#include <common/compat/socket.h>
#include <common/defaults.h>
#include <common/kernel-consumer/kernel-consumer.h>
-#include <common/ust-consumer/ust-consumer.h>
#include <common/futex.h>
+#include <common/relayd/relayd.h>
+#include <common/utils.h>
#include "lttng-sessiond.h"
#include "channel.h"
+#include "consumer.h"
#include "context.h"
#include "event.h"
#include "kernel.h"
+#include "kernel-consumer.h"
#include "modprobe.h"
#include "shm.h"
#include "ust-ctl.h"
+#include "ust-consumer.h"
#include "utils.h"
#include "fd-limit.h"
+#include "filter.h"
+#include "health.h"
#define CONSUMERD_FILE "lttng-consumerd"
-struct consumer_data {
- enum lttng_consumer_type type;
-
- pthread_t thread; /* Worker thread interacting with the consumer */
- sem_t sem;
-
- /* Mutex to control consumerd pid assignation */
- pthread_mutex_t pid_mutex;
- pid_t pid;
-
- int err_sock;
- int cmd_sock;
-
- /* consumer error and command Unix socket path */
- char err_unix_sock_path[PATH_MAX];
- char cmd_unix_sock_path[PATH_MAX];
-};
-
/* Const values */
const char default_home_dir[] = DEFAULT_HOME_DIR;
const char default_tracing_group[] = DEFAULT_TRACING_GROUP;
.cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
.err_sock = -1,
.cmd_sock = -1,
+ .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
+ .lock = PTHREAD_MUTEX_INITIALIZER,
};
static struct consumer_data ustconsumer64_data = {
.type = LTTNG_CONSUMER64_UST,
.cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
.err_sock = -1,
.cmd_sock = -1,
+ .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
+ .lock = PTHREAD_MUTEX_INITIALIZER,
};
static struct consumer_data ustconsumer32_data = {
.type = LTTNG_CONSUMER32_UST,
.cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
.err_sock = -1,
.cmd_sock = -1,
+ .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
+ .lock = PTHREAD_MUTEX_INITIALIZER,
};
+/* Shared between threads */
static int dispatch_thread_exit;
/* Global application Unix socket path */
static char client_unix_sock_path[PATH_MAX];
/* global wait shm path for UST */
static char wait_shm_path[PATH_MAX];
+/* Global health check unix path */
+static char health_unix_sock_path[PATH_MAX];
/* Sockets and FDs */
static int client_sock = -1;
static pthread_t client_thread;
static pthread_t kernel_thread;
static pthread_t dispatch_thread;
-
+static pthread_t health_thread;
/*
* UST registration command queue. This queue is tied with a futex and uses a N
static enum consumerd_state ust_consumerd_state;
static enum consumerd_state kernel_consumerd_state;
+/*
+ * Used to keep a unique index for each relayd socket created where this value
+ * is associated with streams on the consumer so it can match the right relayd
+ * to send to.
+ *
+ * This value should be incremented atomically for safety purposes and future
+ * possible concurrent access.
+ */
+static unsigned int relayd_net_seq_idx;
+
+/* Used for the health monitoring of the session daemon. See health.h */
+struct health_state health_thread_cmd;
+struct health_state health_thread_app_manage;
+struct health_state health_thread_app_reg;
+struct health_state health_thread_kernel;
+
static
void setup_consumerd_path(void)
{
*/
static void teardown_kernel_session(struct ltt_session *session)
{
+ int ret;
+ struct lttng_ht_iter iter;
+ struct ltt_kernel_session *ksess;
+ struct consumer_socket *socket;
+
if (!session->kernel_session) {
DBG3("No kernel session when tearing down session");
return;
}
+ ksess = session->kernel_session;
+
DBG("Tearing down kernel session");
+ /*
+ * Destroy relayd associated with the session consumer. This action is
+ * valid since in order to destroy a session we must acquire the session
+ * lock. This means that there CAN NOT be stream(s) being sent to a
+ * consumer since this action also requires the session lock at any time.
+ *
+ * At this point, we are sure that not streams data will be lost after this
+ * command is issued.
+ */
+ if (ksess->consumer && ksess->consumer->type == CONSUMER_DST_NET) {
+ cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter, socket,
+ node.node) {
+ ret = consumer_send_destroy_relayd(socket, ksess->consumer);
+ if (ret < 0) {
+ ERR("Unable to send destroy relayd command to consumer");
+ /* Continue since we MUST delete everything at this point. */
+ }
+ }
+ }
+
/*
* If a custom kernel consumer was registered, close the socket before
* tearing down the complete kernel session structure
*/
- if (kconsumer_data.cmd_sock >= 0 &&
- session->kernel_session->consumer_fd != kconsumer_data.cmd_sock) {
- lttcomm_close_unix_sock(session->kernel_session->consumer_fd);
+ cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter, socket,
+ node.node) {
+ if (socket->fd != kconsumer_data.cmd_sock) {
+ rcu_read_lock();
+ consumer_del_socket(socket, ksess->consumer);
+ lttcomm_close_unix_sock(socket->fd);
+ consumer_destroy_socket(socket);
+ rcu_read_unlock();
+ }
}
- trace_kernel_destroy_session(session->kernel_session);
+ trace_kernel_destroy_session(ksess);
}
/*
static void teardown_ust_session(struct ltt_session *session)
{
int ret;
+ struct lttng_ht_iter iter;
+ struct ltt_ust_session *usess;
+ struct consumer_socket *socket;
if (!session->ust_session) {
DBG3("No UST session when tearing down session");
return;
}
+ usess = session->ust_session;
DBG("Tearing down UST session(s)");
- ret = ust_app_destroy_trace_all(session->ust_session);
+ /*
+ * Destroy relayd associated with the session consumer. This action is
+ * valid since in order to destroy a session we must acquire the session
+ * lock. This means that there CAN NOT be stream(s) being sent to a
+ * consumer since this action also requires the session lock at any time.
+ *
+ * At this point, we are sure that not streams data will be lost after this
+ * command is issued.
+ */
+ if (usess->consumer && usess->consumer->type == CONSUMER_DST_NET) {
+ cds_lfht_for_each_entry(usess->consumer->socks->ht, &iter.iter, socket,
+ node.node) {
+ ret = consumer_send_destroy_relayd(socket, usess->consumer);
+ if (ret < 0) {
+ ERR("Unable to send destroy relayd command to consumer");
+ /* Continue since we MUST delete everything at this point. */
+ }
+ }
+ }
+
+ ret = ust_app_destroy_trace_all(usess);
if (ret) {
ERR("Error in ust_app_destroy_trace_all");
}
- trace_ust_destroy_session(session->ust_session);
+ trace_ust_destroy_session(usess);
}
/*
}
/* Dispatch thread */
- dispatch_thread_exit = 1;
+ CMM_STORE_SHARED(dispatch_thread_exit, 1);
futex_nto1_wake(&ust_cmd_queue.futex);
}
*/
static void cleanup(void)
{
- int ret, i;
+ int ret;
char *cmd;
struct ltt_session *sess, *stmp;
DBG("Closing all UST sockets");
ust_app_clean_list();
- pthread_mutex_destroy(&kconsumer_data.pid_mutex);
-
if (is_root && !opt_no_kernel) {
DBG2("Closing kernel fd");
if (kernel_tracer_fd >= 0) {
DBG("Unloading kernel modules");
modprobe_remove_lttng_all();
}
-
- /*
- * Closing all pipes used for communication between threads.
- */
- for (i = 0; i < 2; i++) {
- if (kernel_poll_pipe[i] >= 0) {
- ret = close(kernel_poll_pipe[i]);
- if (ret) {
- PERROR("close");
- }
- }
- }
- for (i = 0; i < 2; i++) {
- if (thread_quit_pipe[i] >= 0) {
- ret = close(thread_quit_pipe[i]);
- if (ret) {
- PERROR("close");
- }
- }
- }
- for (i = 0; i < 2; i++) {
- if (apps_cmd_pipe[i] >= 0) {
- ret = close(apps_cmd_pipe[i]);
- if (ret) {
- PERROR("close");
- }
- }
- }
+ utils_close_pipe(kernel_poll_pipe);
+ utils_close_pipe(thread_quit_pipe);
+ utils_close_pipe(apps_cmd_pipe);
/* <fun> */
DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
}
}
-/*
- * Send all stream fds of kernel channel to the consumer.
- */
-static int send_kconsumer_channel_streams(struct consumer_data *consumer_data,
- int sock, struct ltt_kernel_channel *channel,
- uid_t uid, gid_t gid)
-{
- int ret;
- struct ltt_kernel_stream *stream;
- struct lttcomm_consumer_msg lkm;
-
- DBG("Sending streams of channel %s to kernel consumer",
- channel->channel->name);
-
- /* Send channel */
- lkm.cmd_type = LTTNG_CONSUMER_ADD_CHANNEL;
- lkm.u.channel.channel_key = channel->fd;
- lkm.u.channel.max_sb_size = channel->channel->attr.subbuf_size;
- lkm.u.channel.mmap_len = 0; /* for kernel */
- DBG("Sending channel %d to consumer", lkm.u.channel.channel_key);
- ret = lttcomm_send_unix_sock(sock, &lkm, sizeof(lkm));
- if (ret < 0) {
- PERROR("send consumer channel");
- goto error;
- }
-
- /* Send streams */
- cds_list_for_each_entry(stream, &channel->stream_list.head, list) {
- if (!stream->fd) {
- continue;
- }
- lkm.cmd_type = LTTNG_CONSUMER_ADD_STREAM;
- lkm.u.stream.channel_key = channel->fd;
- lkm.u.stream.stream_key = stream->fd;
- lkm.u.stream.state = stream->state;
- lkm.u.stream.output = channel->channel->attr.output;
- lkm.u.stream.mmap_len = 0; /* for kernel */
- lkm.u.stream.uid = uid;
- lkm.u.stream.gid = gid;
- strncpy(lkm.u.stream.path_name, stream->pathname, PATH_MAX - 1);
- lkm.u.stream.path_name[PATH_MAX - 1] = '\0';
- DBG("Sending stream %d to consumer", lkm.u.stream.stream_key);
- ret = lttcomm_send_unix_sock(sock, &lkm, sizeof(lkm));
- if (ret < 0) {
- PERROR("send consumer stream");
- goto error;
- }
- ret = lttcomm_send_fds_unix_sock(sock, &stream->fd, 1);
- if (ret < 0) {
- PERROR("send consumer stream ancillary data");
- goto error;
- }
- }
-
- DBG("consumer channel streams sent");
-
- return 0;
-
-error:
- return ret;
-}
-
-/*
- * Send all stream fds of the kernel session to the consumer.
- */
-static int send_kconsumer_session_streams(struct consumer_data *consumer_data,
- struct ltt_kernel_session *session)
-{
- int ret;
- struct ltt_kernel_channel *chan;
- struct lttcomm_consumer_msg lkm;
- int sock = session->consumer_fd;
-
- DBG("Sending metadata stream fd");
-
- /* Extra protection. It's NOT supposed to be set to -1 at this point */
- if (session->consumer_fd < 0) {
- session->consumer_fd = consumer_data->cmd_sock;
- }
-
- if (session->metadata_stream_fd >= 0) {
- /* Send metadata channel fd */
- lkm.cmd_type = LTTNG_CONSUMER_ADD_CHANNEL;
- lkm.u.channel.channel_key = session->metadata->fd;
- lkm.u.channel.max_sb_size = session->metadata->conf->attr.subbuf_size;
- lkm.u.channel.mmap_len = 0; /* for kernel */
- DBG("Sending metadata channel %d to consumer", lkm.u.channel.channel_key);
- ret = lttcomm_send_unix_sock(sock, &lkm, sizeof(lkm));
- if (ret < 0) {
- PERROR("send consumer channel");
- goto error;
- }
-
- /* Send metadata stream fd */
- lkm.cmd_type = LTTNG_CONSUMER_ADD_STREAM;
- lkm.u.stream.channel_key = session->metadata->fd;
- lkm.u.stream.stream_key = session->metadata_stream_fd;
- lkm.u.stream.state = LTTNG_CONSUMER_ACTIVE_STREAM;
- lkm.u.stream.output = DEFAULT_KERNEL_CHANNEL_OUTPUT;
- lkm.u.stream.mmap_len = 0; /* for kernel */
- lkm.u.stream.uid = session->uid;
- lkm.u.stream.gid = session->gid;
- strncpy(lkm.u.stream.path_name, session->metadata->pathname, PATH_MAX - 1);
- lkm.u.stream.path_name[PATH_MAX - 1] = '\0';
- DBG("Sending metadata stream %d to consumer", lkm.u.stream.stream_key);
- ret = lttcomm_send_unix_sock(sock, &lkm, sizeof(lkm));
- if (ret < 0) {
- PERROR("send consumer stream");
- goto error;
- }
- ret = lttcomm_send_fds_unix_sock(sock, &session->metadata_stream_fd, 1);
- if (ret < 0) {
- PERROR("send consumer stream");
- goto error;
- }
- }
-
- cds_list_for_each_entry(chan, &session->channel_list.head, list) {
- ret = send_kconsumer_channel_streams(consumer_data, sock, chan,
- session->uid, session->gid);
- if (ret < 0) {
- goto error;
- }
- }
-
- DBG("consumer fds (metadata and channel streams) sent");
-
- return 0;
-
-error:
- return ret;
-}
-
/*
* Notify UST applications using the shm mmap futex.
*/
{
int ret = 0;
struct ltt_session *session;
+ struct ltt_kernel_session *ksess;
struct ltt_kernel_channel *channel;
DBG("Updating kernel streams for channel fd %d", fd);
session_unlock(session);
continue;
}
+ ksess = session->kernel_session;
- /* This is not suppose to be -1 but this is an extra security check */
- if (session->kernel_session->consumer_fd < 0) {
- session->kernel_session->consumer_fd = consumer_data->cmd_sock;
- }
-
- cds_list_for_each_entry(channel,
- &session->kernel_session->channel_list.head, list) {
+ cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
if (channel->fd == fd) {
DBG("Channel found, updating kernel streams");
ret = kernel_open_channel_stream(channel);
* that tracing is started so it is safe to send our updated
* stream fds.
*/
- if (session->kernel_session->consumer_fds_sent == 1) {
- ret = send_kconsumer_channel_streams(consumer_data,
- session->kernel_session->consumer_fd, channel,
- session->uid, session->gid);
- if (ret < 0) {
- goto error;
+ if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
+ struct lttng_ht_iter iter;
+ struct consumer_socket *socket;
+
+
+ cds_lfht_for_each_entry(ksess->consumer->socks->ht,
+ &iter.iter, socket, node.node) {
+ /* Code flow error */
+ assert(socket->fd >= 0);
+
+ pthread_mutex_lock(socket->lock);
+ ret = kernel_consumer_send_channel_stream(socket->fd,
+ channel, ksess);
+ pthread_mutex_unlock(socket->lock);
+ if (ret < 0) {
+ goto error;
+ }
}
}
goto error;
*/
static void *thread_manage_kernel(void *data)
{
- int ret, i, pollfd, update_poll_flag = 1;
+ int ret, i, pollfd, update_poll_flag = 1, err = -1;
uint32_t revents, nb_fd;
char tmp;
struct lttng_poll_event events;
DBG("Thread manage kernel started");
+ health_code_update(&health_thread_kernel);
+
ret = create_thread_poll_set(&events, 2);
if (ret < 0) {
goto error_poll_create;
}
while (1) {
+ health_code_update(&health_thread_kernel);
+
if (update_poll_flag == 1) {
/*
* Reset number of fd in the poll set. Always 2 since there is the thread
/* Poll infinite value of time */
restart:
+ health_poll_update(&health_thread_kernel);
ret = lttng_poll_wait(&events, -1);
+ health_poll_update(&health_thread_kernel);
if (ret < 0) {
/*
* Restart interrupted system call.
revents = LTTNG_POLL_GETEV(&events, i);
pollfd = LTTNG_POLL_GETFD(&events, i);
+ health_code_update(&health_thread_kernel);
+
/* Thread quit pipe has been closed. Killing thread. */
ret = check_thread_quit_pipe(pollfd, revents);
if (ret) {
- goto error;
+ err = 0;
+ goto exit;
}
/* Check for data on kernel pipe */
}
}
+exit:
error:
lttng_poll_clean(&events);
error_poll_create:
+ if (err) {
+ health_error(&health_thread_kernel);
+ ERR("Health error occurred in %s", __func__);
+ }
+ health_exit(&health_thread_kernel);
DBG("Kernel thread dying");
return NULL;
}
*/
static void *thread_manage_consumer(void *data)
{
- int sock = -1, i, ret, pollfd;
+ int sock = -1, i, ret, pollfd, err = -1;
uint32_t revents, nb_fd;
enum lttcomm_return_code code;
struct lttng_poll_event events;
DBG("[thread] Manage consumer started");
+ health_code_update(&consumer_data->health);
+
ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
if (ret < 0) {
goto error_listen;
nb_fd = LTTNG_POLL_GETNB(&events);
+ health_code_update(&consumer_data->health);
+
/* Inifinite blocking call, waiting for transmission */
restart:
+ health_poll_update(&consumer_data->health);
ret = lttng_poll_wait(&events, -1);
+ health_poll_update(&consumer_data->health);
if (ret < 0) {
/*
* Restart interrupted system call.
revents = LTTNG_POLL_GETEV(&events, i);
pollfd = LTTNG_POLL_GETFD(&events, i);
+ health_code_update(&consumer_data->health);
+
/* Thread quit pipe has been closed. Killing thread. */
ret = check_thread_quit_pipe(pollfd, revents);
if (ret) {
- goto error;
+ err = 0;
+ goto exit;
}
/* Event on the registration socket */
goto error;
}
+ health_code_update(&consumer_data->health);
+
DBG2("Receiving code from consumer err_sock");
/* Getting status code from kconsumerd */
goto error;
}
+ health_code_update(&consumer_data->health);
+
if (code == CONSUMERD_COMMAND_SOCK_READY) {
consumer_data->cmd_sock =
lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
goto error;
}
+ health_code_update(&consumer_data->health);
+
/* Update number of fd */
nb_fd = LTTNG_POLL_GETNB(&events);
/* Inifinite blocking call, waiting for transmission */
restart_poll:
+ health_poll_update(&consumer_data->health);
ret = lttng_poll_wait(&events, -1);
+ health_poll_update(&consumer_data->health);
if (ret < 0) {
/*
* Restart interrupted system call.
revents = LTTNG_POLL_GETEV(&events, i);
pollfd = LTTNG_POLL_GETFD(&events, i);
+ health_code_update(&consumer_data->health);
+
/* Thread quit pipe has been closed. Killing thread. */
ret = check_thread_quit_pipe(pollfd, revents);
if (ret) {
- goto error;
+ err = 0;
+ goto exit;
}
/* Event on the kconsumerd socket */
}
}
+ health_code_update(&consumer_data->health);
+
/* Wait for any kconsumerd error */
ret = lttcomm_recv_unix_sock(sock, &code,
sizeof(enum lttcomm_return_code));
ERR("consumer return code : %s", lttcomm_get_readable_code(-code));
+exit:
error:
/* Immediately set the consumerd state to stopped */
if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
lttng_poll_clean(&events);
error_poll:
error_listen:
+ if (err) {
+ health_error(&consumer_data->health);
+ ERR("Health error occurred in %s", __func__);
+ }
+ health_exit(&consumer_data->health);
DBG("consumer thread cleanup completed");
return NULL;
*/
static void *thread_manage_apps(void *data)
{
- int i, ret, pollfd;
+ int i, ret, pollfd, err = -1;
uint32_t revents, nb_fd;
struct ust_command ust_cmd;
struct lttng_poll_event events;
rcu_register_thread();
rcu_thread_online();
+ health_code_update(&health_thread_app_manage);
+
ret = create_thread_poll_set(&events, 2);
if (ret < 0) {
goto error_poll_create;
goto error;
}
+ health_code_update(&health_thread_app_manage);
+
while (1) {
/* Zeroed the events structure */
lttng_poll_reset(&events);
/* Inifinite blocking call, waiting for transmission */
restart:
+ health_poll_update(&health_thread_app_manage);
ret = lttng_poll_wait(&events, -1);
+ health_poll_update(&health_thread_app_manage);
if (ret < 0) {
/*
* Restart interrupted system call.
revents = LTTNG_POLL_GETEV(&events, i);
pollfd = LTTNG_POLL_GETFD(&events, i);
+ health_code_update(&health_thread_app_manage);
+
/* Thread quit pipe has been closed. Killing thread. */
ret = check_thread_quit_pipe(pollfd, revents);
if (ret) {
- goto error;
+ err = 0;
+ goto exit;
}
/* Inspect the apps cmd pipe */
goto error;
}
+ health_code_update(&health_thread_app_manage);
+
/* Register applicaton to the session daemon */
ret = ust_app_register(&ust_cmd.reg_msg,
ust_cmd.sock);
break;
}
+ health_code_update(&health_thread_app_manage);
+
/*
* Validate UST version compatibility.
*/
update_ust_app(ust_cmd.sock);
}
+ health_code_update(&health_thread_app_manage);
+
ret = ust_app_register_done(ust_cmd.sock);
if (ret < 0) {
/*
ust_cmd.sock);
}
+ health_code_update(&health_thread_app_manage);
+
break;
}
} else {
break;
}
}
+
+ health_code_update(&health_thread_app_manage);
}
}
+exit:
error:
lttng_poll_clean(&events);
error_poll_create:
+ if (err) {
+ health_error(&health_thread_app_manage);
+ ERR("Health error occurred in %s", __func__);
+ }
+ health_exit(&health_thread_app_manage);
DBG("Application communication apps thread cleanup complete");
rcu_thread_offline();
rcu_unregister_thread();
DBG("[thread] Dispatch UST command started");
- while (!dispatch_thread_exit) {
+ while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
/* Atomically prepare the queue futex */
futex_nto1_prepare(&ust_cmd_queue.futex);
*/
static void *thread_registration_apps(void *data)
{
- int sock = -1, i, ret, pollfd;
+ int sock = -1, i, ret, pollfd, err = -1;
uint32_t revents, nb_fd;
struct lttng_poll_event events;
/*
/* Inifinite blocking call, waiting for transmission */
restart:
+ health_poll_update(&health_thread_app_reg);
ret = lttng_poll_wait(&events, -1);
+ health_poll_update(&health_thread_app_reg);
if (ret < 0) {
/*
* Restart interrupted system call.
}
for (i = 0; i < nb_fd; i++) {
+ health_code_update(&health_thread_app_reg);
+
/* 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 = check_thread_quit_pipe(pollfd, revents);
if (ret) {
- goto error;
+ err = 0;
+ goto exit;
}
/* Event on the registration socket */
sock = -1;
continue;
}
+ health_code_update(&health_thread_app_reg);
ret = lttcomm_recv_unix_sock(sock, &ust_cmd->reg_msg,
sizeof(struct ust_register_msg));
if (ret < 0 || ret < sizeof(struct ust_register_msg)) {
sock = -1;
continue;
}
+ health_code_update(&health_thread_app_reg);
ust_cmd->sock = sock;
sock = -1;
}
}
+exit:
error:
+ if (err) {
+ health_error(&health_thread_app_reg);
+ ERR("Health error occurred in %s", __func__);
+ }
+ health_exit(&health_thread_app_reg);
+
/* Notify that the registration thread is gone */
notify_ust_apps(0);
void *status;
int ret;
- if (consumer_data->pid != 0) {
+ /* Consumer pid must be a real one. */
+ if (consumer_data->pid > 0) {
ret = kill(consumer_data->pid, SIGTERM);
if (ret) {
ERR("Error killing consumer daemon");
return ret;
}
+/*
+ * Compute health status of each consumer. If one of them is zero (bad
+ * state), we return 0.
+ */
+static int check_consumer_health(void)
+{
+ int ret;
+
+ ret = health_check_state(&kconsumer_data.health) &&
+ health_check_state(&ustconsumer32_data.health) &&
+ health_check_state(&ustconsumer64_data.health);
+
+ DBG3("Health consumer check %d", ret);
+
+ return ret;
+}
+
/*
* Check version of the lttng-modules.
*/
static int init_kernel_tracing(struct ltt_kernel_session *session)
{
int ret = 0;
+ struct lttng_ht_iter iter;
+ struct consumer_socket *socket;
- if (session->consumer_fds_sent == 0) {
- /*
- * Assign default kernel consumer socket if no consumer assigned to the
- * kernel session. At this point, it's NOT supposed to be -1 but this is
- * an extra security check.
- */
- if (session->consumer_fd < 0) {
- session->consumer_fd = kconsumer_data.cmd_sock;
- }
+ assert(session);
- ret = send_kconsumer_session_streams(&kconsumer_data, session);
- if (ret < 0) {
- ret = LTTCOMM_KERN_CONSUMER_FAIL;
- goto error;
- }
+ if (session->consumer_fds_sent == 0 && session->consumer != NULL) {
+ cds_lfht_for_each_entry(session->consumer->socks->ht, &iter.iter,
+ socket, node.node) {
+ /* Code flow error */
+ assert(socket->fd >= 0);
- session->consumer_fds_sent = 1;
+ pthread_mutex_lock(socket->lock);
+ ret = kernel_consumer_send_session(socket->fd, session);
+ pthread_mutex_unlock(socket->lock);
+ if (ret < 0) {
+ ret = LTTCOMM_KERN_CONSUMER_FAIL;
+ goto error;
+ }
+ }
}
error:
}
/*
- * Create an UST session and add it to the session ust list.
+ * Create a socket to the relayd using the URI.
+ *
+ * On success, the relayd_sock pointer is set to the created socket.
+ * Else, it is untouched and an lttcomm error code is returned.
*/
-static int create_ust_session(struct ltt_session *session,
- struct lttng_domain *domain)
+static int create_connect_relayd(struct consumer_output *output,
+ const char *session_name, struct lttng_uri *uri,
+ struct lttcomm_sock **relayd_sock)
{
- struct ltt_ust_session *lus = NULL;
int ret;
+ struct lttcomm_sock *sock;
- switch (domain->type) {
- case LTTNG_DOMAIN_UST:
- break;
- default:
- ret = LTTCOMM_UNKNOWN_DOMAIN;
+ /* Create socket object from URI */
+ sock = lttcomm_alloc_sock_from_uri(uri);
+ if (sock == NULL) {
+ ret = LTTCOMM_FATAL;
goto error;
}
- DBG("Creating UST session");
-
- lus = trace_ust_create_session(session->path, session->id, domain);
- if (lus == NULL) {
- ret = LTTCOMM_UST_SESS_FAIL;
+ ret = lttcomm_create_sock(sock);
+ if (ret < 0) {
+ ret = LTTCOMM_FATAL;
goto error;
}
- ret = run_as_mkdir_recursive(lus->pathname, S_IRWXU | S_IRWXG,
- session->uid, session->gid);
+ /* Connect to relayd so we can proceed with a session creation. */
+ ret = relayd_connect(sock);
if (ret < 0) {
- if (ret != -EEXIST) {
- ERR("Trace directory creation error");
- ret = LTTCOMM_UST_SESS_FAIL;
- goto error;
- }
+ ERR("Unable to reach lttng-relayd");
+ ret = LTTCOMM_RELAYD_SESSION_FAIL;
+ goto free_sock;
}
- /* The domain type dictate different actions on session creation */
- switch (domain->type) {
- case LTTNG_DOMAIN_UST:
- /* No ustctl for the global UST domain */
- break;
- default:
- ERR("Unknown UST domain on create session %d", domain->type);
- goto error;
+ /* Create socket for control stream. */
+ if (uri->stype == LTTNG_STREAM_CONTROL) {
+ DBG3("Creating relayd stream socket from URI");
+
+ /* Check relayd version */
+ ret = relayd_version_check(sock, LTTNG_UST_COMM_MAJOR, 0);
+ if (ret < 0) {
+ ret = LTTCOMM_RELAYD_VERSION_FAIL;
+ goto close_sock;
+ }
+ } else if (uri->stype == LTTNG_STREAM_DATA) {
+ DBG3("Creating relayd data socket from URI");
+ } else {
+ /* Command is not valid */
+ ERR("Relayd invalid stream type: %d", uri->stype);
+ ret = LTTCOMM_INVALID;
+ goto close_sock;
}
- lus->uid = session->uid;
- lus->gid = session->gid;
- session->ust_session = lus;
+
+ *relayd_sock = sock;
return LTTCOMM_OK;
+close_sock:
+ if (sock) {
+ (void) relayd_close(sock);
+ }
+free_sock:
+ if (sock) {
+ lttcomm_destroy_sock(sock);
+ }
error:
- free(lus);
return ret;
}
/*
- * Create a kernel tracer session then create the default channel.
+ * Connect to the relayd using URI and send the socket to the right consumer.
*/
-static int create_kernel_session(struct ltt_session *session)
+static int send_socket_relayd_consumer(int domain, struct ltt_session *session,
+ struct lttng_uri *relayd_uri, struct consumer_output *consumer,
+ int consumer_fd)
{
int ret;
+ struct lttcomm_sock *sock = NULL;
- DBG("Creating kernel session");
+ /* Set the network sequence index if not set. */
+ if (consumer->net_seq_index == -1) {
+ /*
+ * Increment net_seq_idx because we are about to transfer the
+ * new relayd socket to the consumer.
+ */
+ uatomic_inc(&relayd_net_seq_idx);
+ /* Assign unique key so the consumer can match streams */
+ consumer->net_seq_index = uatomic_read(&relayd_net_seq_idx);
+ }
- ret = kernel_create_session(session, kernel_tracer_fd);
- if (ret < 0) {
- ret = LTTCOMM_KERN_SESS_FAIL;
- goto error;
+ /* Connect to relayd and make version check if uri is the control. */
+ ret = create_connect_relayd(consumer, session->name, relayd_uri, &sock);
+ if (ret != LTTCOMM_OK) {
+ goto close_sock;
}
- /* Set kernel consumer socket fd */
- if (kconsumer_data.cmd_sock >= 0) {
- session->kernel_session->consumer_fd = kconsumer_data.cmd_sock;
+ /* If the control socket is connected, network session is ready */
+ if (relayd_uri->stype == LTTNG_STREAM_CONTROL) {
+ session->net_handle = 1;
}
- ret = run_as_mkdir_recursive(session->kernel_session->trace_path,
- S_IRWXU | S_IRWXG, session->uid, session->gid);
+ /* Send relayd socket to consumer. */
+ ret = consumer_send_relayd_socket(consumer_fd, sock,
+ consumer, relayd_uri->stype);
if (ret < 0) {
- if (ret != -EEXIST) {
- ERR("Trace directory creation error");
- goto error;
- }
+ ret = LTTCOMM_ENABLE_CONSUMER_FAIL;
+ goto close_sock;
+ }
+
+ ret = LTTCOMM_OK;
+
+ /*
+ * Close socket which was dup on the consumer side. The session daemon does
+ * NOT keep track of the relayd socket(s) once transfer to the consumer.
+ */
+
+close_sock:
+ if (sock) {
+ (void) relayd_close(sock);
+ lttcomm_destroy_sock(sock);
}
- session->kernel_session->uid = session->uid;
- session->kernel_session->gid = session->gid;
-error:
return ret;
}
/*
- * Check if the UID or GID match the session. Root user has access to all
- * sessions.
+ * Send both relayd sockets to a specific consumer and domain. This is a
+ * helper function to facilitate sending the information to the consumer for a
+ * session.
*/
-static int session_access_ok(struct ltt_session *session, uid_t uid, gid_t gid)
+static int send_sockets_relayd_consumer(int domain,
+ struct ltt_session *session, struct consumer_output *consumer, int fd)
{
- if (uid != session->uid && gid != session->gid && uid != 0) {
- return 0;
- } else {
+ int ret;
+
+ /* Sending control relayd socket. */
+ ret = send_socket_relayd_consumer(domain, session,
+ &consumer->dst.net.control, consumer, fd);
+ if (ret != LTTCOMM_OK) {
+ goto error;
+ }
+
+ /* Sending data relayd socket. */
+ ret = send_socket_relayd_consumer(domain, session,
+ &consumer->dst.net.data, consumer, fd);
+ if (ret != LTTCOMM_OK) {
+ goto error;
+ }
+
+error:
+ return ret;
+}
+
+/*
+ * Setup relayd connections for a tracing session. First creates the socket to
+ * the relayd and send them to the right domain consumer. Consumer type MUST be
+ * network.
+ */
+static int setup_relayd(struct ltt_session *session)
+{
+ int ret = LTTCOMM_OK;
+ struct ltt_ust_session *usess;
+ struct ltt_kernel_session *ksess;
+ struct consumer_socket *socket;
+ struct lttng_ht_iter iter;
+
+ assert(session);
+
+ usess = session->ust_session;
+ ksess = session->kernel_session;
+
+ DBG2("Setting relayd for session %s", session->name);
+
+ if (usess && usess->consumer->type == CONSUMER_DST_NET &&
+ usess->consumer->enabled) {
+ /* For each consumer socket, send relayd sockets */
+ cds_lfht_for_each_entry(usess->consumer->socks->ht, &iter.iter,
+ socket, node.node) {
+ /* Code flow error */
+ assert(socket->fd >= 0);
+
+ pthread_mutex_lock(socket->lock);
+ send_sockets_relayd_consumer(LTTNG_DOMAIN_UST, session,
+ usess->consumer, socket->fd);
+ pthread_mutex_unlock(socket->lock);
+ if (ret != LTTCOMM_OK) {
+ goto error;
+ }
+ }
+ } else if (ksess && ksess->consumer->type == CONSUMER_DST_NET &&
+ ksess->consumer->enabled) {
+ cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
+ socket, node.node) {
+ /* Code flow error */
+ assert(socket->fd >= 0);
+
+ pthread_mutex_lock(socket->lock);
+ send_sockets_relayd_consumer(LTTNG_DOMAIN_KERNEL, session,
+ ksess->consumer, socket->fd);
+ pthread_mutex_unlock(socket->lock);
+ if (ret != LTTCOMM_OK) {
+ goto error;
+ }
+ }
+ }
+
+error:
+ return ret;
+}
+
+/*
+ * Copy consumer output from the tracing session to the domain session. The
+ * function also applies the right modification on a per domain basis for the
+ * trace files destination directory.
+ */
+static int copy_session_consumer(int domain, struct ltt_session *session)
+{
+ int ret;
+ const char *dir_name;
+ struct consumer_output *consumer;
+
+ switch (domain) {
+ case LTTNG_DOMAIN_KERNEL:
+ DBG3("Copying tracing session consumer output in kernel session");
+ session->kernel_session->consumer =
+ consumer_copy_output(session->consumer);
+ /* Ease our life a bit for the next part */
+ consumer = session->kernel_session->consumer;
+ dir_name = DEFAULT_KERNEL_TRACE_DIR;
+ break;
+ case LTTNG_DOMAIN_UST:
+ DBG3("Copying tracing session consumer output in UST session");
+ session->ust_session->consumer =
+ consumer_copy_output(session->consumer);
+ /* Ease our life a bit for the next part */
+ consumer = session->ust_session->consumer;
+ dir_name = DEFAULT_UST_TRACE_DIR;
+ break;
+ default:
+ ret = LTTCOMM_UNKNOWN_DOMAIN;
+ goto error;
+ }
+
+ /* Append correct directory to subdir */
+ strncat(consumer->subdir, dir_name, sizeof(consumer->subdir));
+ DBG3("Copy session consumer subdir %s", consumer->subdir);
+
+ /* Add default trace directory name */
+ if (consumer->type == CONSUMER_DST_LOCAL) {
+ strncat(consumer->dst.trace_path, dir_name,
+ sizeof(consumer->dst.trace_path));
+ }
+
+ ret = LTTCOMM_OK;
+
+error:
+ return ret;
+}
+
+/*
+ * Create an UST session and add it to the session ust list.
+ */
+static int create_ust_session(struct ltt_session *session,
+ struct lttng_domain *domain)
+{
+ int ret;
+ struct ltt_ust_session *lus = NULL;
+
+ assert(session);
+ assert(session->consumer);
+
+ switch (domain->type) {
+ case LTTNG_DOMAIN_UST:
+ break;
+ default:
+ ERR("Unknown UST domain on create session %d", domain->type);
+ ret = LTTCOMM_UNKNOWN_DOMAIN;
+ goto error;
+ }
+
+ DBG("Creating UST session");
+
+ lus = trace_ust_create_session(session->path, session->id, domain);
+ if (lus == NULL) {
+ ret = LTTCOMM_UST_SESS_FAIL;
+ goto error;
+ }
+
+ if (session->consumer->type == CONSUMER_DST_LOCAL) {
+ ret = run_as_mkdir_recursive(lus->pathname, S_IRWXU | S_IRWXG,
+ session->uid, session->gid);
+ if (ret < 0) {
+ if (ret != -EEXIST) {
+ ERR("Trace directory creation error");
+ ret = LTTCOMM_UST_SESS_FAIL;
+ goto error;
+ }
+ }
+ }
+
+ lus->uid = session->uid;
+ lus->gid = session->gid;
+ session->ust_session = lus;
+
+ /* Copy session output to the newly created UST session */
+ ret = copy_session_consumer(domain->type, session);
+ if (ret != LTTCOMM_OK) {
+ goto error;
+ }
+
+ return LTTCOMM_OK;
+
+error:
+ free(lus);
+ session->ust_session = NULL;
+ return ret;
+}
+
+/*
+ * Create a kernel tracer session then create the default channel.
+ */
+static int create_kernel_session(struct ltt_session *session)
+{
+ int ret;
+
+ DBG("Creating kernel session");
+
+ ret = kernel_create_session(session, kernel_tracer_fd);
+ if (ret < 0) {
+ ret = LTTCOMM_KERN_SESS_FAIL;
+ goto error;
+ }
+
+ /* Copy session output to the newly created Kernel session */
+ ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
+ if (ret != LTTCOMM_OK) {
+ goto error;
+ }
+
+ /* Create directory(ies) on local filesystem. */
+ if (session->consumer->type == CONSUMER_DST_LOCAL) {
+ ret = run_as_mkdir_recursive(
+ session->kernel_session->consumer->dst.trace_path,
+ S_IRWXU | S_IRWXG, session->uid, session->gid);
+ if (ret < 0) {
+ if (ret != -EEXIST) {
+ ERR("Trace directory creation error");
+ goto error;
+ }
+ }
+ }
+
+ session->kernel_session->uid = session->uid;
+ session->kernel_session->gid = session->gid;
+
+ return LTTCOMM_OK;
+
+error:
+ trace_kernel_destroy_session(session->kernel_session);
+ session->kernel_session = NULL;
+ return ret;
+}
+
+/*
+ * Check if the UID or GID match the session. Root user has access to all
+ * sessions.
+ */
+static int session_access_ok(struct ltt_session *session, uid_t uid, gid_t gid)
+{
+ if (uid != session->uid && gid != session->gid && uid != 0) {
+ return 0;
+ } else {
return 1;
}
}
+/*
+ * Count number of session permitted by uid/gid.
+ */
static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
{
unsigned int i = 0;
tmp[i].loglevel_type = LTTNG_EVENT_LOGLEVEL_SINGLE;
break;
}
+ if (uevent->filter) {
+ tmp[i].filter = 1;
+ }
i++;
}
return ret;
}
+/*
+ * Command LTTNG_SET_FILTER processed by the client thread.
+ */
+static int cmd_set_filter(struct ltt_session *session, int domain,
+ char *channel_name, char *event_name,
+ struct lttng_filter_bytecode *bytecode)
+{
+ int ret;
+
+ switch (domain) {
+ case LTTNG_DOMAIN_KERNEL:
+ ret = LTTCOMM_FATAL;
+ break;
+ case LTTNG_DOMAIN_UST:
+ {
+ struct ltt_ust_session *usess = session->ust_session;
+
+ ret = filter_ust_set(usess, domain, bytecode, event_name, channel_name);
+ if (ret != LTTCOMM_OK) {
+ goto error;
+ }
+ break;
+ }
+#if 0
+ case LTTNG_DOMAIN_UST_EXEC_NAME:
+ case LTTNG_DOMAIN_UST_PID:
+ case LTTNG_DOMAIN_UST_PID_FOLLOW_CHILDREN:
+#endif
+ default:
+ ret = LTTCOMM_UND;
+ goto error;
+ }
+
+ ret = LTTCOMM_OK;
+
+error:
+ return ret;
+
+}
+
/*
* Command LTTNG_ENABLE_EVENT processed by the client thread.
*/
int ret;
struct ltt_kernel_session *ksession;
struct ltt_ust_session *usess;
+ struct ltt_kernel_channel *kchan;
- /* Short cut */
+ /* Ease our life a bit ;) */
ksession = session->kernel_session;
usess = session->ust_session;
session->enabled = 1;
+ ret = setup_relayd(session);
+ if (ret != LTTCOMM_OK) {
+ ERR("Error setting up relayd for session %s", session->name);
+ goto error;
+ }
+
/* Kernel tracing */
if (ksession != NULL) {
- struct ltt_kernel_channel *kchan;
-
/* Open kernel metadata */
if (ksession->metadata == NULL) {
- ret = kernel_open_metadata(ksession, ksession->trace_path);
+ ret = kernel_open_metadata(ksession,
+ ksession->consumer->dst.trace_path);
if (ret < 0) {
ret = LTTCOMM_KERN_META_FAIL;
goto error;
if (ksession != NULL) {
DBG("Stop kernel tracing");
- /* Flush all buffers before stopping */
- ret = kernel_metadata_flush_buffer(ksession->metadata_stream_fd);
- if (ret < 0) {
- ERR("Kernel metadata flush failed");
+ /* Flush metadata if exist */
+ if (ksession->metadata_stream_fd >= 0) {
+ ret = kernel_metadata_flush_buffer(ksession->metadata_stream_fd);
+ if (ret < 0) {
+ ERR("Kernel metadata flush failed");
+ }
}
+ /* Flush all buffers before stopping */
cds_list_for_each_entry(kchan, &ksession->channel_list.head, list) {
ret = kernel_flush_buffer(kchan);
if (ret < 0) {
}
/*
- * Command LTTNG_CREATE_SESSION processed by the client thread.
+ * Command LTTNG_CREATE_SESSION_URI processed by the client thread.
*/
-static int cmd_create_session(char *name, char *path, lttng_sock_cred *creds)
+static int cmd_create_session_uri(char *name, struct lttng_uri *ctrl_uri,
+ struct lttng_uri *data_uri, unsigned int enable_consumer,
+ lttng_sock_cred *creds)
{
int ret;
+ char *path = NULL;
+ struct ltt_session *session;
+ struct consumer_output *consumer;
+
+ /* Verify if the session already exist */
+ session = session_find_by_name(name);
+ if (session != NULL) {
+ ret = LTTCOMM_EXIST_SESS;
+ goto error;
+ }
+
+ /* TODO: validate URIs */
+
+ /* Create default consumer output */
+ consumer = consumer_create_output(CONSUMER_DST_LOCAL);
+ if (consumer == NULL) {
+ ret = LTTCOMM_FATAL;
+ goto error;
+ }
+ strncpy(consumer->subdir, ctrl_uri->subdir, sizeof(consumer->subdir));
+ DBG2("Consumer subdir set to %s", consumer->subdir);
+
+ switch (ctrl_uri->dtype) {
+ case LTTNG_DST_IPV4:
+ case LTTNG_DST_IPV6:
+ /* Set control URI into consumer output object */
+ ret = consumer_set_network_uri(consumer, ctrl_uri);
+ if (ret < 0) {
+ ret = LTTCOMM_FATAL;
+ goto error;
+ }
+
+ /* Set data URI into consumer output object */
+ ret = consumer_set_network_uri(consumer, data_uri);
+ if (ret < 0) {
+ ret = LTTCOMM_FATAL;
+ goto error;
+ }
+
+ /* Empty path since the session is network */
+ path = "";
+ break;
+ case LTTNG_DST_PATH:
+ /* Very volatile pointer. Only used for the create session. */
+ path = ctrl_uri->dst.path;
+ strncpy(consumer->dst.trace_path, path,
+ sizeof(consumer->dst.trace_path));
+ break;
+ }
+
+ /* Set if the consumer is enabled or not */
+ consumer->enabled = enable_consumer;
ret = session_create(name, path, LTTNG_SOCK_GET_UID_CRED(creds),
LTTNG_SOCK_GET_GID_CRED(creds));
if (ret != LTTCOMM_OK) {
- goto error;
+ goto consumer_error;
}
- ret = LTTCOMM_OK;
+ /* Get the newly created session pointer back */
+ session = session_find_by_name(name);
+ assert(session);
+
+ /* Assign consumer to session */
+ session->consumer = consumer;
+
+ return LTTCOMM_OK;
+
+consumer_error:
+ consumer_destroy_output(consumer);
+error:
+ return ret;
+}
+
+/*
+ * Command LTTNG_CREATE_SESSION processed by the client thread.
+ */
+static int cmd_create_session(char *name, char *path, lttng_sock_cred *creds)
+{
+ int ret;
+ struct lttng_uri uri;
+
+ /* Zeroed temporary URI */
+ memset(&uri, 0, sizeof(uri));
+
+ uri.dtype = LTTNG_DST_PATH;
+ uri.utype = LTTNG_URI_DST;
+ strncpy(uri.dst.path, path, sizeof(uri.dst.path));
+
+ /* TODO: Strip date-time from path and put it in uri's subdir */
+
+ ret = cmd_create_session_uri(name, &uri, NULL, 1, creds);
+ if (ret != LTTCOMM_OK) {
+ goto error;
+ }
error:
return ret;
{
int ret;
+ /* Safety net */
+ assert(session);
+
/* Clean kernel session teardown */
teardown_kernel_session(session);
/* UST session teardown */
char *sock_path)
{
int ret, sock;
+ struct consumer_socket *socket;
switch (domain) {
case LTTNG_DOMAIN_KERNEL:
goto error;
}
- session->kernel_session->consumer_fd = sock;
+ socket = consumer_allocate_socket(sock);
+ if (socket == NULL) {
+ ret = LTTCOMM_FATAL;
+ close(sock);
+ goto error;
+ }
+
+ socket->lock = zmalloc(sizeof(pthread_mutex_t));
+ if (socket->lock == NULL) {
+ PERROR("zmalloc pthread mutex");
+ ret = LTTCOMM_FATAL;
+ goto error;
+ }
+ pthread_mutex_init(socket->lock, NULL);
+
+ rcu_read_lock();
+ consumer_add_socket(socket, session->kernel_session->consumer);
+ rcu_read_unlock();
+
+ pthread_mutex_lock(&kconsumer_data.pid_mutex);
+ kconsumer_data.pid = -1;
+ pthread_mutex_unlock(&kconsumer_data.pid_mutex);
+
break;
default:
/* TODO: Userspace tracing */
}
break;
}
- default:
- ret = -LTTCOMM_UND;
- goto error;
- }
+ default:
+ ret = -LTTCOMM_UND;
+ goto error;
+ }
+
+ ret = nb_event;
+
+error:
+ return ret;
+}
+
+/*
+ * Command LTTNG_SET_CONSUMER_URI processed by the client thread.
+ */
+static int cmd_set_consumer_uri(int domain, struct ltt_session *session,
+ struct lttng_uri *uri)
+{
+ int ret;
+ struct ltt_kernel_session *ksess = session->kernel_session;
+ struct ltt_ust_session *usess = session->ust_session;
+ struct consumer_output *consumer;
+
+ /* Can't enable consumer after session started. */
+ if (session->enabled) {
+ ret = LTTCOMM_TRACE_ALREADY_STARTED;
+ goto error;
+ }
+
+ switch (domain) {
+ case LTTNG_DOMAIN_KERNEL:
+ {
+ struct lttng_ht_iter iter;
+ struct consumer_socket *socket;
+
+ /* Code flow error if we don't have a kernel session here. */
+ assert(ksess);
+
+ /* Create consumer output if none exists */
+ consumer = ksess->tmp_consumer;
+ if (consumer == NULL) {
+ consumer = consumer_copy_output(ksess->consumer);
+ if (consumer == NULL) {
+ ret = LTTCOMM_FATAL;
+ goto error;
+ }
+ /* Reassign new pointer */
+ ksess->tmp_consumer = consumer;
+ }
+
+ switch (uri->dtype) {
+ case LTTNG_DST_IPV4:
+ case LTTNG_DST_IPV6:
+ DBG2("Setting network URI for kernel session %s", session->name);
+
+ /* Set URI into consumer output object */
+ ret = consumer_set_network_uri(consumer, uri);
+ if (ret < 0) {
+ ret = LTTCOMM_FATAL;
+ goto error;
+ }
+
+ /* On a new subdir, reappend the default trace dir. */
+ if (strlen(uri->subdir) != 0) {
+ strncat(consumer->subdir, DEFAULT_KERNEL_TRACE_DIR,
+ sizeof(consumer->subdir));
+ }
+
+ cds_lfht_for_each_entry(consumer->socks->ht, &iter.iter,
+ socket, node.node) {
+ /* Code flow error */
+ assert(socket->fd >= 0);
+
+ pthread_mutex_lock(socket->lock);
+ ret = send_socket_relayd_consumer(domain, session, uri, consumer,
+ socket->fd);
+ pthread_mutex_unlock(socket->lock);
+ if (ret != LTTCOMM_OK) {
+ goto error;
+ }
+ }
+
+ break;
+ case LTTNG_DST_PATH:
+ DBG2("Setting trace directory path from URI to %s", uri->dst.path);
+ memset(consumer->dst.trace_path, 0,
+ sizeof(consumer->dst.trace_path));
+ strncpy(consumer->dst.trace_path, uri->dst.path,
+ sizeof(consumer->dst.trace_path));
+ /* Append default kernel trace dir */
+ strncat(consumer->dst.trace_path, DEFAULT_KERNEL_TRACE_DIR,
+ sizeof(consumer->dst.trace_path));
+ break;
+ }
+
+ /* All good! */
+ break;
+ }
+ case LTTNG_DOMAIN_UST:
+ /* Code flow error if we don't have a kernel session here. */
+ assert(usess);
+
+ /* Create consumer output if none exists */
+ consumer = usess->tmp_consumer;
+ if (consumer == NULL) {
+ consumer = consumer_copy_output(usess->consumer);
+ if (consumer == NULL) {
+ ret = LTTCOMM_FATAL;
+ goto error;
+ }
+ /* Reassign new pointer */
+ usess->tmp_consumer = consumer;
+ }
+
+ switch (uri->dtype) {
+ case LTTNG_DST_IPV4:
+ case LTTNG_DST_IPV6:
+ {
+ struct consumer_socket *socket;
+
+ DBG2("Setting network URI for UST session %s", session->name);
+
+ /* Set URI into consumer object */
+ ret = consumer_set_network_uri(consumer, uri);
+ if (ret < 0) {
+ ret = LTTCOMM_FATAL;
+ goto error;
+ }
+
+ /* On a new subdir, reappend the default trace dir. */
+ if (strlen(uri->subdir) != 0) {
+ strncat(consumer->subdir, DEFAULT_UST_TRACE_DIR,
+ sizeof(consumer->subdir));
+ }
+
+ rcu_read_lock();
+ socket = consumer_find_socket(uatomic_read(&ust_consumerd64_fd),
+ consumer);
+ if (socket != NULL) {
+ pthread_mutex_lock(socket->lock);
+ ret = send_socket_relayd_consumer(domain, session, uri,
+ consumer, socket->fd);
+ pthread_mutex_unlock(socket->lock);
+ if (ret != LTTCOMM_OK) {
+ goto error;
+ }
+ }
+
+ socket = consumer_find_socket(uatomic_read(&ust_consumerd32_fd),
+ consumer);
+ if (socket != NULL) {
+ pthread_mutex_lock(socket->lock);
+ ret = send_socket_relayd_consumer(domain, session, uri,
+ consumer, socket->fd);
+ pthread_mutex_unlock(socket->lock);
+ if (ret != LTTCOMM_OK) {
+ goto error;
+ }
+ }
+ rcu_read_unlock();
+ break;
+ }
+ case LTTNG_DST_PATH:
+ DBG2("Setting trace directory path from URI to %s", uri->dst.path);
+ memset(consumer->dst.trace_path, 0,
+ sizeof(consumer->dst.trace_path));
+ strncpy(consumer->dst.trace_path, uri->dst.path,
+ sizeof(consumer->dst.trace_path));
+ /* Append default UST trace dir */
+ strncat(consumer->dst.trace_path, DEFAULT_UST_TRACE_DIR,
+ sizeof(consumer->dst.trace_path));
+ break;
+ }
+ break;
+ }
+
+ /* All good! */
+ ret = LTTCOMM_OK;
+
+error:
+ return ret;
+}
+
+/*
+ * Command LTTNG_DISABLE_CONSUMER processed by the client thread.
+ */
+static int cmd_disable_consumer(int domain, struct ltt_session *session)
+{
+ int ret;
+ struct ltt_kernel_session *ksess = session->kernel_session;
+ struct ltt_ust_session *usess = session->ust_session;
+ struct consumer_output *consumer;
+
+ if (session->enabled) {
+ /* Can't disable consumer on an already started session */
+ ret = LTTCOMM_TRACE_ALREADY_STARTED;
+ goto error;
+ }
+
+ switch (domain) {
+ case LTTNG_DOMAIN_KERNEL:
+ /* Code flow error if we don't have a kernel session here. */
+ assert(ksess);
+
+ DBG("Disabling kernel consumer");
+ consumer = ksess->consumer;
+
+ break;
+ case LTTNG_DOMAIN_UST:
+ /* Code flow error if we don't have a UST session here. */
+ assert(usess);
+
+ DBG("Disabling UST consumer");
+ consumer = usess->consumer;
+
+ break;
+ default:
+ ret = LTTCOMM_UNKNOWN_DOMAIN;
+ goto error;
+ }
+
+ assert(consumer);
+ consumer->enabled = 0;
+
+ /* Success at this point */
+ ret = LTTCOMM_OK;
+
+error:
+ return ret;
+}
+
+/*
+ * Command LTTNG_ENABLE_CONSUMER processed by the client thread.
+ */
+static int cmd_enable_consumer(int domain, struct ltt_session *session)
+{
+ int ret;
+ struct ltt_kernel_session *ksess = session->kernel_session;
+ struct ltt_ust_session *usess = session->ust_session;
+ struct consumer_output *tmp_out;
+
+ /* Can't enable consumer after session started. */
+ if (session->enabled) {
+ ret = LTTCOMM_TRACE_ALREADY_STARTED;
+ goto error;
+ }
+
+ switch (domain) {
+ case LTTNG_DOMAIN_KERNEL:
+ /* Code flow error if we don't have a kernel session here. */
+ assert(ksess);
+
+ /*
+ * Check if we have already sent fds to the consumer. In that case,
+ * the enable-consumer command can't be used because a start trace
+ * had previously occured.
+ */
+ if (ksess->consumer_fds_sent) {
+ ret = LTTCOMM_ENABLE_CONSUMER_FAIL;
+ goto error;
+ }
+
+ tmp_out = ksess->tmp_consumer;
+ if (tmp_out == NULL) {
+ /* No temp. consumer output exists. Using the current one. */
+ DBG3("No temporary consumer. Using default");
+ ret = LTTCOMM_OK;
+ goto error;
+ }
+
+ switch (tmp_out->type) {
+ case CONSUMER_DST_LOCAL:
+ DBG2("Consumer output is local. Creating directory(ies)");
+
+ /* Create directory(ies) */
+ ret = run_as_mkdir_recursive(tmp_out->dst.trace_path,
+ S_IRWXU | S_IRWXG, session->uid, session->gid);
+ if (ret < 0) {
+ if (ret != -EEXIST) {
+ ERR("Trace directory creation error");
+ ret = LTTCOMM_FATAL;
+ goto error;
+ }
+ }
+ break;
+ case CONSUMER_DST_NET:
+ DBG2("Consumer output is network. Validating URIs");
+ /* Validate if we have both control and data path set. */
+ if (!tmp_out->dst.net.control_isset) {
+ ret = LTTCOMM_URI_CTRL_MISS;
+ goto error;
+ }
+
+ if (!tmp_out->dst.net.data_isset) {
+ ret = LTTCOMM_URI_DATA_MISS;
+ goto error;
+ }
+
+ /* Check established network session state */
+ if (session->net_handle == 0) {
+ ret = LTTCOMM_ENABLE_CONSUMER_FAIL;
+ ERR("Session network handle is not set on enable-consumer");
+ goto error;
+ }
+
+ /* Append default kernel trace dir to subdir */
+ strncat(ksess->consumer->subdir, DEFAULT_KERNEL_TRACE_DIR,
+ sizeof(ksess->consumer->subdir));
+
+ break;
+ }
+
+ /*
+ * @session-lock
+ * This is race free for now since the session lock is acquired before
+ * ending up in this function. No other threads can access this kernel
+ * session without this lock hence freeing the consumer output object
+ * is valid.
+ */
+ consumer_destroy_output(ksess->consumer);
+ ksess->consumer = tmp_out;
+ ksess->tmp_consumer = NULL;
+
+ break;
+ case LTTNG_DOMAIN_UST:
+ /* Code flow error if we don't have a UST session here. */
+ assert(usess);
+
+ /*
+ * Check if we have already sent fds to the consumer. In that case,
+ * the enable-consumer command can't be used because a start trace
+ * had previously occured.
+ */
+ if (usess->start_trace) {
+ ret = LTTCOMM_ENABLE_CONSUMER_FAIL;
+ goto error;
+ }
+
+ tmp_out = usess->tmp_consumer;
+ if (tmp_out == NULL) {
+ /* No temp. consumer output exists. Using the current one. */
+ DBG3("No temporary consumer. Using default");
+ ret = LTTCOMM_OK;
+ goto error;
+ }
+
+ switch (tmp_out->type) {
+ case CONSUMER_DST_LOCAL:
+ DBG2("Consumer output is local. Creating directory(ies)");
+
+ /* Create directory(ies) */
+ ret = run_as_mkdir_recursive(tmp_out->dst.trace_path,
+ S_IRWXU | S_IRWXG, session->uid, session->gid);
+ if (ret < 0) {
+ if (ret != -EEXIST) {
+ ERR("Trace directory creation error");
+ ret = LTTCOMM_FATAL;
+ goto error;
+ }
+ }
+ break;
+ case CONSUMER_DST_NET:
+ DBG2("Consumer output is network. Validating URIs");
+ /* Validate if we have both control and data path set. */
+ if (!tmp_out->dst.net.control_isset) {
+ ret = LTTCOMM_URI_CTRL_MISS;
+ goto error;
+ }
+
+ if (!tmp_out->dst.net.data_isset) {
+ ret = LTTCOMM_URI_DATA_MISS;
+ goto error;
+ }
+
+ /* Check established network session state */
+ if (session->net_handle == 0) {
+ ret = LTTCOMM_ENABLE_CONSUMER_FAIL;
+ DBG2("Session network handle is not set on enable-consumer");
+ goto error;
+ }
+
+ if (tmp_out->net_seq_index == -1) {
+ ret = LTTCOMM_ENABLE_CONSUMER_FAIL;
+ DBG2("Network index is not set on the consumer");
+ goto error;
+ }
+
+ /* Append default kernel trace dir to subdir */
+ strncat(usess->consumer->subdir, DEFAULT_UST_TRACE_DIR,
+ sizeof(usess->consumer->subdir));
+
+ break;
+ }
+
+ /*
+ * @session-lock
+ * This is race free for now since the session lock is acquired before
+ * ending up in this function. No other threads can access this kernel
+ * session without this lock hence freeing the consumer output object
+ * is valid.
+ */
+ consumer_destroy_output(usess->consumer);
+ usess->consumer = tmp_out;
+ usess->tmp_consumer = NULL;
+
+ break;
+ }
- ret = nb_event;
+ /* Success at this point */
+ ret = LTTCOMM_OK;
error:
return ret;
* is set and ready for transmission before returning.
*
* Return any error encountered or 0 for success.
+ *
+ * "sock" is only used for special-case var. len data.
*/
-static int process_client_msg(struct command_ctx *cmd_ctx)
+static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
+ int *sock_error)
{
int ret = LTTCOMM_OK;
int need_tracing_session = 1;
DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
+ *sock_error = 0;
+
switch (cmd_ctx->lsm->cmd_type) {
case LTTNG_CREATE_SESSION:
+ case LTTNG_CREATE_SESSION_URI:
case LTTNG_DESTROY_SESSION:
case LTTNG_LIST_SESSIONS:
case LTTNG_LIST_DOMAINS:
/* Commands that DO NOT need a session. */
switch (cmd_ctx->lsm->cmd_type) {
case LTTNG_CREATE_SESSION:
+ case LTTNG_CREATE_SESSION_URI:
case LTTNG_CALIBRATE:
case LTTNG_LIST_SESSIONS:
case LTTNG_LIST_TRACEPOINTS:
/* Need a session for kernel command */
if (need_tracing_session) {
+ struct consumer_socket *socket;
+
if (cmd_ctx->session->kernel_session == NULL) {
ret = create_kernel_session(cmd_ctx->session);
if (ret < 0) {
} else {
pthread_mutex_unlock(&kconsumer_data.pid_mutex);
}
+
+ /* Set kernel consumer socket fd */
+ if (kconsumer_data.cmd_sock >= 0) {
+ rcu_read_lock();
+ socket = consumer_find_socket(kconsumer_data.cmd_sock,
+ cmd_ctx->session->kernel_session->consumer);
+ rcu_read_unlock();
+ if (socket == NULL) {
+ socket = consumer_allocate_socket(kconsumer_data.cmd_sock);
+ if (socket == NULL) {
+ goto error;
+ }
+
+ socket->lock = &kconsumer_data.lock;
+ rcu_read_lock();
+ consumer_add_socket(socket,
+ cmd_ctx->session->kernel_session->consumer);
+ rcu_read_unlock();
+ }
+ }
}
break;
}
if (need_tracing_session) {
+ struct consumer_socket *socket;
+
if (cmd_ctx->session->ust_session == NULL) {
ret = create_ust_session(cmd_ctx->session,
&cmd_ctx->lsm->domain);
goto error;
}
}
+
/* Start the UST consumer daemons */
/* 64-bit */
pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
ret = start_consumerd(&ustconsumer64_data);
if (ret < 0) {
ret = LTTCOMM_UST_CONSUMER64_FAIL;
- ust_consumerd64_fd = -EINVAL;
+ uatomic_set(&ust_consumerd64_fd, -EINVAL);
goto error;
}
- ust_consumerd64_fd = ustconsumer64_data.cmd_sock;
+ uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
} else {
pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
}
+
+ /*
+ * Setup socket for consumer 64 bit. No need for atomic access
+ * since it was set above and can ONLY be set in this thread.
+ */
+ if (ust_consumerd64_fd >= 0) {
+ rcu_read_lock();
+ socket = consumer_find_socket(uatomic_read(&ust_consumerd64_fd),
+ cmd_ctx->session->ust_session->consumer);
+ rcu_read_unlock();
+ if (socket == NULL) {
+ socket = consumer_allocate_socket(ust_consumerd64_fd);
+ if (socket == NULL) {
+ goto error;
+ }
+ socket->lock = &ustconsumer32_data.lock;
+
+ rcu_read_lock();
+ consumer_add_socket(socket,
+ cmd_ctx->session->ust_session->consumer);
+ rcu_read_unlock();
+ }
+ DBG3("UST consumer 64 bit socket set to %d", socket->fd);
+ }
+
/* 32-bit */
if (consumerd32_bin[0] != '\0' &&
ustconsumer32_data.pid == 0 &&
ret = start_consumerd(&ustconsumer32_data);
if (ret < 0) {
ret = LTTCOMM_UST_CONSUMER32_FAIL;
- ust_consumerd32_fd = -EINVAL;
+ uatomic_set(&ust_consumerd32_fd, -EINVAL);
goto error;
}
- ust_consumerd32_fd = ustconsumer32_data.cmd_sock;
+ uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
} else {
pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
}
+
+ /*
+ * Setup socket for consumer 64 bit. No need for atomic access
+ * since it was set above and can ONLY be set in this thread.
+ */
+ if (ust_consumerd32_fd >= 0) {
+ rcu_read_lock();
+ socket = consumer_find_socket(uatomic_read(&ust_consumerd64_fd),
+ cmd_ctx->session->ust_session->consumer);
+ rcu_read_unlock();
+ if (socket == NULL) {
+ socket = consumer_allocate_socket(ust_consumerd32_fd);
+ if (socket == NULL) {
+ goto error;
+ }
+ socket->lock = &ustconsumer32_data.lock;
+
+ rcu_read_lock();
+ consumer_add_socket(socket,
+ cmd_ctx->session->ust_session->consumer);
+ rcu_read_unlock();
+ }
+ DBG3("UST consumer 32 bit socket set to %d", socket->fd);
+ }
}
break;
}
cmd_ctx->lsm->u.disable.channel_name);
break;
}
+ case LTTNG_DISABLE_CONSUMER:
+ {
+ ret = cmd_disable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
+ break;
+ }
case LTTNG_ENABLE_CHANNEL:
{
ret = cmd_enable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
&cmd_ctx->lsm->u.channel.chan);
break;
}
+ case LTTNG_ENABLE_CONSUMER:
+ {
+ ret = cmd_enable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
+ break;
+ }
case LTTNG_ENABLE_EVENT:
{
ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
ret = LTTCOMM_OK;
break;
}
-
+ case LTTNG_SET_CONSUMER_URI:
+ {
+ ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
+ &cmd_ctx->lsm->u.uri);
+ break;
+ }
case LTTNG_START_TRACE:
{
ret = cmd_start_trace(cmd_ctx->session);
cmd_ctx->lsm->session.path, &cmd_ctx->creds);
break;
}
+ case LTTNG_CREATE_SESSION_URI:
+ {
+ ret = cmd_create_session_uri(cmd_ctx->lsm->session.name,
+ &cmd_ctx->lsm->u.create_uri.ctrl_uri,
+ &cmd_ctx->lsm->u.create_uri.data_uri,
+ cmd_ctx->lsm->u.create_uri.enable_consumer, &cmd_ctx->creds);
+ break;
+ }
case LTTNG_DESTROY_SESSION:
{
ret = cmd_destroy_session(cmd_ctx->session,
cmd_ctx->lsm->u.reg.path);
break;
}
+ case LTTNG_SET_FILTER:
+ {
+ struct lttng_filter_bytecode *bytecode;
+
+ if (cmd_ctx->lsm->u.filter.bytecode_len > 65336) {
+ ret = LTTCOMM_FILTER_INVAL;
+ goto error;
+ }
+ bytecode = zmalloc(cmd_ctx->lsm->u.filter.bytecode_len);
+ if (!bytecode) {
+ ret = LTTCOMM_FILTER_NOMEM;
+ goto error;
+ }
+ /* Receive var. len. data */
+ DBG("Receiving var len data from client ...");
+ ret = lttcomm_recv_unix_sock(sock, bytecode,
+ cmd_ctx->lsm->u.filter.bytecode_len);
+ if (ret <= 0) {
+ DBG("Nothing recv() from client var len data... continuing");
+ *sock_error = 1;
+ ret = LTTCOMM_FILTER_INVAL;
+ goto error;
+ }
+
+ if (bytecode->len + sizeof(*bytecode)
+ != cmd_ctx->lsm->u.filter.bytecode_len) {
+ free(bytecode);
+ ret = LTTCOMM_FILTER_INVAL;
+ goto error;
+ }
+
+ ret = cmd_set_filter(cmd_ctx->session, cmd_ctx->lsm->domain.type,
+ cmd_ctx->lsm->u.filter.channel_name,
+ cmd_ctx->lsm->u.filter.event_name,
+ bytecode);
+ break;
+ }
default:
ret = LTTCOMM_UND;
break;
return ret;
}
+/*
+ * Thread managing health check socket.
+ */
+static 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 lttcomm_health_msg msg;
+ struct lttcomm_health_data reply;
+
+ DBG("[thread] Manage health check started");
+
+ rcu_register_thread();
+
+ /* 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;
+ }
+
+ ret = lttcomm_listen_unix_sock(sock);
+ if (ret < 0) {
+ goto error;
+ }
+
+ /*
+ * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
+ * more will be added to this poll set.
+ */
+ ret = create_thread_poll_set(&events, 2);
+ if (ret < 0) {
+ goto error;
+ }
+
+ /* Add the application registration socket */
+ ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
+ if (ret < 0) {
+ goto error;
+ }
+
+ while (1) {
+ DBG("Health check ready");
+
+ nb_fd = LTTNG_POLL_GETNB(&events);
+
+ /* 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;
+ }
+
+ for (i = 0; i < nb_fd; i++) {
+ /* 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 = check_thread_quit_pipe(pollfd, revents);
+ if (ret) {
+ err = 0;
+ goto exit;
+ }
+
+ /* Event on the registration socket */
+ if (pollfd == sock) {
+ if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
+ ERR("Health socket poll error");
+ goto error;
+ }
+ }
+ }
+
+ new_sock = lttcomm_accept_unix_sock(sock);
+ if (new_sock < 0) {
+ goto error;
+ }
+
+ 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();
+
+ switch (msg.component) {
+ case LTTNG_HEALTH_CMD:
+ reply.ret_code = health_check_state(&health_thread_cmd);
+ break;
+ case LTTNG_HEALTH_APP_MANAGE:
+ reply.ret_code = health_check_state(&health_thread_app_manage);
+ break;
+ case LTTNG_HEALTH_APP_REG:
+ reply.ret_code = health_check_state(&health_thread_app_reg);
+ break;
+ case LTTNG_HEALTH_KERNEL:
+ reply.ret_code = health_check_state(&health_thread_kernel);
+ break;
+ case LTTNG_HEALTH_CONSUMER:
+ reply.ret_code = check_consumer_health();
+ break;
+ case LTTNG_HEALTH_ALL:
+ reply.ret_code =
+ health_check_state(&health_thread_app_manage) &&
+ health_check_state(&health_thread_app_reg) &&
+ health_check_state(&health_thread_cmd) &&
+ health_check_state(&health_thread_kernel) &&
+ check_consumer_health();
+ break;
+ default:
+ reply.ret_code = LTTCOMM_UND;
+ break;
+ }
+
+ /*
+ * Flip ret value since 0 is a success and 1 indicates a bad health for
+ * the client where in the sessiond it is the opposite. Again, this is
+ * just to make things easier for us poor developer which enjoy a lot
+ * lazyness.
+ */
+ if (reply.ret_code == 0 || reply.ret_code == 1) {
+ reply.ret_code = !reply.ret_code;
+ }
+
+ DBG2("Health check return value %d", 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");
+ }
+ }
+ if (new_sock >= 0) {
+ ret = close(new_sock);
+ if (ret) {
+ PERROR("close");
+ }
+ }
+
+ lttng_poll_clean(&events);
+
+ rcu_unregister_thread();
+ return NULL;
+}
+
/*
* This thread manage all clients request using the unix client socket for
* communication.
*/
static void *thread_manage_clients(void *data)
{
- int sock = -1, ret, i, pollfd;
+ int sock = -1, ret, i, pollfd, err = -1;
+ int sock_error;
uint32_t revents, nb_fd;
struct command_ctx *cmd_ctx = NULL;
struct lttng_poll_event events;
rcu_register_thread();
+ health_code_update(&health_thread_cmd);
+
ret = lttcomm_listen_unix_sock(client_sock);
if (ret < 0) {
goto error;
kill(ppid, SIGUSR1);
}
+ health_code_update(&health_thread_cmd);
+
while (1) {
DBG("Accepting client command ...");
/* Inifinite blocking call, waiting for transmission */
restart:
+ health_poll_update(&health_thread_cmd);
ret = lttng_poll_wait(&events, -1);
+ health_poll_update(&health_thread_cmd);
if (ret < 0) {
/*
* Restart interrupted system call.
revents = LTTNG_POLL_GETEV(&events, i);
pollfd = LTTNG_POLL_GETFD(&events, i);
+ health_code_update(&health_thread_cmd);
+
/* Thread quit pipe has been closed. Killing thread. */
ret = check_thread_quit_pipe(pollfd, revents);
if (ret) {
- goto error;
+ err = 0;
+ goto exit;
}
/* Event on the registration socket */
DBG("Wait for client response");
+ health_code_update(&health_thread_cmd);
+
sock = lttcomm_accept_unix_sock(client_sock);
if (sock < 0) {
goto error;
cmd_ctx->llm = NULL;
cmd_ctx->session = NULL;
+ health_code_update(&health_thread_cmd);
+
/*
* Data is received from the lttng client. The struct
* lttcomm_session_msg (lsm) contains the command and data request of
continue;
}
+ health_code_update(&health_thread_cmd);
+
// TODO: Validate cmd_ctx including sanity check for
// security purpose.
* informations for the client. The command context struct contains
* everything this function may needs.
*/
- ret = process_client_msg(cmd_ctx);
+ ret = process_client_msg(cmd_ctx, sock, &sock_error);
rcu_thread_offline();
if (ret < 0) {
+ if (sock_error) {
+ ret = close(sock);
+ if (ret) {
+ PERROR("close");
+ }
+ sock = -1;
+ }
/*
* TODO: Inform client somehow of the fatal error. At
* this point, ret < 0 means that a zmalloc failed
- * (ENOMEM). Error detected but still accept command.
+ * (ENOMEM). Error detected but still accept
+ * command, unless a socket error has been
+ * detected.
*/
clean_command_ctx(&cmd_ctx);
continue;
}
+ health_code_update(&health_thread_cmd);
+
DBG("Sending response (size: %d, retcode: %s)",
cmd_ctx->lttng_msg_size,
lttng_strerror(-cmd_ctx->llm->ret_code));
sock = -1;
clean_command_ctx(&cmd_ctx);
+
+ health_code_update(&health_thread_cmd);
}
+exit:
error:
+ if (err) {
+ health_error(&health_thread_cmd);
+ ERR("Health error occurred in %s", __func__);
+ }
+ health_exit(&health_thread_cmd);
+
DBG("Client thread dying");
unlink(client_unix_sock_path);
if (client_sock >= 0) {
return ret;
}
-/*
- * Create the pipe used to wake up the kernel thread.
- * Closed in cleanup().
- */
-static int create_kernel_poll_pipe(void)
-{
- int ret, i;
-
- ret = pipe(kernel_poll_pipe);
- if (ret < 0) {
- PERROR("kernel poll pipe");
- goto error;
- }
-
- for (i = 0; i < 2; i++) {
- ret = fcntl(kernel_poll_pipe[i], F_SETFD, FD_CLOEXEC);
- if (ret < 0) {
- PERROR("fcntl kernel_poll_pipe");
- goto error;
- }
- }
-
-error:
- return ret;
-}
-
-/*
- * Create the application command pipe to wake thread_manage_apps.
- * Closed in cleanup().
- */
-static int create_apps_cmd_pipe(void)
-{
- int ret, i;
-
- ret = pipe(apps_cmd_pipe);
- if (ret < 0) {
- PERROR("apps cmd pipe");
- goto error;
- }
-
- for (i = 0; i < 2; i++) {
- ret = fcntl(apps_cmd_pipe[i], F_SETFD, FD_CLOEXEC);
- if (ret < 0) {
- PERROR("fcntl apps_cmd_pipe");
- goto error;
- }
- }
-
-error:
- return ret;
-}
-
/*
* Create the lttng run directory needed for all global sockets and pipe.
*/
DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
}
+ if (strlen(health_unix_sock_path) == 0) {
+ snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
+ DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
+ }
+
/* Setup kernel consumerd path */
snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
snprintf(wait_shm_path, PATH_MAX,
DEFAULT_HOME_APPS_WAIT_SHM_PATH, geteuid());
}
+
+ /* Set health check Unix path */
+ if (strlen(health_unix_sock_path) == 0) {
+ snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
+ DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
+ }
}
/* Set consumer initial state */
}
/* Setup the kernel pipe for waking up the kernel thread */
- if ((ret = create_kernel_poll_pipe()) < 0) {
+ if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
goto exit;
}
/* Setup the thread apps communication pipe. */
- if ((ret = create_apps_cmd_pipe()) < 0) {
+ if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
goto exit;
}
/* Set up max poll set size */
lttng_poll_set_max_size();
+ /*
+ * Set network sequence index to 1 for streams to match a relayd socket on
+ * the consumer side.
+ */
+ uatomic_set(&relayd_net_seq_idx, 1);
+
+ /* Init all health thread counters. */
+ health_init(&health_thread_cmd);
+ health_init(&health_thread_kernel);
+ health_init(&health_thread_app_manage);
+ health_init(&health_thread_app_reg);
+
+ /*
+ * Init health counters of the consumer thread. We do a quick hack here to
+ * the state of the consumer health is fine even if the thread is not
+ * started. This is simply to ease our life and has no cost what so ever.
+ */
+ health_init(&kconsumer_data.health);
+ health_poll_update(&kconsumer_data.health);
+ health_init(&ustconsumer32_data.health);
+ health_poll_update(&ustconsumer32_data.health);
+ health_init(&ustconsumer64_data.health);
+ health_poll_update(&ustconsumer64_data.health);
+
+ /* Create thread to manage the client socket */
+ ret = pthread_create(&health_thread, NULL,
+ thread_manage_health, (void *) NULL);
+ if (ret != 0) {
+ PERROR("pthread_create health");
+ goto exit_health;
+ }
+
/* Create thread to manage the client socket */
ret = pthread_create(&client_thread, NULL,
thread_manage_clients, (void *) NULL);
}
exit_client:
+exit_health:
exit:
/*
* cleanup() is called when no other thread is running.