/*
* Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca>
* Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
+ * Copyright (C) 2017 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
*
* 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,
* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
*/
-#define _GNU_SOURCE
+#define _LGPL_SOURCE
#include <assert.h>
#include <poll.h>
#include <pthread.h>
#include <unistd.h>
#include <sys/stat.h>
+#include <bin/lttng-consumerd/health-consumerd.h>
#include <common/common.h>
#include <common/kernel-ctl/kernel-ctl.h>
#include <common/sessiond-comm/sessiond-comm.h>
#include <common/sessiond-comm/relayd.h>
#include <common/compat/fcntl.h>
+#include <common/compat/endian.h>
#include <common/pipe.h>
#include <common/relayd/relayd.h>
#include <common/utils.h>
-#include <common/consumer-stream.h>
+#include <common/consumer/consumer-stream.h>
#include <common/index/index.h>
-#include <common/consumer-timer.h>
+#include <common/consumer/consumer-timer.h>
#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
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
*
ret = kernctl_snapshot_get_produced(infd, pos);
if (ret != 0) {
- perror("kernctl_snapshot_get_produced");
- ret = -errno;
+ PERROR("kernctl_snapshot_get_produced");
}
return ret;
ret = kernctl_snapshot_get_consumed(infd, pos);
if (ret != 0) {
- perror("kernctl_snapshot_get_consumed");
- ret = -errno;
+ PERROR("kernctl_snapshot_get_consumed");
}
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.
*
* 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,
+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);
}
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.
*/
path, stream->name, 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 stream");
- ret = -errno;
+ /*
+ * 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;
}
&stream->max_sb_size);
if (ret < 0) {
ERR("Getting kernel max_sb_size");
- ret = -errno;
goto end_unlock;
}
}
- /*
- * The original value is sent back if max stream size is larger than
- * the possible size of the snapshot. Also, we asume that the session
- * daemon should never send a maximum stream size that is lower than
- * subbuffer size.
- */
- consumed_pos = consumer_get_consumed_maxsize(consumed_pos,
- produced_pos, max_stream_size);
+ consumed_pos = consumer_get_consume_start_pos(consumed_pos,
+ produced_pos, nb_packets_per_stream,
+ stream->max_sb_size);
while (consumed_pos < produced_pos) {
ssize_t read_len;
unsigned long len, padded_len;
+ 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;
}
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;
error_put_subbuf:
ret = kernctl_put_subbuf(stream->wait_fd);
if (ret < 0) {
- ret = -errno;
ERR("Snapshot kernctl_put_subbuf error path");
}
end_unlock:
/*
* Read the whole metadata available for a snapshot.
+ * RCU read-side lock must be held across this function to ensure existence of
+ * metadata_channel.
*
* 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);
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);
/* Flag once that we have a valid relayd for the stream. */
if (relayd_id != (uint64_t) -1ULL) {
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->tracefile_count_current,
metadata_stream->uid, metadata_stream->gid, NULL);
if (ret < 0) {
- goto error;
+ goto error_snapshot;
}
metadata_stream->out_fd = ret;
}
do {
- ret_read = lttng_kconsumer_read_subbuffer(metadata_stream, ctx);
+ health_code_update();
+
+ ret_read = lttng_kconsumer_read_subbuffer(metadata_stream, ctx, NULL);
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;
}
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;
}
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)) {
if (ret > 0) {
}
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();
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;
+ 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;
}
+
+ 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.tracefile_size,
msg.u.channel.tracefile_count, 0,
msg.u.channel.monitor,
- msg.u.channel.live_timer_interval);
+ msg.u.channel.live_timer_interval,
+ NULL, NULL);
if (new_channel == NULL) {
lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
goto end_nosignal;
goto end_nosignal;
};
+ health_code_update();
+
if (ctx->on_recv_channel != NULL) {
ret_recv = ctx->on_recv_channel(new_channel);
if (ret_recv == 0) {
} else {
ret = consumer_add_channel(new_channel, ctx);
}
- consumer_timer_live_start(new_channel, msg.u.channel.live_timer_interval);
+ 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) {
* 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;
}
- if (ret_code != LTTNG_OK) {
+
+ health_code_update();
+
+ if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
/* Channel was not found. */
goto end_nosignal;
}
/* Blocking call */
- if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) {
- rcu_read_unlock();
- return -EINTR;
+ health_poll_entry();
+ ret = lttng_consumer_poll_socket(consumer_sockpoll);
+ health_poll_exit();
+ if (ret) {
+ goto error_fatal;
}
+ health_code_update();
+
/* Get stream file descriptor from socket */
ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
if (ret != sizeof(fd)) {
return ret;
}
+ 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
goto end_nosignal;
}
+ health_code_update();
+
new_stream = consumer_allocate_stream(channel->key,
fd,
LTTNG_CONSUMER_ACTIVE_STREAM,
msg.u.stream.cpu,
&alloc_ret,
channel->type,
- channel->monitor);
+ channel->monitor,
+ msg.u.stream.trace_archive_id);
if (new_stream == NULL) {
switch (alloc_ret) {
case -ENOMEM:
new_stream->chan = channel;
new_stream->wait_fd = fd;
+ 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) {
+ goto end_nosignal;
+ }
break;
case CONSUMER_CHANNEL_MMAP:
new_stream->output = LTTNG_EVENT_MMAP;
*/
new_stream->hangup_flush_done = 0;
+ health_code_update();
+
if (ctx->on_recv_stream) {
ret = ctx->on_recv_stream(new_stream);
if (ret < 0) {
}
}
+ health_code_update();
+
if (new_stream->metadata_flag) {
channel->metadata_stream = new_stream;
}
consumer_stream_free(new_stream);
goto end_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) {
+ goto end_nosignal;
+ }
+ }
}
/* 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();
+
ret = lttng_pipe_write(stream_pipe, &new_stream, sizeof(new_stream));
if (ret < 0) {
ERR("Consumer write %s stream to pipe %d",
goto end_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);
+ break;
+ }
+ 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 end_nosignal;
+ }
+
+ health_code_update();
+
+ /*
+ * We should not send this message if we don't monitor the
+ * streams in this channel.
+ */
+ if (!channel->monitor) {
+ break;
+ }
+
+ 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 end_nosignal;
+ }
+ channel->streams_sent_to_relayd = true;
+ }
break;
}
case LTTNG_CONSUMER_UPDATE_STREAM:
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;
}
/*
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. */
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) {
}
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;
+ 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;
+ }
}
}
+ health_code_update();
ret = consumer_send_status_msg(sock, ret_code);
if (ret < 0) {
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;
}
+ health_code_update();
+
+ /* Stop right now if no channel was found. */
+ if (!channel) {
+ goto end_nosignal;
+ }
+
/*
* This command should ONLY be issued for channel with streams set in
* no monitor mode.
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;
+
+ 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.pathname,
+ msg.u.rotate_channel.relayd_id,
+ msg.u.rotate_channel.metadata,
+ msg.u.rotate_channel.new_chunk_id,
+ 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_ROTATE_RENAME:
+ {
+ DBG("Consumer rename session %" PRIu64 " after rotation, old path = \"%s\", new path = \"%s\"",
+ msg.u.rotate_rename.session_id,
+ msg.u.rotate_rename.old_path,
+ msg.u.rotate_rename.new_path);
+ ret = lttng_consumer_rotate_rename(msg.u.rotate_rename.old_path,
+ msg.u.rotate_rename.new_path,
+ msg.u.rotate_rename.uid,
+ msg.u.rotate_rename.gid,
+ msg.u.rotate_rename.relayd_id);
+ if (ret < 0) {
+ ERR("Rotate rename failed");
+ ret_code = LTTCOMM_CONSUMERD_ROTATE_RENAME_FAILED;
+ }
+
+ 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_CHECK_ROTATION_PENDING_LOCAL:
+ {
+ int pending;
+ uint32_t pending_reply;
+
+ DBG("Perform local check of pending rotation for session id %" PRIu64,
+ msg.u.check_rotation_pending_local.session_id);
+ pending = lttng_consumer_check_rotation_pending_local(
+ msg.u.check_rotation_pending_local.session_id,
+ msg.u.check_rotation_pending_local.chunk_id);
+ if (pending < 0) {
+ ERR("Local rotation pending check failed with code %i", pending);
+ ret_code = LTTCOMM_CONSUMERD_ROTATION_PENDING_LOCAL_FAILED;
+ } else {
+ pending_reply = !!pending;
+ }
+
+ 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 (pending < 0) {
+ /*
+ * An error occured while running the command;
+ * don't send the 'pending' flag as the sessiond
+ * will not read it.
+ */
+ break;
+ }
+
+ /* Send back returned value to session daemon */
+ ret = lttcomm_send_unix_sock(sock, &pending_reply,
+ sizeof(pending_reply));
+ if (ret < 0) {
+ PERROR("Failed to send rotation pending return code");
+ goto error_fatal;
+ }
+ break;
+ }
+ case LTTNG_CONSUMER_CHECK_ROTATION_PENDING_RELAY:
+ {
+ int pending;
+ uint32_t pending_reply;
+
+ DBG("Perform relayd check of pending rotation for session id %" PRIu64,
+ msg.u.check_rotation_pending_relay.session_id);
+ pending = lttng_consumer_check_rotation_pending_relay(
+ msg.u.check_rotation_pending_relay.session_id,
+ msg.u.check_rotation_pending_relay.relayd_id,
+ msg.u.check_rotation_pending_relay.chunk_id);
+ if (pending < 0) {
+ ERR("Relayd rotation pending check failed with code %i", pending);
+ ret_code = LTTCOMM_CONSUMERD_ROTATION_PENDING_RELAY_FAILED;
+ } else {
+ pending_reply = !!pending;
+ }
+
+ 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 (pending < 0) {
+ /*
+ * An error occured while running the command;
+ * don't send the 'pending' flag as the sessiond
+ * will not read it.
+ */
+ break;
+ }
+
+ /* Send back returned value to session daemon */
+ ret = lttcomm_send_unix_sock(sock, &pending_reply,
+ sizeof(pending_reply));
+ if (ret < 0) {
+ PERROR("Failed to send rotation pending return code");
+ goto error_fatal;
+ }
+ break;
+ }
+ case LTTNG_CONSUMER_MKDIR:
+ {
+ DBG("Consumer mkdir %s in session %" PRIu64,
+ msg.u.mkdir.path,
+ msg.u.mkdir.session_id);
+ ret = lttng_consumer_mkdir(msg.u.mkdir.path,
+ msg.u.mkdir.uid,
+ msg.u.mkdir.gid,
+ msg.u.mkdir.relayd_id);
+ if (ret < 0) {
+ ERR("consumer mkdir failed");
+ ret_code = LTTCOMM_CONSUMERD_MKDIR_FAILED;
+ }
+
+ 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;
+ }
default:
goto end_nosignal;
}
* 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;
error_fatal:
*
* Return 0 on success or else a negative value.
*/
-static int get_index_values(struct lttng_packet_index *index, int infd)
+static int get_index_values(struct ctf_packet_index *index, int infd)
{
int ret;
}
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.
*/
ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
- struct lttng_consumer_local_data *ctx)
+ struct lttng_consumer_local_data *ctx, bool *rotated)
{
unsigned long len, subbuf_size, padding;
- int err, write_index = 1;
+ int err, write_index = 1, rotation_ret;
ssize_t ret = 0;
int infd = stream->wait_fd;
- struct lttng_packet_index index;
+ 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, rotated);
+ 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) {
*/
DBG("Reserving sub buffer failed (everything is normal, "
"it is due to concurrency)");
- ret = -errno;
- goto end;
+ ret = err;
+ goto error;
}
/* 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;
- goto end;
+ 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 error;
}
if (!stream->metadata_flag) {
ret = get_index_values(&index, infd);
if (ret < 0) {
- goto end;
+ 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) {
/* Get subbuffer size without padding */
err = kernctl_get_subbuf_size(infd, &subbuf_size);
if (err != 0) {
- perror("Getting sub-buffer len failed.");
- ret = -errno;
- goto end;
+ 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 error;
}
/* Make sure the tracer is not gone mad on us! */
(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;
err = kernctl_put_next_subbuf(infd);
if (err != 0) {
- 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 = -errno;
- goto end;
+ ret = err;
+ goto error;
}
/* Write index if needed. */
if (!write_index) {
- goto end;
+ goto rotate;
+ }
+
+ 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 end;
+ goto error;
}
-end:
+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, rotated);
+ 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;
}
stream->tracefile_size_current = 0;
if (!stream->metadata_flag) {
- ret = index_create_file(stream->chan->pathname,
+ struct lttng_index_file *index_file;
+
+ index_file = lttng_index_file_create(stream->chan->pathname,
stream->name, stream->uid, stream->gid,
stream->chan->tracefile_size,
- stream->tracefile_count_current);
- if (ret < 0) {
+ stream->tracefile_count_current,
+ CTF_INDEX_MAJOR, CTF_INDEX_MINOR);
+ if (!index_file) {
goto error;
}
- stream->index_fd = ret;
+ assert(!stream->index_file);
+ stream->index_file = index_file;
}
}
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;
projects / lttng-tools.git / blobdiff