X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fcommon%2Fust-consumer%2Fust-consumer.c;h=e5143cd4a5f38afbc8e77189135c68fff25e4e96;hp=82d56ebb796910b28b86f417f7ec0f64a08de42b;hb=a2814ea7573bf5edd5323d6f89c48ff14105db69;hpb=f8f3885cc52af9d3c951da78989d6f4a25270411 diff --git a/src/common/ust-consumer/ust-consumer.c b/src/common/ust-consumer/ust-consumer.c index 82d56ebb7..e5143cd4a 100644 --- a/src/common/ust-consumer/ust-consumer.c +++ b/src/common/ust-consumer/ust-consumer.c @@ -1,21 +1,13 @@ /* - * 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. */ +#include #define _LGPL_SOURCE #include #include @@ -31,6 +23,7 @@ #include #include #include +#include #include #include @@ -50,7 +43,6 @@ 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 @@ -72,6 +64,7 @@ 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); } @@ -118,26 +111,6 @@ 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 *root_shm_path, 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, root_shm_path, 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. @@ -154,14 +127,14 @@ static struct lttng_consumer_stream *allocate_stream(int cpu, int key, assert(channel); assert(ctx); - stream = consumer_allocate_stream(channel->key, + stream = consumer_allocate_stream( + 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, @@ -184,7 +157,7 @@ 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) { @@ -206,28 +179,21 @@ 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) { @@ -239,7 +205,9 @@ static int send_stream_to_thread(struct lttng_consumer_stream *stream, } else { consumer_del_stream_for_data(stream); } + goto error; } + error: return ret; } @@ -266,6 +234,7 @@ end: /* * 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. */ @@ -275,6 +244,7 @@ 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); @@ -321,6 +291,8 @@ 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. @@ -359,12 +331,17 @@ static int create_ust_streams(struct lttng_consumer_channel *channel, 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; } @@ -410,9 +387,8 @@ error_shm_open: return -1; } -static int open_ust_stream_fd(struct lttng_consumer_channel *channel, - struct ustctl_consumer_channel_attr *attr, - int cpu) +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; @@ -426,7 +402,7 @@ static int open_ust_stream_fd(struct lttng_consumer_channel *channel, } return run_as_open(shm_path, O_RDWR | O_CREAT | O_EXCL, S_IRUSR | S_IWUSR, - channel->uid, channel->gid); + session_credentials->uid, session_credentials->gid); error_shm_path: return -1; @@ -449,6 +425,7 @@ static int create_ust_channel(struct lttng_consumer_channel *channel, assert(channel); assert(attr); assert(ust_chanp); + assert(channel->buffer_credentials.is_set); DBG3("Creating channel to ustctl with attr: [overwrite: %d, " "subbuf_size: %" PRIu64 ", num_subbuf: %" PRIu64 ", " @@ -467,7 +444,8 @@ static int create_ust_channel(struct lttng_consumer_channel *channel, goto error_alloc; } for (i = 0; i < nr_stream_fds; i++) { - stream_fds[i] = open_ust_stream_fd(channel, attr, i); + stream_fds[i] = open_ust_stream_fd(channel, i, + &channel->buffer_credentials.value); if (stream_fds[i] < 0) { ret = -1; goto error_open; @@ -502,7 +480,8 @@ error_open: ERR("Cannot get stream shm path"); } closeret = run_as_unlink(shm_path, - channel->uid, channel->gid); + channel->buffer_credentials.value.uid, + channel->buffer_credentials.value.gid); if (closeret) { PERROR("unlink %s", shm_path); } @@ -511,7 +490,9 @@ error_open: /* Try to rmdir all directories under shm_path root. */ if (channel->root_shm_path[0]) { (void) run_as_rmdir_recursive(channel->root_shm_path, - channel->uid, channel->gid); + channel->buffer_credentials.value.uid, + channel->buffer_credentials.value.gid, + LTTNG_DIRECTORY_HANDLE_SKIP_NON_EMPTY_FLAG); } free(stream_fds); error_alloc: @@ -543,11 +524,11 @@ 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) { @@ -567,6 +548,8 @@ 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) { /* @@ -642,7 +625,7 @@ 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) { @@ -685,7 +668,9 @@ 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; } @@ -721,14 +706,8 @@ 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: @@ -768,10 +747,19 @@ static int flush_channel(uint64_t chan_key) health_code_update(); 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: @@ -823,7 +811,6 @@ 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. */ @@ -831,6 +818,7 @@ 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); @@ -849,13 +837,48 @@ 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); @@ -918,6 +941,10 @@ 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) { /* @@ -925,7 +952,7 @@ 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)); @@ -943,6 +970,7 @@ 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; @@ -950,14 +978,15 @@ 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); @@ -968,13 +997,6 @@ 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(); @@ -1002,22 +1024,17 @@ 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 { @@ -1043,18 +1060,49 @@ 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); @@ -1066,23 +1114,27 @@ 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) { - /* Are we at a position _before_ the first available packet ? */ - bool before_first_packet = true; - 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) { @@ -1091,24 +1143,13 @@ 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); } /* @@ -1147,10 +1188,11 @@ 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; - int lost_packet = 0; + const char *subbuf_addr; + struct lttng_buffer_view subbuf_view; health_code_update(); @@ -1164,15 +1206,7 @@ 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; - - /* - * Start accounting lost packets only when we - * already have extracted packets (to match the - * content of the final snapshot). - */ - if (!before_first_packet) { - lost_packet = 1; - } + stream->chan->lost_packets++; continue; } @@ -1188,8 +1222,16 @@ 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(ctx, + stream, &subbuf_view, padded_len - len, + NULL); if (use_relayd) { if (read_len != len) { ret = -EPERM; @@ -1208,16 +1250,6 @@ static int snapshot_channel(uint64_t key, char *path, uint64_t relayd_id, goto error_close_stream; } consumed_pos += stream->max_sb_size; - - /* - * Only account lost packets located between - * succesfully extracted packets (do not account before - * and after since they are not visible in the - * resulting snapshot). - */ - stream->chan->lost_packets += lost_packet; - lost_packet = 0; - before_first_packet = false; } /* Simply close the stream so we can use it on the next snapshot. */ @@ -1236,7 +1268,6 @@ error_close_stream: consumer_stream_close(stream); error_unlock: pthread_mutex_unlock(&stream->lock); -error: rcu_read_unlock(); return ret; } @@ -1354,7 +1385,7 @@ 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); @@ -1422,24 +1453,37 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, { int ret; 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 = msg.u.ask_channel.buffer_credentials.uid, + .gid = 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 @@ -1455,6 +1499,7 @@ 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)); + attr.blocking_timeout= msg.u.ask_channel.blocking_timeout; /* Match channel buffer type to the UST abi. */ switch (msg.u.ask_channel.output) { @@ -1487,7 +1532,7 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, health_code_update(); - ret = ask_channel(ctx, sock, channel, &attr); + ret = ask_channel(ctx, channel, &attr); if (ret < 0) { goto end_channel_error; } @@ -1501,8 +1546,17 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, 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(); @@ -1525,6 +1579,9 @@ 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; } @@ -1555,13 +1612,14 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, if (!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, channel, ctx, + &relayd_err); if (ret < 0) { if (relayd_err) { /* @@ -1570,13 +1628,13 @@ 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(); @@ -1586,7 +1644,7 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, * so don't send them to the data thread. */ if (!channel->monitor) { - goto end_msg_sessiond; + goto end_get_channel; } ret = send_streams_to_thread(channel, ctx); @@ -1595,11 +1653,16 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, * 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)); +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: { @@ -1670,7 +1733,7 @@ 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(); @@ -1681,14 +1744,14 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, * checked whether the channel can be found. */ ret_code = LTTCOMM_CONSUMERD_SUCCESS; - goto end_msg_sessiond; + 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(); @@ -1698,7 +1761,7 @@ 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(); @@ -1707,11 +1770,15 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, len, version, 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: { @@ -1725,27 +1792,35 @@ 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 *channel; + uint64_t key = msg.u.snapshot_channel.key; + + channel = consumer_find_channel(key); + if (!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) { + ret = snapshot_metadata(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_SNAPSHOT_FAILED; + } + } else { + ret = snapshot_channel(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_SNAPSHOT_FAILED; + } } } - health_code_update(); ret = consumer_send_status_msg(sock, ret_code); if (ret < 0) { @@ -1857,20 +1932,260 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, break; } - default: + case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE: + { + int channel_monitor_pipe; + + ret_code = LTTCOMM_CONSUMERD_SUCCESS; + /* Successfully received the command's type. */ + ret = consumer_send_status_msg(sock, ret_code); + if (ret < 0) { + goto error_fatal; + } + + ret = lttcomm_recv_fds_unix_sock(sock, &channel_monitor_pipe, + 1); + if (ret != sizeof(channel_monitor_pipe)) { + ERR("Failed to receive channel monitor pipe"); + goto error_fatal; + } + + DBG("Received channel monitor pipe (%d)", channel_monitor_pipe); + ret = consumer_timer_thread_set_channel_monitor_pipe( + channel_monitor_pipe); + if (!ret) { + int flags; + + ret_code = LTTCOMM_CONSUMERD_SUCCESS; + /* Set the pipe as non-blocking. */ + ret = fcntl(channel_monitor_pipe, F_GETFL, 0); + if (ret == -1) { + PERROR("fcntl get flags of the channel monitoring pipe"); + goto error_fatal; + } + flags = ret; + + ret = fcntl(channel_monitor_pipe, F_SETFL, + flags | O_NONBLOCK); + if (ret == -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 *channel; + uint64_t key = msg.u.rotate_channel.key; + + channel = consumer_find_channel(key); + if (!channel) { + DBG("Channel %" PRIu64 " not found", key); + ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND; + } else { + /* + * Sample the rotate position of all the streams in + * this channel. + */ + ret = lttng_consumer_rotate_channel(channel, key, + msg.u.rotate_channel.relayd_id, + msg.u.rotate_channel.metadata, + ctx); + if (ret < 0) { + ERR("Rotate channel failed"); + ret_code = LTTCOMM_CONSUMERD_ROTATION_FAIL; + } + + health_code_update(); + } + ret = consumer_send_status_msg(sock, ret_code); + if (ret < 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 (channel) { + ret = lttng_consumer_rotate_ready_streams( + channel, key, ctx); + if (ret < 0) { + ERR("Rotate channel failed"); + } + } break; +end_rotate_channel_nosignal: + goto end_nosignal; } + case LTTNG_CONSUMER_CLEAR_CHANNEL: + { + struct lttng_consumer_channel *channel; + uint64_t key = msg.u.clear_channel.key; -end_nosignal: - rcu_read_unlock(); + channel = consumer_find_channel(key); + if (!channel) { + DBG("Channel %" PRIu64 " not found", key); + ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND; + } else { + ret = lttng_consumer_clear_channel(channel); + if (ret) { + ERR("Clear channel failed key %" PRIu64, key); + ret_code = ret; + } - health_code_update(); + health_code_update(); + } + ret = consumer_send_status_msg(sock, ret_code); + if (ret < 0) { + /* Somehow, the session daemon is not responding anymore. */ + goto end_nosignal; + } + break; + } + case LTTNG_CONSUMER_INIT: + { + ret_code = lttng_consumer_init_command(ctx, + msg.u.init.sessiond_uuid); + health_code_update(); + ret = consumer_send_status_msg(sock, ret_code); + if (ret < 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 = msg.u.create_trace_chunk.credentials.value.uid, + .gid = 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; + + /* Acnowledge the reception of the command. */ + ret = consumer_send_status_msg(sock, + LTTCOMM_CONSUMERD_SUCCESS); + if (ret < 0) { + /* Somehow, the session daemon is not responding anymore. */ + goto end_nosignal; + } + + /* + * Receive trace chunk domain dirfd. + */ + ret = lttcomm_recv_fds_unix_sock(sock, &chunk_dirfd, 1); + if (ret != 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; + } + default: + break; + } + +end_nosignal: /* * Return 1 to indicate success since the 0 value can be a socket * shutdown during the recv() or send() call. */ - return 1; + ret = 1; + goto end; end_msg_sessiond: /* @@ -1882,11 +2197,9 @@ end_msg_sessiond: if (ret < 0) { goto error_fatal; } - rcu_read_unlock(); - - health_code_update(); + ret = 1; + goto end; - return 1; end_channel_error: if (channel) { /* @@ -1901,53 +2214,54 @@ end_channel_error: /* Stop everything if session daemon can not be notified. */ goto error_fatal; } - rcu_read_unlock(); - - health_code_update(); + ret = 1; + goto end; - return 1; error_fatal: - rcu_read_unlock(); /* This will issue a consumer stop. */ - return -1; + ret = -1; + goto end; + +end: + rcu_read_unlock(); + health_code_update(); + return ret; } -/* - * 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); } /* @@ -1989,6 +2303,14 @@ 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) { @@ -2032,6 +2354,7 @@ void lttng_ustconsumer_del_channel(struct lttng_consumer_channel *chan) assert(chan); assert(chan->uchan); + assert(chan->buffer_credentials.is_set); if (chan->switch_timer_enabled == 1) { consumer_timer_switch_stop(chan); @@ -2050,7 +2373,9 @@ void lttng_ustconsumer_del_channel(struct lttng_consumer_channel *chan) if (ret) { ERR("Cannot get stream shm path"); } - ret = run_as_unlink(shm_path, chan->uid, chan->gid); + ret = run_as_unlink(shm_path, + chan->buffer_credentials.value.uid, + chan->buffer_credentials.value.gid); if (ret) { PERROR("unlink %s", shm_path); } @@ -2062,13 +2387,16 @@ 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, - chan->uid, chan->gid); + chan->buffer_credentials.value.uid, + chan->buffer_credentials.value.gid, + LTTNG_DIRECTORY_HANDLE_SKIP_NON_EMPTY_FLAG); } free(chan->stream_fds); } @@ -2109,62 +2437,69 @@ static int get_index_values(struct ctf_packet_index *index, struct ustctl_consumer_stream *ustream) { int ret; + uint64_t packet_size, content_size, timestamp_begin, timestamp_end, + events_discarded, stream_id, stream_instance_id, + packet_seq_num; - ret = ustctl_get_timestamp_begin(ustream, &index->timestamp_begin); + ret = ustctl_get_timestamp_begin(ustream, ×tamp_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); + ret = ustctl_get_timestamp_end(ustream, ×tamp_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); + ret = ustctl_get_events_discarded(ustream, &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); + ret = ustctl_get_content_size(ustream, &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); + ret = ustctl_get_packet_size(ustream, &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); + ret = ustctl_get_stream_id(ustream, &stream_id); if (ret < 0) { PERROR("ustctl_get_stream_id"); goto error; } - index->stream_id = htobe64(index->stream_id); - ret = ustctl_get_instance_id(ustream, &index->stream_instance_id); + ret = ustctl_get_instance_id(ustream, &stream_instance_id); if (ret < 0) { PERROR("ustctl_get_instance_id"); goto error; } - index->stream_instance_id = htobe64(index->stream_instance_id); - ret = ustctl_get_sequence_number(ustream, &index->packet_seq_num); + ret = ustctl_get_sequence_number(ustream, &packet_seq_num); if (ret < 0) { PERROR("ustctl_get_sequence_number"); goto error; } - index->packet_seq_num = htobe64(index->packet_seq_num); + + *index = (typeof(*index)) { + .offset = index->offset, + .packet_size = htobe64(packet_size), + .content_size = htobe64(content_size), + .timestamp_begin = htobe64(timestamp_begin), + .timestamp_end = htobe64(timestamp_end), + .events_discarded = htobe64(events_discarded), + .stream_id = htobe64(stream_id), + .stream_instance_id = htobe64(stream_instance_id), + .packet_seq_num = htobe64(packet_seq_num), + }; error: return ret; @@ -2241,6 +2576,13 @@ int commit_one_metadata_packet(struct lttng_consumer_stream *stream) 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; @@ -2255,38 +2597,59 @@ end: * interacting with sessiond, else we cause a deadlock with live * awaiting on metadata to be pushed out. * + * The RCU read side lock must be held by the caller. + * * Return 0 if new metadatda is available, EAGAIN if the metadata stream * is empty or a negative value on error. */ int lttng_ustconsumer_sync_metadata(struct lttng_consumer_local_data *ctx, - struct lttng_consumer_stream *metadata) + struct lttng_consumer_stream *metadata_stream) { int ret; int retry = 0; + struct lttng_consumer_channel *metadata_channel; assert(ctx); - assert(metadata); + assert(metadata_stream); - pthread_mutex_unlock(&metadata->lock); + 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) { goto end; } - pthread_mutex_lock(&metadata->lock); - ret = commit_one_metadata_packet(metadata); + /* + * 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); + ret = 0; + goto end; + } + + ret = commit_one_metadata_packet(metadata_stream); if (ret <= 0) { goto end; } else if (ret > 0) { retry = 1; } - ustctl_flush_buffer(metadata->ustream, 1); - ret = ustctl_snapshot(metadata->ustream); + ret = ustctl_snapshot(metadata_stream->ustream); if (ret < 0) { if (errno != EAGAIN) { ERR("Sync metadata, taking UST snapshot"); @@ -2418,7 +2781,7 @@ end: /* * Read subbuffer from the given stream. * - * Stream lock MUST be acquired. + * Stream and channel locks MUST be acquired by the caller. * * Return 0 on success else a negative value. */ @@ -2426,10 +2789,12 @@ int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream *stream, struct lttng_consumer_local_data *ctx) { unsigned long len, subbuf_size, padding; - int err, write_index = 1; + int err, write_index = 1, rotation_ret; long ret = 0; struct ustctl_consumer_stream *ustream; struct ctf_packet_index index; + const char *subbuf_addr; + struct lttng_buffer_view subbuf_view; assert(stream); assert(stream->ustream); @@ -2458,7 +2823,21 @@ 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; + goto error; + } + } + + /* + * 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 extracting data"); + rotation_ret = lttng_consumer_rotate_stream(ctx, stream); + if (rotation_ret < 0) { + ERR("Stream rotation error"); + ret = -1; + goto error; } } @@ -2473,9 +2852,8 @@ retry: if (stream->metadata_flag) { ret = commit_one_metadata_packet(stream); if (ret <= 0) { - goto end; + goto error; } - ustctl_flush_buffer(stream->ustream, 1); goto retry; } @@ -2488,7 +2866,7 @@ retry: */ DBG("Reserving sub buffer failed (everything is normal, " "it is due to concurrency) [ret: %d]", err); - goto end; + goto error; } assert(stream->chan->output == CONSUMER_CHANNEL_MMAP); @@ -2496,14 +2874,18 @@ retry: index.offset = htobe64(stream->out_fd_offset); ret = get_index_values(&index, ustream); if (ret < 0) { - goto end; + err = ustctl_put_subbuf(ustream); + assert(err == 0); + goto error; } /* Update the stream's sequence and discarded events count. */ ret = update_stream_stats(stream); if (ret < 0) { PERROR("kernctl_get_events_discarded"); - goto end; + err = ustctl_put_subbuf(ustream); + assert(err == 0); + goto error; } } else { write_index = 0; @@ -2521,11 +2903,22 @@ retry: assert(len >= subbuf_size); padding = len - subbuf_size; + + ret = get_current_subbuf_addr(stream, &subbuf_addr); + if (ret) { + write_index = 0; + goto error_put_subbuf; + } + + subbuf_view = lttng_buffer_view_init(subbuf_addr, 0, len); + /* write the subbuffer to the tracefile */ - ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size, padding, &index); + ret = lttng_consumer_on_read_subbuffer_mmap( + ctx, stream, &subbuf_view, 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. + * 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)) { @@ -2542,6 +2935,7 @@ retry: ret, len, subbuf_size); write_index = 0; } +error_put_subbuf: err = ustctl_put_next_subbuf(ustream); assert(err == 0); @@ -2552,13 +2946,13 @@ retry: if (!stream->metadata_flag) { ret = notify_if_more_data(stream, ctx); if (ret < 0) { - goto end; + goto error; } } /* Write index if needed. */ if (!write_index) { - goto end; + goto rotate; } if (stream->chan->live_timer_interval && !stream->metadata_flag) { @@ -2583,17 +2977,35 @@ retry: } if (err < 0) { - goto end; + goto error; } } assert(!stream->metadata_flag); err = consumer_stream_write_index(stream, &index); if (err < 0) { - goto end; + goto error; } -end: +rotate: + /* + * After extracting the packet, we check if the stream is now ready to be + * rotated and perform the action immediately. + */ + 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) { + ERR("Stream rotation error"); + ret = -1; + goto error; + } + } else if (rotation_ret < 0) { + ERR("Checking if stream is ready to rotate"); + ret = -1; + goto error; + } +error: return ret; } @@ -2608,30 +3020,16 @@ 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) { - struct lttng_index_file *index_file; - - 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 (!index_file) { - goto error; - } - stream->index_file = index_file; - } } ret = 0; @@ -2711,7 +3109,7 @@ 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) { @@ -2837,7 +3235,7 @@ 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);