X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fcommon%2Fconsumer%2Fconsumer.c;h=8d9fa06478cddc7a83904e76664aa21550cf527c;hp=3755932ab59cf8ad177607b5ad64c95a9c76bcae;hb=514775d9bca89b3bd072c58e779201682304c57d;hpb=9a2fcf78ea85b9ad74405db4605338896d70b46f diff --git a/src/common/consumer/consumer.c b/src/common/consumer/consumer.c index 3755932ab..8d9fa0647 100644 --- a/src/common/consumer/consumer.c +++ b/src/common/consumer/consumer.c @@ -1,22 +1,13 @@ /* - * Copyright (C) 2011 - Julien Desfossez - * Mathieu Desnoyers - * 2012 - David Goulet + * Copyright (C) 2011 Julien Desfossez + * Copyright (C) 2011 Mathieu Desnoyers + * Copyright (C) 2012 David Goulet * - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License, version 2 only, - * as published by the Free Software Foundation. + * SPDX-License-Identifier: GPL-2.0-only * - * This program is distributed in the hope that it will be useful, but WITHOUT - * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or - * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for - * more details. - * - * You should have received a copy of the GNU General Public License along - * with this program; if not, write to the Free Software Foundation, Inc., - * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. */ +#include "common/index/ctf-index.h" #define _LGPL_SOURCE #include #include @@ -33,6 +24,7 @@ #include #include #include +#include #include #include #include @@ -48,6 +40,10 @@ #include #include #include +#include +#include +#include +#include struct lttng_consumer_global_data consumer_data = { .stream_count = 0, @@ -86,6 +82,20 @@ int consumer_quit; static struct lttng_ht *metadata_ht; static struct lttng_ht *data_ht; +static const char *get_consumer_domain(void) +{ + switch (consumer_data.type) { + case LTTNG_CONSUMER_KERNEL: + return DEFAULT_KERNEL_TRACE_DIR; + case LTTNG_CONSUMER64_UST: + /* Fall-through. */ + case LTTNG_CONSUMER32_UST: + return DEFAULT_UST_TRACE_DIR; + default: + abort(); + } +} + /* * Notify a thread lttng pipe to poll back again. This usually means that some * global state has changed so we just send back the thread in a poll wait @@ -323,6 +333,7 @@ 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); } @@ -357,7 +368,6 @@ 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); @@ -388,12 +398,25 @@ void consumer_del_channel(struct lttng_consumer_channel *channel) 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(consumer_data.channel_ht, &iter); + assert(!ret); + iter.iter.node = &channel->channels_by_session_id_ht_node.node; + ret = lttng_ht_del(consumer_data.channels_by_session_id_ht, + &iter); + assert(!ret); + rcu_read_unlock(); + } + + channel->is_deleted = true; call_rcu(&channel->node.head, free_channel_rcu); end: pthread_mutex_unlock(&channel->lock); @@ -464,14 +487,13 @@ 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); - 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 +513,8 @@ 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); } /* @@ -541,90 +561,12 @@ void consumer_del_stream_for_metadata(struct lttng_consumer_stream *stream) consumer_stream_destroy(stream, metadata_ht); } -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) +void consumer_stream_update_channel_attributes( + struct lttng_consumer_stream *stream, + struct lttng_consumer_channel *channel) { - 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; - 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; + stream->channel_read_only_attributes.tracefile_size = + channel->tracefile_size; } /* @@ -684,11 +626,6 @@ void consumer_add_data_stream(struct lttng_consumer_stream *stream) pthread_mutex_unlock(&consumer_data.lock); } -void consumer_del_data_stream(struct lttng_consumer_stream *stream) -{ - consumer_del_stream(stream, data_ht); -} - /* * Add relayd socket to global consumer data hashtable. RCU read side lock MUST * be acquired before calling this. @@ -795,10 +732,14 @@ 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); + 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; } @@ -840,6 +781,8 @@ 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 { @@ -907,6 +850,7 @@ 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 @@ -933,6 +877,86 @@ error: return outfd; } +/* + * 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); + assert(stream->metadata_flag); + assert(stream->chan->trace_chunk); + + 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); + 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); + + 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. @@ -941,10 +965,9 @@ 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, @@ -952,10 +975,22 @@ 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( + 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) { @@ -967,13 +1002,12 @@ 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); @@ -1020,15 +1054,29 @@ 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 +1100,10 @@ int consumer_add_channel(struct lttng_consumer_channel *channel, rcu_read_lock(); lttng_ht_add_unique_u64(consumer_data.channel_ht, &channel->node); + lttng_ht_add_u64(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); @@ -1102,16 +1153,13 @@ 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; @@ -1199,6 +1247,7 @@ void lttng_consumer_cleanup(void) { struct lttng_ht_iter iter; struct lttng_consumer_channel *channel; + unsigned int trace_chunks_left; rcu_read_lock(); @@ -1210,6 +1259,7 @@ void lttng_consumer_cleanup(void) rcu_read_unlock(); lttng_ht_destroy(consumer_data.channel_ht); + lttng_ht_destroy(consumer_data.channels_by_session_id_ht); cleanup_relayd_ht(); @@ -1221,6 +1271,29 @@ void lttng_consumer_cleanup(void) * it. */ lttng_ht_destroy(consumer_data.stream_list_ht); + + /* + * Trace chunks in the registry may still exist if the session + * daemon has encountered an internal error and could not + * tear down its sessions and/or trace chunks properly. + * + * Release the session daemon's implicit reference to any remaining + * trace chunk and print an error if any trace chunk was found. Note + * that there are _no_ legitimate cases for trace chunks to be left, + * it is a leak. However, it can happen following a crash of the + * session daemon and not emptying the registry would cause an assertion + * to hit. + */ + trace_chunks_left = lttng_trace_chunk_registry_put_each_chunk( + consumer_data.chunk_registry); + if (trace_chunks_left) { + ERR("%u trace chunks are leaked by lttng-consumerd. " + "This can be caused by an internal error of the session daemon.", + trace_chunks_left); + } + /* Run all callbacks freeing each chunk. */ + rcu_barrier(); + lttng_trace_chunk_registry_destroy(consumer_data.chunk_registry); } /* @@ -1303,7 +1376,7 @@ void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream, struct lttng_consumer_local_data *lttng_consumer_create( enum lttng_consumer_type type, ssize_t (*buffer_ready)(struct lttng_consumer_stream *stream, - struct lttng_consumer_local_data *ctx), + struct lttng_consumer_local_data *ctx, bool locked_by_caller), int (*recv_channel)(struct lttng_consumer_channel *channel), int (*recv_stream)(struct lttng_consumer_stream *stream), int (*update_stream)(uint64_t stream_key, uint32_t state)) @@ -1465,7 +1538,7 @@ void lttng_consumer_destroy(struct lttng_consumer_local_data *ctx) */ static int write_relayd_metadata_id(int fd, struct lttng_consumer_stream *stream, - struct consumer_relayd_sock_pair *relayd, unsigned long padding) + unsigned long padding) { ssize_t ret; struct lttcomm_relayd_metadata_payload hdr; @@ -1503,29 +1576,30 @@ 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(); + 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) { @@ -1536,39 +1610,9 @@ ssize_t lttng_consumer_on_read_subbuffer_mmap( } } - /* get the offset inside the fd to mmap */ - switch (consumer_data.type) { - case LTTNG_CONSUMER_KERNEL: - mmap_base = stream->mmap_base; - ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset); - if (ret < 0) { - PERROR("tracer ctl get_mmap_read_offset"); - goto end; - } - break; - case LTTNG_CONSUMER32_UST: - case LTTNG_CONSUMER64_UST: - mmap_base = lttng_ustctl_get_mmap_base(stream); - if (!mmap_base) { - ERR("read mmap get mmap base for stream %s", stream->name); - ret = -EPERM; - goto end; - } - ret = lttng_ustctl_get_mmap_read_offset(stream, &mmap_offset); - if (ret != 0) { - PERROR("tracer ctl get_mmap_read_offset"); - ret = -EINVAL; - goto end; - } - break; - default: - ERR("Unknown consumer_data type"); - assert(0); - } - /* Handle stream on the relayd if the output is on the network */ if (relayd) { - unsigned long netlen = len; + unsigned long netlen = subbuf_content_size; /* * Lock the control socket for the complete duration of the function @@ -1600,16 +1644,16 @@ ssize_t lttng_consumer_on_read_subbuffer_mmap( /* Write metadata stream id before payload */ if (stream->metadata_flag) { - ret = write_relayd_metadata_id(outfd, stream, relayd, padding); + ret = write_relayd_metadata_id(outfd, stream, padding); if (ret < 0) { relayd_hang_up = 1; 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) { @@ -1623,49 +1667,26 @@ 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. @@ -1676,7 +1697,7 @@ 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 @@ -1686,7 +1707,8 @@ ssize_t lttng_consumer_on_read_subbuffer_mmap( DBG("Consumer mmap write detected relayd hang up"); } else { /* Unhandled error, print it and stop function right now. */ - PERROR("Error in write mmap (ret %zd != len %lu)", ret, len); + PERROR("Error in write mmap (ret %zd != write_len %zu)", ret, + write_len); } goto write_error; } @@ -1695,9 +1717,9 @@ ssize_t lttng_consumer_on_read_subbuffer_mmap( /* This call is useless on a socket so better save a syscall. */ if (!relayd) { /* This won't block, but will start writeout asynchronously */ - lttng_sync_file_range(outfd, stream->out_fd_offset, len, + lttng_sync_file_range(outfd, stream->out_fd_offset, write_len, SYNC_FILE_RANGE_WRITE); - stream->out_fd_offset += len; + stream->out_fd_offset += write_len; lttng_consumer_sync_trace_file(stream, orig_offset); } @@ -1707,7 +1729,8 @@ 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: @@ -1730,8 +1753,7 @@ end: ssize_t lttng_consumer_on_read_subbuffer_splice( struct lttng_consumer_local_data *ctx, struct lttng_consumer_stream *stream, unsigned long len, - unsigned long padding, - struct ctf_packet_index *index) + unsigned long padding) { ssize_t ret = 0, written = 0, ret_splice = 0; loff_t offset = 0; @@ -1789,7 +1811,7 @@ ssize_t lttng_consumer_on_read_subbuffer_splice( } stream->reset_metadata_flag = 0; } - ret = write_relayd_metadata_id(splice_pipe[1], stream, relayd, + ret = write_relayd_metadata_id(splice_pipe[1], stream, padding); if (ret < 0) { written = ret; @@ -1826,37 +1848,15 @@ 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) { @@ -1933,7 +1933,8 @@ 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; } @@ -1961,6 +1962,25 @@ end: return written; } +/* + * Sample the snapshot positions for a specific fd + * + * Returns 0 on success, < 0 on error + */ +int lttng_consumer_sample_snapshot_positions(struct lttng_consumer_stream *stream) +{ + switch (consumer_data.type) { + case LTTNG_CONSUMER_KERNEL: + return lttng_kconsumer_sample_snapshot_positions(stream); + case LTTNG_CONSUMER32_UST: + case LTTNG_CONSUMER64_UST: + return lttng_ustconsumer_sample_snapshot_positions(stream); + default: + ERR("Unknown consumer_data type"); + assert(0); + return -ENOSYS; + } +} /* * Take a snapshot for a specific fd * @@ -2002,6 +2022,27 @@ int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream *stream, } } +/* + * Get the consumed position (free-running counter position in bytes). + * + * Returns 0 on success, < 0 on error + */ +int lttng_consumer_get_consumed_snapshot(struct lttng_consumer_stream *stream, + unsigned long *pos) +{ + switch (consumer_data.type) { + case LTTNG_CONSUMER_KERNEL: + return lttng_kconsumer_get_consumed_snapshot(stream, pos); + case LTTNG_CONSUMER32_UST: + case LTTNG_CONSUMER64_UST: + return lttng_ustconsumer_get_consumed_snapshot(stream, pos); + default: + ERR("Unknown consumer_data type"); + assert(0); + return -ENOSYS; + } +} + int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx, int sock, struct pollfd *consumer_sockpoll) { @@ -2018,6 +2059,7 @@ int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx, } } +static void lttng_consumer_close_all_metadata(void) { switch (consumer_data.type) { @@ -2050,7 +2092,8 @@ void lttng_consumer_close_all_metadata(void) void consumer_del_metadata_stream(struct lttng_consumer_stream *stream, struct lttng_ht *ht) { - struct lttng_consumer_channel *free_chan = NULL; + struct lttng_consumer_channel *channel = NULL; + bool free_channel = false; assert(stream); /* @@ -2062,11 +2105,17 @@ void consumer_del_metadata_stream(struct lttng_consumer_stream *stream, DBG3("Consumer delete metadata stream %d", stream->wait_fd); pthread_mutex_lock(&consumer_data.lock); - pthread_mutex_lock(&stream->chan->lock); + /* + * Note that this assumes that a stream's channel is never changed and + * that the stream's lock doesn't need to be taken to sample its + * channel. + */ + channel = stream->chan; + pthread_mutex_lock(&channel->lock); pthread_mutex_lock(&stream->lock); - if (stream->chan->metadata_cache) { + if (channel->metadata_cache) { /* Only applicable to userspace consumers. */ - pthread_mutex_lock(&stream->chan->metadata_cache->lock); + pthread_mutex_lock(&channel->metadata_cache->lock); } /* Remove any reference to that stream. */ @@ -2078,30 +2127,33 @@ 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(&channel->lock); pthread_mutex_unlock(&consumer_data.lock); - if (free_chan) { - consumer_del_channel(free_chan); + if (free_channel) { + consumer_del_channel(channel); } + lttng_trace_chunk_put(stream->trace_chunk); + stream->trace_chunk = NULL; consumer_stream_free(stream); } @@ -2155,7 +2207,7 @@ void consumer_add_metadata_stream(struct lttng_consumer_stream *stream) lttng_ht_add_unique_u64(ht, &stream->node); - lttng_ht_add_unique_u64(consumer_data.stream_per_chan_id_ht, + lttng_ht_add_u64(consumer_data.stream_per_chan_id_ht, &stream->node_channel_id); /* @@ -2299,11 +2351,6 @@ 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; @@ -2376,7 +2423,7 @@ restart: do { health_code_update(); - len = ctx->on_buffer_ready(stream, ctx); + len = ctx->on_buffer_ready(stream, ctx, false); /* * We don't check the return value here since if we get * a negative len, it means an error occurred thus we @@ -2403,7 +2450,7 @@ restart: do { health_code_update(); - len = ctx->on_buffer_ready(stream, ctx); + len = ctx->on_buffer_ready(stream, ctx, false); /* * We don't check the return value here since if we get * a negative len, it means an error occurred thus we @@ -2458,6 +2505,8 @@ void *consumer_thread_data_poll(void *data) struct lttng_consumer_stream **local_stream = NULL, *new_stream = NULL; /* local view of consumer_data.fds_count */ 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; @@ -2497,18 +2546,15 @@ void *consumer_thread_data_poll(void *data) free(local_stream); local_stream = NULL; - /* - * Allocate for all fds +1 for the consumer_data_pipe and +1 for - * wake up pipe. - */ - pollfd = zmalloc((consumer_data.stream_count + 2) * sizeof(struct pollfd)); + /* Allocate for all fds */ + pollfd = zmalloc((consumer_data.stream_count + nb_pipes_fd) * sizeof(struct pollfd)); if (pollfd == NULL) { PERROR("pollfd malloc"); pthread_mutex_unlock(&consumer_data.lock); goto end; } - local_stream = zmalloc((consumer_data.stream_count + 2) * + local_stream = zmalloc((consumer_data.stream_count + nb_pipes_fd) * sizeof(struct lttng_consumer_stream *)); if (local_stream == NULL) { PERROR("local_stream malloc"); @@ -2536,12 +2582,12 @@ void *consumer_thread_data_poll(void *data) } /* poll on the array of fds */ restart: - DBG("polling on %d fd", nb_fd + 2); + DBG("polling on %d fd", nb_fd + nb_pipes_fd); if (testpoint(consumerd_thread_data_poll)) { goto end; } health_poll_entry(); - num_rdy = poll(pollfd, nb_fd + 2, -1); + num_rdy = poll(pollfd, nb_fd + nb_pipes_fd, -1); health_poll_exit(); DBG("poll num_rdy : %d", num_rdy); if (num_rdy == -1) { @@ -2620,7 +2666,7 @@ void *consumer_thread_data_poll(void *data) if (pollfd[i].revents & POLLPRI) { DBG("Urgent read on fd %d", pollfd[i].fd); high_prio = 1; - len = ctx->on_buffer_ready(local_stream[i], ctx); + len = ctx->on_buffer_ready(local_stream[i], ctx, false); /* it's ok to have an unavailable sub-buffer */ if (len < 0 && len != -EAGAIN && len != -ENODATA) { /* Clean the stream and free it. */ @@ -2651,7 +2697,7 @@ void *consumer_thread_data_poll(void *data) local_stream[i]->hangup_flush_done || local_stream[i]->has_data) { DBG("Normal read on fd %d", pollfd[i].fd); - len = ctx->on_buffer_ready(local_stream[i], ctx); + len = ctx->on_buffer_ready(local_stream[i], ctx, false); /* it's ok to have an unavailable sub-buffer */ if (len < 0 && len != -EAGAIN && len != -ENODATA) { /* Clean the stream and free it. */ @@ -2894,11 +2940,6 @@ 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; @@ -3212,7 +3253,7 @@ 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; @@ -3269,62 +3310,163 @@ 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; - - pthread_mutex_lock(&stream->lock); - if (stream->metadata_flag) { - pthread_mutex_lock(&stream->metadata_rdv_lock); - } + size_t i; + int ret = 0; + const size_t count = lttng_dynamic_array_get_count( + &stream->read_subbuffer_ops.post_consume_cbs); - switch (consumer_data.type) { - case LTTNG_CONSUMER_KERNEL: - ret = lttng_kconsumer_read_subbuffer(stream, ctx); - break; - case LTTNG_CONSUMER32_UST: - case LTTNG_CONSUMER64_UST: - ret = lttng_ustconsumer_read_subbuffer(stream, ctx); - break; - default: - ERR("Unknown consumer_data type"); - assert(0); - ret = -ENOSYS; - break; - } + 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); - if (stream->metadata_flag) { - pthread_cond_broadcast(&stream->metadata_rdv); - pthread_mutex_unlock(&stream->metadata_rdv_lock); + ret = op(stream, subbuffer, ctx); + if (ret) { + goto end; + } } - pthread_mutex_unlock(&stream->lock); +end: return ret; } -int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream) +ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream, + struct lttng_consumer_local_data *ctx, + bool locked_by_caller) { - switch (consumer_data.type) { - case LTTNG_CONSUMER_KERNEL: - return lttng_kconsumer_on_recv_stream(stream); - case LTTNG_CONSUMER32_UST: - case LTTNG_CONSUMER64_UST: - return lttng_ustconsumer_on_recv_stream(stream); - default: - ERR("Unknown consumer_data type"); - assert(0); - return -ENOSYS; - } -} + ssize_t ret, written_bytes = 0; + int rotation_ret; + struct stream_subbuffer subbuffer = {}; -/* - * Allocate and set consumer data hash tables. - */ -int lttng_consumer_init(void) -{ - consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); - if (!consumer_data.channel_ht) { - goto error; + 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; + } + } + + ret = stream->read_subbuffer_ops.get_next_subbuffer(stream, &subbuffer); + if (ret) { + if (ret == -ENODATA) { + /* Not an error. */ + ret = 0; + goto sleep_stream; + } + goto end; + } + + ret = stream->read_subbuffer_ops.pre_consume_subbuffer( + stream, &subbuffer); + if (ret) { + goto error_put_subbuf; + } + + written_bytes = stream->read_subbuffer_ops.consume_subbuffer( + ctx, stream, &subbuffer); + 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) { + case LTTNG_CONSUMER_KERNEL: + return lttng_kconsumer_on_recv_stream(stream); + case LTTNG_CONSUMER32_UST: + case LTTNG_CONSUMER64_UST: + return lttng_ustconsumer_on_recv_stream(stream); + default: + ERR("Unknown consumer_data type"); + assert(0); + return -ENOSYS; + } +} + +/* + * Allocate and set consumer data hash tables. + */ +int lttng_consumer_init(void) +{ + consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); + if (!consumer_data.channel_ht) { + goto error; + } + + consumer_data.channels_by_session_id_ht = + lttng_ht_new(0, LTTNG_HT_TYPE_U64); + if (!consumer_data.channels_by_session_id_ht) { + goto error; } consumer_data.relayd_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); @@ -3352,6 +3494,11 @@ int lttng_consumer_init(void) goto error; } + consumer_data.chunk_registry = lttng_trace_chunk_registry_create(); + if (!consumer_data.chunk_registry) { + goto error; + } + return 0; error: @@ -3462,7 +3609,6 @@ 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; @@ -3490,7 +3636,6 @@ 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; @@ -3504,6 +3649,11 @@ 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); @@ -3517,6 +3667,7 @@ 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! */ @@ -3540,34 +3691,6 @@ 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. * @@ -3634,28 +3757,11 @@ int consumer_data_pending(uint64_t id) /* 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; - } - } - 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 @@ -3673,9 +3779,27 @@ 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); @@ -3684,24 +3808,25 @@ 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) { @@ -3783,3 +3908,1188 @@ unsigned long consumer_get_consume_start_pos(unsigned long consumed_pos, } return start_pos; } + +/* + * Sample the rotate position for all the streams of a channel. If a stream + * 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(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_stream *stream; + struct lttng_ht_iter iter; + struct lttng_ht *ht = consumer_data.stream_per_chan_id_ht; + struct lttng_dynamic_array stream_rotation_positions; + uint64_t next_chunk_id, stream_count = 0; + enum lttng_trace_chunk_status chunk_status; + const bool is_local_trace = relayd_id == -1ULL; + struct consumer_relayd_sock_pair *relayd = NULL; + bool rotating_to_new_chunk = true; + /* 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); + + lttng_dynamic_array_init(&stream_rotation_positions, + sizeof(struct relayd_stream_rotation_position), NULL); + lttng_dynamic_pointer_array_init(&streams_packet_to_open, NULL); + + rcu_read_lock(); + + pthread_mutex_lock(&channel->lock); + assert(channel->trace_chunk); + chunk_status = lttng_trace_chunk_get_id(channel->trace_chunk, + &next_chunk_id); + if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { + ret = -1; + goto end_unlock_channel; + } + + cds_lfht_for_each_entry_duplicate(ht->ht, + ht->hash_fct(&channel->key, lttng_ht_seed), + ht->match_fct, &channel->key, &iter.iter, + stream, node_channel_id.node) { + unsigned long produced_pos = 0, consumed_pos = 0; + + health_code_update(); + + /* + * Lock stream because we are about to change its state. + */ + pthread_mutex_lock(&stream->lock); + + if (stream->trace_chunk == stream->chan->trace_chunk) { + rotating_to_new_chunk = false; + } + + /* + * Do not flush an empty packet when rotating from a NULL trace + * chunk. The stream has no means to output data, and the prior + * rotation which rotated to NULL performed that side-effect already. + */ + if (stream->trace_chunk) { + /* + * For metadata stream, do an active flush, which does not + * produce empty packets. For data streams, empty-flush; + * ensures we have at least one packet in each stream per trace + * chunk, even if no data was produced. + */ + ret = consumer_stream_flush_buffer( + stream, stream->metadata_flag ? 1 : 0); + if (ret < 0) { + ERR("Failed to flush stream %" PRIu64 " during channel rotation", + stream->key); + goto end_unlock_stream; + } + } + + ret = lttng_consumer_take_snapshot(stream); + if (ret < 0 && ret != -ENODATA && ret != -EAGAIN) { + ERR("Failed to sample snapshot position during channel rotation"); + goto end_unlock_stream; + } + if (!ret) { + ret = lttng_consumer_get_produced_snapshot(stream, + &produced_pos); + if (ret < 0) { + ERR("Failed to sample produced position during channel rotation"); + goto end_unlock_stream; + } + + ret = lttng_consumer_get_consumed_snapshot(stream, + &consumed_pos); + if (ret < 0) { + ERR("Failed to sample consumed position during channel rotation"); + goto end_unlock_stream; + } + } + /* + * Align produced position on the start-of-packet boundary of the first + * packet going into the next trace chunk. + */ + produced_pos = ALIGN_FLOOR(produced_pos, stream->max_sb_size); + if (consumed_pos == produced_pos) { + DBG("Set rotate ready for stream %" PRIu64 " produced = %lu consumed = %lu", + stream->key, produced_pos, consumed_pos); + stream->rotate_ready = true; + } else { + DBG("Different consumed and produced positions " + "for stream %" PRIu64 " produced = %lu consumed = %lu", + stream->key, produced_pos, consumed_pos); + } + /* + * 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); + } + stream = NULL; + + 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(); + } + } + + 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; +} + +static +int consumer_clear_buffer(struct lttng_consumer_stream *stream) +{ + int ret = 0; + unsigned long consumed_pos_before, consumed_pos_after; + + 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_before); + if (ret < 0) { + ERR("Consumed snapshot position"); + goto end; + } + + switch (consumer_data.type) { + case LTTNG_CONSUMER_KERNEL: + ret = kernctl_buffer_clear(stream->wait_fd); + if (ret < 0) { + ERR("Failed to clear kernel stream (ret = %d)", ret); + goto end; + } + break; + case LTTNG_CONSUMER32_UST: + case LTTNG_CONSUMER64_UST: + lttng_ustconsumer_clear_buffer(stream); + break; + default: + ERR("Unknown consumer_data type"); + abort(); + } + + ret = lttng_consumer_sample_snapshot_positions(stream); + if (ret < 0) { + ERR("Taking snapshot positions"); + goto end; + } + ret = lttng_consumer_get_consumed_snapshot(stream, &consumed_pos_after); + if (ret < 0) { + ERR("Consumed snapshot position"); + goto end; + } + DBG("clear: before: %lu after: %lu", consumed_pos_before, consumed_pos_after); +end: + return ret; +} + +static +int consumer_clear_stream(struct lttng_consumer_stream *stream) +{ + int ret; + + ret = consumer_stream_flush_buffer(stream, 1); + if (ret < 0) { + ERR("Failed to flush stream %" PRIu64 " during channel clear", + stream->key); + ret = LTTCOMM_CONSUMERD_FATAL; + goto error; + } + + ret = consumer_clear_buffer(stream); + if (ret < 0) { + ERR("Failed to clear stream %" PRIu64 " during channel clear", + stream->key); + ret = LTTCOMM_CONSUMERD_FATAL; + goto error; + } + + ret = LTTCOMM_CONSUMERD_SUCCESS; +error: + return ret; +} + +static +int consumer_clear_unmonitored_channel(struct lttng_consumer_channel *channel) +{ + int ret; + struct lttng_consumer_stream *stream; + + rcu_read_lock(); + pthread_mutex_lock(&channel->lock); + cds_list_for_each_entry(stream, &channel->streams.head, send_node) { + health_code_update(); + pthread_mutex_lock(&stream->lock); + ret = consumer_clear_stream(stream); + if (ret) { + goto error_unlock; + } + pthread_mutex_unlock(&stream->lock); + } + pthread_mutex_unlock(&channel->lock); + rcu_read_unlock(); + return 0; + +error_unlock: + pthread_mutex_unlock(&stream->lock); + pthread_mutex_unlock(&channel->lock); + rcu_read_unlock(); + return ret; +} + +/* + * Check if a stream is ready to be rotated after extracting it. + * + * Return 1 if it is ready for rotation, 0 if it is not, a negative value on + * error. Stream lock must be held. + */ +int lttng_consumer_stream_is_rotate_ready(struct lttng_consumer_stream *stream) +{ + DBG("Check is rotate ready for stream %" PRIu64 + " ready %u rotate_position %" PRIu64 + " last_sequence_number %" PRIu64, + stream->key, stream->rotate_ready, + stream->rotate_position, stream->last_sequence_number); + if (stream->rotate_ready) { + return 1; + } + + /* + * If packet seq num is unavailable, it means we are interacting + * with a pre-2.8 lttng-modules which does not implement the + * sequence number. Rotation should never be used by sessiond in this + * scenario. + */ + if (stream->sequence_number_unavailable) { + ERR("Internal error: rotation used on stream %" PRIu64 + " with unavailable sequence number", + stream->key); + return -1; + } + + if (stream->rotate_position == -1ULL || + stream->last_sequence_number == -1ULL) { + return 0; + } + + /* + * Rotate position not reached yet. The stream rotate position is + * the position of the next packet belonging to the next trace chunk, + * but consumerd considers rotation ready when reaching the last + * packet of the current chunk, hence the "rotate_position - 1". + */ + + DBG("Check is rotate ready for stream %" PRIu64 + " last_sequence_number %" PRIu64 + " rotate_position %" PRIu64, + stream->key, stream->last_sequence_number, + stream->rotate_position); + if (stream->last_sequence_number >= stream->rotate_position - 1) { + return 1; + } + + return 0; +} + +/* + * Reset the state for a stream after a rotation occurred. + */ +void lttng_consumer_reset_stream_rotate_state(struct lttng_consumer_stream *stream) +{ + DBG("lttng_consumer_reset_stream_rotate_state for stream %" PRIu64, + stream->key); + stream->rotate_position = -1ULL; + stream->rotate_ready = false; +} + +/* + * Perform the rotation a local stream file. + */ +static +int rotate_local_stream(struct lttng_consumer_local_data *ctx, + struct lttng_consumer_stream *stream) +{ + int ret = 0; + + 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; + } + + 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 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) +{ + int ret; + + DBG("Consumer rotate stream %" PRIu64, stream->key); + + /* + * 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->net_seq_idx == (uint64_t) -1ULL) { + ret = rotate_local_stream(ctx, stream); + if (ret < 0) { + ERR("Failed to rotate stream, ret = %i", ret); + goto error; + } + } + + if (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; + +error: + return ret; +} + +/* + * Rotate all the ready streams now. + * + * 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(struct lttng_consumer_channel *channel, + uint64_t key, struct lttng_consumer_local_data *ctx) +{ + int ret; + struct lttng_consumer_stream *stream; + struct lttng_ht_iter iter; + struct lttng_ht *ht = consumer_data.stream_per_chan_id_ht; + + rcu_read_lock(); + + DBG("Consumer rotate ready streams in channel %" PRIu64, key); + + 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); + pthread_mutex_unlock(&stream->lock); + pthread_mutex_unlock(&stream->chan->lock); + if (ret) { + goto end; + } + } + + ret = 0; + +end: + rcu_read_unlock(); + return ret; +} + +enum lttcomm_return_code lttng_consumer_init_command( + struct lttng_consumer_local_data *ctx, + const lttng_uuid sessiond_uuid) +{ + enum lttcomm_return_code ret; + char uuid_str[LTTNG_UUID_STR_LEN]; + + if (ctx->sessiond_uuid.is_set) { + ret = LTTCOMM_CONSUMERD_ALREADY_SET; + goto end; + } + + 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; +} + +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; + 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; + + 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)"; + } + } + + /* Local protocol error. */ + 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( + 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(consumer_data.channels_by_session_id_ht->ht, + consumer_data.channels_by_session_id_ht->hash_fct( + &session_id, lttng_ht_seed), + 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; +} + +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) +{ + 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( + 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(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; +} + +enum lttcomm_return_code lttng_consumer_trace_chunk_exists( + const uint64_t *relayd_id, uint64_t session_id, + uint64_t chunk_id) +{ + int ret; + enum lttcomm_return_code ret_code; + char relayd_id_buffer[MAX_INT_DEC_LEN(*relayd_id)]; + const char *relayd_id_str = "(none)"; + const bool is_local_trace = !relayd_id; + struct consumer_relayd_sock_pair *relayd = NULL; + bool chunk_exists_local, chunk_exists_remote; + + if (relayd_id) { + int ret; + + /* 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)"; + } + } + + DBG("Consumer trace chunk exists command: relayd_id = %s" + ", chunk_id = %" PRIu64, relayd_id_str, + chunk_id); + ret = lttng_trace_chunk_registry_chunk_exists( + consumer_data.chunk_registry, session_id, + chunk_id, &chunk_exists_local); + if (ret) { + /* Internal error. */ + ERR("Failed to query the existence of a trace chunk"); + ret_code = LTTCOMM_CONSUMERD_FATAL; + goto end; + } + DBG("Trace chunk %s locally", + chunk_exists_local ? "exists" : "does not exist"); + if (chunk_exists_local) { + ret_code = LTTCOMM_CONSUMERD_TRACE_CHUNK_EXISTS_LOCAL; + goto end; + } else if (is_local_trace) { + ret_code = LTTCOMM_CONSUMERD_UNKNOWN_TRACE_CHUNK; + goto end; + } + + rcu_read_lock(); + relayd = consumer_find_relayd(*relayd_id); + if (!relayd) { + ERR("Failed to find relayd %" PRIu64, *relayd_id); + ret_code = LTTCOMM_CONSUMERD_INVALID_PARAMETERS; + goto end_rcu_unlock; + } + DBG("Looking up existence of trace chunk on relay daemon"); + pthread_mutex_lock(&relayd->ctrl_sock_mutex); + ret = relayd_trace_chunk_exists(&relayd->control_sock, chunk_id, + &chunk_exists_remote); + pthread_mutex_unlock(&relayd->ctrl_sock_mutex); + if (ret < 0) { + ERR("Failed to look-up the existence of trace chunk on relay daemon"); + ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL; + goto end_rcu_unlock; + } + + ret_code = chunk_exists_remote ? + LTTCOMM_CONSUMERD_TRACE_CHUNK_EXISTS_REMOTE : + LTTCOMM_CONSUMERD_UNKNOWN_TRACE_CHUNK; + DBG("Trace chunk %s on relay daemon", + chunk_exists_remote ? "exists" : "does not exist"); + +end_rcu_unlock: + rcu_read_unlock(); +end: + return ret_code; +} + +static +int consumer_clear_monitored_channel(struct lttng_consumer_channel *channel) +{ + struct lttng_ht *ht; + struct lttng_consumer_stream *stream; + struct lttng_ht_iter iter; + int ret; + + ht = consumer_data.stream_per_chan_id_ht; + + rcu_read_lock(); + cds_lfht_for_each_entry_duplicate(ht->ht, + ht->hash_fct(&channel->key, lttng_ht_seed), + ht->match_fct, &channel->key, + &iter.iter, stream, node_channel_id.node) { + /* + * Protect against teardown with mutex. + */ + pthread_mutex_lock(&stream->lock); + if (cds_lfht_is_node_deleted(&stream->node.node)) { + goto next; + } + ret = consumer_clear_stream(stream); + if (ret) { + goto error_unlock; + } + next: + pthread_mutex_unlock(&stream->lock); + } + rcu_read_unlock(); + return LTTCOMM_CONSUMERD_SUCCESS; + +error_unlock: + pthread_mutex_unlock(&stream->lock); + rcu_read_unlock(); + return ret; +} + +int lttng_consumer_clear_channel(struct lttng_consumer_channel *channel) +{ + int ret; + + DBG("Consumer clear channel %" PRIu64, channel->key); + + if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA) { + /* + * Nothing to do for the metadata channel/stream. + * Snapshot mechanism already take care of the metadata + * handling/generation, and monitored channels only need to + * have their data stream cleared.. + */ + ret = LTTCOMM_CONSUMERD_SUCCESS; + goto end; + } + + if (!channel->monitor) { + ret = consumer_clear_unmonitored_channel(channel); + } else { + ret = consumer_clear_monitored_channel(channel); + } +end: + return ret; +} + +enum lttcomm_return_code lttng_consumer_open_channel_packets( + struct lttng_consumer_channel *channel) +{ + struct lttng_consumer_stream *stream; + enum lttcomm_return_code ret = LTTCOMM_CONSUMERD_SUCCESS; + + if (channel->metadata_stream) { + ERR("Open channel packets command attempted on a metadata channel"); + ret = LTTCOMM_CONSUMERD_INVALID_PARAMETERS; + goto end; + } + + 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; +}