#include <sys/reboot.h>
#include <linux/reboot.h>
#include <sys/wait.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <sys/mount.h>
+#include <sys/syscall.h>
+#include <fcntl.h>
#include <dirent.h>
#include <errno.h>
#include <unistd.h>
#include <stdlib.h>
#include <string.h>
+#include "missing.h"
#include "log.h"
#include "umount.h"
#include "util.h"
+#include "virt.h"
+#include "watchdog.h"
#define TIMEOUT_USEC (5 * USEC_PER_SEC)
#define FINALIZE_ATTEMPTS 50
static bool ignore_proc(pid_t pid) {
+ char buf[PATH_MAX];
+ FILE *f;
+ char c;
+ size_t count;
+ uid_t uid;
+ int r;
+
+ /* We are PID 1, let's not commit suicide */
if (pid == 1)
return true;
- /* TODO: add more ignore rules here: device-mapper, etc */
+ r = get_process_uid(pid, &uid);
+ if (r < 0)
+ return true; /* not really, but better safe than sorry */
- return false;
-}
+ /* Non-root processes otherwise are always subject to be killed */
+ if (uid != 0)
+ return false;
-static bool is_kernel_thread(pid_t pid)
-{
- char buf[PATH_MAX];
- FILE *f;
- char c;
- size_t count;
+ snprintf(buf, sizeof(buf), "/proc/%lu/cmdline", (unsigned long) pid);
+ char_array_0(buf);
- snprintf(buf, sizeof(buf), "/proc/%lu/cmdline", (unsigned long)pid);
f = fopen(buf, "re");
if (!f)
return true; /* not really, but has the desired effect */
count = fread(&c, 1, 1, f);
fclose(f);
- return count != 1;
+
+ /* Kernel threads have an empty cmdline */
+ if (count <= 0)
+ return true;
+
+ /* Processes with argv[0][0] = '@' we ignore from the killing
+ * spree.
+ *
+ * http://www.freedesktop.org/wiki/Software/systemd/RootStorageDaemons */
+ if (count == 1 && c == '@')
+ return true;
+
+ return false;
}
static int killall(int sign) {
struct dirent *d;
unsigned int n_processes = 0;
- if ((dir = opendir("/proc")) == NULL)
+ dir = opendir("/proc");
+ if (!dir)
return -errno;
while ((d = readdir(dir))) {
if (parse_pid(d->d_name, &pid) < 0)
continue;
- if (is_kernel_thread(pid))
- continue;
-
if (ignore_proc(pid))
continue;
sigprocmask(SIG_SETMASK, &oldmask, NULL);
}
+static int prepare_new_root(void) {
+ static const char dirs[] =
+ "/run/initramfs/oldroot\0"
+ "/run/initramfs/proc\0"
+ "/run/initramfs/sys\0"
+ "/run/initramfs/dev\0"
+ "/run/initramfs/run\0";
+
+ const char *dir;
+
+ if (mount("/run/initramfs", "/run/initramfs", NULL, MS_BIND, NULL) < 0) {
+ log_error("Failed to mount bind /run/initramfs on /run/initramfs: %m");
+ return -errno;
+ }
+
+ if (mount(NULL, "/run/initramfs", NULL, MS_PRIVATE, NULL) < 0) {
+ log_error("Failed to make /run/initramfs private mount: %m");
+ return -errno;
+ }
+
+ NULSTR_FOREACH(dir, dirs)
+ if (mkdir_p(dir, 0755) < 0 && errno != EEXIST) {
+ log_error("Failed to mkdir %s: %m", dir);
+ return -errno;
+ }
+
+ if (mount("/sys", "/run/initramfs/sys", NULL, MS_BIND, NULL) < 0) {
+ log_error("Failed to mount bind /sys on /run/initramfs/sys: %m");
+ return -errno;
+ }
+
+ if (mount("/proc", "/run/initramfs/proc", NULL, MS_BIND, NULL) < 0) {
+ log_error("Failed to mount bind /proc on /run/initramfs/proc: %m");
+ return -errno;
+ }
+
+ if (mount("/dev", "/run/initramfs/dev", NULL, MS_BIND, NULL) < 0) {
+ log_error("Failed to mount bind /dev on /run/initramfs/dev: %m");
+ return -errno;
+ }
+
+ if (mount("/run", "/run/initramfs/run", NULL, MS_BIND, NULL) < 0) {
+ log_error("Failed to mount bind /run on /run/initramfs/run: %m");
+ return -errno;
+ }
+
+ return 0;
+}
+
+static int pivot_to_new_root(void) {
+ int fd;
+
+ chdir("/run/initramfs");
+
+ /*
+ In case some evil process made "/" MS_SHARED
+ It works for pivot_root, but the ref count for the root device
+ is not decreasing :-/
+ */
+ if (mount(NULL, "/", NULL, MS_PRIVATE, NULL) < 0) {
+ log_error("Failed to make \"/\" private mount %m");
+ return -errno;
+ }
+
+ if (pivot_root(".", "oldroot") < 0) {
+ log_error("pivot failed: %m");
+ /* only chroot if pivot root succeded */
+ return -errno;
+ }
+
+ chroot(".");
+ log_info("Successfully changed into root pivot.");
+
+ fd = open("/dev/console", O_RDWR);
+ if (fd < 0)
+ log_error("Failed to open /dev/console: %m");
+ else {
+ make_stdio(fd);
+
+ /* Initialize the controlling terminal */
+ setsid();
+ ioctl(STDIN_FILENO, TIOCSCTTY, NULL);
+ }
+
+ return 0;
+}
+
int main(int argc, char *argv[]) {
int cmd, r;
unsigned retries;
bool need_umount = true, need_swapoff = true, need_loop_detach = true, need_dm_detach = true;
- bool killed_everbody = false, in_container;
+ bool killed_everbody = false, in_container, use_watchdog = false;
log_parse_environment();
log_set_target(LOG_TARGET_CONSOLE); /* syslog will die if not gone yet */
log_open();
+ umask(0022);
+
if (getpid() != 1) {
log_error("Not executed by init (pid 1).");
r = -EPERM;
goto error;
}
+ use_watchdog = !!getenv("WATCHDOG_USEC");
+
/* lock us into memory */
if (mlockall(MCL_CURRENT|MCL_FUTURE) != 0)
log_warning("Cannot lock process memory: %m");
for (retries = 0; retries < FINALIZE_ATTEMPTS; retries++) {
bool changed = false;
+ if (use_watchdog)
+ watchdog_ping();
+
if (need_umount) {
log_info("Unmounting file systems.");
r = umount_all(&changed);
log_error("Failed to detach DM devices: %s", strerror(-r));
}
- if (!need_umount && !need_swapoff && !need_loop_detach && !need_dm_detach)
+ 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 */
break;
+ }
/* If in this iteration we didn't manage to
* unmount/deactivate anything, we either kill more
/* If we are in a container, just exit, this will kill our
* container for good. */
- if (in_container)
+ if (in_container) {
+ log_error("Exiting container.");
exit(0);
+ }
+
+ if (access("/run/initramfs/shutdown", X_OK) == 0) {
+
+ if (prepare_new_root() >= 0 &&
+ pivot_to_new_root() >= 0) {
+ execv("/shutdown", argv);
+ log_error("Failed to execute shutdown binary: %m");
+ }
+ }
sync();