/*
* Thread managing health check socket.
*/
void *thread_manage_health(void *data)
{
int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
uint32_t revents, nb_fd;
struct lttng_poll_event events;
struct health_comm_msg msg;
struct health_comm_reply reply;
int is_root;
DBG("[thread] Manage health check started");
setup_health_path();
rcu_register_thread();
/* We might hit an error path before this is created. */
lttng_poll_init(&events);
/* Create unix socket */
sock = lttcomm_create_unix_sock(health_unix_sock_path);
if (sock < 0) {
ERR("Unable to create health check Unix socket");
ret = -1;
goto error;
}
is_root = !getuid();
if (is_root) {
/* lttng health client socket path permissions */
ret = chown(health_unix_sock_path, 0,
utils_get_group_id(tracing_group_name));
if (ret < 0) {
ERR("Unable to set group on %s", health_unix_sock_path);
PERROR("chown");
ret = -1;
goto error;
}
ret = chmod(health_unix_sock_path,
S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
if (ret < 0) {
ERR("Unable to set permissions on %s", health_unix_sock_path);
PERROR("chmod");
ret = -1;
goto error;
}
}
/*
* Set the CLOEXEC flag. Return code is useless because either way, the
* show must go on.
*/
(void) utils_set_fd_cloexec(sock);
ret = lttcomm_listen_unix_sock(sock);
if (ret < 0) {
goto error;
}
/* Size is set to 1 for the consumer_channel pipe */
ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
if (ret < 0) {
ERR("Poll set creation failed");
goto error;
}
ret = lttng_poll_add(&events, health_quit_pipe[0], LPOLLIN);
if (ret < 0) {
goto error;
}
/* Add the application registration socket */
ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
if (ret < 0) {
goto error;
}
/* Perform prior memory accesses before decrementing ready */
cmm_smp_mb__before_uatomic_dec();
uatomic_dec(<tng_consumer_ready);
while (1) {
DBG("Health check ready");
/* Inifinite blocking call, waiting for transmission */
restart:
ret = lttng_poll_wait(&events, -1);
if (ret < 0) {
/*
* Restart interrupted system call.
*/
if (errno == EINTR) {
goto restart;
}
goto error;
}
nb_fd = ret;
for (i = 0; i < nb_fd; i++) {
/* Fetch once the poll data */
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_health_quit_pipe(pollfd, revents);
if (ret) {
err = 0;
goto exit;
}
/* Event on the registration socket */
if (pollfd == sock) {
if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
&& !(revents & LPOLLIN)) {
ERR("Health socket poll error");
goto error;
}
}
}
new_sock = lttcomm_accept_unix_sock(sock);
if (new_sock < 0) {
goto error;
}
/*
* Set the CLOEXEC flag. Return code is useless because either way, the
* show must go on.
*/
(void) utils_set_fd_cloexec(new_sock);
DBG("Receiving data from client for health...");
ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
if (ret <= 0) {
DBG("Nothing recv() from client... continuing");
ret = close(new_sock);
if (ret) {
PERROR("close");
}
new_sock = -1;
continue;
}
rcu_thread_online();
assert(msg.cmd == HEALTH_CMD_CHECK);
memset(&reply, 0, sizeof(reply));
for (i = 0; i < NR_HEALTH_CONSUMERD_TYPES; i++) {
/*
* health_check_state return 0 if thread is in
* error.
*/
if (!health_check_state(health_consumerd, i)) {
reply.ret_code |= 1ULL << i;
}
}
DBG("Health check return value %" PRIx64, reply.ret_code);
ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
if (ret < 0) {
ERR("Failed to send health data back to client");
}
/* End of transmission */
ret = close(new_sock);
if (ret) {
PERROR("close");
}
new_sock = -1;
}
exit:
error:
if (err) {
ERR("Health error occurred in %s", __func__);
}
DBG("Health check thread dying");
unlink(health_unix_sock_path);
if (sock >= 0) {
ret = close(sock);
if (ret) {
PERROR("close");
}
}
lttng_poll_clean(&events);
rcu_unregister_thread();
return NULL;
}
void *thread_ht_cleanup(void *data)
{
int ret, i, pollfd, err = -1;
ssize_t size_ret;
uint32_t revents, nb_fd;
struct lttng_poll_event events;
DBG("[ht-thread] startup.");
rcu_register_thread();
rcu_thread_online();
health_register(health_sessiond, HEALTH_SESSIOND_TYPE_HT_CLEANUP);
health_code_update();
ret = sessiond_set_thread_pollset(&events, 2);
if (ret < 0) {
goto error_poll_create;
}
/* Add pipe to the pollset. */
ret = lttng_poll_add(&events, ht_cleanup_pipe[0], LPOLLIN | LPOLLERR);
if (ret < 0) {
goto error;
}
health_code_update();
while (1) {
DBG3("[ht-thread] Polling on %d fds.",
LTTNG_POLL_GETNB(&events));
/* Inifinite blocking call, waiting for transmission */
restart:
health_poll_entry();
ret = lttng_poll_wait(&events, -1);
health_poll_exit();
if (ret < 0) {
/*
* Restart interrupted system call.
*/
if (errno == EINTR) {
goto restart;
}
goto error;
}
nb_fd = ret;
for (i = 0; i < nb_fd; i++) {
struct lttng_ht *ht;
health_code_update();
/* Fetch once the poll data */
revents = LTTNG_POLL_GETEV(&events, i);
pollfd = LTTNG_POLL_GETFD(&events, i);
/* Thread quit pipe has been closed. Killing thread. */
ret = sessiond_check_thread_quit_pipe(pollfd, revents);
if (ret) {
err = 0;
goto exit;
}
assert(pollfd == ht_cleanup_pipe[0]);
if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
ERR("ht cleanup pipe error");
goto error;
} else if (!(revents & LPOLLIN)) {
/* No POLLIN and not a catched error, stop the thread. */
ERR("ht cleanup failed. revent: %u", revents);
goto error;
}
/* Get socket from dispatch thread. */
size_ret = lttng_read(ht_cleanup_pipe[0], &ht,
sizeof(ht));
if (size_ret < sizeof(ht)) {
PERROR("ht cleanup notify pipe");
goto error;
}
health_code_update();
/*
* The whole point of this thread is to call
* lttng_ht_destroy from a context that is NOT:
* 1) a read-side RCU lock,
* 2) a call_rcu thread.
*/
lttng_ht_destroy(ht);
health_code_update();
}
}
exit:
error:
lttng_poll_clean(&events);
error_poll_create:
utils_close_pipe(ht_cleanup_pipe);
ht_cleanup_pipe[0] = ht_cleanup_pipe[1] = -1;
DBG("[ust-thread] cleanup complete.");
if (err) {
health_error();
ERR("Health error occurred in %s", __func__);
}
health_unregister(health_sessiond);
rcu_thread_offline();
rcu_unregister_thread();
return NULL;
}
/*
* Thread managing health check socket.
*/
static void *thread_manage_health(void *data)
{
const bool is_root = (getuid() == 0);
int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
uint32_t revents, nb_fd;
struct lttng_poll_event events;
struct health_comm_msg msg;
struct health_comm_reply reply;
/* Thread-specific quit pipe. */
struct thread_notifiers *notifiers = data;
const int quit_pipe_read_fd = lttng_pipe_get_readfd(
notifiers->quit_pipe);
DBG("[thread] Manage health check started");
rcu_register_thread();
/*
* Created with a size of two for:
* - client socket
* - thread quit pipe
*/
ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
if (ret < 0) {
goto error;
}
/* Create unix socket */
sock = lttcomm_create_unix_sock(config.health_unix_sock_path.value);
if (sock < 0) {
ERR("Unable to create health check Unix socket");
goto error;
}
if (is_root) {
/* lttng health client socket path permissions */
gid_t gid;
ret = utils_get_group_id(config.tracing_group_name.value, true, &gid);
if (ret) {
/* Default to root group. */
gid = 0;
}
ret = chown(config.health_unix_sock_path.value, 0, gid);
if (ret < 0) {
ERR("Unable to set group on %s", config.health_unix_sock_path.value);
PERROR("chown");
goto error;
}
ret = chmod(config.health_unix_sock_path.value,
S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
if (ret < 0) {
ERR("Unable to set permissions on %s", config.health_unix_sock_path.value);
PERROR("chmod");
goto error;
}
}
/*
* Set the CLOEXEC flag. Return code is useless because either way, the
* show must go on.
*/
(void) utils_set_fd_cloexec(sock);
ret = lttcomm_listen_unix_sock(sock);
if (ret < 0) {
goto error;
}
ret = lttng_poll_add(&events, quit_pipe_read_fd, LPOLLIN | LPOLLERR);
if (ret < 0) {
goto error;
}
/* Add the application registration socket */
ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
if (ret < 0) {
goto error;
}
mark_thread_as_ready(notifiers);
while (1) {
DBG("Health check ready");
/* Infinite blocking call, waiting for transmission */
restart:
ret = lttng_poll_wait(&events, -1);
if (ret < 0) {
/*
* Restart interrupted system call.
*/
if (errno == EINTR) {
goto restart;
}
goto error;
}
nb_fd = ret;
for (i = 0; i < nb_fd; i++) {
/* Fetch once the poll data */
revents = LTTNG_POLL_GETEV(&events, i);
pollfd = LTTNG_POLL_GETFD(&events, i);
/* Event on the registration socket */
if (pollfd == sock) {
if (revents & LPOLLIN) {
continue;
} else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
ERR("Health socket poll error");
goto error;
} else {
ERR("Unexpected poll events %u for sock %d", revents, pollfd);
goto error;
}
} else {
/* Event on the thread's quit pipe. */
err = 0;
goto exit;
}
}
new_sock = lttcomm_accept_unix_sock(sock);
if (new_sock < 0) {
goto error;
}
/*
* Set the CLOEXEC flag. Return code is useless because either way, the
* show must go on.
*/
(void) utils_set_fd_cloexec(new_sock);
DBG("Receiving data from client for health...");
ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
if (ret <= 0) {
DBG("Nothing recv() from client... continuing");
ret = close(new_sock);
if (ret) {
PERROR("close");
}
continue;
}
rcu_thread_online();
memset(&reply, 0, sizeof(reply));
for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
/*
* health_check_state returns 0 if health is
* bad.
*/
if (!health_check_state(health_sessiond, i)) {
reply.ret_code |= 1ULL << i;
}
}
DBG2("Health check return value %" PRIx64, reply.ret_code);
ret = lttcomm_send_unix_sock(new_sock, (void *) &reply,
sizeof(reply));
if (ret < 0) {
ERR("Failed to send health data back to client");
}
/* End of transmission */
ret = close(new_sock);
if (ret) {
PERROR("close");
}
}
exit:
error:
if (err) {
ERR("Health error occurred in %s", __func__);
}
DBG("Health check thread dying");
unlink(config.health_unix_sock_path.value);
if (sock >= 0) {
ret = close(sock);
if (ret) {
PERROR("close");
}
}
lttng_poll_clean(&events);
rcu_unregister_thread();
return NULL;
}
/*
* This thread manage application notify communication.
*/
void *ust_thread_manage_notify(void *data)
{
int i, ret, pollfd, err = -1;
ssize_t size_ret;
uint32_t revents, nb_fd;
struct lttng_poll_event events;
DBG("[ust-thread] Manage application notify command");
rcu_register_thread();
rcu_thread_online();
health_register(health_sessiond,
HEALTH_SESSIOND_TYPE_APP_MANAGE_NOTIFY);
if (testpoint(sessiond_thread_app_manage_notify)) {
goto error_testpoint;
}
health_code_update();
ret = sessiond_set_thread_pollset(&events, 2);
if (ret < 0) {
goto error_poll_create;
}
/* Add notify pipe to the pollset. */
ret = lttng_poll_add(&events, apps_cmd_notify_pipe[0], LPOLLIN | LPOLLERR);
if (ret < 0) {
goto error;
}
health_code_update();
while (1) {
DBG3("[ust-thread] Manage notify polling on %d fds",
LTTNG_POLL_GETNB(&events));
/* Inifinite blocking call, waiting for transmission */
restart:
health_poll_entry();
ret = lttng_poll_wait(&events, -1);
health_poll_exit();
if (ret < 0) {
/*
* Restart interrupted system call.
*/
if (errno == EINTR) {
goto restart;
}
goto error;
}
nb_fd = ret;
for (i = 0; i < nb_fd; i++) {
health_code_update();
/* Fetch once the poll data */
revents = LTTNG_POLL_GETEV(&events, i);
pollfd = LTTNG_POLL_GETFD(&events, i);
/* Thread quit pipe has been closed. Killing thread. */
ret = sessiond_check_thread_quit_pipe(pollfd, revents);
if (ret) {
err = 0;
goto exit;
}
/* Inspect the apps cmd pipe */
if (pollfd == apps_cmd_notify_pipe[0]) {
int sock;
if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
ERR("Apps notify command pipe error");
goto error;
} else if (!(revents & LPOLLIN)) {
/* No POLLIN and not a catched error, stop the thread. */
ERR("Notify command pipe failed. revent: %u", revents);
goto error;
}
/* Get socket from dispatch thread. */
size_ret = lttng_read(apps_cmd_notify_pipe[0],
&sock, sizeof(sock));
if (size_ret < sizeof(sock)) {
PERROR("read apps notify pipe");
goto error;
}
health_code_update();
ret = lttng_poll_add(&events, sock,
LPOLLIN | LPOLLERR | LPOLLHUP | LPOLLRDHUP);
if (ret < 0) {
/*
* It's possible we've reached the max poll fd allowed.
* Let's close the socket but continue normal execution.
*/
ret = close(sock);
if (ret) {
PERROR("close notify socket %d", sock);
}
lttng_fd_put(LTTNG_FD_APPS, 1);
continue;
}
DBG3("UST thread notify added sock %d to pollset", sock);
} else {
/*
* At this point, we know that a registered application
* triggered the event.
*/
if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
/* Removing from the poll set */
ret = lttng_poll_del(&events, pollfd);
if (ret < 0) {
goto error;
}
/* The socket is closed after a grace period here. */
ust_app_notify_sock_unregister(pollfd);
} else if (revents & (LPOLLIN | LPOLLPRI)) {
ret = ust_app_recv_notify(pollfd);
if (ret < 0) {
/*
* If the notification failed either the application is
* dead or an internal error happened. In both cases,
* we can only continue here. If the application is
* dead, an unregistration will follow or else the
* application will notice that we are not responding
* on that socket and will close it.
*/
continue;
}
} else {
ERR("Unknown poll events %u for sock %d", revents, pollfd);
continue;
}
health_code_update();
}
}
}
exit:
error:
lttng_poll_clean(&events);
error_poll_create:
error_testpoint:
utils_close_pipe(apps_cmd_notify_pipe);
apps_cmd_notify_pipe[0] = apps_cmd_notify_pipe[1] = -1;
DBG("Application notify communication apps thread cleanup complete");
if (err) {
health_error();
ERR("Health error occurred in %s", __func__);
}
health_unregister(health_sessiond);
rcu_thread_offline();
rcu_unregister_thread();
return NULL;
}
