* Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
*/
-#define _GNU_SOURCE
+#define _LGPL_SOURCE
#include <assert.h>
#include <stdio.h>
#include <stdlib.h>
#include <common/common.h>
#include <common/defaults.h>
+#include <common/compat/endian.h>
#include <common/sessiond-comm/relayd.h>
-#include <common/index/lttng-index.h>
+#include <common/index/ctf-index.h>
#include "relayd.h"
goto alloc_error;
}
+ memset(&header, 0, sizeof(header));
header.cmd = htobe32(cmd);
header.data_size = htobe64(size);
memcpy(buf + sizeof(header), data, size);
}
+ DBG3("Relayd sending command %d of size %" PRIu64, (int) cmd, buf_size);
ret = rsock->sock.ops->sendmsg(&rsock->sock, buf, buf_size, flags);
if (ret < 0) {
+ PERROR("Failed to send command %d of size %" PRIu64,
+ (int) cmd, buf_size);
ret = -errno;
goto error;
}
-
- DBG3("Relayd sending command %d of size %" PRIu64, cmd, buf_size);
-
error:
free(buf);
alloc_error:
return ret;
}
+static int recv_reply_ignore(struct lttcomm_relayd_sock *rsock, size_t size)
+{
+ int ret;
+ size_t cutfoff = 128;
+
+ /*
+ * To prevent ever growing size of recv_reply to ignore, if the number
+ * of bytes we want to ignore is bigger than `cutoff`, consume half of
+ * the cutoff. We might block on it but still, most of bytes to ignore
+ * should already be ready to consume at this point.
+ *
+ * This kind of scenario can easily happen on stopped session with a
+ * live_timer since no actual receive is done on the socket that would
+ * discard the `ignore` portion.
+ *
+ * TCP guarantee in-order transmission both on send and receive so this
+ * is safe to do.
+ */
+ if (rsock->bytes_to_ignore_on_recv >= cutfoff) {
+ size_t to_discard = cutfoff / 2;
+
+ ret = rsock->sock.ops->recvmsg(&rsock->sock, NULL, to_discard, MSG_TRUNC);
+ if (ret <= 0 || ret != to_discard) {
+ if (ret == 0) {
+ /* Orderly shutdown. */
+ DBG("Socket %d has performed an orderly shutdown", rsock->sock.fd);
+ } else {
+ DBG("Receiving reply to discard failed on sock %d for size %zu with ret %d",
+ rsock->sock.fd, to_discard, ret);
+ }
+ ret = -1;
+ goto error;
+ }
+
+ DBG("Force discard of %zu bytes for socket %d", to_discard, rsock->sock.fd);
+ rsock->bytes_to_ignore_on_recv -= to_discard;
+ }
+
+ DBG3("Relayd ignore reply of %zu bytes for socket %d.", size, rsock->sock.fd);
+ /* Do not wait for the current reply to be ignored */
+ rsock->bytes_to_ignore_on_recv += size;
+ ret = 0;
+
+error:
+ return ret;
+}
+
/*
* Receive reply data on socket. This MUST be call after send_command or else
* could result in unexpected behavior(s).
return -ECONNRESET;
}
+ /*
+ * We have to consume the bytes that are marked to ignore.
+ */
+ if (rsock->bytes_to_ignore_on_recv != 0) {
+ ret = rsock->sock.ops->recvmsg(&rsock->sock, NULL, rsock->bytes_to_ignore_on_recv, MSG_TRUNC);
+ if (ret <= 0 || ret != rsock->bytes_to_ignore_on_recv) {
+ if (ret == 0) {
+ /* Orderly shutdown. */
+ DBG("Socket %d has performed an orderly shutdown", rsock->sock.fd);
+ } else {
+ DBG("Receiving reply to skip failed on sock %d for size %zu with ret %d",
+ rsock->sock.fd, rsock->bytes_to_ignore_on_recv, ret);
+ }
+ ret = -1;
+ goto error;
+ }
+ DBG("Discarded %zu bytes on sock %d", rsock->bytes_to_ignore_on_recv, rsock->sock.fd);
+ rsock->bytes_to_ignore_on_recv = 0;
+ }
+
DBG3("Relayd waiting for reply of size %zu", size);
ret = rsock->sock.ops->recvmsg(&rsock->sock, data, size, 0);
*/
static int relayd_create_session_2_4(struct lttcomm_relayd_sock *rsock,
uint64_t *session_id, char *session_name, char *hostname,
- int session_live_timer)
+ int session_live_timer, unsigned int snapshot)
{
int ret;
struct lttcomm_relayd_create_session_2_4 msg;
- strncpy(msg.session_name, session_name, sizeof(msg.session_name));
- strncpy(msg.hostname, hostname, sizeof(msg.hostname));
+ if (lttng_strncpy(msg.session_name, session_name,
+ sizeof(msg.session_name))) {
+ ret = -1;
+ goto error;
+ }
+ if (lttng_strncpy(msg.hostname, hostname, sizeof(msg.hostname))) {
+ ret = -1;
+ goto error;
+ }
msg.live_timer = htobe32(session_live_timer);
+ msg.snapshot = htobe32(snapshot);
/* Send command */
ret = send_command(rsock, RELAYD_CREATE_SESSION, &msg, sizeof(msg), 0);
* a lttng error code from the relayd.
*/
int relayd_create_session(struct lttcomm_relayd_sock *rsock, uint64_t *session_id,
- char *session_name, char *hostname, int session_live_timer)
+ char *session_name, char *hostname, int session_live_timer,
+ unsigned int snapshot)
{
int ret;
struct lttcomm_relayd_status_session reply;
case 2:
case 3:
ret = relayd_create_session_2_1(rsock, session_id);
+ break;
case 4:
default:
- ret = relayd_create_session_2_4(rsock, session_id,
- session_name, hostname,
- session_live_timer);
+ ret = relayd_create_session_2_4(rsock, session_id, session_name,
+ hostname, session_live_timer, snapshot);
+ break;
}
if (ret < 0) {
/* Compat with relayd 2.1 */
if (rsock->minor == 1) {
- strncpy(msg.channel_name, channel_name, sizeof(msg.channel_name));
- strncpy(msg.pathname, pathname, sizeof(msg.pathname));
+ memset(&msg, 0, sizeof(msg));
+ if (lttng_strncpy(msg.channel_name, channel_name,
+ sizeof(msg.channel_name))) {
+ ret = -1;
+ goto error;
+ }
+ if (lttng_strncpy(msg.pathname, pathname,
+ sizeof(msg.pathname))) {
+ ret = -1;
+ goto error;
+ }
/* Send command */
ret = send_command(rsock, RELAYD_ADD_STREAM, (void *) &msg, sizeof(msg), 0);
goto error;
}
} else {
+ memset(&msg_2_2, 0, sizeof(msg_2_2));
/* Compat with relayd 2.2+ */
- strncpy(msg_2_2.channel_name, channel_name, sizeof(msg_2_2.channel_name));
- strncpy(msg_2_2.pathname, pathname, sizeof(msg_2_2.pathname));
+ if (lttng_strncpy(msg_2_2.channel_name, channel_name,
+ sizeof(msg_2_2.channel_name))) {
+ ret = -1;
+ goto error;
+ }
+ if (lttng_strncpy(msg_2_2.pathname, pathname,
+ sizeof(msg_2_2.pathname))) {
+ ret = -1;
+ goto error;
+ }
msg_2_2.tracefile_size = htobe64(tracefile_size);
msg_2_2.tracefile_count = htobe64(tracefile_count);
return ret;
}
+/*
+ * Add stream on the relayd. Send part.
+ *
+ * On success return 0 else return ret_code negative value.
+ */
+int relayd_add_stream_send(struct lttcomm_relayd_sock *rsock, const char *channel_name,
+ const char *pathname, uint64_t tracefile_size, uint64_t tracefile_count)
+{
+ int ret;
+ struct lttcomm_relayd_add_stream msg;
+ struct lttcomm_relayd_add_stream_2_2 msg_2_2;
+
+ /* Code flow error. Safety net. */
+ assert(rsock);
+ assert(channel_name);
+ assert(pathname);
+
+ DBG("Relayd adding stream for channel name %s. Part send", channel_name);
+
+ /* Compat with relayd 2.1 */
+ if (rsock->minor == 1) {
+ memset(&msg, 0, sizeof(msg));
+ if (lttng_strncpy(msg.channel_name, channel_name,
+ sizeof(msg.channel_name))) {
+ ret = -1;
+ goto error;
+ }
+ if (lttng_strncpy(msg.pathname, pathname,
+ sizeof(msg.pathname))) {
+ ret = -1;
+ goto error;
+ }
+
+ /* Send command */
+ ret = send_command(rsock, RELAYD_ADD_STREAM, (void *) &msg, sizeof(msg), 0);
+ if (ret < 0) {
+ goto error;
+ }
+ } else {
+ memset(&msg_2_2, 0, sizeof(msg_2_2));
+ /* Compat with relayd 2.2+ */
+ if (lttng_strncpy(msg_2_2.channel_name, channel_name,
+ sizeof(msg_2_2.channel_name))) {
+ ret = -1;
+ goto error;
+ }
+ if (lttng_strncpy(msg_2_2.pathname, pathname,
+ sizeof(msg_2_2.pathname))) {
+ ret = -1;
+ goto error;
+ }
+ msg_2_2.tracefile_size = htobe64(tracefile_size);
+ msg_2_2.tracefile_count = htobe64(tracefile_count);
+
+ /* Send command */
+ ret = send_command(rsock, RELAYD_ADD_STREAM, (void *) &msg_2_2, sizeof(msg_2_2), 0);
+ if (ret < 0) {
+ goto error;
+ }
+ }
+
+ DBG("Relayd add stream sent for channel name %s.", channel_name);
+ ret = 0;
+
+error:
+ return ret;
+}
+
+int relayd_add_stream_rcv(struct lttcomm_relayd_sock *rsock, uint64_t *_stream_id)
+{
+ int ret;
+ struct lttcomm_relayd_status_stream reply;
+
+ /* Code flow error. Safety net. */
+ assert(rsock);
+
+ /* Waiting for reply */
+ ret = recv_reply(rsock, (void *) &reply, sizeof(reply));
+ if (ret < 0) {
+ goto error;
+ }
+
+ /* Back to host bytes order. */
+ reply.handle = be64toh(reply.handle);
+ reply.ret_code = be32toh(reply.ret_code);
+
+ /* Return session id or negative ret code. */
+ if (reply.ret_code != LTTNG_OK) {
+ ret = -1;
+ ERR("Relayd add stream replied error %d", reply.ret_code);
+ } else {
+ /* Success */
+ ret = 0;
+ *_stream_id = reply.handle;
+ }
+
+ DBG("Relayd stream added successfully with handle %" PRIu64,
+ reply.handle);
+
+error:
+ return ret;
+}
+
+/*
+ * Inform the relay that all the streams for the current channel has been sent.
+ *
+ * On success return 0 else return ret_code negative value.
+ */
+int relayd_streams_sent(struct lttcomm_relayd_sock *rsock)
+{
+ int ret;
+ struct lttcomm_relayd_generic_reply reply;
+
+ /* Code flow error. Safety net. */
+ assert(rsock);
+
+ DBG("Relayd sending streams sent.");
+
+ /* This feature was introduced in 2.4, ignore it for earlier versions. */
+ if (rsock->minor < 4) {
+ ret = 0;
+ goto end;
+ }
+
+ /* Send command */
+ ret = send_command(rsock, RELAYD_STREAMS_SENT, NULL, 0, 0);
+ if (ret < 0) {
+ goto error;
+ }
+
+ /* Waiting for reply */
+ ret = recv_reply(rsock, (void *) &reply, sizeof(reply));
+ if (ret < 0) {
+ goto error;
+ }
+
+ /* Back to host bytes order. */
+ reply.ret_code = be32toh(reply.ret_code);
+
+ /* Return session id or negative ret code. */
+ if (reply.ret_code != LTTNG_OK) {
+ ret = -1;
+ ERR("Relayd streams sent replied error %d", reply.ret_code);
+ goto error;
+ } else {
+ /* Success */
+ ret = 0;
+ }
+
+ DBG("Relayd streams sent success");
+
+error:
+end:
+ return ret;
+}
+
/*
* Check version numbers on the relayd.
* If major versions are compatible, we assign minor_to_use to the
* minor version of the procotol we are going to use for this session.
*
- * Return 0 if compatible else negative value.
+ * Return 0 if the two daemons are compatible, LTTNG_ERR_RELAYD_VERSION_FAIL
+ * otherwise, or a negative value on network errors.
*/
int relayd_version_check(struct lttcomm_relayd_sock *rsock)
{
DBG("Relayd version check for major.minor %u.%u", rsock->major,
rsock->minor);
+ memset(&msg, 0, sizeof(msg));
/* Prepare network byte order before transmission. */
msg.major = htobe32(rsock->major);
msg.minor = htobe32(rsock->minor);
*/
if (msg.major != rsock->major) {
/* Not compatible */
- ret = -1;
+ ret = LTTNG_ERR_RELAYD_VERSION_FAIL;
DBG2("Relayd version is NOT compatible. Relayd version %u != %u (us)",
msg.major, rsock->major);
goto error;
goto end;
}
+ /*
+ * This ensure that we do not close the socket while the lttng-relayd
+ * expects to be able to send a response that we skipped.
+ * While we loose some time to receive everything, this keep the
+ * protocol intact from the point of view of lttng-relayd.
+ */
+ if (rsock->bytes_to_ignore_on_recv != 0) {
+ ret = rsock->sock.ops->recvmsg(&rsock->sock, NULL, rsock->bytes_to_ignore_on_recv, MSG_TRUNC);
+ if (ret <= 0 || ret != rsock->bytes_to_ignore_on_recv) {
+ if (ret == 0) {
+ /* Orderly shutdown. */
+ DBG("Socket %d has performed an orderly shutdown", rsock->sock.fd);
+ } else {
+ DBG("Receiving reply to skip failed on sock %d for size %zu with ret %d",
+ rsock->sock.fd, rsock->bytes_to_ignore_on_recv, ret);
+ }
+ }
+ DBG("Discarded %zu bytes on sock %d", rsock->bytes_to_ignore_on_recv, rsock->sock.fd);
+ rsock->bytes_to_ignore_on_recv = 0;
+ }
+
DBG3("Relayd closing socket %d", rsock->sock.fd);
if (rsock->sock.ops) {
DBG("Relayd closing stream id %" PRIu64, stream_id);
+ memset(&msg, 0, sizeof(msg));
msg.stream_id = htobe64(stream_id);
msg.last_net_seq_num = htobe64(last_net_seq_num);
goto error;
}
- /* Receive response */
- ret = recv_reply(rsock, (void *) &reply, sizeof(reply));
+ /*
+ * Discard response since we do not really care for it and that TCP
+ * guarantee in-order delivery. As for error handling, there is not much
+ * to do at this point (closing).
+ **/
+ ret = recv_reply_ignore(rsock, sizeof(reply));
if (ret < 0) {
goto error;
}
- reply.ret_code = be32toh(reply.ret_code);
-
- /* Return session id or negative ret code. */
- if (reply.ret_code != LTTNG_OK) {
- ret = -1;
- ERR("Relayd close stream replied error %d", reply.ret_code);
- } else {
- /* Success */
- ret = 0;
- }
-
DBG("Relayd close stream id %" PRIu64 " successfully", stream_id);
error:
DBG("Relayd data pending for stream id %" PRIu64, stream_id);
+ memset(&msg, 0, sizeof(msg));
msg.stream_id = htobe64(stream_id);
msg.last_net_seq_num = htobe64(last_net_seq_num);
DBG("Relayd checking quiescent control state");
+ memset(&msg, 0, sizeof(msg));
msg.stream_id = htobe64(metadata_stream_id);
/* Send command */
DBG("Relayd begin data pending");
+ memset(&msg, 0, sizeof(msg));
msg.session_id = htobe64(id);
/* Send command */
int relayd_end_data_pending(struct lttcomm_relayd_sock *rsock, uint64_t id,
unsigned int *is_data_inflight)
{
- int ret;
+ int ret, recv_ret;
struct lttcomm_relayd_end_data_pending msg;
struct lttcomm_relayd_generic_reply reply;
DBG("Relayd end data pending");
+ memset(&msg, 0, sizeof(msg));
msg.session_id = htobe64(id);
/* Send command */
goto error;
}
- reply.ret_code = be32toh(reply.ret_code);
- if (reply.ret_code < 0) {
- ret = reply.ret_code;
+ recv_ret = be32toh(reply.ret_code);
+ if (recv_ret < 0) {
+ ret = recv_ret;
goto error;
}
- *is_data_inflight = reply.ret_code;
+ *is_data_inflight = recv_ret;
- DBG("Relayd end data pending is data inflight: %d", reply.ret_code);
+ DBG("Relayd end data pending is data inflight: %d", recv_ret);
return 0;
* Send index to the relayd.
*/
int relayd_send_index(struct lttcomm_relayd_sock *rsock,
- struct lttng_packet_index *index, uint64_t relay_stream_id,
+ struct ctf_packet_index *index, uint64_t relay_stream_id,
uint64_t net_seq_num)
{
int ret;
DBG("Relayd sending index for stream ID %" PRIu64, relay_stream_id);
+ memset(&msg, 0, sizeof(msg));
msg.relay_stream_id = htobe64(relay_stream_id);
msg.net_seq_num = htobe64(net_seq_num);
msg.events_discarded = index->events_discarded;
msg.stream_id = index->stream_id;
+ if (rsock->minor >= 8) {
+ msg.stream_instance_id = index->stream_instance_id;
+ msg.packet_seq_num = index->packet_seq_num;
+ }
+
/* Send command */
- ret = send_command(rsock, RELAYD_SEND_INDEX, &msg, sizeof(msg), 0);
+ ret = send_command(rsock, RELAYD_SEND_INDEX, &msg,
+ lttcomm_relayd_index_len(lttng_to_index_major(rsock->major,
+ rsock->minor),
+ lttng_to_index_minor(rsock->major, rsock->minor)),
+ 0);
+ if (ret < 0) {
+ goto error;
+ }
+
+ /*
+ * Ignore the response. TCP guarantee in-order arrival and the overall
+ * protocol do not rely on hard ordering between the control and data
+ * socket for index.
+ * Indexes are sent either at the end of the buffer consumption or
+ * during the live timer.
+ */
+ ret = recv_reply_ignore(rsock, sizeof(reply));
+ if (ret < 0) {
+ goto error;
+ }
+
+error:
+ return ret;
+}
+
+/*
+ * Ask the relay to reset the metadata trace file (regeneration).
+ */
+int relayd_reset_metadata(struct lttcomm_relayd_sock *rsock,
+ uint64_t stream_id, uint64_t version)
+{
+ int ret;
+ struct lttcomm_relayd_reset_metadata msg;
+ struct lttcomm_relayd_generic_reply reply;
+
+ /* Code flow error. Safety net. */
+ assert(rsock);
+
+ /* Should have been prevented by the sessiond. */
+ if (rsock->minor < 8) {
+ ERR("Metadata regeneration unsupported before 2.8");
+ ret = -1;
+ goto error;
+ }
+
+ DBG("Relayd reset metadata stream id %" PRIu64, stream_id);
+
+ memset(&msg, 0, sizeof(msg));
+ msg.stream_id = htobe64(stream_id);
+ msg.version = htobe64(version);
+
+ /* Send command */
+ ret = send_command(rsock, RELAYD_RESET_METADATA, (void *) &msg, sizeof(msg), 0);
if (ret < 0) {
goto error;
}
/* Return session id or negative ret code. */
if (reply.ret_code != LTTNG_OK) {
ret = -1;
- ERR("Relayd send index replied error %d", reply.ret_code);
+ ERR("Relayd reset metadata replied error %d", reply.ret_code);
} else {
/* Success */
ret = 0;
}
+ DBG("Relayd reset metadata stream id %" PRIu64 " successfully", stream_id);
+
error:
return ret;
}