#include <common/futex.h>
#include <common/relayd/relayd.h>
#include <common/utils.h>
+#include <common/daemonize.h>
#include "lttng-sessiond.h"
#include "buffer-registry.h"
static const char *opt_pidfile;
static int opt_sig_parent;
static int opt_verbose_consumer;
-static int opt_daemon;
+static int opt_daemon, opt_background;
static int opt_no_kernel;
-static int is_root; /* Set to 1 if the daemon is running as root */
static pid_t ppid; /* Parent PID for --sig-parent option */
static pid_t child_ppid; /* Internal parent PID use with daemonize. */
static char *rundir;
+static int lockfile_fd = -1;
/* Set to 1 when a SIGUSR1 signal is received. */
static int recv_child_signal;
/* JUL TCP port for registration. Used by the JUL thread. */
unsigned int jul_tcp_port = DEFAULT_JUL_TCP_PORT;
+/* Am I root or not. */
+int is_root; /* Set to 1 if the daemon is running as root */
+
+/*
+ * Whether sessiond is ready for commands/health check requests.
+ * NR_LTTNG_SESSIOND_READY must match the number of calls to
+ * lttng_sessiond_notify_ready().
+ */
+#define NR_LTTNG_SESSIOND_READY 2
+int lttng_sessiond_ready = NR_LTTNG_SESSIOND_READY;
+
+/* Notify parents that we are ready for cmd and health check */
+static
+void lttng_sessiond_notify_ready(void)
+{
+ if (uatomic_sub_return(<tng_sessiond_ready, 1) == 0) {
+ /*
+ * Notify parent pid that we are ready to accept command
+ * for client side. This ppid is the one from the
+ * external process that spawned us.
+ */
+ if (opt_sig_parent) {
+ kill(ppid, SIGUSR1);
+ }
+
+ /*
+ * Notify the parent of the fork() process that we are
+ * ready.
+ */
+ if (opt_daemon || opt_background) {
+ kill(child_ppid, SIGUSR1);
+ }
+ }
+}
+
static
void setup_consumerd_path(void)
{
}
}
+/*
+ * Generate the full lock file path using the rundir.
+ *
+ * Return the snprintf() return value thus a negative value is an error.
+ */
+static int generate_lock_file_path(char *path, size_t len)
+{
+ int ret;
+
+ assert(path);
+ assert(rundir);
+
+ /* Build lockfile path from rundir. */
+ ret = snprintf(path, len, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE, rundir);
+ if (ret < 0) {
+ PERROR("snprintf lockfile path");
+ }
+
+ return ret;
+}
+
/*
* Cleanup the daemon
*/
DBG("Removing directory %s", path);
(void) rmdir(path);
- /*
- * We do NOT rmdir rundir because there are other processes
- * using it, for instance lttng-relayd, which can start in
- * parallel with this teardown.
- */
-
- free(rundir);
-
DBG("Cleaning up all sessions");
/* Destroy session list mutex */
close_consumer_sockets();
+
+ /*
+ * Cleanup lock file by deleting it and finaly closing it which will
+ * release the file system lock.
+ */
+ if (lockfile_fd >= 0) {
+ char lockfile_path[PATH_MAX];
+
+ ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path));
+ if (ret > 0) {
+ ret = remove(lockfile_path);
+ if (ret < 0) {
+ PERROR("remove lock file");
+ }
+ ret = close(lockfile_fd);
+ if (ret < 0) {
+ PERROR("close lock file");
+ }
+ }
+ }
+
+ /*
+ * We do NOT rmdir rundir because there are other processes
+ * using it, for instance lttng-relayd, which can start in
+ * parallel with this teardown.
+ */
+
+ free(rundir);
+
/* <fun> */
DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
"Matthew, BEET driven development works!%c[%dm",
*/
lttng_poll_init(&events);
- if (testpoint(thread_manage_kernel)) {
+ if (testpoint(sessiond_thread_manage_kernel)) {
goto error_testpoint;
}
health_code_update();
- if (testpoint(thread_manage_kernel_before_loop)) {
+ if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
goto error_testpoint;
}
health_code_update();
+ if (!revents) {
+ /* No activity for this FD (poll implementation). */
+ continue;
+ }
+
/* Thread quit pipe has been closed. Killing thread. */
ret = sessiond_check_thread_quit_pipe(pollfd, revents);
if (ret) {
*/
static void *thread_manage_consumer(void *data)
{
- int sock = -1, i, ret, pollfd, err = -1;
+ int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
uint32_t revents, nb_fd;
enum lttcomm_return_code code;
struct lttng_poll_event events;
restart:
health_poll_entry();
- if (testpoint(thread_manage_consumer)) {
+ if (testpoint(sessiond_thread_manage_consumer)) {
goto error;
}
health_code_update();
+ if (!revents) {
+ /* No activity for this FD (poll implementation). */
+ continue;
+ }
+
/* Thread quit pipe has been closed. Killing thread. */
ret = sessiond_check_thread_quit_pipe(pollfd, revents);
if (ret) {
}
health_code_update();
-
if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
/* Connect both socket, command and metadata. */
consumer_data->cmd_sock =
/* Infinite blocking call, waiting for transmission */
restart_poll:
while (1) {
+ health_code_update();
+
+ /* Exit the thread because the thread quit pipe has been triggered. */
+ if (should_quit) {
+ /* Not a health error. */
+ err = 0;
+ goto exit;
+ }
+
health_poll_entry();
ret = lttng_poll_wait(&events, -1);
health_poll_exit();
health_code_update();
- /* Thread quit pipe has been closed. Killing thread. */
- ret = sessiond_check_thread_quit_pipe(pollfd, revents);
- if (ret) {
- err = 0;
- goto exit;
+ if (!revents) {
+ /* No activity for this FD (poll implementation). */
+ continue;
}
+ /*
+ * Thread quit pipe has been triggered, flag that we should stop
+ * but continue the current loop to handle potential data from
+ * consumer.
+ */
+ should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
+
if (pollfd == sock) {
/* Event on the consumerd socket */
if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
ERR("Handling metadata request");
goto error;
}
- break;
- } else {
- ERR("Unknown pollfd");
- goto error;
}
+ /* No need for an else branch all FDs are tested prior. */
}
health_code_update();
}
}
consumer_data->cmd_sock = -1;
}
- if (*consumer_data->metadata_sock.fd_ptr >= 0) {
+ if (consumer_data->metadata_sock.fd_ptr &&
+ *consumer_data->metadata_sock.fd_ptr >= 0) {
ret = close(*consumer_data->metadata_sock.fd_ptr);
if (ret) {
PERROR("close");
}
}
-
if (sock >= 0) {
ret = close(sock);
if (ret) {
pthread_mutex_unlock(&consumer_data->lock);
/* Cleanup metadata socket mutex. */
- pthread_mutex_destroy(consumer_data->metadata_sock.lock);
- free(consumer_data->metadata_sock.lock);
-
+ if (consumer_data->metadata_sock.lock) {
+ pthread_mutex_destroy(consumer_data->metadata_sock.lock);
+ free(consumer_data->metadata_sock.lock);
+ }
lttng_poll_clean(&events);
error_poll:
if (err) {
health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
- if (testpoint(thread_manage_apps)) {
+ if (testpoint(sessiond_thread_manage_apps)) {
goto error_testpoint;
}
goto error;
}
- if (testpoint(thread_manage_apps_before_loop)) {
+ if (testpoint(sessiond_thread_manage_apps_before_loop)) {
goto error;
}
health_code_update();
+ if (!revents) {
+ /* No activity for this FD (poll implementation). */
+ continue;
+ }
+
/* Thread quit pipe has been closed. Killing thread. */
ret = sessiond_check_thread_quit_pipe(pollfd, revents);
if (ret) {
}
DBG("Apps with sock %d added to poll set", sock);
-
- health_code_update();
-
- break;
}
} else {
/*
/* Socket closed on remote end. */
ust_app_unregister(pollfd);
- break;
}
}
uint32_t revents = LTTNG_POLL_GETEV(&events, i);
int pollfd = LTTNG_POLL_GETFD(&events, i);
+ if (!revents) {
+ /* No activity for this FD (poll implementation). */
+ continue;
+ }
+
cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
&wait_queue->head, head) {
if (pollfd == wait_node->app->sock &&
health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
+ if (testpoint(sessiond_thread_app_reg_dispatch)) {
+ goto error_testpoint;
+ }
+
health_code_update();
CDS_INIT_LIST_HEAD(&wait_queue.head);
free(wait_node);
}
+error_testpoint:
DBG("Dispatch thread dying");
if (err) {
health_error();
health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
- if (testpoint(thread_registration_apps)) {
+ if (testpoint(sessiond_thread_registration_apps)) {
goto error_testpoint;
}
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 = sessiond_check_thread_quit_pipe(pollfd, revents);
if (ret) {
exit:
error:
- if (err) {
- health_error();
- ERR("Health error occurred in %s", __func__);
- }
-
/* Notify that the registration thread is gone */
notify_ust_apps(0);
error_create_poll:
error_testpoint:
DBG("UST Registration thread cleanup complete");
+ if (err) {
+ health_error();
+ ERR("Health error occurred in %s", __func__);
+ }
health_unregister(health_sessiond);
return NULL;
if (ret != 0) {
errno = ret;
if (ret == ETIMEDOUT) {
+ int pth_ret;
+
/*
* Call has timed out so we kill the kconsumerd_thread and return
* an error.
*/
ERR("Condition timed out. The consumer thread was never ready."
" Killing it");
- ret = pthread_cancel(consumer_data->thread);
- if (ret < 0) {
+ pth_ret = pthread_cancel(consumer_data->thread);
+ if (pth_ret < 0) {
PERROR("pthread_cancel consumer thread");
}
} else {
PERROR("pthread_cond_wait failed consumer thread");
}
+ /* Caller is expecting a negative value on failure. */
+ ret = -1;
goto error;
}
consumer_to_use = consumerd32_bin;
} else {
DBG("Could not find any valid consumerd executable");
+ ret = -EINVAL;
break;
}
DBG("Using kernel consumer at: %s", consumer_to_use);
- execl(consumer_to_use,
+ ret = execl(consumer_to_use,
"lttng-consumerd", verbosity, "-k",
"--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
"--consumerd-err-sock", consumer_data->err_unix_sock_path,
if (consumerd64_libdir[0] != '\0') {
free(tmpnew);
}
- if (ret) {
- goto error;
- }
break;
}
case LTTNG_CONSUMER32_UST:
if (consumerd32_libdir[0] != '\0') {
free(tmpnew);
}
- if (ret) {
- goto error;
- }
break;
}
default:
exit(EXIT_FAILURE);
}
if (errno != 0) {
- PERROR("kernel start consumer exec");
+ PERROR("Consumer execl()");
}
+ /* Reaching this point, we got a failure on our execl(). */
exit(EXIT_FAILURE);
} else if (pid > 0) {
ret = pid;
}
/* 32-bit */
+ pthread_mutex_lock(&ustconsumer32_data.pid_mutex);
if (consumerd32_bin[0] != '\0' &&
ustconsumer32_data.pid == 0 &&
cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
struct lttng_event *events;
ssize_t nb_events;
+ session_lock_list();
nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
+ session_unlock_list();
if (nb_events < 0) {
/* Return value is a negative lttng_error_code. */
ret = -nb_events;
struct lttng_event_field *fields;
ssize_t nb_fields;
+ session_lock_list();
nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
&fields);
+ session_unlock_list();
if (nb_fields < 0) {
/* Return value is a negative lttng_error_code. */
ret = -nb_fields;
case LTTNG_LIST_CHANNELS:
{
int nb_chan;
- struct lttng_channel *channels;
+ struct lttng_channel *channels = NULL;
nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
cmd_ctx->session, &channels);
goto error;
}
+ lttng_sessiond_notify_ready();
+
while (1) {
DBG("Health check ready");
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 = sessiond_check_thread_quit_pipe(pollfd, revents);
if (ret) {
rcu_thread_online();
- reply.ret_code = 0;
+ memset(&reply, 0, sizeof(reply));
for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
/*
* health_check_state returns 0 if health is
health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
- if (testpoint(thread_manage_clients)) {
- goto error_testpoint;
- }
-
health_code_update();
ret = lttcomm_listen_unix_sock(client_sock);
goto error;
}
- /*
- * Notify parent pid that we are ready to accept command for client side.
- * This ppid is the one from the external process that spawned us.
- */
- if (opt_sig_parent) {
- kill(ppid, SIGUSR1);
- }
+ lttng_sessiond_notify_ready();
- /* Notify the parent of the fork() process that we are ready. */
- if (opt_daemon) {
- kill(child_ppid, SIGUSR1);
+ /* This testpoint is after we signal readiness to the parent. */
+ if (testpoint(sessiond_thread_manage_clients)) {
+ goto error;
}
- if (testpoint(thread_manage_clients_before_loop)) {
+ if (testpoint(sessiond_thread_manage_clients_before_loop)) {
goto error;
}
health_code_update();
+ if (!revents) {
+ /* No activity for this FD (poll implementation). */
+ continue;
+ }
+
/* Thread quit pipe has been closed. Killing thread. */
ret = sessiond_check_thread_quit_pipe(pollfd, revents);
if (ret) {
error_listen:
error_create_poll:
-error_testpoint:
unlink(client_unix_sock_path);
if (client_sock >= 0) {
ret = close(client_sock);
fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
+ fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n");
fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
fprintf(stderr, " -V, --version Show version number.\n");
fprintf(stderr, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
{ "no-kernel", 0, 0, 'N' },
{ "pidfile", 1, 0, 'p' },
{ "jul-tcp-port", 1, 0, 'J' },
+ { "background", 0, 0, 'b' },
{ NULL, 0, 0, 0 }
};
while (1) {
int option_index = 0;
- c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:J:",
+ c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:J:b",
long_options, &option_index);
if (c == -1) {
break;
case 'd':
opt_daemon = 1;
break;
+ case 'b':
+ opt_background = 1;
+ break;
case 'g':
tracing_group_name = optarg;
break;
return;
}
+/*
+ * Create lockfile using the rundir and return its fd.
+ */
+static int create_lockfile(void)
+{
+ int ret;
+ char lockfile_path[PATH_MAX];
+
+ ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path));
+ if (ret < 0) {
+ goto error;
+ }
+
+ ret = utils_create_lock_file(lockfile_path);
+error:
+ return ret;
+}
+
/*
* Write JUL TCP port using the rundir.
*/
return;
}
-/*
- * Daemonize this process by forking and making the parent wait for the child
- * to signal it indicating readiness. Once received, the parent successfully
- * quits.
- *
- * The child process undergoes the same action that daemon(3) does meaning
- * setsid, chdir, and dup /dev/null into 0, 1 and 2.
- *
- * Return 0 on success else -1 on error.
- */
-static int daemonize(void)
-{
- int ret;
- pid_t pid;
-
- /* Get parent pid of this process. */
- child_ppid = getppid();
-
- pid = fork();
- if (pid < 0) {
- PERROR("fork");
- goto error;
- } else if (pid == 0) {
- int fd;
- pid_t sid;
-
- /* Child */
-
- /*
- * Get the newly created parent pid so we can signal that process when
- * we are ready to operate.
- */
- child_ppid = getppid();
-
- sid = setsid();
- if (sid < 0) {
- PERROR("setsid");
- goto error;
- }
-
- /* Try to change directory to /. If we can't well at least notify. */
- ret = chdir("/");
- if (ret < 0) {
- PERROR("chdir");
- }
-
- fd = open(_PATH_DEVNULL, O_RDWR, 0);
- if (fd < 0) {
- PERROR("open %s", _PATH_DEVNULL);
- /* Let 0, 1 and 2 open since we can't bind them to /dev/null. */
- } else {
- (void) dup2(fd, STDIN_FILENO);
- (void) dup2(fd, STDOUT_FILENO);
- (void) dup2(fd, STDERR_FILENO);
- if (fd > 2) {
- ret = close(fd);
- if (ret < 0) {
- PERROR("close");
- }
- }
- }
- goto end;
- } else {
- /* Parent */
-
- /*
- * Waiting for child to notify this parent that it can exit. Note that
- * sleep() is interrupted before the 1 second delay as soon as the
- * signal is received, so it will not cause visible delay for the
- * user.
- */
- while (!CMM_LOAD_SHARED(recv_child_signal)) {
- int status;
- pid_t ret;
-
- /*
- * Check if child exists without blocking. If so, we have to stop
- * this parent process and return an error.
- */
- ret = waitpid(pid, &status, WNOHANG);
- if (ret < 0 || (ret != 0 && WIFEXITED(status))) {
- /* The child exited somehow or was not valid. */
- goto error;
- }
- sleep(1);
- }
-
- /*
- * From this point on, the parent can exit and the child is now an
- * operationnal session daemon ready to serve clients and applications.
- */
- exit(EXIT_SUCCESS);
- }
-
-end:
- return 0;
-
-error:
- return -1;
-}
-
/*
* main
*/
}
/* Daemonize */
- if (opt_daemon) {
+ if (opt_daemon || opt_background) {
int i;
- ret = daemonize();
+ ret = lttng_daemonize(&child_ppid, &recv_child_signal,
+ !opt_background);
if (ret < 0) {
goto error;
}
if (is_root) {
rundir = strdup(DEFAULT_LTTNG_RUNDIR);
+ if (!rundir) {
+ ret = -ENOMEM;
+ goto error;
+ }
/* Create global run dir with root access */
ret = create_lttng_rundir(rundir);
}
}
+ lockfile_fd = create_lockfile();
+ if (lockfile_fd < 0) {
+ goto error;
+ }
+
/* Set consumer initial state */
kernel_consumerd_state = CONSUMER_STOPPED;
ust_consumerd_state = CONSUMER_STOPPED;
ret = pthread_create(&apps_notify_thread, NULL,
ust_thread_manage_notify, (void *) NULL);
if (ret != 0) {
- PERROR("pthread_create apps");
+ PERROR("pthread_create notify");
goto exit_apps_notify;
}
ret = pthread_create(&jul_reg_thread, NULL,
jul_thread_manage_registration, (void *) NULL);
if (ret != 0) {
- PERROR("pthread_create apps");
+ PERROR("pthread_create JUL");
goto exit_jul_reg;
}
projects / lttng-tools.git / blobdiff