X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fcommon%2Fkernel-consumer%2Fkernel-consumer.c;h=196da6633353a44ea6f73821494642e0cd7539a3;hp=a68438253c9d694a6d29d7267b1da50548ad4172;hb=a9d36096fd1bfbeb7c55a5cfc97523512237f0d4;hpb=29decac336fd0d72ae00105b7e3d0cdece2ccd25 diff --git a/src/common/kernel-consumer/kernel-consumer.c b/src/common/kernel-consumer/kernel-consumer.c index a68438253..196da6633 100644 --- a/src/common/kernel-consumer/kernel-consumer.c +++ b/src/common/kernel-consumer/kernel-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 @@ -29,21 +30,25 @@ #include #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 @@ -56,14 +61,30 @@ int lttng_kconsumer_take_snapshot(struct lttng_consumer_stream *stream) int infd = stream->wait_fd; ret = kernctl_snapshot(infd); - if (ret != 0) { - errno = -ret; - perror("Getting sub-buffer snapshot."); + /* + * -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 * @@ -77,8 +98,7 @@ int lttng_kconsumer_get_produced_snapshot(struct lttng_consumer_stream *stream, ret = kernctl_snapshot_get_produced(infd, pos); if (ret != 0) { - errno = -ret; - perror("kernctl_snapshot_get_produced"); + PERROR("kernctl_snapshot_get_produced"); } return ret; @@ -97,113 +117,63 @@ int lttng_kconsumer_get_consumed_snapshot(struct lttng_consumer_stream *stream, ret = kernctl_snapshot_get_consumed(infd, pos); if (ret != 0) { - errno = -ret; - perror("kernctl_snapshot_get_consumed"); - } - - return ret; -} - -/* - * Find a relayd and send the stream - * - * Returns 0 on success, < 0 on error - */ -static -int send_relayd_stream(struct lttng_consumer_stream *stream, char *path) -{ - struct consumer_relayd_sock_pair *relayd; - int ret = 0; - char *stream_path; - - if (path != NULL) { - stream_path = path; - } else { - stream_path = stream->chan->pathname; - } - /* The stream is not metadata. Get relayd reference if exists. */ - rcu_read_lock(); - relayd = consumer_find_relayd(stream->net_seq_idx); - if (relayd != NULL) { - /* Add stream on the relayd */ - pthread_mutex_lock(&relayd->ctrl_sock_mutex); - ret = relayd_add_stream(&relayd->control_sock, - stream->name, stream_path, - &stream->relayd_stream_id, - stream->chan->tracefile_size, - stream->chan->tracefile_count); - pthread_mutex_unlock(&relayd->ctrl_sock_mutex); - if (ret < 0) { - goto end; - } - uatomic_inc(&relayd->refcount); - } else if (stream->net_seq_idx != (uint64_t) -1ULL) { - ERR("Network sequence index %" PRIu64 " unknown. Not adding stream.", - stream->net_seq_idx); - ret = -1; - goto end; + PERROR("kernctl_snapshot_get_consumed"); } -end: - rcu_read_unlock(); return ret; } -/* - * Find a relayd and close the stream - */ -static -void close_relayd_stream(struct lttng_consumer_stream *stream) -{ - struct consumer_relayd_sock_pair *relayd; - - /* The stream is not metadata. Get relayd reference if exists. */ - rcu_read_lock(); - relayd = consumer_find_relayd(stream->net_seq_idx); - if (relayd != NULL) { - consumer_stream_relayd_close(stream, relayd); - } - rcu_read_unlock(); -} - /* * 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, struct lttng_consumer_local_data *ctx) +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 %lu", key); + DBG("Kernel consumer snapshot channel %" PRIu64, key); rcu_read_lock(); - channel = consumer_find_channel(key); - if (!channel) { - ERR("No channel found for key %lu", 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->stream_no_monitor_list.head, - no_monitor_node) { + cds_list_for_each_entry(stream, &channel->streams.head, send_node) { + unsigned long consumed_pos, produced_pos; + + health_code_update(); + /* * Lock stream because we are about to change its state. */ pthread_mutex_lock(&stream->lock); + assert(channel->trace_chunk); + if (!lttng_trace_chunk_get(channel->trace_chunk)) { + /* + * Can't happen barring an internal error as the channel + * holds a reference to the trace chunk. + */ + ERR("Failed to acquire reference to channel's trace chunk"); + ret = -1; + goto 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. @@ -211,31 +181,35 @@ int lttng_kconsumer_snapshot_channel(uint64_t key, char *path, stream->net_seq_idx = relayd_id; channel->relayd_id = relayd_id; if (relayd_id != (uint64_t) -1ULL) { - ret = send_relayd_stream(stream, path); + ret = consumer_send_relayd_stream(stream, path); if (ret < 0) { ERR("sending stream to relayd"); goto end_unlock; } - DBG("Stream %s sent to the relayd", stream->name); } else { - ret = utils_create_stream_file(path, stream->name, - stream->chan->tracefile_size, stream->tracefile_count_current, - stream->uid, stream->gid); + 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 (%lu)", path, - stream->name, stream->key); + DBG("Kernel consumer snapshot stream (%" PRIu64 ")", + stream->key); } - ret = kernctl_buffer_flush(stream->wait_fd); + ret = kernctl_buffer_flush_empty(stream->wait_fd); if (ret < 0) { - ERR("Failed to flush kernel metadata stream"); + /* + * 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("Failed to flush kernel stream"); + goto end_unlock; + } goto end_unlock; } @@ -257,29 +231,27 @@ 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"); - goto end_unlock; - } - } + consumed_pos = consumer_get_consume_start_pos(consumed_pos, + produced_pos, nb_packets_per_stream, + stream->max_sb_size); - while (consumed_pos < produced_pos) { + while ((long) (consumed_pos - produced_pos) < 0) { ssize_t read_len; unsigned long len, padded_len; + 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"); goto end_unlock; } DBG("Kernel consumer get subbuf failed. Skipping it."); consumed_pos += stream->max_sb_size; + stream->chan->lost_packets++; continue; } @@ -296,7 +268,7 @@ int lttng_kconsumer_snapshot_channel(uint64_t key, char *path, } read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len, - padded_len - len); + padded_len - len, NULL); /* * We write the padded len in local tracefiles but the data len * when using a relay. Display the error but continue processing @@ -323,16 +295,20 @@ int lttng_kconsumer_snapshot_channel(uint64_t key, char *path, } if (relayd_id == (uint64_t) -1ULL) { - ret = close(stream->out_fd); - if (ret < 0) { - PERROR("Kernel consumer snapshot close out_fd"); - goto end_unlock; + if (stream->out_fd >= 0) { + ret = close(stream->out_fd); + if (ret < 0) { + PERROR("Kernel consumer snapshot close out_fd"); + goto end_unlock; + } + stream->out_fd = -1; } - stream->out_fd = -1; } else { 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); } @@ -354,75 +330,93 @@ 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) { - struct lttng_consumer_channel *metadata_channel; + int ret, use_relayd = 0; + ssize_t ret_read; struct lttng_consumer_stream *metadata_stream; - int ret; + + assert(ctx); DBG("Kernel consumer snapshot metadata with key %" PRIu64 " at path %s", key, path); rcu_read_lock(); - metadata_channel = consumer_find_channel(key); - if (!metadata_channel) { - ERR("Snapshot kernel metadata channel not found for key %lu", key); - ret = -1; - goto end; - } - 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) { - ret = send_relayd_stream(metadata_stream, path); + use_relayd = 1; + } + + if (use_relayd) { + ret = consumer_send_relayd_stream(metadata_stream, path); if (ret < 0) { - ERR("sending stream to relayd"); + goto error_snapshot; } - DBG("Stream %s sent to the relayd", metadata_stream->name); } 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); + ret = consumer_stream_create_output_files(metadata_stream, + false); if (ret < 0) { - goto end; + goto error_snapshot; } - metadata_stream->out_fd = ret; } - ret = 0; - while (ret >= 0) { - ret = lttng_kconsumer_read_subbuffer(metadata_stream, ctx); - if (ret < 0) { - if (ret != -EPERM) { - ERR("Kernel snapshot reading subbuffer"); - goto end; + do { + health_code_update(); + + ret_read = lttng_kconsumer_read_subbuffer(metadata_stream, ctx); + if (ret_read < 0) { + if (ret_read != -EAGAIN) { + ERR("Kernel snapshot reading metadata subbuffer (ret: %zd)", + ret_read); + ret = ret_read; + goto error_snapshot; } - /* "ret" is negative at this point so we will exit the loop. */ + /* ret_read is negative at this point so we will exit the loop. */ continue; } - } + } while (ret_read >= 0); - if (relayd_id == (uint64_t) -1ULL) { - ret = close(metadata_stream->out_fd); - if (ret < 0) { - PERROR("Kernel consumer snapshot close out_fd"); - goto end; - } - metadata_stream->out_fd = -1; - } else { + if (use_relayd) { close_relayd_stream(metadata_stream); metadata_stream->net_seq_idx = (uint64_t) -1ULL; + } else { + if (metadata_stream->out_fd >= 0) { + ret = close(metadata_stream->out_fd); + if (ret < 0) { + PERROR("Kernel consumer snapshot metadata close out_fd"); + /* + * Don't go on error here since the snapshot was successful at this + * point but somehow the close failed. + */ + } + metadata_stream->out_fd = -1; + lttng_trace_chunk_put(metadata_stream->trace_chunk); + metadata_stream->trace_chunk = NULL; + } } ret = 0; -end: +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; rcu_read_unlock(); return ret; } @@ -436,28 +430,26 @@ int lttng_kconsumer_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; + health_code_update(); + ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg)); if (ret != sizeof(msg)) { - lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD); if (ret > 0) { + lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD); ret = -1; } 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 command */ + assert(msg.cmd_type != LTTNG_CONSUMER_STOP); + + health_code_update(); /* relayd needs RCU read-side protection */ rcu_read_lock(); @@ -466,15 +458,19 @@ 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.sock, msg.u.relayd_sock.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(); /* First send a status message before receiving the fds. */ ret = consumer_send_status_msg(sock, ret_code); @@ -482,14 +478,22 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, /* Somehow, the session daemon is not responding anymore. */ goto error_fatal; } + + health_code_update(); + 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.monitor, + msg.u.channel.live_timer_interval, + NULL, NULL); if (new_channel == NULL) { lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR); goto end_nosignal; @@ -518,6 +522,8 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, goto end_nosignal; }; + health_code_update(); + if (ctx->on_recv_channel != NULL) { ret_recv = ctx->on_recv_channel(new_channel); if (ret_recv == 0) { @@ -528,6 +534,23 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, } else { ret = consumer_add_channel(new_channel, ctx); } + 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(); /* If we received an error in add_channel, we need to report it. */ if (ret < 0) { @@ -559,39 +582,44 @@ 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(); + /* First send a status message before receiving the fds. */ ret = consumer_send_status_msg(sock, ret_code); if (ret < 0) { - /* - * Somehow, the session daemon is not responding - * anymore. - */ - goto error_fatal; + /* Somehow, the session daemon is not responding anymore. */ + goto error_add_stream_fatal; } - if (ret_code != LTTNG_OK) { - /* - * Channel was not found. - */ - goto end_nosignal; + + health_code_update(); + + if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) { + /* Channel was not found. */ + goto error_add_stream_nosignal; } - /* block */ - if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) { - rcu_read_unlock(); - return -EINTR; + /* Blocking call */ + health_poll_entry(); + ret = lttng_consumer_poll_socket(consumer_sockpoll); + health_poll_exit(); + if (ret) { + goto error_add_stream_fatal; } + health_code_update(); + /* Get stream file descriptor from socket */ 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(); + /* * Send status code to session daemon only if the recv works. If the * above recv() failed, the session daemon is notified through the @@ -600,20 +628,22 @@ 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 end_nosignal; + goto error_add_stream_nosignal; } + health_code_update(); + + pthread_mutex_lock(&channel->lock); new_stream = consumer_allocate_stream(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); + channel->type, + channel->monitor); if (new_stream == NULL) { switch (alloc_ret) { case -ENOMEM: @@ -622,20 +652,38 @@ 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; + 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); switch (channel->output) { case CONSUMER_CHANNEL_SPLICE: new_stream->output = LTTNG_EVENT_SPLICE; + ret = utils_create_pipe(new_stream->splice_pipe); + if (ret < 0) { + pthread_mutex_unlock(&channel->lock); + goto error_add_stream_nosignal; + } break; case CONSUMER_CHANNEL_MMAP: new_stream->output = LTTNG_EVENT_MMAP; break; default: ERR("Stream output unknown %d", channel->output); - goto end_nosignal; + pthread_mutex_unlock(&channel->lock); + goto error_add_stream_nosignal; } /* @@ -658,13 +706,19 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, */ new_stream->hangup_flush_done = 0; + health_code_update(); + + pthread_mutex_lock(&new_stream->lock); if (ctx->on_recv_stream) { ret = ctx->on_recv_stream(new_stream); if (ret < 0) { - consumer_del_stream(new_stream, NULL); - goto end_nosignal; + pthread_mutex_unlock(&new_stream->lock); + pthread_mutex_unlock(&channel->lock); + consumer_stream_free(new_stream); + goto error_add_stream_nosignal; } } + health_code_update(); if (new_stream->metadata_flag) { channel->metadata_stream = new_stream; @@ -675,36 +729,131 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, DBG("Kernel consumer add stream %s in no monitor mode with " "relayd id %" PRIu64, new_stream->name, new_stream->net_seq_idx); - cds_list_add(&new_stream->no_monitor_node, - &channel->stream_no_monitor_list.head); - break; + cds_list_add(&new_stream->send_node, &channel->streams.head); + pthread_mutex_unlock(&new_stream->lock); + pthread_mutex_unlock(&channel->lock); + goto end_add_stream; } - ret = send_relayd_stream(new_stream, NULL); - if (ret < 0) { - consumer_del_stream(new_stream, NULL); - goto end_nosignal; + /* Send stream to relayd if the stream has an ID. */ + if (new_stream->net_seq_idx != (uint64_t) -1ULL) { + 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 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) { + consumer_add_metadata_stream(new_stream); stream_pipe = ctx->consumer_metadata_pipe; } else { + consumer_add_data_stream(new_stream); stream_pipe = ctx->consumer_data_pipe; } + /* Visible to other threads */ + new_stream->globally_visible = 1; + + health_code_update(); + ret = lttng_pipe_write(stream_pipe, &new_stream, sizeof(new_stream)); if (ret < 0) { ERR("Consumer write %s stream to pipe %d", new_stream->metadata_flag ? "metadata" : "data", lttng_pipe_get_writefd(stream_pipe)); - consumer_del_stream(new_stream, NULL); - goto end_nosignal; + if (new_stream->metadata_flag) { + consumer_del_stream_for_metadata(new_stream); + } else { + consumer_del_stream_for_data(new_stream); + } + goto error_add_stream_nosignal; + } + + 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: + { + struct lttng_consumer_channel *channel; + + /* + * Get stream's channel reference. Needed when adding the stream to the + * global hash table. + */ + channel = consumer_find_channel(msg.u.sent_streams.channel_key); + if (!channel) { + /* + * We could not find the channel. Can happen if cpu hotplug + * happens while tearing down. + */ + ERR("Unable to find channel key %" PRIu64, + msg.u.sent_streams.channel_key); + ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND; + } + + health_code_update(); + + /* + * Send status code to session daemon. + */ + ret = consumer_send_status_msg(sock, ret_code); + if (ret < 0 || ret_code != LTTCOMM_CONSUMERD_SUCCESS) { + /* Somehow, the session daemon is not responding anymore. */ + goto error_streams_sent_nosignal; + } + + health_code_update(); + + /* + * We should not send this message if we don't monitor the + * streams in this channel. + */ + if (!channel->monitor) { + goto end_error_streams_sent; } - DBG("Kernel consumer ADD_STREAM %s (fd: %d) with relayd id %" PRIu64, - new_stream->name, fd, new_stream->relayd_stream_id); + health_code_update(); + /* Send stream to relayd if the stream has an ID. */ + if (msg.u.sent_streams.net_seq_idx != (uint64_t) -1ULL) { + ret = consumer_send_relayd_streams_sent( + msg.u.sent_streams.net_seq_idx); + if (ret < 0) { + 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: { @@ -722,7 +871,7 @@ 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; } /* @@ -739,6 +888,8 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, consumer_flag_relayd_for_destroy(relayd); } + health_code_update(); + ret = consumer_send_status_msg(sock, ret_code); if (ret < 0) { /* Somehow, the session daemon is not responding anymore. */ @@ -756,6 +907,8 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, ret = consumer_data_pending(id); + health_code_update(); + /* Send back returned value to session daemon */ ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret)); if (ret < 0) { @@ -771,23 +924,37 @@ 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, 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); if (ret < 0) { @@ -804,13 +971,22 @@ 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(); + 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; } /* @@ -826,62 +1002,636 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, assert(!uatomic_sub_return(&channel->refcount, 1)); consumer_del_channel(channel); - - goto end_nosignal; - } - default: +end_destroy_channel: 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; -end_nosignal: - rcu_read_unlock(); + DBG("Kernel consumer discarded events 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. - */ - return 1; + channel = consumer_find_channel(key); + if (!channel) { + ERR("Kernel consumer discarded events channel %" + PRIu64 " not found", key); + count = 0; + } else { + count = channel->discarded_events; + } -error_fatal: - rcu_read_unlock(); - /* This will issue a consumer stop. */ - return -1; -} + health_code_update(); -/* - * 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) -{ - unsigned long len, subbuf_size, padding; - int err; - ssize_t ret = 0; - int infd = stream->wait_fd; + /* 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; + } - DBG("In read_subbuffer (infd : %d)", infd); - /* Get the next subbuffer */ - err = kernctl_get_next_subbuf(infd); - if (err != 0) { - ret = err; - /* - * 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)"); - goto end; + 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; + + DBG("Kernel consumer lost packets command for session id %" + PRIu64 ", channel key %" PRIu64, id, key); + + channel = consumer_find_channel(key); + if (!channel) { + ERR("Kernel consumer lost packets channel %" + PRIu64 " not found", key); + count = 0; + } else { + count = channel->lost_packets; + } + + health_code_update(); + + /* 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; + } + + 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; + } + ret = consumer_send_status_msg(sock, ret_code); + if (ret < 0) { + goto error_fatal; + } + break; + } + case LTTNG_CONSUMER_ROTATE_CHANNEL: + { + struct lttng_consumer_channel *channel; + uint64_t key = msg.u.rotate_channel.key; + + 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 end_nosignal; + } + 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; + } + case LTTNG_CONSUMER_INIT: + { + ret_code = lttng_consumer_init_command(ctx, + msg.u.init.sessiond_uuid); + health_code_update(); + ret = consumer_send_status_msg(sock, ret_code); + if (ret < 0) { + /* Somehow, the session daemon is not responding anymore. */ + goto end_nosignal; + } + break; + } + case LTTNG_CONSUMER_CREATE_TRACE_CHUNK: + { + const struct lttng_credentials credentials = { + .uid = msg.u.create_trace_chunk.credentials.value.uid, + .gid = msg.u.create_trace_chunk.credentials.value.gid, + }; + const bool is_local_trace = + !msg.u.create_trace_chunk.relayd_id.is_set; + const uint64_t relayd_id = + msg.u.create_trace_chunk.relayd_id.value; + const char *chunk_override_name = + *msg.u.create_trace_chunk.override_name ? + msg.u.create_trace_chunk.override_name : + NULL; + LTTNG_OPTIONAL(struct lttng_directory_handle) chunk_directory_handle = + LTTNG_OPTIONAL_INIT; + + /* + * 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); + ret = lttng_directory_handle_init_from_dirfd( + &chunk_directory_handle.value, + chunk_dirfd); + if (ret) { + 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; + } + chunk_directory_handle.is_set = true; + } + + 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.is_set ? + &chunk_directory_handle.value : + NULL); + + if (chunk_directory_handle.is_set) { + lttng_directory_handle_fini( + &chunk_directory_handle.value); + } + 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; + + 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); + goto end_msg_sessiond; + } + 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; + } + +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) { + goto error_fatal; + } + ret = 1; +end: + health_code_update(); + rcu_read_unlock(); + return ret; +} + +/* + * 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; + + ret = kernctl_get_timestamp_begin(infd, &index->timestamp_begin); + if (ret < 0) { + PERROR("kernctl_get_timestamp_begin"); + goto error; + } + index->timestamp_begin = htobe64(index->timestamp_begin); + + ret = kernctl_get_timestamp_end(infd, &index->timestamp_end); + if (ret < 0) { + PERROR("kernctl_get_timestamp_end"); + goto error; + } + index->timestamp_end = htobe64(index->timestamp_end); + + ret = kernctl_get_events_discarded(infd, &index->events_discarded); + if (ret < 0) { + PERROR("kernctl_get_events_discarded"); + goto error; + } + index->events_discarded = htobe64(index->events_discarded); + + ret = kernctl_get_content_size(infd, &index->content_size); + if (ret < 0) { + PERROR("kernctl_get_content_size"); + goto error; + } + index->content_size = htobe64(index->content_size); + + ret = kernctl_get_packet_size(infd, &index->packet_size); + if (ret < 0) { + PERROR("kernctl_get_packet_size"); + goto error; + } + index->packet_size = htobe64(index->packet_size); + + ret = kernctl_get_stream_id(infd, &index->stream_id); + if (ret < 0) { + PERROR("kernctl_get_stream_id"); + goto error; + } + index->stream_id = htobe64(index->stream_id); + + ret = kernctl_get_instance_id(infd, &index->stream_instance_id); + if (ret < 0) { + if (ret == -ENOTTY) { + /* Command not implemented by lttng-modules. */ + index->stream_instance_id = -1ULL; + } else { + PERROR("kernctl_get_instance_id"); + goto error; + } + } + index->stream_instance_id = htobe64(index->stream_instance_id); + + ret = kernctl_get_sequence_number(infd, &index->packet_seq_num); + if (ret < 0) { + if (ret == -ENOTTY) { + /* Command not implemented by lttng-modules. */ + index->packet_seq_num = -1ULL; + ret = 0; + } else { + PERROR("kernctl_get_sequence_number"); + goto error; + } + } + index->packet_seq_num = htobe64(index->packet_seq_num); + +error: + return ret; +} +/* + * Sync metadata meaning request them to the session daemon and snapshot to the + * metadata thread can consumer them. + * + * Metadata stream lock MUST be acquired. + * + * Return 0 if new metadatda is available, EAGAIN if the metadata stream + * is empty or a negative value on error. + */ +int lttng_kconsumer_sync_metadata(struct lttng_consumer_stream *metadata) +{ + int ret; + + assert(metadata); + + ret = kernctl_buffer_flush(metadata->wait_fd); + if (ret < 0) { + ERR("Failed to flush kernel stream"); + goto end; + } + + ret = kernctl_snapshot(metadata->wait_fd); + if (ret < 0) { + if (ret != -EAGAIN) { + ERR("Sync metadata, taking kernel snapshot failed."); + goto end; + } + DBG("Sync metadata, no new kernel metadata"); + /* No new metadata, exit. */ + ret = ENODATA; + goto end; + } + +end: + return ret; +} + +static +int update_stream_stats(struct lttng_consumer_stream *stream) +{ + int ret; + uint64_t seq, discarded; + + ret = kernctl_get_sequence_number(stream->wait_fd, &seq); + if (ret < 0) { + if (ret == -ENOTTY) { + /* Command not implemented by lttng-modules. */ + seq = -1ULL; + } else { + PERROR("kernctl_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 = kernctl_get_events_discarded(stream->wait_fd, &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; +} + +/* + * Check if the local version of the metadata stream matches with the version + * of the metadata stream in the kernel. If it was updated, set the reset flag + * on the stream. + */ +static +int metadata_stream_check_version(int infd, struct lttng_consumer_stream *stream) +{ + int ret; + uint64_t cur_version; + + ret = kernctl_get_metadata_version(infd, &cur_version); + if (ret < 0) { + if (ret == -ENOTTY) { + /* + * LTTng-modules does not implement this + * command. + */ + ret = 0; + goto end; + } + ERR("Failed to get the metadata version"); + goto end; + } + + if (stream->metadata_version == cur_version) { + ret = 0; + goto end; + } + + DBG("New metadata version detected"); + stream->metadata_version = cur_version; + stream->reset_metadata_flag = 1; + ret = 0; + +end: + return ret; +} + +/* + * Consume data on a file descriptor and write it on a trace file. + * The stream and channel locks must be held by the caller. + */ +ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream, + struct lttng_consumer_local_data *ctx) +{ + unsigned long len, subbuf_size, padding; + int err, write_index = 1, rotation_ret; + ssize_t ret = 0; + int infd = stream->wait_fd; + struct ctf_packet_index index; + + DBG("In read_subbuffer (infd : %d)", infd); + + /* + * If the stream was flagged to be ready for rotation before we extract the + * next packet, rotate it now. + */ + if (stream->rotate_ready) { + DBG("Rotate stream before extracting data"); + rotation_ret = lttng_consumer_rotate_stream(ctx, stream); + if (rotation_ret < 0) { + ERR("Stream rotation error"); + ret = -1; + goto error; + } + } + + /* 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 = err; + goto error; } /* Get the full subbuffer size including padding */ err = kernctl_get_padded_subbuf_size(infd, &len); if (err != 0) { - errno = -err; - perror("Getting sub-buffer len failed."); + PERROR("Getting sub-buffer len failed."); + err = kernctl_put_subbuf(infd); + if (err != 0) { + if (err == -EFAULT) { + PERROR("Error in unreserving sub buffer\n"); + } else if (err == -EIO) { + /* Should never happen with newer LTTng versions */ + PERROR("Reader has been pushed by the writer, last sub-buffer corrupted."); + } + ret = err; + goto error; + } ret = err; - goto end; + goto error; + } + + if (!stream->metadata_flag) { + ret = get_index_values(&index, infd); + if (ret < 0) { + err = kernctl_put_subbuf(infd); + if (err != 0) { + if (err == -EFAULT) { + PERROR("Error in unreserving sub buffer\n"); + } else if (err == -EIO) { + /* Should never happen with newer LTTng versions */ + PERROR("Reader has been pushed by the writer, last sub-buffer corrupted."); + } + ret = err; + goto error; + } + goto error; + } + ret = update_stream_stats(stream); + if (ret < 0) { + err = kernctl_put_subbuf(infd); + if (err != 0) { + if (err == -EFAULT) { + PERROR("Error in unreserving sub buffer\n"); + } else if (err == -EIO) { + /* Should never happen with newer LTTng versions */ + PERROR("Reader has been pushed by the writer, last sub-buffer corrupted."); + } + ret = err; + goto error; + } + goto error; + } + } else { + write_index = 0; + ret = metadata_stream_check_version(infd, stream); + if (ret < 0) { + err = kernctl_put_subbuf(infd); + if (err != 0) { + if (err == -EFAULT) { + PERROR("Error in unreserving sub buffer\n"); + } else if (err == -EIO) { + /* Should never happen with newer LTTng versions */ + PERROR("Reader has been pushed by the writer, last sub-buffer corrupted."); + } + ret = err; + goto error; + } + goto error; + } } switch (stream->chan->output) { @@ -896,7 +1646,7 @@ ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream, /* splice the subbuffer to the tracefile */ ret = lttng_consumer_on_read_subbuffer_splice(ctx, stream, subbuf_size, - padding); + 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. @@ -908,16 +1658,27 @@ ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream, */ 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) { - errno = -err; - perror("Getting sub-buffer len failed."); + PERROR("Getting sub-buffer len failed."); + err = kernctl_put_subbuf(infd); + if (err != 0) { + if (err == -EFAULT) { + PERROR("Error in unreserving sub buffer\n"); + } else if (err == -EIO) { + /* Should never happen with newer LTTng versions */ + PERROR("Reader has been pushed by the writer, last sub-buffer corrupted."); + } + ret = err; + goto error; + } ret = err; - goto end; + goto error; } /* Make sure the tracer is not gone mad on us! */ @@ -927,7 +1688,7 @@ ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream, /* write the subbuffer to the tracefile */ ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size, - padding); + 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_ @@ -937,33 +1698,87 @@ ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream, (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) { /* * Display the error but continue processing to try to release the - * subbuffer + * subbuffer. This is a DBG statement since this is possible to + * happen without being a critical error. */ - ERR("Error writing to tracefile " + 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 = -1; + ret = -EPERM; } err = kernctl_put_next_subbuf(infd); if (err != 0) { - errno = -err; - if (errno == EFAULT) { - perror("Error in unreserving sub buffer\n"); - } else if (errno == EIO) { + if (err == -EFAULT) { + PERROR("Error in unreserving sub buffer\n"); + } else if (err == -EIO) { /* Should never happen with newer LTTng versions */ - perror("Reader has been pushed by the writer, last sub-buffer corrupted."); + PERROR("Reader has been pushed by the writer, last sub-buffer corrupted."); } + ret = err; + goto error; + } - ret = -err; - goto end; + /* Write index if needed. */ + if (!write_index) { + goto rotate; } -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_kernel_index(stream); + } else { + pthread_mutex_unlock(&stream->metadata_timer_lock); + } + if (err < 0) { + goto error; + } + } + + err = consumer_stream_write_index(stream, &index); + if (err < 0) { + goto error; + } + +rotate: + /* + * After extracting the packet, we check if the stream is now ready to be + * rotated and perform the action immediately. + */ + rotation_ret = lttng_consumer_stream_is_rotate_ready(stream); + if (rotation_ret == 1) { + rotation_ret = lttng_consumer_rotate_stream(ctx, stream); + if (rotation_ret < 0) { + ERR("Stream rotation error"); + ret = -1; + goto error; + } + } else if (rotation_ret < 0) { + ERR("Checking if stream is ready to rotate"); + ret = -1; + goto error; + } + +error: return ret; } @@ -974,18 +1789,15 @@ 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); - 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->output == LTTNG_EVENT_MMAP) { @@ -994,7 +1806,6 @@ int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream) ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len); if (ret != 0) { - errno = -ret; PERROR("kernctl_get_mmap_len"); goto error_close_fd; } @@ -1013,11 +1824,12 @@ int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream) return 0; error_close_fd: - { + if (stream->out_fd >= 0) { int err; err = close(stream->out_fd); assert(!err); + stream->out_fd = -1; } error: return ret; @@ -1037,6 +1849,11 @@ int lttng_kconsumer_data_pending(struct lttng_consumer_stream *stream) assert(stream); + if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) { + ret = 0; + goto end; + } + ret = kernctl_get_next_subbuf(stream->wait_fd); if (ret == 0) { /* There is still data so let's put back this subbuffer. */