/* Close thread quit pipes */
utils_close_pipe(thread_quit_pipe);
- /* Close relay cmd pipes */
- utils_close_pipe(relay_cmd_pipe);
-
uri_free(control_uri);
uri_free(data_uri);
}
{
if (stream->close_flag && stream->prev_seq == stream->last_net_seq_num) {
+ /*
+ * We are about to close the stream so set the data pending flag to 1
+ * which will make the end data pending command skip the stream which
+ * is now closed and ready. Note that after proceeding to a file close,
+ * the written file is ready for reading.
+ */
+ stream->data_pending_check_done = 1;
return 1;
}
return 0;
while (1) {
DBG("Listener accepting connections");
- nb_fd = LTTNG_POLL_GETNB(&events);
-
restart:
ret = lttng_poll_wait(&events, -1);
if (ret < 0) {
goto error;
}
+ nb_fd = ret;
+
DBG("Relay new connection received");
for (i = 0; i < nb_fd; i++) {
/* Fetch once the poll data */
DBG("Relay deleting session %" PRIu64, cmd->session->id);
- lttcomm_destroy_sock(cmd->session->sock);
-
rcu_read_lock();
cds_lfht_for_each_entry(streams_ht->ht, &iter.iter, node, node) {
node = lttng_ht_iter_get_node_ulong(&iter);
stream = caa_container_of(node,
struct relay_stream, stream_n);
if (stream->session == cmd->session) {
- close(stream->fd);
+ ret = close(stream->fd);
+ if (ret < 0) {
+ PERROR("close stream fd on delete session");
+ }
ret = lttng_ht_del(streams_ht, &iter);
assert(!ret);
call_rcu(&stream->rcu_node,
free(cmd->session);
}
+/*
+ * Handle the RELAYD_CREATE_SESSION command.
+ *
+ * On success, send back the session id or else return a negative value.
+ */
+static
+int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr,
+ struct relay_command *cmd)
+{
+ int ret = 0, send_ret;
+ struct relay_session *session;
+ struct lttcomm_relayd_status_session reply;
+
+ assert(recv_hdr);
+ assert(cmd);
+
+ memset(&reply, 0, sizeof(reply));
+
+ session = zmalloc(sizeof(struct relay_session));
+ if (session == NULL) {
+ PERROR("relay session zmalloc");
+ ret = -1;
+ goto error;
+ }
+
+ session->id = ++last_relay_session_id;
+ session->sock = cmd->sock;
+ cmd->session = session;
+
+ reply.session_id = htobe64(session->id);
+
+ DBG("Created session %" PRIu64, session->id);
+
+error:
+ if (ret < 0) {
+ reply.ret_code = htobe32(LTTNG_ERR_FATAL);
+ } else {
+ reply.ret_code = htobe32(LTTNG_OK);
+ }
+
+ send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
+ if (send_ret < 0) {
+ ERR("Relayd sending session id");
+ }
+
+ return ret;
+}
+
/*
* relay_add_stream: allocate a new stream for a session
*/
char *path = NULL, *root_path = NULL;
int ret, send_ret;
- if (!session || session->version_check_done == 0) {
+ if (!session || cmd->version_check_done == 0) {
ERR("Trying to add a stream before version check");
ret = -1;
goto end_no_session;
/* FIXME : use data_size for something ? */
ret = cmd->sock->ops->recvmsg(cmd->sock, &stream_info,
- sizeof(struct lttcomm_relayd_add_stream), MSG_WAITALL);
+ sizeof(struct lttcomm_relayd_add_stream), 0);
if (ret < sizeof(struct lttcomm_relayd_add_stream)) {
ERR("Relay didn't receive valid add_stream struct size : %d", ret);
ret = -1;
DBG("Close stream received");
- if (!session || session->version_check_done == 0) {
+ if (!session || cmd->version_check_done == 0) {
ERR("Trying to close a stream before version check");
ret = -1;
goto end_no_session;
}
ret = cmd->sock->ops->recvmsg(cmd->sock, &stream_info,
- sizeof(struct lttcomm_relayd_close_stream), MSG_WAITALL);
+ sizeof(struct lttcomm_relayd_close_stream), 0);
if (ret < sizeof(struct lttcomm_relayd_close_stream)) {
ERR("Relay didn't receive valid add_stream struct size : %d", ret);
ret = -1;
if (close_stream_check(stream)) {
int delret;
- close(stream->fd);
+ delret = close(stream->fd);
+ if (delret < 0) {
+ PERROR("close stream");
+ }
delret = lttng_ht_del(streams_ht, &iter);
assert(!delret);
call_rcu(&stream->rcu_node,
if (data_buffer_size < data_size) {
/* In case the realloc fails, we can free the memory */
- char *tmp_data_ptr = data_buffer;
- data_buffer = realloc(data_buffer, data_size);
- if (!data_buffer) {
+ char *tmp_data_ptr;
+
+ tmp_data_ptr = realloc(data_buffer, data_size);
+ if (!tmp_data_ptr) {
ERR("Allocating data buffer");
- free(tmp_data_ptr);
+ free(data_buffer);
ret = -1;
goto end;
}
+ data_buffer = tmp_data_ptr;
data_buffer_size = data_size;
}
memset(data_buffer, 0, data_size);
DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size);
- ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size,
- MSG_WAITALL);
+ ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0);
if (ret < 0 || ret != data_size) {
ret = -1;
ERR("Relay didn't receive the whole metadata");
struct relay_command *cmd)
{
int ret;
- struct lttcomm_relayd_version reply;
- struct relay_session *session;
+ struct lttcomm_relayd_version reply, msg;
- if (cmd->session == NULL) {
- session = zmalloc(sizeof(struct relay_session));
- if (session == NULL) {
- PERROR("relay session zmalloc");
- ret = -1;
- goto end;
- }
- session->id = ++last_relay_session_id;
- DBG("Created session %" PRIu64, session->id);
- cmd->session = session;
- } else {
- session = cmd->session;
+ assert(cmd);
+
+ cmd->version_check_done = 1;
+
+ /* Get version from the other side. */
+ ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
+ if (ret < 0 || ret != sizeof(msg)) {
+ ret = -1;
+ ERR("Relay failed to receive the version values.");
+ goto end;
}
- session->version_check_done = 1;
- ret = sscanf(VERSION, "%u.%u", &reply.major, &reply.minor);
+ /*
+ * For now, we just ignore the received version but after 2.1 stable
+ * release, a check must be done to see if we either adapt to the other
+ * side version (which MUST be lower than us) or keep the latest data
+ * structure considering that the other side will adapt.
+ */
+
+ ret = sscanf(VERSION, "%10u.%10u", &reply.major, &reply.minor);
if (ret < 2) {
ERR("Error in scanning version");
ret = -1;
}
/*
- * Check for data availability for a given stream id from the session daemon.
+ * Check for data pending for a given stream id from the session daemon.
*/
static
-int relay_data_available(struct lttcomm_relayd_hdr *recv_hdr,
+int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
struct relay_command *cmd, struct lttng_ht *streams_ht)
{
struct relay_session *session = cmd->session;
- struct lttcomm_relayd_data_available msg;
+ struct lttcomm_relayd_data_pending msg;
struct lttcomm_relayd_generic_reply reply;
struct relay_stream *stream;
int ret;
struct lttng_ht_iter iter;
uint64_t last_net_seq_num, stream_id;
- DBG("Data available command received");
+ DBG("Data pending command received");
- if (!session || session->version_check_done == 0) {
+ if (!session || cmd->version_check_done == 0) {
ERR("Trying to check for data before version check");
ret = -1;
goto end_no_session;
}
- ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), MSG_WAITALL);
+ ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
if (ret < sizeof(msg)) {
- ERR("Relay didn't receive valid data_available struct size : %d", ret);
+ ERR("Relay didn't receive valid data_pending struct size : %d", ret);
ret = -1;
goto end_no_session;
}
stream = caa_container_of(node, struct relay_stream, stream_n);
assert(stream);
- DBG("Data available for stream id %" PRIu64 " prev_seq %" PRIu64
+ DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64
" and last_seq %" PRIu64, stream_id, stream->prev_seq,
last_net_seq_num);
/* Avoid wrapping issue */
if (((int64_t) (stream->prev_seq - last_net_seq_num)) <= 0) {
- /* Data has in fact been written and is available */
- ret = 1;
- } else {
- /* Data still being streamed. */
+ /* Data has in fact been written and is NOT pending */
ret = 0;
+ } else {
+ /* Data still being streamed thus pending */
+ ret = 1;
}
+ /* Pending check is now done. */
+ stream->data_pending_check_done = 1;
+
end_unlock:
rcu_read_unlock();
reply.ret_code = htobe32(ret);
ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
if (ret < 0) {
- ERR("Relay data available ret code failed");
+ ERR("Relay data pending ret code failed");
}
end_no_session:
reply.ret_code = htobe32(LTTNG_OK);
ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
if (ret < 0) {
- ERR("Relay data available ret code failed");
+ ERR("Relay data quiescent control ret code failed");
}
return ret;
}
+/*
+ * Initialize a data pending command. This means that a client is about to ask
+ * for data pending for each stream he/she holds. Simply iterate over all
+ * streams of a session and set the data_pending_check_done flag.
+ *
+ * This command returns to the client a LTTNG_OK code.
+ */
+static
+int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
+ struct relay_command *cmd, struct lttng_ht *streams_ht)
+{
+ int ret;
+ struct lttng_ht_iter iter;
+ struct lttcomm_relayd_begin_data_pending msg;
+ struct lttcomm_relayd_generic_reply reply;
+ struct relay_stream *stream;
+ uint64_t session_id;
+
+ assert(recv_hdr);
+ assert(cmd);
+ assert(streams_ht);
+
+ DBG("Init streams for data pending");
+
+ if (!cmd->session || cmd->version_check_done == 0) {
+ ERR("Trying to check for data before version check");
+ ret = -1;
+ goto end_no_session;
+ }
+
+ ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
+ if (ret < sizeof(msg)) {
+ ERR("Relay didn't receive valid begin data_pending struct size: %d",
+ ret);
+ ret = -1;
+ goto end_no_session;
+ }
+
+ session_id = be64toh(msg.session_id);
+
+ /*
+ * Iterate over all streams to set the begin data pending flag. For now, the
+ * streams are indexed by stream handle so we have to iterate over all
+ * streams to find the one associated with the right session_id.
+ */
+ rcu_read_lock();
+ cds_lfht_for_each_entry(streams_ht->ht, &iter.iter, stream, stream_n.node) {
+ if (stream->session->id == session_id) {
+ stream->data_pending_check_done = 0;
+ DBG("Set begin data pending flag to stream %" PRIu64,
+ stream->stream_handle);
+ }
+ }
+ rcu_read_unlock();
+
+ /* All good, send back reply. */
+ reply.ret_code = htobe32(LTTNG_OK);
+
+ ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
+ if (ret < 0) {
+ ERR("Relay begin data pending send reply failed");
+ }
+
+end_no_session:
+ return ret;
+}
+
+/*
+ * End data pending command. This will check, for a given session id, if each
+ * stream associated with it has its data_pending_check_done flag set. If not,
+ * this means that the client lost track of the stream but the data is still
+ * being streamed on our side. In this case, we inform the client that data is
+ * inflight.
+ *
+ * Return to the client if there is data in flight or not with a ret_code.
+ */
+static
+int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
+ struct relay_command *cmd, struct lttng_ht *streams_ht)
+{
+ int ret;
+ struct lttng_ht_iter iter;
+ struct lttcomm_relayd_end_data_pending msg;
+ struct lttcomm_relayd_generic_reply reply;
+ struct relay_stream *stream;
+ uint64_t session_id;
+ uint32_t is_data_inflight = 0;
+
+ assert(recv_hdr);
+ assert(cmd);
+ assert(streams_ht);
+
+ DBG("End data pending command");
+
+ if (!cmd->session || cmd->version_check_done == 0) {
+ ERR("Trying to check for data before version check");
+ ret = -1;
+ goto end_no_session;
+ }
+
+ ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
+ if (ret < sizeof(msg)) {
+ ERR("Relay didn't receive valid end data_pending struct size: %d",
+ ret);
+ ret = -1;
+ goto end_no_session;
+ }
+
+ session_id = be64toh(msg.session_id);
+
+ /* Iterate over all streams to see if the begin data pending flag is set. */
+ rcu_read_lock();
+ cds_lfht_for_each_entry(streams_ht->ht, &iter.iter, stream, stream_n.node) {
+ if (stream->session->id == session_id &&
+ !stream->data_pending_check_done) {
+ is_data_inflight = 1;
+ DBG("Data is still in flight for stream %" PRIu64,
+ stream->stream_handle);
+ break;
+ }
+ }
+ rcu_read_unlock();
+
+ /* All good, send back reply. */
+ reply.ret_code = htobe32(is_data_inflight);
+
+ ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
+ if (ret < 0) {
+ ERR("Relay end data pending send reply failed");
+ }
+
+end_no_session:
+ return ret;
+}
+
/*
* relay_process_control: Process the commands received on the control socket
*/
int ret = 0;
switch (be32toh(recv_hdr->cmd)) {
- /*
case RELAYD_CREATE_SESSION:
ret = relay_create_session(recv_hdr, cmd);
break;
- */
case RELAYD_ADD_STREAM:
ret = relay_add_stream(recv_hdr, cmd, streams_ht);
break;
case RELAYD_CLOSE_STREAM:
ret = relay_close_stream(recv_hdr, cmd, streams_ht);
break;
- case RELAYD_DATA_AVAILABLE:
- ret = relay_data_available(recv_hdr, cmd, streams_ht);
+ case RELAYD_DATA_PENDING:
+ ret = relay_data_pending(recv_hdr, cmd, streams_ht);
break;
case RELAYD_QUIESCENT_CONTROL:
ret = relay_quiescent_control(recv_hdr, cmd);
break;
+ case RELAYD_BEGIN_DATA_PENDING:
+ ret = relay_begin_data_pending(recv_hdr, cmd, streams_ht);
+ break;
+ case RELAYD_END_DATA_PENDING:
+ ret = relay_end_data_pending(recv_hdr, cmd, streams_ht);
+ break;
case RELAYD_UPDATE_SYNC_INFO:
default:
ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd));
uint32_t data_size;
ret = cmd->sock->ops->recvmsg(cmd->sock, &data_hdr,
- sizeof(struct lttcomm_relayd_data_hdr), MSG_WAITALL);
+ sizeof(struct lttcomm_relayd_data_hdr), 0);
if (ret <= 0) {
ERR("Connections seems to be closed");
ret = -1;
data_size = be32toh(data_hdr.data_size);
if (data_buffer_size < data_size) {
- char *tmp_data_ptr = data_buffer;
- data_buffer = realloc(data_buffer, data_size);
- if (!data_buffer) {
+ char *tmp_data_ptr;
+
+ tmp_data_ptr = realloc(data_buffer, data_size);
+ if (!tmp_data_ptr) {
ERR("Allocating data buffer");
- free(tmp_data_ptr);
+ free(data_buffer);
ret = -1;
goto end_unlock;
}
+ data_buffer = tmp_data_ptr;
data_buffer_size = data_size;
}
memset(data_buffer, 0, data_size);
DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64,
data_size, stream_id, net_seq_num);
- ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, MSG_WAITALL);
+ ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0);
if (ret <= 0) {
ret = -1;
goto end_unlock;
/* Check if we need to close the FD */
if (close_stream_check(stream)) {
+ int cret;
struct lttng_ht_iter iter;
- close(stream->fd);
+ cret = close(stream->fd);
+ if (cret < 0) {
+ PERROR("close stream process data");
+ }
iter.iter.node = &stream->stream_n.node;
ret = lttng_ht_del(streams_ht, &iter);
assert(!ret);
}
while (1) {
- /* Zeroed the events structure */
- lttng_poll_reset(&events);
-
- nb_fd = LTTNG_POLL_GETNB(&events);
-
/* Infinite blocking call, waiting for transmission */
restart:
DBG3("Relayd worker thread polling...");
goto error;
}
+ nb_fd = ret;
+
for (i = 0; i < nb_fd; i++) {
/* Fetch once the poll data */
revents = LTTNG_POLL_GETEV(&events, i);
if (relay_connection->type == RELAY_CONTROL) {
ret = relay_connection->sock->ops->recvmsg(
relay_connection->sock, &recv_hdr,
- sizeof(struct lttcomm_relayd_hdr), MSG_WAITALL);
+ sizeof(struct lttcomm_relayd_hdr), 0);
/* connection closed */
if (ret <= 0) {
relay_cleanup_poll_connection(&events, pollfd);
streams_ht_error:
lttng_ht_destroy(relay_connections_ht);
relay_connections_ht_error:
+ /* Close relay cmd pipes */
+ utils_close_pipe(relay_cmd_pipe);
if (err) {
DBG("Thread exited with error");
}
/* Parse arguments */
progname = argv[0];
- if ((ret = parse_args(argc, argv) < 0)) {
+ if ((ret = parse_args(argc, argv)) < 0) {
goto exit;
}
projects / lttng-tools.git / blobdiff