/*
* 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,
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) {
+ /*
+ * -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
*
}
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;
unsigned long consumed_pos, produced_pos;
health_code_update();
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);
+
+ ret = kernctl_buffer_flush_empty(stream->wait_fd);
+ if (ret < 0) {
+ /*
+ * Doing a buffer flush which does not take into
+ * account empty packets. This is not perfect
+ * for stream intersection, but required as a
+ * fall-back when "flush_empty" is not
+ * implemented by lttng-modules.
+ */
+ ret = kernctl_buffer_flush(stream->wait_fd);
if (ret < 0) {
- ERR("sending streams sent to relayd");
+ ERR("Failed to flush kernel stream");
goto end_unlock;
}
- }
-
- ret = kernctl_buffer_flush(stream->wait_fd);
- if (ret < 0) {
- ERR("Failed to flush kernel stream");
goto end_unlock;
}
while (consumed_pos < produced_pos) {
ssize_t read_len;
unsigned long len, padded_len;
- int lost_packet = 0;
health_code_update();
}
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;
- }
+ stream->chan->lost_packets++;
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) {
*
* Returns 0 on success, < 0 on error
*/
-int lttng_kconsumer_snapshot_metadata(uint64_t key, char *path,
+static int lttng_kconsumer_snapshot_metadata(uint64_t key, char *path,
uint64_t relayd_id, struct lttng_consumer_local_data *ctx)
{
int ret, use_relayd = 0;
if (!metadata_channel) {
ERR("Kernel snapshot metadata not found for key %" PRIu64, key);
ret = -1;
- goto error;
+ goto error_no_channel;
}
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 {
health_code_update();
- ret_read = lttng_kconsumer_read_subbuffer(metadata_stream, ctx);
+ 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:
+error_no_channel:
rcu_read_unlock();
return ret;
}
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:
} else {
ret = consumer_add_channel(new_channel, ctx);
}
- if (CONSUMER_CHANNEL_TYPE_DATA) {
+ if (msg.u.channel.type == CONSUMER_CHANNEL_TYPE_DATA && !ret) {
+ int monitor_start_ret;
+
+ DBG("Consumer starting monitor timer");
consumer_timer_live_start(new_channel,
msg.u.channel.live_timer_interval);
+ monitor_start_ret = consumer_timer_monitor_start(
+ new_channel,
+ msg.u.channel.monitor_timer_interval);
+ if (monitor_start_ret < 0) {
+ ERR("Starting channel monitoring timer failed");
+ goto end_nosignal;
+ }
+
}
health_code_update();
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;
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();
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:
if (ret < 0) {
goto end_nosignal;
}
+ channel->streams_sent_to_relayd = true;
}
break;
}
}
case LTTNG_CONSUMER_DISCARDED_EVENTS:
{
- uint64_t ret;
+ 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;
if (!channel) {
ERR("Kernel consumer discarded events channel %"
PRIu64 " not found", key);
- ret = 0;
+ count = 0;
} else {
- ret = channel->discarded_events;
+ count = channel->discarded_events;
}
health_code_update();
/* Send back returned value to session daemon */
- ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret));
+ ret = lttcomm_send_unix_sock(sock, &count, sizeof(count));
if (ret < 0) {
PERROR("send discarded events");
goto error_fatal;
}
case LTTNG_CONSUMER_LOST_PACKETS:
{
- uint64_t ret;
+ 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;
if (!channel) {
ERR("Kernel consumer lost packets channel %"
PRIu64 " not found", key);
- ret = 0;
+ count = 0;
} else {
- ret = channel->lost_packets;
+ count = channel->lost_packets;
}
health_code_update();
/* Send back returned value to session daemon */
- ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret));
+ 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_SET_CHANNEL_ROTATE_PIPE:
+ {
+ int channel_rotate_pipe;
+ int flags;
+
+ 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_rotate_pipe, 1);
+ if (ret != (ssize_t) sizeof(channel_rotate_pipe)) {
+ ERR("Failed to receive channel rotate pipe");
+ goto error_fatal;
+ }
+
+ DBG("Received channel rotate pipe (%d)", channel_rotate_pipe);
+ ctx->channel_rotate_pipe = channel_rotate_pipe;
+ /* Set the pipe as non-blocking. */
+ ret = fcntl(channel_rotate_pipe, F_GETFL, 0);
+ if (ret == -1) {
+ PERROR("fcntl get flags of the channel rotate pipe");
+ goto error_fatal;
+ }
+ flags = ret;
+
+ ret = fcntl(channel_rotate_pipe, F_SETFL, flags | O_NONBLOCK);
+ if (ret == -1) {
+ PERROR("fcntl set O_NONBLOCK flag of the channel rotate pipe");
+ goto error_fatal;
+ }
+ DBG("Channel rotate pipe set as non-blocking");
+ ret_code = LTTCOMM_CONSUMERD_SUCCESS;
+ ret = consumer_send_status_msg(sock, ret_code);
+ if (ret < 0) {
+ goto error_fatal;
+ }
+ break;
+ }
+ case LTTNG_CONSUMER_ROTATE_CHANNEL:
+ {
+ DBG("Consumer rotate channel %" PRIu64, msg.u.rotate_channel.key);
+
+ /*
+ * Sample the rotate position of all the streams in this channel.
+ */
+ ret = lttng_consumer_rotate_channel(msg.u.rotate_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_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;
+ }
+
+ /* Rotate the streams that are ready right now. */
+ ret = lttng_consumer_rotate_ready_streams(
+ msg.u.rotate_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_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;
+ }
+ break;
+ }
+ case LTTNG_CONSUMER_ROTATE_PENDING_RELAY:
+ {
+ int pending;
+ uint32_t pending_reply;
+
+ DBG("Consumer rotate pending on relay for session %" PRIu64,
+ msg.u.rotate_pending_relay.session_id);
+ pending = lttng_consumer_rotate_pending_relay(
+ msg.u.rotate_pending_relay.session_id,
+ msg.u.rotate_pending_relay.relayd_id,
+ msg.u.rotate_pending_relay.chunk_id);
+ if (pending < 0) {
+ ERR("Rotate pending relay failed");
+ ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
+ } 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("send data pending ret 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_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;
+ }
+ break;
+ }
default:
goto end_nosignal;
}
if (ret == -ENOTTY) {
/* Command not implemented by lttng-modules. */
index->stream_instance_id = -1ULL;
- ret = 0;
} else {
PERROR("kernctl_get_instance_id");
goto error;
if (ret == -ENOTTY) {
/* Command not implemented by lttng-modules. */
seq = -1ULL;
- ret = 0;
} else {
PERROR("kernctl_get_sequence_number");
goto end;
* 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 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 = err;
- goto end;
+ goto error;
}
/* Get the full subbuffer size including padding */
PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
}
ret = err;
- goto end;
+ goto error;
}
ret = err;
- goto end;
+ goto error;
}
if (!stream->metadata_flag) {
PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
}
ret = err;
- goto end;
+ goto error;
}
- goto end;
+ goto error;
}
ret = update_stream_stats(stream);
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;
}
} else {
write_index = 0;
ret = metadata_stream_check_version(infd, stream);
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;
}
}
PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
}
ret = err;
- goto end;
+ goto error;
}
ret = err;
- goto end;
+ goto error;
}
/* Make sure the tracer is not gone mad on us! */
PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
}
ret = err;
- goto end;
+ goto error;
}
/* Write index if needed. */
if (!write_index) {
- goto end;
+ goto rotate;
}
if (stream->chan->live_timer_interval && !stream->metadata_flag) {
pthread_mutex_unlock(&stream->metadata_timer_lock);
}
if (err < 0) {
- goto end;
+ 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;
}
if (!index_file) {
goto error;
}
+ assert(!stream->index_file);
stream->index_file = index_file;
}
}
projects / lttng-tools.git / blobdiff