/*
* 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 <lttng/ust-ctl.h>
#include <poll.h>
#include <urcu/list.h>
#include <signal.h>
+#include <bin/lttng-consumerd/health-consumerd.h>
#include <common/common.h>
#include <common/sessiond-comm/sessiond-comm.h>
#include <common/relayd/relayd.h>
#include <common/compat/fcntl.h>
-#include <common/consumer-metadata-cache.h>
-#include <common/consumer-stream.h>
-#include <common/consumer-timer.h>
+#include <common/compat/endian.h>
+#include <common/consumer/consumer-metadata-cache.h>
+#include <common/consumer/consumer-stream.h>
+#include <common/consumer/consumer-timer.h>
#include <common/utils.h>
+#include <common/index/index.h>
#include "ust-consumer.h"
+#define INT_MAX_STR_LEN 12 /* includes \0 */
+
extern struct lttng_consumer_global_data consumer_data;
extern int consumer_poll_timeout;
extern volatile int consumer_quit;
cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
send_node) {
+
+ health_code_update();
+
cds_list_del(&stream->send_node);
ustctl_destroy_stream(stream->ustream);
free(stream);
*/
if (channel->uchan) {
lttng_ustconsumer_del_channel(channel);
+ lttng_ustconsumer_free_channel(channel);
}
free(channel);
}
const char *pathname, const char *name, uid_t uid, gid_t gid,
uint64_t relayd_id, uint64_t key, enum lttng_event_output output,
uint64_t tracefile_size, uint64_t tracefile_count,
- uint64_t session_id_per_pid, unsigned int monitor)
+ uint64_t session_id_per_pid, unsigned int monitor,
+ unsigned int live_timer_interval,
+ const char *root_shm_path, const char *shm_path)
{
assert(pathname);
assert(name);
return consumer_allocate_channel(key, session_id, pathname, name, uid,
gid, relayd_id, output, tracefile_size,
- tracefile_count, session_id_per_pid, monitor);
+ tracefile_count, session_id_per_pid, monitor,
+ live_timer_interval, root_shm_path, shm_path);
}
/*
/* Get the right pipe where the stream will be sent. */
if (stream->metadata_flag) {
+ ret = consumer_add_metadata_stream(stream);
+ if (ret) {
+ ERR("Consumer add metadata stream %" PRIu64 " failed.",
+ stream->key);
+ goto error;
+ }
stream_pipe = ctx->consumer_metadata_pipe;
} else {
+ ret = consumer_add_data_stream(stream);
+ if (ret) {
+ ERR("Consumer add stream %" PRIu64 " failed.",
+ stream->key);
+ goto error;
+ }
stream_pipe = ctx->consumer_data_pipe;
}
+ /*
+ * From this point on, the stream's ownership has been moved away from
+ * the channel and becomes globally visible.
+ */
+ stream->globally_visible = 1;
+
ret = lttng_pipe_write(stream_pipe, &stream, sizeof(stream));
if (ret < 0) {
ERR("Consumer write %s stream to pipe %d",
stream->metadata_flag ? "metadata" : "data",
lttng_pipe_get_writefd(stream_pipe));
+ if (stream->metadata_flag) {
+ consumer_del_stream_for_metadata(stream);
+ } else {
+ consumer_del_stream_for_data(stream);
+ }
}
+error:
+ return ret;
+}
+
+static
+int get_stream_shm_path(char *stream_shm_path, const char *shm_path, int cpu)
+{
+ char cpu_nr[INT_MAX_STR_LEN]; /* int max len */
+ int ret;
+ strncpy(stream_shm_path, shm_path, PATH_MAX);
+ stream_shm_path[PATH_MAX - 1] = '\0';
+ ret = snprintf(cpu_nr, INT_MAX_STR_LEN, "%i", cpu);
+ if (ret < 0) {
+ PERROR("snprintf");
+ goto end;
+ }
+ strncat(stream_shm_path, cpu_nr,
+ PATH_MAX - strlen(stream_shm_path) - 1);
+ ret = 0;
+end:
return ret;
}
int wait_fd;
int ust_metadata_pipe[2];
+ health_code_update();
+
if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA && channel->monitor) {
ret = utils_create_pipe_cloexec_nonblock(ust_metadata_pipe);
if (ret < 0) {
/* Keep stream reference when creating metadata. */
if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA) {
channel->metadata_stream = stream;
- stream->ust_metadata_poll_pipe[0] = ust_metadata_pipe[0];
- stream->ust_metadata_poll_pipe[1] = ust_metadata_pipe[1];
+ if (channel->monitor) {
+ /* Set metadata poll pipe if we created one */
+ memcpy(stream->ust_metadata_poll_pipe,
+ ust_metadata_pipe,
+ sizeof(ust_metadata_pipe));
+ }
}
}
return ret;
}
+/*
+ * create_posix_shm is never called concurrently within a process.
+ */
+static
+int create_posix_shm(void)
+{
+ char tmp_name[NAME_MAX];
+ int shmfd, ret;
+
+ ret = snprintf(tmp_name, NAME_MAX, "/ust-shm-consumer-%d", getpid());
+ if (ret < 0) {
+ PERROR("snprintf");
+ return -1;
+ }
+ /*
+ * Allocate shm, and immediately unlink its shm oject, keeping
+ * only the file descriptor as a reference to the object.
+ * We specifically do _not_ use the / at the beginning of the
+ * pathname so that some OS implementations can keep it local to
+ * the process (POSIX leaves this implementation-defined).
+ */
+ shmfd = shm_open(tmp_name, O_CREAT | O_EXCL | O_RDWR, 0700);
+ if (shmfd < 0) {
+ PERROR("shm_open");
+ goto error_shm_open;
+ }
+ ret = shm_unlink(tmp_name);
+ if (ret < 0 && errno != ENOENT) {
+ PERROR("shm_unlink");
+ goto error_shm_release;
+ }
+ return shmfd;
+
+error_shm_release:
+ ret = close(shmfd);
+ if (ret) {
+ PERROR("close");
+ }
+error_shm_open:
+ return -1;
+}
+
+static int open_ust_stream_fd(struct lttng_consumer_channel *channel,
+ struct ustctl_consumer_channel_attr *attr,
+ int cpu)
+{
+ char shm_path[PATH_MAX];
+ int ret;
+
+ if (!channel->shm_path[0]) {
+ return create_posix_shm();
+ }
+ ret = get_stream_shm_path(shm_path, channel->shm_path, cpu);
+ if (ret) {
+ goto error_shm_path;
+ }
+ return run_as_open(shm_path,
+ O_RDWR | O_CREAT | O_EXCL, S_IRUSR | S_IWUSR,
+ channel->uid, channel->gid);
+
+error_shm_path:
+ return -1;
+}
+
/*
* Create an UST channel with the given attributes and send it to the session
* daemon using the ust ctl API.
*
* Return 0 on success or else a negative value.
*/
-static int create_ust_channel(struct ustctl_consumer_channel_attr *attr,
- struct ustctl_consumer_channel **chanp)
+static int create_ust_channel(struct lttng_consumer_channel *channel,
+ struct ustctl_consumer_channel_attr *attr,
+ struct ustctl_consumer_channel **ust_chanp)
{
- int ret;
- struct ustctl_consumer_channel *channel;
+ int ret, nr_stream_fds, i, j;
+ int *stream_fds;
+ struct ustctl_consumer_channel *ust_channel;
+ assert(channel);
assert(attr);
- assert(chanp);
+ assert(ust_chanp);
DBG3("Creating channel to ustctl with attr: [overwrite: %d, "
"subbuf_size: %" PRIu64 ", num_subbuf: %" PRIu64 ", "
attr->num_subbuf, attr->switch_timer_interval,
attr->read_timer_interval, attr->output, attr->type);
- channel = ustctl_create_channel(attr);
- if (!channel) {
+ if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA)
+ nr_stream_fds = 1;
+ else
+ nr_stream_fds = ustctl_get_nr_stream_per_channel();
+ stream_fds = zmalloc(nr_stream_fds * sizeof(*stream_fds));
+ if (!stream_fds) {
+ ret = -1;
+ goto error_alloc;
+ }
+ for (i = 0; i < nr_stream_fds; i++) {
+ stream_fds[i] = open_ust_stream_fd(channel, attr, i);
+ if (stream_fds[i] < 0) {
+ ret = -1;
+ goto error_open;
+ }
+ }
+ ust_channel = ustctl_create_channel(attr, stream_fds, nr_stream_fds);
+ if (!ust_channel) {
ret = -1;
goto error_create;
}
-
- *chanp = channel;
+ channel->nr_stream_fds = nr_stream_fds;
+ channel->stream_fds = stream_fds;
+ *ust_chanp = ust_channel;
return 0;
error_create:
+error_open:
+ for (j = i - 1; j >= 0; j--) {
+ int closeret;
+
+ closeret = close(stream_fds[j]);
+ if (closeret) {
+ PERROR("close");
+ }
+ if (channel->shm_path[0]) {
+ char shm_path[PATH_MAX];
+
+ closeret = get_stream_shm_path(shm_path,
+ channel->shm_path, j);
+ if (closeret) {
+ ERR("Cannot get stream shm path");
+ }
+ closeret = run_as_unlink(shm_path,
+ channel->uid, channel->gid);
+ if (closeret) {
+ PERROR("unlink %s", shm_path);
+ }
+ }
+ }
+ /* Try to rmdir all directories under shm_path root. */
+ if (channel->root_shm_path[0]) {
+ (void) run_as_rmdir_recursive(channel->root_shm_path,
+ channel->uid, channel->gid);
+ }
+ free(stream_fds);
+error_alloc:
return ret;
}
struct lttng_consumer_channel *channel,
struct lttng_consumer_local_data *ctx, int *relayd_error)
{
- int ret, ret_code = LTTNG_OK;
+ int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS;
struct lttng_consumer_stream *stream;
+ uint64_t net_seq_idx = -1ULL;
assert(channel);
assert(ctx);
if (channel->relayd_id != (uint64_t) -1ULL) {
cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
+
+ health_code_update();
+
/* Try to send the stream to the relayd if one is available. */
ret = consumer_send_relayd_stream(stream, stream->chan->pathname);
if (ret < 0) {
if (relayd_error) {
*relayd_error = 1;
}
- ret_code = LTTNG_ERR_RELAYD_CONNECT_FAIL;
+ ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
+ }
+ if (net_seq_idx == -1ULL) {
+ net_seq_idx = stream->net_seq_idx;
}
}
}
/* Inform sessiond that we are about to send channel and streams. */
ret = consumer_send_status_msg(sock, ret_code);
- if (ret < 0 || ret_code != LTTNG_OK) {
+ if (ret < 0 || ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
/*
* Either the session daemon is not responding or the relayd died so we
* stop now.
/* The channel was sent successfully to the sessiond at this point. */
cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
+
+ health_code_update();
+
/* Send stream to session daemon. */
ret = send_sessiond_stream(sock, stream);
if (ret < 0) {
return 0;
error:
- if (ret_code != LTTNG_OK) {
+ if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
ret = -1;
}
return ret;
channel->nb_init_stream_left = 0;
/* The reply msg status is handled in the following call. */
- ret = create_ust_channel(attr, &channel->uchan);
+ ret = create_ust_channel(channel, attr, &channel->uchan);
if (ret < 0) {
goto end;
}
/* Send streams to the corresponding thread. */
cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
send_node) {
+
+ health_code_update();
+
/* Sending the stream to the thread. */
ret = send_stream_to_thread(stream, ctx);
if (ret < 0) {
* If we are unable to send the stream to the thread, there is
* a big problem so just stop everything.
*/
+ /* Remove node from the channel stream list. */
+ cds_list_del(&stream->send_node);
goto error;
}
/* Remove node from the channel stream list. */
cds_list_del(&stream->send_node);
- /*
- * From this point on, the stream's ownership has been moved away from
- * the channel and becomes globally visible.
- */
- stream->globally_visible = 1;
}
error:
cds_lfht_for_each_entry_duplicate(ht->ht,
ht->hash_fct(&channel->key, lttng_ht_seed), ht->match_fct,
&channel->key, &iter.iter, stream, node_channel_id.node) {
- ustctl_flush_buffer(stream->ustream, 1);
+
+ health_code_update();
+
+ pthread_mutex_lock(&stream->lock);
+ if (!stream->quiescent) {
+ ustctl_flush_buffer(stream->ustream, 0);
+ stream->quiescent = true;
+ }
+ pthread_mutex_unlock(&stream->lock);
+ }
+error:
+ rcu_read_unlock();
+ return ret;
+}
+
+/*
+ * Clear quiescent state from channel's streams using the given key to
+ * retrieve the channel.
+ *
+ * Return 0 on success else an LTTng error code.
+ */
+static int clear_quiescent_channel(uint64_t chan_key)
+{
+ int ret = 0;
+ struct lttng_consumer_channel *channel;
+ struct lttng_consumer_stream *stream;
+ struct lttng_ht *ht;
+ struct lttng_ht_iter iter;
+
+ DBG("UST consumer clear quiescent channel key %" PRIu64, chan_key);
+
+ rcu_read_lock();
+ channel = consumer_find_channel(chan_key);
+ if (!channel) {
+ ERR("UST consumer clear quiescent channel %" PRIu64 " not found", chan_key);
+ ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
+ goto error;
+ }
+
+ ht = consumer_data.stream_per_chan_id_ht;
+
+ /* For each stream of the channel id, clear quiescent state. */
+ cds_lfht_for_each_entry_duplicate(ht->ht,
+ ht->hash_fct(&channel->key, lttng_ht_seed), ht->match_fct,
+ &channel->key, &iter.iter, stream, node_channel_id.node) {
+
+ health_code_update();
+
+ pthread_mutex_lock(&stream->lock);
+ stream->quiescent = false;
+ pthread_mutex_unlock(&stream->lock);
}
error:
rcu_read_unlock();
goto error_unlock;
}
- if (channel->switch_timer_enabled == 1) {
- DBG("Deleting timer on metadata channel");
- consumer_timer_switch_stop(channel);
- }
-
- if (channel->metadata_stream) {
- ret = ustctl_stream_close_wakeup_fd(channel->metadata_stream->ustream);
- if (ret < 0) {
- ERR("UST consumer unable to close fd of metadata (ret: %d)", ret);
- ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
- goto error_unlock;
- }
- if (channel->monitor) {
- /* close the read-side in consumer_del_metadata_stream */
- ret = close(channel->metadata_stream->ust_metadata_poll_pipe[1]);
- if (ret < 0) {
- PERROR("Close UST metadata write-side poll pipe");
- }
- }
- }
+ lttng_ustconsumer_close_metadata(channel);
error_unlock:
pthread_mutex_unlock(&channel->lock);
ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
goto error;
}
+ ret = consumer_send_relayd_streams_sent(
+ metadata->metadata_stream->net_seq_idx);
+ if (ret < 0) {
+ ret = LTTCOMM_CONSUMERD_RELAYD_FAIL;
+ goto error;
+ }
}
ret = send_streams_to_thread(metadata, ctx);
* the stream is still in the local stream list of the channel. This call
* will make sure to clean that list.
*/
- cds_list_del(&metadata->metadata_stream->send_node);
consumer_stream_destroy(metadata->metadata_stream, NULL);
+ cds_list_del(&metadata->metadata_stream->send_node);
+ metadata->metadata_stream = NULL;
error_no_stream:
end:
return ret;
metadata_channel = consumer_find_channel(key);
if (!metadata_channel) {
- ERR("UST snapshot metadata channel not found for key %lu", key);
+ ERR("UST snapshot metadata channel not found for key %" PRIu64,
+ key);
ret = -1;
goto error;
}
assert(!metadata_channel->monitor);
+ health_code_update();
+
/*
* Ask the sessiond if we have new metadata waiting and update the
* consumer metadata cache.
*/
- ret = lttng_ustconsumer_request_metadata(ctx, metadata_channel, 0);
+ ret = lttng_ustconsumer_request_metadata(ctx, metadata_channel, 0, 1);
if (ret < 0) {
goto error;
}
+ health_code_update();
+
/*
* The metadata stream is NOT created in no monitor mode when the channel
* is created on a sessiond ask channel command.
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);
+ metadata_stream->uid, metadata_stream->gid, NULL);
if (ret < 0) {
goto error_stream;
}
metadata_stream->tracefile_size_current = 0;
}
- pthread_mutex_lock(&metadata_channel->metadata_cache->lock);
-
do {
+ health_code_update();
+
ret = lttng_consumer_read_subbuffer(metadata_stream, ctx);
if (ret < 0) {
- goto error_unlock;
+ goto error_stream;
}
} while (ret > 0);
-error_unlock:
- pthread_mutex_unlock(&metadata_channel->metadata_cache->lock);
-
error_stream:
/*
* Clean up the stream completly because the next snapshot will use a new
* metadata stream.
*/
- cds_list_del(&metadata_stream->send_node);
consumer_stream_destroy(metadata_stream, NULL);
+ cds_list_del(&metadata_stream->send_node);
metadata_channel->metadata_stream = NULL;
error:
* Returns 0 on success, < 0 on error
*/
static int snapshot_channel(uint64_t key, char *path, uint64_t relayd_id,
- uint64_t max_stream_size, struct lttng_consumer_local_data *ctx)
+ uint64_t nb_packets_per_stream, struct lttng_consumer_local_data *ctx)
{
int ret;
unsigned use_relayd = 0;
channel = consumer_find_channel(key);
if (!channel) {
- ERR("UST snapshot channel not found for key %lu", key);
+ ERR("UST snapshot channel not found for key %" PRIu64, key);
ret = -1;
goto error;
}
assert(!channel->monitor);
- DBG("UST consumer snapshot channel %lu", key);
+ DBG("UST consumer snapshot channel %" PRIu64, key);
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. */
pthread_mutex_lock(&stream->lock);
stream->net_seq_idx = relayd_id;
ret = utils_create_stream_file(path, stream->name,
stream->chan->tracefile_size,
stream->tracefile_count_current,
- stream->uid, stream->gid);
+ stream->uid, stream->gid, NULL);
if (ret < 0) {
goto error_unlock;
}
DBG("UST 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) {
+ goto error_unlock;
+ }
+ }
- ustctl_flush_buffer(stream->ustream, 1);
+ /*
+ * If tracing is active, we want to perform a "full" buffer flush.
+ * Else, if quiescent, it has already been done by the prior stop.
+ */
+ if (!stream->quiescent) {
+ ustctl_flush_buffer(stream->ustream, 0);
+ }
ret = lttng_ustconsumer_take_snapshot(stream);
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
+ * the possible size of the snapshot. Also, we assume 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("UST consumer taking snapshot at pos %lu", consumed_pos);
}
DBG("UST 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;
}
}
read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len,
- padded_len - len);
+ padded_len - len, NULL);
if (use_relayd) {
if (read_len != len) {
- ret = -1;
+ ret = -EPERM;
goto error_put_subbuf;
}
} else {
if (read_len != padded_len) {
- ret = -1;
+ ret = -EPERM;
goto error_put_subbuf;
}
}
goto error_close_stream;
}
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;
}
/* Simply close the stream so we can use it on the next snapshot. */
}
/*
- * Receive the metadata updates from the sessiond.
+ * Receive the metadata updates from the sessiond. Supports receiving
+ * overlapping metadata, but is needs to always belong to a contiguous
+ * range starting from 0.
+ * Be careful about the locks held when calling this function: it needs
+ * the metadata cache flush to concurrently progress in order to
+ * complete.
*/
int lttng_ustconsumer_recv_metadata(int sock, uint64_t key, uint64_t offset,
- uint64_t len, struct lttng_consumer_channel *channel,
- int timer)
+ uint64_t len, uint64_t version,
+ struct lttng_consumer_channel *channel, int timer, int wait)
{
- int ret, ret_code = LTTNG_OK;
+ int ret, ret_code = LTTCOMM_CONSUMERD_SUCCESS;
char *metadata_str;
DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key, len);
goto end;
}
+ health_code_update();
+
/* Receive metadata string. */
ret = lttcomm_recv_unix_sock(sock, metadata_str, len);
if (ret < 0) {
goto end_free;
}
+ health_code_update();
+
pthread_mutex_lock(&channel->metadata_cache->lock);
- ret = consumer_metadata_cache_write(channel, offset, len, metadata_str);
+ ret = consumer_metadata_cache_write(channel, offset, len, version,
+ metadata_str);
if (ret < 0) {
/* Unable to handle metadata. Notify session daemon. */
ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
}
pthread_mutex_unlock(&channel->metadata_cache->lock);
+ if (!wait) {
+ goto end_free;
+ }
while (consumer_metadata_cache_flushed(channel, offset + len, timer)) {
DBG("Waiting for metadata to be flushed");
+
+ health_code_update();
+
usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME);
}
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;
struct lttng_consumer_channel *channel = NULL;
+ health_code_update();
+
ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
if (ret != sizeof(msg)) {
DBG("Consumer received unexpected message size %zd (expects %zu)",
}
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 */
+ assert(msg.cmd_type != LTTNG_CONSUMER_STOP);
+
+ health_code_update();
/* relayd needs RCU read-side lock */
rcu_read_lock();
/* Session daemon status message are handled in the following call. */
ret = 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_DESTROY_RELAYD:
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;
}
/*
msg.u.ask_channel.tracefile_size,
msg.u.ask_channel.tracefile_count,
msg.u.ask_channel.session_id_per_pid,
- msg.u.ask_channel.monitor);
+ msg.u.ask_channel.monitor,
+ msg.u.ask_channel.live_timer_interval,
+ msg.u.ask_channel.root_shm_path,
+ msg.u.ask_channel.shm_path);
if (!channel) {
goto end_channel_error;
}
+ /*
+ * Assign UST application UID to the channel. This value is ignored for
+ * per PID buffers. This is specific to UST thus setting this after the
+ * allocation.
+ */
+ channel->ust_app_uid = msg.u.ask_channel.ust_app_uid;
+
/* Build channel attributes from received message. */
attr.subbuf_size = msg.u.ask_channel.subbuf_size;
attr.num_subbuf = msg.u.ask_channel.num_subbuf;
attr.switch_timer_interval = msg.u.ask_channel.switch_timer_interval;
attr.read_timer_interval = msg.u.ask_channel.read_timer_interval;
attr.chan_id = msg.u.ask_channel.chan_id;
- attr.output = msg.u.ask_channel.output;
memcpy(attr.uuid, msg.u.ask_channel.uuid, sizeof(attr.uuid));
+ /* Match channel buffer type to the UST abi. */
+ switch (msg.u.ask_channel.output) {
+ case LTTNG_EVENT_MMAP:
+ default:
+ attr.output = LTTNG_UST_MMAP;
+ break;
+ }
+
/* Translate and save channel type. */
switch (msg.u.ask_channel.type) {
case LTTNG_UST_CHAN_PER_CPU:
goto error_fatal;
};
+ health_code_update();
+
ret = ask_channel(ctx, sock, channel, &attr);
if (ret < 0) {
goto end_channel_error;
}
consumer_timer_switch_start(channel, attr.switch_timer_interval);
attr.switch_timer_interval = 0;
+ } else {
+ int monitor_start_ret;
+
+ consumer_timer_live_start(channel,
+ msg.u.ask_channel.live_timer_interval);
+ monitor_start_ret = consumer_timer_monitor_start(
+ channel,
+ msg.u.ask_channel.monitor_timer_interval);
+ if (monitor_start_ret < 0) {
+ ERR("Starting channel monitoring timer failed");
+ goto end_channel_error;
+ }
}
+ health_code_update();
+
/*
* Add the channel to the internal state AFTER all streams were created
* and successfully sent to session daemon. This way, all streams must
}
consumer_metadata_cache_destroy(channel);
}
+ if (channel->live_timer_enabled == 1) {
+ consumer_timer_live_stop(channel);
+ }
+ if (channel->monitor_timer_enabled == 1) {
+ consumer_timer_monitor_stop(channel);
+ }
goto end_channel_error;
}
+ health_code_update();
+
/*
* Channel and streams are now created. Inform the session daemon that
* everything went well and should wait to receive the channel and
channel = consumer_find_channel(key);
if (!channel) {
ERR("UST consumer get channel key %" PRIu64 " not found", key);
- ret_code = LTTNG_ERR_UST_CHAN_NOT_FOUND;
+ ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
goto end_msg_sessiond;
}
+ health_code_update();
+
/* Send everything to sessiond. */
ret = send_sessiond_channel(sock, channel, ctx, &relayd_err);
if (ret < 0) {
goto error_fatal;
}
+ health_code_update();
+
/*
* In no monitor mode, the streams ownership is kept inside the channel
* so don't send them to the data thread.
goto end_msg_sessiond;
}
+ case LTTNG_CONSUMER_CLEAR_QUIESCENT_CHANNEL:
+ {
+ int ret;
+
+ ret = clear_quiescent_channel(
+ msg.u.clear_quiescent_channel.key);
+ if (ret != 0) {
+ ret_code = ret;
+ }
+
+ goto end_msg_sessiond;
+ }
case LTTNG_CONSUMER_PUSH_METADATA:
{
int ret;
uint64_t len = msg.u.push_metadata.len;
uint64_t key = msg.u.push_metadata.key;
uint64_t offset = msg.u.push_metadata.target_offset;
+ uint64_t version = msg.u.push_metadata.version;
struct lttng_consumer_channel *channel;
DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key,
channel = consumer_find_channel(key);
if (!channel) {
- ERR("UST consumer push metadata %" PRIu64 " not found", key);
- ret_code = LTTNG_ERR_UST_CHAN_NOT_FOUND;
+ /*
+ * This is possible if the metadata creation on the consumer side
+ * is in flight vis-a-vis a concurrent push metadata from the
+ * session daemon. Simply return that the channel failed and the
+ * session daemon will handle that message correctly considering
+ * that this race is acceptable thus the DBG() statement here.
+ */
+ DBG("UST consumer push metadata %" PRIu64 " not found", key);
+ ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL;
+ goto end_msg_sessiond;
+ }
+
+ health_code_update();
+
+ if (!len) {
+ /*
+ * There is nothing to receive. We have simply
+ * checked whether the channel can be found.
+ */
+ ret_code = LTTCOMM_CONSUMERD_SUCCESS;
goto end_msg_sessiond;
}
/* Tell session daemon we are ready to receive the metadata. */
- ret = consumer_send_status_msg(sock, LTTNG_OK);
+ ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS);
if (ret < 0) {
/* Somehow, the session daemon is not responding anymore. */
goto error_fatal;
}
+ health_code_update();
+
/* Wait for more data. */
- if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) {
+ health_poll_entry();
+ ret = lttng_consumer_poll_socket(consumer_sockpoll);
+ health_poll_exit();
+ if (ret) {
goto error_fatal;
}
+ health_code_update();
+
ret = lttng_ustconsumer_recv_metadata(sock, key, offset,
- len, channel, 0);
+ len, version, channel, 0, 1);
if (ret < 0) {
/* error receiving from sessiond */
goto error_fatal;
ctx);
if (ret < 0) {
ERR("Snapshot metadata failed");
- ret_code = LTTNG_ERR_UST_META_FAIL;
+ ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
}
} else {
ret = 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_UST_CHAN_FAIL;
+ ret_code = LTTCOMM_CONSUMERD_CHANNEL_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;
}
+ health_code_update();
break;
}
- default:
+ case LTTNG_CONSUMER_DISCARDED_EVENTS:
+ {
+ int ret = 0;
+ uint64_t discarded_events;
+ struct lttng_ht_iter iter;
+ struct lttng_ht *ht;
+ struct lttng_consumer_stream *stream;
+ uint64_t id = msg.u.discarded_events.session_id;
+ uint64_t key = msg.u.discarded_events.channel_key;
+
+ DBG("UST consumer discarded events command for session id %"
+ PRIu64, id);
+ rcu_read_lock();
+ pthread_mutex_lock(&consumer_data.lock);
+
+ ht = consumer_data.stream_list_ht;
+
+ /*
+ * We only need a reference to the channel, but they are not
+ * directly indexed, so we just use the first matching stream
+ * to extract the information we need, we default to 0 if not
+ * found (no events are dropped if the channel is not yet in
+ * use).
+ */
+ discarded_events = 0;
+ cds_lfht_for_each_entry_duplicate(ht->ht,
+ ht->hash_fct(&id, lttng_ht_seed),
+ ht->match_fct, &id,
+ &iter.iter, stream, node_session_id.node) {
+ if (stream->chan->key == key) {
+ discarded_events = stream->chan->discarded_events;
+ break;
+ }
+ }
+ pthread_mutex_unlock(&consumer_data.lock);
+ rcu_read_unlock();
+
+ DBG("UST consumer discarded events command for session id %"
+ PRIu64 ", channel key %" PRIu64, id, key);
+
+ health_code_update();
+
+ /* Send back returned value to session daemon */
+ ret = lttcomm_send_unix_sock(sock, &discarded_events, sizeof(discarded_events));
+ if (ret < 0) {
+ PERROR("send discarded events");
+ goto error_fatal;
+ }
+
break;
}
+ case LTTNG_CONSUMER_LOST_PACKETS:
+ {
+ int ret;
+ uint64_t lost_packets;
+ struct lttng_ht_iter iter;
+ struct lttng_ht *ht;
+ struct lttng_consumer_stream *stream;
+ uint64_t id = msg.u.lost_packets.session_id;
+ uint64_t key = msg.u.lost_packets.channel_key;
-end_nosignal:
- rcu_read_unlock();
+ DBG("UST consumer lost packets command for session id %"
+ PRIu64, id);
+ rcu_read_lock();
+ pthread_mutex_lock(&consumer_data.lock);
- /*
- * Return 1 to indicate success since the 0 value can be a socket
- * shutdown during the recv() or send() call.
- */
- return 1;
+ ht = consumer_data.stream_list_ht;
-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;
- }
- rcu_read_unlock();
- return 1;
-end_channel_error:
- if (channel) {
/*
- * Free channel here since no one has a reference to it. We don't
- * free after that because a stream can store this pointer.
+ * We only need a reference to the channel, but they are not
+ * directly indexed, so we just use the first matching stream
+ * to extract the information we need, we default to 0 if not
+ * found (no packets lost if the channel is not yet in use).
*/
- destroy_channel(channel);
- }
+ lost_packets = 0;
+ cds_lfht_for_each_entry_duplicate(ht->ht,
+ ht->hash_fct(&id, lttng_ht_seed),
+ ht->match_fct, &id,
+ &iter.iter, stream, node_session_id.node) {
+ if (stream->chan->key == key) {
+ lost_packets = stream->chan->lost_packets;
+ break;
+ }
+ }
+ pthread_mutex_unlock(&consumer_data.lock);
+ rcu_read_unlock();
+
+ DBG("UST consumer lost packets command for session id %"
+ PRIu64 ", channel key %" PRIu64, id, key);
+
+ health_code_update();
+
+ /* Send back returned value to session daemon */
+ ret = lttcomm_send_unix_sock(sock, &lost_packets,
+ sizeof(lost_packets));
+ 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;
+ }
+ goto end_msg_sessiond;
+ }
+ default:
+ break;
+ }
+
+end_nosignal:
+ rcu_read_unlock();
+
+ health_code_update();
+
+ /*
+ * Return 1 to indicate success since the 0 value can be a socket
+ * shutdown during the recv() or send() call.
+ */
+ return 1;
+
+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;
+ }
+ rcu_read_unlock();
+
+ health_code_update();
+
+ return 1;
+end_channel_error:
+ if (channel) {
+ /*
+ * Free channel here since no one has a reference to it. We don't
+ * free after that because a stream can store this pointer.
+ */
+ destroy_channel(channel);
+ }
/* We have to send a status channel message indicating an error. */
ret = consumer_send_status_channel(sock, NULL);
if (ret < 0) {
goto error_fatal;
}
rcu_read_unlock();
+
+ health_code_update();
+
return 1;
error_fatal:
rcu_read_unlock();
}
/*
- * Take a snapshot for a specific fd
+ * Take a snapshot for a specific stream.
*
* Returns 0 on success, < 0 on error
*/
return ustctl_snapshot(stream->ustream);
}
+/*
+ * Sample consumed and produced positions for a specific stream.
+ *
+ * Returns 0 on success, < 0 on error.
+ */
+int lttng_ustconsumer_sample_snapshot_positions(
+ struct lttng_consumer_stream *stream)
+{
+ assert(stream);
+ assert(stream->ustream);
+
+ return ustctl_snapshot_sample_positions(stream->ustream);
+}
+
/*
* Get the produced position
*
return ustctl_snapshot_get_consumed(stream->ustream, pos);
}
+void lttng_ustconsumer_flush_buffer(struct lttng_consumer_stream *stream,
+ int producer)
+{
+ assert(stream);
+ assert(stream->ustream);
+
+ ustctl_flush_buffer(stream->ustream, producer);
+}
+
+int lttng_ustconsumer_get_current_timestamp(
+ struct lttng_consumer_stream *stream, uint64_t *ts)
+{
+ assert(stream);
+ assert(stream->ustream);
+ assert(ts);
+
+ return ustctl_get_current_timestamp(stream->ustream, ts);
+}
+
+int lttng_ustconsumer_get_sequence_number(
+ struct lttng_consumer_stream *stream, uint64_t *seq)
+{
+ assert(stream);
+ assert(stream->ustream);
+ assert(seq);
+
+ return ustctl_get_sequence_number(stream->ustream, seq);
+}
+
/*
- * Called when the stream signal the consumer that it has hang up.
+ * Called when the stream signals the consumer that it has hung up.
*/
void lttng_ustconsumer_on_stream_hangup(struct lttng_consumer_stream *stream)
{
assert(stream);
assert(stream->ustream);
- ustctl_flush_buffer(stream->ustream, 0);
+ pthread_mutex_lock(&stream->lock);
+ if (!stream->quiescent) {
+ ustctl_flush_buffer(stream->ustream, 0);
+ stream->quiescent = true;
+ }
+ pthread_mutex_unlock(&stream->lock);
stream->hangup_flush_done = 1;
}
void lttng_ustconsumer_del_channel(struct lttng_consumer_channel *chan)
{
+ int i;
+
assert(chan);
assert(chan->uchan);
if (chan->switch_timer_enabled == 1) {
consumer_timer_switch_stop(chan);
}
+ for (i = 0; i < chan->nr_stream_fds; i++) {
+ int ret;
+
+ ret = close(chan->stream_fds[i]);
+ if (ret) {
+ PERROR("close");
+ }
+ if (chan->shm_path[0]) {
+ char shm_path[PATH_MAX];
+
+ ret = get_stream_shm_path(shm_path, chan->shm_path, i);
+ if (ret) {
+ ERR("Cannot get stream shm path");
+ }
+ ret = run_as_unlink(shm_path, chan->uid, chan->gid);
+ if (ret) {
+ PERROR("unlink %s", shm_path);
+ }
+ }
+ }
+}
+
+void lttng_ustconsumer_free_channel(struct lttng_consumer_channel *chan)
+{
+ assert(chan);
+ assert(chan->uchan);
+
consumer_metadata_cache_destroy(chan);
ustctl_destroy_channel(chan->uchan);
+ /* Try to rmdir all directories under shm_path root. */
+ if (chan->root_shm_path[0]) {
+ (void) run_as_rmdir_recursive(chan->root_shm_path,
+ chan->uid, chan->gid);
+ }
+ free(chan->stream_fds);
}
void lttng_ustconsumer_del_stream(struct lttng_consumer_stream *stream)
ustctl_destroy_stream(stream->ustream);
}
+int lttng_ustconsumer_get_wakeup_fd(struct lttng_consumer_stream *stream)
+{
+ assert(stream);
+ assert(stream->ustream);
+
+ return ustctl_stream_get_wakeup_fd(stream->ustream);
+}
+
+int lttng_ustconsumer_close_wakeup_fd(struct lttng_consumer_stream *stream)
+{
+ assert(stream);
+ assert(stream->ustream);
+
+ return ustctl_stream_close_wakeup_fd(stream->ustream);
+}
+
+/*
+ * Populate index values of a UST 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,
+ struct ustctl_consumer_stream *ustream)
+{
+ int ret;
+
+ ret = ustctl_get_timestamp_begin(ustream, &index->timestamp_begin);
+ if (ret < 0) {
+ PERROR("ustctl_get_timestamp_begin");
+ goto error;
+ }
+ index->timestamp_begin = htobe64(index->timestamp_begin);
+
+ ret = ustctl_get_timestamp_end(ustream, &index->timestamp_end);
+ if (ret < 0) {
+ PERROR("ustctl_get_timestamp_end");
+ goto error;
+ }
+ index->timestamp_end = htobe64(index->timestamp_end);
+
+ ret = ustctl_get_events_discarded(ustream, &index->events_discarded);
+ if (ret < 0) {
+ PERROR("ustctl_get_events_discarded");
+ goto error;
+ }
+ index->events_discarded = htobe64(index->events_discarded);
+
+ ret = ustctl_get_content_size(ustream, &index->content_size);
+ if (ret < 0) {
+ PERROR("ustctl_get_content_size");
+ goto error;
+ }
+ index->content_size = htobe64(index->content_size);
+
+ ret = ustctl_get_packet_size(ustream, &index->packet_size);
+ if (ret < 0) {
+ PERROR("ustctl_get_packet_size");
+ goto error;
+ }
+ index->packet_size = htobe64(index->packet_size);
+
+ ret = ustctl_get_stream_id(ustream, &index->stream_id);
+ if (ret < 0) {
+ PERROR("ustctl_get_stream_id");
+ goto error;
+ }
+ index->stream_id = htobe64(index->stream_id);
+
+ ret = ustctl_get_instance_id(ustream, &index->stream_instance_id);
+ if (ret < 0) {
+ PERROR("ustctl_get_instance_id");
+ goto error;
+ }
+ index->stream_instance_id = htobe64(index->stream_instance_id);
+
+ ret = ustctl_get_sequence_number(ustream, &index->packet_seq_num);
+ if (ret < 0) {
+ PERROR("ustctl_get_sequence_number");
+ goto error;
+ }
+ index->packet_seq_num = htobe64(index->packet_seq_num);
+
+error:
+ return ret;
+}
+
+static
+void metadata_stream_reset_cache(struct lttng_consumer_stream *stream,
+ struct consumer_metadata_cache *cache)
+{
+ DBG("Metadata stream update to version %" PRIu64,
+ cache->version);
+ stream->ust_metadata_pushed = 0;
+ stream->metadata_version = cache->version;
+ stream->reset_metadata_flag = 1;
+}
+
+/*
+ * Check if the version of the metadata stream and metadata cache match.
+ * If the cache got updated, reset the metadata stream.
+ * The stream lock and metadata cache lock MUST be held.
+ * Return 0 on success, a negative value on error.
+ */
+static
+int metadata_stream_check_version(struct lttng_consumer_stream *stream)
+{
+ int ret = 0;
+ struct consumer_metadata_cache *cache = stream->chan->metadata_cache;
+
+ if (cache->version == stream->metadata_version) {
+ goto end;
+ }
+ metadata_stream_reset_cache(stream, cache);
+
+end:
+ return ret;
+}
+
+/*
+ * Write up to one packet from the metadata cache to the channel.
+ *
+ * Returns the number of bytes pushed in the cache, or a negative value
+ * on error.
+ */
+static
+int commit_one_metadata_packet(struct lttng_consumer_stream *stream)
+{
+ ssize_t write_len;
+ int ret;
+
+ pthread_mutex_lock(&stream->chan->metadata_cache->lock);
+ ret = metadata_stream_check_version(stream);
+ if (ret < 0) {
+ goto end;
+ }
+ if (stream->chan->metadata_cache->max_offset
+ == stream->ust_metadata_pushed) {
+ ret = 0;
+ goto end;
+ }
+
+ write_len = ustctl_write_one_packet_to_channel(stream->chan->uchan,
+ &stream->chan->metadata_cache->data[stream->ust_metadata_pushed],
+ stream->chan->metadata_cache->max_offset
+ - stream->ust_metadata_pushed);
+ assert(write_len != 0);
+ if (write_len < 0) {
+ ERR("Writing one metadata packet");
+ ret = -1;
+ goto end;
+ }
+ stream->ust_metadata_pushed += write_len;
+
+ assert(stream->chan->metadata_cache->max_offset >=
+ stream->ust_metadata_pushed);
+ ret = write_len;
+
+end:
+ pthread_mutex_unlock(&stream->chan->metadata_cache->lock);
+ return ret;
+}
+
+
+/*
+ * Sync metadata meaning request them to the session daemon and snapshot to the
+ * metadata thread can consumer them.
+ *
+ * Metadata stream lock is held here, but we need to release it when
+ * interacting with sessiond, else we cause a deadlock with live
+ * awaiting on metadata to be pushed out.
+ *
+ * Return 0 if new metadatda is available, EAGAIN if the metadata stream
+ * is empty or a negative value on error.
+ */
+int lttng_ustconsumer_sync_metadata(struct lttng_consumer_local_data *ctx,
+ struct lttng_consumer_stream *metadata)
+{
+ int ret;
+ int retry = 0;
+
+ assert(ctx);
+ assert(metadata);
+
+ pthread_mutex_unlock(&metadata->lock);
+ /*
+ * Request metadata from the sessiond, but don't wait for the flush
+ * because we locked the metadata thread.
+ */
+ ret = lttng_ustconsumer_request_metadata(ctx, metadata->chan, 0, 0);
+ pthread_mutex_lock(&metadata->lock);
+ if (ret < 0) {
+ goto end;
+ }
+
+ ret = commit_one_metadata_packet(metadata);
+ if (ret <= 0) {
+ goto end;
+ } else if (ret > 0) {
+ retry = 1;
+ }
+
+ ustctl_flush_buffer(metadata->ustream, 1);
+ ret = ustctl_snapshot(metadata->ustream);
+ if (ret < 0) {
+ if (errno != EAGAIN) {
+ ERR("Sync metadata, taking UST snapshot");
+ goto end;
+ }
+ DBG("No new metadata when syncing them.");
+ /* No new metadata, exit. */
+ ret = ENODATA;
+ goto end;
+ }
+
+ /*
+ * After this flush, we still need to extract metadata.
+ */
+ if (retry) {
+ ret = EAGAIN;
+ }
+
+end:
+ return ret;
+}
+
+/*
+ * Return 0 on success else a negative value.
+ */
+static int notify_if_more_data(struct lttng_consumer_stream *stream,
+ struct lttng_consumer_local_data *ctx)
+{
+ int ret;
+ struct ustctl_consumer_stream *ustream;
+
+ assert(stream);
+ assert(ctx);
+
+ ustream = stream->ustream;
+
+ /*
+ * First, we are going to check if there is a new subbuffer available
+ * before reading the stream wait_fd.
+ */
+ /* Get the next subbuffer */
+ ret = ustctl_get_next_subbuf(ustream);
+ if (ret) {
+ /* No more data found, flag the stream. */
+ stream->has_data = 0;
+ ret = 0;
+ goto end;
+ }
+
+ ret = ustctl_put_subbuf(ustream);
+ assert(!ret);
+
+ /* This stream still has data. Flag it and wake up the data thread. */
+ stream->has_data = 1;
+
+ if (stream->monitor && !stream->hangup_flush_done && !ctx->has_wakeup) {
+ ssize_t writelen;
+
+ writelen = lttng_pipe_write(ctx->consumer_wakeup_pipe, "!", 1);
+ if (writelen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
+ ret = writelen;
+ goto end;
+ }
+
+ /* The wake up pipe has been notified. */
+ ctx->has_wakeup = 1;
+ }
+ ret = 0;
+
+end:
+ return ret;
+}
+
+static
+int update_stream_stats(struct lttng_consumer_stream *stream)
+{
+ int ret;
+ uint64_t seq, discarded;
+
+ ret = ustctl_get_sequence_number(stream->ustream, &seq);
+ if (ret < 0) {
+ PERROR("ustctl_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 = ustctl_get_events_discarded(stream->ustream, &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;
+}
+
+/*
+ * Read subbuffer from the given stream.
+ *
+ * Stream lock MUST be acquired.
+ *
+ * Return 0 on success else a negative value.
+ */
int lttng_ustconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
struct lttng_consumer_local_data *ctx)
{
unsigned long len, subbuf_size, padding;
- int err;
+ int err, write_index = 1;
long ret = 0;
- char dummy;
struct ustctl_consumer_stream *ustream;
+ struct ctf_packet_index index;
assert(stream);
assert(stream->ustream);
/* Ease our life for what's next. */
ustream = stream->ustream;
- /* We can consume the 1 byte written into the wait_fd by UST */
- if (stream->monitor && !stream->hangup_flush_done) {
+ /*
+ * We can consume the 1 byte written into the wait_fd by UST. Don't trigger
+ * error if we cannot read this one byte (read returns 0), or if the error
+ * is EAGAIN or EWOULDBLOCK.
+ *
+ * This is only done when the stream is monitored by a thread, before the
+ * flush is done after a hangup and if the stream is not flagged with data
+ * since there might be nothing to consume in the wait fd but still have
+ * data available flagged by the consumer wake up pipe.
+ */
+ if (stream->monitor && !stream->hangup_flush_done && !stream->has_data) {
+ char dummy;
ssize_t readlen;
- do {
- readlen = read(stream->wait_fd, &dummy, 1);
- } while (readlen == -1 && errno == EINTR);
- if (readlen == -1 && errno != EAGAIN && errno != EWOULDBLOCK) {
+ readlen = lttng_read(stream->wait_fd, &dummy, 1);
+ if (readlen < 0 && errno != EAGAIN && errno != EWOULDBLOCK) {
ret = readlen;
goto end;
}
* already been read.
*/
if (stream->metadata_flag) {
- ssize_t write_len;
-
- if (stream->chan->metadata_cache->contiguous
- == stream->ust_metadata_pushed) {
- ret = 0;
+ ret = commit_one_metadata_packet(stream);
+ if (ret <= 0) {
goto end;
}
-
- write_len = ustctl_write_one_packet_to_channel(stream->chan->uchan,
- &stream->chan->metadata_cache->data[stream->ust_metadata_pushed],
- stream->chan->metadata_cache->contiguous
- - stream->ust_metadata_pushed);
- assert(write_len != 0);
- if (write_len < 0) {
- ERR("Writing one metadata packet");
- ret = -1;
- goto end;
- }
- stream->ust_metadata_pushed += write_len;
ustctl_flush_buffer(stream->ustream, 1);
goto retry;
}
goto end;
}
assert(stream->chan->output == CONSUMER_CHANNEL_MMAP);
+
+ if (!stream->metadata_flag) {
+ index.offset = htobe64(stream->out_fd_offset);
+ ret = get_index_values(&index, ustream);
+ if (ret < 0) {
+ err = ustctl_put_subbuf(ustream);
+ assert(err == 0);
+ goto end;
+ }
+
+ /* Update the stream's sequence and discarded events count. */
+ ret = update_stream_stats(stream);
+ if (ret < 0) {
+ PERROR("kernctl_get_events_discarded");
+ err = ustctl_put_subbuf(ustream);
+ assert(err == 0);
+ goto end;
+ }
+ } else {
+ write_index = 0;
+ }
+
/* Get the full padded subbuffer size */
err = ustctl_get_padded_subbuf_size(ustream, &len);
assert(err == 0);
padding = len - subbuf_size;
/* write the subbuffer to the tracefile */
- ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size, padding);
+ ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size, padding, &index);
/*
* The mmap operation should write subbuf_size amount of data when network
* streaming or the full padding (len) size when we are _not_ streaming.
DBG("Error writing to tracefile "
"(ret: %ld != len: %lu != subbuf_size: %lu)",
ret, len, subbuf_size);
+ write_index = 0;
}
err = ustctl_put_next_subbuf(ustream);
assert(err == 0);
+ /*
+ * This will consumer the byte on the wait_fd if and only if there is not
+ * next subbuffer to be acquired.
+ */
+ if (!stream->metadata_flag) {
+ ret = notify_if_more_data(stream, ctx);
+ if (ret < 0) {
+ goto end;
+ }
+ }
+
+ /* Write index if needed. */
+ if (!write_index) {
+ goto 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_ust_index(stream);
+ } else {
+ pthread_mutex_unlock(&stream->metadata_timer_lock);
+ }
+
+ if (err < 0) {
+ goto end;
+ }
+ }
+
+ assert(!stream->metadata_flag);
+ err = consumer_stream_write_index(stream, &index);
+ if (err < 0) {
+ goto end;
+ }
+
end:
return ret;
}
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);
+ stream->uid, stream->gid, NULL);
if (ret < 0) {
goto error;
}
stream->out_fd = ret;
stream->tracefile_size_current = 0;
+
+ if (!stream->metadata_flag) {
+ 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,
+ CTF_INDEX_MAJOR, CTF_INDEX_MINOR);
+ if (!index_file) {
+ goto error;
+ }
+ assert(!stream->index_file);
+ stream->index_file = index_file;
+ }
}
ret = 0;
}
if (stream->chan->type == CONSUMER_CHANNEL_TYPE_METADATA) {
+ uint64_t contiguous, pushed;
+
+ /* Ease our life a bit. */
+ contiguous = stream->chan->metadata_cache->max_offset;
+ pushed = stream->ust_metadata_pushed;
+
/*
* We can simply check whether all contiguously available data
* has been pushed to the ring buffer, since the push operation
* metadata has been consumed from the metadata stream.
*/
DBG("UST consumer metadata pending check: contiguous %" PRIu64 " vs pushed %" PRIu64,
- stream->chan->metadata_cache->contiguous,
- stream->ust_metadata_pushed);
- if (stream->chan->metadata_cache->contiguous
- != stream->ust_metadata_pushed) {
+ contiguous, pushed);
+ assert(((int64_t) (contiguous - pushed)) >= 0);
+ if ((contiguous != pushed) ||
+ (((int64_t) contiguous - pushed) > 0 || contiguous == 0)) {
ret = 1; /* Data is pending */
goto end;
}
return ret;
}
+/*
+ * Stop a given metadata channel timer if enabled and close the wait fd which
+ * is the poll pipe of the metadata stream.
+ *
+ * This MUST be called with the metadata channel acquired.
+ */
+void lttng_ustconsumer_close_metadata(struct lttng_consumer_channel *metadata)
+{
+ int ret;
+
+ assert(metadata);
+ assert(metadata->type == CONSUMER_CHANNEL_TYPE_METADATA);
+
+ DBG("Closing metadata channel key %" PRIu64, metadata->key);
+
+ if (metadata->switch_timer_enabled == 1) {
+ consumer_timer_switch_stop(metadata);
+ }
+
+ if (!metadata->metadata_stream) {
+ goto end;
+ }
+
+ /*
+ * Closing write side so the thread monitoring the stream wakes up if any
+ * and clean the metadata stream.
+ */
+ if (metadata->metadata_stream->ust_metadata_poll_pipe[1] >= 0) {
+ ret = close(metadata->metadata_stream->ust_metadata_poll_pipe[1]);
+ if (ret < 0) {
+ PERROR("closing metadata pipe write side");
+ }
+ metadata->metadata_stream->ust_metadata_poll_pipe[1] = -1;
+ }
+
+end:
+ return;
+}
+
/*
* Close every metadata stream wait fd of the metadata hash table. This
* function MUST be used very carefully so not to run into a race between the
* producer so calling this is safe because we are assured that no state change
* can occur in the metadata thread for the streams in the hash table.
*/
-void lttng_ustconsumer_close_metadata(struct lttng_ht *metadata_ht)
+void lttng_ustconsumer_close_all_metadata(struct lttng_ht *metadata_ht)
{
- int ret;
struct lttng_ht_iter iter;
struct lttng_consumer_stream *stream;
rcu_read_lock();
cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream,
node.node) {
- int fd = stream->wait_fd;
- /*
- * Whatever happens here we have to continue to try to close every
- * streams. Let's report at least the error on failure.
- */
- ret = ustctl_stream_close_wakeup_fd(stream->ustream);
- if (ret) {
- ERR("Unable to close metadata stream fd %d ret %d", fd, ret);
- }
- DBG("Metadata wait fd %d closed", fd);
+ health_code_update();
+
+ pthread_mutex_lock(&stream->chan->lock);
+ lttng_ustconsumer_close_metadata(stream->chan);
+ pthread_mutex_unlock(&stream->chan->lock);
+
}
rcu_read_unlock();
}
* function or any of its callees. Timers have a very strict locking
* semantic with respect to teardown. Failure to respect this semantic
* introduces deadlocks.
+ *
+ * DON'T hold the metadata lock when calling this function, else this
+ * can cause deadlock involving consumer awaiting for metadata to be
+ * pushed out due to concurrent interaction with the session daemon.
*/
int lttng_ustconsumer_request_metadata(struct lttng_consumer_local_data *ctx,
- struct lttng_consumer_channel *channel, int timer)
+ struct lttng_consumer_channel *channel, int timer, int wait)
{
struct lttcomm_metadata_request_msg request;
struct lttcomm_consumer_msg msg;
- enum lttng_error_code ret_code = LTTNG_OK;
- uint64_t len, key, offset;
+ enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
+ uint64_t len, key, offset, version;
int ret;
assert(channel);
assert(channel->metadata_cache);
+ memset(&request, 0, sizeof(request));
+
/* send the metadata request to sessiond */
switch (consumer_data.type) {
case LTTNG_CONSUMER64_UST:
request.session_id = channel->session_id;
request.session_id_per_pid = channel->session_id_per_pid;
- request.uid = channel->uid;
+ /*
+ * Request the application UID here so the metadata of that application can
+ * be sent back. The channel UID corresponds to the user UID of the session
+ * used for the rights on the stream file(s).
+ */
+ request.uid = channel->ust_app_uid;
request.key = channel->key;
+
DBG("Sending metadata request to sessiond, session id %" PRIu64
- ", per-pid %" PRIu64,
- channel->session_id,
- channel->session_id_per_pid);
+ ", per-pid %" PRIu64 ", app UID %u and channek key %" PRIu64,
+ request.session_id, request.session_id_per_pid, request.uid,
+ request.key);
pthread_mutex_lock(&ctx->metadata_socket_lock);
+
+ health_code_update();
+
ret = lttcomm_send_unix_sock(ctx->consumer_metadata_socket, &request,
sizeof(request));
if (ret < 0) {
goto end;
}
+ health_code_update();
+
/* Receive the metadata from sessiond */
ret = lttcomm_recv_unix_sock(ctx->consumer_metadata_socket, &msg,
sizeof(msg));
goto end;
}
+ health_code_update();
+
if (msg.cmd_type == LTTNG_ERR_UND) {
/* No registry found */
(void) consumer_send_status_msg(ctx->consumer_metadata_socket,
len = msg.u.push_metadata.len;
key = msg.u.push_metadata.key;
offset = msg.u.push_metadata.target_offset;
+ version = msg.u.push_metadata.version;
assert(key == channel->key);
if (len == 0) {
DBG("No new metadata to receive for key %" PRIu64, key);
}
+ health_code_update();
+
/* Tell session daemon we are ready to receive the metadata. */
ret = consumer_send_status_msg(ctx->consumer_metadata_socket,
- LTTNG_OK);
+ LTTCOMM_CONSUMERD_SUCCESS);
if (ret < 0 || len == 0) {
/*
* Somehow, the session daemon is not responding anymore or there is
goto end;
}
- ret_code = lttng_ustconsumer_recv_metadata(ctx->consumer_metadata_socket,
- key, offset, len, channel, timer);
- if (ret_code >= 0) {
+ health_code_update();
+
+ ret = lttng_ustconsumer_recv_metadata(ctx->consumer_metadata_socket,
+ key, offset, len, version, channel, timer, wait);
+ if (ret >= 0) {
/*
* Only send the status msg if the sessiond is alive meaning a positive
* ret code.
*/
- (void) consumer_send_status_msg(ctx->consumer_metadata_socket, ret_code);
+ (void) consumer_send_status_msg(ctx->consumer_metadata_socket, ret);
}
ret = 0;
end:
+ health_code_update();
+
pthread_mutex_unlock(&ctx->metadata_socket_lock);
return ret;
}
+
+/*
+ * Return the ustctl call for the get stream id.
+ */
+int lttng_ustconsumer_get_stream_id(struct lttng_consumer_stream *stream,
+ uint64_t *stream_id)
+{
+ assert(stream);
+ assert(stream_id);
+
+ return ustctl_get_stream_id(stream->ustream, stream_id);
+}
projects / lttng-tools.git / blobdiff