X-Git-Url: http://git.efficios.com/?a=blobdiff_plain;f=src%2Fcommon%2Fconsumer%2Fconsumer.c;h=c2a2b1587ae6500039dcdddbbedc6eb41fa047b0;hb=c4d7f609b1ea5dde017b93e21647fe9c91a9dfa6;hp=321703a5157305104f855a3834a55ddf81434bc5;hpb=c7a78aabfa0491974d4ffc188d72eb1e67c7344e;p=lttng-tools.git diff --git a/src/common/consumer/consumer.c b/src/common/consumer/consumer.c index 321703a51..c2a2b1587 100644 --- a/src/common/consumer/consumer.c +++ b/src/common/consumer/consumer.c @@ -47,6 +47,7 @@ #include #include #include +#include struct lttng_consumer_global_data consumer_data = { .stream_count = 0, @@ -457,14 +458,13 @@ static void update_endpoint_status_by_netidx(uint64_t net_seq_idx, * If a local data context is available, notify the threads that the streams' * state have changed. */ -static void cleanup_relayd(struct consumer_relayd_sock_pair *relayd, - struct lttng_consumer_local_data *ctx) +void lttng_consumer_cleanup_relayd(struct consumer_relayd_sock_pair *relayd) { uint64_t netidx; assert(relayd); - DBG("Cleaning up relayd sockets"); + DBG("Cleaning up relayd object ID %"PRIu64, relayd->net_seq_idx); /* Save the net sequence index before destroying the object */ netidx = relayd->net_seq_idx; @@ -484,10 +484,8 @@ static void cleanup_relayd(struct consumer_relayd_sock_pair *relayd, * memory barrier ordering the updates of the end point status from the * read of this status which happens AFTER receiving this notify. */ - if (ctx) { - notify_thread_lttng_pipe(ctx->consumer_data_pipe); - notify_thread_lttng_pipe(ctx->consumer_metadata_pipe); - } + notify_thread_lttng_pipe(relayd->ctx->consumer_data_pipe); + notify_thread_lttng_pipe(relayd->ctx->consumer_metadata_pipe); } /* @@ -570,7 +568,7 @@ struct lttng_consumer_stream *consumer_allocate_stream(uint64_t channel_key, stream->session_id = session_id; stream->monitor = monitor; stream->endpoint_status = CONSUMER_ENDPOINT_ACTIVE; - stream->index_fd = -1; + stream->index_file = NULL; stream->last_sequence_number = -1ULL; pthread_mutex_init(&stream->lock, NULL); pthread_mutex_init(&stream->metadata_timer_lock, NULL); @@ -793,10 +791,13 @@ int consumer_send_relayd_stream(struct lttng_consumer_stream *stream, ret = relayd_add_stream(&relayd->control_sock, stream->name, path, &stream->relayd_stream_id, stream->chan->tracefile_size, stream->chan->tracefile_count); - pthread_mutex_unlock(&relayd->ctrl_sock_mutex); if (ret < 0) { + ERR("Relayd add stream failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); + lttng_consumer_cleanup_relayd(relayd); + pthread_mutex_unlock(&relayd->ctrl_sock_mutex); goto end; } + pthread_mutex_unlock(&relayd->ctrl_sock_mutex); uatomic_inc(&relayd->refcount); stream->sent_to_relayd = 1; @@ -834,10 +835,13 @@ int consumer_send_relayd_streams_sent(uint64_t net_seq_idx) /* Add stream on the relayd */ pthread_mutex_lock(&relayd->ctrl_sock_mutex); ret = relayd_streams_sent(&relayd->control_sock); - pthread_mutex_unlock(&relayd->ctrl_sock_mutex); if (ret < 0) { + ERR("Relayd streams sent failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); + lttng_consumer_cleanup_relayd(relayd); + pthread_mutex_unlock(&relayd->ctrl_sock_mutex); goto end; } + pthread_mutex_unlock(&relayd->ctrl_sock_mutex); } else { ERR("Relayd ID %" PRIu64 " unknown. Can't send streams_sent.", net_seq_idx); @@ -1021,7 +1025,7 @@ struct lttng_consumer_channel *consumer_allocate_channel(uint64_t key, CDS_INIT_LIST_HEAD(&channel->streams.head); - DBG("Allocated channel (key %" PRIu64 ")", channel->key) + DBG("Allocated channel (key %" PRIu64 ")", channel->key); end: return channel; @@ -1070,7 +1074,7 @@ int consumer_add_channel(struct lttng_consumer_channel *channel, */ static int update_poll_array(struct lttng_consumer_local_data *ctx, struct pollfd **pollfd, struct lttng_consumer_stream **local_stream, - struct lttng_ht *ht) + struct lttng_ht *ht, int *nb_inactive_fd) { int i = 0; struct lttng_ht_iter iter; @@ -1082,6 +1086,7 @@ static int update_poll_array(struct lttng_consumer_local_data *ctx, assert(local_stream); DBG("Updating poll fd array"); + *nb_inactive_fd = 0; rcu_read_lock(); cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { /* @@ -1092,9 +1097,14 @@ static int update_poll_array(struct lttng_consumer_local_data *ctx, * just after the check. However, this is OK since the stream(s) will * be deleted once the thread is notified that the end point state has * changed where this function will be called back again. + * + * We track the number of inactive FDs because they still need to be + * closed by the polling thread after a wakeup on the data_pipe or + * metadata_pipe. */ if (stream->state != LTTNG_CONSUMER_ACTIVE_STREAM || stream->endpoint_status == CONSUMER_ENDPOINT_INACTIVE) { + (*nb_inactive_fd)++; continue; } /* @@ -1229,6 +1239,11 @@ void lttng_consumer_should_exit(struct lttng_consumer_local_data *ctx) DBG("Consumer flag that it should quit"); } + +/* + * Flush pending writes to trace output disk file. + */ +static void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream, off_t orig_offset) { @@ -1265,8 +1280,9 @@ void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream, */ ret = posix_fadvise(outfd, orig_offset - stream->max_sb_size, stream->max_sb_size, POSIX_FADV_DONTNEED); - if (ret) { - WARN("posix_fadvise() error (%i)", ret); + if (ret && ret != -ENOSYS) { + errno = ret; + PERROR("posix_fadvise on fd %i", outfd); } } @@ -1524,7 +1540,6 @@ ssize_t lttng_consumer_on_read_subbuffer_mmap( mmap_base = stream->mmap_base; ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset); if (ret < 0) { - ret = -errno; PERROR("tracer ctl get_mmap_read_offset"); goto end; } @@ -1619,21 +1634,16 @@ ssize_t lttng_consumer_on_read_subbuffer_mmap( } outfd = stream->out_fd; - if (stream->index_fd >= 0) { - ret = close(stream->index_fd); - if (ret < 0) { - PERROR("Closing index"); - goto end; - } - stream->index_fd = -1; - ret = index_create_file(stream->chan->pathname, + if (stream->index_file) { + lttng_index_file_put(stream->index_file); + stream->index_file = lttng_index_file_create(stream->chan->pathname, stream->name, stream->uid, stream->gid, stream->chan->tracefile_size, - stream->tracefile_count_current); - if (ret < 0) { + stream->tracefile_count_current, + CTF_INDEX_MAJOR, CTF_INDEX_MINOR); + if (!stream->index_file) { goto end; } - stream->index_fd = ret; } /* Reset current size because we just perform a rotation. */ @@ -1686,8 +1696,8 @@ ssize_t lttng_consumer_on_read_subbuffer_mmap( lttng_sync_file_range(outfd, stream->out_fd_offset, len, SYNC_FILE_RANGE_WRITE); stream->out_fd_offset += len; + lttng_consumer_sync_trace_file(stream, orig_offset); } - lttng_consumer_sync_trace_file(stream, orig_offset); write_error: /* @@ -1695,7 +1705,8 @@ write_error: * cleanup the relayd object and all associated streams. */ if (relayd && relayd_hang_up) { - cleanup_relayd(relayd, ctx); + ERR("Relayd hangup. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); + lttng_consumer_cleanup_relayd(relayd); } end: @@ -1826,22 +1837,16 @@ ssize_t lttng_consumer_on_read_subbuffer_splice( } outfd = stream->out_fd; - if (stream->index_fd >= 0) { - ret = close(stream->index_fd); - if (ret < 0) { - PERROR("Closing index"); - goto end; - } - stream->index_fd = -1; - ret = index_create_file(stream->chan->pathname, + if (stream->index_file) { + lttng_index_file_put(stream->index_file); + stream->index_file = lttng_index_file_create(stream->chan->pathname, stream->name, stream->uid, stream->gid, stream->chan->tracefile_size, - stream->tracefile_count_current); - if (ret < 0) { - written = ret; + stream->tracefile_count_current, + CTF_INDEX_MAJOR, CTF_INDEX_MINOR); + if (!stream->index_file) { goto end; } - stream->index_fd = ret; } /* Reset current size because we just perform a rotation. */ @@ -1916,7 +1921,9 @@ ssize_t lttng_consumer_on_read_subbuffer_splice( stream->output_written += ret_splice; written += ret_splice; } - lttng_consumer_sync_trace_file(stream, orig_offset); + if (!relayd) { + lttng_consumer_sync_trace_file(stream, orig_offset); + } goto end; write_error: @@ -1925,7 +1932,8 @@ write_error: * cleanup the relayd object and all associated streams. */ if (relayd && relayd_hang_up) { - cleanup_relayd(relayd, ctx); + ERR("Relayd hangup. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); + lttng_consumer_cleanup_relayd(relayd); /* Skip splice error so the consumer does not fail */ goto end; } @@ -2056,6 +2064,10 @@ void consumer_del_metadata_stream(struct lttng_consumer_stream *stream, pthread_mutex_lock(&consumer_data.lock); pthread_mutex_lock(&stream->chan->lock); pthread_mutex_lock(&stream->lock); + if (stream->chan->metadata_cache) { + /* Only applicable to userspace consumers. */ + pthread_mutex_lock(&stream->chan->metadata_cache->lock); + } /* Remove any reference to that stream. */ consumer_stream_delete(stream, ht); @@ -2079,6 +2091,9 @@ void consumer_del_metadata_stream(struct lttng_consumer_stream *stream, */ stream->chan->metadata_stream = NULL; + if (stream->chan->metadata_cache) { + pthread_mutex_unlock(&stream->chan->metadata_cache->lock); + } pthread_mutex_unlock(&stream->lock); pthread_mutex_unlock(&stream->chan->lock); pthread_mutex_unlock(&consumer_data.lock); @@ -2141,7 +2156,7 @@ int consumer_add_metadata_stream(struct lttng_consumer_stream *stream) lttng_ht_add_unique_u64(ht, &stream->node); - lttng_ht_add_unique_u64(consumer_data.stream_per_chan_id_ht, + lttng_ht_add_u64(consumer_data.stream_per_chan_id_ht, &stream->node_channel_id); /* @@ -2366,7 +2381,7 @@ restart: len = ctx->on_buffer_ready(stream, ctx); /* * We don't check the return value here since if we get - * a negative len, it means an error occured thus we + * a negative len, it means an error occurred thus we * simply remove it from the poll set and free the * stream. */ @@ -2393,7 +2408,7 @@ restart: len = ctx->on_buffer_ready(stream, ctx); /* * We don't check the return value here since if we get - * a negative len, it means an error occured thus we + * a negative len, it means an error occurred thus we * simply remove it from the poll set and free the * stream. */ @@ -2445,6 +2460,8 @@ void *consumer_thread_data_poll(void *data) struct lttng_consumer_stream **local_stream = NULL, *new_stream = NULL; /* local view of consumer_data.fds_count */ int nb_fd = 0; + /* Number of FDs with CONSUMER_ENDPOINT_INACTIVE but still open. */ + int nb_inactive_fd = 0; struct lttng_consumer_local_data *ctx = data; ssize_t len; @@ -2501,7 +2518,7 @@ void *consumer_thread_data_poll(void *data) goto end; } ret = update_poll_array(ctx, &pollfd, local_stream, - data_ht); + data_ht, &nb_inactive_fd); if (ret < 0) { ERR("Error in allocating pollfd or local_outfds"); lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); @@ -2514,7 +2531,7 @@ void *consumer_thread_data_poll(void *data) pthread_mutex_unlock(&consumer_data.lock); /* No FDs and consumer_quit, consumer_cleanup the thread */ - if (nb_fd == 0 && consumer_quit == 1) { + if (nb_fd == 0 && consumer_quit == 1 && nb_inactive_fd == 0) { err = 0; /* All is OK */ goto end; } @@ -3496,6 +3513,7 @@ int consumer_add_relayd_socket(uint64_t net_seq_idx, int sock_type, * Add relayd socket pair to consumer data hashtable. If object already * exists or on error, the function gracefully returns. */ + relayd->ctx = ctx; add_relayd(relayd); /* All good! */ @@ -3621,11 +3639,14 @@ int consumer_data_pending(uint64_t id) pthread_mutex_lock(&relayd->ctrl_sock_mutex); ret = relayd_begin_data_pending(&relayd->control_sock, relayd->relayd_session_id); - pthread_mutex_unlock(&relayd->ctrl_sock_mutex); if (ret < 0) { /* Communication error thus the relayd so no data pending. */ + ERR("Relayd begin data pending failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); + lttng_consumer_cleanup_relayd(relayd); + pthread_mutex_unlock(&relayd->ctrl_sock_mutex); goto data_not_pending; } + pthread_mutex_unlock(&relayd->ctrl_sock_mutex); } cds_lfht_for_each_entry_duplicate(ht->ht, @@ -3665,6 +3686,13 @@ int consumer_data_pending(uint64_t id) stream->relayd_stream_id, stream->next_net_seq_num - 1); } + if (ret < 0) { + ERR("Relayd data pending failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); + lttng_consumer_cleanup_relayd(relayd); + pthread_mutex_unlock(&relayd->ctrl_sock_mutex); + pthread_mutex_unlock(&stream->lock); + goto data_not_pending; + } pthread_mutex_unlock(&relayd->ctrl_sock_mutex); if (ret == 1) { pthread_mutex_unlock(&stream->lock); @@ -3681,10 +3709,13 @@ int consumer_data_pending(uint64_t id) pthread_mutex_lock(&relayd->ctrl_sock_mutex); ret = relayd_end_data_pending(&relayd->control_sock, relayd->relayd_session_id, &is_data_inflight); - pthread_mutex_unlock(&relayd->ctrl_sock_mutex); if (ret < 0) { + ERR("Relayd end data pending failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); + lttng_consumer_cleanup_relayd(relayd); + pthread_mutex_unlock(&relayd->ctrl_sock_mutex); goto data_not_pending; } + pthread_mutex_unlock(&relayd->ctrl_sock_mutex); if (is_data_inflight) { goto data_pending; }