X-Git-Url: https://www.chiark.greenend.org.uk/ucgi/~ianmdlvl/git?p=elogind.git;a=blobdiff_plain;f=src%2Fshutdown.c;h=11c7544fc7c50cd45c42673edd006d25a1f2f4ba;hp=3c9a53126f208a4e4fcb8fcb9152850eeae29b48;hb=2f8cd170aeb0d748f5af3cefb387d14f67fc286e;hpb=567ea02a89b358090032c8d1ec89286db07ab2ff

diff --git a/src/shutdown.c b/src/shutdown.c
index 3c9a53126..11c7544fc 100644
--- a/src/shutdown.c
+++ b/src/shutdown.c
@@ -38,7 +38,6 @@
 
 #define TIMEOUT_USEC (5 * USEC_PER_SEC)
 #define FINALIZE_ATTEMPTS 50
-#define FINALIZE_CRITICAL_ATTEMPTS 10
 
 static bool ignore_proc(pid_t pid) {
         if (pid == 1)
@@ -69,7 +68,7 @@ static bool is_kernel_thread(pid_t pid)
 static int killall(int sign) {
         DIR *dir;
         struct dirent *d;
-        unsigned int processes = 0;
+        unsigned int n_processes = 0;
 
         if ((dir = opendir("/proc")) == NULL)
                 return -errno;
@@ -87,20 +86,20 @@ static int killall(int sign) {
                         continue;
 
                 if (kill(pid, sign) == 0)
-                        processes++;
+                        n_processes++;
                 else
                         log_warning("Could not kill %d: %m", pid);
         }
 
         closedir(dir);
 
-        return processes;
+        return n_processes;
 }
 
 static int send_signal(int sign) {
         sigset_t mask, oldmask;
         usec_t until;
-        int processes;
+        int n_processes;
         struct timespec ts;
 
         assert_se(sigemptyset(&mask) == 0);
@@ -109,29 +108,32 @@ static int send_signal(int sign) {
                 return -errno;
 
         if (kill(-1, SIGSTOP) < 0)
-                log_warning("Failed kill(-1, SIGSTOP): %m");
+                log_warning("kill(-1, SIGSTOP) failed: %m");
 
-        processes = killall(sign);
+        n_processes = killall(sign);
 
         if (kill(-1, SIGCONT) < 0)
-                log_warning("Failed kill(-1, SIGCONT): %m");
+                log_warning("kill(-1, SIGCONT) failed: %m");
 
-        if (processes <= 0)
+        if (n_processes <= 0)
                 goto finish;
 
         until = now(CLOCK_MONOTONIC) + TIMEOUT_USEC;
         for (;;) {
+                int k;
                 usec_t n = now(CLOCK_MONOTONIC);
+
                 for (;;) {
                         pid_t pid = waitpid(-1, NULL, WNOHANG);
+
                         if (pid == 0)
                                 break;
                         else if (pid < 0 && errno == ECHILD) {
-                                processes = 0;
+                                n_processes = 0;
                                 goto finish;
                         }
 
-                        if (--processes == 0)
+                        if (--n_processes == 0)
                                 goto finish;
                 }
 
@@ -139,14 +141,18 @@ static int send_signal(int sign) {
                         goto finish;
 
                 timespec_store(&ts, until - n);
-                if (sigtimedwait(&mask, NULL, &ts) != SIGCHLD)
-                        log_warning("Failed: sigtimedwait did not return SIGCHLD: %m");
+                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");
+                }
         }
 
 finish:
         sigprocmask(SIG_SETMASK, &oldmask, NULL);
 
-        return processes;
+        return n_processes;
 }
 
 static int rescue_send_signal(int sign) {
@@ -161,21 +167,23 @@ static int rescue_send_signal(int sign) {
                 return -errno;
 
         if (kill(-1, SIGSTOP) < 0)
-                log_warning("Failed kill(-1, SIGSTOP): %m");
+                log_warning("kill(-1, SIGSTOP) failed: %m");
 
         r = kill(-1, sign);
         if (r < 0)
-                log_warning("Failed kill(-1, %d): %m", sign);
+                log_warning("kill(-1, %d) failed: %m", sign);
 
         if (kill(-1, SIGCONT) < 0)
-                log_warning("Failed kill(-1, SIGCONT): %m");
+                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);
+
                 for (;;) {
                         pid_t pid = waitpid(-1, NULL, WNOHANG);
                         if (pid == 0)
@@ -188,8 +196,12 @@ static int rescue_send_signal(int sign) {
                         goto finish;
 
                 timespec_store(&ts, until - n);
-                if (sigtimedwait(&mask, NULL, &ts) != SIGCHLD)
-                        log_warning("Failed: sigtimedwait did not return SIGCHLD: %m");
+                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");
+                }
         }
 
 finish:
@@ -199,11 +211,13 @@ finish:
 }
 
 int main(int argc, char *argv[]) {
-        int cmd, r, retries;
-        bool need_umount = true, need_swapoff = true, need_loop_detach = true;
+        int cmd, r;
+        unsigned retries;
+        bool need_umount = true, need_swapoff = true, need_loop_detach = true, need_dm_detach = true;
+        bool killed_everbody = false;
 
         log_parse_environment();
-        log_set_target(LOG_TARGET_KMSG); /* syslog will die if not gone yet */
+        log_set_target(LOG_TARGET_CONSOLE); /* syslog will die if not gone yet */
         log_open();
 
         if (getpid() != 1) {
@@ -236,99 +250,121 @@ int main(int argc, char *argv[]) {
         if (mlockall(MCL_CURRENT|MCL_FUTURE) != 0)
                 log_warning("Cannot lock process memory: %m");
 
-        log_info("Sending SIGTERM to processes");
+        log_info("Sending SIGTERM to remaining processes...");
         r = send_signal(SIGTERM);
         if (r < 0)
-                log_warning("Cannot send SIGTERM to all process: %s", strerror(r));
+                log_warning("Failed to send SIGTERM to remaining processes: %s", strerror(r));
 
-        log_info("Sending SIGKILL to processes");
+        log_info("Sending SIGKILL to remaining processes...");
         r = send_signal(SIGKILL);
         if (r < 0)
-                log_warning("Cannot send SIGKILL to all process: %s", strerror(r));
+                log_warning("Failed to send SIGKILL to remaining processes: %s", strerror(r));
 
         /* Unmount all mountpoints, swaps, and loopback devices */
-        retries = FINALIZE_ATTEMPTS;
-        while (need_umount || need_swapoff || need_loop_detach) {
+        for (retries = 0; retries < FINALIZE_ATTEMPTS; retries++) {
+                bool changed = false;
+
                 if (need_umount) {
-                        log_info("Unmounting filesystems.");
-                        r = umount_all();
+                        log_info("Unmounting file systems.");
+                        r = umount_all(&changed);
                         if (r == 0)
                                 need_umount = false;
                         else if (r > 0)
-                                log_warning("Not all filesystems unmounted, %d left.", r);
+                                log_info("Not all file systems unmounted, %d left.", r);
                         else
-                                log_error("Error unmounting filesystems: %s", strerror(-r));
+                                log_error("Failed to unmount file systems: %s", strerror(-r));
                 }
 
                 if (need_swapoff) {
                         log_info("Disabling swaps.");
-                        r = swapoff_all();
+                        r = swapoff_all(&changed);
                         if (r == 0)
                                 need_swapoff = false;
                         else if (r > 0)
-                                log_warning("Not all swaps are off, %d left.", r);
+                                log_info("Not all swaps are turned off, %d left.", r);
                         else
-                                log_error("Error turning off swaps: %s", strerror(-r));
+                                log_error("Failed to turn off swaps: %s", strerror(-r));
                 }
 
                 if (need_loop_detach) {
                         log_info("Detaching loop devices.");
-                        r = loopback_detach_all();
+                        r = loopback_detach_all(&changed);
                         if (r == 0)
                                 need_loop_detach = false;
                         else if (r > 0)
-                                log_warning("Not all loop devices detached, %d left.", r);
+                                log_info("Not all loop devices detached, %d left.", r);
                         else
-                                log_error("Error detaching loop devices: %s", strerror(-r));
+                                log_error("Failed to detach loop devices: %s", strerror(-r));
+                }
 
+                if (need_dm_detach) {
+                        log_info("Detaching DM devices.");
+                        r = dm_detach_all(&changed);
+                        if (r == 0)
+                                need_dm_detach = false;
+                        else if (r > 0)
+                                log_warning("Not all DM devices detached, %d left.", r);
+                        else
+                                log_error("Failed to detach DM devices: %s", strerror(-r));
                 }
 
-                if (need_umount || need_swapoff || need_loop_detach) {
-                        retries--;
+                if (!need_umount && !need_swapoff && !need_loop_detach && !need_dm_detach)
+                        /* Yay, done */
+                        break;
 
-                        if (retries <= FINALIZE_CRITICAL_ATTEMPTS) {
-                                log_warning("Approaching critical level to finalize filesystem and devices, try to kill all processes.");
-                                rescue_send_signal(SIGTERM);
-                                rescue_send_signal(SIGKILL);
-                        }
+                /* If in this iteration we didn't manage to
+                 * unmount/deactivate anything, we either kill more
+                 * processes, or simply give up */
+                if (!changed) {
 
-                        if (retries > 0)
-                                log_info("Action still required, %d tries left", retries);
-                        else {
-                                log_error("Tried enough but still action required need_umount=%d, need_swapoff=%d, need_loop_detach=%d", need_umount, need_swapoff, need_loop_detach);
-                                r = -EBUSY;
-                                goto error;
+                        if (killed_everbody) {
+                                /* Hmm, we already killed everybody,
+                                 * let's just give up */
+                                log_error("Cannot finalize remaining file systems and devices, giving up.");
+                                break;
                         }
+
+                        log_warning("Cannot finalize remaining file systems and devices, trying to kill remaining processes.");
+                        rescue_send_signal(SIGTERM);
+                        rescue_send_signal(SIGKILL);
+                        killed_everbody = true;
                 }
+
+                log_debug("Couldn't finalize remaining file systems and devices after %u retries, trying again.", retries+1);
         }
 
+        if (retries >= FINALIZE_ATTEMPTS)
+                log_error("Too many interations, giving up.");
+
         sync();
 
         if (cmd == LINUX_REBOOT_CMD_KEXEC) {
-                /* we cheat and exec kexec to avoid doing all its work */
+                /* We cheat and exec kexec to avoid doing all its work */
                 pid_t pid = fork();
-                if (pid < 0) {
-                        log_error("Could not fork: %m. Falling back to reboot.");
-                        cmd = RB_AUTOBOOT;
-                } else if (pid > 0) {
-                        waitpid(pid, NULL, 0);
-                        log_warning("Failed %s -e -x -f. Falling back to reboot", KEXEC_BINARY_PATH);
-                        cmd = RB_AUTOBOOT;
+
+                if (pid < 0)
+                        log_error("Could not fork: %m. Falling back to normal reboot.");
+                else if (pid > 0) {
+                        wait_for_terminate_and_warn("kexec", pid);
+                        log_warning("kexec failed. Falling back to normal reboot.");
                 } else {
-                        const char *args[5] = {KEXEC_BINARY_PATH, "-e", "-f", "-x", NULL};
+                        /* Child */
+                        const char *args[5] = { "/sbin/kexec", "-e", "-f", "-x", NULL };
                         execv(args[0], (char * const *) args);
                         return EXIT_FAILURE;
                 }
+
+                cmd = RB_AUTOBOOT;
         }
 
         reboot(cmd);
-        r = errno;
+        log_error("Failed to invoke reboot(): %m");
+        r = -errno;
 
   error:
         sync();
-        if (r < 0)
-                r = -r;
-        log_error("Critical error while doing system shutdown: %s", strerror(r));
+        log_error("Critical error while doing system shutdown: %s", strerror(-r));
+
         freeze();
         return EXIT_FAILURE;
 }