X-Git-Url: https://www.chiark.greenend.org.uk/ucgi/~ianmdlvl/git?a=blobdiff_plain;f=src%2Fshutdown.c;h=9f65b1dab0b78aeed203717b75ba3852cc3ceb1e;hb=b8b5e648cacc8d73c55fdffbb3466ecd8146131a;hp=11c7544fc7c50cd45c42673edd006d25a1f2f4ba;hpb=91901329245f070b621a24577393fb8f4ce9bffc;p=elogind.git diff --git a/src/shutdown.c b/src/shutdown.c index 11c7544fc..9f65b1dab 100644 --- a/src/shutdown.c +++ b/src/shutdown.c @@ -24,6 +24,11 @@ #include #include #include +#include +#include +#include +#include +#include #include #include #include @@ -32,37 +37,58 @@ #include #include +#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) { @@ -70,7 +96,8 @@ 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))) { @@ -79,9 +106,6 @@ static int killall(int sign) { if (parse_pid(d->d_name, &pid) < 0) continue; - if (is_kernel_thread(pid)) - continue; - if (ignore_proc(pid)) continue; @@ -96,130 +120,201 @@ static int killall(int sign) { return n_processes; } -static int send_signal(int sign) { - sigset_t mask, oldmask; +static void wait_for_children(int n_processes, sigset_t *mask) { usec_t until; - int n_processes; - struct timespec ts; - - assert_se(sigemptyset(&mask) == 0); - assert_se(sigaddset(&mask, SIGCHLD) == 0); - if (sigprocmask(SIG_BLOCK, &mask, &oldmask) != 0) - return -errno; - - if (kill(-1, SIGSTOP) < 0) - log_warning("kill(-1, SIGSTOP) failed: %m"); - - n_processes = killall(sign); - - if (kill(-1, SIGCONT) < 0) - log_warning("kill(-1, SIGCONT) failed: %m"); - if (n_processes <= 0) - goto finish; + assert(mask); until = now(CLOCK_MONOTONIC) + TIMEOUT_USEC; for (;;) { + struct timespec ts; int k; - usec_t n = now(CLOCK_MONOTONIC); + usec_t n; for (;;) { pid_t pid = waitpid(-1, NULL, WNOHANG); if (pid == 0) break; - else if (pid < 0 && errno == ECHILD) { - n_processes = 0; - goto finish; - } - if (--n_processes == 0) - goto finish; + if (pid < 0 && errno == ECHILD) + return; + + if (n_processes > 0) + if (--n_processes == 0) + return; } + n = now(CLOCK_MONOTONIC); if (n >= until) - goto finish; + return; timespec_store(&ts, until - n); - if ((k = sigtimedwait(&mask, NULL, &ts)) != SIGCHLD) { + + if ((k = sigtimedwait(mask, NULL, &ts)) != SIGCHLD) { + + if (k < 0 && errno != EAGAIN) { + log_error("sigtimedwait() failed: %m"); + return; + } + if (k >= 0) log_warning("sigtimedwait() returned unexpected signal."); - if (k < 0 && errno != EAGAIN) - log_warning("sigtimedwait() failed: %m"); } } +} + +static void send_signal(int sign) { + sigset_t mask, oldmask; + int n_processes; + + assert_se(sigemptyset(&mask) == 0); + assert_se(sigaddset(&mask, SIGCHLD) == 0); + assert_se(sigprocmask(SIG_BLOCK, &mask, &oldmask) == 0); + + if (kill(-1, SIGSTOP) < 0 && errno != ESRCH) + log_warning("kill(-1, SIGSTOP) failed: %m"); + + n_processes = killall(sign); + + if (kill(-1, SIGCONT) < 0 && errno != ESRCH) + log_warning("kill(-1, SIGCONT) failed: %m"); + + if (n_processes <= 0) + goto finish; + + wait_for_children(n_processes, &mask); finish: sigprocmask(SIG_SETMASK, &oldmask, NULL); - - return n_processes; } -static int rescue_send_signal(int sign) { +static void ultimate_send_signal(int sign) { sigset_t mask, oldmask; - usec_t until; - struct timespec ts; int r; - sigemptyset(&mask); - sigaddset(&mask, SIGCHLD); - if (sigprocmask(SIG_BLOCK, &mask, &oldmask) != 0) - return -errno; + assert_se(sigemptyset(&mask) == 0); + assert_se(sigaddset(&mask, SIGCHLD) == 0); + assert_se(sigprocmask(SIG_BLOCK, &mask, &oldmask) == 0); - if (kill(-1, SIGSTOP) < 0) + if (kill(-1, SIGSTOP) < 0 && errno != ESRCH) log_warning("kill(-1, SIGSTOP) failed: %m"); r = kill(-1, sign); - if (r < 0) - log_warning("kill(-1, %d) failed: %m", sign); + if (r < 0 && errno != ESRCH) + log_warning("kill(-1, %s) failed: %m", signal_to_string(sign)); - if (kill(-1, SIGCONT) < 0) + if (kill(-1, SIGCONT) < 0 && errno != ESRCH) log_warning("kill(-1, SIGCONT) failed: %m"); if (r < 0) goto finish; - until = now(CLOCK_MONOTONIC) + TIMEOUT_USEC; - for (;;) { - int k; - usec_t n = now(CLOCK_MONOTONIC); + wait_for_children(0, &mask); - for (;;) { - pid_t pid = waitpid(-1, NULL, WNOHANG); - if (pid == 0) - break; - else if (pid < 0 && errno == ECHILD) - goto finish; - } +finish: + sigprocmask(SIG_SETMASK, &oldmask, NULL); +} - if (n >= until) - goto finish; +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"; - timespec_store(&ts, until - n); - if ((k = sigtimedwait(&mask, NULL, &ts)) != SIGCHLD) { - if (k >= 0) - log_warning("sigtimedwait() returned unexpected signal."); - if (k < 0 && errno != EAGAIN) - log_warning("sigtimedwait() failed: %m"); + 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; } -finish: - sigprocmask(SIG_SETMASK, &oldmask, NULL); + 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; + } - return r; + 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; + 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; @@ -232,6 +327,8 @@ int main(int argc, char *argv[]) { goto error; } + in_container = detect_container(NULL) > 0; + if (streq(argv[1], "reboot")) cmd = RB_AUTOBOOT; else if (streq(argv[1], "poweroff")) @@ -246,24 +343,28 @@ int main(int argc, char *argv[]) { 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"); log_info("Sending SIGTERM to remaining processes..."); - r = send_signal(SIGTERM); - if (r < 0) - log_warning("Failed to send SIGTERM to remaining processes: %s", strerror(r)); + send_signal(SIGTERM); log_info("Sending SIGKILL to remaining processes..."); - r = send_signal(SIGKILL); - if (r < 0) - log_warning("Failed to send SIGKILL to remaining processes: %s", strerror(r)); + send_signal(SIGKILL); + + if (in_container) + need_swapoff = false; /* Unmount all mountpoints, swaps, and loopback devices */ 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); @@ -308,9 +409,12 @@ int main(int argc, char *argv[]) { 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 @@ -325,8 +429,8 @@ int main(int argc, char *argv[]) { } log_warning("Cannot finalize remaining file systems and devices, trying to kill remaining processes."); - rescue_send_signal(SIGTERM); - rescue_send_signal(SIGKILL); + ultimate_send_signal(SIGTERM); + ultimate_send_signal(SIGKILL); killed_everbody = true; } @@ -334,7 +438,25 @@ int main(int argc, char *argv[]) { } if (retries >= FINALIZE_ATTEMPTS) - log_error("Too many interations, giving up."); + log_error("Too many iterations, giving up."); + + execute_directory(SYSTEM_SHUTDOWN_PATH, NULL, NULL); + + /* If we are in a container, just exit, this will kill our + * container for good. */ + 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(); @@ -349,7 +471,7 @@ int main(int argc, char *argv[]) { log_warning("kexec failed. Falling back to normal reboot."); } else { /* Child */ - const char *args[5] = { "/sbin/kexec", "-e", "-f", "-x", NULL }; + const char *args[3] = { "/sbin/kexec", "-e", NULL }; execv(args[0], (char * const *) args); return EXIT_FAILURE; } @@ -362,7 +484,6 @@ int main(int argc, char *argv[]) { r = -errno; error: - sync(); log_error("Critical error while doing system shutdown: %s", strerror(-r)); freeze();