#include "ust-consumer.h"
#include "utils.h"
#include "fd-limit.h"
-#include "health.h"
+#include "health-sessiond.h"
#include "testpoint.h"
#include "ust-thread.h"
+#include "jul-thread.h"
#define CONSUMERD_FILE "lttng-consumerd"
-/* Const values */
-const char default_tracing_group[] = DEFAULT_TRACING_GROUP;
-
const char *progname;
-const char *opt_tracing_group;
+static const char *tracing_group_name = DEFAULT_TRACING_GROUP;
static const char *opt_pidfile;
static int opt_sig_parent;
static int opt_verbose_consumer;
static pthread_t dispatch_thread;
static pthread_t health_thread;
static pthread_t ht_cleanup_thread;
+static pthread_t jul_reg_thread;
/*
* UST registration command queue. This queue is tied with a futex and uses a N
/* Set in main() with the current page size. */
long page_size;
+/* Application health monitoring */
+struct health_app *health_sessiond;
+
+/* JUL TCP port for registration. Used by the JUL thread. */
+unsigned int jul_tcp_port = DEFAULT_JUL_TCP_PORT;
+
static
void setup_consumerd_path(void)
{
return 0;
}
-/*
- * Return group ID of the tracing group or -1 if not found.
- */
-static gid_t allowed_group(void)
-{
- struct group *grp;
-
- if (opt_tracing_group) {
- grp = getgrnam(opt_tracing_group);
- } else {
- grp = getgrnam(default_tracing_group);
- }
- if (!grp) {
- return -1;
- } else {
- return grp->gr_gid;
- }
-}
-
/*
* Init thread quit pipe.
*
static void cleanup(void)
{
int ret;
- char *cmd = NULL;
struct ltt_session *sess, *stmp;
+ char path[PATH_MAX];
DBG("Cleaning up");
}
}
- DBG("Removing %s directory", rundir);
- ret = asprintf(&cmd, "rm -rf %s", rundir);
- if (ret < 0) {
- ERR("asprintf failed. Something is really wrong!");
- }
+ DBG("Removing sessiond and consumerd content of directory %s", rundir);
+
+ /* sessiond */
+ snprintf(path, PATH_MAX,
+ "%s/%s",
+ rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
+ DBG("Removing %s", path);
+ (void) unlink(path);
+
+ snprintf(path, PATH_MAX, "%s/%s", rundir,
+ DEFAULT_LTTNG_SESSIOND_JULPORT_FILE);
+ DBG("Removing %s", path);
+ (void) unlink(path);
+
+ /* kconsumerd */
+ snprintf(path, PATH_MAX,
+ DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
+ rundir);
+ DBG("Removing %s", path);
+ (void) unlink(path);
+
+ snprintf(path, PATH_MAX,
+ DEFAULT_KCONSUMERD_PATH,
+ rundir);
+ DBG("Removing directory %s", path);
+ (void) rmdir(path);
+
+ /* ust consumerd 32 */
+ snprintf(path, PATH_MAX,
+ DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
+ rundir);
+ DBG("Removing %s", path);
+ (void) unlink(path);
+
+ snprintf(path, PATH_MAX,
+ DEFAULT_USTCONSUMERD32_PATH,
+ rundir);
+ DBG("Removing directory %s", path);
+ (void) rmdir(path);
+
+ /* ust consumerd 64 */
+ snprintf(path, PATH_MAX,
+ DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
+ rundir);
+ DBG("Removing %s", path);
+ (void) unlink(path);
+
+ snprintf(path, PATH_MAX,
+ DEFAULT_USTCONSUMERD64_PATH,
+ rundir);
+ 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.
+ */
- /* Remove lttng run directory */
- ret = system(cmd);
- if (ret < 0) {
- ERR("Unable to clean %s", rundir);
- }
- free(cmd);
free(rundir);
DBG("Cleaning up all sessions");
DBG("[thread] Thread manage kernel started");
- health_register(HEALTH_TYPE_KERNEL);
+ health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
/*
* This first step of the while is to clean this structure which could free
WARN("Kernel thread died unexpectedly. "
"Kernel tracing can continue but CPU hotplug is disabled.");
}
- health_unregister();
+ health_unregister(health_sessiond);
DBG("Kernel thread dying");
return NULL;
}
DBG("[thread] Manage consumer started");
- health_register(HEALTH_TYPE_CONSUMER);
+ health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
health_code_update();
health_error();
ERR("Health error occurred in %s", __func__);
}
- health_unregister();
+ health_unregister(health_sessiond);
DBG("consumer thread cleanup completed");
return NULL;
rcu_register_thread();
rcu_thread_online();
- health_register(HEALTH_TYPE_APP_MANAGE);
+ health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
if (testpoint(thread_manage_apps)) {
goto error_testpoint;
goto error;
}
- /*
- * Set socket timeout for both receiving and ending.
- * app_socket_timeout is in seconds, whereas
- * lttcomm_setsockopt_rcv_timeout and
- * 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);
-
DBG("Apps with sock %d added to poll set", sock);
health_code_update();
health_error();
ERR("Health error occurred in %s", __func__);
}
- health_unregister();
+ health_unregister(health_sessiond);
DBG("Application communication apps thread cleanup complete");
rcu_thread_offline();
rcu_unregister_thread();
.count = 0,
};
- health_register(HEALTH_TYPE_APP_REG_DISPATCH);
+ health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
health_code_update();
if (ret < 0) {
PERROR("close ust sock dispatch %d", ust_cmd->sock);
}
- lttng_fd_put(1, LTTNG_FD_APPS);
+ lttng_fd_put(LTTNG_FD_APPS, 1);
free(ust_cmd);
goto error;
}
if (ret < 0) {
PERROR("close ust sock dispatch %d", ust_cmd->sock);
}
- lttng_fd_put(1, LTTNG_FD_APPS);
+ lttng_fd_put(LTTNG_FD_APPS, 1);
free(wait_node);
free(ust_cmd);
continue;
if (ret < 0) {
PERROR("close ust sock dispatch %d", ust_cmd->sock);
}
- lttng_fd_put(1, LTTNG_FD_APPS);
+ lttng_fd_put(LTTNG_FD_APPS, 1);
}
free(ust_cmd);
}
health_error();
ERR("Health error occurred in %s", __func__);
}
- health_unregister();
+ health_unregister(health_sessiond);
return NULL;
}
DBG("[thread] Manage application registration started");
- health_register(HEALTH_TYPE_APP_REG);
+ health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
if (testpoint(thread_registration_apps)) {
goto error_testpoint;
goto error;
}
+ /*
+ * Set socket timeout for both receiving and ending.
+ * app_socket_timeout is in seconds, whereas
+ * lttcomm_setsockopt_rcv_timeout and
+ * 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);
+
/*
* Set the CLOEXEC flag. Return code is useless because
* either way, the show must go on.
error_create_poll:
error_testpoint:
DBG("UST Registration thread cleanup complete");
- health_unregister();
+ health_unregister(health_sessiond);
return NULL;
}
"lttng-consumerd", verbosity, "-k",
"--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
"--consumerd-err-sock", consumer_data->err_unix_sock_path,
+ "--group", tracing_group_name,
NULL);
break;
case LTTNG_CONSUMER64_UST:
ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
"--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
"--consumerd-err-sock", consumer_data->err_unix_sock_path,
+ "--group", tracing_group_name,
NULL);
if (consumerd64_libdir[0] != '\0') {
free(tmpnew);
ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
"--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
"--consumerd-err-sock", consumer_data->err_unix_sock_path,
+ "--group", tracing_group_name,
NULL);
if (consumerd32_libdir[0] != '\0') {
free(tmpnew);
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(HEALTH_TYPE_CONSUMER);
-
- DBG3("Health consumer check %d", ret);
-
- return ret;
-}
-
/*
* Setup necessary data for kernel tracer action.
*/
consumer = session->kernel_session->consumer;
dir_name = DEFAULT_KERNEL_TRACE_DIR;
break;
+ case LTTNG_DOMAIN_JUL:
case LTTNG_DOMAIN_UST:
DBG3("Copying tracing session consumer output in UST session");
if (session->ust_session->consumer) {
assert(session->consumer);
switch (domain->type) {
+ case LTTNG_DOMAIN_JUL:
case LTTNG_DOMAIN_UST:
break;
default:
}
break;
+ case LTTNG_DOMAIN_JUL:
case LTTNG_DOMAIN_UST:
{
if (!ust_app_supported()) {
if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
switch (cmd_ctx->lsm->domain.type) {
+ case LTTNG_DOMAIN_JUL:
case LTTNG_DOMAIN_UST:
if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
ret = LTTNG_ERR_NO_USTCONSUMERD;
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;
+ struct health_comm_msg msg;
+ struct health_comm_reply reply;
DBG("[thread] Manage health check started");
goto error;
}
+ if (is_root) {
+ /* lttng health client socket path permissions */
+ ret = chown(health_unix_sock_path, 0,
+ utils_get_group_id(tracing_group_name));
+ if (ret < 0) {
+ ERR("Unable to set group on %s", health_unix_sock_path);
+ PERROR("chown");
+ ret = -1;
+ goto error;
+ }
+
+ ret = chmod(health_unix_sock_path,
+ S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
+ if (ret < 0) {
+ ERR("Unable to set permissions on %s", health_unix_sock_path);
+ PERROR("chmod");
+ ret = -1;
+ goto error;
+ }
+ }
+
/*
* Set the CLOEXEC flag. Return code is useless because either way, the
* show must go on.
rcu_thread_online();
- switch (msg.component) {
- case LTTNG_HEALTH_CMD:
- reply.ret_code = health_check_state(HEALTH_TYPE_CMD);
- break;
- case LTTNG_HEALTH_APP_MANAGE:
- reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE);
- break;
- case LTTNG_HEALTH_APP_REG:
- reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG);
- break;
- case LTTNG_HEALTH_KERNEL:
- reply.ret_code = health_check_state(HEALTH_TYPE_KERNEL);
- break;
- case LTTNG_HEALTH_CONSUMER:
- reply.ret_code = check_consumer_health();
- break;
- case LTTNG_HEALTH_HT_CLEANUP:
- reply.ret_code = health_check_state(HEALTH_TYPE_HT_CLEANUP);
- break;
- case LTTNG_HEALTH_APP_MANAGE_NOTIFY:
- reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY);
- break;
- case LTTNG_HEALTH_APP_REG_DISPATCH:
- reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG_DISPATCH);
- break;
- case LTTNG_HEALTH_ALL:
- reply.ret_code =
- health_check_state(HEALTH_TYPE_APP_MANAGE) &&
- health_check_state(HEALTH_TYPE_APP_REG) &&
- health_check_state(HEALTH_TYPE_CMD) &&
- health_check_state(HEALTH_TYPE_KERNEL) &&
- check_consumer_health() &&
- health_check_state(HEALTH_TYPE_HT_CLEANUP) &&
- health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY) &&
- health_check_state(HEALTH_TYPE_APP_REG_DISPATCH);
- break;
- default:
- reply.ret_code = LTTNG_ERR_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;
+ reply.ret_code = 0;
+ for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
+ /*
+ * health_check_state returns 0 if health is
+ * bad.
+ */
+ if (!health_check_state(health_sessiond, i)) {
+ reply.ret_code |= 1ULL << i;
+ }
}
- DBG2("Health check return value %d", reply.ret_code);
+ DBG2("Health check return value %" PRIx64, reply.ret_code);
ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
if (ret < 0) {
rcu_register_thread();
- health_register(HEALTH_TYPE_CMD);
+ health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
if (testpoint(thread_manage_clients)) {
goto error_testpoint;
ERR("Health error occurred in %s", __func__);
}
- health_unregister();
+ health_unregister(health_sessiond);
DBG("Client thread dying");
fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
fprintf(stderr, " --no-kernel Disable kernel tracer\n");
+ fprintf(stderr, " --jul-tcp-port JUL application registration TCP port\n");
}
/*
{ "verbose-consumer", 0, 0, 'Z' },
{ "no-kernel", 0, 0, 'N' },
{ "pidfile", 1, 0, 'p' },
+ { "jul-tcp-port", 1, 0, 'J' },
{ 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:",
+ c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:J:",
long_options, &option_index);
if (c == -1) {
break;
opt_daemon = 1;
break;
case 'g':
- opt_tracing_group = optarg;
+ tracing_group_name = optarg;
break;
case 'h':
usage();
case 'p':
opt_pidfile = optarg;
break;
+ case 'J': /* JUL TCP port. */
+ {
+ unsigned long v;
+
+ errno = 0;
+ v = strtoul(optarg, NULL, 0);
+ if (errno != 0 || !isdigit(optarg[0])) {
+ ERR("Wrong value in --jul-tcp-port parameter: %s", optarg);
+ return -1;
+ }
+ if (v == 0 || v >= 65535) {
+ ERR("Port overflow in --jul-tcp-port parameter: %s", optarg);
+ return -1;
+ }
+ jul_tcp_port = (uint32_t) v;
+ DBG3("JUL TCP port set to non default: %u", jul_tcp_port);
+ break;
+ }
default:
/* Unknown option or other error.
* Error is printed by getopt, just return */
int ret;
gid_t gid;
- ret = allowed_group();
- if (ret < 0) {
- WARN("No tracing group detected");
- /* Setting gid to 0 if no tracing group is found */
- gid = 0;
- } else {
- gid = ret;
- }
+ gid = utils_get_group_id(tracing_group_name);
/* Set lttng run dir */
ret = chown(rundir, 0, gid);
PERROR("chown");
}
- /* Ensure all applications and tracing group can search the run dir */
- ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH);
+ /*
+ * Ensure all applications and tracing group can search the run
+ * dir. Allow everyone to read the directory, since it does not
+ * buy us anything to hide its content.
+ */
+ ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
if (ret < 0) {
ERR("Unable to set permissions on %s", rundir);
PERROR("chmod");
}
/* kconsumer error socket path */
- ret = chown(kconsumer_data.err_unix_sock_path, 0, gid);
+ ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
if (ret < 0) {
ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
PERROR("chown");
}
/* 64-bit ustconsumer error socket path */
- ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid);
+ ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
if (ret < 0) {
ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
PERROR("chown");
}
/* 32-bit ustconsumer compat32 error socket path */
- ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid);
+ ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
if (ret < 0) {
ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
PERROR("chown");
int ret;
char path[PATH_MAX];
- switch (consumer_data->type) {
+ switch (consumer_data->type) {
case LTTNG_CONSUMER_KERNEL:
snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
break;
DBG2("Creating consumer directory: %s", path);
- ret = mkdir(path, S_IRWXU);
+ ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
if (ret < 0) {
if (errno != EEXIST) {
PERROR("mkdir");
}
ret = -1;
}
+ if (is_root) {
+ ret = chown(path, 0, utils_get_group_id(tracing_group_name));
+ if (ret < 0) {
+ ERR("Unable to set group on %s", path);
+ PERROR("chown");
+ goto error;
+ }
+ }
/* Create the kconsumerd error unix socket */
consumer_data->err_sock =
return;
}
+/*
+ * Write JUL TCP port using the rundir.
+ */
+static void write_julport(void)
+{
+ int ret;
+ char path[PATH_MAX];
+
+ assert(rundir);
+
+ ret = snprintf(path, sizeof(path), "%s/"
+ DEFAULT_LTTNG_SESSIOND_JULPORT_FILE, rundir);
+ if (ret < 0) {
+ PERROR("snprintf julport path");
+ goto error;
+ }
+
+ /*
+ * Create TCP JUL port file in rundir. Return value is of no importance.
+ * The execution will continue even though we are not able to write the
+ * file.
+ */
+ (void) utils_create_pid_file(jul_tcp_port, path);
+
+error:
+ return;
+}
+
/*
* main
*/
*/
ust_app_ht_alloc();
+ /* Initialize JUL domain subsystem. */
+ if ((ret = jul_init()) < 0) {
+ /* ENOMEM at this point. */
+ goto error;
+ }
+
/* After this point, we can safely call cleanup() with "goto exit" */
/*
}
write_pidfile();
+ write_julport();
/* Initialize communication library */
lttcomm_init();
* Initialize the health check subsystem. This call should set the
* appropriate time values.
*/
- health_init();
+ health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
+ if (!health_sessiond) {
+ PERROR("health_app_create error");
+ goto exit_health_sessiond_cleanup;
+ }
/* Create thread to manage the client socket */
ret = pthread_create(&ht_cleanup_thread, NULL,
goto exit_apps_notify;
}
+ /* Create JUL registration thread. */
+ ret = pthread_create(&jul_reg_thread, NULL,
+ jul_thread_manage_registration, (void *) NULL);
+ if (ret != 0) {
+ PERROR("pthread_create apps");
+ goto exit_jul_reg;
+ }
+
/* Don't start this thread if kernel tracing is not requested nor root */
if (is_root && !opt_no_kernel) {
/* Create kernel thread to manage kernel event */
}
exit_kernel:
+ ret = pthread_join(jul_reg_thread, &status);
+ if (ret != 0) {
+ PERROR("pthread_join JUL");
+ goto error; /* join error, exit without cleanup */
+ }
+
+exit_jul_reg:
ret = pthread_join(apps_notify_thread, &status);
if (ret != 0) {
PERROR("pthread_join apps notify");
goto error; /* join error, exit without cleanup */
}
+
exit_apps:
ret = pthread_join(reg_apps_thread, &status);
if (ret != 0) {
goto error; /* join error, exit without cleanup */
}
exit_ht_cleanup:
+ health_app_destroy(health_sessiond);
+exit_health_sessiond_cleanup:
exit:
/*
* cleanup() is called when no other thread is running.