X-Git-Url: http://git.efficios.com/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fcommon%2Fkernel-consumer%2Fkernel-consumer.c;h=e06f3b3e079aae44f494ba6d4409f02120018af4;hp=061123d61160d32016249aa9856b8f0e4a43b5c6;hb=6f9449c22eef59294cf1e1dc3610a5cbf14baec0;hpb=2336629e66d2d2a7bdf85523b32f87aa49d05238 diff --git a/src/common/kernel-consumer/kernel-consumer.c b/src/common/kernel-consumer/kernel-consumer.c index 061123d61..e06f3b3e0 100644 --- a/src/common/kernel-consumer/kernel-consumer.c +++ b/src/common/kernel-consumer/kernel-consumer.c @@ -1,22 +1,13 @@ /* - * Copyright (C) 2011 - Julien Desfossez - * Mathieu Desnoyers + * Copyright (C) 2011 Julien Desfossez + * Copyright (C) 2011 Mathieu Desnoyers + * Copyright (C) 2017 Jérémie Galarneau * - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License, version 2 only, - * as published by the Free Software Foundation. + * SPDX-License-Identifier: GPL-2.0-only * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License along - * with this program; if not, write to the Free Software Foundation, Inc., - * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. */ -#define _GNU_SOURCE +#define _LGPL_SOURCE #include #include #include @@ -35,18 +26,22 @@ #include #include #include +#include #include #include #include -#include +#include #include -#include +#include +#include +#include +#include +#include #include "kernel-consumer.h" extern struct lttng_consumer_global_data consumer_data; extern int consumer_poll_timeout; -extern volatile int consumer_quit; /* * Take a snapshot for a specific fd @@ -59,14 +54,30 @@ int lttng_kconsumer_take_snapshot(struct lttng_consumer_stream *stream) int infd = stream->wait_fd; ret = kernctl_snapshot(infd); - if (ret != 0) { - perror("Getting sub-buffer snapshot."); - ret = -errno; + /* + * -EAGAIN is not an error, it just means that there is no data to + * be read. + */ + if (ret != 0 && ret != -EAGAIN) { + PERROR("Getting sub-buffer snapshot."); } return ret; } +/* + * Sample consumed and produced positions for a specific fd. + * + * Returns 0 on success, < 0 on error. + */ +int lttng_kconsumer_sample_snapshot_positions( + struct lttng_consumer_stream *stream) +{ + assert(stream); + + return kernctl_snapshot_sample_positions(stream->wait_fd); +} + /* * Get the produced position * @@ -80,8 +91,7 @@ int lttng_kconsumer_get_produced_snapshot(struct lttng_consumer_stream *stream, ret = kernctl_snapshot_get_produced(infd, pos); if (ret != 0) { - perror("kernctl_snapshot_get_produced"); - ret = -errno; + PERROR("kernctl_snapshot_get_produced"); } return ret; @@ -100,46 +110,61 @@ int lttng_kconsumer_get_consumed_snapshot(struct lttng_consumer_stream *stream, ret = kernctl_snapshot_get_consumed(infd, pos); if (ret != 0) { - perror("kernctl_snapshot_get_consumed"); - ret = -errno; + PERROR("kernctl_snapshot_get_consumed"); } return ret; } +static +int get_current_subbuf_addr(struct lttng_consumer_stream *stream, + const char **addr) +{ + int ret; + unsigned long mmap_offset; + const char *mmap_base = stream->mmap_base; + + ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset); + if (ret < 0) { + PERROR("Failed to get mmap read offset"); + goto error; + } + + *addr = mmap_base + mmap_offset; +error: + return ret; +} + /* * Take a snapshot of all the stream of a channel + * RCU read-side lock must be held across this function to ensure existence of + * channel. The channel lock must be held by the caller. * * Returns 0 on success, < 0 on error */ -int lttng_kconsumer_snapshot_channel(uint64_t key, char *path, - uint64_t relayd_id, uint64_t max_stream_size, +static int lttng_kconsumer_snapshot_channel( + struct lttng_consumer_channel *channel, + uint64_t key, char *path, uint64_t relayd_id, + uint64_t nb_packets_per_stream, struct lttng_consumer_local_data *ctx) { int ret; - unsigned long consumed_pos, produced_pos; - struct lttng_consumer_channel *channel; struct lttng_consumer_stream *stream; DBG("Kernel consumer snapshot channel %" PRIu64, key); rcu_read_lock(); - channel = consumer_find_channel(key); - if (!channel) { - ERR("No channel found for key %" PRIu64, key); - ret = -1; - goto end; - } - /* Splice is not supported yet for channel snapshot. */ if (channel->output != CONSUMER_CHANNEL_MMAP) { - ERR("Unsupported output %d", channel->output); + ERR("Unsupported output type for channel \"%s\": mmap output is required to record a snapshot", + channel->name); ret = -1; goto end; } cds_list_for_each_entry(stream, &channel->streams.head, send_node) { + unsigned long consumed_pos, produced_pos; health_code_update(); @@ -148,6 +173,19 @@ int lttng_kconsumer_snapshot_channel(uint64_t key, char *path, */ pthread_mutex_lock(&stream->lock); + assert(channel->trace_chunk); + if (!lttng_trace_chunk_get(channel->trace_chunk)) { + /* + * Can't happen barring an internal error as the channel + * holds a reference to the trace chunk. + */ + ERR("Failed to acquire reference to channel's trace chunk"); + ret = -1; + goto end_unlock; + } + assert(!stream->trace_chunk); + stream->trace_chunk = channel->trace_chunk; + /* * Assign the received relayd ID so we can use it for streaming. The streams * are not visible to anyone so this is OK to change it. @@ -161,33 +199,29 @@ int lttng_kconsumer_snapshot_channel(uint64_t key, char *path, goto end_unlock; } } else { - ret = utils_create_stream_file(path, stream->name, - stream->chan->tracefile_size, - stream->tracefile_count_current, - stream->uid, stream->gid, NULL); + ret = consumer_stream_create_output_files(stream, + false); if (ret < 0) { - ERR("utils_create_stream_file"); goto end_unlock; } - - stream->out_fd = ret; - stream->tracefile_size_current = 0; - - DBG("Kernel consumer snapshot stream %s/%s (%" PRIu64 ")", - path, stream->name, stream->key); + DBG("Kernel consumer snapshot stream (%" PRIu64 ")", + stream->key); } - if (relayd_id != -1ULL) { - ret = consumer_send_relayd_streams_sent(relayd_id); + + ret = kernctl_buffer_flush_empty(stream->wait_fd); + if (ret < 0) { + /* + * Doing a buffer flush which does not take into + * account empty packets. This is not perfect + * for stream intersection, but required as a + * fall-back when "flush_empty" is not + * implemented by lttng-modules. + */ + ret = kernctl_buffer_flush(stream->wait_fd); if (ret < 0) { - ERR("sending streams sent to relayd"); + ERR("Failed to flush kernel stream"); goto end_unlock; } - } - - ret = kernctl_buffer_flush(stream->wait_fd); - if (ret < 0) { - ERR("Failed to flush kernel stream"); - ret = -errno; goto end_unlock; } @@ -209,61 +243,53 @@ int lttng_kconsumer_snapshot_channel(uint64_t key, char *path, goto end_unlock; } - if (stream->max_sb_size == 0) { - ret = kernctl_get_max_subbuf_size(stream->wait_fd, - &stream->max_sb_size); - if (ret < 0) { - ERR("Getting kernel max_sb_size"); - ret = -errno; - goto end_unlock; - } - } + consumed_pos = consumer_get_consume_start_pos(consumed_pos, + produced_pos, nb_packets_per_stream, + stream->max_sb_size); - /* - * 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) { + while ((long) (consumed_pos - produced_pos) < 0) { ssize_t read_len; unsigned long len, padded_len; + const char *subbuf_addr; + struct lttng_buffer_view subbuf_view; health_code_update(); - DBG("Kernel consumer taking snapshot at pos %lu", consumed_pos); ret = kernctl_get_subbuf(stream->wait_fd, &consumed_pos); if (ret < 0) { - if (errno != EAGAIN) { + if (ret != -EAGAIN) { PERROR("kernctl_get_subbuf snapshot"); - ret = -errno; goto end_unlock; } DBG("Kernel consumer get subbuf failed. Skipping it."); consumed_pos += stream->max_sb_size; + stream->chan->lost_packets++; continue; } ret = kernctl_get_subbuf_size(stream->wait_fd, &len); if (ret < 0) { ERR("Snapshot kernctl_get_subbuf_size"); - ret = -errno; goto error_put_subbuf; } ret = kernctl_get_padded_subbuf_size(stream->wait_fd, &padded_len); if (ret < 0) { ERR("Snapshot kernctl_get_padded_subbuf_size"); - ret = -errno; goto error_put_subbuf; } - read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len, - padded_len - len, NULL); + ret = get_current_subbuf_addr(stream, &subbuf_addr); + if (ret) { + goto error_put_subbuf; + } + + subbuf_view = lttng_buffer_view_init( + subbuf_addr, 0, padded_len); + read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, + stream, &subbuf_view, + padded_len - len); /* * We write the padded len in local tracefiles but the data len * when using a relay. Display the error but continue processing @@ -284,7 +310,6 @@ int lttng_kconsumer_snapshot_channel(uint64_t key, char *path, ret = kernctl_put_subbuf(stream->wait_fd); if (ret < 0) { ERR("Snapshot kernctl_put_subbuf"); - ret = -errno; goto end_unlock; } consumed_pos += stream->max_sb_size; @@ -303,6 +328,8 @@ int lttng_kconsumer_snapshot_channel(uint64_t key, char *path, close_relayd_stream(stream); stream->net_seq_idx = (uint64_t) -1ULL; } + lttng_trace_chunk_put(stream->trace_chunk); + stream->trace_chunk = NULL; pthread_mutex_unlock(&stream->lock); } @@ -313,7 +340,6 @@ int lttng_kconsumer_snapshot_channel(uint64_t key, char *path, error_put_subbuf: ret = kernctl_put_subbuf(stream->wait_fd); if (ret < 0) { - ret = -errno; ERR("Snapshot kernctl_put_subbuf error path"); } end_unlock: @@ -325,15 +351,18 @@ end: /* * Read the whole metadata available for a snapshot. + * RCU read-side lock must be held across this function to ensure existence of + * metadata_channel. The channel lock must be held by the caller. * * Returns 0 on success, < 0 on error */ -int lttng_kconsumer_snapshot_metadata(uint64_t key, char *path, - uint64_t relayd_id, struct lttng_consumer_local_data *ctx) +static int lttng_kconsumer_snapshot_metadata( + struct lttng_consumer_channel *metadata_channel, + uint64_t key, char *path, uint64_t relayd_id, + struct lttng_consumer_local_data *ctx) { int ret, use_relayd = 0; ssize_t ret_read; - struct lttng_consumer_channel *metadata_channel; struct lttng_consumer_stream *metadata_stream; assert(ctx); @@ -343,16 +372,13 @@ int lttng_kconsumer_snapshot_metadata(uint64_t key, char *path, rcu_read_lock(); - metadata_channel = consumer_find_channel(key); - if (!metadata_channel) { - ERR("Kernel snapshot metadata not found for key %" PRIu64, key); - ret = -1; - goto error; - } - metadata_stream = metadata_channel->metadata_stream; assert(metadata_stream); + pthread_mutex_lock(&metadata_stream->lock); + assert(metadata_channel->trace_chunk); + assert(metadata_stream->trace_chunk); + /* Flag once that we have a valid relayd for the stream. */ if (relayd_id != (uint64_t) -1ULL) { use_relayd = 1; @@ -361,28 +387,26 @@ int lttng_kconsumer_snapshot_metadata(uint64_t key, char *path, if (use_relayd) { ret = consumer_send_relayd_stream(metadata_stream, path); if (ret < 0) { - goto error; + goto error_snapshot; } } else { - ret = utils_create_stream_file(path, metadata_stream->name, - metadata_stream->chan->tracefile_size, - metadata_stream->tracefile_count_current, - metadata_stream->uid, metadata_stream->gid, NULL); + ret = consumer_stream_create_output_files(metadata_stream, + false); if (ret < 0) { - goto error; + goto error_snapshot; } - metadata_stream->out_fd = ret; } do { health_code_update(); - ret_read = lttng_kconsumer_read_subbuffer(metadata_stream, ctx); + ret_read = lttng_consumer_read_subbuffer(metadata_stream, ctx, true); if (ret_read < 0) { if (ret_read != -EAGAIN) { ERR("Kernel snapshot reading metadata subbuffer (ret: %zd)", ret_read); - goto error; + ret = ret_read; + goto error_snapshot; } /* ret_read is negative at this point so we will exit the loop. */ continue; @@ -403,15 +427,17 @@ int lttng_kconsumer_snapshot_metadata(uint64_t key, char *path, */ } metadata_stream->out_fd = -1; + lttng_trace_chunk_put(metadata_stream->trace_chunk); + metadata_stream->trace_chunk = NULL; } } ret = 0; - +error_snapshot: + pthread_mutex_unlock(&metadata_stream->lock); cds_list_del(&metadata_stream->send_node); consumer_stream_destroy(metadata_stream, NULL); metadata_channel->metadata_stream = NULL; -error: rcu_read_unlock(); return ret; } @@ -441,17 +467,8 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, health_code_update(); - 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; - } + /* Deprecated command */ + assert(msg.cmd_type != LTTNG_CONSUMER_STOP); health_code_update(); @@ -462,16 +479,17 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, case LTTNG_CONSUMER_ADD_RELAYD_SOCKET: { /* Session daemon status message are handled in the following call. */ - ret = consumer_add_relayd_socket(msg.u.relayd_sock.net_index, + consumer_add_relayd_socket(msg.u.relayd_sock.net_index, msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll, &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id, - msg.u.relayd_sock.relayd_session_id); + msg.u.relayd_sock.relayd_session_id); goto end_nosignal; } case LTTNG_CONSUMER_ADD_CHANNEL: { struct lttng_consumer_channel *new_channel; int ret_recv; + const uint64_t chunk_id = msg.u.channel.chunk_id.value; health_code_update(); @@ -486,13 +504,18 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, DBG("consumer_add_channel %" PRIu64, msg.u.channel.channel_key); new_channel = consumer_allocate_channel(msg.u.channel.channel_key, - msg.u.channel.session_id, msg.u.channel.pathname, - msg.u.channel.name, msg.u.channel.uid, msg.u.channel.gid, + msg.u.channel.session_id, + msg.u.channel.chunk_id.is_set ? + &chunk_id : NULL, + msg.u.channel.pathname, + msg.u.channel.name, msg.u.channel.relayd_id, msg.u.channel.output, msg.u.channel.tracefile_size, msg.u.channel.tracefile_count, 0, msg.u.channel.monitor, - msg.u.channel.live_timer_interval); + msg.u.channel.live_timer_interval, + msg.u.channel.is_live, + NULL, NULL); if (new_channel == NULL) { lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR); goto end_nosignal; @@ -533,9 +556,20 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, } else { ret = consumer_add_channel(new_channel, ctx); } - if (CONSUMER_CHANNEL_TYPE_DATA) { + if (msg.u.channel.type == CONSUMER_CHANNEL_TYPE_DATA && !ret) { + int monitor_start_ret; + + DBG("Consumer starting monitor timer"); consumer_timer_live_start(new_channel, msg.u.channel.live_timer_interval); + monitor_start_ret = consumer_timer_monitor_start( + new_channel, + msg.u.channel.monitor_timer_interval); + if (monitor_start_ret < 0) { + ERR("Starting channel monitoring timer failed"); + goto end_nosignal; + } + } health_code_update(); @@ -570,7 +604,7 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, * happens while tearing down. */ ERR("Unable to find channel key %" PRIu64, msg.u.stream.channel_key); - ret_code = LTTNG_ERR_KERN_CHAN_NOT_FOUND; + ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND; } health_code_update(); @@ -579,23 +613,22 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, ret = consumer_send_status_msg(sock, ret_code); if (ret < 0) { /* Somehow, the session daemon is not responding anymore. */ - goto error_fatal; + goto error_add_stream_fatal; } health_code_update(); if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) { /* Channel was not found. */ - goto end_nosignal; + goto error_add_stream_nosignal; } /* Blocking call */ health_poll_entry(); ret = lttng_consumer_poll_socket(consumer_sockpoll); health_poll_exit(); - if (ret < 0) { - rcu_read_unlock(); - return -EINTR; + if (ret) { + goto error_add_stream_fatal; } health_code_update(); @@ -604,8 +637,7 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1); if (ret != sizeof(fd)) { lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD); - rcu_read_unlock(); - return ret; + goto end; } health_code_update(); @@ -618,19 +650,20 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, ret = consumer_send_status_msg(sock, ret_code); if (ret < 0) { /* Somehow, the session daemon is not responding anymore. */ - goto end_nosignal; + goto error_add_stream_nosignal; } health_code_update(); - new_stream = consumer_allocate_stream(channel->key, + pthread_mutex_lock(&channel->lock); + new_stream = consumer_stream_create( + channel, + channel->key, fd, - LTTNG_CONSUMER_ACTIVE_STREAM, channel->name, - channel->uid, - channel->gid, channel->relayd_id, channel->session_id, + channel->trace_chunk, msg.u.stream.cpu, &alloc_ret, channel->type, @@ -643,23 +676,22 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR); break; } - goto end_nosignal; + pthread_mutex_unlock(&channel->lock); + goto error_add_stream_nosignal; } - new_stream->chan = channel; new_stream->wait_fd = fd; - switch (channel->output) { - case CONSUMER_CHANNEL_SPLICE: - new_stream->output = LTTNG_EVENT_SPLICE; - break; - case CONSUMER_CHANNEL_MMAP: - new_stream->output = LTTNG_EVENT_MMAP; - break; - default: - ERR("Stream output unknown %d", channel->output); - goto end_nosignal; + ret = kernctl_get_max_subbuf_size(new_stream->wait_fd, + &new_stream->max_sb_size); + if (ret < 0) { + pthread_mutex_unlock(&channel->lock); + ERR("Failed to get kernel maximal subbuffer size"); + goto error_add_stream_nosignal; } + consumer_stream_update_channel_attributes(new_stream, + channel); + /* * We've just assigned the channel to the stream so increment the * refcount right now. We don't need to increment the refcount for @@ -682,14 +714,16 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, health_code_update(); + pthread_mutex_lock(&new_stream->lock); if (ctx->on_recv_stream) { ret = ctx->on_recv_stream(new_stream); if (ret < 0) { + pthread_mutex_unlock(&new_stream->lock); + pthread_mutex_unlock(&channel->lock); consumer_stream_free(new_stream); - goto end_nosignal; + goto error_add_stream_nosignal; } } - health_code_update(); if (new_stream->metadata_flag) { @@ -702,7 +736,9 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, "relayd id %" PRIu64, new_stream->name, new_stream->net_seq_idx); cds_list_add(&new_stream->send_node, &channel->streams.head); - break; + pthread_mutex_unlock(&new_stream->lock); + pthread_mutex_unlock(&channel->lock); + goto end_add_stream; } /* Send stream to relayd if the stream has an ID. */ @@ -710,33 +746,40 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, ret = consumer_send_relayd_stream(new_stream, new_stream->chan->pathname); if (ret < 0) { + pthread_mutex_unlock(&new_stream->lock); + pthread_mutex_unlock(&channel->lock); consumer_stream_free(new_stream); - goto end_nosignal; + goto error_add_stream_nosignal; + } + + /* + * If adding an extra stream to an already + * existing channel (e.g. cpu hotplug), we need + * to send the "streams_sent" command to relayd. + */ + if (channel->streams_sent_to_relayd) { + ret = consumer_send_relayd_streams_sent( + new_stream->net_seq_idx); + if (ret < 0) { + pthread_mutex_unlock(&new_stream->lock); + pthread_mutex_unlock(&channel->lock); + goto error_add_stream_nosignal; + } } } + pthread_mutex_unlock(&new_stream->lock); + pthread_mutex_unlock(&channel->lock); /* Get the right pipe where the stream will be sent. */ if (new_stream->metadata_flag) { - ret = consumer_add_metadata_stream(new_stream); - if (ret) { - ERR("Consumer add metadata stream %" PRIu64 " failed. Continuing", - new_stream->key); - consumer_stream_free(new_stream); - goto end_nosignal; - } + consumer_add_metadata_stream(new_stream); stream_pipe = ctx->consumer_metadata_pipe; } else { - ret = consumer_add_data_stream(new_stream); - if (ret) { - ERR("Consumer add stream %" PRIu64 " failed. Continuing", - new_stream->key); - consumer_stream_free(new_stream); - goto end_nosignal; - } + consumer_add_data_stream(new_stream); stream_pipe = ctx->consumer_data_pipe; } - /* Vitible to other threads */ + /* Visible to other threads */ new_stream->globally_visible = 1; health_code_update(); @@ -751,12 +794,17 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, } else { consumer_del_stream_for_data(new_stream); } - goto end_nosignal; + goto error_add_stream_nosignal; } - DBG("Kernel consumer ADD_STREAM %s (fd: %d) with relayd id %" PRIu64, - new_stream->name, fd, new_stream->relayd_stream_id); + DBG("Kernel consumer ADD_STREAM %s (fd: %d) %s with relayd id %" PRIu64, + new_stream->name, fd, new_stream->chan->pathname, new_stream->relayd_stream_id); +end_add_stream: break; +error_add_stream_nosignal: + goto end_nosignal; +error_add_stream_fatal: + goto error_fatal; } case LTTNG_CONSUMER_STREAMS_SENT: { @@ -774,7 +822,7 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, */ ERR("Unable to find channel key %" PRIu64, msg.u.sent_streams.channel_key); - ret_code = LTTNG_ERR_KERN_CHAN_NOT_FOUND; + ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND; } health_code_update(); @@ -783,9 +831,9 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, * Send status code to session daemon. */ ret = consumer_send_status_msg(sock, ret_code); - if (ret < 0) { + if (ret < 0 || ret_code != LTTCOMM_CONSUMERD_SUCCESS) { /* Somehow, the session daemon is not responding anymore. */ - goto end_nosignal; + goto error_streams_sent_nosignal; } health_code_update(); @@ -795,7 +843,7 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, * streams in this channel. */ if (!channel->monitor) { - break; + goto end_error_streams_sent; } health_code_update(); @@ -804,10 +852,14 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, ret = consumer_send_relayd_streams_sent( msg.u.sent_streams.net_seq_idx); if (ret < 0) { - goto end_nosignal; + goto error_streams_sent_nosignal; } + channel->streams_sent_to_relayd = true; } +end_error_streams_sent: break; +error_streams_sent_nosignal: + goto end_nosignal; } case LTTNG_CONSUMER_UPDATE_STREAM: { @@ -825,7 +877,7 @@ int lttng_kconsumer_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; } /* @@ -878,26 +930,36 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, } case LTTNG_CONSUMER_SNAPSHOT_CHANNEL: { - if (msg.u.snapshot_channel.metadata == 1) { - ret = lttng_kconsumer_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_KERN_META_FAIL; - } + struct lttng_consumer_channel *channel; + uint64_t key = msg.u.snapshot_channel.key; + + channel = consumer_find_channel(key); + if (!channel) { + ERR("Channel %" PRIu64 " not found", key); + ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND; } else { - ret = lttng_kconsumer_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_KERN_CHAN_FAIL; + pthread_mutex_lock(&channel->lock); + if (msg.u.snapshot_channel.metadata == 1) { + ret = lttng_kconsumer_snapshot_metadata(channel, key, + msg.u.snapshot_channel.pathname, + msg.u.snapshot_channel.relayd_id, ctx); + if (ret < 0) { + ERR("Snapshot metadata failed"); + ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED; + } + } else { + ret = lttng_kconsumer_snapshot_channel(channel, key, + msg.u.snapshot_channel.pathname, + msg.u.snapshot_channel.relayd_id, + msg.u.snapshot_channel.nb_packets_per_stream, + ctx); + if (ret < 0) { + ERR("Snapshot channel failed"); + ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED; + } } + pthread_mutex_unlock(&channel->lock); } - health_code_update(); ret = consumer_send_status_msg(sock, ret_code); @@ -915,7 +977,7 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, channel = consumer_find_channel(key); if (!channel) { ERR("Kernel consumer destroy channel %" PRIu64 " not found", key); - ret_code = LTTNG_ERR_KERN_CHAN_NOT_FOUND; + ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND; } health_code_update(); @@ -923,11 +985,16 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, ret = consumer_send_status_msg(sock, ret_code); if (ret < 0) { /* Somehow, the session daemon is not responding anymore. */ - goto end_nosignal; + goto end_destroy_channel; } health_code_update(); + /* Stop right now if no channel was found. */ + if (!channel) { + goto end_destroy_channel; + } + /* * This command should ONLY be issued for channel with streams set in * no monitor mode. @@ -941,83 +1008,340 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, assert(!uatomic_sub_return(&channel->refcount, 1)); consumer_del_channel(channel); - +end_destroy_channel: goto end_nosignal; } - default: - goto end_nosignal; + case LTTNG_CONSUMER_DISCARDED_EVENTS: + { + ssize_t ret; + uint64_t count; + struct lttng_consumer_channel *channel; + uint64_t id = msg.u.discarded_events.session_id; + uint64_t key = msg.u.discarded_events.channel_key; + + DBG("Kernel consumer discarded events command for session id %" + PRIu64 ", channel key %" PRIu64, id, key); + + channel = consumer_find_channel(key); + if (!channel) { + ERR("Kernel consumer discarded events channel %" + PRIu64 " not found", key); + count = 0; + } else { + count = channel->discarded_events; + } + + health_code_update(); + + /* Send back returned value to session daemon */ + ret = lttcomm_send_unix_sock(sock, &count, sizeof(count)); + if (ret < 0) { + PERROR("send discarded events"); + goto error_fatal; + } + + break; } + case LTTNG_CONSUMER_LOST_PACKETS: + { + ssize_t ret; + uint64_t count; + struct lttng_consumer_channel *channel; + uint64_t id = msg.u.lost_packets.session_id; + uint64_t key = msg.u.lost_packets.channel_key; -end_nosignal: - rcu_read_unlock(); + DBG("Kernel consumer lost packets command for session id %" + PRIu64 ", channel key %" PRIu64, id, key); - /* - * Return 1 to indicate success since the 0 value can be a socket - * shutdown during the recv() or send() call. - */ - health_code_update(); - return 1; + channel = consumer_find_channel(key); + if (!channel) { + ERR("Kernel consumer lost packets channel %" + PRIu64 " not found", key); + count = 0; + } else { + count = channel->lost_packets; + } -error_fatal: - rcu_read_unlock(); - /* This will issue a consumer stop. */ - return -1; -} + health_code_update(); -/* - * Populate index values of a kernel 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, int infd) -{ - int ret; + /* Send back returned value to session daemon */ + ret = lttcomm_send_unix_sock(sock, &count, sizeof(count)); + if (ret < 0) { + PERROR("send lost packets"); + goto error_fatal; + } - ret = kernctl_get_timestamp_begin(infd, &index->timestamp_begin); - if (ret < 0) { - PERROR("kernctl_get_timestamp_begin"); - goto error; + break; } - index->timestamp_begin = htobe64(index->timestamp_begin); + case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE: + { + int channel_monitor_pipe; - ret = kernctl_get_timestamp_end(infd, &index->timestamp_end); - if (ret < 0) { - PERROR("kernctl_get_timestamp_end"); - goto error; + 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; + } + ret = consumer_send_status_msg(sock, ret_code); + if (ret < 0) { + goto error_fatal; + } + break; } - index->timestamp_end = htobe64(index->timestamp_end); + case LTTNG_CONSUMER_ROTATE_CHANNEL: + { + struct lttng_consumer_channel *channel; + uint64_t key = msg.u.rotate_channel.key; - ret = kernctl_get_events_discarded(infd, &index->events_discarded); - if (ret < 0) { - PERROR("kernctl_get_events_discarded"); - goto error; + DBG("Consumer rotate channel %" PRIu64, key); + + channel = consumer_find_channel(key); + if (!channel) { + ERR("Channel %" PRIu64 " not found", key); + ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND; + } else { + /* + * Sample the rotate position of all the streams in this channel. + */ + ret = lttng_consumer_rotate_channel(channel, key, + msg.u.rotate_channel.relayd_id, + msg.u.rotate_channel.metadata, + ctx); + if (ret < 0) { + ERR("Rotate channel failed"); + ret_code = LTTCOMM_CONSUMERD_ROTATION_FAIL; + } + + health_code_update(); + } + ret = consumer_send_status_msg(sock, ret_code); + if (ret < 0) { + /* Somehow, the session daemon is not responding anymore. */ + goto error_rotate_channel; + } + if (channel) { + /* Rotate the streams that are ready right now. */ + ret = lttng_consumer_rotate_ready_streams( + channel, key, ctx); + if (ret < 0) { + ERR("Rotate ready streams failed"); + } + } + break; +error_rotate_channel: + goto end_nosignal; } - index->events_discarded = htobe64(index->events_discarded); + case LTTNG_CONSUMER_CLEAR_CHANNEL: + { + struct lttng_consumer_channel *channel; + uint64_t key = msg.u.clear_channel.key; - ret = kernctl_get_content_size(infd, &index->content_size); - if (ret < 0) { - PERROR("kernctl_get_content_size"); - goto error; + channel = consumer_find_channel(key); + if (!channel) { + DBG("Channel %" PRIu64 " not found", key); + ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND; + } else { + ret = lttng_consumer_clear_channel(channel); + if (ret) { + ERR("Clear channel failed"); + ret_code = ret; + } + + health_code_update(); + } + ret = consumer_send_status_msg(sock, ret_code); + if (ret < 0) { + /* Somehow, the session daemon is not responding anymore. */ + goto end_nosignal; + } + + break; + } + case LTTNG_CONSUMER_INIT: + { + ret_code = lttng_consumer_init_command(ctx, + msg.u.init.sessiond_uuid); + health_code_update(); + ret = consumer_send_status_msg(sock, ret_code); + if (ret < 0) { + /* Somehow, the session daemon is not responding anymore. */ + goto end_nosignal; + } + break; } - index->content_size = htobe64(index->content_size); + case LTTNG_CONSUMER_CREATE_TRACE_CHUNK: + { + const struct lttng_credentials credentials = { + .uid = msg.u.create_trace_chunk.credentials.value.uid, + .gid = msg.u.create_trace_chunk.credentials.value.gid, + }; + const bool is_local_trace = + !msg.u.create_trace_chunk.relayd_id.is_set; + const uint64_t relayd_id = + msg.u.create_trace_chunk.relayd_id.value; + const char *chunk_override_name = + *msg.u.create_trace_chunk.override_name ? + msg.u.create_trace_chunk.override_name : + NULL; + struct lttng_directory_handle *chunk_directory_handle = NULL; - ret = kernctl_get_packet_size(infd, &index->packet_size); - if (ret < 0) { - PERROR("kernctl_get_packet_size"); - goto error; + /* + * The session daemon will only provide a chunk directory file + * descriptor for local traces. + */ + if (is_local_trace) { + int chunk_dirfd; + + /* Acnowledge the reception of the command. */ + ret = consumer_send_status_msg(sock, + LTTCOMM_CONSUMERD_SUCCESS); + if (ret < 0) { + /* Somehow, the session daemon is not responding anymore. */ + goto end_nosignal; + } + + ret = lttcomm_recv_fds_unix_sock(sock, &chunk_dirfd, 1); + if (ret != sizeof(chunk_dirfd)) { + ERR("Failed to receive trace chunk directory file descriptor"); + goto error_fatal; + } + + DBG("Received trace chunk directory fd (%d)", + chunk_dirfd); + chunk_directory_handle = lttng_directory_handle_create_from_dirfd( + chunk_dirfd); + if (!chunk_directory_handle) { + ERR("Failed to initialize chunk directory handle from directory file descriptor"); + if (close(chunk_dirfd)) { + PERROR("Failed to close chunk directory file descriptor"); + } + goto error_fatal; + } + } + + ret_code = lttng_consumer_create_trace_chunk( + !is_local_trace ? &relayd_id : NULL, + msg.u.create_trace_chunk.session_id, + msg.u.create_trace_chunk.chunk_id, + (time_t) msg.u.create_trace_chunk + .creation_timestamp, + chunk_override_name, + msg.u.create_trace_chunk.credentials.is_set ? + &credentials : + NULL, + chunk_directory_handle); + lttng_directory_handle_put(chunk_directory_handle); + goto end_msg_sessiond; + } + case LTTNG_CONSUMER_CLOSE_TRACE_CHUNK: + { + enum lttng_trace_chunk_command_type close_command = + msg.u.close_trace_chunk.close_command.value; + const uint64_t relayd_id = + msg.u.close_trace_chunk.relayd_id.value; + struct lttcomm_consumer_close_trace_chunk_reply reply; + char path[LTTNG_PATH_MAX]; + + ret_code = lttng_consumer_close_trace_chunk( + msg.u.close_trace_chunk.relayd_id.is_set ? + &relayd_id : + NULL, + msg.u.close_trace_chunk.session_id, + msg.u.close_trace_chunk.chunk_id, + (time_t) msg.u.close_trace_chunk.close_timestamp, + msg.u.close_trace_chunk.close_command.is_set ? + &close_command : + NULL, path); + reply.ret_code = ret_code; + reply.path_length = strlen(path) + 1; + ret = lttcomm_send_unix_sock(sock, &reply, sizeof(reply)); + if (ret != sizeof(reply)) { + goto error_fatal; + } + ret = lttcomm_send_unix_sock(sock, path, reply.path_length); + if (ret != reply.path_length) { + goto error_fatal; + } + goto end_nosignal; + } + case LTTNG_CONSUMER_TRACE_CHUNK_EXISTS: + { + const uint64_t relayd_id = + msg.u.trace_chunk_exists.relayd_id.value; + + ret_code = lttng_consumer_trace_chunk_exists( + msg.u.trace_chunk_exists.relayd_id.is_set ? + &relayd_id : NULL, + msg.u.trace_chunk_exists.session_id, + msg.u.trace_chunk_exists.chunk_id); + goto end_msg_sessiond; + } + default: + goto end_nosignal; } - index->packet_size = htobe64(index->packet_size); - ret = kernctl_get_stream_id(infd, &index->stream_id); +end_nosignal: + /* + * Return 1 to indicate success since the 0 value can be a socket + * shutdown during the recv() or send() call. + */ + ret = 1; + goto end; +error_fatal: + /* This will issue a consumer stop. */ + ret = -1; + goto end; +end_msg_sessiond: + /* + * The returned value here is not useful since either way we'll return 1 to + * the caller because the session daemon socket management is done + * elsewhere. Returning a negative code or 0 will shutdown the consumer. + */ + ret = consumer_send_status_msg(sock, ret_code); if (ret < 0) { - PERROR("kernctl_get_stream_id"); - goto error; + goto error_fatal; } - index->stream_id = htobe64(index->stream_id); - -error: + ret = 1; +end: + health_code_update(); + rcu_read_unlock(); return ret; } + /* * Sync metadata meaning request them to the session daemon and snapshot to the * metadata thread can consumer them. @@ -1041,7 +1365,7 @@ int lttng_kconsumer_sync_metadata(struct lttng_consumer_stream *metadata) ret = kernctl_snapshot(metadata->wait_fd); if (ret < 0) { - if (errno != EAGAIN) { + if (ret != -EAGAIN) { ERR("Sync metadata, taking kernel snapshot failed."); goto end; } @@ -1055,153 +1379,227 @@ end: return ret; } -/* - * Consume data on a file descriptor and write it on a trace file. - */ -ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream, - struct lttng_consumer_local_data *ctx) +static +int extract_common_subbuffer_info(struct lttng_consumer_stream *stream, + struct stream_subbuffer *subbuf) { - unsigned long len, subbuf_size, padding; - int err, write_index = 1; - ssize_t ret = 0; - int infd = stream->wait_fd; - struct ctf_packet_index index; + int ret; - DBG("In read_subbuffer (infd : %d)", infd); + ret = kernctl_get_subbuf_size( + stream->wait_fd, &subbuf->info.data.subbuf_size); + if (ret) { + goto end; + } - /* Get the next subbuffer */ - err = kernctl_get_next_subbuf(infd); - if (err != 0) { - /* - * This is a debug message even for single-threaded consumer, - * because poll() have more relaxed criterions than get subbuf, - * so get_subbuf may fail for short race windows where poll() - * would issue wakeups. - */ - DBG("Reserving sub buffer failed (everything is normal, " - "it is due to concurrency)"); - ret = -errno; + ret = kernctl_get_padded_subbuf_size( + stream->wait_fd, &subbuf->info.data.padded_subbuf_size); + if (ret) { + goto end; + } + +end: + return ret; +} + +static +int extract_metadata_subbuffer_info(struct lttng_consumer_stream *stream, + struct stream_subbuffer *subbuf) +{ + int ret; + + ret = extract_common_subbuffer_info(stream, subbuf); + if (ret) { goto end; } - /* Get the full subbuffer size including padding */ - err = kernctl_get_padded_subbuf_size(infd, &len); - if (err != 0) { - perror("Getting sub-buffer len failed."); - ret = -errno; + ret = kernctl_get_metadata_version( + stream->wait_fd, &subbuf->info.metadata.version); + if (ret) { goto end; } - if (!stream->metadata_flag) { - ret = get_index_values(&index, infd); - if (ret < 0) { +end: + return ret; +} + +static +int extract_data_subbuffer_info(struct lttng_consumer_stream *stream, + struct stream_subbuffer *subbuf) +{ + int ret; + + ret = extract_common_subbuffer_info(stream, subbuf); + if (ret) { + goto end; + } + + ret = kernctl_get_packet_size( + stream->wait_fd, &subbuf->info.data.packet_size); + if (ret < 0) { + PERROR("Failed to get sub-buffer packet size"); + goto end; + } + + ret = kernctl_get_content_size( + stream->wait_fd, &subbuf->info.data.content_size); + if (ret < 0) { + PERROR("Failed to get sub-buffer content size"); + goto end; + } + + ret = kernctl_get_timestamp_begin( + stream->wait_fd, &subbuf->info.data.timestamp_begin); + if (ret < 0) { + PERROR("Failed to get sub-buffer begin timestamp"); + goto end; + } + + ret = kernctl_get_timestamp_end( + stream->wait_fd, &subbuf->info.data.timestamp_end); + if (ret < 0) { + PERROR("Failed to get sub-buffer end timestamp"); + goto end; + } + + ret = kernctl_get_events_discarded( + stream->wait_fd, &subbuf->info.data.events_discarded); + if (ret) { + PERROR("Failed to get sub-buffer events discarded count"); + goto end; + } + + ret = kernctl_get_sequence_number(stream->wait_fd, + &subbuf->info.data.sequence_number.value); + if (ret) { + /* May not be supported by older LTTng-modules. */ + if (ret != -ENOTTY) { + PERROR("Failed to get sub-buffer sequence number"); goto end; } } else { - write_index = 0; + subbuf->info.data.sequence_number.is_set = true; } - switch (stream->chan->output) { - case CONSUMER_CHANNEL_SPLICE: - /* - * XXX: The lttng-modules splice "actor" does not handle copying - * partial pages hence only using the subbuffer size without the - * padding makes the splice fail. - */ - subbuf_size = len; - padding = 0; + ret = kernctl_get_stream_id( + stream->wait_fd, &subbuf->info.data.stream_id); + if (ret < 0) { + PERROR("Failed to get stream id"); + goto end; + } - /* splice the subbuffer to the tracefile */ - ret = lttng_consumer_on_read_subbuffer_splice(ctx, stream, subbuf_size, - padding, &index); - /* - * XXX: Splice does not support network streaming so the return value - * is simply checked against subbuf_size and not like the mmap() op. - */ - if (ret != subbuf_size) { - /* - * display the error but continue processing to try - * to release the subbuffer - */ - ERR("Error splicing to tracefile (ret: %zd != len: %lu)", - ret, subbuf_size); - write_index = 0; - } - break; - case CONSUMER_CHANNEL_MMAP: - /* Get subbuffer size without padding */ - err = kernctl_get_subbuf_size(infd, &subbuf_size); - if (err != 0) { - perror("Getting sub-buffer len failed."); - ret = -errno; + ret = kernctl_get_instance_id(stream->wait_fd, + &subbuf->info.data.stream_instance_id.value); + if (ret) { + /* May not be supported by older LTTng-modules. */ + if (ret != -ENOTTY) { + PERROR("Failed to get stream instance id"); goto end; } + } else { + subbuf->info.data.stream_instance_id.is_set = true; + } +end: + return ret; +} + +static +int get_subbuffer_common(struct lttng_consumer_stream *stream, + struct stream_subbuffer *subbuffer) +{ + int ret; + + ret = kernctl_get_next_subbuf(stream->wait_fd); + if (ret) { + goto end; + } - /* Make sure the tracer is not gone mad on us! */ - assert(len >= subbuf_size); + ret = stream->read_subbuffer_ops.extract_subbuffer_info( + stream, subbuffer); +end: + return ret; +} - padding = len - subbuf_size; +static +int get_next_subbuffer_splice(struct lttng_consumer_stream *stream, + struct stream_subbuffer *subbuffer) +{ + int ret; - /* write the subbuffer to the tracefile */ - ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size, - padding, &index); - /* - * The mmap operation should write subbuf_size amount of data when - * network streaming or the full padding (len) size when we are _not_ - * streaming. - */ - if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) || - (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) { - /* - * Display the error but continue processing to try to release the - * subbuffer. This is a DBG statement since this is possible to - * happen without being a critical error. - */ - DBG("Error writing to tracefile " - "(ret: %zd != len: %lu != subbuf_size: %lu)", - ret, len, subbuf_size); - write_index = 0; - } - break; - default: - ERR("Unknown output method"); - ret = -EPERM; + ret = get_subbuffer_common(stream, subbuffer); + if (ret) { + goto end; } - err = kernctl_put_next_subbuf(infd); - if (err != 0) { - if (errno == EFAULT) { - perror("Error in unreserving sub buffer\n"); - } else if (errno == EIO) { - /* Should never happen with newer LTTng versions */ - perror("Reader has been pushed by the writer, last sub-buffer corrupted."); - } - ret = -errno; + subbuffer->buffer.fd = stream->wait_fd; +end: + return ret; +} + +static +int get_next_subbuffer_mmap(struct lttng_consumer_stream *stream, + struct stream_subbuffer *subbuffer) +{ + int ret; + const char *addr; + + ret = get_subbuffer_common(stream, subbuffer); + if (ret) { goto end; } - /* Write index if needed. */ - if (!write_index) { + ret = get_current_subbuf_addr(stream, &addr); + if (ret) { goto end; } - if (stream->chan->live_timer_interval && !stream->metadata_flag) { - /* - * In live, block until all the metadata is sent. - */ - err = consumer_stream_sync_metadata(ctx, stream->session_id); - if (err < 0) { - goto end; + subbuffer->buffer.buffer = lttng_buffer_view_init( + addr, 0, subbuffer->info.data.padded_subbuf_size); +end: + return ret; +} + +static +int put_next_subbuffer(struct lttng_consumer_stream *stream, + struct stream_subbuffer *subbuffer) +{ + const int ret = kernctl_put_next_subbuf(stream->wait_fd); + + if (ret) { + if (ret == -EFAULT) { + PERROR("Error in unreserving sub buffer"); + } else if (ret == -EIO) { + /* Should never happen with newer LTTng versions */ + PERROR("Reader has been pushed by the writer, last sub-buffer corrupted"); } } - err = consumer_stream_write_index(stream, &index); - if (err < 0) { - goto end; + return ret; +} + +static void lttng_kconsumer_set_stream_ops( + struct lttng_consumer_stream *stream) +{ + if (stream->chan->output == CONSUMER_CHANNEL_MMAP) { + stream->read_subbuffer_ops.get_next_subbuffer = + get_next_subbuffer_mmap; + } else { + stream->read_subbuffer_ops.get_next_subbuffer = + get_next_subbuffer_splice; } -end: - return ret; + if (stream->metadata_flag) { + stream->read_subbuffer_ops.extract_subbuffer_info = + extract_metadata_subbuffer_info; + } else { + stream->read_subbuffer_ops.extract_subbuffer_info = + extract_data_subbuffer_info; + if (stream->chan->is_live) { + stream->read_subbuffer_ops.send_live_beacon = + consumer_flush_kernel_index; + } + } + + stream->read_subbuffer_ops.put_next_subbuffer = put_next_subbuffer; } int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream) @@ -1211,29 +1609,15 @@ int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream) assert(stream); /* - * Don't create anything if this is set for streaming or should not be - * monitored. + * Don't create anything if this is set for streaming or if there is + * no current trace chunk on the parent channel. */ - if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor) { - ret = utils_create_stream_file(stream->chan->pathname, stream->name, - stream->chan->tracefile_size, stream->tracefile_count_current, - stream->uid, stream->gid, NULL); - if (ret < 0) { + if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor && + stream->chan->trace_chunk) { + ret = consumer_stream_create_output_files(stream, true); + if (ret) { goto error; } - stream->out_fd = ret; - stream->tracefile_size_current = 0; - - if (!stream->metadata_flag) { - ret = index_create_file(stream->chan->pathname, - stream->name, stream->uid, stream->gid, - stream->chan->tracefile_size, - stream->tracefile_count_current); - if (ret < 0) { - goto error; - } - stream->index_fd = ret; - } } if (stream->output == LTTNG_EVENT_MMAP) { @@ -1243,7 +1627,6 @@ int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream) ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len); if (ret != 0) { PERROR("kernctl_get_mmap_len"); - ret = -errno; goto error_close_fd; } stream->mmap_len = (size_t) mmap_len; @@ -1257,6 +1640,8 @@ int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream) } } + lttng_kconsumer_set_stream_ops(stream); + /* we return 0 to let the library handle the FD internally */ return 0;