* 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"
ret = kernctl_snapshot(infd);
if (ret != 0) {
- perror("Getting sub-buffer snapshot.");
+ PERROR("Getting sub-buffer snapshot.");
ret = -errno;
}
ret = kernctl_snapshot_get_produced(infd, pos);
if (ret != 0) {
- perror("kernctl_snapshot_get_produced");
+ PERROR("kernctl_snapshot_get_produced");
ret = -errno;
}
ret = kernctl_snapshot_get_consumed(infd, pos);
if (ret != 0) {
- perror("kernctl_snapshot_get_consumed");
+ PERROR("kernctl_snapshot_get_consumed");
ret = -errno;
}
* 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,
+ uint64_t relayd_id, uint64_t nb_packets_per_stream,
struct lttng_consumer_local_data *ctx)
{
int ret;
}
cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
+ /* Are we at a position _before_ the first available packet ? */
+ bool before_first_packet = true;
+
+ health_code_update();
+
/*
* Lock stream because we are about to change its state.
*/
DBG("Kernel consumer snapshot stream %s/%s (%" PRIu64 ")",
path, stream->name, stream->key);
}
+ if (relayd_id != -1ULL) {
+ ret = consumer_send_relayd_streams_sent(relayd_id);
+ if (ret < 0) {
+ ERR("sending streams sent to relayd");
+ goto end_unlock;
+ }
+ }
ret = kernctl_buffer_flush(stream->wait_fd);
if (ret < 0) {
}
}
- /*
- * 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;
+ int lost_packet = 0;
+
+ health_code_update();
DBG("Kernel consumer taking snapshot at pos %lu", consumed_pos);
}
DBG("Kernel consumer get subbuf failed. Skipping it.");
consumed_pos += stream->max_sb_size;
+
+ /*
+ * Start accounting lost packets only when we
+ * already have extracted packets (to match the
+ * content of the final snapshot).
+ */
+ if (!before_first_packet) {
+ lost_packet = 1;
+ }
continue;
}
goto end_unlock;
}
consumed_pos += stream->max_sb_size;
+
+ /*
+ * Only account lost packets located between
+ * succesfully extracted packets (do not account before
+ * and after since they are not visible in the
+ * resulting snapshot).
+ */
+ stream->chan->lost_packets += lost_packet;
+ lost_packet = 0;
+ before_first_packet = false;
}
if (relayd_id == (uint64_t) -1ULL) {
}
do {
+ health_code_update();
+
ret_read = lttng_kconsumer_read_subbuffer(metadata_stream, ctx);
if (ret_read < 0) {
if (ret_read != -EAGAIN) {
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();
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) {
msg.u.channel.live_timer_interval);
}
+ health_code_update();
+
/* If we received an error in add_channel, we need to report it. */
if (ret < 0) {
ret = consumer_send_status_msg(sock, ret);
* 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,
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;
}
/* Vitible 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->name, fd, 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;
+ }
+ }
+ break;
+ }
case LTTNG_CONSUMER_UPDATE_STREAM:
{
rcu_read_unlock();
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) {
msg.u.snapshot_channel.relayd_id, ctx);
if (ret < 0) {
ERR("Snapshot metadata failed");
- ret_code = LTTNG_ERR_KERN_META_FAIL;
+ ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
}
} 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,
+ msg.u.snapshot_channel.nb_packets_per_stream,
ctx);
if (ret < 0) {
ERR("Snapshot channel failed");
- ret_code = LTTNG_ERR_KERN_CHAN_FAIL;
+ 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. */
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:
+ {
+ uint64_t ret;
+ 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);
+ ret = 0;
+ } else {
+ ret = channel->discarded_events;
+ }
+
+ health_code_update();
+
+ /* Send back returned value to session daemon */
+ ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret));
+ if (ret < 0) {
+ PERROR("send discarded events");
+ goto error_fatal;
+ }
+
+ break;
+ }
+ case LTTNG_CONSUMER_LOST_PACKETS:
+ {
+ uint64_t ret;
+ 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);
+ ret = 0;
+ } else {
+ ret = channel->lost_packets;
+ }
+
+ health_code_update();
+
+ /* Send back returned value to session daemon */
+ ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret));
+ if (ret < 0) {
+ PERROR("send lost packets");
+ goto error_fatal;
+ }
+
+ 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) {
+ 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) {
+ PERROR("kernctl_get_sequence_number");
+ goto error;
+ }
+ index->packet_seq_num = htobe64(index->packet_seq_num);
+
error:
return ret;
}
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) {
+ 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 occured. We assume only one wrap-around
+ * has occured.
+ */
+ 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) {
+ 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.
*/
int err, write_index = 1;
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);
/* Get the full subbuffer size including padding */
err = kernctl_get_padded_subbuf_size(infd, &len);
if (err != 0) {
- perror("Getting sub-buffer len failed.");
+ PERROR("Getting sub-buffer len failed.");
+ err = kernctl_put_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;
+ goto end;
+ }
ret = -errno;
goto end;
}
if (!stream->metadata_flag) {
ret = get_index_values(&index, infd);
+ if (ret < 0) {
+ err = kernctl_put_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;
+ goto end;
+ }
+ goto end;
+ }
+ ret = update_stream_stats(stream);
if (ret < 0) {
goto end;
}
} else {
write_index = 0;
+ ret = metadata_stream_check_version(infd, stream);
+ if (ret < 0) {
+ goto end;
+ }
}
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.");
+ PERROR("Getting sub-buffer len failed.");
+ err = kernctl_put_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;
+ goto end;
+ }
ret = -errno;
goto end;
}
(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");
+ 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.");
+ PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
}
ret = -errno;
goto end;
/*
* 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 end;
}
projects / lttng-tools.git / blobdiff