X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fcommon%2Fust-consumer%2Fust-consumer.c;h=5bcca4e17a3d94622f423d8a30e410e05473d741;hp=431b94626736d042d92db89db0774e7e941cc3e7;hb=b8086166dbf40d53a4a1e3f0cddbf291fac97bd3;hpb=331744e34f56a5aec69b05d356d6901e67926acc diff --git a/src/common/ust-consumer/ust-consumer.c b/src/common/ust-consumer/ust-consumer.c index 431b94626..5bcca4e17 100644 --- a/src/common/ust-consumer/ust-consumer.c +++ b/src/common/ust-consumer/ust-consumer.c @@ -37,7 +37,9 @@ #include #include #include +#include #include +#include #include "ust-consumer.h" @@ -112,13 +114,16 @@ error: */ static struct lttng_consumer_channel *allocate_channel(uint64_t session_id, const char *pathname, const char *name, uid_t uid, gid_t gid, - int relayd_id, uint64_t key, enum lttng_event_output output) + 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) { assert(pathname); assert(name); - return consumer_allocate_channel(key, session_id, pathname, name, uid, gid, - relayd_id, output); + return consumer_allocate_channel(key, session_id, pathname, name, uid, + gid, relayd_id, output, tracefile_size, + tracefile_count, session_id_per_pid, monitor); } /* @@ -147,7 +152,8 @@ static struct lttng_consumer_stream *allocate_stream(int cpu, int key, channel->session_id, cpu, &alloc_ret, - channel->type); + channel->type, + channel->monitor); if (stream == NULL) { switch (alloc_ret) { case -ENOENT: @@ -183,60 +189,26 @@ error: static int send_stream_to_thread(struct lttng_consumer_stream *stream, struct lttng_consumer_local_data *ctx) { - int ret, stream_pipe; + int ret; + struct lttng_pipe *stream_pipe; /* Get the right pipe where the stream will be sent. */ if (stream->metadata_flag) { - stream_pipe = ctx->consumer_metadata_pipe[1]; + stream_pipe = ctx->consumer_metadata_pipe; } else { - stream_pipe = ctx->consumer_data_pipe[1]; + stream_pipe = ctx->consumer_data_pipe; } - do { - ret = write(stream_pipe, &stream, sizeof(stream)); - } while (ret < 0 && errno == EINTR); + ret = lttng_pipe_write(stream_pipe, &stream, sizeof(stream)); if (ret < 0) { - PERROR("Consumer write %s stream to pipe %d", - stream->metadata_flag ? "metadata" : "data", stream_pipe); + ERR("Consumer write %s stream to pipe %d", + stream->metadata_flag ? "metadata" : "data", + lttng_pipe_get_writefd(stream_pipe)); } return ret; } -/* - * Search for a relayd object related to the stream. If found, send the stream - * to the relayd. - * - * On success, returns 0 else a negative value. - */ -static int send_stream_to_relayd(struct lttng_consumer_stream *stream) -{ - int ret = 0; - struct consumer_relayd_sock_pair *relayd; - - assert(stream); - - relayd = consumer_find_relayd(stream->net_seq_idx); - if (relayd != NULL) { - pthread_mutex_lock(&relayd->ctrl_sock_mutex); - /* Add stream on the relayd */ - ret = relayd_add_stream(&relayd->control_sock, stream->name, - stream->chan->pathname, &stream->relayd_stream_id); - pthread_mutex_unlock(&relayd->ctrl_sock_mutex); - if (ret < 0) { - goto error; - } - } else if (stream->net_seq_idx != (uint64_t) -1ULL) { - ERR("Network sequence index %" PRIu64 " unknown. Not adding stream.", - stream->net_seq_idx); - ret = -1; - goto error; - } - -error: - return ret; -} - /* * Create streams for the given channel using liblttng-ust-ctl. * @@ -258,8 +230,18 @@ static int create_ust_streams(struct lttng_consumer_channel *channel, */ while ((ustream = ustctl_create_stream(channel->uchan, cpu))) { int wait_fd; + int ust_metadata_pipe[2]; - wait_fd = ustctl_stream_get_wait_fd(ustream); + if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA && channel->monitor) { + ret = utils_create_pipe_cloexec_nonblock(ust_metadata_pipe); + if (ret < 0) { + ERR("Create ust metadata poll pipe"); + goto error; + } + wait_fd = ust_metadata_pipe[0]; + } else { + wait_fd = ustctl_stream_get_wait_fd(ustream); + } /* Allocate consumer stream object. */ stream = allocate_stream(cpu, wait_fd, channel, ctx, &ret); @@ -274,6 +256,14 @@ static int create_ust_streams(struct lttng_consumer_channel *channel, */ stream->wait_fd = wait_fd; + /* + * Increment channel refcount since the channel reference has now been + * assigned in the allocation process above. + */ + if (stream->chan->monitor) { + uatomic_inc(&stream->chan->refcount); + } + /* * Order is important this is why a list is used. On error, the caller * should clean this list. @@ -305,6 +295,8 @@ 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]; } } @@ -363,7 +355,7 @@ static int send_sessiond_stream(int sock, struct lttng_consumer_stream *stream) assert(stream); assert(sock >= 0); - DBG2("UST consumer sending stream %" PRIu64 " to sessiond", stream->key); + DBG("UST consumer sending stream %" PRIu64 " to sessiond", stream->key); /* Send stream to session daemon. */ ret = ustctl_send_stream_to_sessiond(sock, stream->ustream); @@ -384,7 +376,7 @@ static int send_sessiond_channel(int sock, struct lttng_consumer_channel *channel, struct lttng_consumer_local_data *ctx, int *relayd_error) { - int ret; + int ret, ret_code = LTTNG_OK; struct lttng_consumer_stream *stream; assert(channel); @@ -393,6 +385,33 @@ static int send_sessiond_channel(int sock, DBG("UST consumer sending channel %s to sessiond", channel->name); + if (channel->relayd_id != (uint64_t) -1ULL) { + cds_list_for_each_entry(stream, &channel->streams.head, send_node) { + /* Try to send the stream to the relayd if one is available. */ + ret = consumer_send_relayd_stream(stream, stream->chan->pathname); + if (ret < 0) { + /* + * Flag that the relayd was the problem here probably due to a + * communicaton error on the socket. + */ + if (relayd_error) { + *relayd_error = 1; + } + ret_code = LTTNG_ERR_RELAYD_CONNECT_FAIL; + } + } + } + + /* Inform sessiond that we are about to send channel and streams. */ + ret = consumer_send_status_msg(sock, ret_code); + if (ret < 0 || ret_code != LTTNG_OK) { + /* + * Either the session daemon is not responding or the relayd died so we + * stop now. + */ + goto error; + } + /* Send channel to sessiond. */ ret = ustctl_send_channel_to_sessiond(sock, channel->uchan); if (ret < 0) { @@ -406,19 +425,6 @@ static int send_sessiond_channel(int sock, /* The channel was sent successfully to the sessiond at this point. */ cds_list_for_each_entry(stream, &channel->streams.head, send_node) { - /* Try to send the stream to the relayd if one is available. */ - ret = send_stream_to_relayd(stream); - if (ret < 0) { - /* - * Flag that the relayd was the problem here probably due to a - * communicaton error on the socket. - */ - if (relayd_error) { - *relayd_error = 1; - } - goto error; - } - /* Send stream to session daemon. */ ret = send_sessiond_stream(sock, stream); if (ret < 0) { @@ -437,6 +443,9 @@ static int send_sessiond_channel(int sock, return 0; error: + if (ret_code != LTTNG_OK) { + ret = -1; + } return ret; } @@ -476,22 +485,27 @@ static int ask_channel(struct lttng_consumer_local_data *ctx, int sock, /* The reply msg status is handled in the following call. */ ret = create_ust_channel(attr, &channel->uchan); if (ret < 0) { - goto error; + goto end; } channel->wait_fd = ustctl_channel_get_wait_fd(channel->uchan); - if (ret < 0) { - goto error; + /* + * For the snapshots (no monitor), we create the metadata streams + * on demand, not during the channel creation. + */ + if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA && !channel->monitor) { + ret = 0; + goto end; } /* Open all streams for this channel. */ ret = create_ust_streams(channel, ctx); if (ret < 0) { - goto error; + goto end; } -error: +end: return ret; } @@ -524,40 +538,14 @@ static int send_streams_to_thread(struct lttng_consumer_channel *channel, /* Remove node from the channel stream list. */ cds_list_del(&stream->send_node); - } - -error: - return ret; -} -/* - * Write metadata to the given channel using ustctl to convert the string to - * the ringbuffer. - * Called only from consumer_metadata_cache_write. - * The metadata cache lock MUST be acquired to write in the cache. - * - * Return 0 on success else a negative value. - */ -int lttng_ustconsumer_push_metadata(struct lttng_consumer_channel *metadata, - const char *metadata_str, uint64_t target_offset, uint64_t len) -{ - int ret; - - assert(metadata); - assert(metadata_str); - - DBG("UST consumer writing metadata to channel %s", metadata->name); - - assert(target_offset <= metadata->metadata_cache->max_offset); - ret = ustctl_write_metadata_to_channel(metadata->uchan, - metadata_str + target_offset, len); - if (ret < 0) { - ERR("ustctl write metadata fail with ret %d, len %ld", ret, len); - goto error; + /* + * From this point on, the stream's ownership has been moved away from + * the channel and becomes globally visible. + */ + stream->globally_visible = 1; } - ustctl_flush_buffer(metadata->metadata_stream->ustream, 1); - error: return ret; } @@ -575,11 +563,12 @@ static int flush_channel(uint64_t chan_key) struct lttng_ht *ht; struct lttng_ht_iter iter; - DBG("UST consumer flush channel key %lu", chan_key); + DBG("UST consumer flush channel key %" PRIu64, chan_key); + rcu_read_lock(); channel = consumer_find_channel(chan_key); if (!channel) { - ERR("UST consumer flush channel %lu not found", chan_key); + ERR("UST consumer flush channel %" PRIu64 " not found", chan_key); ret = LTTNG_ERR_UST_CHAN_NOT_FOUND; goto error; } @@ -587,49 +576,73 @@ static int flush_channel(uint64_t chan_key) ht = consumer_data.stream_per_chan_id_ht; /* For each stream of the channel id, flush it. */ - rcu_read_lock(); cds_lfht_for_each_entry_duplicate(ht->ht, ht->hash_fct(&channel->key, lttng_ht_seed), ht->match_fct, &channel->key, &iter.iter, stream, node_channel_id.node) { - ustctl_flush_buffer(stream->ustream, 1); + ustctl_flush_buffer(stream->ustream, 1); } - rcu_read_unlock(); - error: + rcu_read_unlock(); return ret; } /* * 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. */ static int close_metadata(uint64_t chan_key) { - int ret; + int ret = 0; struct lttng_consumer_channel *channel; - DBG("UST consumer close metadata key %lu", chan_key); + DBG("UST consumer close metadata key %" PRIu64, chan_key); channel = consumer_find_channel(chan_key); if (!channel) { - ERR("UST consumer close metadata %lu not found", chan_key); + /* + * This is possible if the metadata thread has issue a delete because + * the endpoint point of the stream hung up. There is no way the + * session daemon can know about it thus use a DBG instead of an actual + * error. + */ + DBG("UST consumer close metadata %" PRIu64 " not found", chan_key); ret = LTTNG_ERR_UST_CHAN_NOT_FOUND; goto error; } - ret = ustctl_stream_close_wakeup_fd(channel->metadata_stream->ustream); - if (ret < 0) { - ERR("UST consumer unable to close fd of metadata (ret: %d)", ret); - ret = LTTCOMM_CONSUMERD_ERROR_METADATA; - goto error; + pthread_mutex_lock(&consumer_data.lock); + pthread_mutex_lock(&channel->lock); + + if (cds_lfht_is_node_deleted(&channel->node.node)) { + goto error_unlock; } + if (channel->switch_timer_enabled == 1) { DBG("Deleting timer on metadata channel"); consumer_timer_switch_stop(channel); } - consumer_metadata_cache_destroy(channel); + if (channel->metadata_stream) { + ret = ustctl_stream_close_wakeup_fd(channel->metadata_stream->ustream); + if (ret < 0) { + ERR("UST consumer unable to close fd of metadata (ret: %d)", ret); + ret = LTTCOMM_CONSUMERD_ERROR_METADATA; + goto error_unlock; + } + if (channel->monitor) { + /* close the read-side in consumer_del_metadata_stream */ + ret = close(channel->metadata_stream->ust_metadata_poll_pipe[1]); + if (ret < 0) { + PERROR("Close UST metadata write-side poll pipe"); + } + } + } + +error_unlock: + pthread_mutex_unlock(&channel->lock); + pthread_mutex_unlock(&consumer_data.lock); error: return ret; } @@ -644,13 +657,23 @@ static int setup_metadata(struct lttng_consumer_local_data *ctx, uint64_t key) int ret; struct lttng_consumer_channel *metadata; - DBG("UST consumer setup metadata key %lu", key); + DBG("UST consumer setup metadata key %" PRIu64, key); metadata = consumer_find_channel(key); if (!metadata) { ERR("UST consumer push metadata %" PRIu64 " not found", key); ret = LTTNG_ERR_UST_CHAN_NOT_FOUND; - goto error; + goto end; + } + + /* + * In no monitor mode, the metadata channel has no stream(s) so skip the + * ownership transfer to the metadata thread. + */ + if (!metadata->monitor) { + DBG("Metadata channel in no monitor"); + ret = 0; + goto end; } /* @@ -660,14 +683,17 @@ static int setup_metadata(struct lttng_consumer_local_data *ctx, uint64_t key) if (cds_list_empty(&metadata->streams.head)) { ERR("Metadata channel key %" PRIu64 ", no stream available.", key); ret = LTTCOMM_CONSUMERD_ERROR_METADATA; - goto error; + goto error_no_stream; } /* Send metadata stream to relayd if needed. */ - ret = send_stream_to_relayd(metadata->metadata_stream); - if (ret < 0) { - ret = LTTCOMM_CONSUMERD_ERROR_METADATA; - goto error; + if (metadata->metadata_stream->net_seq_idx != (uint64_t) -1ULL) { + ret = consumer_send_relayd_stream(metadata->metadata_stream, + metadata->pathname); + if (ret < 0) { + ret = LTTCOMM_CONSUMERD_ERROR_METADATA; + goto error; + } } ret = send_streams_to_thread(metadata, ctx); @@ -683,8 +709,270 @@ static int setup_metadata(struct lttng_consumer_local_data *ctx, uint64_t key) assert(cds_list_empty(&metadata->streams.head)); ret = 0; + goto end; + +error: + /* + * Delete metadata channel on error. At this point, the metadata stream can + * NOT be monitored by the metadata thread thus having the guarantee that + * the stream is still in the local stream list of the channel. This call + * will make sure to clean that list. + */ + cds_list_del(&metadata->metadata_stream->send_node); + consumer_stream_destroy(metadata->metadata_stream, NULL); +error_no_stream: +end: + return ret; +} + +/* + * Snapshot the whole metadata. + * + * Returns 0 on success, < 0 on error + */ +static int snapshot_metadata(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); + assert(ctx); + + DBG("UST consumer snapshot metadata with key %" PRIu64 " at path %s", + key, path); + + rcu_read_lock(); + + metadata_channel = consumer_find_channel(key); + if (!metadata_channel) { + ERR("UST snapshot metadata channel not found for key %lu", key); + ret = -1; + goto error; + } + assert(!metadata_channel->monitor); + + /* + * Ask the sessiond if we have new metadata waiting and update the + * consumer metadata cache. + */ + ret = lttng_ustconsumer_request_metadata(ctx, metadata_channel, 0); + if (ret < 0) { + goto error; + } + + /* + * The metadata stream is NOT created in no monitor mode when the channel + * is created on a sessiond ask channel command. + */ + ret = create_ust_streams(metadata_channel, ctx); + if (ret < 0) { + goto error; + } + + metadata_stream = metadata_channel->metadata_stream; + assert(metadata_stream); + + 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); + if (ret < 0) { + goto error_stream; + } + metadata_stream->out_fd = ret; + metadata_stream->tracefile_size_current = 0; + } + + pthread_mutex_lock(&metadata_channel->metadata_cache->lock); + + do { + ret = lttng_consumer_read_subbuffer(metadata_stream, ctx); + if (ret < 0) { + goto error_unlock; + } + } while (ret > 0); + +error_unlock: + pthread_mutex_unlock(&metadata_channel->metadata_cache->lock); + +error_stream: + /* + * Clean up the stream completly because the next snapshot will use a new + * metadata stream. + */ + cds_list_del(&metadata_stream->send_node); + consumer_stream_destroy(metadata_stream, NULL); + metadata_channel->metadata_stream = NULL; + +error: + rcu_read_unlock(); + return ret; +} + +/* + * Take a snapshot of all the stream of a channel. + * + * Returns 0 on success, < 0 on error + */ +static int snapshot_channel(uint64_t key, char *path, uint64_t relayd_id, + uint64_t max_stream_size, 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); + assert(ctx); + + rcu_read_lock(); + + if (relayd_id != (uint64_t) -1ULL) { + use_relayd = 1; + } + + channel = consumer_find_channel(key); + if (!channel) { + ERR("UST snapshot channel not found for key %lu", key); + ret = -1; + goto error; + } + assert(!channel->monitor); + DBG("UST consumer snapshot channel %lu", key); + + cds_list_for_each_entry(stream, &channel->streams.head, send_node) { + /* Lock stream because we are about to change its state. */ + pthread_mutex_lock(&stream->lock); + stream->net_seq_idx = relayd_id; + + if (use_relayd) { + ret = consumer_send_relayd_stream(stream, path); + if (ret < 0) { + goto error_unlock; + } + } else { + ret = utils_create_stream_file(path, stream->name, + stream->chan->tracefile_size, + stream->tracefile_count_current, + stream->uid, stream->gid); + 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); + } + + ustctl_flush_buffer(stream->ustream, 1); + + ret = lttng_ustconsumer_take_snapshot(stream); + if (ret < 0) { + ERR("Taking UST snapshot"); + goto error_unlock; + } + + ret = lttng_ustconsumer_get_produced_snapshot(stream, &produced_pos); + if (ret < 0) { + ERR("Produced UST snapshot position"); + goto error_unlock; + } + + ret = lttng_ustconsumer_get_consumed_snapshot(stream, &consumed_pos); + if (ret < 0) { + ERR("Consumerd UST snapshot position"); + goto error_unlock; + } + + /* + * The original value is sent back if max stream size is larger than + * the possible size of the snapshot. Also, we asume that the session + * daemon should never send a maximum stream size that is lower than + * subbuffer size. + */ + consumed_pos = consumer_get_consumed_maxsize(consumed_pos, + produced_pos, max_stream_size); + + while (consumed_pos < produced_pos) { + ssize_t read_len; + unsigned long len, padded_len; + + DBG("UST consumer taking snapshot at pos %lu", consumed_pos); + + ret = ustctl_get_subbuf(stream->ustream, &consumed_pos); + if (ret < 0) { + if (ret != -EAGAIN) { + PERROR("ustctl_get_subbuf snapshot"); + goto error_close_stream; + } + DBG("UST consumer get subbuf failed. Skipping it."); + consumed_pos += stream->max_sb_size; + continue; + } + + ret = ustctl_get_subbuf_size(stream->ustream, &len); + if (ret < 0) { + ERR("Snapshot ustctl_get_subbuf_size"); + goto error_put_subbuf; + } + + ret = ustctl_get_padded_subbuf_size(stream->ustream, &padded_len); + if (ret < 0) { + ERR("Snapshot ustctl_get_padded_subbuf_size"); + goto error_put_subbuf; + } + + read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len, + padded_len - len); + if (use_relayd) { + if (read_len != len) { + ret = -1; + goto error_put_subbuf; + } + } else { + if (read_len != padded_len) { + ret = -1; + goto error_put_subbuf; + } + } + + ret = ustctl_put_subbuf(stream->ustream); + if (ret < 0) { + ERR("Snapshot ustctl_put_subbuf"); + goto error_close_stream; + } + consumed_pos += stream->max_sb_size; + } + + /* Simply close the stream so we can use it on the next snapshot. */ + consumer_stream_close(stream); + pthread_mutex_unlock(&stream->lock); + } + + rcu_read_unlock(); + return 0; +error_put_subbuf: + if (ustctl_put_subbuf(stream->ustream) < 0) { + ERR("Snapshot ustctl_put_subbuf"); + } +error_close_stream: + consumer_stream_close(stream); +error_unlock: + pthread_mutex_unlock(&stream->lock); error: + rcu_read_unlock(); return ret; } @@ -692,12 +980,13 @@ error: * Receive the metadata updates from the sessiond. */ int lttng_ustconsumer_recv_metadata(int sock, uint64_t key, uint64_t offset, - uint64_t len, struct lttng_consumer_channel *channel) + uint64_t len, struct lttng_consumer_channel *channel, + int timer) { int ret, ret_code = LTTNG_OK; char *metadata_str; - DBG("UST consumer push metadata key %lu of len %lu", key, len); + DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key, len); metadata_str = zmalloc(len * sizeof(char)); if (!metadata_str) { @@ -719,10 +1008,17 @@ int lttng_ustconsumer_recv_metadata(int sock, uint64_t key, uint64_t offset, if (ret < 0) { /* Unable to handle metadata. Notify session daemon. */ ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA; + /* + * Skip metadata flush on write error since the offset and len might + * 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; } pthread_mutex_unlock(&channel->metadata_cache->lock); - while (consumer_metadata_cache_flushed(channel, offset + len)) { + while (consumer_metadata_cache_flushed(channel, offset + len, timer)) { DBG("Waiting for metadata to be flushed"); usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME); } @@ -750,11 +1046,14 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, if (ret != sizeof(msg)) { DBG("Consumer received unexpected message size %zd (expects %zu)", ret, sizeof(msg)); - lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD); /* * 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; + } return ret; } if (msg.cmd_type == LTTNG_CONSUMER_STOP) { @@ -830,6 +1129,7 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, sizeof(is_data_pending)); if (ret < 0) { DBG("Error when sending the data pending ret code: %d", ret); + goto error_fatal; } /* @@ -848,7 +1148,11 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, 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, - (enum lttng_event_output) msg.u.ask_channel.output); + (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); if (!channel) { goto end_channel_error; } @@ -860,23 +1164,20 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, attr.switch_timer_interval = msg.u.ask_channel.switch_timer_interval; attr.read_timer_interval = msg.u.ask_channel.read_timer_interval; attr.chan_id = msg.u.ask_channel.chan_id; + attr.output = msg.u.ask_channel.output; memcpy(attr.uuid, msg.u.ask_channel.uuid, sizeof(attr.uuid)); - /* Translate the event output type to UST. */ - switch (channel->output) { - case LTTNG_EVENT_SPLICE: - /* Splice not supported so fallback on mmap(). */ - case LTTNG_EVENT_MMAP: - default: - attr.output = CONSUMER_CHANNEL_MMAP; - break; - }; - /* Translate and save channel type. */ switch (msg.u.ask_channel.type) { case LTTNG_UST_CHAN_PER_CPU: channel->type = CONSUMER_CHANNEL_TYPE_DATA; attr.type = LTTNG_UST_CHAN_PER_CPU; + /* + * Set refcount to 1 for owner. Below, we will + * pass ownership to the + * consumer_thread_channel_poll() thread. + */ + channel->refcount = 1; break; case LTTNG_UST_CHAN_METADATA: channel->type = CONSUMER_CHANNEL_TYPE_METADATA; @@ -892,17 +1193,34 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, goto end_channel_error; } + if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) { + ret = consumer_metadata_cache_allocate(channel); + if (ret < 0) { + ERR("Allocating metadata cache"); + goto end_channel_error; + } + consumer_timer_switch_start(channel, attr.switch_timer_interval); + attr.switch_timer_interval = 0; + } + /* * Add the channel to the internal state AFTER all streams were created * and successfully sent to session daemon. This way, all streams must * be ready before this channel is visible to the threads. + * 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) { + if (channel->switch_timer_enabled == 1) { + consumer_timer_switch_stop(channel); + } + consumer_metadata_cache_destroy(channel); + } goto end_channel_error; } - /* * Channel and streams are now created. Inform the session daemon that * everything went well and should wait to receive the channel and @@ -911,20 +1229,9 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, ret = consumer_send_status_channel(sock, channel); if (ret < 0) { /* - * There is probably a problem on the socket so the poll will get - * it and clean everything up. + * There is probably a problem on the socket. */ - goto end_nosignal; - } - - if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) { - ret = consumer_metadata_cache_allocate(channel); - if (ret < 0) { - ERR("Allocating metadata cache"); - goto end_channel_error; - } - consumer_timer_switch_start(channel, attr.switch_timer_interval); - attr.switch_timer_interval = 0; + goto error_fatal; } break; @@ -937,18 +1244,11 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, channel = consumer_find_channel(key); if (!channel) { - ERR("UST consumer get channel key %lu not found", key); + ERR("UST consumer get channel key %" PRIu64 " not found", key); ret_code = LTTNG_ERR_UST_CHAN_NOT_FOUND; goto end_msg_sessiond; } - /* Inform sessiond that we are about to send channel and streams. */ - ret = consumer_send_status_msg(sock, LTTNG_OK); - if (ret < 0) { - /* Somehow, the session daemon is not responding anymore. */ - goto end_nosignal; - } - /* Send everything to sessiond. */ ret = send_sessiond_channel(sock, channel, ctx, &relayd_err); if (ret < 0) { @@ -956,10 +1256,10 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, /* * We were unable to send to the relayd the stream so avoid * sending back a fatal error to the thread since this is OK - * and the consumer can continue its work. + * and the consumer can continue its work. The above call + * has sent the error status message to the sessiond. */ - ret_code = LTTNG_ERR_RELAYD_CONNECT_FAIL; - goto end_msg_sessiond; + goto end_nosignal; } /* * The communicaton was broken hence there is a bad state between @@ -968,6 +1268,14 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, goto error_fatal; } + /* + * 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; + } + ret = send_streams_to_thread(channel, ctx); if (ret < 0) { /* @@ -978,23 +1286,18 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, } /* List MUST be empty after or else it could be reused. */ assert(cds_list_empty(&channel->streams.head)); - goto end_msg_sessiond; } case LTTNG_CONSUMER_DESTROY_CHANNEL: { uint64_t key = msg.u.destroy_channel.key; - struct lttng_consumer_channel *channel; - - channel = consumer_find_channel(key); - if (!channel) { - ERR("UST consumer get channel key %lu not found", key); - ret_code = LTTNG_ERR_UST_CHAN_NOT_FOUND; - goto end_msg_sessiond; - } - - destroy_channel(channel); + /* + * Only called if streams have not been sent to stream + * manager thread. However, channel has been sent to + * channel manager thread. + */ + notify_thread_del_channel(ctx, key); goto end_msg_sessiond; } case LTTNG_CONSUMER_CLOSE_METADATA: @@ -1027,12 +1330,14 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, uint64_t offset = msg.u.push_metadata.target_offset; struct lttng_consumer_channel *channel; - DBG("UST consumer push metadata key %lu of len %lu", key, len); + DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key, + len); channel = consumer_find_channel(key); if (!channel) { - ERR("UST consumer push metadata %lu not found", key); + ERR("UST consumer push metadata %" PRIu64 " not found", key); ret_code = LTTNG_ERR_UST_CHAN_NOT_FOUND; + goto end_msg_sessiond; } /* Tell session daemon we are ready to receive the metadata. */ @@ -1044,14 +1349,14 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, /* Wait for more data. */ if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) { - goto end_nosignal; + goto error_fatal; } ret = lttng_ustconsumer_recv_metadata(sock, key, offset, - len, channel); + len, channel, 0); if (ret < 0) { /* error receiving from sessiond */ - goto end_nosignal; + goto error_fatal; } else { ret_code = ret; goto end_msg_sessiond; @@ -1067,6 +1372,36 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, } goto end_msg_sessiond; } + 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 = LTTNG_ERR_UST_META_FAIL; + } + } else { + ret = snapshot_channel(msg.u.snapshot_channel.key, + msg.u.snapshot_channel.pathname, + msg.u.snapshot_channel.relayd_id, + msg.u.snapshot_channel.max_stream_size, + ctx); + if (ret < 0) { + ERR("Snapshot channel failed"); + ret_code = LTTNG_ERR_UST_CHAN_FAIL; + } + } + + ret = consumer_send_status_msg(sock, ret_code); + if (ret < 0) { + /* Somehow, the session daemon is not responding anymore. */ + goto end_nosignal; + } + break; + } default: break; } @@ -1086,7 +1421,10 @@ end_msg_sessiond: * the caller because the session daemon socket management is done * elsewhere. Returning a negative code or 0 will shutdown the consumer. */ - (void) consumer_send_status_msg(sock, ret_code); + ret = consumer_send_status_msg(sock, ret_code); + if (ret < 0) { + goto error_fatal; + } rcu_read_unlock(); return 1; end_channel_error: @@ -1164,6 +1502,21 @@ int lttng_ustconsumer_get_produced_snapshot( return ustctl_snapshot_get_produced(stream->ustream, pos); } +/* + * Get the consumed position + * + * Returns 0 on success, < 0 on error + */ +int lttng_ustconsumer_get_consumed_snapshot( + struct lttng_consumer_stream *stream, unsigned long *pos) +{ + assert(stream); + assert(stream->ustream); + assert(pos); + + return ustctl_snapshot_get_consumed(stream->ustream, pos); +} + /* * Called when the stream signal the consumer that it has hang up. */ @@ -1181,6 +1534,10 @@ void lttng_ustconsumer_del_channel(struct lttng_consumer_channel *chan) assert(chan); assert(chan->uchan); + if (chan->switch_timer_enabled == 1) { + consumer_timer_switch_stop(chan); + } + consumer_metadata_cache_destroy(chan); ustctl_destroy_channel(chan->uchan); } @@ -1189,6 +1546,9 @@ void lttng_ustconsumer_del_stream(struct lttng_consumer_stream *stream) assert(stream); assert(stream->ustream); + if (stream->chan->switch_timer_enabled == 1) { + consumer_timer_switch_stop(stream->chan); + } ustctl_destroy_stream(stream->ustream); } @@ -1205,28 +1565,57 @@ int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream *stream, assert(stream->ustream); assert(ctx); - DBG2("In UST read_subbuffer (wait_fd: %d, name: %s)", stream->wait_fd, + 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; /* We can consume the 1 byte written into the wait_fd by UST */ - if (!stream->hangup_flush_done) { + if (stream->monitor && !stream->hangup_flush_done) { ssize_t readlen; do { readlen = read(stream->wait_fd, &dummy, 1); } while (readlen == -1 && errno == EINTR); - if (readlen == -1) { + if (readlen == -1 && 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) { + ssize_t write_len; + + if (stream->chan->metadata_cache->contiguous + == stream->ust_metadata_pushed) { + 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); + assert(write_len != 0); + if (write_len < 0) { + ERR("Writing one metadata packet"); + ret = -1; + goto end; + } + stream->ust_metadata_pushed += write_len; + ustctl_flush_buffer(stream->ustream, 1); + goto retry; + } + ret = err; /* ustctl_get_next_subbuf returns negative, caller expect positive. */ /* * This is a debug message even for single-threaded consumer, @@ -1268,7 +1657,7 @@ int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream *stream, * happen and it is OK with the code flow. */ DBG("Error writing to tracefile " - "(ret: %zd != len: %lu != subbuf_size: %lu)", + "(ret: %ld != len: %lu != subbuf_size: %lu)", ret, len, subbuf_size); } err = ustctl_put_next_subbuf(ustream); @@ -1280,35 +1669,27 @@ end: /* * Called when a stream is created. + * + * Return 0 on success or else a negative value. */ int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream *stream) { int ret; - char full_path[PATH_MAX]; - - /* Opening the tracefile in write mode */ - if (stream->net_seq_idx != (uint64_t) -1ULL) { - goto end; - } - ret = snprintf(full_path, sizeof(full_path), "%s/%s", - stream->chan->pathname, stream->name); - if (ret < 0) { - PERROR("snprintf on_recv_stream"); - goto error; - } + assert(stream); - ret = run_as_open(full_path, O_WRONLY | O_CREAT | O_TRUNC, - S_IRWXU | S_IRWXG | S_IRWXO, stream->uid, stream->gid); - if (ret < 0) { - PERROR("open stream path %s", full_path); - goto error; + /* 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); + if (ret < 0) { + goto error; + } + stream->out_fd = ret; + stream->tracefile_size_current = 0; } - stream->out_fd = ret; - -end: - /* we return 0 to let the library handle the FD internally */ - return 0; + ret = 0; error: return ret; @@ -1331,15 +1712,44 @@ int lttng_ustconsumer_data_pending(struct lttng_consumer_stream *stream) DBG("UST consumer checking data pending"); - ret = ustctl_get_next_subbuf(stream->ustream); - if (ret == 0) { - /* There is still data so let's put back this subbuffer. */ - ret = ustctl_put_subbuf(stream->ustream); - assert(ret == 0); - ret = 1; /* Data is pending */ + if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) { + ret = 0; goto end; } + if (stream->chan->type == CONSUMER_CHANNEL_TYPE_METADATA) { + /* + * We can simply check whether all contiguously available data + * has been pushed to the ring buffer, since the push operation + * is performed within get_next_subbuf(), and because both + * get_next_subbuf() and put_next_subbuf() are issued atomically + * thanks to the stream lock within + * lttng_ustconsumer_read_subbuffer(). This basically means that + * whetnever ust_metadata_pushed is incremented, the associated + * metadata has been consumed from the metadata stream. + */ + DBG("UST consumer metadata pending check: contiguous %" PRIu64 " vs pushed %" PRIu64, + stream->chan->metadata_cache->contiguous, + stream->ust_metadata_pushed); + if (stream->chan->metadata_cache->contiguous + != stream->ust_metadata_pushed) { + ret = 1; /* Data is pending */ + goto end; + } + } else { + ret = ustctl_get_next_subbuf(stream->ustream); + if (ret == 0) { + /* + * There is still data so let's put back this + * subbuffer. + */ + ret = ustctl_put_subbuf(stream->ustream); + assert(ret == 0); + ret = 1; /* Data is pending */ + goto end; + } + } + /* Data is NOT pending so ready to be read. */ ret = 0; @@ -1395,8 +1805,14 @@ void lttng_ustconsumer_close_stream_wakeup(struct lttng_consumer_stream *stream) } } +/* + * Please refer to consumer-timer.c before adding any lock within this + * function or any of its callees. Timers have a very strict locking + * semantic with respect to teardown. Failure to respect this semantic + * introduces deadlocks. + */ int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx, - struct lttng_consumer_channel *channel) + struct lttng_consumer_channel *channel, int timer) { struct lttcomm_metadata_request_msg request; struct lttcomm_consumer_msg msg; @@ -1421,11 +1837,15 @@ int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx, } request.session_id = channel->session_id; + request.session_id_per_pid = channel->session_id_per_pid; request.uid = channel->uid; request.key = channel->key; - DBG("Sending metadata request to sessiond, session %" PRIu64, - channel->session_id); + DBG("Sending metadata request to sessiond, session id %" PRIu64 + ", per-pid %" PRIu64, + channel->session_id, + channel->session_id_per_pid); + pthread_mutex_lock(&ctx->metadata_socket_lock); ret = lttcomm_send_unix_sock(ctx->consumer_metadata_socket, &request, sizeof(request)); if (ret < 0) { @@ -1437,7 +1857,7 @@ int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx, ret = lttcomm_recv_unix_sock(ctx->consumer_metadata_socket, &msg, sizeof(msg)); if (ret != sizeof(msg)) { - DBG("Consumer received unexpected message size %d (expects %lu)", + DBG("Consumer received unexpected message size %d (expects %zu)", ret, sizeof(msg)); lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD); /* @@ -1480,10 +1900,17 @@ int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx, } ret_code = lttng_ustconsumer_recv_metadata(ctx->consumer_metadata_socket, - key, offset, len, channel); - (void) consumer_send_status_msg(ctx->consumer_metadata_socket, ret_code); + key, offset, len, channel, timer); + if (ret_code >= 0) { + /* + * Only send the status msg if the sessiond is alive meaning a positive + * ret code. + */ + (void) consumer_send_status_msg(ctx->consumer_metadata_socket, ret_code); + } ret = 0; end: + pthread_mutex_unlock(&ctx->metadata_socket_lock); return ret; }