X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fcommon%2Fust-consumer%2Fust-consumer.c;h=147fe8aafe899a092a91ee8be58357786d7f074f;hp=3133835be78091700eaa64e16e114626ca8327f2;hb=e9404c27e7cc9d841785e6c4292c1add19fbc1cc;hpb=d01178b6f6465443d7e6e1015aa7054e9d093e91 diff --git a/src/common/ust-consumer/ust-consumer.c b/src/common/ust-consumer/ust-consumer.c index 3133835be..147fe8aaf 100644 --- a/src/common/ust-consumer/ust-consumer.c +++ b/src/common/ust-consumer/ust-consumer.c @@ -1,6 +1,7 @@ /* * Copyright (C) 2011 - Julien Desfossez * 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, @@ -16,7 +17,7 @@ * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. */ -#define _GNU_SOURCE +#define _LGPL_SOURCE #include #include #include @@ -32,17 +33,22 @@ #include #include +#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; @@ -62,6 +68,9 @@ static void destroy_channel(struct lttng_consumer_channel *channel) cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head, send_node) { + + health_code_update(); + cds_list_del(&stream->send_node); ustctl_destroy_stream(stream->ustream); free(stream); @@ -73,6 +82,7 @@ static void destroy_channel(struct lttng_consumer_channel *channel) */ if (channel->uchan) { lttng_ustconsumer_del_channel(channel); + lttng_ustconsumer_free_channel(channel); } free(channel); } @@ -116,14 +126,17 @@ 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) + 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); + tracefile_count, session_id_per_pid, monitor, + live_timer_interval, root_shm_path, shm_path); } /* @@ -194,18 +207,61 @@ 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; + } 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; + } 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. + */ + stream->globally_visible = 1; + ret = lttng_pipe_write(stream_pipe, &stream, sizeof(stream)); if (ret < 0) { ERR("Consumer write %s stream to pipe %d", stream->metadata_flag ? "metadata" : "data", lttng_pipe_get_writefd(stream_pipe)); + if (stream->metadata_flag) { + consumer_del_stream_for_metadata(stream); + } else { + consumer_del_stream_for_data(stream); + } } +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; } @@ -230,8 +286,20 @@ 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]; + + health_code_update(); - 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); @@ -285,6 +353,12 @@ 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; + if (channel->monitor) { + /* Set metadata poll pipe if we created one */ + memcpy(stream->ust_metadata_poll_pipe, + ust_metadata_pipe, + sizeof(ust_metadata_pipe)); + } } } @@ -295,20 +369,87 @@ error_alloc: return ret; } +/* + * create_posix_shm is never called concurrently within a process. + */ +static +int create_posix_shm(void) +{ + char tmp_name[NAME_MAX]; + int shmfd, ret; + + ret = snprintf(tmp_name, NAME_MAX, "/ust-shm-consumer-%d", getpid()); + if (ret < 0) { + PERROR("snprintf"); + return -1; + } + /* + * Allocate shm, and immediately unlink its shm oject, keeping + * only the file descriptor as a reference to the object. + * We specifically do _not_ use the / at the beginning of the + * pathname so that some OS implementations can keep it local to + * the process (POSIX leaves this implementation-defined). + */ + shmfd = shm_open(tmp_name, O_CREAT | O_EXCL | O_RDWR, 0700); + if (shmfd < 0) { + PERROR("shm_open"); + goto error_shm_open; + } + ret = shm_unlink(tmp_name); + if (ret < 0 && errno != ENOENT) { + PERROR("shm_unlink"); + goto error_shm_release; + } + return shmfd; + +error_shm_release: + ret = close(shmfd); + if (ret) { + PERROR("close"); + } +error_shm_open: + return -1; +} + +static int open_ust_stream_fd(struct lttng_consumer_channel *channel, + struct ustctl_consumer_channel_attr *attr, + int cpu) +{ + char shm_path[PATH_MAX]; + int ret; + + if (!channel->shm_path[0]) { + return create_posix_shm(); + } + 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, + channel->uid, channel->gid); + +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); DBG3("Creating channel to ustctl with attr: [overwrite: %d, " "subbuf_size: %" PRIu64 ", num_subbuf: %" PRIu64 ", " @@ -317,17 +458,64 @@ 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, attr, i); + 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, + channel->uid, channel->gid); + 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, + channel->uid, channel->gid); + } + free(stream_fds); +error_alloc: return ret; } @@ -343,7 +531,7 @@ 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); @@ -364,8 +552,9 @@ static int send_sessiond_channel(int sock, struct lttng_consumer_channel *channel, struct lttng_consumer_local_data *ctx, int *relayd_error) { - int ret, ret_code = LTTNG_OK; + int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS; struct lttng_consumer_stream *stream; + uint64_t net_seq_idx = -1ULL; assert(channel); assert(ctx); @@ -373,24 +562,32 @@ static int send_sessiond_channel(int sock, DBG("UST consumer sending channel %s to sessiond", channel->name); - 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; + if (channel->relayd_id != (uint64_t) -1ULL) { + cds_list_for_each_entry(stream, &channel->streams.head, send_node) { + + health_code_update(); + + /* 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 = LTTCOMM_CONSUMERD_RELAYD_FAIL; + } + if (net_seq_idx == -1ULL) { + net_seq_idx = stream->net_seq_idx; } - 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) { + if (ret < 0 || ret_code != LTTCOMM_CONSUMERD_SUCCESS) { /* * Either the session daemon is not responding or the relayd died so we * stop now. @@ -411,6 +608,9 @@ 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) { + + health_code_update(); + /* Send stream to session daemon. */ ret = send_sessiond_stream(sock, stream); if (ret < 0) { @@ -429,7 +629,7 @@ static int send_sessiond_channel(int sock, return 0; error: - if (ret_code != LTTNG_OK) { + if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) { ret = -1; } return ret; @@ -469,7 +669,7 @@ 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; } @@ -512,6 +712,9 @@ static int send_streams_to_thread(struct lttng_consumer_channel *channel, /* Send streams to the corresponding thread. */ cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head, send_node) { + + health_code_update(); + /* Sending the stream to the thread. */ ret = send_stream_to_thread(stream, ctx); if (ret < 0) { @@ -519,17 +722,14 @@ 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); - /* - * From this point on, the stream's ownership has been moved away from - * the channel and becomes globally visible. - */ - stream->globally_visible = 1; } error: @@ -537,48 +737,56 @@ error: } /* - * 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. + * Flush channel's streams using the given key to retrieve the channel. * - * Return 0 on success else a negative value. + * Return 0 on success else an LTTng error code. */ -int lttng_ustconsumer_push_metadata(struct lttng_consumer_channel *metadata, - const char *metadata_str, uint64_t target_offset, uint64_t len) +static int flush_channel(uint64_t chan_key) { - int ret; - - assert(metadata); - assert(metadata_str); + 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 writing metadata to channel %s", metadata->name); + DBG("UST consumer flush channel key %" PRIu64, chan_key); - if (!metadata->metadata_stream) { - ret = 0; + rcu_read_lock(); + channel = consumer_find_channel(chan_key); + if (!channel) { + ERR("UST consumer flush channel %" PRIu64 " not found", chan_key); + ret = LTTNG_ERR_UST_CHAN_NOT_FOUND; goto error; } - 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 %" PRIu64, ret, len); - goto error; - } + ht = consumer_data.stream_per_chan_id_ht; + + /* For each stream of the channel id, flush it. */ + 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(metadata->metadata_stream->ustream, 1); + health_code_update(); + pthread_mutex_lock(&stream->lock); + if (!stream->quiescent) { + ustctl_flush_buffer(stream->ustream, 0); + stream->quiescent = true; + } + pthread_mutex_unlock(&stream->lock); + } error: + rcu_read_unlock(); return ret; } /* - * Flush channel's streams using the given key to retrieve the channel. + * 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 flush_channel(uint64_t chan_key) +static int clear_quiescent_channel(uint64_t chan_key) { int ret = 0; struct lttng_consumer_channel *channel; @@ -586,23 +794,28 @@ static int flush_channel(uint64_t chan_key) struct lttng_ht *ht; struct lttng_ht_iter iter; - DBG("UST consumer flush channel key %" PRIu64, chan_key); + DBG("UST consumer clear quiescent channel key %" PRIu64, chan_key); rcu_read_lock(); channel = consumer_find_channel(chan_key); if (!channel) { - ERR("UST consumer flush channel %" PRIu64 " not found", chan_key); + 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, flush it. */ + /* 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) { - ustctl_flush_buffer(stream->ustream, 1); + + health_code_update(); + + pthread_mutex_lock(&stream->lock); + stream->quiescent = false; + pthread_mutex_unlock(&stream->lock); } error: rcu_read_unlock(); @@ -636,26 +849,16 @@ static int close_metadata(uint64_t chan_key) } 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); - } - - 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; - } - } + lttng_ustconsumer_close_metadata(channel); error_unlock: + pthread_mutex_unlock(&channel->lock); pthread_mutex_unlock(&consumer_data.lock); error: return ret; @@ -701,11 +904,19 @@ static int setup_metadata(struct lttng_consumer_local_data *ctx, uint64_t key) } /* Send metadata stream to relayd if needed. */ - ret = consumer_send_relayd_stream(metadata->metadata_stream, - metadata->pathname); - 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 = consumer_send_relayd_streams_sent( + metadata->metadata_stream->net_seq_idx); + if (ret < 0) { + ret = LTTCOMM_CONSUMERD_RELAYD_FAIL; + goto error; + } } ret = send_streams_to_thread(metadata, ctx); @@ -730,8 +941,9 @@ error: * 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); + cds_list_del(&metadata->metadata_stream->send_node); + metadata->metadata_stream = NULL; error_no_stream: end: return ret; @@ -746,8 +958,6 @@ static int snapshot_metadata(uint64_t key, char *path, uint64_t relayd_id, struct lttng_consumer_local_data *ctx) { int ret = 0; - ssize_t write_len; - uint64_t total_len = 0; struct lttng_consumer_channel *metadata_channel; struct lttng_consumer_stream *metadata_stream; @@ -761,21 +971,26 @@ static int snapshot_metadata(uint64_t key, char *path, uint64_t relayd_id, metadata_channel = consumer_find_channel(key); if (!metadata_channel) { - ERR("UST snapshot metadata channel not found for key %lu", key); + ERR("UST snapshot metadata channel not found for key %" PRIu64, + key); ret = -1; goto error; } assert(!metadata_channel->monitor); + health_code_update(); + /* * Ask the sessiond if we have new metadata waiting and update the * consumer metadata cache. */ - ret = lttng_ustconsumer_request_metadata(ctx, metadata_channel); + ret = lttng_ustconsumer_request_metadata(ctx, metadata_channel, 0, 1); if (ret < 0) { goto error; } + health_code_update(); + /* * The metadata stream is NOT created in no monitor mode when the channel * is created on a sessiond ask channel command. @@ -798,7 +1013,7 @@ static int snapshot_metadata(uint64_t key, char *path, uint64_t relayd_id, 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); + metadata_stream->uid, metadata_stream->gid, NULL); if (ret < 0) { goto error_stream; } @@ -806,42 +1021,22 @@ static int snapshot_metadata(uint64_t key, char *path, uint64_t relayd_id, metadata_stream->tracefile_size_current = 0; } - pthread_mutex_lock(&metadata_channel->metadata_cache->lock); - while (total_len < metadata_channel->metadata_cache->total_bytes_written) { - /* - * Write at most one packet of metadata into the channel - * to avoid blocking here. - */ - write_len = ustctl_write_one_packet_to_channel(metadata_channel->uchan, - metadata_channel->metadata_cache->data, - metadata_channel->metadata_cache->total_bytes_written); - if (write_len < 0) { - ERR("UST consumer snapshot writing metadata packet"); - ret = -1; - goto error_unlock; - } - total_len += write_len; + do { + health_code_update(); - DBG("Written %" PRIu64 " bytes to metadata (left: %" PRIu64 ")", - write_len, - metadata_channel->metadata_cache->total_bytes_written - write_len); - ustctl_flush_buffer(metadata_stream->ustream, 1); ret = lttng_consumer_read_subbuffer(metadata_stream, ctx); if (ret < 0) { - goto error_unlock; + goto error_stream; } - } - -error_unlock: - pthread_mutex_unlock(&metadata_channel->metadata_cache->lock); + } while (ret > 0); 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); + cds_list_del(&metadata_stream->send_node); metadata_channel->metadata_stream = NULL; error: @@ -855,7 +1050,7 @@ error: * Returns 0 on success, < 0 on error */ static int snapshot_channel(uint64_t key, char *path, uint64_t relayd_id, - struct lttng_consumer_local_data *ctx) + uint64_t nb_packets_per_stream, struct lttng_consumer_local_data *ctx) { int ret; unsigned use_relayd = 0; @@ -874,14 +1069,19 @@ static int snapshot_channel(uint64_t key, char *path, uint64_t relayd_id, channel = consumer_find_channel(key); if (!channel) { - ERR("UST snapshot channel not found for key %lu", key); + ERR("UST snapshot channel not found for key %" PRIu64, key); ret = -1; goto error; } assert(!channel->monitor); - DBG("UST consumer snapshot channel %lu", key); + 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); stream->net_seq_idx = relayd_id; @@ -895,7 +1095,7 @@ static int snapshot_channel(uint64_t key, char *path, uint64_t relayd_id, ret = utils_create_stream_file(path, stream->name, stream->chan->tracefile_size, stream->tracefile_count_current, - stream->uid, stream->gid); + stream->uid, stream->gid, NULL); if (ret < 0) { goto error_unlock; } @@ -905,8 +1105,20 @@ static int snapshot_channel(uint64_t key, char *path, uint64_t relayd_id, 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); + if (ret < 0) { + goto error_unlock; + } + } - 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) { @@ -926,9 +1138,22 @@ static int snapshot_channel(uint64_t key, char *path, uint64_t relayd_id, goto error_unlock; } + /* + * The original value is sent back if max stream size is larger than + * the possible size of the snapshot. Also, we assume that the session + * daemon should never send a maximum stream size that is lower than + * subbuffer size. + */ + consumed_pos = consumer_get_consume_start_pos(consumed_pos, + produced_pos, nb_packets_per_stream, + stream->max_sb_size); + while (consumed_pos < produced_pos) { ssize_t read_len; unsigned long len, padded_len; + int lost_packet = 0; + + health_code_update(); DBG("UST consumer taking snapshot at pos %lu", consumed_pos); @@ -940,6 +1165,15 @@ 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; + } continue; } @@ -956,15 +1190,15 @@ static int snapshot_channel(uint64_t key, char *path, uint64_t relayd_id, } read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len, - padded_len - len); + padded_len - len, NULL); if (use_relayd) { if (read_len != len) { - ret = -1; + ret = -EPERM; goto error_put_subbuf; } } else { if (read_len != padded_len) { - ret = -1; + ret = -EPERM; goto error_put_subbuf; } } @@ -975,6 +1209,16 @@ 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. */ @@ -999,12 +1243,18 @@ error: } /* - * 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) + uint64_t len, uint64_t version, + struct lttng_consumer_channel *channel, int timer, int wait) { - int ret, ret_code = LTTNG_OK; + int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS; char *metadata_str; DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key, len); @@ -1016,6 +1266,8 @@ int lttng_ustconsumer_recv_metadata(int sock, uint64_t key, uint64_t offset, goto end; } + health_code_update(); + /* Receive metadata string. */ ret = lttcomm_recv_unix_sock(sock, metadata_str, len); if (ret < 0) { @@ -1024,19 +1276,11 @@ int lttng_ustconsumer_recv_metadata(int sock, uint64_t key, uint64_t offset, goto end_free; } - /* - * XXX: The consumer data lock is acquired before calling metadata cache - * write which calls push metadata that MUST be protected by the consumer - * lock in order to be able to check the validity of the metadata stream of - * the channel. - * - * Note that this will be subject to change to better fine grained locking - * and ultimately try to get rid of this global consumer data lock. - */ - pthread_mutex_lock(&consumer_data.lock); + health_code_update(); pthread_mutex_lock(&channel->metadata_cache->lock); - ret = consumer_metadata_cache_write(channel, offset, len, metadata_str); + ret = consumer_metadata_cache_write(channel, offset, len, version, + metadata_str); if (ret < 0) { /* Unable to handle metadata. Notify session daemon. */ ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA; @@ -1046,14 +1290,18 @@ int lttng_ustconsumer_recv_metadata(int sock, uint64_t key, uint64_t offset, * waiting for the metadata cache to be flushed. */ pthread_mutex_unlock(&channel->metadata_cache->lock); - pthread_mutex_unlock(&consumer_data.lock); goto end_free; } pthread_mutex_unlock(&channel->metadata_cache->lock); - pthread_mutex_unlock(&consumer_data.lock); - while (consumer_metadata_cache_flushed(channel, offset + len)) { + if (!wait) { + goto end_free; + } + while (consumer_metadata_cache_flushed(channel, offset + len, timer)) { DBG("Waiting for metadata to be flushed"); + + health_code_update(); + usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME); } @@ -1072,10 +1320,12 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, int sock, struct pollfd *consumer_sockpoll) { ssize_t ret; - enum lttng_error_code ret_code = LTTNG_OK; + 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)", @@ -1090,17 +1340,13 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, } return ret; } - if (msg.cmd_type == LTTNG_CONSUMER_STOP) { - /* - * Notify the session daemon that the command is completed. - * - * On transport layer error, the function call will print an error - * message so handling the returned code is a bit useless since we - * return an error code anyway. - */ - (void) consumer_send_status_msg(sock, ret_code); - return -ENOENT; - } + + health_code_update(); + + /* deprecated */ + assert(msg.cmd_type != LTTNG_CONSUMER_STOP); + + health_code_update(); /* relayd needs RCU read-side lock */ rcu_read_lock(); @@ -1111,7 +1357,8 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, /* Session daemon status message are handled in the following call. */ ret = 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.sock, msg.u.relayd_sock.session_id, + msg.u.relayd_sock.relayd_session_id); goto end_nosignal; } case LTTNG_CONSUMER_DESTROY_RELAYD: @@ -1125,7 +1372,7 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, relayd = consumer_find_relayd(index); if (relayd == NULL) { DBG("Unable to find relayd %" PRIu64, index); - ret_code = LTTNG_ERR_NO_CONSUMER; + ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL; } /* @@ -1186,11 +1433,21 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, 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.monitor, + msg.u.ask_channel.live_timer_interval, + msg.u.ask_channel.root_shm_path, + msg.u.ask_channel.shm_path); if (!channel) { goto end_channel_error; } + /* + * 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 + * allocation. + */ + channel->ust_app_uid = msg.u.ask_channel.ust_app_uid; + /* Build channel attributes from received message. */ attr.subbuf_size = msg.u.ask_channel.subbuf_size; attr.num_subbuf = msg.u.ask_channel.num_subbuf; @@ -1198,9 +1455,16 @@ 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)); + /* Match channel buffer type to the UST abi. */ + switch (msg.u.ask_channel.output) { + case LTTNG_EVENT_MMAP: + default: + attr.output = LTTNG_UST_MMAP; + break; + } + /* Translate and save channel type. */ switch (msg.u.ask_channel.type) { case LTTNG_UST_CHAN_PER_CPU: @@ -1222,6 +1486,8 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, goto error_fatal; }; + health_code_update(); + ret = ask_channel(ctx, sock, channel, &attr); if (ret < 0) { goto end_channel_error; @@ -1235,8 +1501,22 @@ 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(); + /* * Add the channel to the internal state AFTER all streams were created * and successfully sent to session daemon. This way, all streams must @@ -1252,9 +1532,17 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, } consumer_metadata_cache_destroy(channel); } + 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; } + health_code_update(); + /* * Channel and streams are now created. Inform the session daemon that * everything went well and should wait to receive the channel and @@ -1279,10 +1567,12 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, channel = consumer_find_channel(key); if (!channel) { ERR("UST consumer get channel key %" PRIu64 " not found", key); - ret_code = LTTNG_ERR_UST_CHAN_NOT_FOUND; + ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND; goto end_msg_sessiond; } + health_code_update(); + /* Send everything to sessiond. */ ret = send_sessiond_channel(sock, channel, ctx, &relayd_err); if (ret < 0) { @@ -1302,6 +1592,8 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, goto error_fatal; } + health_code_update(); + /* * In no monitor mode, the streams ownership is kept inside the channel * so don't send them to the data thread. @@ -1356,12 +1648,25 @@ 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; + uint64_t version = msg.u.push_metadata.version; struct lttng_consumer_channel *channel; DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key, @@ -1369,25 +1674,50 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, channel = consumer_find_channel(key); if (!channel) { - ERR("UST consumer push metadata %" PRIu64 " not found", key); - ret_code = LTTNG_ERR_UST_CHAN_NOT_FOUND; + /* + * This is possible if the metadata creation on the consumer side + * is in flight vis-a-vis a concurrent push metadata from the + * session daemon. Simply return that the channel failed and the + * session daemon will handle that message correctly considering + * that this race is acceptable thus the DBG() statement here. + */ + DBG("UST consumer push metadata %" PRIu64 " not found", key); + ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL; + goto end_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_msg_sessiond; } /* Tell session daemon we are ready to receive the metadata. */ - ret = consumer_send_status_msg(sock, LTTNG_OK); + ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS); if (ret < 0) { /* Somehow, the session daemon is not responding anymore. */ goto error_fatal; } + health_code_update(); + /* Wait for more data. */ - if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) { + health_poll_entry(); + ret = lttng_consumer_poll_socket(consumer_sockpoll); + health_poll_exit(); + if (ret) { goto error_fatal; } + health_code_update(); + ret = lttng_ustconsumer_recv_metadata(sock, key, offset, - len, channel); + len, version, channel, 0, 1); if (ret < 0) { /* error receiving from sessiond */ goto error_fatal; @@ -1415,37 +1745,189 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, ctx); if (ret < 0) { ERR("Snapshot metadata failed"); - ret_code = LTTNG_ERR_UST_META_FAIL; + ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA; } } 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 = LTTNG_ERR_UST_CHAN_FAIL; + ret_code = LTTCOMM_CONSUMERD_CHANNEL_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_nosignal; } + health_code_update(); break; } - default: - 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; -end_nosignal: - rcu_read_unlock(); + DBG("UST consumer discarded events command for session id %" + PRIu64, id); + rcu_read_lock(); + pthread_mutex_lock(&consumer_data.lock); - /* - * Return 1 to indicate success since the 0 value can be a socket - * shutdown during the recv() or send() call. - */ + 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(); + + /* 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_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; + } + 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; end_msg_sessiond: @@ -1459,6 +1941,9 @@ end_msg_sessiond: goto error_fatal; } rcu_read_unlock(); + + health_code_update(); + return 1; end_channel_error: if (channel) { @@ -1475,6 +1960,9 @@ end_channel_error: goto error_fatal; } rcu_read_unlock(); + + health_code_update(); + return 1; error_fatal: rcu_read_unlock(); @@ -1508,7 +1996,7 @@ void *lttng_ustctl_get_mmap_base(struct lttng_consumer_stream *stream) } /* - * Take a snapshot for a specific fd + * Take a snapshot for a specific stream. * * Returns 0 on success, < 0 on error */ @@ -1520,6 +2008,20 @@ int lttng_ustconsumer_take_snapshot(struct lttng_consumer_stream *stream) return ustctl_snapshot(stream->ustream); } +/* + * Sample consumed and produced positions for a specific stream. + * + * Returns 0 on success, < 0 on error. + */ +int lttng_ustconsumer_sample_snapshot_positions( + struct lttng_consumer_stream *stream) +{ + assert(stream); + assert(stream->ustream); + + return ustctl_snapshot_sample_positions(stream->ustream); +} + /* * Get the produced position * @@ -1550,28 +2052,97 @@ int lttng_ustconsumer_get_consumed_snapshot( return ustctl_snapshot_get_consumed(stream->ustream, pos); } +void lttng_ustconsumer_flush_buffer(struct lttng_consumer_stream *stream, + int producer) +{ + assert(stream); + assert(stream->ustream); + + ustctl_flush_buffer(stream->ustream, producer); +} + +int lttng_ustconsumer_get_current_timestamp( + struct lttng_consumer_stream *stream, uint64_t *ts) +{ + assert(stream); + assert(stream->ustream); + assert(ts); + + 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); 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, chan->uid, chan->gid); + if (ret) { + PERROR("unlink %s", shm_path); + } + } + } +} + +void lttng_ustconsumer_free_channel(struct lttng_consumer_channel *chan) +{ + assert(chan); + assert(chan->uchan); + 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); + } + free(chan->stream_fds); } void lttng_ustconsumer_del_stream(struct lttng_consumer_stream *stream) @@ -1585,41 +2156,401 @@ void lttng_ustconsumer_del_stream(struct lttng_consumer_stream *stream) ustctl_destroy_stream(stream->ustream); } +int lttng_ustconsumer_get_wakeup_fd(struct lttng_consumer_stream *stream) +{ + assert(stream); + assert(stream->ustream); + + return ustctl_stream_get_wakeup_fd(stream->ustream); +} + +int lttng_ustconsumer_close_wakeup_fd(struct lttng_consumer_stream *stream) +{ + assert(stream); + assert(stream->ustream); + + 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); + + ret = ustctl_get_instance_id(ustream, &index->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); + if (ret < 0) { + PERROR("ustctl_get_sequence_number"); + goto error; + } + index->packet_seq_num = htobe64(index->packet_seq_num); + +error: + return ret; +} + +static +void metadata_stream_reset_cache(struct lttng_consumer_stream *stream, + struct consumer_metadata_cache *cache) +{ + DBG("Metadata stream update to version %" PRIu64, + cache->version); + stream->ust_metadata_pushed = 0; + stream->metadata_version = cache->version; + stream->reset_metadata_flag = 1; +} + +/* + * Check if the version of the metadata stream and metadata cache match. + * If the cache got updated, reset the metadata stream. + * The stream lock and metadata cache lock MUST be held. + * Return 0 on success, a negative value on error. + */ +static +int metadata_stream_check_version(struct lttng_consumer_stream *stream) +{ + int ret = 0; + struct consumer_metadata_cache *cache = stream->chan->metadata_cache; + + if (cache->version == stream->metadata_version) { + goto end; + } + metadata_stream_reset_cache(stream, cache); + +end: + 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. + */ +static +int commit_one_metadata_packet(struct lttng_consumer_stream *stream) +{ + ssize_t write_len; + int ret; + + pthread_mutex_lock(&stream->chan->metadata_cache->lock); + ret = metadata_stream_check_version(stream); + if (ret < 0) { + goto end; + } + if (stream->chan->metadata_cache->max_offset + == 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->max_offset + - 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; + + assert(stream->chan->metadata_cache->max_offset >= + stream->ust_metadata_pushed); + ret = write_len; + +end: + pthread_mutex_unlock(&stream->chan->metadata_cache->lock); + return ret; +} + + +/* + * Sync metadata meaning request them to the session daemon and snapshot to the + * metadata thread can consumer them. + * + * 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. + */ +int lttng_ustconsumer_sync_metadata(struct lttng_consumer_local_data *ctx, + struct lttng_consumer_stream *metadata) +{ + int ret; + int retry = 0; + + assert(ctx); + assert(metadata); + + pthread_mutex_unlock(&metadata->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); + pthread_mutex_lock(&metadata->lock); + if (ret < 0) { + goto end; + } + + ret = commit_one_metadata_packet(metadata); + if (ret <= 0) { + goto end; + } else if (ret > 0) { + retry = 1; + } + + ustctl_flush_buffer(metadata->ustream, 1); + ret = ustctl_snapshot(metadata->ustream); + 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; + goto end; + } + + /* + * After this flush, we still need to extract metadata. + */ + if (retry) { + ret = EAGAIN; + } + +end: + return ret; +} + +/* + * Return 0 on success else a negative value. + */ +static int notify_if_more_data(struct lttng_consumer_stream *stream, + struct lttng_consumer_local_data *ctx) +{ + int ret; + struct ustctl_consumer_stream *ustream; + + assert(stream); + assert(ctx); + + ustream = stream->ustream; + + /* + * First, we are going to check if there is a new subbuffer available + * before reading the stream wait_fd. + */ + /* Get the next subbuffer */ + ret = ustctl_get_next_subbuf(ustream); + if (ret) { + /* No more data found, flag the stream. */ + stream->has_data = 0; + ret = 0; + goto end; + } + + ret = ustctl_put_subbuf(ustream); + assert(!ret); + + /* This stream still has data. Flag it and wake up the data thread. */ + stream->has_data = 1; + + if (stream->monitor && !stream->hangup_flush_done && !ctx->has_wakeup) { + ssize_t writelen; + + writelen = lttng_pipe_write(ctx->consumer_wakeup_pipe, "!", 1); + if (writelen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) { + ret = writelen; + goto end; + } + + /* The wake up pipe has been notified. */ + ctx->has_wakeup = 1; + } + ret = 0; + +end: + return ret; +} + +static +int update_stream_stats(struct lttng_consumer_stream *stream) +{ + int ret; + uint64_t seq, discarded; + + ret = ustctl_get_sequence_number(stream->ustream, &seq); + if (ret < 0) { + PERROR("ustctl_get_sequence_number"); + goto end; + } + /* + * Start the sequence when we extract the first packet in case we don't + * start at 0 (for example if a consumer is not connected to the + * session immediately after the beginning). + */ + if (stream->last_sequence_number == -1ULL) { + stream->last_sequence_number = seq; + } else if (seq > stream->last_sequence_number) { + stream->chan->lost_packets += seq - + stream->last_sequence_number - 1; + } else { + /* seq <= last_sequence_number */ + ERR("Sequence number inconsistent : prev = %" PRIu64 + ", current = %" PRIu64, + stream->last_sequence_number, seq); + ret = -1; + goto end; + } + stream->last_sequence_number = seq; + + ret = ustctl_get_events_discarded(stream->ustream, &discarded); + if (ret < 0) { + PERROR("kernctl_get_events_discarded"); + goto end; + } + if (discarded < stream->last_discarded_events) { + /* + * Overflow has occurred. We assume only one wrap-around + * has occurred. + */ + stream->chan->discarded_events += + (1ULL << (CAA_BITS_PER_LONG - 1)) - + stream->last_discarded_events + discarded; + } else { + stream->chan->discarded_events += discarded - + stream->last_discarded_events; + } + stream->last_discarded_events = discarded; + ret = 0; + +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) { unsigned long len, subbuf_size, padding; - int err; + int err, write_index = 1; long ret = 0; - char dummy; struct ustctl_consumer_stream *ustream; + struct ctf_packet_index index; assert(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) { + /* + * 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. + */ + if (stream->monitor && !stream->hangup_flush_done && !stream->has_data) { + char dummy; ssize_t readlen; - do { - readlen = read(stream->wait_fd, &dummy, 1); - } while (readlen == -1 && errno == EINTR); - if (readlen == -1) { + 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; + } + ret = err; /* ustctl_get_next_subbuf returns negative, caller expect positive. */ /* * This is a debug message even for single-threaded consumer, @@ -1632,6 +2563,28 @@ int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream *stream, 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) { + err = ustctl_put_subbuf(ustream); + assert(err == 0); + goto end; + } + + /* Update the stream's sequence and discarded events count. */ + ret = update_stream_stats(stream); + if (ret < 0) { + PERROR("kernctl_get_events_discarded"); + err = ustctl_put_subbuf(ustream); + assert(err == 0); + goto end; + } + } else { + write_index = 0; + } + /* Get the full padded subbuffer size */ err = ustctl_get_padded_subbuf_size(ustream, &len); assert(err == 0); @@ -1645,7 +2598,7 @@ int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream *stream, padding = len - subbuf_size; /* write the subbuffer to the tracefile */ - ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size, padding); + 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. @@ -1663,10 +2616,59 @@ int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream *stream, DBG("Error writing to tracefile " "(ret: %ld != len: %lu != subbuf_size: %lu)", ret, len, subbuf_size); + write_index = 0; } 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; + } + } + + /* Write index if needed. */ + if (!write_index) { + goto end; + } + + if (stream->chan->live_timer_interval && !stream->metadata_flag) { + /* + * In live, block until all the metadata is sent. + */ + pthread_mutex_lock(&stream->metadata_timer_lock); + assert(!stream->missed_metadata_flush); + stream->waiting_on_metadata = true; + pthread_mutex_unlock(&stream->metadata_timer_lock); + + err = consumer_stream_sync_metadata(ctx, stream->session_id); + + pthread_mutex_lock(&stream->metadata_timer_lock); + stream->waiting_on_metadata = false; + if (stream->missed_metadata_flush) { + stream->missed_metadata_flush = false; + pthread_mutex_unlock(&stream->metadata_timer_lock); + (void) consumer_flush_ust_index(stream); + } else { + pthread_mutex_unlock(&stream->metadata_timer_lock); + } + + if (err < 0) { + goto end; + } + } + + assert(!stream->metadata_flag); + err = consumer_stream_write_index(stream, &index); + if (err < 0) { + goto end; + } + end: return ret; } @@ -1686,12 +2688,27 @@ int lttng_ustconsumer_on_recv_stream(struct lttng_consumer_stream *stream) 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); + stream->uid, stream->gid, NULL); if (ret < 0) { 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; + } + assert(!stream->index_file); + stream->index_file = index_file; + } } ret = 0; @@ -1716,15 +2733,50 @@ 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) { + uint64_t contiguous, pushed; + + /* Ease our life a bit. */ + contiguous = stream->chan->metadata_cache->max_offset; + pushed = stream->ust_metadata_pushed; + + /* + * 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, + contiguous, pushed); + assert(((int64_t) (contiguous - pushed)) >= 0); + if ((contiguous != pushed) || + (((int64_t) contiguous - pushed) > 0 || contiguous == 0)) { + 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; @@ -1732,6 +2784,45 @@ end: return ret; } +/* + * 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. + */ +void lttng_ustconsumer_close_metadata(struct lttng_consumer_channel *metadata) +{ + int ret; + + assert(metadata); + assert(metadata->type == CONSUMER_CHANNEL_TYPE_METADATA); + + DBG("Closing metadata channel key %" PRIu64, metadata->key); + + if (metadata->switch_timer_enabled == 1) { + consumer_timer_switch_stop(metadata); + } + + if (!metadata->metadata_stream) { + goto end; + } + + /* + * Closing write side so the thread monitoring the stream wakes up if any + * and clean the metadata stream. + */ + if (metadata->metadata_stream->ust_metadata_poll_pipe[1] >= 0) { + ret = close(metadata->metadata_stream->ust_metadata_poll_pipe[1]); + if (ret < 0) { + PERROR("closing metadata pipe write side"); + } + metadata->metadata_stream->ust_metadata_poll_pipe[1] = -1; + } + +end: + return; +} + /* * Close every metadata stream wait fd of the metadata hash table. This * function MUST be used very carefully so not to run into a race between the @@ -1741,9 +2832,8 @@ end: * producer so calling this is safe because we are assured that no state change * can occur in the metadata thread for the streams in the hash table. */ -void lttng_ustconsumer_close_metadata(struct lttng_ht *metadata_ht) +void lttng_ustconsumer_close_all_metadata(struct lttng_ht *metadata_ht) { - int ret; struct lttng_ht_iter iter; struct lttng_consumer_stream *stream; @@ -1755,17 +2845,13 @@ void lttng_ustconsumer_close_metadata(struct lttng_ht *metadata_ht) rcu_read_lock(); cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) { - int fd = stream->wait_fd; - /* - * Whatever happens here we have to continue to try to close every - * streams. Let's report at least the error on failure. - */ - ret = ustctl_stream_close_wakeup_fd(stream->ustream); - if (ret) { - ERR("Unable to close metadata stream fd %d ret %d", fd, ret); - } - DBG("Metadata wait fd %d closed", fd); + health_code_update(); + + pthread_mutex_lock(&stream->chan->lock); + lttng_ustconsumer_close_metadata(stream->chan); + pthread_mutex_unlock(&stream->chan->lock); + } rcu_read_unlock(); } @@ -1780,18 +2866,30 @@ 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. + * + * 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) + struct lttng_consumer_channel *channel, int timer, int wait) { struct lttcomm_metadata_request_msg request; struct lttcomm_consumer_msg msg; - enum lttng_error_code ret_code = LTTNG_OK; - uint64_t len, key, offset; + enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS; + uint64_t len, key, offset, version; int ret; assert(channel); assert(channel->metadata_cache); + memset(&request, 0, sizeof(request)); + /* send the metadata request to sessiond */ switch (consumer_data.type) { case LTTNG_CONSUMER64_UST: @@ -1807,12 +2905,22 @@ 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 the application UID here so the metadata of that application can + * be sent back. The channel UID corresponds to the user UID of the session + * used for the rights on the stream file(s). + */ + request.uid = channel->ust_app_uid; request.key = channel->key; + DBG("Sending metadata request to sessiond, session id %" PRIu64 - ", per-pid %" PRIu64, - channel->session_id, - channel->session_id_per_pid); + ", per-pid %" PRIu64 ", app UID %u and channek key %" PRIu64, + request.session_id, request.session_id_per_pid, request.uid, + request.key); + + pthread_mutex_lock(&ctx->metadata_socket_lock); + + health_code_update(); ret = lttcomm_send_unix_sock(ctx->consumer_metadata_socket, &request, sizeof(request)); @@ -1821,6 +2929,8 @@ int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx, goto end; } + health_code_update(); + /* Receive the metadata from sessiond */ ret = lttcomm_recv_unix_sock(ctx->consumer_metadata_socket, &msg, sizeof(msg)); @@ -1835,6 +2945,8 @@ int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx, goto end; } + health_code_update(); + if (msg.cmd_type == LTTNG_ERR_UND) { /* No registry found */ (void) consumer_send_status_msg(ctx->consumer_metadata_socket, @@ -1850,15 +2962,18 @@ 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) { DBG("No new metadata to receive for key %" PRIu64, key); } + health_code_update(); + /* Tell session daemon we are ready to receive the metadata. */ ret = consumer_send_status_msg(ctx->consumer_metadata_socket, - LTTNG_OK); + LTTCOMM_CONSUMERD_SUCCESS); if (ret < 0 || len == 0) { /* * Somehow, the session daemon is not responding anymore or there is @@ -1867,17 +2982,34 @@ int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx, goto end; } - ret_code = lttng_ustconsumer_recv_metadata(ctx->consumer_metadata_socket, - key, offset, len, channel); - if (ret_code >= 0) { + health_code_update(); + + ret = lttng_ustconsumer_recv_metadata(ctx->consumer_metadata_socket, + key, offset, len, version, channel, timer, wait); + if (ret >= 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); + (void) consumer_send_status_msg(ctx->consumer_metadata_socket, ret); } ret = 0; end: + health_code_update(); + + pthread_mutex_unlock(&ctx->metadata_socket_lock); return ret; } + +/* + * Return the ustctl call for the get stream id. + */ +int lttng_ustconsumer_get_stream_id(struct lttng_consumer_stream *stream, + uint64_t *stream_id) +{ + assert(stream); + assert(stream_id); + + return ustctl_get_stream_id(stream->ustream, stream_id); +}