.type = LTTNG_CONSUMER_UNKNOWN,
};
-/* timeout parameter, to control the polling thread grace period. */
-int consumer_poll_timeout = -1;
-
/*
* Flag to inform the polling thread to quit when all fd hung up. Updated by
* the consumer_thread_receive_fds when it notices that all fds has hung up.
volatile int consumer_quit;
/*
- * The following two hash tables are visible by all threads which are separated
- * in different source files.
- *
* Global hash table containing respectively metadata and data streams. The
* stream element in this ht should only be updated by the metadata poll thread
* for the metadata and the data poll thread for the data.
*/
-struct lttng_ht *metadata_ht;
-struct lttng_ht *data_ht;
+static struct lttng_ht *metadata_ht;
+static struct lttng_ht *data_ht;
/*
* Notify a thread pipe to poll back again. This usually means that some global
rcu_read_unlock();
}
+/*
+ * Return a channel object for the given key.
+ *
+ * RCU read side lock MUST be acquired before calling this function and
+ * protects the channel ptr.
+ */
static struct lttng_consumer_channel *consumer_find_channel(int key)
{
struct lttng_ht_iter iter;
return NULL;
}
- rcu_read_lock();
-
lttng_ht_lookup(consumer_data.channel_ht, (void *)((unsigned long) key),
&iter);
node = lttng_ht_iter_get_node_ulong(&iter);
channel = caa_container_of(node, struct lttng_consumer_channel, node);
}
- rcu_read_unlock();
-
return channel;
}
call_rcu(&relayd->node.head, consumer_rcu_free_relayd);
}
+/*
+ * Iterate over the relayd hash table and destroy each element. Finally,
+ * destroy the whole hash table.
+ */
+static void cleanup_relayd_ht(void)
+{
+ struct lttng_ht_iter iter;
+ struct consumer_relayd_sock_pair *relayd;
+
+ rcu_read_lock();
+
+ cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd,
+ node.node) {
+ destroy_relayd(relayd);
+ }
+
+ lttng_ht_destroy(consumer_data.relayd_ht);
+
+ rcu_read_unlock();
+}
+
/*
* Update the end point status of all streams having the given network sequence
* index (relayd index).
goto free_stream;
}
- pthread_mutex_lock(&stream->lock);
pthread_mutex_lock(&consumer_data.lock);
+ pthread_mutex_lock(&stream->lock);
switch (consumer_data.type) {
case LTTNG_CONSUMER_KERNEL:
end:
consumer_data.need_update = 1;
- pthread_mutex_unlock(&consumer_data.lock);
pthread_mutex_unlock(&stream->lock);
+ pthread_mutex_unlock(&consumer_data.lock);
if (free_chan) {
consumer_del_channel(free_chan);
goto end;
}
+ rcu_read_lock();
+
/*
* Get stream's channel reference. Needed when adding the stream to the
* global hash table.
stream->path_name, stream->key, stream->shm_fd, stream->wait_fd,
(unsigned long long) stream->mmap_len, stream->out_fd,
stream->net_seq_idx, stream->session_id);
+
+ rcu_read_unlock();
return stream;
error:
+ rcu_read_unlock();
free(stream);
end:
return NULL;
DBG3("Adding consumer stream %d", stream->key);
pthread_mutex_lock(&consumer_data.lock);
+ pthread_mutex_lock(&stream->lock);
rcu_read_lock();
/* Steal stream identifier to avoid having streams with the same key */
consumer_data.need_update = 1;
rcu_read_unlock();
+ pthread_mutex_unlock(&stream->lock);
pthread_mutex_unlock(&consumer_data.lock);
return ret;
int ret;
struct lttng_ht_iter iter;
+ DBG("Consumer delete channel key %d", channel->key);
+
pthread_mutex_lock(&consumer_data.lock);
switch (consumer_data.type) {
* changed where this function will be called back again.
*/
if (stream->state != LTTNG_CONSUMER_ACTIVE_STREAM ||
- stream->endpoint_status) {
+ stream->endpoint_status == CONSUMER_ENDPOINT_INACTIVE) {
continue;
}
DBG("Active FD %d", stream->wait_fd);
}
/*
- * Close all the tracefiles and stream fds, should be called when all instances
- * are destroyed.
+ * Close all the tracefiles and stream fds and MUST be called when all
+ * instances are destroyed i.e. when all threads were joined and are ended.
*/
void lttng_consumer_cleanup(void)
{
rcu_read_unlock();
lttng_ht_destroy(consumer_data.channel_ht);
+
+ cleanup_relayd_ht();
+
+ /*
+ * This HT contains streams that are freed by either the metadata thread or
+ * the data thread so we do *nothing* on the hash table and simply destroy
+ * it.
+ */
+ lttng_ht_destroy(consumer_data.stream_list_ht);
}
/*
* core function for writing trace buffers to either the local filesystem or
* the network.
*
+ * It must be called with the stream lock held.
+ *
* Careful review MUST be put if any changes occur!
*
* Returns the number of bytes written
/* RCU lock for the relayd pointer */
rcu_read_lock();
- pthread_mutex_lock(&stream->lock);
-
/* Flag that the current stream if set for network streaming. */
if (stream->net_seq_idx != -1) {
relayd = consumer_find_relayd(stream->net_seq_idx);
if (relayd && stream->metadata_flag) {
pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
}
- pthread_mutex_unlock(&stream->lock);
rcu_read_unlock();
return written;
/*
* Splice the data from the ring buffer to the tracefile.
*
+ * It must be called with the stream lock held.
+ *
* Returns the number of bytes spliced.
*/
ssize_t lttng_consumer_on_read_subbuffer_splice(
/* RCU lock for the relayd pointer */
rcu_read_lock();
- pthread_mutex_lock(&stream->lock);
-
/* Flag that the current stream if set for network streaming. */
if (stream->net_seq_idx != -1) {
relayd = consumer_find_relayd(stream->net_seq_idx);
if (relayd && stream->metadata_flag) {
pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
}
- pthread_mutex_unlock(&stream->lock);
rcu_read_unlock();
return written;
*/
static void destroy_data_stream_ht(struct lttng_ht *ht)
{
- int ret;
struct lttng_ht_iter iter;
struct lttng_consumer_stream *stream;
rcu_read_lock();
cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
- ret = lttng_ht_del(ht, &iter);
- assert(!ret);
-
- call_rcu(&stream->node.head, consumer_free_stream);
+ /*
+ * Ignore return value since we are currently cleaning up so any error
+ * can't be handled.
+ */
+ (void) consumer_del_stream(stream, ht);
}
rcu_read_unlock();
*/
static void destroy_stream_ht(struct lttng_ht *ht)
{
- int ret;
struct lttng_ht_iter iter;
struct lttng_consumer_stream *stream;
rcu_read_lock();
cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
- ret = lttng_ht_del(ht, &iter);
- assert(!ret);
-
- call_rcu(&stream->node.head, consumer_free_stream);
+ /*
+ * Ignore return value since we are currently cleaning up so any error
+ * can't be handled.
+ */
+ (void) consumer_del_metadata_stream(stream, ht);
}
rcu_read_unlock();
goto free_stream;
}
+ pthread_mutex_lock(&consumer_data.lock);
pthread_mutex_lock(&stream->lock);
- pthread_mutex_lock(&consumer_data.lock);
switch (consumer_data.type) {
case LTTNG_CONSUMER_KERNEL:
if (stream->mmap_base != NULL) {
}
end:
- pthread_mutex_unlock(&consumer_data.lock);
pthread_mutex_unlock(&stream->lock);
+ pthread_mutex_unlock(&consumer_data.lock);
if (free_chan) {
consumer_del_channel(free_chan);
{
int ret = 0;
struct consumer_relayd_sock_pair *relayd;
+ struct lttng_ht_iter iter;
+ struct lttng_ht_node_ulong *node;
assert(stream);
assert(ht);
DBG3("Adding metadata stream %d to hash table", stream->wait_fd);
pthread_mutex_lock(&consumer_data.lock);
+ pthread_mutex_lock(&stream->lock);
/*
* From here, refcounts are updated so be _careful_ when returning an error
*/
rcu_read_lock();
+
+ /*
+ * Lookup the stream just to make sure it does not exist in our internal
+ * state. This should NEVER happen.
+ */
+ lttng_ht_lookup(ht, (void *)((unsigned long) stream->wait_fd), &iter);
+ node = lttng_ht_iter_get_node_ulong(&iter);
+ assert(!node);
+
/* Find relayd and, if one is found, increment refcount. */
relayd = consumer_find_relayd(stream->net_seq_idx);
if (relayd != NULL) {
uatomic_dec(&stream->chan->nb_init_streams);
}
- /* Steal stream identifier to avoid having streams with the same key */
- consumer_steal_stream_key(stream->key, ht);
-
lttng_ht_add_unique_ulong(ht, &stream->node);
/*
rcu_read_unlock();
+ pthread_mutex_unlock(&stream->lock);
pthread_mutex_unlock(&consumer_data.lock);
return ret;
}
rcu_read_lock();
cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) {
/* Validate delete flag of the stream */
- if (stream->endpoint_status != CONSUMER_ENDPOINT_INACTIVE) {
+ if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) {
continue;
}
/* Delete it right now */
rcu_read_lock();
cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) {
/* Validate delete flag of the stream */
- if (!stream->endpoint_status) {
+ if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) {
continue;
}
/*
rcu_register_thread();
+ metadata_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
+ if (!metadata_ht) {
+ /* ENOMEM at this point. Better to bail out. */
+ goto error;
+ }
+
DBG("Thread metadata poll started");
/* Size is set to 1 for the consumer_metadata pipe */
* since their might be data to consume.
*/
lttng_poll_del(&events, ctx->consumer_metadata_pipe[0]);
- close(ctx->consumer_metadata_pipe[0]);
+ ret = close(ctx->consumer_metadata_pipe[0]);
+ if (ret < 0) {
+ PERROR("close metadata pipe");
+ }
continue;
} else if (revents & LPOLLIN) {
do {
DBG("Metadata poll thread exiting");
lttng_poll_clean(&events);
- if (metadata_ht) {
- destroy_stream_ht(metadata_ht);
- }
+ destroy_stream_ht(metadata_ht);
rcu_unregister_thread();
return NULL;
data_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
if (data_ht == NULL) {
+ /* ENOMEM at this point. Better to bail out. */
goto end;
}
/* poll on the array of fds */
restart:
DBG("polling on %d fd", nb_fd + 1);
- num_rdy = poll(pollfd, nb_fd + 1, consumer_poll_timeout);
+ num_rdy = poll(pollfd, nb_fd + 1, -1);
DBG("poll num_rdy : %d", num_rdy);
if (num_rdy == -1) {
/*
* only tracked fd in the poll set. The thread will take care of closing
* the read side.
*/
- close(ctx->consumer_metadata_pipe[1]);
-
- if (data_ht) {
- destroy_data_stream_ht(data_ht);
+ ret = close(ctx->consumer_metadata_pipe[1]);
+ if (ret < 0) {
+ PERROR("close data pipe");
}
+ destroy_data_stream_ht(data_ht);
+
rcu_unregister_thread();
return NULL;
}
*/
void *consumer_thread_sessiond_poll(void *data)
{
- int sock, client_socket, ret;
+ int sock = -1, client_socket, ret;
/*
* structure to poll for incoming data on communication socket avoids
* making blocking sockets.
goto end;
}
+ /* This socket is not useful anymore. */
+ ret = close(client_socket);
+ if (ret < 0) {
+ PERROR("close client_socket");
+ }
+ client_socket = -1;
+
/* update the polling structure to poll on the established socket */
consumer_sockpoll[1].fd = sock;
consumer_sockpoll[1].events = POLLIN | POLLPRI;
*/
consumer_quit = 1;
- /*
- * 2s of grace period, if no polling events occur during
- * this period, the polling thread will exit even if there
- * are still open FDs (should not happen, but safety mechanism).
- */
- consumer_poll_timeout = LTTNG_CONSUMER_POLL_TIMEOUT;
-
/*
* Notify the data poll thread to poll back again and test the
* consumer_quit state that we just set so to quit gracefully.
*/
notify_thread_pipe(ctx->consumer_data_pipe[1]);
+ /* Cleaning up possibly open sockets. */
+ if (sock >= 0) {
+ ret = close(sock);
+ if (ret < 0) {
+ PERROR("close sock sessiond poll");
+ }
+ }
+ if (client_socket >= 0) {
+ ret = close(sock);
+ if (ret < 0) {
+ PERROR("close client_socket sessiond poll");
+ }
+ }
+
rcu_unregister_thread();
return NULL;
}
ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream,
struct lttng_consumer_local_data *ctx)
{
+ ssize_t ret;
+
+ pthread_mutex_lock(&stream->lock);
+
switch (consumer_data.type) {
case LTTNG_CONSUMER_KERNEL:
- return lttng_kconsumer_read_subbuffer(stream, ctx);
+ ret = lttng_kconsumer_read_subbuffer(stream, ctx);
+ break;
case LTTNG_CONSUMER32_UST:
case LTTNG_CONSUMER64_UST:
- return lttng_ustconsumer_read_subbuffer(stream, ctx);
+ ret = lttng_ustconsumer_read_subbuffer(stream, ctx);
+ break;
default:
ERR("Unknown consumer_data type");
assert(0);
- return -ENOSYS;
+ ret = -ENOSYS;
+ break;
}
+
+ pthread_mutex_unlock(&stream->lock);
+ return ret;
}
int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream)
consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
consumer_data.relayd_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
consumer_data.stream_list_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
-
- metadata_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
- assert(metadata_ht);
- data_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
- assert(data_ht);
}
/*
struct lttng_consumer_local_data *ctx, int sock,
struct pollfd *consumer_sockpoll, struct lttcomm_sock *relayd_sock)
{
- int fd, ret = -1;
+ int fd = -1, ret = -1;
+ enum lttng_error_code ret_code = LTTNG_OK;
struct consumer_relayd_sock_pair *relayd;
DBG("Consumer adding relayd socket (idx: %d)", net_seq_idx);
+ /* 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;
+ }
+
/* Get relayd reference if exists. */
relayd = consumer_find_relayd(net_seq_idx);
if (relayd == NULL) {
if (ret != sizeof(fd)) {
lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
ret = -1;
+ fd = -1; /* Just in case it gets set with an invalid value. */
+ goto error;
+ }
+
+ /* We have the fds without error. Send status back. */
+ ret = consumer_send_status_msg(sock, ret_code);
+ if (ret < 0) {
+ /* Somehow, the session daemon is not responding anymore. */
goto error;
}
/* Copy received lttcomm socket */
lttcomm_copy_sock(&relayd->control_sock, relayd_sock);
ret = lttcomm_create_sock(&relayd->control_sock);
+ /* Immediately try to close the created socket if valid. */
+ if (relayd->control_sock.fd >= 0) {
+ if (close(relayd->control_sock.fd)) {
+ PERROR("close relayd control socket");
+ }
+ }
+ /* Handle create_sock error. */
if (ret < 0) {
goto error;
}
- /* Close the created socket fd which is useless */
- close(relayd->control_sock.fd);
-
/* Assign new file descriptor */
relayd->control_sock.fd = fd;
+
+ /*
+ * Create a session on the relayd and store the returned id. No need to
+ * grab the socket lock since the relayd object is not yet visible.
+ */
+ ret = relayd_create_session(&relayd->control_sock,
+ &relayd->session_id);
+ if (ret < 0) {
+ goto error;
+ }
+
break;
case LTTNG_STREAM_DATA:
/* Copy received lttcomm socket */
lttcomm_copy_sock(&relayd->data_sock, relayd_sock);
ret = lttcomm_create_sock(&relayd->data_sock);
+ /* Immediately try to close the created socket if valid. */
+ if (relayd->data_sock.fd >= 0) {
+ if (close(relayd->data_sock.fd)) {
+ PERROR("close relayd data socket");
+ }
+ }
+ /* Handle create_sock error. */
if (ret < 0) {
goto error;
}
- /* Close the created socket fd which is useless */
- close(relayd->data_sock.fd);
-
/* Assign new file descriptor */
relayd->data_sock.fd = fd;
break;
add_relayd(relayd);
/* All good! */
- ret = 0;
+ return 0;
error:
+ /* Close received socket if valid. */
+ if (fd >= 0) {
+ if (close(fd)) {
+ PERROR("close received socket");
+ }
+ }
return ret;
}
ht = consumer_data.stream_list_ht;
cds_lfht_for_each_entry_duplicate(ht->ht,
- ht->hash_fct((void *)((unsigned long) id), 0x42UL),
+ ht->hash_fct((void *)((unsigned long) id), lttng_ht_seed),
ht->match_fct, (void *)((unsigned long) id),
&iter.iter, stream, node_session_id.node) {
/* If this call fails, the stream is being used hence data pending. */
rcu_read_unlock();
return 1;
}
+
+/*
+ * Send a ret code status message to the sessiond daemon.
+ *
+ * Return the sendmsg() return value.
+ */
+int consumer_send_status_msg(int sock, int ret_code)
+{
+ struct lttcomm_consumer_status_msg msg;
+
+ msg.ret_code = ret_code;
+
+ return lttcomm_send_unix_sock(sock, &msg, sizeof(msg));
+}