#include <sys/stat.h>
#include <sys/types.h>
#include <sys/wait.h>
+#include <sys/resource.h>
#include <inttypes.h>
#include <urcu/futex.h>
#include <urcu/uatomic.h>
+#include <urcu/rculist.h>
#include <unistd.h>
#include <fcntl.h>
+#include <strings.h>
+#include <ctype.h>
#include <lttng/lttng.h>
#include <common/common.h>
#include <common/sessiond-comm/relayd.h>
#include <common/uri.h>
#include <common/utils.h>
+#include <common/align.h>
#include <common/config/session-config.h>
-#include <urcu/rculist.h>
+#include <common/dynamic-buffer.h>
+#include <common/buffer-view.h>
+#include <common/string-utils/format.h>
+#include <common/fd-tracker/fd-tracker.h>
+#include <common/fd-tracker/utils.h>
+#include "backward-compatibility-group-by.h"
#include "cmd.h"
+#include "connection.h"
#include "ctf-trace.h"
+#include "health-relayd.h"
#include "index.h"
-#include "utils.h"
-#include "lttng-relayd.h"
#include "live.h"
-#include "health-relayd.h"
-#include "testpoint.h"
-#include "viewer-stream.h"
+#include "lttng-relayd.h"
#include "session.h"
+#include "sessiond-trace-chunks.h"
#include "stream.h"
-#include "connection.h"
+#include "tcp_keep_alive.h"
+#include "testpoint.h"
#include "tracefile-array.h"
+#include "utils.h"
+#include "version.h"
+#include "viewer-stream.h"
+
+static const char *help_msg =
+#ifdef LTTNG_EMBED_HELP
+#include <lttng-relayd.8.h>
+#else
+NULL
+#endif
+;
+
+enum relay_connection_status {
+ RELAY_CONNECTION_STATUS_OK,
+ /* An error occurred while processing an event on the connection. */
+ RELAY_CONNECTION_STATUS_ERROR,
+ /* Connection closed/shutdown cleanly. */
+ RELAY_CONNECTION_STATUS_CLOSED,
+};
/* command line options */
-char *opt_output_path;
-static int opt_daemon, opt_background;
+char *opt_output_path, *opt_working_directory;
+static int opt_daemon, opt_background, opt_print_version, opt_allow_clear = 1;
+enum relay_group_output_by opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_UNKNOWN;
/*
* We need to wait for listener and live listener threads, as well as
*/
static struct relay_conn_queue relay_conn_queue;
-/* buffer allocated at startup, used to store the trace data */
-static char *data_buffer;
-static unsigned int data_buffer_size;
+/* Cap of file desriptors to be in simultaneous use by the relay daemon. */
+static unsigned int lttng_opt_fd_pool_size = -1;
/* Global relay stream hash table. */
struct lttng_ht *relay_streams_ht;
/* Relayd health monitoring */
struct health_app *health_relayd;
+struct sessiond_trace_chunk_registry *sessiond_trace_chunk_registry;
+
+/* Global fd tracker. */
+struct fd_tracker *the_fd_tracker;
+
static struct option long_options[] = {
{ "control-port", 1, 0, 'C', },
{ "data-port", 1, 0, 'D', },
{ "daemonize", 0, 0, 'd', },
{ "background", 0, 0, 'b', },
{ "group", 1, 0, 'g', },
+ { "fd-pool-size", 1, 0, '\0', },
{ "help", 0, 0, 'h', },
{ "output", 1, 0, 'o', },
{ "verbose", 0, 0, 'v', },
{ "config", 1, 0, 'f' },
+ { "version", 0, 0, 'V' },
+ { "working-directory", 1, 0, 'w', },
+ { "group-output-by-session", 0, 0, 's', },
+ { "group-output-by-host", 0, 0, 'p', },
+ { "disallow-clear", 0, 0, 'x' },
{ NULL, 0, 0, 0, },
};
-static const char *config_ignore_options[] = { "help", "config" };
+static const char *config_ignore_options[] = { "help", "config", "version" };
+
+static void print_version(void) {
+ fprintf(stdout, "%s\n", VERSION);
+}
+
+static void relayd_config_log(void)
+{
+ DBG("LTTng-relayd " VERSION " - " VERSION_NAME "%s%s",
+ GIT_VERSION[0] == '\0' ? "" : " - " GIT_VERSION,
+ EXTRA_VERSION_NAME[0] == '\0' ? "" : " - " EXTRA_VERSION_NAME);
+ if (EXTRA_VERSION_DESCRIPTION[0] != '\0') {
+ DBG("LTTng-relayd extra version description:\n\t" EXTRA_VERSION_DESCRIPTION "\n");
+ }
+ if (EXTRA_VERSION_PATCHES[0] != '\0') {
+ DBG("LTTng-relayd extra patches:\n\t" EXTRA_VERSION_PATCHES "\n");
+ }
+}
/*
* Take an option from the getopt output and set it in the right variable to be
switch (opt) {
case 0:
- fprintf(stderr, "option %s", optname);
- if (arg) {
- fprintf(stderr, " with arg %s\n", arg);
+ if (!strcmp(optname, "fd-pool-size")) {
+ unsigned long v;
+
+ errno = 0;
+ v = strtoul(arg, NULL, 0);
+ if (errno != 0 || !isdigit(arg[0])) {
+ ERR("Wrong value in --fd-pool-size parameter: %s", arg);
+ ret = -1;
+ goto end;
+ }
+ if (v >= UINT_MAX) {
+ ERR("File descriptor cap overflow in --fd-pool-size parameter: %s", arg);
+ ret = -1;
+ goto end;
+ }
+ lttng_opt_fd_pool_size = (unsigned int) v;
+ } else {
+ fprintf(stderr, "unknown option %s", optname);
+ if (arg) {
+ fprintf(stderr, " with arg %s\n", arg);
+ }
}
break;
case 'C':
}
break;
case 'h':
- ret = utils_show_man_page(8, "lttng-relayd");
+ ret = utils_show_help(8, "lttng-relayd", help_msg);
if (ret) {
- ERR("Cannot view man page lttng-relayd(8)");
+ ERR("Cannot show --help for `lttng-relayd`");
perror("exec");
}
exit(EXIT_FAILURE);
+ case 'V':
+ opt_print_version = 1;
+ break;
case 'o':
if (lttng_is_setuid_setgid()) {
WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
}
}
break;
+ case 'w':
+ if (lttng_is_setuid_setgid()) {
+ WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
+ "-w, --working-directory");
+ } else {
+ ret = asprintf(&opt_working_directory, "%s", arg);
+ if (ret < 0) {
+ ret = -errno;
+ PERROR("asprintf opt_working_directory");
+ goto end;
+ }
+ }
+ break;
+
case 'v':
/* Verbose level can increase using multiple -v */
if (arg) {
}
}
break;
+ case 's':
+ if (opt_group_output_by != RELAYD_GROUP_OUTPUT_BY_UNKNOWN) {
+ ERR("Cannot set --group-output-by-session, another --group-output-by argument is present");
+ exit(EXIT_FAILURE);
+ }
+ opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_SESSION;
+ break;
+ case 'p':
+ if (opt_group_output_by != RELAYD_GROUP_OUTPUT_BY_UNKNOWN) {
+ ERR("Cannot set --group-output-by-host, another --group-output-by argument is present");
+ exit(EXIT_FAILURE);
+ }
+ opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_HOST;
+ break;
+ case 'x':
+ /* Disallow clear */
+ opt_allow_clear = 0;
+ break;
default:
/* Unknown option or other error.
* Error is printed by getopt, just return */
return ret;
}
+static int parse_env_options(void)
+{
+ int ret = 0;
+ char *value = NULL;
+
+ value = lttng_secure_getenv(DEFAULT_LTTNG_RELAYD_WORKING_DIRECTORY_ENV);
+ if (value) {
+ opt_working_directory = strdup(value);
+ if (!opt_working_directory) {
+ ERR("Failed to allocate working directory string (\"%s\")",
+ value);
+ ret = -1;
+ }
+ }
+ return ret;
+}
+
+static int set_fd_pool_size(void)
+{
+ int ret = 0;
+ struct rlimit rlimit;
+
+ ret = getrlimit(RLIMIT_NOFILE, &rlimit);
+ if (ret) {
+ PERROR("Failed to get file descriptor limit");
+ ret = -1;
+ goto end;
+ }
+
+ DBG("File descriptor count limits are %" PRIu64 " (soft) and %" PRIu64 " (hard)",
+ (uint64_t) rlimit.rlim_cur,
+ (uint64_t) rlimit.rlim_max);
+ if (lttng_opt_fd_pool_size == -1) {
+ /* Use default value (soft limit - reserve). */
+ if (rlimit.rlim_cur < DEFAULT_RELAYD_MIN_FD_POOL_SIZE) {
+ ERR("The process' file number limit is too low (%" PRIu64 "). The process' file number limit must be set to at least %i.",
+ (uint64_t) rlimit.rlim_cur, DEFAULT_RELAYD_MIN_FD_POOL_SIZE);
+ ret = -1;
+ goto end;
+ }
+ lttng_opt_fd_pool_size = rlimit.rlim_cur -
+ DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE;
+ goto end;
+ }
+
+ if (lttng_opt_fd_pool_size < DEFAULT_RELAYD_MIN_FD_POOL_SIZE) {
+ ERR("File descriptor pool size must be set to at least %d",
+ DEFAULT_RELAYD_MIN_FD_POOL_SIZE);
+ ret = -1;
+ goto end;
+ }
+
+ if (lttng_opt_fd_pool_size > rlimit.rlim_cur) {
+ ERR("File descriptor pool size argument (%u) exceeds the process' soft limit (%" PRIu64 ").",
+ lttng_opt_fd_pool_size, (uint64_t) rlimit.rlim_cur);
+ ret = -1;
+ goto end;
+ }
+
+ DBG("File descriptor pool size argument (%u) adjusted to %u to accomodate transient fd uses",
+ lttng_opt_fd_pool_size,
+ lttng_opt_fd_pool_size - DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE);
+ lttng_opt_fd_pool_size -= DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE;
+end:
+ return ret;
+}
+
static int set_options(int argc, char **argv)
{
int c, ret = 0, option_index = 0, retval = 0;
goto exit;
}
}
+ ret = set_fd_pool_size();
+ if (ret) {
+ retval = -1;
+ goto exit;
+ }
+
+ if (opt_group_output_by == RELAYD_GROUP_OUTPUT_BY_UNKNOWN) {
+ opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_HOST;
+ }
+ if (opt_allow_clear) {
+ /* Check if env variable exists. */
+ const char *value = lttng_secure_getenv(DEFAULT_LTTNG_RELAYD_DISALLOW_CLEAR_ENV);
+ if (value) {
+ ret = config_parse_value(value);
+ if (ret < 0) {
+ ERR("Invalid value for %s specified", DEFAULT_LTTNG_RELAYD_DISALLOW_CLEAR_ENV);
+ retval = -1;
+ goto exit;
+ }
+ opt_allow_clear = !ret;
+ }
+ }
exit:
free(optstring);
static void print_global_objects(void)
{
- rcu_register_thread();
-
print_viewer_streams();
print_relay_streams();
print_sessions();
+}
- rcu_unregister_thread();
+static int noop_close(void *data, int *fds)
+{
+ return 0;
+}
+
+static void untrack_stdio(void)
+{
+ int fds[] = { fileno(stdout), fileno(stderr) };
+
+ /*
+ * noop_close is used since we don't really want to close
+ * the stdio output fds; we merely want to stop tracking them.
+ */
+ (void) fd_tracker_close_unsuspendable_fd(the_fd_tracker,
+ fds, 2, noop_close, NULL);
}
/*
if (sessions_ht)
lttng_ht_destroy(sessions_ht);
- /* free the dynamically allocated opt_output_path */
free(opt_output_path);
+ free(opt_working_directory);
+ if (health_relayd) {
+ health_app_destroy(health_relayd);
+ }
/* Close thread quit pipes */
- utils_close_pipe(thread_quit_pipe);
+ if (health_quit_pipe[0] != -1) {
+ (void) fd_tracker_util_pipe_close(
+ the_fd_tracker, health_quit_pipe);
+ }
+ if (thread_quit_pipe[0] != -1) {
+ (void) fd_tracker_util_pipe_close(
+ the_fd_tracker, thread_quit_pipe);
+ }
+ if (sessiond_trace_chunk_registry) {
+ sessiond_trace_chunk_registry_destroy(
+ sessiond_trace_chunk_registry);
+ }
+ if (the_fd_tracker) {
+ untrack_stdio();
+ /*
+ * fd_tracker_destroy() will log the contents of the fd-tracker
+ * if a leak is detected.
+ */
+ fd_tracker_destroy(the_fd_tracker);
+ }
uri_free(control_uri);
uri_free(data_uri);
static void sighandler(int sig)
{
switch (sig) {
- case SIGPIPE:
- DBG("SIGPIPE caught");
- return;
case SIGINT:
DBG("SIGINT caught");
if (lttng_relay_stop_threads()) {
return ret;
}
- sa.sa_handler = sighandler;
sa.sa_mask = sigset;
sa.sa_flags = 0;
+
+ sa.sa_handler = sighandler;
if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
PERROR("sigaction");
return ret;
return ret;
}
- if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
+ if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
PERROR("sigaction");
return ret;
}
- if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
+ sa.sa_handler = SIG_IGN;
+ if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
PERROR("sigaction");
return ret;
}
*/
static int init_thread_quit_pipe(void)
{
- int ret;
-
- ret = utils_create_pipe_cloexec(thread_quit_pipe);
+ return fd_tracker_util_pipe_open_cloexec(
+ the_fd_tracker, "Quit pipe", thread_quit_pipe);
+}
- return ret;
+/*
+ * Init health quit pipe.
+ *
+ * Return -1 on error or 0 if all pipes are created.
+ */
+static int init_health_quit_pipe(void)
+{
+ return fd_tracker_util_pipe_open_cloexec(the_fd_tracker,
+ "Health quit pipe", health_quit_pipe);
}
/*
* Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
*/
-static int create_thread_poll_set(struct lttng_poll_event *events, int size)
+static int create_named_thread_poll_set(struct lttng_poll_event *events,
+ int size, const char *name)
{
int ret;
goto error;
}
- ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
- if (ret < 0) {
- goto error;
- }
+ ret = fd_tracker_util_poll_create(the_fd_tracker,
+ name, events, 1, LTTNG_CLOEXEC);
/* Add quit pipe */
ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
return 0;
}
+static int create_sock(void *data, int *out_fd)
+{
+ int ret;
+ struct lttcomm_sock *sock = data;
+
+ ret = lttcomm_create_sock(sock);
+ if (ret < 0) {
+ goto end;
+ }
+
+ *out_fd = sock->fd;
+end:
+ return ret;
+}
+
+static int close_sock(void *data, int *in_fd)
+{
+ struct lttcomm_sock *sock = data;
+
+ return sock->ops->close(sock);
+}
+
+static int accept_sock(void *data, int *out_fd)
+{
+ int ret = 0;
+ /* Socks is an array of in_sock, out_sock. */
+ struct lttcomm_sock **socks = data;
+ struct lttcomm_sock *in_sock = socks[0];
+
+ socks[1] = in_sock->ops->accept(in_sock);
+ if (!socks[1]) {
+ ret = -1;
+ goto end;
+ }
+ *out_fd = socks[1]->fd;
+end:
+ return ret;
+}
+
/*
* Create and init socket from uri.
*/
-static struct lttcomm_sock *relay_socket_create(struct lttng_uri *uri)
+static struct lttcomm_sock *relay_socket_create(struct lttng_uri *uri,
+ const char *name)
{
- int ret;
+ int ret, sock_fd;
struct lttcomm_sock *sock = NULL;
+ char uri_str[PATH_MAX];
+ char *formated_name = NULL;
sock = lttcomm_alloc_sock_from_uri(uri);
if (sock == NULL) {
goto error;
}
- ret = lttcomm_create_sock(sock);
- if (ret < 0) {
- goto error;
+ /*
+ * Don't fail to create the socket if the name can't be built as it is
+ * only used for debugging purposes.
+ */
+ ret = uri_to_str_url(uri, uri_str, sizeof(uri_str));
+ uri_str[sizeof(uri_str) - 1] = '\0';
+ if (ret >= 0) {
+ ret = asprintf(&formated_name, "%s socket @ %s", name,
+ uri_str);
+ if (ret < 0) {
+ formated_name = NULL;
+ }
}
- DBG("Listening on sock %d", sock->fd);
+
+ ret = fd_tracker_open_unsuspendable_fd(the_fd_tracker, &sock_fd,
+ (const char **) (formated_name ? &formated_name : NULL),
+ 1, create_sock, sock);
+ free(formated_name);
+ DBG("Listening on %s socket %d", name, sock->fd);
ret = sock->ops->bind(sock);
if (ret < 0) {
+ PERROR("Failed to bind socket");
goto error;
}
return NULL;
}
+static
+struct lttcomm_sock *accept_relayd_sock(struct lttcomm_sock *listening_sock,
+ const char *name)
+{
+ int out_fd, ret;
+ struct lttcomm_sock *socks[2] = { listening_sock, NULL };
+ struct lttcomm_sock *new_sock = NULL;
+
+ ret = fd_tracker_open_unsuspendable_fd(
+ the_fd_tracker, &out_fd,
+ (const char **) &name,
+ 1, accept_sock, &socks);
+ if (ret) {
+ goto end;
+ }
+ new_sock = socks[1];
+ DBG("%s accepted, socket %d", name, new_sock->fd);
+end:
+ return new_sock;
+}
+
/*
* This thread manages the listening for new connections on the network
*/
health_code_update();
- control_sock = relay_socket_create(control_uri);
+ control_sock = relay_socket_create(control_uri, "Control listener");
if (!control_sock) {
goto error_sock_control;
}
- data_sock = relay_socket_create(data_uri);
+ data_sock = relay_socket_create(data_uri, "Data listener");
if (!data_sock) {
goto error_sock_relay;
}
* Pass 3 as size here for the thread quit pipe, control and
* data socket.
*/
- ret = create_thread_poll_set(&events, 3);
+ ret = create_named_thread_poll_set(&events, 3, "Listener thread epoll");
if (ret < 0) {
goto error_create_poll;
}
revents = LTTNG_POLL_GETEV(&events, i);
pollfd = LTTNG_POLL_GETFD(&events, i);
- if (!revents) {
- /*
- * No activity for this FD (poll
- * implementation).
- */
- continue;
- }
-
/* Thread quit pipe has been closed. Killing thread. */
ret = check_thread_quit_pipe(pollfd, revents);
if (ret) {
*/
int val = 1;
struct relay_connection *new_conn;
- struct lttcomm_sock *newsock;
+ struct lttcomm_sock *newsock = NULL;
enum connection_type type;
if (pollfd == data_sock->fd) {
type = RELAY_DATA;
- newsock = data_sock->ops->accept(data_sock);
- DBG("Relay data connection accepted, socket %d",
- newsock->fd);
+ newsock = accept_relayd_sock(data_sock,
+ "Data socket to relayd");
} else {
assert(pollfd == control_sock->fd);
type = RELAY_CONTROL;
lttcomm_destroy_sock(newsock);
goto error;
}
+
+ ret = socket_apply_keep_alive_config(newsock->fd);
+ if (ret < 0) {
+ ERR("Failed to apply TCP keep-alive configuration on socket (%i)",
+ newsock->fd);
+ lttcomm_destroy_sock(newsock);
+ goto error;
+ }
+
new_conn = connection_create(newsock, type);
if (!new_conn) {
lttcomm_destroy_sock(newsock);
error:
error_poll_add:
error_testpoint:
- lttng_poll_clean(&events);
+ (void) fd_tracker_util_poll_clean(the_fd_tracker, &events);
error_create_poll:
if (data_sock->fd >= 0) {
- ret = data_sock->ops->close(data_sock);
+ int data_sock_fd = data_sock->fd;
+
+ ret = fd_tracker_close_unsuspendable_fd(the_fd_tracker,
+ &data_sock_fd, 1, close_sock,
+ data_sock);
if (ret) {
- PERROR("close");
+ PERROR("Failed to close the data listener socket file descriptor");
}
+ data_sock->fd = -1;
}
lttcomm_destroy_sock(data_sock);
error_sock_relay:
if (control_sock->fd >= 0) {
- ret = control_sock->ops->close(control_sock);
+ int control_sock_fd = control_sock->fd;
+
+ ret = fd_tracker_close_unsuspendable_fd(the_fd_tracker,
+ &control_sock_fd, 1, close_sock,
+ control_sock);
if (ret) {
- PERROR("close");
+ PERROR("Failed to close the control listener socket file descriptor");
}
+ control_sock->fd = -1;
}
lttcomm_destroy_sock(control_sock);
error_sock_control:
health_code_update();
- while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
+ for (;;) {
health_code_update();
/* Atomically prepare the queue futex */
futex_nto1_prepare(&relay_conn_queue.futex);
+ if (CMM_LOAD_SHARED(dispatch_thread_exit)) {
+ break;
+ }
+
do {
health_code_update();
return NULL;
}
-/*
- * Set index data from the control port to a given index object.
- */
-static int set_index_control_data(struct relay_index *index,
- struct lttcomm_relayd_index *data,
- struct relay_connection *conn)
+static bool session_streams_have_index(const struct relay_session *session)
{
- struct ctf_packet_index index_data;
-
- /*
- * The index on disk is encoded in big endian, so we don't need
- * to convert the data received on the network. The data_offset
- * value is NEVER modified here and is updated by the data
- * thread.
- */
- index_data.packet_size = data->packet_size;
- index_data.content_size = data->content_size;
- index_data.timestamp_begin = data->timestamp_begin;
- index_data.timestamp_end = data->timestamp_end;
- index_data.events_discarded = data->events_discarded;
- index_data.stream_id = data->stream_id;
-
- if (conn->minor >= 8) {
- index->index_data.stream_instance_id = data->stream_instance_id;
- index->index_data.packet_seq_num = data->packet_seq_num;
- }
-
- return relay_index_set_data(index, &index_data);
+ return session->minor >= 4 && !session->snapshot;
}
/*
*
* 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_connection *conn)
+static int relay_create_session(const struct lttcomm_relayd_hdr *recv_hdr,
+ struct relay_connection *conn,
+ const struct lttng_buffer_view *payload)
{
- int ret = 0, send_ret;
- struct relay_session *session;
- struct lttcomm_relayd_status_session reply;
- char session_name[LTTNG_NAME_MAX];
- char hostname[LTTNG_HOST_NAME_MAX];
+ int ret = 0;
+ ssize_t send_ret;
+ struct relay_session *session = NULL;
+ struct lttcomm_relayd_create_session_reply_2_11 reply = {};
+ char session_name[LTTNG_NAME_MAX] = {};
+ char hostname[LTTNG_HOST_NAME_MAX] = {};
uint32_t live_timer = 0;
bool snapshot = false;
-
- memset(session_name, 0, LTTNG_NAME_MAX);
- memset(hostname, 0, LTTNG_HOST_NAME_MAX);
-
- memset(&reply, 0, sizeof(reply));
-
- switch (conn->minor) {
- case 1:
- case 2:
- case 3:
- break;
- case 4: /* LTTng sessiond 2.4 */
- default:
- ret = cmd_create_session_2_4(conn, session_name,
+ bool session_name_contains_creation_timestamp = false;
+ /* Left nil for peers < 2.11. */
+ char base_path[LTTNG_PATH_MAX] = {};
+ lttng_uuid sessiond_uuid = {};
+ LTTNG_OPTIONAL(uint64_t) id_sessiond = {};
+ LTTNG_OPTIONAL(uint64_t) current_chunk_id = {};
+ LTTNG_OPTIONAL(time_t) creation_time = {};
+ struct lttng_dynamic_buffer reply_payload;
+
+ lttng_dynamic_buffer_init(&reply_payload);
+
+ if (conn->minor < 4) {
+ /* From 2.1 to 2.3 */
+ ret = 0;
+ } else if (conn->minor >= 4 && conn->minor < 11) {
+ /* From 2.4 to 2.10 */
+ ret = cmd_create_session_2_4(payload, session_name,
hostname, &live_timer, &snapshot);
+ } else {
+ bool has_current_chunk;
+ uint64_t current_chunk_id_value;
+ time_t creation_time_value;
+ uint64_t id_sessiond_value;
+
+ /* From 2.11 to ... */
+ ret = cmd_create_session_2_11(payload, session_name, hostname,
+ base_path, &live_timer, &snapshot, &id_sessiond_value,
+ sessiond_uuid, &has_current_chunk,
+ ¤t_chunk_id_value, &creation_time_value,
+ &session_name_contains_creation_timestamp);
+ if (lttng_uuid_is_nil(sessiond_uuid)) {
+ /* The nil UUID is reserved for pre-2.11 clients. */
+ ERR("Illegal nil UUID announced by peer in create session command");
+ ret = -1;
+ goto send_reply;
+ }
+ LTTNG_OPTIONAL_SET(&id_sessiond, id_sessiond_value);
+ LTTNG_OPTIONAL_SET(&creation_time, creation_time_value);
+ if (has_current_chunk) {
+ LTTNG_OPTIONAL_SET(¤t_chunk_id,
+ current_chunk_id_value);
+ }
}
+
if (ret < 0) {
goto send_reply;
}
- session = session_create(session_name, hostname, live_timer,
- snapshot, conn->major, conn->minor);
+ session = session_create(session_name, hostname, base_path, live_timer,
+ snapshot, sessiond_uuid,
+ id_sessiond.is_set ? &id_sessiond.value : NULL,
+ current_chunk_id.is_set ? ¤t_chunk_id.value : NULL,
+ creation_time.is_set ? &creation_time.value : NULL,
+ conn->major, conn->minor,
+ session_name_contains_creation_timestamp);
if (!session) {
ret = -1;
goto send_reply;
conn->session = session;
DBG("Created session %" PRIu64, session->id);
- reply.session_id = htobe64(session->id);
+ reply.generic.session_id = htobe64(session->id);
send_reply:
if (ret < 0) {
- reply.ret_code = htobe32(LTTNG_ERR_FATAL);
+ reply.generic.ret_code = htobe32(LTTNG_ERR_FATAL);
} else {
- reply.ret_code = htobe32(LTTNG_OK);
+ reply.generic.ret_code = htobe32(LTTNG_OK);
}
- send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
- if (send_ret < 0) {
- ERR("Relayd sending session id");
- ret = send_ret;
+ if (conn->minor < 11) {
+ /* From 2.1 to 2.10 */
+ ret = lttng_dynamic_buffer_append(&reply_payload,
+ &reply.generic, sizeof(reply.generic));
+ if (ret) {
+ ERR("Failed to append \"create session\" command reply header to payload buffer");
+ ret = -1;
+ goto end;
+ }
+ } else {
+ const uint32_t output_path_length =
+ session ? strlen(session->output_path) + 1 : 0;
+
+ reply.output_path_length = htobe32(output_path_length);
+ ret = lttng_dynamic_buffer_append(
+ &reply_payload, &reply, sizeof(reply));
+ if (ret) {
+ ERR("Failed to append \"create session\" command reply header to payload buffer");
+ goto end;
+ }
+
+ if (output_path_length) {
+ ret = lttng_dynamic_buffer_append(&reply_payload,
+ session->output_path,
+ output_path_length);
+ if (ret) {
+ ERR("Failed to append \"create session\" command reply path to payload buffer");
+ goto end;
+ }
+ }
}
+ send_ret = conn->sock->ops->sendmsg(conn->sock, reply_payload.data,
+ reply_payload.size, 0);
+ if (send_ret < (ssize_t) reply_payload.size) {
+ ERR("Failed to send \"create session\" command reply of %zu bytes (ret = %zd)",
+ reply_payload.size, send_ret);
+ ret = -1;
+ }
+end:
+ if (ret < 0 && session) {
+ session_put(session);
+ }
+ lttng_dynamic_buffer_reset(&reply_payload);
return ret;
}
pthread_mutex_unlock(&session->lock);
}
+static int conform_channel_path(char *channel_path)
+{
+ int ret = 0;
+
+ if (strstr("../", channel_path)) {
+ ERR("Refusing channel path as it walks up the path hierarchy: \"%s\"",
+ channel_path);
+ ret = -1;
+ goto end;
+ }
+
+ if (*channel_path == '/') {
+ const size_t len = strlen(channel_path);
+
+ /*
+ * Channel paths from peers prior to 2.11 are expressed as an
+ * absolute path that is, in reality, relative to the relay
+ * daemon's output directory. Remove the leading slash so it
+ * is correctly interpreted as a relative path later on.
+ *
+ * len (and not len - 1) is used to copy the trailing NULL.
+ */
+ bcopy(channel_path + 1, channel_path, len);
+ }
+end:
+ return ret;
+}
+
/*
* relay_add_stream: allocate a new stream for a session
*/
-static int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr,
- struct relay_connection *conn)
+static int relay_add_stream(const struct lttcomm_relayd_hdr *recv_hdr,
+ struct relay_connection *conn,
+ const struct lttng_buffer_view *payload)
{
int ret;
ssize_t send_ret;
uint64_t stream_handle = -1ULL;
char *path_name = NULL, *channel_name = NULL;
uint64_t tracefile_size = 0, tracefile_count = 0;
+ LTTNG_OPTIONAL(uint64_t) stream_chunk_id = {};
- if (!session || conn->version_check_done == 0) {
+ if (!session || !conn->version_check_done) {
ERR("Trying to add a stream before version check");
ret = -1;
goto end_no_session;
}
- switch (session->minor) {
- case 1: /* LTTng sessiond 2.1. Allocates path_name and channel_name. */
- ret = cmd_recv_stream_2_1(conn, &path_name,
+ if (session->minor == 1) {
+ /* For 2.1 */
+ ret = cmd_recv_stream_2_1(payload, &path_name,
&channel_name);
- break;
- case 2: /* LTTng sessiond 2.2. Allocates path_name and channel_name. */
- default:
- ret = cmd_recv_stream_2_2(conn, &path_name,
+ } else if (session->minor > 1 && session->minor < 11) {
+ /* From 2.2 to 2.10 */
+ ret = cmd_recv_stream_2_2(payload, &path_name,
&channel_name, &tracefile_size, &tracefile_count);
- break;
+ } else {
+ /* From 2.11 to ... */
+ ret = cmd_recv_stream_2_11(payload, &path_name,
+ &channel_name, &tracefile_size, &tracefile_count,
+ &stream_chunk_id.value);
+ stream_chunk_id.is_set = true;
}
+
if (ret < 0) {
goto send_reply;
}
+ if (conform_channel_path(path_name)) {
+ goto send_reply;
+ }
+
+ /*
+ * Backward compatibility for --group-output-by-session.
+ * Prior to lttng 2.11, the complete path is passed by the stream.
+ * Starting at 2.11, lttng-relayd uses chunk. When dealing with producer
+ * >=2.11 the chunk is responsible for the output path. When dealing
+ * with producer < 2.11 the chunk output_path is the root output path
+ * and the stream carries the complete path (path_name).
+ * To support --group-output-by-session with older producer (<2.11), we
+ * need to craft the path based on the stream path.
+ */
+ if (opt_group_output_by == RELAYD_GROUP_OUTPUT_BY_SESSION) {
+ if (conn->minor < 4) {
+ /*
+ * From 2.1 to 2.3, the session_name is not passed on
+ * the RELAYD_CREATE_SESSION command. The session name
+ * is necessary to detect the presence of a base_path
+ * inside the stream path. Without it we cannot perform
+ * a valid group-output-by-session transformation.
+ */
+ WARN("Unable to perform a --group-by-session transformation for session %" PRIu64
+ " for stream with path \"%s\" as it is produced by a peer using a protocol older than v2.4",
+ session->id, path_name);
+ } else if (conn->minor >= 4 && conn->minor < 11) {
+ char *group_by_session_path_name;
+
+ assert(session->session_name[0] != '\0');
+
+ group_by_session_path_name =
+ backward_compat_group_by_session(
+ path_name,
+ session->session_name);
+ if (!group_by_session_path_name) {
+ ERR("Failed to apply group by session to stream of session %" PRIu64,
+ session->id);
+ goto send_reply;
+ }
+
+ DBG("Transformed session path from \"%s\" to \"%s\" to honor per-session name grouping",
+ path_name, group_by_session_path_name);
+
+ free(path_name);
+ path_name = group_by_session_path_name;
+ }
+ }
+
trace = ctf_trace_get_by_path_or_create(session, path_name);
if (!trace) {
goto send_reply;
}
- /* This stream here has one reference on the trace. */
+ /* This stream here has one reference on the trace. */
pthread_mutex_lock(&last_relay_stream_id_lock);
stream_handle = ++last_relay_stream_id;
pthread_mutex_unlock(&last_relay_stream_id_lock);
/* We pass ownership of path_name and channel_name. */
stream = stream_create(trace, stream_handle, path_name,
- channel_name, tracefile_size, tracefile_count);
+ channel_name, tracefile_size, tracefile_count);
path_name = NULL;
channel_name = NULL;
send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
sizeof(struct lttcomm_relayd_status_stream), 0);
- if (send_ret < 0) {
- ERR("Relay sending stream id");
- ret = (int) send_ret;
+ if (send_ret < (ssize_t) sizeof(reply)) {
+ ERR("Failed to send \"add stream\" command reply (ret = %zd)",
+ send_ret);
+ ret = -1;
}
end_no_session:
/*
* relay_close_stream: close a specific stream
*/
-static int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr,
- struct relay_connection *conn)
+static int relay_close_stream(const struct lttcomm_relayd_hdr *recv_hdr,
+ struct relay_connection *conn,
+ const struct lttng_buffer_view *payload)
{
- int ret, send_ret;
+ int ret;
+ ssize_t send_ret;
struct relay_session *session = conn->session;
struct lttcomm_relayd_close_stream stream_info;
struct lttcomm_relayd_generic_reply reply;
DBG("Close stream received");
- if (!session || conn->version_check_done == 0) {
+ if (!session || !conn->version_check_done) {
ERR("Trying to close a stream before version check");
ret = -1;
goto end_no_session;
}
- ret = conn->sock->ops->recvmsg(conn->sock, &stream_info,
- sizeof(struct lttcomm_relayd_close_stream), 0);
- if (ret < sizeof(struct lttcomm_relayd_close_stream)) {
- if (ret == 0) {
- /* Orderly shutdown. Not necessary to print an error. */
- DBG("Socket %d did an orderly shutdown", conn->sock->fd);
- } else {
- ERR("Relay didn't receive valid add_stream struct size : %d", ret);
- }
+ if (payload->size < sizeof(stream_info)) {
+ ERR("Unexpected payload size in \"relay_close_stream\": expected >= %zu bytes, got %zu bytes",
+ sizeof(stream_info), payload->size);
ret = -1;
goto end_no_session;
}
+ memcpy(&stream_info, payload->data, sizeof(stream_info));
+ stream_info.stream_id = be64toh(stream_info.stream_id);
+ stream_info.last_net_seq_num = be64toh(stream_info.last_net_seq_num);
- stream = stream_get_by_id(be64toh(stream_info.stream_id));
+ stream = stream_get_by_id(stream_info.stream_id);
if (!stream) {
ret = -1;
goto end;
* pending check.
*/
pthread_mutex_lock(&stream->lock);
- stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num);
+ stream->last_net_seq_num = stream_info.last_net_seq_num;
pthread_mutex_unlock(&stream->lock);
/*
vstream = viewer_stream_get_by_id(stream->stream_handle);
if (vstream) {
- if (vstream->metadata_sent == stream->metadata_received) {
+ if (stream->no_new_metadata_notified) {
/*
* Since all the metadata has been sent to the
* viewer and that we have a request to close
}
}
stream_put(stream);
+ ret = 0;
end:
memset(&reply, 0, sizeof(reply));
}
send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
sizeof(struct lttcomm_relayd_generic_reply), 0);
- if (send_ret < 0) {
- ERR("Relay sending stream id");
- ret = send_ret;
+ if (send_ret < (ssize_t) sizeof(reply)) {
+ ERR("Failed to send \"close stream\" command reply (ret = %zd)",
+ send_ret);
+ ret = -1;
}
end_no_session:
* relay_reset_metadata: reset a metadata stream
*/
static
-int relay_reset_metadata(struct lttcomm_relayd_hdr *recv_hdr,
- struct relay_connection *conn)
+int relay_reset_metadata(const struct lttcomm_relayd_hdr *recv_hdr,
+ struct relay_connection *conn,
+ const struct lttng_buffer_view *payload)
{
- int ret, send_ret;
+ int ret;
+ ssize_t send_ret;
struct relay_session *session = conn->session;
struct lttcomm_relayd_reset_metadata stream_info;
struct lttcomm_relayd_generic_reply reply;
DBG("Reset metadata received");
- if (!session || conn->version_check_done == 0) {
+ if (!session || !conn->version_check_done) {
ERR("Trying to reset a metadata stream before version check");
ret = -1;
goto end_no_session;
}
- ret = conn->sock->ops->recvmsg(conn->sock, &stream_info,
- sizeof(struct lttcomm_relayd_reset_metadata), 0);
- if (ret < sizeof(struct lttcomm_relayd_reset_metadata)) {
- if (ret == 0) {
- /* Orderly shutdown. Not necessary to print an error. */
- DBG("Socket %d did an orderly shutdown", conn->sock->fd);
- } else {
- ERR("Relay didn't receive valid reset_metadata struct "
- "size : %d", ret);
- }
+ if (payload->size < sizeof(stream_info)) {
+ ERR("Unexpected payload size in \"relay_reset_metadata\": expected >= %zu bytes, got %zu bytes",
+ sizeof(stream_info), payload->size);
ret = -1;
goto end_no_session;
}
- DBG("Update metadata to version %" PRIu64, be64toh(stream_info.version));
+ memcpy(&stream_info, payload->data, sizeof(stream_info));
+ stream_info.stream_id = be64toh(stream_info.stream_id);
+ stream_info.version = be64toh(stream_info.version);
+
+ DBG("Update metadata to version %" PRIu64, stream_info.version);
/* Unsupported for live sessions for now. */
if (session->live_timer != 0) {
goto end;
}
- stream = stream_get_by_id(be64toh(stream_info.stream_id));
+ stream = stream_get_by_id(stream_info.stream_id);
if (!stream) {
ret = -1;
goto end;
goto end_unlock;
}
- ret = utils_rotate_stream_file(stream->path_name, stream->channel_name,
- 0, 0, -1, -1, stream->stream_fd->fd, NULL,
- &stream->stream_fd->fd);
+ ret = stream_reset_file(stream);
if (ret < 0) {
- ERR("Failed to rotate metadata file %s of channel %s",
- stream->path_name, stream->channel_name);
+ ERR("Failed to reset metadata stream %" PRIu64
+ ": stream_path = %s, channel = %s",
+ stream->stream_handle, stream->path_name,
+ stream->channel_name);
goto end_unlock;
}
-
end_unlock:
pthread_mutex_unlock(&stream->lock);
stream_put(stream);
}
send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
sizeof(struct lttcomm_relayd_generic_reply), 0);
- if (send_ret < 0) {
- ERR("Relay sending reset metadata reply");
- ret = send_ret;
+ if (send_ret < (ssize_t) sizeof(reply)) {
+ ERR("Failed to send \"reset metadata\" command reply (ret = %zd)",
+ send_ret);
+ ret = -1;
}
end_no_session:
static void relay_unknown_command(struct relay_connection *conn)
{
struct lttcomm_relayd_generic_reply reply;
- int ret;
+ ssize_t send_ret;
memset(&reply, 0, sizeof(reply));
reply.ret_code = htobe32(LTTNG_ERR_UNK);
- ret = conn->sock->ops->sendmsg(conn->sock, &reply,
- sizeof(struct lttcomm_relayd_generic_reply), 0);
- if (ret < 0) {
- ERR("Relay sending unknown command");
+ send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
+ if (send_ret < sizeof(reply)) {
+ ERR("Failed to send \"unknown command\" command reply (ret = %zd)", send_ret);
}
}
* relay_start: send an acknowledgment to the client to tell if we are
* ready to receive data. We are ready if a session is established.
*/
-static int relay_start(struct lttcomm_relayd_hdr *recv_hdr,
- struct relay_connection *conn)
+static int relay_start(const struct lttcomm_relayd_hdr *recv_hdr,
+ struct relay_connection *conn,
+ const struct lttng_buffer_view *payload)
{
- int ret = htobe32(LTTNG_OK);
+ int ret = 0;
+ ssize_t send_ret;
struct lttcomm_relayd_generic_reply reply;
struct relay_session *session = conn->session;
}
memset(&reply, 0, sizeof(reply));
- reply.ret_code = ret;
- ret = conn->sock->ops->sendmsg(conn->sock, &reply,
- sizeof(struct lttcomm_relayd_generic_reply), 0);
- if (ret < 0) {
- ERR("Relay sending start ack");
- }
-
- return ret;
-}
-
-/*
- * Append padding to the file pointed by the file descriptor fd.
- */
-static int write_padding_to_file(int fd, uint32_t size)
-{
- ssize_t ret = 0;
- char *zeros;
-
- if (size == 0) {
- goto end;
- }
-
- zeros = zmalloc(size);
- if (zeros == NULL) {
- PERROR("zmalloc zeros for padding");
+ reply.ret_code = htobe32(LTTNG_OK);
+ send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
+ sizeof(reply), 0);
+ if (send_ret < (ssize_t) sizeof(reply)) {
+ ERR("Failed to send \"relay_start\" command reply (ret = %zd)",
+ send_ret);
ret = -1;
- goto end;
- }
-
- ret = lttng_write(fd, zeros, size);
- if (ret < size) {
- PERROR("write padding to file");
}
- free(zeros);
-
-end:
return ret;
}
/*
* relay_recv_metadata: receive the metadata for the session.
*/
-static int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr,
- struct relay_connection *conn)
+static int relay_recv_metadata(const struct lttcomm_relayd_hdr *recv_hdr,
+ struct relay_connection *conn,
+ const struct lttng_buffer_view *payload)
{
int ret = 0;
- ssize_t size_ret;
struct relay_session *session = conn->session;
- struct lttcomm_relayd_metadata_payload *metadata_struct;
+ struct lttcomm_relayd_metadata_payload metadata_payload_header;
struct relay_stream *metadata_stream;
- uint64_t data_size, payload_size;
+ uint64_t metadata_payload_size;
+ struct lttng_buffer_view packet_view;
if (!session) {
ERR("Metadata sent before version check");
goto end;
}
- data_size = payload_size = be64toh(recv_hdr->data_size);
- if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
+ if (recv_hdr->data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
ERR("Incorrect data size");
ret = -1;
goto end;
}
- payload_size -= sizeof(struct lttcomm_relayd_metadata_payload);
+ metadata_payload_size = recv_hdr->data_size -
+ sizeof(struct lttcomm_relayd_metadata_payload);
- if (data_buffer_size < data_size) {
- /* In case the realloc fails, we can free the memory */
- char *tmp_data_ptr;
-
- tmp_data_ptr = realloc(data_buffer, data_size);
- if (!tmp_data_ptr) {
- ERR("Allocating data buffer");
- 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);
- size_ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, data_size, 0);
- if (size_ret < 0 || size_ret != data_size) {
- if (size_ret == 0) {
- /* Orderly shutdown. Not necessary to print an error. */
- DBG("Socket %d did an orderly shutdown", conn->sock->fd);
- } else {
- ERR("Relay didn't receive the whole metadata");
- }
- ret = -1;
- goto end;
- }
- metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer;
+ memcpy(&metadata_payload_header, payload->data,
+ sizeof(metadata_payload_header));
+ metadata_payload_header.stream_id = be64toh(
+ metadata_payload_header.stream_id);
+ metadata_payload_header.padding_size = be32toh(
+ metadata_payload_header.padding_size);
- metadata_stream = stream_get_by_id(be64toh(metadata_struct->stream_id));
+ metadata_stream = stream_get_by_id(metadata_payload_header.stream_id);
if (!metadata_stream) {
ret = -1;
goto end;
}
- pthread_mutex_lock(&metadata_stream->lock);
-
- size_ret = lttng_write(metadata_stream->stream_fd->fd, metadata_struct->payload,
- payload_size);
- if (size_ret < payload_size) {
- ERR("Relay error writing metadata on file");
+ packet_view = lttng_buffer_view_from_view(payload,
+ sizeof(metadata_payload_header), metadata_payload_size);
+ if (!packet_view.data) {
+ ERR("Invalid metadata packet length announced by header");
ret = -1;
goto end_put;
}
- size_ret = write_padding_to_file(metadata_stream->stream_fd->fd,
- be32toh(metadata_struct->padding_size));
- if (size_ret < 0) {
+ pthread_mutex_lock(&metadata_stream->lock);
+ ret = stream_write(metadata_stream, &packet_view,
+ metadata_payload_header.padding_size);
+ pthread_mutex_unlock(&metadata_stream->lock);
+ if (ret){
+ ret = -1;
goto end_put;
}
-
- metadata_stream->metadata_received +=
- payload_size + be32toh(metadata_struct->padding_size);
- DBG2("Relay metadata written. Updated metadata_received %" PRIu64,
- metadata_stream->metadata_received);
-
end_put:
- pthread_mutex_unlock(&metadata_stream->lock);
stream_put(metadata_stream);
end:
return ret;
/*
* relay_send_version: send relayd version number
*/
-static int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr,
- struct relay_connection *conn)
+static int relay_send_version(const struct lttcomm_relayd_hdr *recv_hdr,
+ struct relay_connection *conn,
+ const struct lttng_buffer_view *payload)
{
int ret;
+ ssize_t send_ret;
struct lttcomm_relayd_version reply, msg;
+ bool compatible = true;
- conn->version_check_done = 1;
+ conn->version_check_done = true;
/* Get version from the other side. */
- ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
- if (ret < 0 || ret != sizeof(msg)) {
- if (ret == 0) {
- /* Orderly shutdown. Not necessary to print an error. */
- DBG("Socket %d did an orderly shutdown", conn->sock->fd);
- } else {
- ERR("Relay failed to receive the version values.");
- }
+ if (payload->size < sizeof(msg)) {
+ ERR("Unexpected payload size in \"relay_send_version\": expected >= %zu bytes, got %zu bytes",
+ sizeof(msg), payload->size);
ret = -1;
goto end;
}
+ memcpy(&msg, payload->data, sizeof(msg));
+ msg.major = be32toh(msg.major);
+ msg.minor = be32toh(msg.minor);
+
memset(&reply, 0, sizeof(reply));
reply.major = RELAYD_VERSION_COMM_MAJOR;
reply.minor = RELAYD_VERSION_COMM_MINOR;
/* Major versions must be the same */
- if (reply.major != be32toh(msg.major)) {
+ if (reply.major != msg.major) {
DBG("Incompatible major versions (%u vs %u), deleting session",
- reply.major, be32toh(msg.major));
- connection_put(conn);
- ret = 0;
- goto end;
+ reply.major, msg.major);
+ compatible = false;
}
conn->major = reply.major;
/* We adapt to the lowest compatible version */
- if (reply.minor <= be32toh(msg.minor)) {
+ if (reply.minor <= msg.minor) {
conn->minor = reply.minor;
} else {
- conn->minor = be32toh(msg.minor);
+ conn->minor = msg.minor;
}
reply.major = htobe32(reply.major);
reply.minor = htobe32(reply.minor);
- ret = conn->sock->ops->sendmsg(conn->sock, &reply,
- sizeof(struct lttcomm_relayd_version), 0);
- if (ret < 0) {
- ERR("Relay sending version");
+ send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
+ sizeof(reply), 0);
+ if (send_ret < (ssize_t) sizeof(reply)) {
+ ERR("Failed to send \"send version\" command reply (ret = %zd)",
+ send_ret);
+ ret = -1;
+ goto end;
+ } else {
+ ret = 0;
+ }
+
+ if (!compatible) {
+ ret = -1;
+ goto end;
}
DBG("Version check done using protocol %u.%u", conn->major,
/*
* Check for data pending for a given stream id from the session daemon.
*/
-static int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
- struct relay_connection *conn)
+static int relay_data_pending(const struct lttcomm_relayd_hdr *recv_hdr,
+ struct relay_connection *conn,
+ const struct lttng_buffer_view *payload)
{
struct relay_session *session = conn->session;
struct lttcomm_relayd_data_pending msg;
struct lttcomm_relayd_generic_reply reply;
struct relay_stream *stream;
+ ssize_t send_ret;
int ret;
- uint64_t last_net_seq_num, stream_id;
+ uint64_t stream_seq;
DBG("Data pending command received");
- if (!session || conn->version_check_done == 0) {
+ if (!session || !conn->version_check_done) {
ERR("Trying to check for data before version check");
ret = -1;
goto end_no_session;
}
- ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
- if (ret < sizeof(msg)) {
- if (ret == 0) {
- /* Orderly shutdown. Not necessary to print an error. */
- DBG("Socket %d did an orderly shutdown", conn->sock->fd);
- } else {
- ERR("Relay didn't receive valid data_pending struct size : %d",
- ret);
- }
+ if (payload->size < sizeof(msg)) {
+ ERR("Unexpected payload size in \"relay_data_pending\": expected >= %zu bytes, got %zu bytes",
+ sizeof(msg), payload->size);
ret = -1;
goto end_no_session;
}
+ memcpy(&msg, payload->data, sizeof(msg));
+ msg.stream_id = be64toh(msg.stream_id);
+ msg.last_net_seq_num = be64toh(msg.last_net_seq_num);
- stream_id = be64toh(msg.stream_id);
- last_net_seq_num = be64toh(msg.last_net_seq_num);
-
- stream = stream_get_by_id(stream_id);
+ stream = stream_get_by_id(msg.stream_id);
if (stream == NULL) {
ret = -1;
goto end;
pthread_mutex_lock(&stream->lock);
- DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64
- " and last_seq %" PRIu64, stream_id, stream->prev_seq,
- last_net_seq_num);
+ if (session_streams_have_index(session)) {
+ /*
+ * Ensure that both the index and stream data have been
+ * flushed up to the requested point.
+ */
+ stream_seq = min(stream->prev_data_seq, stream->prev_index_seq);
+ } else {
+ stream_seq = stream->prev_data_seq;
+ }
+ DBG("Data pending for stream id %" PRIu64 ": prev_data_seq %" PRIu64
+ ", prev_index_seq %" PRIu64
+ ", and last_seq %" PRIu64, msg.stream_id,
+ stream->prev_data_seq, stream->prev_index_seq,
+ msg.last_net_seq_num);
/* Avoid wrapping issue */
- if (((int64_t) (stream->prev_seq - last_net_seq_num)) >= 0) {
+ if (((int64_t) (stream_seq - msg.last_net_seq_num)) >= 0) {
/* Data has in fact been written and is NOT pending */
ret = 0;
} else {
memset(&reply, 0, sizeof(reply));
reply.ret_code = htobe32(ret);
- ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
- if (ret < 0) {
- ERR("Relay data pending ret code failed");
+ send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
+ if (send_ret < (ssize_t) sizeof(reply)) {
+ ERR("Failed to send \"data pending\" command reply (ret = %zd)",
+ send_ret);
+ ret = -1;
}
end_no_session:
* the control socket has been handled. So, this is why we simply return
* OK here.
*/
-static int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr,
- struct relay_connection *conn)
+static int relay_quiescent_control(const struct lttcomm_relayd_hdr *recv_hdr,
+ struct relay_connection *conn,
+ const struct lttng_buffer_view *payload)
{
int ret;
- uint64_t stream_id;
+ ssize_t send_ret;
struct relay_stream *stream;
struct lttcomm_relayd_quiescent_control msg;
struct lttcomm_relayd_generic_reply reply;
DBG("Checking quiescent state on control socket");
- if (!conn->session || conn->version_check_done == 0) {
+ if (!conn->session || !conn->version_check_done) {
ERR("Trying to check for data before version check");
ret = -1;
goto end_no_session;
}
- ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
- if (ret < sizeof(msg)) {
- if (ret == 0) {
- /* Orderly shutdown. Not necessary to print an error. */
- DBG("Socket %d did an orderly shutdown", conn->sock->fd);
- } else {
- ERR("Relay didn't receive valid begin data_pending struct size: %d",
- ret);
- }
+ if (payload->size < sizeof(msg)) {
+ ERR("Unexpected payload size in \"relay_quiescent_control\": expected >= %zu bytes, got %zu bytes",
+ sizeof(msg), payload->size);
ret = -1;
goto end_no_session;
}
+ memcpy(&msg, payload->data, sizeof(msg));
+ msg.stream_id = be64toh(msg.stream_id);
- stream_id = be64toh(msg.stream_id);
- stream = stream_get_by_id(stream_id);
+ stream = stream_get_by_id(msg.stream_id);
if (!stream) {
goto reply;
}
pthread_mutex_lock(&stream->lock);
stream->data_pending_check_done = true;
pthread_mutex_unlock(&stream->lock);
- DBG("Relay quiescent control pending flag set to %" PRIu64, stream_id);
+
+ DBG("Relay quiescent control pending flag set to %" PRIu64, msg.stream_id);
stream_put(stream);
reply:
memset(&reply, 0, sizeof(reply));
reply.ret_code = htobe32(LTTNG_OK);
- ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
- if (ret < 0) {
- ERR("Relay data quiescent control ret code failed");
+ send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
+ if (send_ret < (ssize_t) sizeof(reply)) {
+ ERR("Failed to send \"quiescent control\" command reply (ret = %zd)",
+ send_ret);
+ ret = -1;
+ } else {
+ ret = 0;
}
end_no_session:
*
* This command returns to the client a LTTNG_OK code.
*/
-static int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
- struct relay_connection *conn)
+static int relay_begin_data_pending(const struct lttcomm_relayd_hdr *recv_hdr,
+ struct relay_connection *conn,
+ const struct lttng_buffer_view *payload)
{
int ret;
+ ssize_t send_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(conn);
DBG("Init streams for data pending");
- if (!conn->session || conn->version_check_done == 0) {
+ if (!conn->session || !conn->version_check_done) {
ERR("Trying to check for data before version check");
ret = -1;
goto end_no_session;
}
- ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
- if (ret < sizeof(msg)) {
- if (ret == 0) {
- /* Orderly shutdown. Not necessary to print an error. */
- DBG("Socket %d did an orderly shutdown", conn->sock->fd);
- } else {
- ERR("Relay didn't receive valid begin data_pending struct size: %d",
- ret);
- }
+ if (payload->size < sizeof(msg)) {
+ ERR("Unexpected payload size in \"relay_begin_data_pending\": expected >= %zu bytes, got %zu bytes",
+ sizeof(msg), payload->size);
ret = -1;
goto end_no_session;
}
-
- session_id = be64toh(msg.session_id);
+ memcpy(&msg, payload->data, sizeof(msg));
+ msg.session_id = be64toh(msg.session_id);
/*
* Iterate over all streams to set the begin data pending flag.
if (!stream_get(stream)) {
continue;
}
- if (stream->trace->session->id == session_id) {
+ if (stream->trace->session->id == msg.session_id) {
pthread_mutex_lock(&stream->lock);
stream->data_pending_check_done = false;
pthread_mutex_unlock(&stream->lock);
/* All good, send back reply. */
reply.ret_code = htobe32(LTTNG_OK);
- ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
- if (ret < 0) {
- ERR("Relay begin data pending send reply failed");
+ send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
+ if (send_ret < (ssize_t) sizeof(reply)) {
+ ERR("Failed to send \"begin data pending\" command reply (ret = %zd)",
+ send_ret);
+ ret = -1;
+ } else {
+ ret = 0;
}
end_no_session:
*
* 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_connection *conn)
+static int relay_end_data_pending(const struct lttcomm_relayd_hdr *recv_hdr,
+ struct relay_connection *conn,
+ const struct lttng_buffer_view *payload)
{
int ret;
+ ssize_t send_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;
DBG("End data pending command");
- if (!conn->session || conn->version_check_done == 0) {
+ if (!conn->session || !conn->version_check_done) {
ERR("Trying to check for data before version check");
ret = -1;
goto end_no_session;
}
- ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
- if (ret < sizeof(msg)) {
- if (ret == 0) {
- /* Orderly shutdown. Not necessary to print an error. */
- DBG("Socket %d did an orderly shutdown", conn->sock->fd);
- } else {
- ERR("Relay didn't receive valid end data_pending struct size: %d",
- ret);
- }
+ if (payload->size < sizeof(msg)) {
+ ERR("Unexpected payload size in \"relay_end_data_pending\": expected >= %zu bytes, got %zu bytes",
+ sizeof(msg), payload->size);
ret = -1;
goto end_no_session;
}
-
- session_id = be64toh(msg.session_id);
+ memcpy(&msg, payload->data, sizeof(msg));
+ msg.session_id = be64toh(msg.session_id);
/*
* Iterate over all streams to see if the begin data pending
if (!stream_get(stream)) {
continue;
}
- if (stream->trace->session->id != session_id) {
+ if (stream->trace->session->id != msg.session_id) {
stream_put(stream);
continue;
}
pthread_mutex_lock(&stream->lock);
if (!stream->data_pending_check_done) {
- if (!stream->closed || !(((int64_t) (stream->prev_seq - stream->last_net_seq_num)) >= 0)) {
+ uint64_t stream_seq;
+
+ if (session_streams_have_index(conn->session)) {
+ /*
+ * Ensure that both the index and stream data have been
+ * flushed up to the requested point.
+ */
+ stream_seq = min(stream->prev_data_seq, stream->prev_index_seq);
+ } else {
+ stream_seq = stream->prev_data_seq;
+ }
+ if (!stream->closed || !(((int64_t) (stream_seq - stream->last_net_seq_num)) >= 0)) {
is_data_inflight = 1;
DBG("Data is still in flight for stream %" PRIu64,
stream->stream_handle);
/* All good, send back reply. */
reply.ret_code = htobe32(is_data_inflight);
- ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
- if (ret < 0) {
- ERR("Relay end data pending send reply failed");
+ send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
+ if (send_ret < (ssize_t) sizeof(reply)) {
+ ERR("Failed to send \"end data pending\" command reply (ret = %zd)",
+ send_ret);
+ ret = -1;
+ } else {
+ ret = 0;
}
end_no_session:
*
* Return 0 on success else a negative value.
*/
-static int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr,
- struct relay_connection *conn)
+static int relay_recv_index(const struct lttcomm_relayd_hdr *recv_hdr,
+ struct relay_connection *conn,
+ const struct lttng_buffer_view *payload)
{
- int ret, send_ret;
+ int ret;
+ ssize_t send_ret;
struct relay_session *session = conn->session;
struct lttcomm_relayd_index index_info;
- struct relay_index *index;
struct lttcomm_relayd_generic_reply reply;
struct relay_stream *stream;
- uint64_t net_seq_num;
+ size_t msg_len;
assert(conn);
DBG("Relay receiving index");
- if (!session || conn->version_check_done == 0) {
+ if (!session || !conn->version_check_done) {
ERR("Trying to close a stream before version check");
ret = -1;
goto end_no_session;
}
- ret = conn->sock->ops->recvmsg(conn->sock, &index_info,
- sizeof(index_info), 0);
- if (ret < sizeof(index_info)) {
- if (ret == 0) {
- /* Orderly shutdown. Not necessary to print an error. */
- DBG("Socket %d did an orderly shutdown", conn->sock->fd);
- } else {
- ERR("Relay didn't receive valid index struct size : %d", ret);
- }
+ msg_len = lttcomm_relayd_index_len(
+ lttng_to_index_major(conn->major, conn->minor),
+ lttng_to_index_minor(conn->major, conn->minor));
+ if (payload->size < msg_len) {
+ ERR("Unexpected payload size in \"relay_recv_index\": expected >= %zu bytes, got %zu bytes",
+ msg_len, payload->size);
ret = -1;
goto end_no_session;
}
+ memcpy(&index_info, payload->data, msg_len);
+ index_info.relay_stream_id = be64toh(index_info.relay_stream_id);
+ index_info.net_seq_num = be64toh(index_info.net_seq_num);
+ index_info.packet_size = be64toh(index_info.packet_size);
+ index_info.content_size = be64toh(index_info.content_size);
+ index_info.timestamp_begin = be64toh(index_info.timestamp_begin);
+ index_info.timestamp_end = be64toh(index_info.timestamp_end);
+ index_info.events_discarded = be64toh(index_info.events_discarded);
+ index_info.stream_id = be64toh(index_info.stream_id);
- net_seq_num = be64toh(index_info.net_seq_num);
+ if (conn->minor >= 8) {
+ index_info.stream_instance_id =
+ be64toh(index_info.stream_instance_id);
+ index_info.packet_seq_num = be64toh(index_info.packet_seq_num);
+ } else {
+ index_info.stream_instance_id = -1ULL;
+ index_info.packet_seq_num = -1ULL;
+ }
- stream = stream_get_by_id(be64toh(index_info.relay_stream_id));
+ stream = stream_get_by_id(index_info.relay_stream_id);
if (!stream) {
ERR("stream_get_by_id not found");
ret = -1;
goto end;
}
- pthread_mutex_lock(&stream->lock);
-
- /* Live beacon handling */
- if (index_info.packet_size == 0) {
- DBG("Received live beacon for stream %" PRIu64,
- stream->stream_handle);
-
- /*
- * Only flag a stream inactive when it has already
- * received data and no indexes are in flight.
- */
- if (stream->index_received_seqcount > 0
- && stream->indexes_in_flight == 0) {
- stream->beacon_ts_end =
- be64toh(index_info.timestamp_end);
- }
- ret = 0;
- goto end_stream_put;
- } else {
- stream->beacon_ts_end = -1ULL;
- }
- if (stream->ctf_stream_id == -1ULL) {
- stream->ctf_stream_id = be64toh(index_info.stream_id);
- }
- index = relay_index_get_by_id_or_create(stream, net_seq_num);
- if (!index) {
- ret = -1;
- ERR("relay_index_get_by_id_or_create index NULL");
- goto end_stream_put;
- }
- if (set_index_control_data(index, &index_info, conn)) {
- ERR("set_index_control_data error");
- relay_index_put(index);
- ret = -1;
+ pthread_mutex_lock(&stream->lock);
+ ret = stream_add_index(stream, &index_info);
+ pthread_mutex_unlock(&stream->lock);
+ if (ret) {
goto end_stream_put;
}
- ret = relay_index_try_flush(index);
- if (ret == 0) {
- tracefile_array_commit_seq(stream->tfa);
- stream->index_received_seqcount++;
- } else if (ret > 0) {
- /* no flush. */
- ret = 0;
- } else {
- ERR("relay_index_try_flush error %d", ret);
- relay_index_put(index);
- ret = -1;
- }
end_stream_put:
- pthread_mutex_unlock(&stream->lock);
stream_put(stream);
-
end:
-
memset(&reply, 0, sizeof(reply));
if (ret < 0) {
reply.ret_code = htobe32(LTTNG_ERR_UNK);
reply.ret_code = htobe32(LTTNG_OK);
}
send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
- if (send_ret < 0) {
- ERR("Relay sending close index id reply");
- ret = send_ret;
+ if (send_ret < (ssize_t) sizeof(reply)) {
+ ERR("Failed to send \"recv index\" command reply (ret = %zd)", send_ret);
+ ret = -1;
}
end_no_session:
*
* Return 0 on success else a negative value.
*/
-static int relay_streams_sent(struct lttcomm_relayd_hdr *recv_hdr,
- struct relay_connection *conn)
+static int relay_streams_sent(const struct lttcomm_relayd_hdr *recv_hdr,
+ struct relay_connection *conn,
+ const struct lttng_buffer_view *payload)
{
- int ret, send_ret;
+ int ret;
+ ssize_t send_ret;
struct lttcomm_relayd_generic_reply reply;
assert(conn);
DBG("Relay receiving streams_sent");
- if (!conn->session || conn->version_check_done == 0) {
+ if (!conn->session || !conn->version_check_done) {
ERR("Trying to close a stream before version check");
ret = -1;
goto end_no_session;
memset(&reply, 0, sizeof(reply));
reply.ret_code = htobe32(LTTNG_OK);
send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
- if (send_ret < 0) {
- ERR("Relay sending sent_stream reply");
- ret = send_ret;
+ if (send_ret < (ssize_t) sizeof(reply)) {
+ ERR("Failed to send \"streams sent\" command reply (ret = %zd)",
+ send_ret);
+ ret = -1;
} else {
/* Success. */
ret = 0;
}
/*
- * Process the commands received on the control socket
+ * relay_rotate_session_stream: rotate a stream to a new tracefile for the
+ * session rotation feature (not the tracefile rotation feature).
*/
-static int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr,
- struct relay_connection *conn)
+static int relay_rotate_session_streams(
+ const struct lttcomm_relayd_hdr *recv_hdr,
+ struct relay_connection *conn,
+ const struct lttng_buffer_view *payload)
{
int ret = 0;
-
- switch (be32toh(recv_hdr->cmd)) {
- case RELAYD_CREATE_SESSION:
- ret = relay_create_session(recv_hdr, conn);
- break;
- case RELAYD_ADD_STREAM:
- ret = relay_add_stream(recv_hdr, conn);
- break;
- case RELAYD_START_DATA:
- ret = relay_start(recv_hdr, conn);
- break;
- case RELAYD_SEND_METADATA:
- ret = relay_recv_metadata(recv_hdr, conn);
- break;
- case RELAYD_VERSION:
- ret = relay_send_version(recv_hdr, conn);
- break;
- case RELAYD_CLOSE_STREAM:
- ret = relay_close_stream(recv_hdr, conn);
- break;
- case RELAYD_DATA_PENDING:
- ret = relay_data_pending(recv_hdr, conn);
- break;
- case RELAYD_QUIESCENT_CONTROL:
- ret = relay_quiescent_control(recv_hdr, conn);
- break;
- case RELAYD_BEGIN_DATA_PENDING:
- ret = relay_begin_data_pending(recv_hdr, conn);
- break;
- case RELAYD_END_DATA_PENDING:
- ret = relay_end_data_pending(recv_hdr, conn);
- break;
- case RELAYD_SEND_INDEX:
- ret = relay_recv_index(recv_hdr, conn);
- break;
- case RELAYD_STREAMS_SENT:
- ret = relay_streams_sent(recv_hdr, conn);
- break;
- case RELAYD_RESET_METADATA:
- ret = relay_reset_metadata(recv_hdr, conn);
- break;
- case RELAYD_UPDATE_SYNC_INFO:
- default:
- ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd));
- relay_unknown_command(conn);
+ uint32_t i;
+ ssize_t send_ret;
+ enum lttng_error_code reply_code = LTTNG_ERR_UNK;
+ struct relay_session *session = conn->session;
+ struct lttcomm_relayd_rotate_streams rotate_streams;
+ struct lttcomm_relayd_generic_reply reply = {};
+ struct relay_stream *stream = NULL;
+ const size_t header_len = sizeof(struct lttcomm_relayd_rotate_streams);
+ struct lttng_trace_chunk *next_trace_chunk = NULL;
+ struct lttng_buffer_view stream_positions;
+ char chunk_id_buf[MAX_INT_DEC_LEN(uint64_t)];
+ const char *chunk_id_str = "none";
+
+ if (!session || !conn->version_check_done) {
+ ERR("Trying to rotate a stream before version check");
ret = -1;
- goto end;
+ goto end_no_reply;
}
-end:
- return ret;
-}
-
-/*
- * Handle index for a data stream.
- *
- * Called with the stream lock held.
- *
- * Return 0 on success else a negative value.
- */
-static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num,
- int rotate_index)
-{
- int ret = 0;
- uint64_t data_offset;
- struct relay_index *index;
-
- /* Get data offset because we are about to update the index. */
- data_offset = htobe64(stream->tracefile_size_current);
-
- DBG("handle_index_data: stream %" PRIu64 " net_seq_num %" PRIu64 " data offset %" PRIu64,
- stream->stream_handle, net_seq_num, stream->tracefile_size_current);
+ if (session->major == 2 && session->minor < 11) {
+ ERR("Unsupported feature before 2.11");
+ ret = -1;
+ goto end_no_reply;
+ }
- /*
- * Lookup for an existing index for that stream id/sequence
- * number. If it exists, the control thread has already received the
- * data for it, thus we need to write it to disk.
- */
- index = relay_index_get_by_id_or_create(stream, net_seq_num);
- if (!index) {
+ if (payload->size < header_len) {
+ ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected >= %zu bytes, got %zu bytes",
+ header_len, payload->size);
ret = -1;
- goto end;
+ goto end_no_reply;
}
- if (rotate_index || !stream->index_fd) {
- int fd;
+ memcpy(&rotate_streams, payload->data, header_len);
- /* Put ref on previous index_fd. */
- if (stream->index_fd) {
- stream_fd_put(stream->index_fd);
- stream->index_fd = NULL;
+ /* Convert header to host endianness. */
+ rotate_streams = (typeof(rotate_streams)) {
+ .stream_count = be32toh(rotate_streams.stream_count),
+ .new_chunk_id = (typeof(rotate_streams.new_chunk_id)) {
+ .is_set = !!rotate_streams.new_chunk_id.is_set,
+ .value = be64toh(rotate_streams.new_chunk_id.value),
}
+ };
- fd = index_create_file(stream->path_name, stream->channel_name,
- -1, -1, stream->tracefile_size,
- tracefile_array_get_file_index_head(stream->tfa));
- if (fd < 0) {
+ if (rotate_streams.new_chunk_id.is_set) {
+ /*
+ * Retrieve the trace chunk the stream must transition to. As
+ * per the protocol, this chunk should have been created
+ * before this command is received.
+ */
+ next_trace_chunk = sessiond_trace_chunk_registry_get_chunk(
+ sessiond_trace_chunk_registry,
+ session->sessiond_uuid, session->id,
+ rotate_streams.new_chunk_id.value);
+ if (!next_trace_chunk) {
+ char uuid_str[LTTNG_UUID_STR_LEN];
+
+ lttng_uuid_to_str(session->sessiond_uuid, uuid_str);
+ ERR("Unknown next trace chunk in ROTATE_STREAMS command: sessiond_uuid = {%s}, session_id = %" PRIu64
+ ", trace_chunk_id = %" PRIu64,
+ uuid_str, session->id,
+ rotate_streams.new_chunk_id.value);
+ reply_code = LTTNG_ERR_INVALID_PROTOCOL;
ret = -1;
- /* Put self-ref for this index due to error. */
- relay_index_put(index);
goto end;
}
- stream->index_fd = stream_fd_create(fd);
- if (!stream->index_fd) {
- ret = -1;
- if (close(fd)) {
- PERROR("Error closing FD %d", fd);
- }
- /* Put self-ref for this index due to error. */
- relay_index_put(index);
- /* Will put the local ref. */
- goto end;
+
+ ret = snprintf(chunk_id_buf, sizeof(chunk_id_buf), "%" PRIu64,
+ rotate_streams.new_chunk_id.value);
+ if (ret < 0 || ret >= sizeof(chunk_id_buf)) {
+ chunk_id_str = "formatting error";
+ } else {
+ chunk_id_str = chunk_id_buf;
}
}
- if (relay_index_set_fd(index, stream->index_fd, data_offset)) {
+ DBG("Rotate %" PRIu32 " streams of session \"%s\" to chunk \"%s\"",
+ rotate_streams.stream_count, session->session_name,
+ chunk_id_str);
+
+ stream_positions = lttng_buffer_view_from_view(payload,
+ sizeof(rotate_streams), -1);
+ if (!stream_positions.data ||
+ stream_positions.size <
+ (rotate_streams.stream_count *
+ sizeof(struct lttcomm_relayd_stream_rotation_position))) {
+ reply_code = LTTNG_ERR_INVALID_PROTOCOL;
ret = -1;
- /* Put self-ref for this index due to error. */
- relay_index_put(index);
goto end;
}
- ret = relay_index_try_flush(index);
- if (ret == 0) {
- tracefile_array_commit_seq(stream->tfa);
- stream->index_received_seqcount++;
- } else if (ret > 0) {
- /* No flush. */
- ret = 0;
- } else {
- /* Put self-ref for this index due to error. */
- relay_index_put(index);
- ret = -1;
- }
-end:
- return ret;
-}
+ for (i = 0; i < rotate_streams.stream_count; i++) {
+ struct lttcomm_relayd_stream_rotation_position *position_comm =
+ &((typeof(position_comm)) stream_positions.data)[i];
+ const struct lttcomm_relayd_stream_rotation_position pos = {
+ .stream_id = be64toh(position_comm->stream_id),
+ .rotate_at_seq_num = be64toh(
+ position_comm->rotate_at_seq_num),
+ };
-/*
- * relay_process_data: Process the data received on the data socket
+ stream = stream_get_by_id(pos.stream_id);
+ if (!stream) {
+ reply_code = LTTNG_ERR_INVALID;
+ ret = -1;
+ goto end;
+ }
+
+ pthread_mutex_lock(&stream->lock);
+ ret = stream_set_pending_rotation(stream, next_trace_chunk,
+ pos.rotate_at_seq_num);
+ pthread_mutex_unlock(&stream->lock);
+ if (ret) {
+ reply_code = LTTNG_ERR_FILE_CREATION_ERROR;
+ goto end;
+ }
+
+ stream_put(stream);
+ stream = NULL;
+ }
+
+ reply_code = LTTNG_OK;
+ ret = 0;
+end:
+ if (stream) {
+ stream_put(stream);
+ }
+
+ reply.ret_code = htobe32((uint32_t) reply_code);
+ send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
+ sizeof(struct lttcomm_relayd_generic_reply), 0);
+ if (send_ret < (ssize_t) sizeof(reply)) {
+ ERR("Failed to send \"rotate session stream\" command reply (ret = %zd)",
+ send_ret);
+ ret = -1;
+ }
+end_no_reply:
+ lttng_trace_chunk_put(next_trace_chunk);
+ return ret;
+}
+
+
+
+/*
+ * relay_create_trace_chunk: create a new trace chunk
*/
-static int relay_process_data(struct relay_connection *conn)
+static int relay_create_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr,
+ struct relay_connection *conn,
+ const struct lttng_buffer_view *payload)
{
- int ret = 0, rotate_index = 0;
- ssize_t size_ret;
- struct relay_stream *stream;
- struct lttcomm_relayd_data_hdr data_hdr;
- uint64_t stream_id;
- uint64_t net_seq_num;
- uint32_t data_size;
- struct relay_session *session;
- bool new_stream = false, close_requested = false;
- size_t chunk_size = RECV_DATA_BUFFER_SIZE;
- size_t recv_off = 0;
- char data_buffer[chunk_size];
+ int ret = 0;
+ ssize_t send_ret;
+ struct relay_session *session = conn->session;
+ struct lttcomm_relayd_create_trace_chunk *msg;
+ struct lttcomm_relayd_generic_reply reply = {};
+ struct lttng_buffer_view header_view;
+ struct lttng_buffer_view chunk_name_view;
+ struct lttng_trace_chunk *chunk = NULL, *published_chunk = NULL;
+ enum lttng_error_code reply_code = LTTNG_OK;
+ enum lttng_trace_chunk_status chunk_status;
+ const char *new_path;
+
+ if (!session || !conn->version_check_done) {
+ ERR("Trying to create a trace chunk before version check");
+ ret = -1;
+ goto end_no_reply;
+ }
+
+ if (session->major == 2 && session->minor < 11) {
+ ERR("Chunk creation command is unsupported before 2.11");
+ ret = -1;
+ goto end_no_reply;
+ }
- ret = conn->sock->ops->recvmsg(conn->sock, &data_hdr,
- sizeof(struct lttcomm_relayd_data_hdr), 0);
- if (ret <= 0) {
- if (ret == 0) {
- /* Orderly shutdown. Not necessary to print an error. */
- DBG("Socket %d did an orderly shutdown", conn->sock->fd);
+ header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
+ if (!header_view.data) {
+ ERR("Failed to receive payload of chunk creation command");
+ ret = -1;
+ goto end_no_reply;
+ }
+
+ /* Convert to host endianness. */
+ msg = (typeof(msg)) header_view.data;
+ msg->chunk_id = be64toh(msg->chunk_id);
+ msg->creation_timestamp = be64toh(msg->creation_timestamp);
+ msg->override_name_length = be32toh(msg->override_name_length);
+
+ if (session->current_trace_chunk &&
+ !lttng_trace_chunk_get_name_overridden(session->current_trace_chunk)) {
+ chunk_status = lttng_trace_chunk_rename_path(session->current_trace_chunk,
+ DEFAULT_CHUNK_TMP_OLD_DIRECTORY);
+ if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
+ ERR("Failed to rename old chunk");
+ ret = -1;
+ reply_code = LTTNG_ERR_UNK;
+ goto end;
+ }
+ }
+ session->ongoing_rotation = true;
+ if (!session->current_trace_chunk) {
+ if (!session->has_rotated) {
+ new_path = "";
} else {
- ERR("Unable to receive data header on sock %d", conn->sock->fd);
+ new_path = NULL;
}
+ } else {
+ new_path = DEFAULT_CHUNK_TMP_NEW_DIRECTORY;
+ }
+ chunk = lttng_trace_chunk_create(
+ msg->chunk_id, msg->creation_timestamp, new_path);
+ if (!chunk) {
+ ERR("Failed to create trace chunk in trace chunk creation command");
ret = -1;
+ reply_code = LTTNG_ERR_NOMEM;
goto end;
}
- stream_id = be64toh(data_hdr.stream_id);
- stream = stream_get_by_id(stream_id);
- if (!stream) {
- ERR("relay_process_data: Cannot find stream %" PRIu64, stream_id);
+ if (msg->override_name_length) {
+ const char *name;
+
+ chunk_name_view = lttng_buffer_view_from_view(payload,
+ sizeof(*msg),
+ msg->override_name_length);
+ name = chunk_name_view.data;
+ if (!name || name[msg->override_name_length - 1]) {
+ ERR("Failed to receive payload of chunk creation command");
+ ret = -1;
+ reply_code = LTTNG_ERR_INVALID;
+ goto end;
+ }
+
+ chunk_status = lttng_trace_chunk_override_name(
+ chunk, chunk_name_view.data);
+ switch (chunk_status) {
+ case LTTNG_TRACE_CHUNK_STATUS_OK:
+ break;
+ case LTTNG_TRACE_CHUNK_STATUS_INVALID_ARGUMENT:
+ ERR("Failed to set the name of new trace chunk in trace chunk creation command (invalid name)");
+ reply_code = LTTNG_ERR_INVALID;
+ ret = -1;
+ goto end;
+ default:
+ ERR("Failed to set the name of new trace chunk in trace chunk creation command (unknown error)");
+ reply_code = LTTNG_ERR_UNK;
+ ret = -1;
+ goto end;
+ }
+ }
+
+ chunk_status = lttng_trace_chunk_set_credentials_current_user(chunk);
+ if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
+ reply_code = LTTNG_ERR_UNK;
ret = -1;
goto end;
}
- session = stream->trace->session;
- data_size = be32toh(data_hdr.data_size);
- net_seq_num = be64toh(data_hdr.net_seq_num);
+ assert(conn->session->output_directory);
+ chunk_status = lttng_trace_chunk_set_as_owner(chunk,
+ conn->session->output_directory);
+ if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
+ reply_code = LTTNG_ERR_UNK;
+ ret = -1;
+ goto end;
+ }
- DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64,
- data_size, stream_id, net_seq_num);
+ published_chunk = sessiond_trace_chunk_registry_publish_chunk(
+ sessiond_trace_chunk_registry,
+ conn->session->sessiond_uuid,
+ conn->session->id,
+ chunk);
+ if (!published_chunk) {
+ char uuid_str[LTTNG_UUID_STR_LEN];
- pthread_mutex_lock(&stream->lock);
+ lttng_uuid_to_str(conn->session->sessiond_uuid, uuid_str);
+ ERR("Failed to publish chunk: sessiond_uuid = %s, session_id = %" PRIu64 ", chunk_id = %" PRIu64,
+ uuid_str,
+ conn->session->id,
+ msg->chunk_id);
+ ret = -1;
+ reply_code = LTTNG_ERR_NOMEM;
+ goto end;
+ }
- /* Check if a rotation is needed. */
- if (stream->tracefile_size > 0 &&
- (stream->tracefile_size_current + data_size) >
- stream->tracefile_size) {
- uint64_t old_id, new_id;
+ pthread_mutex_lock(&conn->session->lock);
+ if (conn->session->pending_closure_trace_chunk) {
+ /*
+ * Invalid; this means a second create_trace_chunk command was
+ * received before a close_trace_chunk.
+ */
+ ERR("Invalid trace chunk close command received; a trace chunk is already waiting for a trace chunk close command");
+ reply_code = LTTNG_ERR_INVALID_PROTOCOL;
+ ret = -1;
+ goto end_unlock_session;
+ }
+ conn->session->pending_closure_trace_chunk =
+ conn->session->current_trace_chunk;
+ conn->session->current_trace_chunk = published_chunk;
+ published_chunk = NULL;
+ if (!conn->session->pending_closure_trace_chunk) {
+ session->ongoing_rotation = false;
+ }
+end_unlock_session:
+ pthread_mutex_unlock(&conn->session->lock);
+end:
+ reply.ret_code = htobe32((uint32_t) reply_code);
+ send_ret = conn->sock->ops->sendmsg(conn->sock,
+ &reply,
+ sizeof(struct lttcomm_relayd_generic_reply),
+ 0);
+ if (send_ret < (ssize_t) sizeof(reply)) {
+ ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)",
+ send_ret);
+ ret = -1;
+ }
+end_no_reply:
+ lttng_trace_chunk_put(chunk);
+ lttng_trace_chunk_put(published_chunk);
+ return ret;
+}
- old_id = tracefile_array_get_file_index_head(stream->tfa);
- tracefile_array_file_rotate(stream->tfa);
+/*
+ * relay_close_trace_chunk: close a trace chunk
+ */
+static int relay_close_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr,
+ struct relay_connection *conn,
+ const struct lttng_buffer_view *payload)
+{
+ int ret = 0, buf_ret;
+ ssize_t send_ret;
+ struct relay_session *session = conn->session;
+ struct lttcomm_relayd_close_trace_chunk *msg;
+ struct lttcomm_relayd_close_trace_chunk_reply reply = {};
+ struct lttng_buffer_view header_view;
+ struct lttng_trace_chunk *chunk = NULL;
+ enum lttng_error_code reply_code = LTTNG_OK;
+ enum lttng_trace_chunk_status chunk_status;
+ uint64_t chunk_id;
+ LTTNG_OPTIONAL(enum lttng_trace_chunk_command_type) close_command = {};
+ time_t close_timestamp;
+ char closed_trace_chunk_path[LTTNG_PATH_MAX];
+ size_t path_length = 0;
+ const char *chunk_name = NULL;
+ struct lttng_dynamic_buffer reply_payload;
+ const char *new_path;
+
+ lttng_dynamic_buffer_init(&reply_payload);
+
+ if (!session || !conn->version_check_done) {
+ ERR("Trying to close a trace chunk before version check");
+ ret = -1;
+ goto end_no_reply;
+ }
- /* new_id is updated by utils_rotate_stream_file. */
- new_id = old_id;
+ if (session->major == 2 && session->minor < 11) {
+ ERR("Chunk close command is unsupported before 2.11");
+ ret = -1;
+ goto end_no_reply;
+ }
- ret = utils_rotate_stream_file(stream->path_name,
- stream->channel_name, stream->tracefile_size,
- stream->tracefile_count, -1,
- -1, stream->stream_fd->fd,
- &new_id, &stream->stream_fd->fd);
- if (ret < 0) {
- ERR("Rotating stream output file");
- goto end_stream_unlock;
+ header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
+ if (!header_view.data) {
+ ERR("Failed to receive payload of chunk close command");
+ ret = -1;
+ goto end_no_reply;
+ }
+
+ /* Convert to host endianness. */
+ msg = (typeof(msg)) header_view.data;
+ chunk_id = be64toh(msg->chunk_id);
+ close_timestamp = (time_t) be64toh(msg->close_timestamp);
+ close_command = (typeof(close_command)){
+ .value = be32toh(msg->close_command.value),
+ .is_set = msg->close_command.is_set,
+ };
+
+ chunk = sessiond_trace_chunk_registry_get_chunk(
+ sessiond_trace_chunk_registry,
+ conn->session->sessiond_uuid,
+ conn->session->id,
+ chunk_id);
+ if (!chunk) {
+ char uuid_str[LTTNG_UUID_STR_LEN];
+
+ lttng_uuid_to_str(conn->session->sessiond_uuid, uuid_str);
+ ERR("Failed to find chunk to close: sessiond_uuid = %s, session_id = %" PRIu64 ", chunk_id = %" PRIu64,
+ uuid_str,
+ conn->session->id,
+ msg->chunk_id);
+ ret = -1;
+ reply_code = LTTNG_ERR_NOMEM;
+ goto end;
+ }
+
+ pthread_mutex_lock(&session->lock);
+ if (close_command.is_set &&
+ close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_DELETE) {
+ /*
+ * Clear command. It is a protocol error to ask for a
+ * clear on a relay which does not allow it. Querying
+ * the configuration allows figuring out whether
+ * clearing is allowed before doing the clear.
+ */
+ if (!opt_allow_clear) {
+ ret = -1;
+ reply_code = LTTNG_ERR_INVALID_PROTOCOL;
+ goto end_unlock_session;
+ }
+ }
+ if (session->pending_closure_trace_chunk &&
+ session->pending_closure_trace_chunk != chunk) {
+ ERR("Trace chunk close command for session \"%s\" does not target the trace chunk pending closure",
+ session->session_name);
+ reply_code = LTTNG_ERR_INVALID_PROTOCOL;
+ ret = -1;
+ goto end_unlock_session;
+ }
+
+ if (session->current_trace_chunk && session->current_trace_chunk != chunk &&
+ !lttng_trace_chunk_get_name_overridden(session->current_trace_chunk)) {
+ if (close_command.is_set &&
+ close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_DELETE &&
+ !session->has_rotated) {
+ /* New chunk stays in session output directory. */
+ new_path = "";
+ } else {
+ /* Use chunk name for new chunk. */
+ new_path = NULL;
+ }
+ /* Rename new chunk path. */
+ chunk_status = lttng_trace_chunk_rename_path(session->current_trace_chunk,
+ new_path);
+ if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
+ ret = -1;
+ goto end;
+ }
+ session->ongoing_rotation = false;
+ }
+ if ((!close_command.is_set ||
+ close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_NO_OPERATION) &&
+ !lttng_trace_chunk_get_name_overridden(chunk)) {
+ const char *old_path;
+
+ if (!session->has_rotated) {
+ old_path = "";
+ } else {
+ old_path = NULL;
+ }
+ /* We need to move back the .tmp_old_chunk to its rightful place. */
+ chunk_status = lttng_trace_chunk_rename_path(chunk, old_path);
+ if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
+ ret = -1;
+ goto end;
+ }
+ }
+ chunk_status = lttng_trace_chunk_set_close_timestamp(
+ chunk, close_timestamp);
+ if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
+ ERR("Failed to set trace chunk close timestamp");
+ ret = -1;
+ reply_code = LTTNG_ERR_UNK;
+ goto end_unlock_session;
+ }
+
+ if (close_command.is_set) {
+ chunk_status = lttng_trace_chunk_set_close_command(
+ chunk, close_command.value);
+ if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
+ ret = -1;
+ reply_code = LTTNG_ERR_INVALID;
+ goto end_unlock_session;
+ }
+ }
+ chunk_status = lttng_trace_chunk_get_name(chunk, &chunk_name, NULL);
+ if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
+ ERR("Failed to get chunk name");
+ ret = -1;
+ reply_code = LTTNG_ERR_UNK;
+ goto end_unlock_session;
+ }
+ if (!session->has_rotated && !session->snapshot) {
+ ret = lttng_strncpy(closed_trace_chunk_path,
+ session->output_path,
+ sizeof(closed_trace_chunk_path));
+ if (ret) {
+ ERR("Failed to send trace chunk path: path length of %zu bytes exceeds the maximal allowed length of %zu bytes",
+ strlen(session->output_path),
+ sizeof(closed_trace_chunk_path));
+ reply_code = LTTNG_ERR_NOMEM;
+ ret = -1;
+ goto end_unlock_session;
+ }
+ } else {
+ if (session->snapshot) {
+ ret = snprintf(closed_trace_chunk_path,
+ sizeof(closed_trace_chunk_path),
+ "%s/%s", session->output_path,
+ chunk_name);
+ } else {
+ ret = snprintf(closed_trace_chunk_path,
+ sizeof(closed_trace_chunk_path),
+ "%s/" DEFAULT_ARCHIVED_TRACE_CHUNKS_DIRECTORY
+ "/%s",
+ session->output_path, chunk_name);
+ }
+ if (ret < 0 || ret == sizeof(closed_trace_chunk_path)) {
+ ERR("Failed to format closed trace chunk resulting path");
+ reply_code = ret < 0 ? LTTNG_ERR_UNK : LTTNG_ERR_NOMEM;
+ ret = -1;
+ goto end_unlock_session;
}
+ }
+ if (close_command.is_set &&
+ close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED) {
+ session->has_rotated = true;
+ }
+ DBG("Reply chunk path on close: %s", closed_trace_chunk_path);
+ path_length = strlen(closed_trace_chunk_path) + 1;
+ if (path_length > UINT32_MAX) {
+ ERR("Closed trace chunk path exceeds the maximal length allowed by the protocol");
+ ret = -1;
+ reply_code = LTTNG_ERR_INVALID_PROTOCOL;
+ goto end_unlock_session;
+ }
+
+ if (session->current_trace_chunk == chunk) {
/*
- * Reset current size because we just performed a stream
- * rotation.
+ * After a trace chunk close command, no new streams
+ * referencing the chunk may be created. Hence, on the
+ * event that no new trace chunk have been created for
+ * the session, the reference to the current trace chunk
+ * is released in order to allow it to be reclaimed when
+ * the last stream releases its reference to it.
*/
- stream->tracefile_size_current = 0;
- rotate_index = 1;
+ lttng_trace_chunk_put(session->current_trace_chunk);
+ session->current_trace_chunk = NULL;
+ }
+ lttng_trace_chunk_put(session->pending_closure_trace_chunk);
+ session->pending_closure_trace_chunk = NULL;
+end_unlock_session:
+ pthread_mutex_unlock(&session->lock);
+
+end:
+ reply.generic.ret_code = htobe32((uint32_t) reply_code);
+ reply.path_length = htobe32((uint32_t) path_length);
+ buf_ret = lttng_dynamic_buffer_append(
+ &reply_payload, &reply, sizeof(reply));
+ if (buf_ret) {
+ ERR("Failed to append \"close trace chunk\" command reply header to payload buffer");
+ goto end_no_reply;
+ }
+
+ if (reply_code == LTTNG_OK) {
+ buf_ret = lttng_dynamic_buffer_append(&reply_payload,
+ closed_trace_chunk_path, path_length);
+ if (buf_ret) {
+ ERR("Failed to append \"close trace chunk\" command reply path to payload buffer");
+ goto end_no_reply;
+ }
+ }
+
+ send_ret = conn->sock->ops->sendmsg(conn->sock,
+ reply_payload.data,
+ reply_payload.size,
+ 0);
+ if (send_ret < reply_payload.size) {
+ ERR("Failed to send \"close trace chunk\" command reply of %zu bytes (ret = %zd)",
+ reply_payload.size, send_ret);
+ ret = -1;
+ goto end_no_reply;
}
+end_no_reply:
+ lttng_trace_chunk_put(chunk);
+ lttng_dynamic_buffer_reset(&reply_payload);
+ return ret;
+}
+/*
+ * relay_trace_chunk_exists: check if a trace chunk exists
+ */
+static int relay_trace_chunk_exists(const struct lttcomm_relayd_hdr *recv_hdr,
+ struct relay_connection *conn,
+ const struct lttng_buffer_view *payload)
+{
+ int ret = 0;
+ ssize_t send_ret;
+ struct relay_session *session = conn->session;
+ struct lttcomm_relayd_trace_chunk_exists *msg;
+ struct lttcomm_relayd_trace_chunk_exists_reply reply = {};
+ struct lttng_buffer_view header_view;
+ uint64_t chunk_id;
+ bool chunk_exists;
+
+ if (!session || !conn->version_check_done) {
+ ERR("Trying to close a trace chunk before version check");
+ ret = -1;
+ goto end_no_reply;
+ }
+
+ if (session->major == 2 && session->minor < 11) {
+ ERR("Chunk close command is unsupported before 2.11");
+ ret = -1;
+ goto end_no_reply;
+ }
+
+ header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
+ if (!header_view.data) {
+ ERR("Failed to receive payload of chunk close command");
+ ret = -1;
+ goto end_no_reply;
+ }
+
+ /* Convert to host endianness. */
+ msg = (typeof(msg)) header_view.data;
+ chunk_id = be64toh(msg->chunk_id);
+
+ ret = sessiond_trace_chunk_registry_chunk_exists(
+ sessiond_trace_chunk_registry,
+ conn->session->sessiond_uuid,
+ conn->session->id,
+ chunk_id, &chunk_exists);
/*
- * Index are handled in protocol version 2.4 and above. Also,
- * snapshot and index are NOT supported.
+ * If ret is not 0, send the reply and report the error to the caller.
+ * It is a protocol (or internal) error and the session/connection
+ * should be torn down.
*/
- if (session->minor >= 4 && !session->snapshot) {
- ret = handle_index_data(stream, net_seq_num, rotate_index);
- if (ret < 0) {
- ERR("handle_index_data: fail stream %" PRIu64 " net_seq_num %" PRIu64 " ret %d",
- stream->stream_handle, net_seq_num, ret);
+ reply = (typeof(reply)){
+ .generic.ret_code = htobe32((uint32_t)
+ (ret == 0 ? LTTNG_OK : LTTNG_ERR_INVALID_PROTOCOL)),
+ .trace_chunk_exists = ret == 0 ? chunk_exists : 0,
+ };
+ send_ret = conn->sock->ops->sendmsg(
+ conn->sock, &reply, sizeof(reply), 0);
+ if (send_ret < (ssize_t) sizeof(reply)) {
+ ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)",
+ send_ret);
+ ret = -1;
+ }
+end_no_reply:
+ return ret;
+}
+
+/*
+ * relay_get_configuration: query whether feature is available
+ */
+static int relay_get_configuration(const struct lttcomm_relayd_hdr *recv_hdr,
+ struct relay_connection *conn,
+ const struct lttng_buffer_view *payload)
+{
+ int ret = 0;
+ ssize_t send_ret;
+ struct lttcomm_relayd_get_configuration *msg;
+ struct lttcomm_relayd_get_configuration_reply reply = {};
+ struct lttng_buffer_view header_view;
+ uint64_t query_flags = 0;
+ uint64_t result_flags = 0;
+
+ header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
+ if (!header_view.data) {
+ ERR("Failed to receive payload of chunk close command");
+ ret = -1;
+ goto end_no_reply;
+ }
+
+ /* Convert to host endianness. */
+ msg = (typeof(msg)) header_view.data;
+ query_flags = be64toh(msg->query_flags);
+
+ if (query_flags) {
+ ret = LTTNG_ERR_INVALID_PROTOCOL;
+ goto reply;
+ }
+ if (opt_allow_clear) {
+ result_flags |= LTTCOMM_RELAYD_CONFIGURATION_FLAG_CLEAR_ALLOWED;
+ }
+ ret = 0;
+reply:
+ reply = (typeof(reply)){
+ .generic.ret_code = htobe32((uint32_t)
+ (ret == 0 ? LTTNG_OK : LTTNG_ERR_INVALID_PROTOCOL)),
+ .relayd_configuration_flags = htobe64(result_flags),
+ };
+ send_ret = conn->sock->ops->sendmsg(
+ conn->sock, &reply, sizeof(reply), 0);
+ if (send_ret < (ssize_t) sizeof(reply)) {
+ ERR("Failed to send \"get configuration\" command reply (ret = %zd)",
+ send_ret);
+ ret = -1;
+ }
+end_no_reply:
+ return ret;
+}
+
+#define DBG_CMD(cmd_name, conn) \
+ DBG3("Processing \"%s\" command for socket %i", cmd_name, conn->sock->fd);
+
+static int relay_process_control_command(struct relay_connection *conn,
+ const struct lttcomm_relayd_hdr *header,
+ const struct lttng_buffer_view *payload)
+{
+ int ret = 0;
+
+ switch (header->cmd) {
+ case RELAYD_CREATE_SESSION:
+ DBG_CMD("RELAYD_CREATE_SESSION", conn);
+ ret = relay_create_session(header, conn, payload);
+ break;
+ case RELAYD_ADD_STREAM:
+ DBG_CMD("RELAYD_ADD_STREAM", conn);
+ ret = relay_add_stream(header, conn, payload);
+ break;
+ case RELAYD_START_DATA:
+ DBG_CMD("RELAYD_START_DATA", conn);
+ ret = relay_start(header, conn, payload);
+ break;
+ case RELAYD_SEND_METADATA:
+ DBG_CMD("RELAYD_SEND_METADATA", conn);
+ ret = relay_recv_metadata(header, conn, payload);
+ break;
+ case RELAYD_VERSION:
+ DBG_CMD("RELAYD_VERSION", conn);
+ ret = relay_send_version(header, conn, payload);
+ break;
+ case RELAYD_CLOSE_STREAM:
+ DBG_CMD("RELAYD_CLOSE_STREAM", conn);
+ ret = relay_close_stream(header, conn, payload);
+ break;
+ case RELAYD_DATA_PENDING:
+ DBG_CMD("RELAYD_DATA_PENDING", conn);
+ ret = relay_data_pending(header, conn, payload);
+ break;
+ case RELAYD_QUIESCENT_CONTROL:
+ DBG_CMD("RELAYD_QUIESCENT_CONTROL", conn);
+ ret = relay_quiescent_control(header, conn, payload);
+ break;
+ case RELAYD_BEGIN_DATA_PENDING:
+ DBG_CMD("RELAYD_BEGIN_DATA_PENDING", conn);
+ ret = relay_begin_data_pending(header, conn, payload);
+ break;
+ case RELAYD_END_DATA_PENDING:
+ DBG_CMD("RELAYD_END_DATA_PENDING", conn);
+ ret = relay_end_data_pending(header, conn, payload);
+ break;
+ case RELAYD_SEND_INDEX:
+ DBG_CMD("RELAYD_SEND_INDEX", conn);
+ ret = relay_recv_index(header, conn, payload);
+ break;
+ case RELAYD_STREAMS_SENT:
+ DBG_CMD("RELAYD_STREAMS_SENT", conn);
+ ret = relay_streams_sent(header, conn, payload);
+ break;
+ case RELAYD_RESET_METADATA:
+ DBG_CMD("RELAYD_RESET_METADATA", conn);
+ ret = relay_reset_metadata(header, conn, payload);
+ break;
+ case RELAYD_ROTATE_STREAMS:
+ DBG_CMD("RELAYD_ROTATE_STREAMS", conn);
+ ret = relay_rotate_session_streams(header, conn, payload);
+ break;
+ case RELAYD_CREATE_TRACE_CHUNK:
+ DBG_CMD("RELAYD_CREATE_TRACE_CHUNK", conn);
+ ret = relay_create_trace_chunk(header, conn, payload);
+ break;
+ case RELAYD_CLOSE_TRACE_CHUNK:
+ DBG_CMD("RELAYD_CLOSE_TRACE_CHUNK", conn);
+ ret = relay_close_trace_chunk(header, conn, payload);
+ break;
+ case RELAYD_TRACE_CHUNK_EXISTS:
+ DBG_CMD("RELAYD_TRACE_CHUNK_EXISTS", conn);
+ ret = relay_trace_chunk_exists(header, conn, payload);
+ break;
+ case RELAYD_GET_CONFIGURATION:
+ DBG_CMD("RELAYD_GET_CONFIGURATION", conn);
+ ret = relay_get_configuration(header, conn, payload);
+ break;
+ case RELAYD_UPDATE_SYNC_INFO:
+ default:
+ ERR("Received unknown command (%u)", header->cmd);
+ relay_unknown_command(conn);
+ ret = -1;
+ goto end;
+ }
+
+end:
+ return ret;
+}
+
+static enum relay_connection_status relay_process_control_receive_payload(
+ struct relay_connection *conn)
+{
+ int ret = 0;
+ enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
+ struct lttng_dynamic_buffer *reception_buffer =
+ &conn->protocol.ctrl.reception_buffer;
+ struct ctrl_connection_state_receive_payload *state =
+ &conn->protocol.ctrl.state.receive_payload;
+ struct lttng_buffer_view payload_view;
+
+ if (state->left_to_receive == 0) {
+ /* Short-circuit for payload-less commands. */
+ goto reception_complete;
+ }
+
+ ret = conn->sock->ops->recvmsg(conn->sock,
+ reception_buffer->data + state->received,
+ state->left_to_receive, MSG_DONTWAIT);
+ if (ret < 0) {
+ if (errno != EAGAIN && errno != EWOULDBLOCK) {
+ PERROR("Unable to receive command payload on sock %d",
+ conn->sock->fd);
+ status = RELAY_CONNECTION_STATUS_ERROR;
+ }
+ goto end;
+ } else if (ret == 0) {
+ DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
+ status = RELAY_CONNECTION_STATUS_CLOSED;
+ goto end;
+ }
+
+ assert(ret > 0);
+ assert(ret <= state->left_to_receive);
+
+ state->left_to_receive -= ret;
+ state->received += ret;
+
+ if (state->left_to_receive > 0) {
+ /*
+ * Can't transition to the protocol's next state, wait to
+ * receive the rest of the header.
+ */
+ DBG3("Partial reception of control connection protocol payload (received %" PRIu64 " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
+ state->received, state->left_to_receive,
+ conn->sock->fd);
+ goto end;
+ }
+
+reception_complete:
+ DBG("Done receiving control command payload: fd = %i, payload size = %" PRIu64 " bytes",
+ conn->sock->fd, state->received);
+ /*
+ * The payload required to process the command has been received.
+ * A view to the reception buffer is forwarded to the various
+ * commands and the state of the control is reset on success.
+ *
+ * Commands are responsible for sending their reply to the peer.
+ */
+ payload_view = lttng_buffer_view_from_dynamic_buffer(reception_buffer,
+ 0, -1);
+ ret = relay_process_control_command(conn,
+ &state->header, &payload_view);
+ if (ret < 0) {
+ status = RELAY_CONNECTION_STATUS_ERROR;
+ goto end;
+ }
+
+ ret = connection_reset_protocol_state(conn);
+ if (ret) {
+ status = RELAY_CONNECTION_STATUS_ERROR;
+ }
+end:
+ return status;
+}
+
+static enum relay_connection_status relay_process_control_receive_header(
+ struct relay_connection *conn)
+{
+ int ret = 0;
+ enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
+ struct lttcomm_relayd_hdr header;
+ struct lttng_dynamic_buffer *reception_buffer =
+ &conn->protocol.ctrl.reception_buffer;
+ struct ctrl_connection_state_receive_header *state =
+ &conn->protocol.ctrl.state.receive_header;
+
+ assert(state->left_to_receive != 0);
+
+ ret = conn->sock->ops->recvmsg(conn->sock,
+ reception_buffer->data + state->received,
+ state->left_to_receive, MSG_DONTWAIT);
+ if (ret < 0) {
+ if (errno != EAGAIN && errno != EWOULDBLOCK) {
+ PERROR("Unable to receive control command header on sock %d",
+ conn->sock->fd);
+ status = RELAY_CONNECTION_STATUS_ERROR;
+ }
+ goto end;
+ } else if (ret == 0) {
+ DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
+ status = RELAY_CONNECTION_STATUS_CLOSED;
+ goto end;
+ }
+
+ assert(ret > 0);
+ assert(ret <= state->left_to_receive);
+
+ state->left_to_receive -= ret;
+ state->received += ret;
+
+ if (state->left_to_receive > 0) {
+ /*
+ * Can't transition to the protocol's next state, wait to
+ * receive the rest of the header.
+ */
+ DBG3("Partial reception of control connection protocol header (received %" PRIu64 " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
+ state->received, state->left_to_receive,
+ conn->sock->fd);
+ goto end;
+ }
+
+ /* Transition to next state: receiving the command's payload. */
+ conn->protocol.ctrl.state_id =
+ CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD;
+ memcpy(&header, reception_buffer->data, sizeof(header));
+ header.circuit_id = be64toh(header.circuit_id);
+ header.data_size = be64toh(header.data_size);
+ header.cmd = be32toh(header.cmd);
+ header.cmd_version = be32toh(header.cmd_version);
+ memcpy(&conn->protocol.ctrl.state.receive_payload.header,
+ &header, sizeof(header));
+
+ DBG("Done receiving control command header: fd = %i, cmd = %" PRIu32 ", cmd_version = %" PRIu32 ", payload size = %" PRIu64 " bytes",
+ conn->sock->fd, header.cmd, header.cmd_version,
+ header.data_size);
+
+ if (header.data_size > DEFAULT_NETWORK_RELAYD_CTRL_MAX_PAYLOAD_SIZE) {
+ ERR("Command header indicates a payload (%" PRIu64 " bytes) that exceeds the maximal payload size allowed on a control connection.",
+ header.data_size);
+ status = RELAY_CONNECTION_STATUS_ERROR;
+ goto end;
+ }
+
+ conn->protocol.ctrl.state.receive_payload.left_to_receive =
+ header.data_size;
+ conn->protocol.ctrl.state.receive_payload.received = 0;
+ ret = lttng_dynamic_buffer_set_size(reception_buffer,
+ header.data_size);
+ if (ret) {
+ status = RELAY_CONNECTION_STATUS_ERROR;
+ goto end;
+ }
+
+ if (header.data_size == 0) {
+ /*
+ * Manually invoke the next state as the poll loop
+ * will not wake-up to allow us to proceed further.
+ */
+ status = relay_process_control_receive_payload(conn);
+ }
+end:
+ return status;
+}
+
+/*
+ * Process the commands received on the control socket
+ */
+static enum relay_connection_status relay_process_control(
+ struct relay_connection *conn)
+{
+ enum relay_connection_status status;
+
+ switch (conn->protocol.ctrl.state_id) {
+ case CTRL_CONNECTION_STATE_RECEIVE_HEADER:
+ status = relay_process_control_receive_header(conn);
+ break;
+ case CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD:
+ status = relay_process_control_receive_payload(conn);
+ break;
+ default:
+ ERR("Unknown control connection protocol state encountered.");
+ abort();
+ }
+
+ return status;
+}
+
+static enum relay_connection_status relay_process_data_receive_header(
+ struct relay_connection *conn)
+{
+ int ret;
+ enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
+ struct data_connection_state_receive_header *state =
+ &conn->protocol.data.state.receive_header;
+ struct lttcomm_relayd_data_hdr header;
+ struct relay_stream *stream;
+
+ assert(state->left_to_receive != 0);
+
+ ret = conn->sock->ops->recvmsg(conn->sock,
+ state->header_reception_buffer + state->received,
+ state->left_to_receive, MSG_DONTWAIT);
+ if (ret < 0) {
+ if (errno != EAGAIN && errno != EWOULDBLOCK) {
+ PERROR("Unable to receive data header on sock %d", conn->sock->fd);
+ status = RELAY_CONNECTION_STATUS_ERROR;
+ }
+ goto end;
+ } else if (ret == 0) {
+ /* Orderly shutdown. Not necessary to print an error. */
+ DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
+ status = RELAY_CONNECTION_STATUS_CLOSED;
+ goto end;
+ }
+
+ assert(ret > 0);
+ assert(ret <= state->left_to_receive);
+
+ state->left_to_receive -= ret;
+ state->received += ret;
+
+ if (state->left_to_receive > 0) {
+ /*
+ * Can't transition to the protocol's next state, wait to
+ * receive the rest of the header.
+ */
+ DBG3("Partial reception of data connection header (received %" PRIu64 " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
+ state->received, state->left_to_receive,
+ conn->sock->fd);
+ goto end;
+ }
+
+ /* Transition to next state: receiving the payload. */
+ conn->protocol.data.state_id = DATA_CONNECTION_STATE_RECEIVE_PAYLOAD;
+
+ memcpy(&header, state->header_reception_buffer, sizeof(header));
+ header.circuit_id = be64toh(header.circuit_id);
+ header.stream_id = be64toh(header.stream_id);
+ header.data_size = be32toh(header.data_size);
+ header.net_seq_num = be64toh(header.net_seq_num);
+ header.padding_size = be32toh(header.padding_size);
+ memcpy(&conn->protocol.data.state.receive_payload.header, &header, sizeof(header));
+
+ conn->protocol.data.state.receive_payload.left_to_receive =
+ header.data_size;
+ conn->protocol.data.state.receive_payload.received = 0;
+ conn->protocol.data.state.receive_payload.rotate_index = false;
+
+ DBG("Received data connection header on fd %i: circuit_id = %" PRIu64 ", stream_id = %" PRIu64 ", data_size = %" PRIu32 ", net_seq_num = %" PRIu64 ", padding_size = %" PRIu32,
+ conn->sock->fd, header.circuit_id,
+ header.stream_id, header.data_size,
+ header.net_seq_num, header.padding_size);
+
+ stream = stream_get_by_id(header.stream_id);
+ if (!stream) {
+ DBG("relay_process_data_receive_payload: Cannot find stream %" PRIu64,
+ header.stream_id);
+ /* Protocol error. */
+ status = RELAY_CONNECTION_STATUS_ERROR;
+ goto end;
+ }
+
+ pthread_mutex_lock(&stream->lock);
+ /* Prepare stream for the reception of a new packet. */
+ ret = stream_init_packet(stream, header.data_size,
+ &conn->protocol.data.state.receive_payload.rotate_index);
+ pthread_mutex_unlock(&stream->lock);
+ if (ret) {
+ ERR("Failed to rotate stream output file");
+ status = RELAY_CONNECTION_STATUS_ERROR;
+ goto end_stream_unlock;
+ }
+
+end_stream_unlock:
+ stream_put(stream);
+end:
+ return status;
+}
+
+static enum relay_connection_status relay_process_data_receive_payload(
+ struct relay_connection *conn)
+{
+ int ret;
+ enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
+ struct relay_stream *stream;
+ struct data_connection_state_receive_payload *state =
+ &conn->protocol.data.state.receive_payload;
+ const size_t chunk_size = RECV_DATA_BUFFER_SIZE;
+ char data_buffer[chunk_size];
+ bool partial_recv = false;
+ bool new_stream = false, close_requested = false, index_flushed = false;
+ uint64_t left_to_receive = state->left_to_receive;
+ struct relay_session *session;
+
+ DBG3("Receiving data for stream id %" PRIu64 " seqnum %" PRIu64 ", %" PRIu64" bytes received, %" PRIu64 " bytes left to receive",
+ state->header.stream_id, state->header.net_seq_num,
+ state->received, left_to_receive);
+
+ stream = stream_get_by_id(state->header.stream_id);
+ if (!stream) {
+ /* Protocol error. */
+ ERR("relay_process_data_receive_payload: cannot find stream %" PRIu64,
+ state->header.stream_id);
+ status = RELAY_CONNECTION_STATUS_ERROR;
+ goto end;
+ }
+
+ pthread_mutex_lock(&stream->lock);
+ session = stream->trace->session;
+ if (!conn->session) {
+ ret = connection_set_session(conn, session);
+ if (ret) {
+ status = RELAY_CONNECTION_STATUS_ERROR;
goto end_stream_unlock;
}
}
- for (recv_off = 0; recv_off < data_size; recv_off += chunk_size) {
- size_t recv_size = min(data_size - recv_off, chunk_size);
+ /*
+ * The size of the "chunk" received on any iteration is bounded by:
+ * - the data left to receive,
+ * - the data immediately available on the socket,
+ * - the on-stack data buffer
+ */
+ while (left_to_receive > 0 && !partial_recv) {
+ size_t recv_size = min(left_to_receive, chunk_size);
+ struct lttng_buffer_view packet_chunk;
- ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, recv_size, 0);
- if (ret <= 0) {
- if (ret == 0) {
- /* Orderly shutdown. Not necessary to print an error. */
- DBG("Socket %d did an orderly shutdown", conn->sock->fd);
- } else {
- ERR("Socket %d error %d", conn->sock->fd, ret);
+ ret = conn->sock->ops->recvmsg(conn->sock, data_buffer,
+ recv_size, MSG_DONTWAIT);
+ if (ret < 0) {
+ if (errno != EAGAIN && errno != EWOULDBLOCK) {
+ PERROR("Socket %d error", conn->sock->fd);
+ status = RELAY_CONNECTION_STATUS_ERROR;
}
- ret = -1;
goto end_stream_unlock;
+ } else if (ret == 0) {
+ /* No more data ready to be consumed on socket. */
+ DBG3("No more data ready for consumption on data socket of stream id %" PRIu64,
+ state->header.stream_id);
+ status = RELAY_CONNECTION_STATUS_CLOSED;
+ break;
+ } else if (ret < (int) recv_size) {
+ /*
+ * All the data available on the socket has been
+ * consumed.
+ */
+ partial_recv = true;
+ recv_size = ret;
}
- /* Write data to stream output fd. */
- size_ret = lttng_write(stream->stream_fd->fd, data_buffer,
- recv_size);
- if (size_ret < recv_size) {
+ packet_chunk = lttng_buffer_view_init(data_buffer,
+ 0, recv_size);
+ assert(packet_chunk.data);
+
+ ret = stream_write(stream, &packet_chunk, 0);
+ if (ret) {
ERR("Relay error writing data to file");
- ret = -1;
+ status = RELAY_CONNECTION_STATUS_ERROR;
goto end_stream_unlock;
}
- DBG2("Relay wrote %zd bytes to tracefile for stream id %" PRIu64,
- size_ret, stream->stream_handle);
+ left_to_receive -= recv_size;
+ state->received += recv_size;
+ state->left_to_receive = left_to_receive;
}
- ret = write_padding_to_file(stream->stream_fd->fd,
- be32toh(data_hdr.padding_size));
- if (ret < 0) {
- ERR("write_padding_to_file: fail stream %" PRIu64 " net_seq_num %" PRIu64 " ret %d",
- stream->stream_handle, net_seq_num, ret);
+ if (state->left_to_receive > 0) {
+ /*
+ * Did not receive all the data expected, wait for more data to
+ * become available on the socket.
+ */
+ DBG3("Partial receive on data connection of stream id %" PRIu64 ", %" PRIu64 " bytes received, %" PRIu64 " bytes left to receive",
+ state->header.stream_id, state->received,
+ state->left_to_receive);
goto end_stream_unlock;
}
- stream->tracefile_size_current +=
- data_size + be32toh(data_hdr.padding_size);
- if (stream->prev_seq == -1ULL) {
+
+ ret = stream_write(stream, NULL, state->header.padding_size);
+ if (ret) {
+ status = RELAY_CONNECTION_STATUS_ERROR;
+ goto end_stream_unlock;
+ }
+
+ if (session_streams_have_index(session)) {
+ ret = stream_update_index(stream, state->header.net_seq_num,
+ state->rotate_index, &index_flushed,
+ state->header.data_size + state->header.padding_size);
+ if (ret < 0) {
+ ERR("Failed to update index: stream %" PRIu64 " net_seq_num %" PRIu64 " ret %d",
+ stream->stream_handle,
+ state->header.net_seq_num, ret);
+ status = RELAY_CONNECTION_STATUS_ERROR;
+ goto end_stream_unlock;
+ }
+ }
+
+ if (stream->prev_data_seq == -1ULL) {
new_stream = true;
}
- stream->prev_seq = net_seq_num;
+ ret = stream_complete_packet(stream, state->header.data_size +
+ state->header.padding_size, state->header.net_seq_num,
+ index_flushed);
+ if (ret) {
+ status = RELAY_CONNECTION_STATUS_ERROR;
+ goto end_stream_unlock;
+ }
+
+ /*
+ * Resetting the protocol state (to RECEIVE_HEADER) will trash the
+ * contents of *state which are aliased (union) to the same location as
+ * the new state. Don't use it beyond this point.
+ */
+ connection_reset_protocol_state(conn);
+ state = NULL;
end_stream_unlock:
close_requested = stream->close_requested;
pthread_mutex_unlock(&stream->lock);
- if (close_requested) {
+ if (close_requested && left_to_receive == 0) {
try_stream_close(stream);
}
uatomic_set(&session->new_streams, 1);
pthread_mutex_unlock(&session->lock);
}
+
stream_put(stream);
end:
- return ret;
+ return status;
+}
+
+/*
+ * relay_process_data: Process the data received on the data socket
+ */
+static enum relay_connection_status relay_process_data(
+ struct relay_connection *conn)
+{
+ enum relay_connection_status status;
+
+ switch (conn->protocol.data.state_id) {
+ case DATA_CONNECTION_STATE_RECEIVE_HEADER:
+ status = relay_process_data_receive_header(conn);
+ break;
+ case DATA_CONNECTION_STATE_RECEIVE_PAYLOAD:
+ status = relay_process_data_receive_payload(conn);
+ break;
+ default:
+ ERR("Unexpected data connection communication state.");
+ abort();
+ }
+
+ return status;
}
static void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
(void) lttng_poll_del(events, pollfd);
- ret = close(pollfd);
+ ret = fd_tracker_close_unsuspendable_fd(the_fd_tracker, &pollfd, 1,
+ fd_tracker_util_close_fd, NULL);
if (ret < 0) {
ERR("Closing pollfd %d", pollfd);
}
struct lttng_poll_event events;
struct lttng_ht *relay_connections_ht;
struct lttng_ht_iter iter;
- struct lttcomm_relayd_hdr recv_hdr;
struct relay_connection *destroy_conn = NULL;
DBG("[thread] Relay worker started");
goto relay_connections_ht_error;
}
- ret = create_thread_poll_set(&events, 2);
+ ret = create_named_thread_poll_set(&events, 2, "Worker thread epoll");
if (ret < 0) {
goto error_poll_create;
}
health_code_update();
- if (!revents) {
- /*
- * No activity for this FD (poll
- * implementation).
- */
- continue;
- }
-
/* Thread quit pipe has been closed. Killing thread. */
ret = check_thread_quit_pipe(pollfd, revents);
if (ret) {
if (ret < 0) {
goto error;
}
- lttng_poll_add(&events, conn->sock->fd,
+ ret = lttng_poll_add(&events,
+ conn->sock->fd,
LPOLLIN | LPOLLRDHUP);
+ if (ret) {
+ ERR("Failed to add new connection file descriptor to poll set");
+ goto error;
+ }
connection_ht_add(relay_connections_ht, conn);
DBG("Connection socket %d added", conn->sock->fd);
} else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
assert(ctrl_conn->type == RELAY_CONTROL);
if (revents & LPOLLIN) {
- ret = ctrl_conn->sock->ops->recvmsg(ctrl_conn->sock,
- &recv_hdr, sizeof(recv_hdr), 0);
- if (ret <= 0) {
- /* Connection closed */
- relay_thread_close_connection(&events, pollfd,
- ctrl_conn);
- } else {
- ret = relay_process_control(&recv_hdr, ctrl_conn);
- if (ret < 0) {
- /* Clear the session on error. */
- relay_thread_close_connection(&events,
- pollfd, ctrl_conn);
+ enum relay_connection_status status;
+
+ status = relay_process_control(ctrl_conn);
+ if (status != RELAY_CONNECTION_STATUS_OK) {
+ /*
+ * On socket error flag the session as aborted to force
+ * the cleanup of its stream otherwise it can leak
+ * during the lifetime of the relayd.
+ *
+ * This prevents situations in which streams can be
+ * left opened because an index was received, the
+ * control connection is closed, and the data
+ * connection is closed (uncleanly) before the packet's
+ * data provided.
+ *
+ * Since the control connection encountered an error,
+ * it is okay to be conservative and close the
+ * session right now as we can't rely on the protocol
+ * being respected anymore.
+ */
+ if (status == RELAY_CONNECTION_STATUS_ERROR) {
+ session_abort(ctrl_conn->session);
}
- seen_control = 1;
+
+ /* Clear the connection on error or close. */
+ relay_thread_close_connection(&events,
+ pollfd,
+ ctrl_conn);
}
+ seen_control = 1;
} else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
relay_thread_close_connection(&events,
pollfd, ctrl_conn);
assert(data_conn->type == RELAY_DATA);
if (revents & LPOLLIN) {
- ret = relay_process_data(data_conn);
- /* Connection closed */
- if (ret < 0) {
+ enum relay_connection_status status;
+
+ status = relay_process_data(data_conn);
+ /* Connection closed or error. */
+ if (status != RELAY_CONNECTION_STATUS_OK) {
+ /*
+ * On socket error flag the session as aborted to force
+ * the cleanup of its stream otherwise it can leak
+ * during the lifetime of the relayd.
+ *
+ * This prevents situations in which streams can be
+ * left opened because an index was received, the
+ * control connection is closed, and the data
+ * connection is closed (uncleanly) before the packet's
+ * data provided.
+ *
+ * Since the data connection encountered an error,
+ * it is okay to be conservative and close the
+ * session right now as we can't rely on the protocol
+ * being respected anymore.
+ */
+ if (status == RELAY_CONNECTION_STATUS_ERROR) {
+ session_abort(data_conn->session);
+ }
relay_thread_close_connection(&events, pollfd,
data_conn);
/*
exit:
error:
- /* Cleanup reamaining connection object. */
+ /* Cleanup remaining connection object. */
rcu_read_lock();
cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter,
destroy_conn,
sock_n.node) {
health_code_update();
+
+ session_abort(destroy_conn->session);
+
/*
* No need to grab another ref, because we own
* destroy_conn.
}
rcu_read_unlock();
- lttng_poll_clean(&events);
+ (void) fd_tracker_util_poll_clean(the_fd_tracker, &events);
error_poll_create:
lttng_ht_destroy(relay_connections_ht);
relay_connections_ht_error:
/* Close relay conn pipes */
- utils_close_pipe(relay_conn_pipe);
+ (void) fd_tracker_util_pipe_close(the_fd_tracker,
+ relay_conn_pipe);
if (err) {
DBG("Thread exited with error");
}
*/
static int create_relay_conn_pipe(void)
{
- int ret;
+ return fd_tracker_util_pipe_open_cloexec(the_fd_tracker,
+ "Relayd connection pipe", relay_conn_pipe);
+}
- ret = utils_create_pipe_cloexec(relay_conn_pipe);
+static int stdio_open(void *data, int *fds)
+{
+ fds[0] = fileno(stdout);
+ fds[1] = fileno(stderr);
+ return 0;
+}
- return ret;
+static int track_stdio(void)
+{
+ int fds[2];
+ const char *names[] = { "stdout", "stderr" };
+
+ return fd_tracker_open_unsuspendable_fd(the_fd_tracker, fds,
+ names, 2, stdio_open, NULL);
}
/*
*/
int main(int argc, char **argv)
{
+ bool thread_is_rcu_registered = false;
int ret = 0, retval = 0;
void *status;
+ char *unlinked_file_directory_path = NULL, *output_path = NULL;
+
+ /* Parse environment variables */
+ ret = parse_env_options();
+ if (ret) {
+ retval = -1;
+ goto exit_options;
+ }
- /* Parse arguments */
+ /*
+ * Parse arguments.
+ * Command line arguments overwrite environment.
+ */
progname = argv[0];
if (set_options(argc, argv)) {
retval = -1;
goto exit_options;
}
+ relayd_config_log();
+
+ if (opt_print_version) {
+ print_version();
+ retval = 0;
+ goto exit_options;
+ }
+
+ ret = fclose(stdin);
+ if (ret) {
+ PERROR("Failed to close stdin");
+ goto exit_options;
+ }
+
+ DBG("Clear command %s", opt_allow_clear ? "allowed" : "disallowed");
+
/* Try to create directory if -o, --output is specified. */
if (opt_output_path) {
if (*opt_output_path != '/') {
/* Daemonize */
if (opt_daemon || opt_background) {
- int i;
-
ret = lttng_daemonize(&child_ppid, &recv_child_signal,
!opt_background);
if (ret < 0) {
retval = -1;
goto exit_options;
}
+ }
- /*
- * We are in the child. Make sure all other file
- * descriptors are closed, in case we are called with
- * more opened file descriptors than the standard ones.
- */
- for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
- (void) close(i);
+ if (opt_working_directory) {
+ ret = utils_change_working_directory(opt_working_directory);
+ if (ret) {
+ /* All errors are already logged. */
+ goto exit_options;
}
}
+ sessiond_trace_chunk_registry = sessiond_trace_chunk_registry_create();
+ if (!sessiond_trace_chunk_registry) {
+ ERR("Failed to initialize session daemon trace chunk registry");
+ retval = -1;
+ goto exit_options;
+ }
+
+ /*
+ * The RCU thread registration (and use, through the fd-tracker's
+ * creation) is done after the daemonization to allow us to not
+ * deal with liburcu's fork() management as the call RCU needs to
+ * be restored.
+ */
+ rcu_register_thread();
+ thread_is_rcu_registered = true;
+
+ output_path = create_output_path("");
+ if (!output_path) {
+ ERR("Failed to get output path");
+ retval = -1;
+ goto exit_options;
+ }
+ ret = asprintf(&unlinked_file_directory_path, "%s/%s", output_path,
+ DEFAULT_UNLINKED_FILES_DIRECTORY);
+ free(output_path);
+ if (ret < 0) {
+ ERR("Failed to format unlinked file directory path");
+ retval = -1;
+ goto exit_options;
+ }
+ the_fd_tracker = fd_tracker_create(
+ unlinked_file_directory_path, lttng_opt_fd_pool_size);
+ free(unlinked_file_directory_path);
+ if (!the_fd_tracker) {
+ retval = -1;
+ goto exit_options;
+ }
+
+ ret = track_stdio();
+ if (ret) {
+ retval = -1;
+ goto exit_options;
+ }
/* Initialize thread health monitoring */
health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES);
if (!health_relayd) {
PERROR("health_app_create error");
retval = -1;
- goto exit_health_app_create;
+ goto exit_options;
}
/* Create thread quit pipe */
if (init_thread_quit_pipe()) {
retval = -1;
- goto exit_init_data;
- }
-
- /* Check if daemon is UID = 0 */
- if (!getuid()) {
- if (control_uri->port < 1024 || data_uri->port < 1024 || live_uri->port < 1024) {
- ERR("Need to be root to use ports < 1024");
- retval = -1;
- goto exit_init_data;
- }
+ goto exit_options;
}
/* Setup the thread apps communication pipe. */
if (create_relay_conn_pipe()) {
retval = -1;
- goto exit_init_data;
+ goto exit_options;
}
/* Init relay command queue. */
cds_wfcq_init(&relay_conn_queue.head, &relay_conn_queue.tail);
- /* Set up max poll set size */
- if (lttng_poll_set_max_size()) {
- retval = -1;
- goto exit_init_data;
- }
-
/* Initialize communication library */
lttcomm_init();
lttcomm_inet_init();
sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
if (!sessions_ht) {
retval = -1;
- goto exit_init_data;
+ goto exit_options;
}
/* tables of streams indexed by stream ID */
relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
if (!relay_streams_ht) {
retval = -1;
- goto exit_init_data;
+ goto exit_options;
}
/* tables of streams indexed by stream ID */
viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
if (!viewer_streams_ht) {
retval = -1;
- goto exit_init_data;
+ goto exit_options;
}
- ret = utils_create_pipe(health_quit_pipe);
+ ret = init_health_quit_pipe();
if (ret) {
retval = -1;
- goto exit_health_quit_pipe;
+ goto exit_options;
}
/* Create thread to manage the client socket */
- ret = pthread_create(&health_thread, NULL,
+ ret = pthread_create(&health_thread, default_pthread_attr(),
thread_manage_health, (void *) NULL);
if (ret) {
errno = ret;
PERROR("pthread_create health");
retval = -1;
- goto exit_health_thread;
+ goto exit_options;
}
/* Setup the dispatcher thread */
- ret = pthread_create(&dispatcher_thread, NULL,
+ ret = pthread_create(&dispatcher_thread, default_pthread_attr(),
relay_thread_dispatcher, (void *) NULL);
if (ret) {
errno = ret;
}
/* Setup the worker thread */
- ret = pthread_create(&worker_thread, NULL,
+ ret = pthread_create(&worker_thread, default_pthread_attr(),
relay_thread_worker, NULL);
if (ret) {
errno = ret;
}
/* Setup the listener thread */
- ret = pthread_create(&listener_thread, NULL,
+ ret = pthread_create(&listener_thread, default_pthread_attr(),
relay_thread_listener, (void *) NULL);
if (ret) {
errno = ret;
PERROR("pthread_join health_thread");
retval = -1;
}
-exit_health_thread:
-
- utils_close_pipe(health_quit_pipe);
-exit_health_quit_pipe:
-
-exit_init_data:
- health_app_destroy(health_relayd);
-exit_health_app_create:
exit_options:
+ /*
+ * Wait for all pending call_rcu work to complete before tearing
+ * down data structures. call_rcu worker may be trying to
+ * perform lookups in those structures.
+ */
+ rcu_barrier();
relayd_cleanup();
/* Ensure all prior call_rcu are done. */
rcu_barrier();
+ if (thread_is_rcu_registered) {
+ rcu_unregister_thread();
+ }
+
if (!retval) {
exit(EXIT_SUCCESS);
} else {