pointer_array_init(&streams_packet_to_open, NULL); - channel = consumer_find_channel(key); - if (!channel) { - ERR("No channel found for key %" PRIu64, key); - ret = -1; - goto end; - } + rcu_read_lock(); pthread_mutex_lock(&channel->lock); - channel->current_chunk_id = new_chunk_id; - - ret = lttng_strncpy(channel->pathname, path, sizeof(channel->pathname)); - if (ret) { - ERR("Failed to copy new path to channel during channel rotation"); + LTTNG_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; } - if (relayd_id == -1ULL) { - /* - * The domain path (/ust or /kernel) has been created before, we - * now need to create the last part of the path: the application/user - * specific section (uid/1000/64-bit). - */ - ret = utils_mkdir_recursive(channel->pathname, S_IRWXU | S_IRWXG, - channel->uid, channel->gid); - if (ret < 0) { - ERR("Failed to create trace directory at %s during rotation", - channel->pathname); - 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(); @@ -4009,262 +4045,609 @@ int lttng_consumer_rotate_channel(uint64_t key, const char *path, */ pthread_mutex_lock(&stream->lock); - ret = lttng_strncpy(stream->channel_read_only_attributes.path, - channel->pathname, - sizeof(stream->channel_read_only_attributes.path)); - if (ret) { - ERR("Failed to sample channel path name during channel rotation"); - goto end_unlock_stream; + if (stream->trace_chunk == stream->chan->trace_chunk) { + 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 a 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. No new data can be produced when a stream has no + * associated trace chunk (e.g. a stop followed by a rotate). + */ + if (stream->trace_chunk) { + bool flush_active; + + if (stream->metadata_flag) { + /* + * Don't produce an empty metadata packet, + * simply close the current one. + * + * Metadata is regenerated on every trace chunk + * switch; there is no concern that no data was + * produced. + */ + flush_active = true; + } else { + /* + * Only flush an empty packet if the "packet + * open" could not be performed on transition + * to a new trace chunk and no packets were + * consumed within the chunk's lifetime. + */ + if (stream->opened_packet_in_current_trace_chunk) { + flush_active = true; + } else { + /* + * Stream could have been full at the + * time of rotation, but then have had + * no activity at all. + * + * It is important to flush a packet + * to prevent 0-length files from being + * produced as most viewers choke on + * them. + * + * Unfortunately viewers will not be + * able to know that tracing was active + * for this stream during this trace + * chunk's lifetime. + */ + ret = sample_stream_positions(stream, &produced_pos, &consumed_pos); + if (ret) { + goto end_unlock_stream; + } + + /* + * Don't flush an empty packet if data + * was produced; it will be consumed + * before the rotation completes. + */ + flush_active = produced_pos != consumed_pos; + if (!flush_active) { + const char *trace_chunk_name; + uint64_t trace_chunk_id; + + chunk_status = lttng_trace_chunk_get_name( + stream->trace_chunk, + &trace_chunk_name, + NULL); + if (chunk_status == LTTNG_TRACE_CHUNK_STATUS_NONE) { + trace_chunk_name = "none"; + } + + /* + * Consumer trace chunks are + * never anonymous. + */ + chunk_status = lttng_trace_chunk_get_id( + stream->trace_chunk, + &trace_chunk_id); + LTTNG_ASSERT(chunk_status == + LTTNG_TRACE_CHUNK_STATUS_OK); + + DBG("Unable to open packet for stream during trace chunk's lifetime. " + "Flushing an empty packet to prevent an empty file from being created: " + "stream id = %" PRIu64 ", trace chunk name = `%s`, trace chunk id = %" PRIu64, + stream->key, trace_chunk_name, trace_chunk_id); + } + } + } + + /* + * Close the current packet before sampling the + * ring buffer positions. + */ + ret = consumer_stream_flush_buffer(stream, flush_active); + 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 = lttng_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); } - channel->nr_stream_rotate_pending++; - - 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++; + } + + stream->opened_packet_in_current_trace_chunk = false; + + if (rotating_to_new_chunk && !stream->metadata_flag) { + /* + * Attempt to flush an empty packet as close to the + * rotation point as possible. In the event where a + * stream remains inactive after the rotation point, + * this ensures that the new trace chunk has a + * beginning timestamp set at the begining of the + * trace chunk instead of only creating an empty + * packet when the trace chunk is stopped. + * + * This indicates to the viewers that the stream + * was being recorded, but more importantly it + * allows viewers to determine a useable trace + * intersection. + * + * This presents a problem in the case where the + * ring-buffer is completely full. + * + * Consider the following scenario: + * - The consumption of data is slow (slow network, + * for instance), + * - The ring buffer is full, + * - A rotation is initiated, + * - The flush below does nothing (no space left to + * open a new packet), + * - The other streams rotate very soon, and new + * data is produced in the new chunk, + * - This stream completes its rotation long after the + * rotation was initiated + * - The session is stopped before any event can be + * produced in this stream's buffers. + * + * The resulting trace chunk will have a single packet + * temporaly at the end of the trace chunk for this + * stream making the stream intersection more narrow + * than it should be. + * + * To work-around this, an empty flush is performed + * after the first consumption of a packet during a + * rotation if open_packet fails. The idea is that + * consuming a packet frees enough space to switch + * packets in this scenario and allows the tracer to + * "stamp" the beginning of the new trace chunk at the + * earliest possible point. + * + * The packet open is performed after the channel + * rotation to ensure that no attempt to open a packet + * is performed in a stream that has no active trace + * chunk. + */ + ret = lttng_dynamic_pointer_array_add_pointer( + &streams_packet_to_open, stream); + if (ret) { + PERROR("Failed to add a stream pointer to array of streams in which to open a packet"); + ret = -1; + goto end_unlock_stream; + } + } pthread_mutex_unlock(&stream->lock); } - pthread_mutex_unlock(&channel->lock); + stream = NULL; - ret = 0; - goto end; + if (!is_local_trace) { + relayd = consumer_find_relayd(relayd_id); + if (!relayd) { + ERR("Failed to find relayd %" PRIu64, relayd_id); + ret = -1; + goto end_unlock_channel; + } + + 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_unlock_channel; + } + } + + for (stream_idx = 0; + stream_idx < lttng_dynamic_pointer_array_get_count( + &streams_packet_to_open); + stream_idx++) { + enum consumer_stream_open_packet_status status; + + stream = lttng_dynamic_pointer_array_get_pointer( + &streams_packet_to_open, stream_idx); + + pthread_mutex_lock(&stream->lock); + status = consumer_stream_open_packet(stream); + pthread_mutex_unlock(&stream->lock); + switch (status) { + case CONSUMER_STREAM_OPEN_PACKET_STATUS_OPENED: + DBG("Opened a packet after a rotation: stream id = %" PRIu64 + ", channel name = %s, session id = %" PRIu64, + stream->key, stream->chan->name, + stream->chan->session_id); + break; + case CONSUMER_STREAM_OPEN_PACKET_STATUS_NO_SPACE: + /* + * Can't open a packet as there is no space left + * in the buffer. A new packet will be opened + * once one has been consumed. + */ + DBG("No space left to open a packet after a rotation: stream id = %" PRIu64 + ", channel name = %s, session id = %" PRIu64, + stream->key, stream->chan->name, + stream->chan->session_id); + break; + case CONSUMER_STREAM_OPEN_PACKET_STATUS_ERROR: + /* Logged by callee. */ + ret = -1; + goto end_unlock_channel; + default: + abort(); + } + } -end_unlock_stream: - pthread_mutex_unlock(&stream->lock); -end_unlock_channel: + pthread_mutex_unlock(&channel->lock); + 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); + lttng_dynamic_pointer_array_reset(&streams_packet_to_open); 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 (the_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: + ret = lttng_ustconsumer_clear_buffer(stream); + if (ret < 0) { + ERR("Failed to clear ust stream (ret = %d)", ret); + goto end; + } + 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; } - - /* Rotate position not reached yet (with check for overflow). */ - if ((long) (consumed_pos - stream->rotate_position) < 0) { - ret = 0; - goto end; - } - ret = 1; - + DBG("clear: before: %lu after: %lu", consumed_pos_before, consumed_pos_after); end: return ret; } -/* - * 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; - stream->rotate_ready = false; -} - -/* - * Perform the rotation a local stream file. - */ -int rotate_local_stream(struct lttng_consumer_local_data *ctx, - struct lttng_consumer_stream *stream) +static +int consumer_clear_stream(struct lttng_consumer_stream *stream) { int ret; - DBG("Rotate local stream: stream key %" PRIu64 ", channel key %" PRIu64 " at path %s", - stream->key, - stream->chan->key, - stream->channel_read_only_attributes.path); - - ret = close(stream->out_fd); + ret = consumer_stream_flush_buffer(stream, 1); if (ret < 0) { - PERROR("Closing trace file (fd %d), stream %" PRIu64, - stream->out_fd, stream->key); - assert(0); + ERR("Failed to flush stream %" PRIu64 " during channel clear", + stream->key); + ret = LTTCOMM_CONSUMERD_FATAL; goto error; } - ret = utils_create_stream_file( - stream->channel_read_only_attributes.path, - stream->name, - stream->channel_read_only_attributes.tracefile_size, - stream->tracefile_count_current, - stream->uid, stream->gid, NULL); + ret = consumer_clear_buffer(stream); if (ret < 0) { - ERR("Rotate create stream file"); + ERR("Failed to clear stream %" PRIu64 " during channel clear", + stream->key); + ret = LTTCOMM_CONSUMERD_FATAL; goto error; } - stream->out_fd = ret; - stream->tracefile_size_current = 0; - if (!stream->metadata_flag) { - struct lttng_index_file *index_file; + ret = LTTCOMM_CONSUMERD_SUCCESS; +error: + return ret; +} - lttng_index_file_put(stream->index_file); +static +int consumer_clear_unmonitored_channel(struct lttng_consumer_channel *channel) +{ + int ret; + struct lttng_consumer_stream *stream; - index_file = lttng_index_file_create( - stream->channel_read_only_attributes.path, - stream->name, stream->uid, stream->gid, - stream->channel_read_only_attributes.tracefile_size, - stream->tracefile_count_current, - CTF_INDEX_MAJOR, CTF_INDEX_MINOR); - if (!index_file) { - ERR("Create index file during rotation"); - goto error; + 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; } - stream->index_file = index_file; - stream->out_fd_offset = 0; + pthread_mutex_unlock(&stream->lock); } + pthread_mutex_unlock(&channel->lock); + rcu_read_unlock(); + return 0; - ret = 0; - goto end; - -error: - ret = -1; -end: +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) +{ + DBG("lttng_consumer_reset_stream_rotate_state for stream %" PRIu64, + stream->key); + stream->rotate_position = -1ULL; + stream->rotate_ready = false; } /* - * Perform the rotation a stream file on the relay. + * Perform the rotation a local stream file. */ -int rotate_relay_stream(struct lttng_consumer_local_data *ctx, +static +int rotate_local_stream(struct lttng_consumer_local_data *ctx, struct lttng_consumer_stream *stream) { - int ret; - struct consumer_relayd_sock_pair *relayd; + int ret = 0; - DBG("Rotate relay stream"); - relayd = consumer_find_relayd(stream->net_seq_idx); - if (!relayd) { - ERR("Failed to find relayd"); - ret = -1; - goto end; + DBG("Rotate local stream: stream key %" PRIu64 ", channel key %" PRIu64, + stream->key, + stream->chan->key); + stream->tracefile_size_current = 0; + stream->tracefile_count_current = 0; + + if (stream->out_fd >= 0) { + ret = close(stream->out_fd); + if (ret) { + PERROR("Failed to close stream out_fd of channel \"%s\"", + stream->chan->name); + } + stream->out_fd = -1; } - pthread_mutex_lock(&relayd->ctrl_sock_mutex); - ret = relayd_rotate_stream(&relayd->control_sock, - stream->relayd_stream_id, - stream->channel_read_only_attributes.path, - stream->chan->current_chunk_id, - stream->last_sequence_number); - pthread_mutex_unlock(&relayd->ctrl_sock_mutex); - if (ret) { - ERR("Rotate relay stream"); + if (stream->index_file) { + lttng_index_file_put(stream->index_file); + stream->index_file = NULL; + } + + if (!stream->trace_chunk) { + goto end; } + ret = consumer_stream_create_output_files(stream, true); end: return ret; } /* * Performs the stream rotation for the rotate session feature if needed. - * It must be called with the stream lock held. + * It must be called with the channel and stream locks held. * * Return 0 on success, a negative number of error. */ int lttng_consumer_rotate_stream(struct lttng_consumer_local_data *ctx, - struct lttng_consumer_stream *stream, bool *rotated) + struct lttng_consumer_stream *stream) { int ret; DBG("Consumer rotate stream %" PRIu64, stream->key); - if (stream->net_seq_idx != (uint64_t) -1ULL) { - ret = rotate_relay_stream(ctx, stream); - } else { - ret = rotate_local_stream(ctx, stream); - } - if (ret < 0) { - ERR("Rotate stream"); + /* + * Update the stream's 'current' chunk to the session's (channel) + * now-current chunk. + */ + lttng_trace_chunk_put(stream->trace_chunk); + if (stream->chan->trace_chunk == stream->trace_chunk) { + /* + * A channel can be rotated and not have a "next" chunk + * to transition to. In that case, the channel's "current chunk" + * has not been closed yet, but it has not been updated to + * a "next" trace chunk either. Hence, the stream, like its + * parent channel, becomes part of no chunk and can't output + * anything until a new trace chunk is created. + */ + stream->trace_chunk = NULL; + } else if (stream->chan->trace_chunk && + !lttng_trace_chunk_get(stream->chan->trace_chunk)) { + ERR("Failed to acquire a reference to channel's trace chunk during stream rotation"); + ret = -1; goto error; + } else { + /* + * Update the stream's trace chunk to its parent channel's + * current trace chunk. + */ + stream->trace_chunk = stream->chan->trace_chunk; } - if (stream->metadata_flag) { - switch (consumer_data.type) { - case LTTNG_CONSUMER_KERNEL: - /* - * Reset the position of what has been read from the metadata - * cache to 0 so we can dump it again. - */ - ret = kernctl_metadata_cache_dump(stream->wait_fd); - if (ret < 0) { - ERR("Failed to dump the kernel metadata cache after rotation"); - goto error; - } - break; - case LTTNG_CONSUMER32_UST: - case LTTNG_CONSUMER64_UST: - /* - * Reset the position pushed from the metadata cache so it - * will write from the beginning on the next push. - */ - stream->ust_metadata_pushed = 0; - break; - default: - ERR("Unknown consumer_data type"); - abort(); + 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; } } - lttng_consumer_reset_stream_rotate_state(stream); - if (rotated) { - *rotated = true; + if (stream->metadata_flag && stream->trace_chunk) { + /* + * If the stream has transitioned to a new trace + * chunk, the metadata should be re-dumped to the + * newest chunk. + * + * However, it is possible for a stream to transition to + * a "no-chunk" state. This can happen if a rotation + * occurs on an inactive session. In such cases, the metadata + * regeneration will happen when the next trace chunk is + * created. + */ + ret = consumer_metadata_stream_dump(stream); + if (ret) { + goto error; + } } + lttng_consumer_reset_stream_rotate_state(stream); ret = 0; @@ -4278,50 +4661,42 @@ error: * This is especially important for low throughput streams that have already * been consumed, we cannot wait for their next packet to perform the * rotation. + * Need to be called with RCU read-side lock held to ensure existence of + * channel. * * Returns 0 on success, < 0 on error */ -int lttng_consumer_rotate_ready_streams(uint64_t key, - struct lttng_consumer_local_data *ctx) +int lttng_consumer_rotate_ready_streams(struct lttng_consumer_channel *channel, + uint64_t key, struct lttng_consumer_local_data *ctx) { int ret; - 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_ht *ht = the_consumer_data.stream_per_chan_id_ht; rcu_read_lock(); DBG("Consumer rotate ready streams in channel %" PRIu64, key); - channel = consumer_find_channel(key); - if (!channel) { - ERR("No channel found for key %" PRIu64, key); - ret = -1; - goto end; - } - 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) { health_code_update(); + pthread_mutex_lock(&stream->chan->lock); pthread_mutex_lock(&stream->lock); if (!stream->rotate_ready) { pthread_mutex_unlock(&stream->lock); + pthread_mutex_unlock(&stream->chan->lock); continue; } DBG("Consumer rotate ready stream %" PRIu64, stream->key); - ret = lttng_consumer_rotate_stream(ctx, stream, NULL); + ret = lttng_consumer_rotate_stream(ctx, stream); pthread_mutex_unlock(&stream->lock); - if (ret) { - goto end; - } - - ret = consumer_post_rotation(stream, ctx); + pthread_mutex_unlock(&stream->chan->lock); if (ret) { goto end; } @@ -4334,128 +4709,551 @@ end: return ret; } -static -int rotate_rename_local(const char *old_path, const char *new_path, - uid_t uid, gid_t gid) +enum lttcomm_return_code lttng_consumer_init_command( + struct lttng_consumer_local_data *ctx, + const lttng_uuid sessiond_uuid) { - int ret; - - assert(old_path); - assert(new_path); - - ret = utils_mkdir_recursive(new_path, S_IRWXU | S_IRWXG, uid, gid); - if (ret < 0) { - ERR("Create directory on rotate"); - goto end; - } + enum lttcomm_return_code ret; + char uuid_str[LTTNG_UUID_STR_LEN]; - ret = rename(old_path, new_path); - if (ret < 0 && errno != ENOENT) { - PERROR("Rename completed rotation chunk"); + if (ctx->sessiond_uuid.is_set) { + ret = LTTCOMM_CONSUMERD_ALREADY_SET; goto end; } - ret = 0; + ctx->sessiond_uuid.is_set = true; + memcpy(ctx->sessiond_uuid.value, sessiond_uuid, sizeof(lttng_uuid)); + ret = LTTCOMM_CONSUMERD_SUCCESS; + lttng_uuid_to_str(sessiond_uuid, uuid_str); + DBG("Received session daemon UUID: %s", uuid_str); end: return ret; } -static -int rotate_rename_relay(const char *old_path, const char *new_path, - uint64_t relayd_id) +enum lttcomm_return_code lttng_consumer_create_trace_chunk( + const uint64_t *relayd_id, uint64_t session_id, + uint64_t chunk_id, + time_t chunk_creation_timestamp, + const char *chunk_override_name, + const struct lttng_credentials *credentials, + struct lttng_directory_handle *chunk_directory_handle) { int ret; - struct consumer_relayd_sock_pair *relayd; + enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS; + 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]; + const char *relayd_id_str = "(none)"; + const char *creation_timestamp_str; + struct lttng_ht_iter iter; + struct lttng_consumer_channel *channel; - relayd = consumer_find_relayd(relayd_id); - if (!relayd) { - ERR("Failed to find relayd while running rotate_rename_relay command"); - ret = -1; - goto end; + if (relayd_id) { + /* Only used for logging purposes. */ + ret = snprintf(relayd_id_buffer, sizeof(relayd_id_buffer), + "%" PRIu64, *relayd_id); + if (ret > 0 && ret < sizeof(relayd_id_buffer)) { + relayd_id_str = relayd_id_buffer; + } else { + relayd_id_str = "(formatting error)"; + } } - pthread_mutex_lock(&relayd->ctrl_sock_mutex); - ret = relayd_rotate_rename(&relayd->control_sock, old_path, new_path); - pthread_mutex_unlock(&relayd->ctrl_sock_mutex); -end: - return ret; + /* Local protocol error. */ + LTTNG_ASSERT(chunk_creation_timestamp); + ret = time_to_iso8601_str(chunk_creation_timestamp, + creation_timestamp_buffer, + sizeof(creation_timestamp_buffer)); + creation_timestamp_str = !ret ? creation_timestamp_buffer : + "(formatting error)"; + + DBG("Consumer create trace chunk command: relay_id = %s" + ", session_id = %" PRIu64 ", chunk_id = %" PRIu64 + ", chunk_override_name = %s" + ", chunk_creation_timestamp = %s", + relayd_id_str, session_id, chunk_id, + chunk_override_name ? : "(none)", + creation_timestamp_str); + + /* + * The trace chunk registry, as used by the consumer daemon, implicitly + * owns the trace chunks. This is only needed in the consumer since + * the consumer has no notion of a session beyond session IDs being + * used to identify other objects. + * + * The lttng_trace_chunk_registry_publish() call below provides a + * reference which is not released; it implicitly becomes the session + * daemon's reference to the chunk in the consumer daemon. + * + * The lifetime of trace chunks in the consumer daemon is managed by + * the session daemon through the LTTNG_CONSUMER_CREATE_TRACE_CHUNK + * and LTTNG_CONSUMER_DESTROY_TRACE_CHUNK commands. + */ + created_chunk = lttng_trace_chunk_create(chunk_id, + chunk_creation_timestamp, NULL); + if (!created_chunk) { + ERR("Failed to create trace chunk"); + ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED; + goto error; + } + + if (chunk_override_name) { + chunk_status = lttng_trace_chunk_override_name(created_chunk, + chunk_override_name); + if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { + ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED; + goto error; + } + } + + if (chunk_directory_handle) { + chunk_status = lttng_trace_chunk_set_credentials(created_chunk, + credentials); + if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { + ERR("Failed to set trace chunk credentials"); + ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED; + goto error; + } + /* + * The consumer daemon has no ownership of the chunk output + * directory. + */ + 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 error; + } + } + + published_chunk = lttng_trace_chunk_registry_publish_chunk( + the_consumer_data.chunk_registry, session_id, + created_chunk); + lttng_trace_chunk_put(created_chunk); + created_chunk = NULL; + if (!published_chunk) { + ERR("Failed to publish trace chunk"); + ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED; + goto error; + } + + rcu_read_lock(); + cds_lfht_for_each_entry_duplicate( + the_consumer_data.channels_by_session_id_ht->ht, + the_consumer_data.channels_by_session_id_ht->hash_fct( + &session_id, lttng_ht_seed), + the_consumer_data.channels_by_session_id_ht->match_fct, + &session_id, &iter.iter, channel, + channels_by_session_id_ht_node.node) { + ret = lttng_consumer_channel_set_trace_chunk(channel, + published_chunk); + if (ret) { + /* + * Roll-back the creation of this chunk. + * + * This is important since the session daemon will + * assume that the creation of this chunk failed and + * will never ask for it to be closed, resulting + * in a leak and an inconsistent state for some + * 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, + 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; + break; + } + } + + 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); + lttng_trace_chunk_put(created_chunk); + return ret_code; } -int lttng_consumer_rotate_rename(const char *old_path, const char *new_path, - uid_t uid, gid_t gid, uint64_t relayd_id) +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, + char *path) { - if (relayd_id != -1ULL) { - return rotate_rename_relay(old_path, new_path, relayd_id); - } else { - return rotate_rename_local(old_path, new_path, uid, gid); + 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; + + if (relayd_id) { + int ret; + + /* Only used for logging purposes. */ + ret = snprintf(relayd_id_buffer, sizeof(relayd_id_buffer), + "%" PRIu64, *relayd_id); + if (ret > 0 && ret < sizeof(relayd_id_buffer)) { + relayd_id_str = relayd_id_buffer; + } 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 + ", close command = %s", + relayd_id_str, session_id, chunk_id, + close_command_name); + + chunk = lttng_trace_chunk_registry_find_chunk( + the_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); + ret_code = LTTCOMM_CONSUMERD_UNKNOWN_TRACE_CHUNK; + goto end; } + + chunk_status = lttng_trace_chunk_set_close_timestamp(chunk, + chunk_close_timestamp); + if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { + ret_code = LTTCOMM_CONSUMERD_CLOSE_TRACE_CHUNK_FAILED; + goto end; + } + + 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 + * it; it is only kept around to compare it (by address) to the + * current chunk found in the session's channels. + */ + rcu_read_lock(); + cds_lfht_for_each_entry(the_consumer_data.channel_ht->ht, &iter.iter, + channel, node.node) { + int ret; + + /* + * Only change the channel's chunk to NULL if it still + * references the chunk being closed. The channel may + * reference a newer channel in the case of a session + * rotation. When a session rotation occurs, the "next" + * chunk is created before the "current" chunk is closed. + */ + if (channel->trace_chunk != chunk) { + continue; + } + ret = lttng_consumer_channel_set_trace_chunk(channel, NULL); + if (ret) { + /* + * Attempt to close the chunk on as many channels as + * possible. + */ + 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; } -int lttng_consumer_rotate_pending_relay(uint64_t session_id, - uint64_t relayd_id, uint64_t chunk_id) +enum lttcomm_return_code lttng_consumer_trace_chunk_exists( + const uint64_t *relayd_id, uint64_t session_id, + uint64_t chunk_id) { int ret; - struct consumer_relayd_sock_pair *relayd; + enum lttcomm_return_code ret_code; + 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) { + /* Only used for logging purposes. */ + ret = snprintf(relayd_id_buffer, sizeof(relayd_id_buffer), + "%" PRIu64, *relayd_id); + if (ret > 0 && ret < sizeof(relayd_id_buffer)) { + relayd_id_str = relayd_id_buffer; + } else { + relayd_id_str = "(formatting error)"; + } + } - relayd = consumer_find_relayd(relayd_id); - if (!relayd) { - ERR("Failed to find relayd"); - ret = -1; + DBG("Consumer trace chunk exists command: relayd_id = %s" + ", chunk_id = %" PRIu64, relayd_id_str, + chunk_id); + ret = lttng_trace_chunk_registry_chunk_exists( + the_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_rotate_pending(&relayd->control_sock, chunk_id); + 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"); +end_rcu_unlock: + rcu_read_unlock(); end: - return ret; + return ret_code; } static -int mkdir_local(const char *path, uid_t uid, gid_t gid) +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; - ret = utils_mkdir_recursive(path, S_IRWXU | S_IRWXG, uid, gid); - if (ret < 0) { - /* utils_mkdir_recursive logs an error. */ + ht = the_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; } - ret = 0; + if (!channel->monitor) { + ret = consumer_clear_unmonitored_channel(channel); + } else { + ret = consumer_clear_monitored_channel(channel); + } end: return ret; } -static -int mkdir_relay(const char *path, uint64_t relayd_id) +enum lttcomm_return_code lttng_consumer_open_channel_packets( + struct lttng_consumer_channel *channel) { - int ret; - struct consumer_relayd_sock_pair *relayd; + struct lttng_consumer_stream *stream; + enum lttcomm_return_code ret = LTTCOMM_CONSUMERD_SUCCESS; - relayd = consumer_find_relayd(relayd_id); - if (!relayd) { - ERR("Failed to find relayd"); - ret = -1; + if (channel->metadata_stream) { + ERR("Open channel packets command attempted on a metadata channel"); + ret = LTTCOMM_CONSUMERD_INVALID_PARAMETERS; goto end; } - pthread_mutex_lock(&relayd->ctrl_sock_mutex); - ret = relayd_mkdir(&relayd->control_sock, path); - pthread_mutex_unlock(&relayd->ctrl_sock_mutex); + rcu_read_lock(); + cds_list_for_each_entry(stream, &channel->streams.head, send_node) { + enum consumer_stream_open_packet_status status; + + pthread_mutex_lock(&stream->lock); + if (cds_lfht_is_node_deleted(&stream->node.node)) { + goto next; + } + status = consumer_stream_open_packet(stream); + switch (status) { + case CONSUMER_STREAM_OPEN_PACKET_STATUS_OPENED: + DBG("Opened a packet in \"open channel packets\" command: stream id = %" PRIu64 + ", channel name = %s, session id = %" PRIu64, + stream->key, stream->chan->name, + stream->chan->session_id); + stream->opened_packet_in_current_trace_chunk = true; + break; + case CONSUMER_STREAM_OPEN_PACKET_STATUS_NO_SPACE: + DBG("No space left to open a packet in \"open channel packets\" command: stream id = %" PRIu64 + ", channel name = %s, session id = %" PRIu64, + stream->key, stream->chan->name, + stream->chan->session_id); + break; + case CONSUMER_STREAM_OPEN_PACKET_STATUS_ERROR: + /* + * Only unexpected internal errors can lead to this + * failing. Report an unknown error. + */ + ERR("Failed to flush empty buffer in \"open channel packets\" command: stream id = %" PRIu64 + ", channel id = %" PRIu64 + ", channel name = %s" + ", session id = %" PRIu64, + stream->key, channel->key, + channel->name, channel->session_id); + ret = LTTCOMM_CONSUMERD_UNKNOWN_ERROR; + goto error_unlock; + default: + abort(); + } + + next: + pthread_mutex_unlock(&stream->lock); + } + +end_rcu_unlock: + rcu_read_unlock(); end: return ret; +error_unlock: + pthread_mutex_unlock(&stream->lock); + goto end_rcu_unlock; } -int lttng_consumer_mkdir(const char *path, uid_t uid, gid_t gid, - uint64_t relayd_id) +void lttng_consumer_sigbus_handle(void *addr) { - if (relayd_id != -1ULL) { - return mkdir_relay(path, relayd_id); - } else { - return mkdir_local(path, uid, gid); - } + lttng_ustconsumer_sigbus_handle(addr); }