void cgroup_bonding_free(CGroupBonding *b, bool trim) {
assert(b);
if (b->unit) {
CGroupBonding *f;
LIST_REMOVE(CGroupBonding, by_unit, b->unit->cgroup_bondings, b);
if (streq(b->controller, SYSTEMD_CGROUP_CONTROLLER)) {
assert_se(f = hashmap_get(b->unit->manager->cgroup_bondings, b->path));
LIST_REMOVE(CGroupBonding, by_path, f, b);
if (f)
hashmap_replace(b->unit->manager->cgroup_bondings, b->path, f);
else
hashmap_remove(b->unit->manager->cgroup_bondings, b->path);
}
}
if (b->realized && b->ours && trim)
cg_trim(b->controller, b->path, false);
free(b->controller);
free(b->path);
free(b);
}
void cgroup_bonding_trim(CGroupBonding *b, bool delete_root) {
assert(b);
if (b->realized && b->ours)
cg_trim(b->controller, b->path, delete_root);
}
int main(int argc, char*argv[]) {
char *path;
char *c, *p;
assert_se(cg_create(SYSTEMD_CGROUP_CONTROLLER, "/test-a") == 0);
assert_se(cg_create(SYSTEMD_CGROUP_CONTROLLER, "/test-a") == 0);
assert_se(cg_create(SYSTEMD_CGROUP_CONTROLLER, "/test-b") == 0);
assert_se(cg_create(SYSTEMD_CGROUP_CONTROLLER, "/test-b/test-c") == 0);
assert_se(cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, "/test-b", 0) == 0);
assert_se(cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, getpid(), &path) == 0);
assert_se(streq(path, "/test-b"));
free(path);
assert_se(cg_attach(SYSTEMD_CGROUP_CONTROLLER, "/test-a", 0) == 0);
assert_se(cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, getpid(), &path) == 0);
assert_se(path_equal(path, "/test-a"));
free(path);
assert_se(cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, "/test-b/test-d", 0) == 0);
assert_se(cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, getpid(), &path) == 0);
assert_se(path_equal(path, "/test-b/test-d"));
free(path);
assert_se(cg_get_path(SYSTEMD_CGROUP_CONTROLLER, "/test-b/test-d", NULL, &path) == 0);
assert_se(path_equal(path, "/sys/fs/cgroup/systemd/test-b/test-d"));
free(path);
assert_se(cg_is_empty(SYSTEMD_CGROUP_CONTROLLER, "/test-a", false) > 0);
assert_se(cg_is_empty(SYSTEMD_CGROUP_CONTROLLER, "/test-b", false) > 0);
assert_se(cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, "/test-a", false) > 0);
assert_se(cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, "/test-b", false) == 0);
assert_se(cg_kill_recursive(SYSTEMD_CGROUP_CONTROLLER, "/test-a", 0, false, false, false, NULL) == 0);
assert_se(cg_kill_recursive(SYSTEMD_CGROUP_CONTROLLER, "/test-b", 0, false, false, false, NULL) > 0);
assert_se(cg_migrate_recursive(SYSTEMD_CGROUP_CONTROLLER, "/test-b", SYSTEMD_CGROUP_CONTROLLER, "/test-a", false, false) > 0);
assert_se(cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, "/test-a", false) == 0);
assert_se(cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, "/test-b", false) > 0);
assert_se(cg_kill_recursive(SYSTEMD_CGROUP_CONTROLLER, "/test-a", 0, false, false, false, NULL) > 0);
assert_se(cg_kill_recursive(SYSTEMD_CGROUP_CONTROLLER, "/test-b", 0, false, false, false, NULL) == 0);
cg_trim(SYSTEMD_CGROUP_CONTROLLER, "/", false);
assert_se(cg_delete(SYSTEMD_CGROUP_CONTROLLER, "/test-b") < 0);
assert_se(cg_delete(SYSTEMD_CGROUP_CONTROLLER, "/test-a") >= 0);
assert_se(cg_split_spec("foobar:/", &c, &p) == 0);
assert(streq(c, "foobar"));
assert(streq(p, "/"));
free(c);
free(p);
assert_se(cg_split_spec("foobar:", &c, &p) < 0);
assert_se(cg_split_spec("foobar:asdfd", &c, &p) < 0);
assert_se(cg_split_spec(":///", &c, &p) < 0);
assert_se(cg_split_spec(":", &c, &p) < 0);
assert_se(cg_split_spec("", &c, &p) < 0);
assert_se(cg_split_spec("fo/obar:/", &c, &p) < 0);
assert_se(cg_split_spec("/", &c, &p) >= 0);
assert(c == NULL);
assert(streq(p, "/"));
free(p);
assert_se(cg_split_spec("foo", &c, &p) >= 0);
assert(streq(c, "foo"));
assert(p == NULL);
free(c);
return 0;
}
int main(int argc, char *argv[]) {
bool need_umount, need_swapoff, need_loop_detach, need_dm_detach;
bool in_container, use_watchdog = false;
_cleanup_free_ char *cgroup = NULL;
char *arguments[3];
unsigned retries;
int cmd, r;
static const char* const dirs[] = {SYSTEM_SHUTDOWN_PATH, NULL};
log_parse_environment();
r = parse_argv(argc, argv);
if (r < 0)
goto error;
/* journald will die if not gone yet. The log target defaults
* to console, but may have been changed by command line options. */
log_close_console(); /* force reopen of /dev/console */
log_open();
umask(0022);
if (getpid() != 1) {
log_error("Not executed by init (PID 1).");
r = -EPERM;
goto error;
}
if (streq(arg_verb, "reboot"))
cmd = RB_AUTOBOOT;
else if (streq(arg_verb, "poweroff"))
cmd = RB_POWER_OFF;
else if (streq(arg_verb, "halt"))
cmd = RB_HALT_SYSTEM;
else if (streq(arg_verb, "kexec"))
cmd = LINUX_REBOOT_CMD_KEXEC;
else if (streq(arg_verb, "exit"))
cmd = 0; /* ignored, just checking that arg_verb is valid */
else {
r = -EINVAL;
log_error("Unknown action '%s'.", arg_verb);
goto error;
}
cg_get_root_path(&cgroup);
use_watchdog = !!getenv("WATCHDOG_USEC");
/* lock us into memory */
mlockall(MCL_CURRENT|MCL_FUTURE);
log_info("Sending SIGTERM to remaining processes...");
broadcast_signal(SIGTERM, true, true);
log_info("Sending SIGKILL to remaining processes...");
broadcast_signal(SIGKILL, true, false);
in_container = detect_container() > 0;
need_umount = !in_container;
need_swapoff = !in_container;
need_loop_detach = !in_container;
need_dm_detach = !in_container;
/* Unmount all mountpoints, swaps, and loopback devices */
for (retries = 0; retries < FINALIZE_ATTEMPTS; retries++) {
bool changed = false;
if (use_watchdog)
watchdog_ping();
/* Let's trim the cgroup tree on each iteration so
that we leave an empty cgroup tree around, so that
container managers get a nice notify event when we
are down */
if (cgroup)
cg_trim(SYSTEMD_CGROUP_CONTROLLER, cgroup, false);
if (need_umount) {
log_info("Unmounting file systems.");
r = umount_all(&changed);
if (r == 0) {
need_umount = false;
log_info("All filesystems unmounted.");
} else if (r > 0)
log_info("Not all file systems unmounted, %d left.", r);
else
log_error_errno(r, "Failed to unmount file systems: %m");
}
if (need_swapoff) {
log_info("Deactivating swaps.");
r = swapoff_all(&changed);
if (r == 0) {
need_swapoff = false;
log_info("All swaps deactivated.");
} else if (r > 0)
log_info("Not all swaps deactivated, %d left.", r);
else
log_error_errno(r, "Failed to deactivate swaps: %m");
}
if (need_loop_detach) {
log_info("Detaching loop devices.");
r = loopback_detach_all(&changed);
if (r == 0) {
need_loop_detach = false;
log_info("All loop devices detached.");
} else if (r > 0)
log_info("Not all loop devices detached, %d left.", r);
else
log_error_errno(r, "Failed to detach loop devices: %m");
}
if (need_dm_detach) {
log_info("Detaching DM devices.");
r = dm_detach_all(&changed);
if (r == 0) {
need_dm_detach = false;
log_info("All DM devices detached.");
} else if (r > 0)
log_info("Not all DM devices detached, %d left.", r);
else
log_error_errno(r, "Failed to detach DM devices: %m");
}
if (!need_umount && !need_swapoff && !need_loop_detach && !need_dm_detach) {
if (retries > 0)
log_info("All filesystems, swaps, loop devices, DM devices detached.");
/* Yay, done */
goto initrd_jump;
}
/* If in this iteration we didn't manage to
* unmount/deactivate anything, we simply give up */
if (!changed) {
log_info("Cannot finalize remaining%s%s%s%s continuing.",
need_umount ? " file systems," : "",
need_swapoff ? " swap devices," : "",
need_loop_detach ? " loop devices," : "",
need_dm_detach ? " DM devices," : "");
goto initrd_jump;
}
log_debug("After %u retries, couldn't finalize remaining %s%s%s%s trying again.",
retries + 1,
need_umount ? " file systems," : "",
need_swapoff ? " swap devices," : "",
need_loop_detach ? " loop devices," : "",
need_dm_detach ? " DM devices," : "");
}
log_error("Too many iterations, giving up.");
initrd_jump:
arguments[0] = NULL;
arguments[1] = arg_verb;
arguments[2] = NULL;
execute_directories(dirs, DEFAULT_TIMEOUT_USEC, arguments);
if (!in_container && !in_initrd() &&
access("/run/initramfs/shutdown", X_OK) == 0) {
r = switch_root_initramfs();
if (r >= 0) {
argv[0] = (char*) "/shutdown";
setsid();
make_console_stdio();
log_info("Successfully changed into root pivot.\n"
"Returning to initrd...");
execv("/shutdown", argv);
log_error_errno(errno, "Failed to execute shutdown binary: %m");
} else
log_error_errno(r, "Failed to switch root to \"/run/initramfs\": %m");
}
if (need_umount || need_swapoff || need_loop_detach || need_dm_detach)
log_error("Failed to finalize %s%s%s%s ignoring",
need_umount ? " file systems," : "",
need_swapoff ? " swap devices," : "",
need_loop_detach ? " loop devices," : "",
need_dm_detach ? " DM devices," : "");
/* The kernel will automaticall flush ATA disks and suchlike
* on reboot(), but the file systems need to be synce'd
* explicitly in advance. So let's do this here, but not
* needlessly slow down containers. */
if (!in_container)
sync();
if (streq(arg_verb, "exit")) {
if (in_container)
exit(arg_exit_code);
else {
/* We cannot exit() on the host, fallback on another
* method. */
cmd = RB_POWER_OFF;
}
}
switch (cmd) {
case LINUX_REBOOT_CMD_KEXEC:
if (!in_container) {
/* We cheat and exec kexec to avoid doing all its work */
pid_t pid;
log_info("Rebooting with kexec.");
pid = fork();
if (pid < 0)
log_error_errno(errno, "Failed to fork: %m");
else if (pid == 0) {
const char * const args[] = {
KEXEC, "-e", NULL
};
/* Child */
execv(args[0], (char * const *) args);
_exit(EXIT_FAILURE);
} else
wait_for_terminate_and_warn("kexec", pid, true);
}
cmd = RB_AUTOBOOT;
/* Fall through */
case RB_AUTOBOOT:
if (!in_container) {
_cleanup_free_ char *param = NULL;
if (read_one_line_file(REBOOT_PARAM_FILE, ¶m) >= 0) {
log_info("Rebooting with argument '%s'.", param);
syscall(SYS_reboot, LINUX_REBOOT_MAGIC1, LINUX_REBOOT_MAGIC2, LINUX_REBOOT_CMD_RESTART2, param);
}
}
log_info("Rebooting.");
break;
case RB_POWER_OFF:
log_info("Powering off.");
break;
case RB_HALT_SYSTEM:
log_info("Halting system.");
break;
default:
assert_not_reached("Unknown magic");
}
reboot(cmd);
if (errno == EPERM && in_container) {
/* If we are in a container, and we lacked
* CAP_SYS_BOOT just exit, this will kill our
* container for good. */
log_info("Exiting container.");
exit(0);
}
r = log_error_errno(errno, "Failed to invoke reboot(): %m");
error:
log_emergency_errno(r, "Critical error while doing system shutdown: %m");
freeze();
}
int main(int argc, char *argv[]) {
bool need_umount, need_swapoff, need_loop_detach, need_dm_detach;
bool in_container, use_watchdog = false, can_initrd;
_cleanup_free_ char *cgroup = NULL;
char *arguments[3];
int cmd, r, umount_log_level = LOG_INFO;
static const char* const dirs[] = {SYSTEM_SHUTDOWN_PATH, NULL};
char *watchdog_device;
/* The log target defaults to console, but the original systemd process will pass its log target in through a
* command line argument, which will override this default. Also, ensure we'll never log to the journal or
* syslog, as these logging daemons are either already dead or will die very soon. */
log_set_target(LOG_TARGET_CONSOLE);
log_set_prohibit_ipc(true);
log_parse_environment();
r = parse_argv(argc, argv);
if (r < 0)
goto error;
log_open();
umask(0022);
if (getpid_cached() != 1) {
log_error("Not executed by init (PID 1).");
r = -EPERM;
goto error;
}
if (streq(arg_verb, "reboot"))
cmd = RB_AUTOBOOT;
else if (streq(arg_verb, "poweroff"))
cmd = RB_POWER_OFF;
else if (streq(arg_verb, "halt"))
cmd = RB_HALT_SYSTEM;
else if (streq(arg_verb, "kexec"))
cmd = LINUX_REBOOT_CMD_KEXEC;
else if (streq(arg_verb, "exit"))
cmd = 0; /* ignored, just checking that arg_verb is valid */
else {
log_error("Unknown action '%s'.", arg_verb);
r = -EINVAL;
goto error;
}
(void) cg_get_root_path(&cgroup);
in_container = detect_container() > 0;
use_watchdog = getenv("WATCHDOG_USEC");
watchdog_device = getenv("WATCHDOG_DEVICE");
if (watchdog_device) {
r = watchdog_set_device(watchdog_device);
if (r < 0)
log_warning_errno(r, "Failed to set watchdog device to %s, ignoring: %m",
watchdog_device);
}
/* Lock us into memory */
(void) mlockall(MCL_CURRENT|MCL_FUTURE);
/* Synchronize everything that is not written to disk yet at this point already. This is a good idea so that
* slow IO is processed here already and the final process killing spree is not impacted by processes
* desperately trying to sync IO to disk within their timeout. Do not remove this sync, data corruption will
* result. */
if (!in_container)
sync_with_progress();
disable_coredumps();
log_info("Sending SIGTERM to remaining processes...");
broadcast_signal(SIGTERM, true, true, arg_timeout);
log_info("Sending SIGKILL to remaining processes...");
broadcast_signal(SIGKILL, true, false, arg_timeout);
need_umount = !in_container;
need_swapoff = !in_container;
need_loop_detach = !in_container;
need_dm_detach = !in_container;
can_initrd = !in_container && !in_initrd() && access("/run/initramfs/shutdown", X_OK) == 0;
/* Unmount all mountpoints, swaps, and loopback devices */
for (;;) {
bool changed = false;
if (use_watchdog)
watchdog_ping();
/* Let's trim the cgroup tree on each iteration so
that we leave an empty cgroup tree around, so that
container managers get a nice notify event when we
are down */
if (cgroup)
cg_trim(SYSTEMD_CGROUP_CONTROLLER, cgroup, false);
if (need_umount) {
log_info("Unmounting file systems.");
r = umount_all(&changed, umount_log_level);
if (r == 0) {
need_umount = false;
log_info("All filesystems unmounted.");
} else if (r > 0)
log_info("Not all file systems unmounted, %d left.", r);
else
log_error_errno(r, "Failed to unmount file systems: %m");
}
if (need_swapoff) {
log_info("Deactivating swaps.");
r = swapoff_all(&changed);
if (r == 0) {
need_swapoff = false;
log_info("All swaps deactivated.");
} else if (r > 0)
log_info("Not all swaps deactivated, %d left.", r);
else
log_error_errno(r, "Failed to deactivate swaps: %m");
}
if (need_loop_detach) {
log_info("Detaching loop devices.");
r = loopback_detach_all(&changed, umount_log_level);
if (r == 0) {
need_loop_detach = false;
log_info("All loop devices detached.");
} else if (r > 0)
log_info("Not all loop devices detached, %d left.", r);
else
log_error_errno(r, "Failed to detach loop devices: %m");
}
if (need_dm_detach) {
log_info("Detaching DM devices.");
r = dm_detach_all(&changed, umount_log_level);
if (r == 0) {
need_dm_detach = false;
log_info("All DM devices detached.");
} else if (r > 0)
log_info("Not all DM devices detached, %d left.", r);
else
log_error_errno(r, "Failed to detach DM devices: %m");
}
if (!need_umount && !need_swapoff && !need_loop_detach && !need_dm_detach) {
log_info("All filesystems, swaps, loop devices and DM devices detached.");
/* Yay, done */
break;
}
if (!changed && umount_log_level == LOG_INFO && !can_initrd) {
/* There are things we cannot get rid of. Loop one more time
* with LOG_ERR to inform the user. Note that we don't need
* to do this if there is a initrd to switch to, because that
* one is likely to get rid of the remounting mounts. If not,
* it will log about them. */
umount_log_level = LOG_ERR;
continue;
}
/* If in this iteration we didn't manage to
* unmount/deactivate anything, we simply give up */
if (!changed) {
log_info("Cannot finalize remaining%s%s%s%s continuing.",
need_umount ? " file systems," : "",
need_swapoff ? " swap devices," : "",
need_loop_detach ? " loop devices," : "",
need_dm_detach ? " DM devices," : "");
break;
}
log_debug("Couldn't finalize remaining %s%s%s%s trying again.",
need_umount ? " file systems," : "",
need_swapoff ? " swap devices," : "",
need_loop_detach ? " loop devices," : "",
need_dm_detach ? " DM devices," : "");
}
/* We're done with the watchdog. */
watchdog_free_device();
arguments[0] = NULL;
arguments[1] = arg_verb;
arguments[2] = NULL;
execute_directories(dirs, DEFAULT_TIMEOUT_USEC, NULL, NULL, arguments, NULL);
(void) rlimit_nofile_safe();
if (can_initrd) {
r = switch_root_initramfs();
if (r >= 0) {
argv[0] = (char*) "/shutdown";
(void) setsid();
(void) make_console_stdio();
log_info("Successfully changed into root pivot.\n"
"Returning to initrd...");
execv("/shutdown", argv);
log_error_errno(errno, "Failed to execute shutdown binary: %m");
} else
log_error_errno(r, "Failed to switch root to \"/run/initramfs\": %m");
}
if (need_umount || need_swapoff || need_loop_detach || need_dm_detach)
log_error("Failed to finalize %s%s%s%s ignoring",
need_umount ? " file systems," : "",
need_swapoff ? " swap devices," : "",
need_loop_detach ? " loop devices," : "",
need_dm_detach ? " DM devices," : "");
/* The kernel will automatically flush ATA disks and suchlike on reboot(), but the file systems need to be
* sync'ed explicitly in advance. So let's do this here, but not needlessly slow down containers. Note that we
* sync'ed things already once above, but we did some more work since then which might have caused IO, hence
* let's do it once more. Do not remove this sync, data corruption will result. */
if (!in_container)
sync_with_progress();
if (streq(arg_verb, "exit")) {
if (in_container)
return arg_exit_code;
cmd = RB_POWER_OFF; /* We cannot exit() on the host, fallback on another method. */
}
switch (cmd) {
case LINUX_REBOOT_CMD_KEXEC:
if (!in_container) {
/* We cheat and exec kexec to avoid doing all its work */
log_info("Rebooting with kexec.");
r = safe_fork("(sd-kexec)", FORK_RESET_SIGNALS|FORK_CLOSE_ALL_FDS|FORK_LOG|FORK_WAIT, NULL);
if (r == 0) {
const char * const args[] = {
KEXEC, "-e", NULL
};
/* Child */
execv(args[0], (char * const *) args);
_exit(EXIT_FAILURE);
}
/* If we are still running, then the kexec can't have worked, let's fall through */
}
cmd = RB_AUTOBOOT;
_fallthrough_;
case RB_AUTOBOOT:
(void) reboot_with_parameter(REBOOT_LOG);
log_info("Rebooting.");
break;
case RB_POWER_OFF:
log_info("Powering off.");
break;
case RB_HALT_SYSTEM:
log_info("Halting system.");
break;
default:
assert_not_reached("Unknown magic");
}
(void) reboot(cmd);
if (errno == EPERM && in_container) {
/* If we are in a container, and we lacked
* CAP_SYS_BOOT just exit, this will kill our
* container for good. */
log_info("Exiting container.");
return EXIT_SUCCESS;
}
r = log_error_errno(errno, "Failed to invoke reboot(): %m");
error:
log_emergency_errno(r, "Critical error while doing system shutdown: %m");
freeze();
}