X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fcommon%2Fust-consumer%2Fust-consumer.c;h=01c73f959edb7320cd5db3a92a61079ef5f64e6d;hp=81f6859124c1353d89dbb36b38f8549761f33d53;hb=594c7c002eda287de5f8c8e1859a8841f6fcc90f;hpb=d7ba13889c8692b14f99238ddf2721ed78df89d2 diff --git a/src/common/ust-consumer/ust-consumer.c b/src/common/ust-consumer/ust-consumer.c index 81f685912..01c73f959 100644 --- a/src/common/ust-consumer/ust-consumer.c +++ b/src/common/ust-consumer/ust-consumer.c @@ -1,22 +1,12 @@ /* - * Copyright (C) 2011 - Julien Desfossez - * Mathieu Desnoyers + * Copyright (C) 2011 Julien Desfossez + * Copyright (C) 2011 Mathieu Desnoyers + * Copyright (C) 2017 Jérémie Galarneau * - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License, version 2 only, - * as published by the Free Software Foundation. + * SPDX-License-Identifier: GPL-2.0-only * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License along - * with this program; if not, write to the Free Software Foundation, Inc., - * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. */ -#define _GNU_SOURCE #define _LGPL_SOURCE #include #include @@ -32,6 +22,8 @@ #include #include #include +#include +#include #include #include @@ -39,17 +31,21 @@ #include #include #include -#include -#include -#include +#include +#include +#include #include #include +#include +#include +#include #include "ust-consumer.h" +#define INT_MAX_STR_LEN 12 /* includes \0 */ + extern struct lttng_consumer_global_data consumer_data; extern int consumer_poll_timeout; -extern volatile int consumer_quit; /* * Free channel object and all streams associated with it. This MUST be used @@ -71,6 +67,7 @@ static void destroy_channel(struct lttng_consumer_channel *channel) cds_list_del(&stream->send_node); ustctl_destroy_stream(stream->ustream); + lttng_trace_chunk_put(stream->trace_chunk); free(stream); } @@ -80,7 +77,13 @@ static void destroy_channel(struct lttng_consumer_channel *channel) */ if (channel->uchan) { lttng_ustconsumer_del_channel(channel); + lttng_ustconsumer_free_channel(channel); + } + + if (channel->trace_chunk) { + lttng_trace_chunk_put(channel->trace_chunk); } + free(channel); } @@ -116,26 +119,6 @@ error: return ret; } -/* - * Allocate and return a consumer channel object. - */ -static struct lttng_consumer_channel *allocate_channel(uint64_t session_id, - const char *pathname, const char *name, uid_t uid, gid_t gid, - uint64_t relayd_id, uint64_t key, enum lttng_event_output output, - uint64_t tracefile_size, uint64_t tracefile_count, - uint64_t session_id_per_pid, unsigned int monitor, - unsigned int live_timer_interval, - const char *shm_path) -{ - assert(pathname); - assert(name); - - return consumer_allocate_channel(key, session_id, pathname, name, uid, - gid, relayd_id, output, tracefile_size, - tracefile_count, session_id_per_pid, monitor, - live_timer_interval, shm_path); -} - /* * Allocate and return a consumer stream object. If _alloc_ret is not NULL, the * error value if applicable is set in it else it is kept untouched. @@ -152,14 +135,14 @@ static struct lttng_consumer_stream *allocate_stream(int cpu, int key, assert(channel); assert(ctx); - stream = consumer_allocate_stream(channel->key, + stream = consumer_stream_create( + channel, + channel->key, key, - LTTNG_CONSUMER_ACTIVE_STREAM, channel->name, - channel->uid, - channel->gid, channel->relayd_id, channel->session_id, + channel->trace_chunk, cpu, &alloc_ret, channel->type, @@ -182,7 +165,7 @@ static struct lttng_consumer_stream *allocate_stream(int cpu, int key, goto error; } - stream->chan = channel; + consumer_stream_update_channel_attributes(stream, channel); error: if (_alloc_ret) { @@ -204,28 +187,21 @@ static int send_stream_to_thread(struct lttng_consumer_stream *stream, /* Get the right pipe where the stream will be sent. */ if (stream->metadata_flag) { - ret = consumer_add_metadata_stream(stream); - if (ret) { - ERR("Consumer add metadata stream %" PRIu64 " failed.", - stream->key); - goto error; - } + consumer_add_metadata_stream(stream); stream_pipe = ctx->consumer_metadata_pipe; } else { - ret = consumer_add_data_stream(stream); - if (ret) { - ERR("Consumer add stream %" PRIu64 " failed.", - stream->key); - goto error; - } + consumer_add_data_stream(stream); stream_pipe = ctx->consumer_data_pipe; } /* * From this point on, the stream's ownership has been moved away from - * the channel and becomes globally visible. + * the channel and it becomes globally visible. Hence, remove it from + * the local stream list to prevent the stream from being both local and + * global. */ stream->globally_visible = 1; + cds_list_del(&stream->send_node); ret = lttng_pipe_write(stream_pipe, &stream, sizeof(stream)); if (ret < 0) { @@ -237,13 +213,36 @@ static int send_stream_to_thread(struct lttng_consumer_stream *stream, } else { consumer_del_stream_for_data(stream); } + goto error; } + error: return ret; } +static +int get_stream_shm_path(char *stream_shm_path, const char *shm_path, int cpu) +{ + char cpu_nr[INT_MAX_STR_LEN]; /* int max len */ + int ret; + + strncpy(stream_shm_path, shm_path, PATH_MAX); + stream_shm_path[PATH_MAX - 1] = '\0'; + ret = snprintf(cpu_nr, INT_MAX_STR_LEN, "%i", cpu); + if (ret < 0) { + PERROR("snprintf"); + goto end; + } + strncat(stream_shm_path, cpu_nr, + PATH_MAX - strlen(stream_shm_path) - 1); + ret = 0; +end: + return ret; +} + /* * Create streams for the given channel using liblttng-ust-ctl. + * The channel lock must be acquired by the caller. * * Return 0 on success else a negative value. */ @@ -253,6 +252,7 @@ static int create_ust_streams(struct lttng_consumer_channel *channel, int ret, cpu = 0; struct ustctl_consumer_stream *ustream; struct lttng_consumer_stream *stream; + pthread_mutex_t *current_stream_lock = NULL; assert(channel); assert(ctx); @@ -299,6 +299,8 @@ static int create_ust_streams(struct lttng_consumer_channel *channel, uatomic_inc(&stream->chan->refcount); } + pthread_mutex_lock(&stream->lock); + current_stream_lock = &stream->lock; /* * Order is important this is why a list is used. On error, the caller * should clean this list. @@ -330,32 +332,67 @@ static int create_ust_streams(struct lttng_consumer_channel *channel, /* Keep stream reference when creating metadata. */ if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA) { channel->metadata_stream = stream; - stream->ust_metadata_poll_pipe[0] = ust_metadata_pipe[0]; - stream->ust_metadata_poll_pipe[1] = ust_metadata_pipe[1]; + if (channel->monitor) { + /* Set metadata poll pipe if we created one */ + memcpy(stream->ust_metadata_poll_pipe, + ust_metadata_pipe, + sizeof(ust_metadata_pipe)); + } } + pthread_mutex_unlock(&stream->lock); + current_stream_lock = NULL; } return 0; error: error_alloc: + if (current_stream_lock) { + pthread_mutex_unlock(current_stream_lock); + } return ret; } +static int open_ust_stream_fd(struct lttng_consumer_channel *channel, int cpu, + const struct lttng_credentials *session_credentials) +{ + char shm_path[PATH_MAX]; + int ret; + + if (!channel->shm_path[0]) { + return shm_create_anonymous("ust-consumer"); + } + ret = get_stream_shm_path(shm_path, channel->shm_path, cpu); + if (ret) { + goto error_shm_path; + } + return run_as_open(shm_path, + O_RDWR | O_CREAT | O_EXCL, S_IRUSR | S_IWUSR, + lttng_credentials_get_uid(session_credentials), + lttng_credentials_get_gid(session_credentials)); + +error_shm_path: + return -1; +} + /* * Create an UST channel with the given attributes and send it to the session * daemon using the ust ctl API. * * Return 0 on success or else a negative value. */ -static int create_ust_channel(struct ustctl_consumer_channel_attr *attr, - struct ustctl_consumer_channel **chanp) +static int create_ust_channel(struct lttng_consumer_channel *channel, + struct ustctl_consumer_channel_attr *attr, + struct ustctl_consumer_channel **ust_chanp) { - int ret; - struct ustctl_consumer_channel *channel; + int ret, nr_stream_fds, i, j; + int *stream_fds; + struct ustctl_consumer_channel *ust_channel; + assert(channel); assert(attr); - assert(chanp); + assert(ust_chanp); + assert(channel->buffer_credentials.is_set); DBG3("Creating channel to ustctl with attr: [overwrite: %d, " "subbuf_size: %" PRIu64 ", num_subbuf: %" PRIu64 ", " @@ -364,17 +401,72 @@ static int create_ust_channel(struct ustctl_consumer_channel_attr *attr, attr->num_subbuf, attr->switch_timer_interval, attr->read_timer_interval, attr->output, attr->type); - channel = ustctl_create_channel(attr); - if (!channel) { + if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA) + nr_stream_fds = 1; + else + nr_stream_fds = ustctl_get_nr_stream_per_channel(); + stream_fds = zmalloc(nr_stream_fds * sizeof(*stream_fds)); + if (!stream_fds) { + ret = -1; + goto error_alloc; + } + for (i = 0; i < nr_stream_fds; i++) { + stream_fds[i] = open_ust_stream_fd(channel, i, + &channel->buffer_credentials.value); + if (stream_fds[i] < 0) { + ret = -1; + goto error_open; + } + } + ust_channel = ustctl_create_channel(attr, stream_fds, nr_stream_fds); + if (!ust_channel) { ret = -1; goto error_create; } - - *chanp = channel; + channel->nr_stream_fds = nr_stream_fds; + channel->stream_fds = stream_fds; + *ust_chanp = ust_channel; return 0; error_create: +error_open: + for (j = i - 1; j >= 0; j--) { + int closeret; + + closeret = close(stream_fds[j]); + if (closeret) { + PERROR("close"); + } + if (channel->shm_path[0]) { + char shm_path[PATH_MAX]; + + closeret = get_stream_shm_path(shm_path, + channel->shm_path, j); + if (closeret) { + ERR("Cannot get stream shm path"); + } + closeret = run_as_unlink(shm_path, + lttng_credentials_get_uid(LTTNG_OPTIONAL_GET_PTR( + channel->buffer_credentials)), + lttng_credentials_get_gid(LTTNG_OPTIONAL_GET_PTR( + channel->buffer_credentials))); + if (closeret) { + PERROR("unlink %s", shm_path); + } + } + } + /* Try to rmdir all directories under shm_path root. */ + if (channel->root_shm_path[0]) { + (void) run_as_rmdir_recursive(channel->root_shm_path, + lttng_credentials_get_uid(LTTNG_OPTIONAL_GET_PTR( + channel->buffer_credentials)), + lttng_credentials_get_gid(LTTNG_OPTIONAL_GET_PTR( + channel->buffer_credentials)), + LTTNG_DIRECTORY_HANDLE_SKIP_NON_EMPTY_FLAG); + } + free(stream_fds); +error_alloc: return ret; } @@ -403,11 +495,11 @@ error: } /* - * Send channel to sessiond. + * Send channel to sessiond and relayd if applicable. * * Return 0 on success or else a negative value. */ -static int send_sessiond_channel(int sock, +static int send_channel_to_sessiond_and_relayd(int sock, struct lttng_consumer_channel *channel, struct lttng_consumer_local_data *ctx, int *relayd_error) { @@ -427,6 +519,8 @@ static int send_sessiond_channel(int sock, health_code_update(); /* Try to send the stream to the relayd if one is available. */ + DBG("Sending stream %" PRIu64 " of channel \"%s\" to relayd", + stream->key, channel->name); ret = consumer_send_relayd_stream(stream, stream->chan->pathname); if (ret < 0) { /* @@ -502,7 +596,7 @@ error: * Return 0 on success or else, a negative value is returned and the channel * MUST be destroyed by consumer_del_channel(). */ -static int ask_channel(struct lttng_consumer_local_data *ctx, int sock, +static int ask_channel(struct lttng_consumer_local_data *ctx, struct lttng_consumer_channel *channel, struct ustctl_consumer_channel_attr *attr) { @@ -528,7 +622,7 @@ static int ask_channel(struct lttng_consumer_local_data *ctx, int sock, channel->nb_init_stream_left = 0; /* The reply msg status is handled in the following call. */ - ret = create_ust_channel(attr, &channel->uchan); + ret = create_ust_channel(channel, attr, &channel->uchan); if (ret < 0) { goto end; } @@ -545,7 +639,9 @@ static int ask_channel(struct lttng_consumer_local_data *ctx, int sock, } /* Open all streams for this channel. */ + pthread_mutex_lock(&channel->lock); ret = create_ust_streams(channel, ctx); + pthread_mutex_unlock(&channel->lock); if (ret < 0) { goto end; } @@ -581,14 +677,8 @@ static int send_streams_to_thread(struct lttng_consumer_channel *channel, * If we are unable to send the stream to the thread, there is * a big problem so just stop everything. */ - /* Remove node from the channel stream list. */ - cds_list_del(&stream->send_node); goto error; } - - /* Remove node from the channel stream list. */ - cds_list_del(&stream->send_node); - } error: @@ -627,7 +717,63 @@ static int flush_channel(uint64_t chan_key) health_code_update(); - ustctl_flush_buffer(stream->ustream, 1); + pthread_mutex_lock(&stream->lock); + + /* + * Protect against concurrent teardown of a stream. + */ + if (cds_lfht_is_node_deleted(&stream->node.node)) { + goto next; + } + + if (!stream->quiescent) { + ustctl_flush_buffer(stream->ustream, 0); + stream->quiescent = true; + } +next: + pthread_mutex_unlock(&stream->lock); + } +error: + rcu_read_unlock(); + return ret; +} + +/* + * Clear quiescent state from channel's streams using the given key to + * retrieve the channel. + * + * Return 0 on success else an LTTng error code. + */ +static int clear_quiescent_channel(uint64_t chan_key) +{ + int ret = 0; + struct lttng_consumer_channel *channel; + struct lttng_consumer_stream *stream; + struct lttng_ht *ht; + struct lttng_ht_iter iter; + + DBG("UST consumer clear quiescent channel key %" PRIu64, chan_key); + + rcu_read_lock(); + channel = consumer_find_channel(chan_key); + if (!channel) { + ERR("UST consumer clear quiescent channel %" PRIu64 " not found", chan_key); + ret = LTTNG_ERR_UST_CHAN_NOT_FOUND; + goto error; + } + + ht = consumer_data.stream_per_chan_id_ht; + + /* For each stream of the channel id, clear quiescent state. */ + 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->lock); + stream->quiescent = false; + pthread_mutex_unlock(&stream->lock); } error: rcu_read_unlock(); @@ -636,7 +782,6 @@ error: /* * Close metadata stream wakeup_fd using the given key to retrieve the channel. - * RCU read side lock MUST be acquired before calling this function. * * Return 0 on success else an LTTng error code. */ @@ -644,6 +789,7 @@ static int close_metadata(uint64_t chan_key) { int ret = 0; struct lttng_consumer_channel *channel; + unsigned int channel_monitor; DBG("UST consumer close metadata key %" PRIu64, chan_key); @@ -662,13 +808,48 @@ static int close_metadata(uint64_t chan_key) pthread_mutex_lock(&consumer_data.lock); pthread_mutex_lock(&channel->lock); - + channel_monitor = channel->monitor; if (cds_lfht_is_node_deleted(&channel->node.node)) { goto error_unlock; } lttng_ustconsumer_close_metadata(channel); + pthread_mutex_unlock(&channel->lock); + pthread_mutex_unlock(&consumer_data.lock); + + /* + * The ownership of a metadata channel depends on the type of + * session to which it belongs. In effect, the monitor flag is checked + * to determine if this metadata channel is in "snapshot" mode or not. + * + * In the non-snapshot case, the metadata channel is created along with + * a single stream which will remain present until the metadata channel + * is destroyed (on the destruction of its session). In this case, the + * metadata stream in "monitored" by the metadata poll thread and holds + * the ownership of its channel. + * + * Closing the metadata will cause the metadata stream's "metadata poll + * pipe" to be closed. Closing this pipe will wake-up the metadata poll + * thread which will teardown the metadata stream which, in return, + * deletes the metadata channel. + * + * In the snapshot case, the metadata stream is created and destroyed + * on every snapshot record. Since the channel doesn't have an owner + * other than the session daemon, it is safe to destroy it immediately + * on reception of the CLOSE_METADATA command. + */ + if (!channel_monitor) { + /* + * The channel and consumer_data locks must be + * released before this call since consumer_del_channel + * re-acquires the channel and consumer_data locks to teardown + * the channel and queue its reclamation by the "call_rcu" + * worker thread. + */ + consumer_del_channel(channel); + } + return ret; error_unlock: pthread_mutex_unlock(&channel->lock); pthread_mutex_unlock(&consumer_data.lock); @@ -731,6 +912,10 @@ static int setup_metadata(struct lttng_consumer_local_data *ctx, uint64_t key) } } + /* + * Ownership of metadata stream is passed along. Freeing is handled by + * the callee. + */ ret = send_streams_to_thread(metadata, ctx); if (ret < 0) { /* @@ -738,7 +923,7 @@ static int setup_metadata(struct lttng_consumer_local_data *ctx, uint64_t key) * a big problem so just stop everything. */ ret = LTTCOMM_CONSUMERD_FATAL; - goto error; + goto send_streams_error; } /* List MUST be empty after or else it could be reused. */ assert(cds_list_empty(&metadata->streams.head)); @@ -756,6 +941,7 @@ error: consumer_stream_destroy(metadata->metadata_stream, NULL); cds_list_del(&metadata->metadata_stream->send_node); metadata->metadata_stream = NULL; +send_streams_error: error_no_stream: end: return ret; @@ -763,14 +949,15 @@ end: /* * Snapshot the whole metadata. + * RCU read-side lock must be held by the caller. * * Returns 0 on success, < 0 on error */ -static int snapshot_metadata(uint64_t key, char *path, uint64_t relayd_id, +static int snapshot_metadata(struct lttng_consumer_channel *metadata_channel, + uint64_t key, char *path, uint64_t relayd_id, struct lttng_consumer_local_data *ctx) { int ret = 0; - struct lttng_consumer_channel *metadata_channel; struct lttng_consumer_stream *metadata_stream; assert(path); @@ -781,13 +968,6 @@ static int snapshot_metadata(uint64_t key, char *path, uint64_t relayd_id, rcu_read_lock(); - metadata_channel = consumer_find_channel(key); - if (!metadata_channel) { - ERR("UST snapshot metadata channel not found for key %" PRIu64, - key); - ret = -1; - goto error; - } assert(!metadata_channel->monitor); health_code_update(); @@ -815,28 +995,23 @@ static int snapshot_metadata(uint64_t key, char *path, uint64_t relayd_id, metadata_stream = metadata_channel->metadata_stream; assert(metadata_stream); + pthread_mutex_lock(&metadata_stream->lock); if (relayd_id != (uint64_t) -1ULL) { metadata_stream->net_seq_idx = relayd_id; ret = consumer_send_relayd_stream(metadata_stream, path); - if (ret < 0) { - goto error_stream; - } } else { - ret = utils_create_stream_file(path, metadata_stream->name, - metadata_stream->chan->tracefile_size, - metadata_stream->tracefile_count_current, - metadata_stream->uid, metadata_stream->gid, NULL); - if (ret < 0) { - goto error_stream; - } - metadata_stream->out_fd = ret; - metadata_stream->tracefile_size_current = 0; + ret = consumer_stream_create_output_files(metadata_stream, + false); + } + pthread_mutex_unlock(&metadata_stream->lock); + if (ret < 0) { + goto error_stream; } do { health_code_update(); - ret = lttng_consumer_read_subbuffer(metadata_stream, ctx); + ret = lttng_consumer_read_subbuffer(metadata_stream, ctx, true); if (ret < 0) { goto error_stream; } @@ -856,18 +1031,49 @@ error: return ret; } +static +int get_current_subbuf_addr(struct lttng_consumer_stream *stream, + const char **addr) +{ + int ret; + unsigned long mmap_offset; + const char *mmap_base; + + mmap_base = ustctl_get_mmap_base(stream->ustream); + if (!mmap_base) { + ERR("Failed to get mmap base for stream `%s`", + stream->name); + ret = -EPERM; + goto error; + } + + ret = ustctl_get_mmap_read_offset(stream->ustream, &mmap_offset); + if (ret != 0) { + ERR("Failed to get mmap offset for stream `%s`", stream->name); + ret = -EINVAL; + goto error; + } + + *addr = mmap_base + mmap_offset; +error: + return ret; + +} + /* * Take a snapshot of all the stream of a channel. + * RCU read-side lock and the channel lock must be held by the caller. * * Returns 0 on success, < 0 on error */ -static int snapshot_channel(uint64_t key, char *path, uint64_t relayd_id, - uint64_t nb_packets_per_stream, struct lttng_consumer_local_data *ctx) +static int snapshot_channel(struct lttng_consumer_channel *channel, + uint64_t key, char *path, uint64_t relayd_id, + uint64_t nb_packets_per_stream, + struct lttng_consumer_local_data *ctx) { int ret; unsigned use_relayd = 0; unsigned long consumed_pos, produced_pos; - struct lttng_consumer_channel *channel; struct lttng_consumer_stream *stream; assert(path); @@ -879,21 +1085,27 @@ static int snapshot_channel(uint64_t key, char *path, uint64_t relayd_id, use_relayd = 1; } - channel = consumer_find_channel(key); - if (!channel) { - ERR("UST snapshot channel not found for key %" PRIu64, key); - ret = -1; - goto error; - } assert(!channel->monitor); DBG("UST consumer snapshot channel %" PRIu64, key); cds_list_for_each_entry(stream, &channel->streams.head, send_node) { - health_code_update(); /* Lock stream because we are about to change its state. */ pthread_mutex_lock(&stream->lock); + assert(channel->trace_chunk); + if (!lttng_trace_chunk_get(channel->trace_chunk)) { + /* + * Can't happen barring an internal error as the channel + * holds a reference to the trace chunk. + */ + ERR("Failed to acquire reference to channel's trace chunk"); + ret = -1; + goto error_unlock; + } + assert(!stream->trace_chunk); + stream->trace_chunk = channel->trace_chunk; + stream->net_seq_idx = relayd_id; if (use_relayd) { @@ -902,27 +1114,22 @@ static int snapshot_channel(uint64_t key, char *path, uint64_t relayd_id, goto error_unlock; } } else { - ret = utils_create_stream_file(path, stream->name, - stream->chan->tracefile_size, - stream->tracefile_count_current, - stream->uid, stream->gid, NULL); - if (ret < 0) { - goto error_unlock; - } - stream->out_fd = ret; - stream->tracefile_size_current = 0; - - DBG("UST consumer snapshot stream %s/%s (%" PRIu64 ")", path, - stream->name, stream->key); - } - if (relayd_id != -1ULL) { - ret = consumer_send_relayd_streams_sent(relayd_id); + ret = consumer_stream_create_output_files(stream, + false); if (ret < 0) { goto error_unlock; } + DBG("UST consumer snapshot stream (%" PRIu64 ")", + stream->key); } - ustctl_flush_buffer(stream->ustream, 1); + /* + * If tracing is active, we want to perform a "full" buffer flush. + * Else, if quiescent, it has already been done by the prior stop. + */ + if (!stream->quiescent) { + ustctl_flush_buffer(stream->ustream, 0); + } ret = lttng_ustconsumer_take_snapshot(stream); if (ret < 0) { @@ -952,9 +1159,11 @@ static int snapshot_channel(uint64_t key, char *path, uint64_t relayd_id, produced_pos, nb_packets_per_stream, stream->max_sb_size); - while (consumed_pos < produced_pos) { + while ((long) (consumed_pos - produced_pos) < 0) { ssize_t read_len; unsigned long len, padded_len; + const char *subbuf_addr; + struct lttng_buffer_view subbuf_view; health_code_update(); @@ -968,6 +1177,7 @@ static int snapshot_channel(uint64_t key, char *path, uint64_t relayd_id, } DBG("UST consumer get subbuf failed. Skipping it."); consumed_pos += stream->max_sb_size; + stream->chan->lost_packets++; continue; } @@ -983,8 +1193,15 @@ static int snapshot_channel(uint64_t key, char *path, uint64_t relayd_id, goto error_put_subbuf; } - read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len, - padded_len - len, NULL); + ret = get_current_subbuf_addr(stream, &subbuf_addr); + if (ret) { + goto error_put_subbuf; + } + + subbuf_view = lttng_buffer_view_init( + subbuf_addr, 0, padded_len); + read_len = lttng_consumer_on_read_subbuffer_mmap( + stream, &subbuf_view, padded_len - len); if (use_relayd) { if (read_len != len) { ret = -EPERM; @@ -1021,20 +1238,36 @@ error_close_stream: consumer_stream_close(stream); error_unlock: pthread_mutex_unlock(&stream->lock); -error: rcu_read_unlock(); return ret; } +static +void metadata_stream_reset_cache_consumed_position( + struct lttng_consumer_stream *stream) +{ + ASSERT_LOCKED(stream->lock); + + DBG("Reset metadata cache of session %" PRIu64, + stream->chan->session_id); + stream->ust_metadata_pushed = 0; +} + /* - * Receive the metadata updates from the sessiond. + * Receive the metadata updates from the sessiond. Supports receiving + * overlapping metadata, but is needs to always belong to a contiguous + * range starting from 0. + * Be careful about the locks held when calling this function: it needs + * the metadata cache flush to concurrently progress in order to + * complete. */ int lttng_ustconsumer_recv_metadata(int sock, uint64_t key, uint64_t offset, - uint64_t len, struct lttng_consumer_channel *channel, - int timer, int wait) + uint64_t len, uint64_t version, + struct lttng_consumer_channel *channel, int timer, int wait) { int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS; char *metadata_str; + enum consumer_metadata_cache_write_status cache_write_status; DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key, len); @@ -1058,8 +1291,41 @@ int lttng_ustconsumer_recv_metadata(int sock, uint64_t key, uint64_t offset, health_code_update(); pthread_mutex_lock(&channel->metadata_cache->lock); - ret = consumer_metadata_cache_write(channel, offset, len, metadata_str); - if (ret < 0) { + cache_write_status = consumer_metadata_cache_write( + channel->metadata_cache, offset, len, version, + metadata_str); + pthread_mutex_unlock(&channel->metadata_cache->lock); + switch (cache_write_status) { + case CONSUMER_METADATA_CACHE_WRITE_STATUS_NO_CHANGE: + /* + * The write entirely overlapped with existing contents of the + * same metadata version (same content); there is nothing to do. + */ + break; + case CONSUMER_METADATA_CACHE_WRITE_STATUS_INVALIDATED: + /* + * The metadata cache was invalidated (previously pushed + * content has been overwritten). Reset the stream's consumed + * metadata position to ensure the metadata poll thread consumes + * the whole cache. + */ + pthread_mutex_lock(&channel->metadata_stream->lock); + metadata_stream_reset_cache_consumed_position( + channel->metadata_stream); + pthread_mutex_unlock(&channel->metadata_stream->lock); + /* Fall-through. */ + case CONSUMER_METADATA_CACHE_WRITE_STATUS_APPENDED_CONTENT: + /* + * In both cases, the metadata poll thread has new data to + * consume. + */ + ret = consumer_metadata_wakeup_pipe(channel); + if (ret) { + ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA; + goto end_free; + } + break; + case CONSUMER_METADATA_CACHE_WRITE_STATUS_ERROR: /* Unable to handle metadata. Notify session daemon. */ ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA; /* @@ -1067,10 +1333,10 @@ int lttng_ustconsumer_recv_metadata(int sock, uint64_t key, uint64_t offset, * not have been updated which could create an infinite loop below when * waiting for the metadata cache to be flushed. */ - pthread_mutex_unlock(&channel->metadata_cache->lock); goto end_free; + default: + abort(); } - pthread_mutex_unlock(&channel->metadata_cache->lock); if (!wait) { goto end_free; @@ -1097,26 +1363,31 @@ end: int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, int sock, struct pollfd *consumer_sockpoll) { - ssize_t ret; + int ret_func; enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS; struct lttcomm_consumer_msg msg; struct lttng_consumer_channel *channel = NULL; health_code_update(); - ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg)); - if (ret != sizeof(msg)) { - DBG("Consumer received unexpected message size %zd (expects %zu)", - ret, sizeof(msg)); - /* - * The ret value might 0 meaning an orderly shutdown but this is ok - * since the caller handles this. - */ - if (ret > 0) { - lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD); - ret = -1; + { + ssize_t ret_recv; + + ret_recv = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg)); + if (ret_recv != sizeof(msg)) { + DBG("Consumer received unexpected message size %zd (expects %zu)", + ret_recv, sizeof(msg)); + /* + * The ret value might 0 meaning an orderly shutdown but this is ok + * since the caller handles this. + */ + if (ret_recv > 0) { + lttng_consumer_send_error(ctx, + LTTCOMM_CONSUMERD_ERROR_RECV_CMD); + ret_recv = -1; + } + return ret_recv; } - return ret; } health_code_update(); @@ -1133,7 +1404,7 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, case LTTNG_CONSUMER_ADD_RELAYD_SOCKET: { /* Session daemon status message are handled in the following call. */ - ret = consumer_add_relayd_socket(msg.u.relayd_sock.net_index, + consumer_add_relayd_socket(msg.u.relayd_sock.net_index, msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll, &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id, msg.u.relayd_sock.relayd_session_id); @@ -1176,7 +1447,8 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, } case LTTNG_CONSUMER_DATA_PENDING: { - int ret, is_data_pending; + int is_data_pending; + ssize_t ret_send; uint64_t id = msg.u.data_pending.session_id; DBG("UST consumer data pending command for id %" PRIu64, id); @@ -1184,10 +1456,11 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, is_data_pending = consumer_data_pending(id); /* Send back returned value to session daemon */ - ret = lttcomm_send_unix_sock(sock, &is_data_pending, + ret_send = lttcomm_send_unix_sock(sock, &is_data_pending, sizeof(is_data_pending)); - if (ret < 0) { - DBG("Error when sending the data pending ret code: %d", ret); + if (ret_send < 0) { + DBG("Error when sending the data pending ret code: %zd", + ret_send); goto error_fatal; } @@ -1199,25 +1472,39 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, } case LTTNG_CONSUMER_ASK_CHANNEL_CREATION: { - int ret; + int ret_ask_channel, ret_add_channel, ret_send; struct ustctl_consumer_channel_attr attr; + const uint64_t chunk_id = msg.u.ask_channel.chunk_id.value; + const struct lttng_credentials buffer_credentials = { + .uid = LTTNG_OPTIONAL_INIT_VALUE(msg.u.ask_channel.buffer_credentials.uid), + .gid = LTTNG_OPTIONAL_INIT_VALUE(msg.u.ask_channel.buffer_credentials.gid), + }; /* Create a plain object and reserve a channel key. */ - channel = allocate_channel(msg.u.ask_channel.session_id, - msg.u.ask_channel.pathname, msg.u.ask_channel.name, - msg.u.ask_channel.uid, msg.u.ask_channel.gid, - msg.u.ask_channel.relayd_id, msg.u.ask_channel.key, + channel = consumer_allocate_channel( + msg.u.ask_channel.key, + msg.u.ask_channel.session_id, + msg.u.ask_channel.chunk_id.is_set ? + &chunk_id : NULL, + msg.u.ask_channel.pathname, + msg.u.ask_channel.name, + msg.u.ask_channel.relayd_id, (enum lttng_event_output) msg.u.ask_channel.output, msg.u.ask_channel.tracefile_size, msg.u.ask_channel.tracefile_count, msg.u.ask_channel.session_id_per_pid, msg.u.ask_channel.monitor, msg.u.ask_channel.live_timer_interval, + msg.u.ask_channel.is_live, + msg.u.ask_channel.root_shm_path, msg.u.ask_channel.shm_path); if (!channel) { goto end_channel_error; } + LTTNG_OPTIONAL_SET(&channel->buffer_credentials, + buffer_credentials); + /* * Assign UST application UID to the channel. This value is ignored for * per PID buffers. This is specific to UST thus setting this after the @@ -1233,23 +1520,21 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, attr.read_timer_interval = msg.u.ask_channel.read_timer_interval; attr.chan_id = msg.u.ask_channel.chan_id; memcpy(attr.uuid, msg.u.ask_channel.uuid, sizeof(attr.uuid)); - strncpy(attr.shm_path, channel->shm_path, - sizeof(attr.shm_path)); - attr.shm_path[sizeof(attr.shm_path) - 1] = '\0'; + attr.blocking_timeout= msg.u.ask_channel.blocking_timeout; /* Match channel buffer type to the UST abi. */ switch (msg.u.ask_channel.output) { case LTTNG_EVENT_MMAP: default: - attr.output = LTTNG_UST_MMAP; + attr.output = LTTNG_UST_ABI_MMAP; break; } /* Translate and save channel type. */ switch (msg.u.ask_channel.type) { - case LTTNG_UST_CHAN_PER_CPU: + case LTTNG_UST_ABI_CHAN_PER_CPU: channel->type = CONSUMER_CHANNEL_TYPE_DATA; - attr.type = LTTNG_UST_CHAN_PER_CPU; + attr.type = LTTNG_UST_ABI_CHAN_PER_CPU; /* * Set refcount to 1 for owner. Below, we will * pass ownership to the @@ -1257,9 +1542,9 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, */ channel->refcount = 1; break; - case LTTNG_UST_CHAN_METADATA: + case LTTNG_UST_ABI_CHAN_METADATA: channel->type = CONSUMER_CHANNEL_TYPE_METADATA; - attr.type = LTTNG_UST_CHAN_METADATA; + attr.type = LTTNG_UST_ABI_CHAN_METADATA; break; default: assert(0); @@ -1268,22 +1553,34 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, health_code_update(); - ret = ask_channel(ctx, sock, channel, &attr); - if (ret < 0) { + ret_ask_channel = ask_channel(ctx, channel, &attr); + if (ret_ask_channel < 0) { goto end_channel_error; } - if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) { - ret = consumer_metadata_cache_allocate(channel); - if (ret < 0) { + if (msg.u.ask_channel.type == LTTNG_UST_ABI_CHAN_METADATA) { + int ret_allocate; + + ret_allocate = consumer_metadata_cache_allocate( + channel); + if (ret_allocate < 0) { ERR("Allocating metadata cache"); goto end_channel_error; } consumer_timer_switch_start(channel, attr.switch_timer_interval); attr.switch_timer_interval = 0; } else { + int monitor_start_ret; + consumer_timer_live_start(channel, msg.u.ask_channel.live_timer_interval); + monitor_start_ret = consumer_timer_monitor_start( + channel, + msg.u.ask_channel.monitor_timer_interval); + if (monitor_start_ret < 0) { + ERR("Starting channel monitoring timer failed"); + goto end_channel_error; + } } health_code_update(); @@ -1295,9 +1592,9 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, * If add_channel succeeds, ownership of the channel is * passed to consumer_thread_channel_poll(). */ - ret = add_channel(channel, ctx); - if (ret < 0) { - if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) { + ret_add_channel = add_channel(channel, ctx); + if (ret_add_channel < 0) { + if (msg.u.ask_channel.type == LTTNG_UST_ABI_CHAN_METADATA) { if (channel->switch_timer_enabled == 1) { consumer_timer_switch_stop(channel); } @@ -1306,6 +1603,9 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, if (channel->live_timer_enabled == 1) { consumer_timer_live_stop(channel); } + if (channel->monitor_timer_enabled == 1) { + consumer_timer_monitor_stop(channel); + } goto end_channel_error; } @@ -1316,8 +1616,8 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, * everything went well and should wait to receive the channel and * streams with ustctl API. */ - ret = consumer_send_status_channel(sock, channel); - if (ret < 0) { + ret_send = consumer_send_status_channel(sock, channel); + if (ret_send < 0) { /* * There is probably a problem on the socket. */ @@ -1330,19 +1630,20 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, { int ret, relayd_err = 0; uint64_t key = msg.u.get_channel.key; - struct lttng_consumer_channel *channel; + struct lttng_consumer_channel *found_channel; - channel = consumer_find_channel(key); - if (!channel) { + found_channel = consumer_find_channel(key); + if (!found_channel) { ERR("UST consumer get channel key %" PRIu64 " not found", key); ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND; - goto end_msg_sessiond; + goto end_get_channel; } health_code_update(); - /* Send everything to sessiond. */ - ret = send_sessiond_channel(sock, channel, ctx, &relayd_err); + /* Send the channel to sessiond (and relayd, if applicable). */ + ret = send_channel_to_sessiond_and_relayd( + sock, found_channel, ctx, &relayd_err); if (ret < 0) { if (relayd_err) { /* @@ -1351,13 +1652,13 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, * and the consumer can continue its work. The above call * has sent the error status message to the sessiond. */ - goto end_nosignal; + goto end_get_channel_nosignal; } /* * The communicaton was broken hence there is a bad state between * the consumer and sessiond so stop everything. */ - goto error_fatal; + goto error_get_channel_fatal; } health_code_update(); @@ -1366,21 +1667,26 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, * In no monitor mode, the streams ownership is kept inside the channel * so don't send them to the data thread. */ - if (!channel->monitor) { - goto end_msg_sessiond; + if (!found_channel->monitor) { + goto end_get_channel; } - ret = send_streams_to_thread(channel, ctx); + ret = send_streams_to_thread(found_channel, ctx); if (ret < 0) { /* * If we are unable to send the stream to the thread, there is * a big problem so just stop everything. */ - goto error_fatal; + goto error_get_channel_fatal; } /* List MUST be empty after or else it could be reused. */ - assert(cds_list_empty(&channel->streams.head)); + assert(cds_list_empty(&found_channel->streams.head)); +end_get_channel: goto end_msg_sessiond; +error_get_channel_fatal: + goto error_fatal; +end_get_channel_nosignal: + goto end_nosignal; } case LTTNG_CONSUMER_DESTROY_CHANNEL: { @@ -1416,19 +1722,32 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, goto end_msg_sessiond; } + case LTTNG_CONSUMER_CLEAR_QUIESCENT_CHANNEL: + { + int ret; + + ret = clear_quiescent_channel( + msg.u.clear_quiescent_channel.key); + if (ret != 0) { + ret_code = ret; + } + + goto end_msg_sessiond; + } case LTTNG_CONSUMER_PUSH_METADATA: { int ret; uint64_t len = msg.u.push_metadata.len; uint64_t key = msg.u.push_metadata.key; uint64_t offset = msg.u.push_metadata.target_offset; - struct lttng_consumer_channel *channel; + uint64_t version = msg.u.push_metadata.version; + struct lttng_consumer_channel *found_channel; DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key, len); - channel = consumer_find_channel(key); - if (!channel) { + found_channel = consumer_find_channel(key); + if (!found_channel) { /* * This is possible if the metadata creation on the consumer side * is in flight vis-a-vis a concurrent push metadata from the @@ -1438,16 +1757,25 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, */ DBG("UST consumer push metadata %" PRIu64 " not found", key); ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL; - goto end_msg_sessiond; + goto end_push_metadata_msg_sessiond; } health_code_update(); + if (!len) { + /* + * There is nothing to receive. We have simply + * checked whether the channel can be found. + */ + ret_code = LTTCOMM_CONSUMERD_SUCCESS; + goto end_push_metadata_msg_sessiond; + } + /* Tell session daemon we are ready to receive the metadata. */ ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS); if (ret < 0) { /* Somehow, the session daemon is not responding anymore. */ - goto error_fatal; + goto error_push_metadata_fatal; } health_code_update(); @@ -1457,20 +1785,24 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, ret = lttng_consumer_poll_socket(consumer_sockpoll); health_poll_exit(); if (ret) { - goto error_fatal; + goto error_push_metadata_fatal; } health_code_update(); - ret = lttng_ustconsumer_recv_metadata(sock, key, offset, - len, channel, 0, 1); + ret = lttng_ustconsumer_recv_metadata(sock, key, offset, len, + version, found_channel, 0, 1); if (ret < 0) { /* error receiving from sessiond */ - goto error_fatal; + goto error_push_metadata_fatal; } else { ret_code = ret; - goto end_msg_sessiond; + goto end_push_metadata_msg_sessiond; } +end_push_metadata_msg_sessiond: + goto end_msg_sessiond; +error_push_metadata_fatal: + goto error_fatal; } case LTTNG_CONSUMER_SETUP_METADATA: { @@ -1484,50 +1816,452 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, } case LTTNG_CONSUMER_SNAPSHOT_CHANNEL: { - if (msg.u.snapshot_channel.metadata) { - ret = snapshot_metadata(msg.u.snapshot_channel.key, - msg.u.snapshot_channel.pathname, - msg.u.snapshot_channel.relayd_id, - ctx); - if (ret < 0) { - ERR("Snapshot metadata failed"); - ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA; - } + struct lttng_consumer_channel *found_channel; + uint64_t key = msg.u.snapshot_channel.key; + int ret_send; + + found_channel = consumer_find_channel(key); + if (!found_channel) { + DBG("UST snapshot channel not found for key %" PRIu64, key); + ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND; } else { - ret = snapshot_channel(msg.u.snapshot_channel.key, - msg.u.snapshot_channel.pathname, - msg.u.snapshot_channel.relayd_id, - msg.u.snapshot_channel.nb_packets_per_stream, - ctx); - if (ret < 0) { - ERR("Snapshot channel failed"); - ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL; + if (msg.u.snapshot_channel.metadata) { + int ret_snapshot; + + ret_snapshot = snapshot_metadata(found_channel, + key, + msg.u.snapshot_channel.pathname, + msg.u.snapshot_channel.relayd_id, + ctx); + if (ret_snapshot < 0) { + ERR("Snapshot metadata failed"); + ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED; + } + } else { + int ret_snapshot; + + ret_snapshot = snapshot_channel(found_channel, + key, + msg.u.snapshot_channel.pathname, + msg.u.snapshot_channel.relayd_id, + msg.u.snapshot_channel + .nb_packets_per_stream, + ctx); + if (ret_snapshot < 0) { + ERR("Snapshot channel failed"); + ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED; + } + } + } + health_code_update(); + ret_send = consumer_send_status_msg(sock, ret_code); + if (ret_send < 0) { + /* Somehow, the session daemon is not responding anymore. */ + goto end_nosignal; + } + health_code_update(); + break; + } + case LTTNG_CONSUMER_DISCARDED_EVENTS: + { + int ret = 0; + uint64_t discarded_events; + struct lttng_ht_iter iter; + struct lttng_ht *ht; + struct lttng_consumer_stream *stream; + uint64_t id = msg.u.discarded_events.session_id; + uint64_t key = msg.u.discarded_events.channel_key; + + DBG("UST consumer discarded events command for session id %" + PRIu64, id); + rcu_read_lock(); + pthread_mutex_lock(&consumer_data.lock); + + ht = consumer_data.stream_list_ht; + + /* + * We only need a reference to the channel, but they are not + * directly indexed, so we just use the first matching stream + * to extract the information we need, we default to 0 if not + * found (no events are dropped if the channel is not yet in + * use). + */ + discarded_events = 0; + 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->chan->key == key) { + discarded_events = stream->chan->discarded_events; + break; } } + pthread_mutex_unlock(&consumer_data.lock); + rcu_read_unlock(); + + DBG("UST consumer discarded events command for session id %" + PRIu64 ", channel key %" PRIu64, id, key); health_code_update(); - ret = consumer_send_status_msg(sock, ret_code); + + /* Send back returned value to session daemon */ + ret = lttcomm_send_unix_sock(sock, &discarded_events, sizeof(discarded_events)); if (ret < 0) { + PERROR("send discarded events"); + goto error_fatal; + } + + break; + } + case LTTNG_CONSUMER_LOST_PACKETS: + { + int ret; + uint64_t lost_packets; + struct lttng_ht_iter iter; + struct lttng_ht *ht; + struct lttng_consumer_stream *stream; + uint64_t id = msg.u.lost_packets.session_id; + uint64_t key = msg.u.lost_packets.channel_key; + + DBG("UST consumer lost packets command for session id %" + PRIu64, id); + rcu_read_lock(); + pthread_mutex_lock(&consumer_data.lock); + + ht = consumer_data.stream_list_ht; + + /* + * We only need a reference to the channel, but they are not + * directly indexed, so we just use the first matching stream + * to extract the information we need, we default to 0 if not + * found (no packets lost if the channel is not yet in use). + */ + lost_packets = 0; + 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->chan->key == key) { + lost_packets = stream->chan->lost_packets; + break; + } + } + pthread_mutex_unlock(&consumer_data.lock); + rcu_read_unlock(); + + DBG("UST consumer lost packets command for session id %" + PRIu64 ", channel key %" PRIu64, id, key); + + health_code_update(); + + /* Send back returned value to session daemon */ + ret = lttcomm_send_unix_sock(sock, &lost_packets, + sizeof(lost_packets)); + if (ret < 0) { + PERROR("send lost packets"); + goto error_fatal; + } + + break; + } + case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE: + { + int channel_monitor_pipe, ret_send, + ret_set_channel_monitor_pipe; + ssize_t ret_recv; + + ret_code = LTTCOMM_CONSUMERD_SUCCESS; + /* Successfully received the command's type. */ + ret_send = consumer_send_status_msg(sock, ret_code); + if (ret_send < 0) { + goto error_fatal; + } + + ret_recv = lttcomm_recv_fds_unix_sock( + sock, &channel_monitor_pipe, 1); + if (ret_recv != sizeof(channel_monitor_pipe)) { + ERR("Failed to receive channel monitor pipe"); + goto error_fatal; + } + + DBG("Received channel monitor pipe (%d)", channel_monitor_pipe); + ret_set_channel_monitor_pipe = + consumer_timer_thread_set_channel_monitor_pipe( + channel_monitor_pipe); + if (!ret_set_channel_monitor_pipe) { + int flags; + int ret_fcntl; + + ret_code = LTTCOMM_CONSUMERD_SUCCESS; + /* Set the pipe as non-blocking. */ + ret_fcntl = fcntl(channel_monitor_pipe, F_GETFL, 0); + if (ret_fcntl == -1) { + PERROR("fcntl get flags of the channel monitoring pipe"); + goto error_fatal; + } + flags = ret_fcntl; + + ret_fcntl = fcntl(channel_monitor_pipe, F_SETFL, + flags | O_NONBLOCK); + if (ret_fcntl == -1) { + PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe"); + goto error_fatal; + } + DBG("Channel monitor pipe set as non-blocking"); + } else { + ret_code = LTTCOMM_CONSUMERD_ALREADY_SET; + } + goto end_msg_sessiond; + } + case LTTNG_CONSUMER_ROTATE_CHANNEL: + { + struct lttng_consumer_channel *found_channel; + uint64_t key = msg.u.rotate_channel.key; + int ret_send_status; + + found_channel = consumer_find_channel(key); + if (!found_channel) { + DBG("Channel %" PRIu64 " not found", key); + ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND; + } else { + int rotate_channel; + + /* + * Sample the rotate position of all the streams in + * this channel. + */ + rotate_channel = lttng_consumer_rotate_channel( + found_channel, key, + msg.u.rotate_channel.relayd_id, + msg.u.rotate_channel.metadata, ctx); + if (rotate_channel < 0) { + ERR("Rotate channel failed"); + ret_code = LTTCOMM_CONSUMERD_ROTATION_FAIL; + } + + health_code_update(); + } + + ret_send_status = consumer_send_status_msg(sock, ret_code); + if (ret_send_status < 0) { + /* Somehow, the session daemon is not responding anymore. */ + goto end_rotate_channel_nosignal; + } + + /* + * Rotate the streams that are ready right now. + * FIXME: this is a second consecutive iteration over the + * streams in a channel, there is probably a better way to + * handle this, but it needs to be after the + * consumer_send_status_msg() call. + */ + if (found_channel) { + int ret_rotate_read_streams; + + ret_rotate_read_streams = + lttng_consumer_rotate_ready_streams( + found_channel, key, + ctx); + if (ret_rotate_read_streams < 0) { + ERR("Rotate channel failed"); + } + } + break; +end_rotate_channel_nosignal: + goto end_nosignal; + } + case LTTNG_CONSUMER_CLEAR_CHANNEL: + { + struct lttng_consumer_channel *found_channel; + uint64_t key = msg.u.clear_channel.key; + int ret_send_status; + + found_channel = consumer_find_channel(key); + if (!found_channel) { + DBG("Channel %" PRIu64 " not found", key); + ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND; + } else { + int ret_clear_channel; + + ret_clear_channel = lttng_consumer_clear_channel( + found_channel); + if (ret_clear_channel) { + ERR("Clear channel failed key %" PRIu64, key); + ret_code = ret_clear_channel; + } + + health_code_update(); + } + ret_send_status = consumer_send_status_msg(sock, ret_code); + if (ret_send_status < 0) { /* Somehow, the session daemon is not responding anymore. */ goto end_nosignal; } + break; + } + case LTTNG_CONSUMER_INIT: + { + int ret_send_status; + + ret_code = lttng_consumer_init_command(ctx, + msg.u.init.sessiond_uuid); health_code_update(); + ret_send_status = consumer_send_status_msg(sock, ret_code); + if (ret_send_status < 0) { + /* Somehow, the session daemon is not responding anymore. */ + goto end_nosignal; + } break; } + case LTTNG_CONSUMER_CREATE_TRACE_CHUNK: + { + const struct lttng_credentials credentials = { + .uid = LTTNG_OPTIONAL_INIT_VALUE(msg.u.create_trace_chunk.credentials.value.uid), + .gid = LTTNG_OPTIONAL_INIT_VALUE(msg.u.create_trace_chunk.credentials.value.gid), + }; + const bool is_local_trace = + !msg.u.create_trace_chunk.relayd_id.is_set; + const uint64_t relayd_id = + msg.u.create_trace_chunk.relayd_id.value; + const char *chunk_override_name = + *msg.u.create_trace_chunk.override_name ? + msg.u.create_trace_chunk.override_name : + NULL; + struct lttng_directory_handle *chunk_directory_handle = NULL; + + /* + * The session daemon will only provide a chunk directory file + * descriptor for local traces. + */ + if (is_local_trace) { + int chunk_dirfd; + int ret_send_status; + ssize_t ret_recv; + + /* Acnowledge the reception of the command. */ + ret_send_status = consumer_send_status_msg( + sock, LTTCOMM_CONSUMERD_SUCCESS); + if (ret_send_status < 0) { + /* Somehow, the session daemon is not responding anymore. */ + goto end_nosignal; + } + + /* + * Receive trace chunk domain dirfd. + */ + ret_recv = lttcomm_recv_fds_unix_sock( + sock, &chunk_dirfd, 1); + if (ret_recv != sizeof(chunk_dirfd)) { + ERR("Failed to receive trace chunk domain directory file descriptor"); + goto error_fatal; + } + + DBG("Received trace chunk domain directory fd (%d)", + chunk_dirfd); + chunk_directory_handle = lttng_directory_handle_create_from_dirfd( + chunk_dirfd); + if (!chunk_directory_handle) { + ERR("Failed to initialize chunk domain directory handle from directory file descriptor"); + if (close(chunk_dirfd)) { + PERROR("Failed to close chunk directory file descriptor"); + } + goto error_fatal; + } + } + + ret_code = lttng_consumer_create_trace_chunk( + !is_local_trace ? &relayd_id : NULL, + msg.u.create_trace_chunk.session_id, + msg.u.create_trace_chunk.chunk_id, + (time_t) msg.u.create_trace_chunk + .creation_timestamp, + chunk_override_name, + msg.u.create_trace_chunk.credentials.is_set ? + &credentials : + NULL, + chunk_directory_handle); + lttng_directory_handle_put(chunk_directory_handle); + goto end_msg_sessiond; + } + case LTTNG_CONSUMER_CLOSE_TRACE_CHUNK: + { + enum lttng_trace_chunk_command_type close_command = + msg.u.close_trace_chunk.close_command.value; + const uint64_t relayd_id = + msg.u.close_trace_chunk.relayd_id.value; + struct lttcomm_consumer_close_trace_chunk_reply reply; + char closed_trace_chunk_path[LTTNG_PATH_MAX]; + int ret; + + ret_code = lttng_consumer_close_trace_chunk( + msg.u.close_trace_chunk.relayd_id.is_set ? + &relayd_id : + NULL, + msg.u.close_trace_chunk.session_id, + msg.u.close_trace_chunk.chunk_id, + (time_t) msg.u.close_trace_chunk.close_timestamp, + msg.u.close_trace_chunk.close_command.is_set ? + &close_command : + NULL, closed_trace_chunk_path); + reply.ret_code = ret_code; + reply.path_length = strlen(closed_trace_chunk_path) + 1; + ret = lttcomm_send_unix_sock(sock, &reply, sizeof(reply)); + if (ret != sizeof(reply)) { + goto error_fatal; + } + ret = lttcomm_send_unix_sock(sock, closed_trace_chunk_path, + reply.path_length); + if (ret != reply.path_length) { + goto error_fatal; + } + goto end_nosignal; + } + case LTTNG_CONSUMER_TRACE_CHUNK_EXISTS: + { + const uint64_t relayd_id = + msg.u.trace_chunk_exists.relayd_id.value; + + ret_code = lttng_consumer_trace_chunk_exists( + msg.u.trace_chunk_exists.relayd_id.is_set ? + &relayd_id : NULL, + msg.u.trace_chunk_exists.session_id, + msg.u.trace_chunk_exists.chunk_id); + goto end_msg_sessiond; + } + case LTTNG_CONSUMER_OPEN_CHANNEL_PACKETS: + { + const uint64_t key = msg.u.open_channel_packets.key; + struct lttng_consumer_channel *found_channel = + consumer_find_channel(key); + + if (found_channel) { + pthread_mutex_lock(&found_channel->lock); + ret_code = lttng_consumer_open_channel_packets( + found_channel); + pthread_mutex_unlock(&found_channel->lock); + } else { + /* + * The channel could have disappeared in per-pid + * buffering mode. + */ + DBG("Channel %" PRIu64 " not found", key); + ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND; + } + + health_code_update(); + goto end_msg_sessiond; + } default: break; } end_nosignal: - rcu_read_unlock(); - - health_code_update(); - /* * Return 1 to indicate success since the 0 value can be a socket * shutdown during the recv() or send() call. */ - return 1; + ret_func = 1; + goto end; end_msg_sessiond: /* @@ -1535,15 +2269,18 @@ end_msg_sessiond: * the caller because the session daemon socket management is done * elsewhere. Returning a negative code or 0 will shutdown the consumer. */ - ret = consumer_send_status_msg(sock, ret_code); - if (ret < 0) { - goto error_fatal; + { + int ret_send_status; + + ret_send_status = consumer_send_status_msg(sock, ret_code); + if (ret_send_status < 0) { + goto error_fatal; + } } - rcu_read_unlock(); - health_code_update(); + ret_func = 1; + goto end; - return 1; end_channel_error: if (channel) { /* @@ -1553,58 +2290,64 @@ end_channel_error: destroy_channel(channel); } /* We have to send a status channel message indicating an error. */ - ret = consumer_send_status_channel(sock, NULL); - if (ret < 0) { - /* Stop everything if session daemon can not be notified. */ - goto error_fatal; + { + int ret_send_status; + + ret_send_status = consumer_send_status_channel(sock, NULL); + if (ret_send_status < 0) { + /* Stop everything if session daemon can not be notified. */ + goto error_fatal; + } } - rcu_read_unlock(); - health_code_update(); + ret_func = 1; + goto end; - return 1; error_fatal: - rcu_read_unlock(); /* This will issue a consumer stop. */ - return -1; + ret_func = -1; + goto end; + +end: + rcu_read_unlock(); + health_code_update(); + return ret_func; } -/* - * Wrapper over the mmap() read offset from ust-ctl library. Since this can be - * compiled out, we isolate it in this library. - */ -int lttng_ustctl_get_mmap_read_offset(struct lttng_consumer_stream *stream, - unsigned long *off) +void lttng_ustctl_flush_buffer(struct lttng_consumer_stream *stream, + int producer_active) { assert(stream); assert(stream->ustream); - return ustctl_get_mmap_read_offset(stream->ustream, off); + ustctl_flush_buffer(stream->ustream, producer_active); } /* - * Wrapper over the mmap() read offset from ust-ctl library. Since this can be - * compiled out, we isolate it in this library. + * Take a snapshot for a specific stream. + * + * Returns 0 on success, < 0 on error */ -void *lttng_ustctl_get_mmap_base(struct lttng_consumer_stream *stream) +int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream *stream) { assert(stream); assert(stream->ustream); - return ustctl_get_mmap_base(stream->ustream); + return ustctl_snapshot(stream->ustream); } /* - * Take a snapshot for a specific fd + * Sample consumed and produced positions for a specific stream. * - * Returns 0 on success, < 0 on error + * Returns 0 on success, < 0 on error. */ -int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream *stream) +int lttng_ustconsumer_sample_snapshot_positions( + struct lttng_consumer_stream *stream) { assert(stream); assert(stream->ustream); - return ustctl_snapshot(stream->ustream); + return ustctl_snapshot_sample_positions(stream->ustream); } /* @@ -1646,6 +2389,14 @@ void lttng_ustconsumer_flush_buffer(struct lttng_consumer_stream *stream, ustctl_flush_buffer(stream->ustream, producer); } +void lttng_ustconsumer_clear_buffer(struct lttng_consumer_stream *stream) +{ + assert(stream); + assert(stream->ustream); + + ustctl_clear_buffer(stream->ustream); +} + int lttng_ustconsumer_get_current_timestamp( struct lttng_consumer_stream *stream, uint64_t *ts) { @@ -1656,28 +2407,88 @@ int lttng_ustconsumer_get_current_timestamp( return ustctl_get_current_timestamp(stream->ustream, ts); } +int lttng_ustconsumer_get_sequence_number( + struct lttng_consumer_stream *stream, uint64_t *seq) +{ + assert(stream); + assert(stream->ustream); + assert(seq); + + return ustctl_get_sequence_number(stream->ustream, seq); +} + /* - * Called when the stream signal the consumer that it has hang up. + * Called when the stream signals the consumer that it has hung up. */ void lttng_ustconsumer_on_stream_hangup(struct lttng_consumer_stream *stream) { assert(stream); assert(stream->ustream); - ustctl_flush_buffer(stream->ustream, 0); + pthread_mutex_lock(&stream->lock); + if (!stream->quiescent) { + ustctl_flush_buffer(stream->ustream, 0); + stream->quiescent = true; + } + pthread_mutex_unlock(&stream->lock); stream->hangup_flush_done = 1; } void lttng_ustconsumer_del_channel(struct lttng_consumer_channel *chan) { + int i; + assert(chan); assert(chan->uchan); + assert(chan->buffer_credentials.is_set); if (chan->switch_timer_enabled == 1) { consumer_timer_switch_stop(chan); } + for (i = 0; i < chan->nr_stream_fds; i++) { + int ret; + + ret = close(chan->stream_fds[i]); + if (ret) { + PERROR("close"); + } + if (chan->shm_path[0]) { + char shm_path[PATH_MAX]; + + ret = get_stream_shm_path(shm_path, chan->shm_path, i); + if (ret) { + ERR("Cannot get stream shm path"); + } + ret = run_as_unlink(shm_path, + lttng_credentials_get_uid(LTTNG_OPTIONAL_GET_PTR( + chan->buffer_credentials)), + lttng_credentials_get_gid(LTTNG_OPTIONAL_GET_PTR( + chan->buffer_credentials))); + if (ret) { + PERROR("unlink %s", shm_path); + } + } + } +} + +void lttng_ustconsumer_free_channel(struct lttng_consumer_channel *chan) +{ + assert(chan); + assert(chan->uchan); + assert(chan->buffer_credentials.is_set); + consumer_metadata_cache_destroy(chan); ustctl_destroy_channel(chan->uchan); + /* Try to rmdir all directories under shm_path root. */ + if (chan->root_shm_path[0]) { + (void) run_as_rmdir_recursive(chan->root_shm_path, + lttng_credentials_get_uid(LTTNG_OPTIONAL_GET_PTR( + chan->buffer_credentials)), + lttng_credentials_get_gid(LTTNG_OPTIONAL_GET_PTR( + chan->buffer_credentials)), + LTTNG_DIRECTORY_HANDLE_SKIP_NON_EMPTY_FLAG); + } + free(chan->stream_fds); } void lttng_ustconsumer_del_stream(struct lttng_consumer_stream *stream) @@ -1707,67 +2518,11 @@ int lttng_ustconsumer_close_wakeup_fd(struct lttng_consumer_stream *stream) return ustctl_stream_close_wakeup_fd(stream->ustream); } -/* - * Populate index values of a UST stream. Values are set in big endian order. - * - * Return 0 on success or else a negative value. - */ -static int get_index_values(struct ctf_packet_index *index, - struct ustctl_consumer_stream *ustream) -{ - int ret; - - ret = ustctl_get_timestamp_begin(ustream, &index->timestamp_begin); - if (ret < 0) { - PERROR("ustctl_get_timestamp_begin"); - goto error; - } - index->timestamp_begin = htobe64(index->timestamp_begin); - - ret = ustctl_get_timestamp_end(ustream, &index->timestamp_end); - if (ret < 0) { - PERROR("ustctl_get_timestamp_end"); - goto error; - } - index->timestamp_end = htobe64(index->timestamp_end); - - ret = ustctl_get_events_discarded(ustream, &index->events_discarded); - if (ret < 0) { - PERROR("ustctl_get_events_discarded"); - goto error; - } - index->events_discarded = htobe64(index->events_discarded); - - ret = ustctl_get_content_size(ustream, &index->content_size); - if (ret < 0) { - PERROR("ustctl_get_content_size"); - goto error; - } - index->content_size = htobe64(index->content_size); - - ret = ustctl_get_packet_size(ustream, &index->packet_size); - if (ret < 0) { - PERROR("ustctl_get_packet_size"); - goto error; - } - index->packet_size = htobe64(index->packet_size); - - ret = ustctl_get_stream_id(ustream, &index->stream_id); - if (ret < 0) { - PERROR("ustctl_get_stream_id"); - goto error; - } - index->stream_id = htobe64(index->stream_id); - -error: - return ret; -} - /* * Write up to one packet from the metadata cache to the channel. * - * Returns the number of bytes pushed in the cache, or a negative value - * on error. + * Returns the number of bytes pushed from the cache into the ring buffer, or a + * negative value on error. */ static int commit_one_metadata_packet(struct lttng_consumer_stream *stream) @@ -1776,28 +2531,66 @@ int commit_one_metadata_packet(struct lttng_consumer_stream *stream) int ret; pthread_mutex_lock(&stream->chan->metadata_cache->lock); - if (stream->chan->metadata_cache->contiguous - == stream->ust_metadata_pushed) { - ret = 0; - goto end; + if (stream->chan->metadata_cache->contents.size == + stream->ust_metadata_pushed) { + /* + * In the context of a user space metadata channel, a + * change in version can be detected in two ways: + * 1) During the pre-consume of the `read_subbuffer` loop, + * 2) When populating the metadata ring buffer (i.e. here). + * + * This function is invoked when there is no metadata + * available in the ring-buffer. If all data was consumed + * up to the size of the metadata cache, there is no metadata + * to insert in the ring-buffer. + * + * However, the metadata version could still have changed (a + * regeneration without any new data will yield the same cache + * size). + * + * The cache's version is checked for a version change and the + * consumed position is reset if one occurred. + * + * This check is only necessary for the user space domain as + * it has to manage the cache explicitly. If this reset was not + * performed, no metadata would be consumed (and no reset would + * occur as part of the pre-consume) until the metadata size + * exceeded the cache size. + */ + if (stream->metadata_version != + stream->chan->metadata_cache->version) { + metadata_stream_reset_cache_consumed_position(stream); + consumer_stream_metadata_set_version(stream, + stream->chan->metadata_cache->version); + } else { + ret = 0; + goto end; + } } write_len = ustctl_write_one_packet_to_channel(stream->chan->uchan, - &stream->chan->metadata_cache->data[stream->ust_metadata_pushed], - stream->chan->metadata_cache->contiguous - - stream->ust_metadata_pushed); + &stream->chan->metadata_cache->contents.data[stream->ust_metadata_pushed], + stream->chan->metadata_cache->contents.size - + stream->ust_metadata_pushed); assert(write_len != 0); if (write_len < 0) { ERR("Writing one metadata packet"); - ret = -1; + ret = write_len; goto end; } stream->ust_metadata_pushed += write_len; - assert(stream->chan->metadata_cache->contiguous >= + assert(stream->chan->metadata_cache->contents.size >= stream->ust_metadata_pushed); ret = write_len; + /* + * Switch packet (but don't open the next one) on every commit of + * a metadata packet. Since the subbuffer is fully filled (with padding, + * if needed), the stream is "quiescent" after this commit. + */ + ustctl_flush_buffer(stream->ustream, 1); + stream->quiescent = true; end: pthread_mutex_unlock(&stream->chan->metadata_cache->lock); return ret; @@ -1808,58 +2601,74 @@ end: * Sync metadata meaning request them to the session daemon and snapshot to the * metadata thread can consumer them. * - * Metadata stream lock MUST be acquired. + * Metadata stream lock is held here, but we need to release it when + * interacting with sessiond, else we cause a deadlock with live + * awaiting on metadata to be pushed out. * - * Return 0 if new metadatda is available, EAGAIN if the metadata stream - * is empty or a negative value on error. + * The RCU read side lock must be held by the caller. */ -int lttng_ustconsumer_sync_metadata(struct lttng_consumer_local_data *ctx, - struct lttng_consumer_stream *metadata) +enum sync_metadata_status lttng_ustconsumer_sync_metadata( + struct lttng_consumer_local_data *ctx, + struct lttng_consumer_stream *metadata_stream) { int ret; - int retry = 0; + enum sync_metadata_status status; + struct lttng_consumer_channel *metadata_channel; assert(ctx); - assert(metadata); + assert(metadata_stream); + metadata_channel = metadata_stream->chan; + pthread_mutex_unlock(&metadata_stream->lock); /* * Request metadata from the sessiond, but don't wait for the flush * because we locked the metadata thread. */ - ret = lttng_ustconsumer_request_metadata(ctx, metadata->chan, 0, 0); + ret = lttng_ustconsumer_request_metadata(ctx, metadata_channel, 0, 0); + pthread_mutex_lock(&metadata_stream->lock); if (ret < 0) { + status = SYNC_METADATA_STATUS_ERROR; goto end; } - ret = commit_one_metadata_packet(metadata); - if (ret <= 0) { + /* + * The metadata stream and channel can be deleted while the + * metadata stream lock was released. The streamed is checked + * for deletion before we use it further. + * + * Note that it is safe to access a logically-deleted stream since its + * existence is still guaranteed by the RCU read side lock. However, + * it should no longer be used. The close/deletion of the metadata + * channel and stream already guarantees that all metadata has been + * consumed. Therefore, there is nothing left to do in this function. + */ + if (consumer_stream_is_deleted(metadata_stream)) { + DBG("Metadata stream %" PRIu64 " was deleted during the metadata synchronization", + metadata_stream->key); + status = SYNC_METADATA_STATUS_NO_DATA; goto end; - } else if (ret > 0) { - retry = 1; } - ustctl_flush_buffer(metadata->ustream, 1); - ret = ustctl_snapshot(metadata->ustream); + ret = commit_one_metadata_packet(metadata_stream); if (ret < 0) { - if (errno != EAGAIN) { - ERR("Sync metadata, taking UST snapshot"); - goto end; - } - DBG("No new metadata when syncing them."); - /* No new metadata, exit. */ - ret = ENODATA; + status = SYNC_METADATA_STATUS_ERROR; + goto end; + } else if (ret > 0) { + status = SYNC_METADATA_STATUS_NEW_DATA; + } else /* ret == 0 */ { + status = SYNC_METADATA_STATUS_NO_DATA; goto end; } - /* - * After this flush, we still need to extract metadata. - */ - if (retry) { - ret = EAGAIN; + ret = ustctl_snapshot(metadata_stream->ustream); + if (ret < 0) { + ERR("Failed to take a snapshot of the metadata ring-buffer positions, ret = %d", ret); + status = SYNC_METADATA_STATUS_ERROR; + goto end; } end: - return ret; + return status; } /* @@ -1913,41 +2722,20 @@ end: return ret; } -/* - * Read subbuffer from the given stream. - * - * Stream lock MUST be acquired. - * - * Return 0 on success else a negative value. - */ -int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream *stream, - struct lttng_consumer_local_data *ctx) +static int consumer_stream_ust_on_wake_up(struct lttng_consumer_stream *stream) { - unsigned long len, subbuf_size, padding; - int err, write_index = 1; - long ret = 0; - struct ustctl_consumer_stream *ustream; - struct ctf_packet_index index; - - assert(stream); - assert(stream->ustream); - assert(ctx); - - DBG("In UST read_subbuffer (wait_fd: %d, name: %s)", stream->wait_fd, - stream->name); - - /* Ease our life for what's next. */ - ustream = stream->ustream; + int ret = 0; /* - * We can consume the 1 byte written into the wait_fd by UST. Don't trigger - * error if we cannot read this one byte (read returns 0), or if the error - * is EAGAIN or EWOULDBLOCK. + * We can consume the 1 byte written into the wait_fd by + * UST. Don't trigger error if we cannot read this one byte + * (read returns 0), or if the error is EAGAIN or EWOULDBLOCK. * - * This is only done when the stream is monitored by a thread, before the - * flush is done after a hangup and if the stream is not flagged with data - * since there might be nothing to consume in the wait fd but still have - * data available flagged by the consumer wake up pipe. + * This is only done when the stream is monitored by a thread, + * before the flush is done after a hangup and if the stream + * is not flagged with data since there might be nothing to + * consume in the wait fd but still have data available + * flagged by the consumer wake up pipe. */ if (stream->monitor && !stream->hangup_flush_done && !stream->has_data) { char dummy; @@ -1956,118 +2744,310 @@ int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream *stream, readlen = lttng_read(stream->wait_fd, &dummy, 1); if (readlen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) { ret = readlen; - goto end; } } -retry: - /* Get the next subbuffer */ - err = ustctl_get_next_subbuf(ustream); - if (err != 0) { - /* - * Populate metadata info if the existing info has - * already been read. - */ - if (stream->metadata_flag) { - ret = commit_one_metadata_packet(stream); - if (ret <= 0) { - goto end; - } - ustctl_flush_buffer(stream->ustream, 1); - goto retry; - } + return ret; +} - ret = err; /* ustctl_get_next_subbuf returns negative, caller expect positive. */ - /* - * This is a debug message even for single-threaded consumer, - * because poll() have more relaxed criterions than get subbuf, - * so get_subbuf may fail for short race windows where poll() - * would issue wakeups. - */ - DBG("Reserving sub buffer failed (everything is normal, " - "it is due to concurrency) [ret: %d]", err); +static int extract_common_subbuffer_info(struct lttng_consumer_stream *stream, + struct stream_subbuffer *subbuf) +{ + int ret; + + ret = ustctl_get_subbuf_size( + stream->ustream, &subbuf->info.data.subbuf_size); + if (ret) { goto end; } - assert(stream->chan->output == CONSUMER_CHANNEL_MMAP); - if (!stream->metadata_flag) { - index.offset = htobe64(stream->out_fd_offset); - ret = get_index_values(&index, ustream); - if (ret < 0) { + ret = ustctl_get_padded_subbuf_size( + stream->ustream, &subbuf->info.data.padded_subbuf_size); + if (ret) { + goto end; + } + +end: + return ret; +} + +static int extract_metadata_subbuffer_info(struct lttng_consumer_stream *stream, + struct stream_subbuffer *subbuf) +{ + int ret; + + ret = extract_common_subbuffer_info(stream, subbuf); + if (ret) { + goto end; + } + + subbuf->info.metadata.version = stream->metadata_version; + +end: + return ret; +} + +static int extract_data_subbuffer_info(struct lttng_consumer_stream *stream, + struct stream_subbuffer *subbuf) +{ + int ret; + + ret = extract_common_subbuffer_info(stream, subbuf); + if (ret) { + goto end; + } + + ret = ustctl_get_packet_size( + stream->ustream, &subbuf->info.data.packet_size); + if (ret < 0) { + PERROR("Failed to get sub-buffer packet size"); + goto end; + } + + ret = ustctl_get_content_size( + stream->ustream, &subbuf->info.data.content_size); + if (ret < 0) { + PERROR("Failed to get sub-buffer content size"); + goto end; + } + + ret = ustctl_get_timestamp_begin( + stream->ustream, &subbuf->info.data.timestamp_begin); + if (ret < 0) { + PERROR("Failed to get sub-buffer begin timestamp"); + goto end; + } + + ret = ustctl_get_timestamp_end( + stream->ustream, &subbuf->info.data.timestamp_end); + if (ret < 0) { + PERROR("Failed to get sub-buffer end timestamp"); + goto end; + } + + ret = ustctl_get_events_discarded( + stream->ustream, &subbuf->info.data.events_discarded); + if (ret) { + PERROR("Failed to get sub-buffer events discarded count"); + goto end; + } + + ret = ustctl_get_sequence_number(stream->ustream, + &subbuf->info.data.sequence_number.value); + if (ret) { + /* May not be supported by older LTTng-modules. */ + if (ret != -ENOTTY) { + PERROR("Failed to get sub-buffer sequence number"); goto end; } } else { - write_index = 0; + subbuf->info.data.sequence_number.is_set = true; } - /* Get the full padded subbuffer size */ - err = ustctl_get_padded_subbuf_size(ustream, &len); - assert(err == 0); + ret = ustctl_get_stream_id( + stream->ustream, &subbuf->info.data.stream_id); + if (ret < 0) { + PERROR("Failed to get stream id"); + goto end; + } - /* Get subbuffer data size (without padding) */ - err = ustctl_get_subbuf_size(ustream, &subbuf_size); - assert(err == 0); + ret = ustctl_get_instance_id(stream->ustream, + &subbuf->info.data.stream_instance_id.value); + if (ret) { + /* May not be supported by older LTTng-modules. */ + if (ret != -ENOTTY) { + PERROR("Failed to get stream instance id"); + goto end; + } + } else { + subbuf->info.data.stream_instance_id.is_set = true; + } +end: + return ret; +} - /* Make sure we don't get a subbuffer size bigger than the padded */ - assert(len >= subbuf_size); +static int get_next_subbuffer_common(struct lttng_consumer_stream *stream, + struct stream_subbuffer *subbuffer) +{ + int ret; + const char *addr; - padding = len - subbuf_size; - /* write the subbuffer to the tracefile */ - ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size, padding, &index); - /* - * The mmap operation should write subbuf_size amount of data when network - * streaming or the full padding (len) size when we are _not_ streaming. - */ - if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) || - (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) { - /* - * Display the error but continue processing to try to release the - * subbuffer. This is a DBG statement since any unexpected kill or - * signal, the application gets unregistered, relayd gets closed or - * anything that affects the buffer lifetime will trigger this error. - * So, for the sake of the user, don't print this error since it can - * happen and it is OK with the code flow. - */ - DBG("Error writing to tracefile " - "(ret: %ld != len: %lu != subbuf_size: %lu)", - ret, len, subbuf_size); - write_index = 0; + ret = stream->read_subbuffer_ops.extract_subbuffer_info( + stream, subbuffer); + if (ret) { + goto end; } - err = ustctl_put_next_subbuf(ustream); - assert(err == 0); - /* - * This will consumer the byte on the wait_fd if and only if there is not - * next subbuffer to be acquired. - */ - if (!stream->metadata_flag) { - ret = notify_if_more_data(stream, ctx); - if (ret < 0) { - goto end; - } + ret = get_current_subbuf_addr(stream, &addr); + if (ret) { + goto end; } - /* Write index if needed. */ - if (!write_index) { + subbuffer->buffer.buffer = lttng_buffer_view_init( + addr, 0, subbuffer->info.data.padded_subbuf_size); + assert(subbuffer->buffer.buffer.data != NULL); +end: + return ret; +} + +static int get_next_subbuffer(struct lttng_consumer_stream *stream, + struct stream_subbuffer *subbuffer) +{ + int ret; + + ret = ustctl_get_next_subbuf(stream->ustream); + if (ret) { goto end; } - if (stream->chan->live_timer_interval && !stream->metadata_flag) { + ret = get_next_subbuffer_common(stream, subbuffer); + if (ret) { + goto end; + } +end: + return ret; +} + +static int get_next_subbuffer_metadata(struct lttng_consumer_stream *stream, + struct stream_subbuffer *subbuffer) +{ + int ret; + bool cache_empty; + bool got_subbuffer; + bool coherent; + bool buffer_empty; + unsigned long consumed_pos, produced_pos; + + do { + ret = ustctl_get_next_subbuf(stream->ustream); + if (ret == 0) { + got_subbuffer = true; + } else { + got_subbuffer = false; + if (ret != -EAGAIN) { + /* Fatal error. */ + goto end; + } + } + /* - * In live, block until all the metadata is sent. + * Determine if the cache is empty and ensure that a sub-buffer + * is made available if the cache is not empty. */ - err = consumer_stream_sync_metadata(ctx, stream->session_id); - if (err < 0) { - goto end; + if (!got_subbuffer) { + ret = commit_one_metadata_packet(stream); + if (ret < 0 && ret != -ENOBUFS) { + goto end; + } else if (ret == 0) { + /* Not an error, the cache is empty. */ + cache_empty = true; + ret = -ENODATA; + goto end; + } else { + cache_empty = false; + } + } else { + pthread_mutex_lock(&stream->chan->metadata_cache->lock); + cache_empty = stream->chan->metadata_cache->contents.size == + stream->ust_metadata_pushed; + pthread_mutex_unlock(&stream->chan->metadata_cache->lock); } + } while (!got_subbuffer); + + /* Populate sub-buffer infos and view. */ + ret = get_next_subbuffer_common(stream, subbuffer); + if (ret) { + goto end; } - assert(!stream->metadata_flag); - err = consumer_stream_write_index(stream, &index); - if (err < 0) { + ret = lttng_ustconsumer_sample_snapshot_positions(stream); + if (ret < 0) { + /* + * -EAGAIN is not expected since we got a sub-buffer and haven't + * pushed the consumption position yet (on put_next). + */ + PERROR("Failed to take a snapshot of metadata buffer positions"); + goto end; + } + + ret = lttng_ustconsumer_get_consumed_snapshot(stream, &consumed_pos); + if (ret) { + PERROR("Failed to get metadata consumed position"); goto end; } + ret = lttng_ustconsumer_get_produced_snapshot(stream, &produced_pos); + if (ret) { + PERROR("Failed to get metadata produced position"); + goto end; + } + + /* Last sub-buffer of the ring buffer ? */ + buffer_empty = (consumed_pos + stream->max_sb_size) == produced_pos; + + /* + * The sessiond registry lock ensures that coherent units of metadata + * are pushed to the consumer daemon at once. Hence, if a sub-buffer is + * acquired, the cache is empty, and it is the only available sub-buffer + * available, it is safe to assume that it is "coherent". + */ + coherent = got_subbuffer && cache_empty && buffer_empty; + + LTTNG_OPTIONAL_SET(&subbuffer->info.metadata.coherent, coherent); +end: + return ret; +} + +static int put_next_subbuffer(struct lttng_consumer_stream *stream, + struct stream_subbuffer *subbuffer) +{ + const int ret = ustctl_put_next_subbuf(stream->ustream); + + assert(ret == 0); + return ret; +} + +static int signal_metadata(struct lttng_consumer_stream *stream, + struct lttng_consumer_local_data *ctx) +{ + ASSERT_LOCKED(stream->metadata_rdv_lock); + return pthread_cond_broadcast(&stream->metadata_rdv) ? -errno : 0; +} + +static int lttng_ustconsumer_set_stream_ops( + struct lttng_consumer_stream *stream) +{ + int ret = 0; + + stream->read_subbuffer_ops.on_wake_up = consumer_stream_ust_on_wake_up; + if (stream->metadata_flag) { + stream->read_subbuffer_ops.get_next_subbuffer = + get_next_subbuffer_metadata; + stream->read_subbuffer_ops.extract_subbuffer_info = + extract_metadata_subbuffer_info; + stream->read_subbuffer_ops.reset_metadata = + metadata_stream_reset_cache_consumed_position; + if (stream->chan->is_live) { + stream->read_subbuffer_ops.on_sleep = signal_metadata; + ret = consumer_stream_enable_metadata_bucketization( + stream); + if (ret) { + goto end; + } + } + } else { + stream->read_subbuffer_ops.get_next_subbuffer = + get_next_subbuffer; + stream->read_subbuffer_ops.extract_subbuffer_info = + extract_data_subbuffer_info; + stream->read_subbuffer_ops.on_sleep = notify_if_more_data; + if (stream->chan->is_live) { + stream->read_subbuffer_ops.send_live_beacon = + consumer_flush_ust_index; + } + } + + stream->read_subbuffer_ops.put_next_subbuffer = put_next_subbuffer; end: return ret; } @@ -2083,28 +3063,19 @@ int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream *stream) assert(stream); - /* Don't create anything if this is set for streaming. */ - if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor) { - ret = utils_create_stream_file(stream->chan->pathname, stream->name, - stream->chan->tracefile_size, stream->tracefile_count_current, - stream->uid, stream->gid, NULL); - if (ret < 0) { + /* + * Don't create anything if this is set for streaming or if there is + * no current trace chunk on the parent channel. + */ + if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor && + stream->chan->trace_chunk) { + ret = consumer_stream_create_output_files(stream, true); + if (ret) { goto error; } - stream->out_fd = ret; - stream->tracefile_size_current = 0; - - if (!stream->metadata_flag) { - ret = index_create_file(stream->chan->pathname, - stream->name, stream->uid, stream->gid, - stream->chan->tracefile_size, - stream->tracefile_count_current); - if (ret < 0) { - goto error; - } - stream->index_fd = ret; - } } + + lttng_ustconsumer_set_stream_ops(stream); ret = 0; error: @@ -2125,6 +3096,7 @@ int lttng_ustconsumer_data_pending(struct lttng_consumer_stream *stream) assert(stream); assert(stream->ustream); + ASSERT_LOCKED(stream->lock); DBG("UST consumer checking data pending"); @@ -2137,7 +3109,9 @@ int lttng_ustconsumer_data_pending(struct lttng_consumer_stream *stream) uint64_t contiguous, pushed; /* Ease our life a bit. */ - contiguous = stream->chan->metadata_cache->contiguous; + pthread_mutex_lock(&stream->chan->metadata_cache->lock); + contiguous = stream->chan->metadata_cache->contents.size; + pthread_mutex_unlock(&stream->chan->metadata_cache->lock); pushed = stream->ust_metadata_pushed; /* @@ -2183,7 +3157,7 @@ end: * Stop a given metadata channel timer if enabled and close the wait fd which * is the poll pipe of the metadata stream. * - * This MUST be called with the metadata channel acquired. + * This MUST be called with the metadata channel lock acquired. */ void lttng_ustconsumer_close_metadata(struct lttng_consumer_channel *metadata) { @@ -2266,6 +3240,10 @@ void lttng_ustconsumer_close_stream_wakeup(struct lttng_consumer_stream *stream) * function or any of its callees. Timers have a very strict locking * semantic with respect to teardown. Failure to respect this semantic * introduces deadlocks. + * + * DON'T hold the metadata lock when calling this function, else this + * can cause deadlock involving consumer awaiting for metadata to be + * pushed out due to concurrent interaction with the session daemon. */ int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx, struct lttng_consumer_channel *channel, int timer, int wait) @@ -2273,7 +3251,7 @@ int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx, struct lttcomm_metadata_request_msg request; struct lttcomm_consumer_msg msg; enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS; - uint64_t len, key, offset; + uint64_t len, key, offset, version; int ret; assert(channel); @@ -2305,7 +3283,7 @@ int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx, request.key = channel->key; DBG("Sending metadata request to sessiond, session id %" PRIu64 - ", per-pid %" PRIu64 ", app UID %u and channek key %" PRIu64, + ", per-pid %" PRIu64 ", app UID %u and channel key %" PRIu64, request.session_id, request.session_id_per_pid, request.uid, request.key); @@ -2353,6 +3331,7 @@ int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx, len = msg.u.push_metadata.len; key = msg.u.push_metadata.key; offset = msg.u.push_metadata.target_offset; + version = msg.u.push_metadata.version; assert(key == channel->key); if (len == 0) { @@ -2375,7 +3354,7 @@ int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx, health_code_update(); ret = lttng_ustconsumer_recv_metadata(ctx->consumer_metadata_socket, - key, offset, len, channel, timer, wait); + key, offset, len, version, channel, timer, wait); if (ret >= 0) { /* * Only send the status msg if the sessiond is alive meaning a positive