#include <common/utils.h>
#include <common/daemonize.h>
#include <common/config/session-config.h>
+#include <common/dynamic-buffer.h>
#include "lttng-sessiond.h"
#include "buffer-registry.h"
static pid_t child_ppid; /* Internal parent PID use with daemonize. */
static char *rundir;
static int lockfile_fd = -1;
+static int opt_print_version;
/* Set to 1 when a SIGUSR1 signal is received. */
static int recv_child_signal;
struct lttng_ht *agent_apps_ht_by_sock = NULL;
/*
- * Whether sessiond is ready for commands/health check requests.
- * NR_LTTNG_SESSIOND_READY must match the number of calls to
- * sessiond_notify_ready().
+ * The initialization of the session daemon is done in multiple phases.
+ *
+ * While all threads are launched near-simultaneously, only some of them
+ * are needed to ensure the session daemon can start to respond to client
+ * requests.
+ *
+ * There are two important guarantees that we wish to offer with respect
+ * to the initialisation of the session daemon:
+ * - When the daemonize/background launcher process exits, the sessiond
+ * is fully able to respond to client requests,
+ * - Auto-loaded sessions are visible to clients.
+ *
+ * In order to achieve this, a number of support threads have to be launched
+ * to allow the "client" thread to function properly. Moreover, since the
+ * "load session" thread needs the client thread, we must provide a way
+ * for the "load session" thread to know that the "client" thread is up
+ * and running.
+ *
+ * Hence, the support threads decrement the lttng_sessiond_ready counter
+ * while the "client" threads waits for it to reach 0. Once the "client" thread
+ * unblocks, it posts the message_thread_ready semaphore which allows the
+ * "load session" thread to progress.
+ *
+ * This implies that the "load session" thread is the last to be initialized
+ * and will explicitly call sessiond_signal_parents(), which signals the parents
+ * that the session daemon is fully initialized.
+ *
+ * The two (2) support threads are:
+ * - agent_thread
+ * - health_thread
*/
-#define NR_LTTNG_SESSIOND_READY 3
-int lttng_sessiond_ready = NR_LTTNG_SESSIOND_READY;
+int lttng_sessiond_ready = 2;
int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
{
/* Notify parents that we are ready for cmd and health check */
LTTNG_HIDDEN
-void sessiond_notify_ready(void)
+void sessiond_signal_parents(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 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);
- }
+ /*
+ * Notify the parent of the fork() process that we are
+ * ready.
+ */
+ if (opt_daemon || opt_background) {
+ kill(child_ppid, SIGUSR1);
}
}
+LTTNG_HIDDEN
+void sessiond_notify_ready(void)
+{
+ /*
+ * The _return variant is used since the implied memory barriers are
+ * required.
+ */
+ (void) uatomic_sub_return(<tng_sessiond_ready, 1);
+}
+
static
void setup_consumerd_path(void)
{
ret = waitpid(consumer_data->pid, &status, 0);
if (ret == -1) {
PERROR("consumerd waitpid pid: %d", consumer_data->pid)
- }
- if (!WIFEXITED(status)) {
+ } else if (!WIFEXITED(status)) {
ERR("consumerd termination with error: %d",
WEXITSTATUS(ret));
}
.count = 0,
};
+ rcu_register_thread();
+
health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
if (testpoint(sessiond_thread_app_reg_dispatch)) {
DBG("[thread] Dispatch UST command started");
- while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
+ for (;;) {
health_code_update();
/* Atomically prepare the queue futex */
futex_nto1_prepare(&ust_cmd_queue.futex);
+ if (CMM_LOAD_SHARED(dispatch_thread_exit)) {
+ break;
+ }
+
do {
struct ust_app *app = NULL;
ust_cmd = NULL;
ERR("Health error occurred in %s", __func__);
}
health_unregister(health_sessiond);
+ rcu_unregister_thread();
return NULL;
}
* lttcomm_setsockopt_snd_timeout expect msec as
* parameter.
*/
- (void) lttcomm_setsockopt_rcv_timeout(sock,
- app_socket_timeout * 1000);
- (void) lttcomm_setsockopt_snd_timeout(sock,
- app_socket_timeout * 1000);
+ if (app_socket_timeout >= 0) {
+ (void) lttcomm_setsockopt_rcv_timeout(sock,
+ app_socket_timeout * 1000);
+ (void) lttcomm_setsockopt_snd_timeout(sock,
+ app_socket_timeout * 1000);
+ }
/*
* Set the CLOEXEC flag. Return code is useless because
} else {
DBG("Could not find any valid consumerd executable");
ret = -EINVAL;
- break;
+ goto error;
}
DBG("Using kernel consumer at: %s", consumer_to_use);
ret = execl(consumer_to_use,
break;
}
default:
- PERROR("unknown consumer type");
+ ERR("unknown consumer type");
exit(EXIT_FAILURE);
}
if (errno != 0) {
kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
if (kernel_tracer_fd < 0) {
DBG("Failed to open %s", module_proc_lttng);
- ret = -1;
goto error_open;
}
case LTTNG_LIST_CHANNELS:
case LTTNG_LIST_EVENTS:
case LTTNG_LIST_SYSCALLS:
- case LTTNG_LIST_TRACKER_PIDS:
+ case LTTNG_LIST_TRACKER_IDS:
case LTTNG_DATA_PENDING:
break;
default:
case LTTNG_CREATE_SESSION:
case LTTNG_CREATE_SESSION_SNAPSHOT:
case LTTNG_CREATE_SESSION_LIVE:
- case LTTNG_CALIBRATE:
case LTTNG_LIST_SESSIONS:
case LTTNG_LIST_TRACEPOINTS:
case LTTNG_LIST_SYSCALLS:
&cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
break;
}
- case LTTNG_TRACK_PID:
+ case LTTNG_TRACK_ID:
{
- ret = cmd_track_pid(cmd_ctx->session,
+ struct lttng_tracker_id id;
+
+ memset(&id, 0, sizeof(id));
+ id.type = cmd_ctx->lsm->u.id_tracker.id_type;
+ switch (id.type) {
+ case LTTNG_ID_ALL:
+ break;
+ case LTTNG_ID_VALUE:
+ id.value = cmd_ctx->lsm->u.id_tracker.u.value;
+ break;
+ case LTTNG_ID_STRING:
+ {
+ size_t var_len = cmd_ctx->lsm->u.id_tracker.u.var_len;
+
+ id.string = zmalloc(var_len);
+ if (!id.string) {
+ ret = LTTNG_ERR_NOMEM;
+ goto error;
+ }
+ DBG("Receiving var len tracker id string from client.");
+ ret = lttcomm_recv_unix_sock(sock, id.string, var_len);
+ if (ret <= 0) {
+ DBG("Nothing received.");
+ *sock_error = 1;
+ free(id.string);
+ ret = LTTNG_ERR_INVALID;
+ goto error;
+ }
+ if (strnlen(id.string, var_len) != var_len - 1) {
+ DBG("Corrupted string.");
+ free(id.string);
+ ret = LTTNG_ERR_INVALID;
+ goto error;
+ }
+ break;
+ }
+ default:
+ ret = LTTNG_ERR_INVALID;
+ goto error;
+ }
+ ret = cmd_track_id(cmd_ctx->session,
+ cmd_ctx->lsm->u.id_tracker.tracker_type,
cmd_ctx->lsm->domain.type,
- cmd_ctx->lsm->u.pid_tracker.pid);
+ &id);
+ free(id.string);
break;
}
- case LTTNG_UNTRACK_PID:
+ case LTTNG_UNTRACK_ID:
{
- ret = cmd_untrack_pid(cmd_ctx->session,
+ struct lttng_tracker_id id;
+
+ memset(&id, 0, sizeof(id));
+ id.type = cmd_ctx->lsm->u.id_tracker.id_type;
+ switch (id.type) {
+ case LTTNG_ID_ALL:
+ break;
+ case LTTNG_ID_VALUE:
+ id.value = cmd_ctx->lsm->u.id_tracker.u.value;
+ break;
+ case LTTNG_ID_STRING:
+ {
+ size_t var_len = cmd_ctx->lsm->u.id_tracker.u.var_len;
+
+ id.string = zmalloc(var_len);
+ if (!id.string) {
+ ret = LTTNG_ERR_NOMEM;
+ goto error;
+ }
+ DBG("Receiving var len tracker id string from client.");
+ ret = lttcomm_recv_unix_sock(sock, id.string, var_len);
+ if (ret <= 0) {
+ DBG("Nothing received.");
+ *sock_error = 1;
+ free(id.string);
+ ret = LTTNG_ERR_INVALID;
+ goto error;
+ }
+ if (strnlen(id.string, var_len) != var_len - 1) {
+ DBG("Corrupted string.");
+ free(id.string);
+ ret = LTTNG_ERR_INVALID;
+ goto error;
+ }
+ break;
+ }
+ default:
+ ret = LTTNG_ERR_INVALID;
+ goto error;
+ }
+ ret = cmd_untrack_id(cmd_ctx->session,
+ cmd_ctx->lsm->u.id_tracker.tracker_type,
cmd_ctx->lsm->domain.type,
- cmd_ctx->lsm->u.pid_tracker.pid);
+ &id);
+ free(id.string);
break;
}
case LTTNG_ENABLE_EVENT:
ret = LTTNG_OK;
break;
}
- case LTTNG_LIST_TRACKER_PIDS:
+ case LTTNG_LIST_TRACKER_IDS:
{
- int32_t *pids = NULL;
- ssize_t nr_pids;
-
- nr_pids = cmd_list_tracker_pids(cmd_ctx->session,
- cmd_ctx->lsm->domain.type, &pids);
- if (nr_pids < 0) {
+ struct lttcomm_tracker_command_header cmd_header;
+ struct lttng_tracker_id *ids = NULL;
+ ssize_t nr_ids, i;
+ struct lttng_dynamic_buffer buf;
+
+ nr_ids = cmd_list_tracker_ids(cmd_ctx->lsm->u.id_tracker.tracker_type,
+ cmd_ctx->session,
+ cmd_ctx->lsm->domain.type, &ids);
+ if (nr_ids < 0) {
/* Return value is a negative lttng_error_code. */
- ret = -nr_pids;
+ ret = -nr_ids;
goto error;
}
- /*
- * Setup lttng message with payload size set to the event list size in
- * bytes and then copy list into the llm payload.
- */
- ret = setup_lttng_msg_no_cmd_header(cmd_ctx, pids,
- sizeof(int32_t) * nr_pids);
- free(pids);
+ lttng_dynamic_buffer_init(&buf);
+ for (i = 0; i < nr_ids; i++) {
+ struct lttng_tracker_id *id = &ids[i];
+ struct lttcomm_tracker_id_header id_hdr;
+ size_t var_data_len = 0;
+ memset(&id_hdr, 0, sizeof(id_hdr));
+ id_hdr.type = id->type;
+ switch (id->type) {
+ case LTTNG_ID_ALL:
+ break;
+ case LTTNG_ID_VALUE:
+ id_hdr.u.value = id->value;
+ break;
+ case LTTNG_ID_STRING:
+ id_hdr.u.var_data_len = var_data_len = strlen(id->string) + 1;
+ break;
+ default:
+ ret = LTTNG_ERR_INVALID;
+ goto error;
+ }
+ ret = lttng_dynamic_buffer_append(&buf, &id_hdr, sizeof(id_hdr));
+ if (ret) {
+ ret = LTTNG_ERR_NOMEM;
+ goto error;
+ }
+ ret = lttng_dynamic_buffer_append(&buf, id->string, var_data_len);
+ if (ret) {
+ ret = LTTNG_ERR_NOMEM;
+ goto error;
+ }
+ free(id->string);
+ }
+
+ cmd_header.nb_tracker_id = nr_ids;
+ ret = setup_lttng_msg(cmd_ctx, buf.data, buf.size, &cmd_header,
+ sizeof(cmd_header));
+ free(ids);
+ lttng_dynamic_buffer_reset(&buf);
if (ret < 0) {
goto setup_error;
}
ret = LTTNG_OK;
break;
}
- case LTTNG_CALIBRATE:
- {
- ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
- &cmd_ctx->lsm->u.calibrate);
- break;
- }
case LTTNG_REGISTER_CONSUMER:
{
struct consumer_data *cdata;
}
lttng_poll_clean(&events);
-
+ stop_threads();
rcu_unregister_thread();
return NULL;
}
goto error;
}
- sessiond_notify_ready();
ret = sem_post(&load_info->message_thread_ready);
if (ret) {
PERROR("sem_post message_thread_ready");
goto error;
}
+ /*
+ * Wait until all support threads are initialized before accepting
+ * commands.
+ */
+ while (uatomic_read(<tng_sessiond_ready) != 0) {
+ fd_set read_fds;
+ struct timeval timeout;
+
+ FD_ZERO(&read_fds);
+ FD_SET(thread_quit_pipe[0], &read_fds);
+ memset(&timeout, 0, sizeof(timeout));
+ timeout.tv_usec = 1000;
+
+ /*
+ * If a support thread failed to launch, it may signal that
+ * we must exit and the sessiond would never be marked as
+ * "ready".
+ *
+ * The timeout is set to 1ms, which serves as a way to
+ * pace down this check.
+ */
+ ret = select(thread_quit_pipe[0] + 1, &read_fds, NULL, NULL,
+ &timeout);
+ if (ret > 0 || (ret < 0 && errno != EINTR)) {
+ goto exit;
+ }
+ }
+ /*
+ * This barrier is paired with the one in sessiond_notify_ready() to
+ * ensure that loads accessing data initialized by the other threads,
+ * on which this thread was waiting, are not performed before this point.
+ *
+ * Note that this could be a 'read' memory barrier, but a full barrier
+ * is used in case the code changes. The performance implications of
+ * this choice are minimal since this is a slow path.
+ */
+ cmm_smp_mb();
+
/* This testpoint is after we signal readiness to the parent. */
if (testpoint(sessiond_thread_manage_clients)) {
goto error;
}
exit(ret ? EXIT_FAILURE : EXIT_SUCCESS);
} else if (string_match(optname, "version") || opt == 'V') {
- fprintf(stdout, "%s\n", VERSION);
- exit(EXIT_SUCCESS);
+ opt_print_version = 1;
} else if (string_match(optname, "sig-parent") || opt == 'S') {
opt_sig_parent = 1;
} else if (string_match(optname, "kconsumerd-err-sock")) {
return ret;
}
+static void sessiond_config_log(void)
+{
+ DBG("LTTng-sessiond " VERSION " - " VERSION_NAME "%s%s",
+ GIT_VERSION[0] == '\0' ? "" : " - " GIT_VERSION,
+ EXTRA_VERSION_NAME[0] == '\0' ? "" : " - " EXTRA_VERSION_NAME);
+ if (EXTRA_VERSION_DESCRIPTION[0] != '\0') {
+ DBG("LTTng-sessiond extra version description:\n\t" EXTRA_VERSION_DESCRIPTION "\n");
+ }
+ if (EXTRA_VERSION_PATCHES[0] != '\0') {
+ DBG("LTTng-sessiond extra patches:\n\t" EXTRA_VERSION_PATCHES "\n");
+ }
+}
+
+static void print_version(void) {
+ fprintf(stdout, "%s\n", VERSION);
+}
+
/*
* daemon configuration loading and argument parsing
*/
static void sighandler(int sig)
{
switch (sig) {
- case SIGPIPE:
- DBG("SIGPIPE caught");
- return;
case SIGINT:
DBG("SIGINT caught");
stop_threads();
return ret;
}
- sa.sa_handler = sighandler;
sa.sa_mask = sigset;
sa.sa_flags = 0;
+
+ sa.sa_handler = sighandler;
if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
PERROR("sigaction");
return ret;
return ret;
}
- if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
+ if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
PERROR("sigaction");
return ret;
}
- if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
+ sa.sa_handler = SIG_IGN;
+ if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
PERROR("sigaction");
return ret;
}
return ret;
}
+static int set_clock_plugin_env(void)
+{
+ int ret = 0;
+ const char *original_env_value;
+ char *full_path = NULL;
+ char *new_env_value = NULL;
+
+ original_env_value = getenv("LTTNG_UST_CLOCK_PLUGIN");
+ if (!original_env_value) {
+ goto end;
+ }
+
+ full_path = utils_expand_path(original_env_value);
+ if (!full_path) {
+ ERR("Failed to expand LTTNG_UST_CLOCK_PLUGIN path \"%s\"",
+ original_env_value);
+ ret = -1;
+ goto end;
+ }
+ ret = asprintf(&new_env_value, "LTTNG_UST_CLOCK_PLUGIN=%s",
+ full_path);
+ free(full_path);
+ if (ret < 0) {
+ PERROR("asprintf");
+ goto end;
+ }
+
+ DBG("Updating environment: %s", new_env_value);
+ ret = putenv(new_env_value);
+ if (ret) {
+ free(new_env_value);
+ PERROR("putenv of LTTNG_UST_CLOCK_PLUGIN");
+ goto end;
+ }
+end:
+ return ret;
+}
+
/*
* main
*/
goto exit_options;
}
+ sessiond_config_log();
+
+ if (opt_print_version) {
+ print_version();
+ retval = 0;
+ goto exit_options;
+ }
+
+ ret = set_clock_plugin_env();
+ if (ret) {
+ retval = -1;
+ goto exit_options;
+ }
+
/* Daemonize */
if (opt_daemon || opt_background) {
int i;
exit_health:
exit_init_data:
+ /*
+ * Wait for all pending call_rcu work to complete before tearing
+ * down data structures. call_rcu worker may be trying to
+ * perform lookups in those structures.
+ */
+ rcu_barrier();
/*
* sessiond_cleanup() is called when no other thread is running, except
* the ht_cleanup thread, which is needed to destroy the hash tables.