src/shutdown.c

   1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
   2
   3 /***
   4   This file is part of systemd.
   5
   6   Copyright 2010 ProFUSION embedded systems
   7
   8   systemd is free software; you can redistribute it and/or modify it
   9   under the terms of the GNU General Public License as published by
  10   the Free Software Foundation; either version 2 of the License, or
  11   (at your option) any later version.
  12
  13   systemd is distributed in the hope that it will be useful, but
  14   WITHOUT ANY WARRANTY; without even the implied warranty of
  15   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  16   General Public License for more details.
  17
  18   You should have received a copy of the GNU General Public License
  19   along with systemd; If not, see <http://www.gnu.org/licenses/>.
  20 ***/
  21
  22 #include <sys/mman.h>
  23 #include <sys/types.h>
  24 #include <sys/reboot.h>
  25 #include <linux/reboot.h>
  26 #include <sys/wait.h>
  27 #include <sys/types.h>
  28 #include <sys/stat.h>
  29 #include <sys/mount.h>
  30 #include <sys/syscall.h>
  31 #include <fcntl.h>
  32 #include <dirent.h>
  33 #include <errno.h>
  34 #include <unistd.h>
  35 #include <signal.h>
  36 #include <stdbool.h>
  37 #include <stdlib.h>
  38 #include <string.h>
  39
  40 #include "missing.h"
  41 #include "log.h"
  42 #include "umount.h"
  43 #include "util.h"
  44 #include "virt.h"
  45
  46 #define TIMEOUT_USEC (5 * USEC_PER_SEC)
  47 #define FINALIZE_ATTEMPTS 50
  48
  49 static bool ignore_proc(pid_t pid) {
  50         char buf[PATH_MAX];
  51         FILE *f;
  52         char c;
  53         size_t count;
  54         uid_t uid;
  55         int r;
  56
  57         /* We are PID 1, let's not commit suicide */
  58         if (pid == 1)
  59                 return true;
  60
  61         r = get_process_uid(pid, &uid);
  62         if (r < 0)
  63                 return true; /* not really, but better safe than sorry */
  64
  65         /* Non-root processes otherwise are always subject to be killed */
  66         if (uid != 0)
  67                 return false;
  68
  69         snprintf(buf, sizeof(buf), "/proc/%lu/cmdline", (unsigned long) pid);
  70         char_array_0(buf);
  71
  72         f = fopen(buf, "re");
  73         if (!f)
  74                 return true; /* not really, but has the desired effect */
  75
  76         count = fread(&c, 1, 1, f);
  77         fclose(f);
  78
  79         /* Kernel threads have an empty cmdline */
  80         if (count <= 0)
  81                 return true;
  82
  83         /* Processes with argv[0][0] = '@' we ignore from the killing
  84          * spree.
  85          *
  86          * http://www.freedesktop.org/wiki/Software/systemd/RootStorageDaemons */
  87         if (count == 1 && c == '@')
  88                 return true;
  89
  90         return false;
  91 }
  92
  93 static int killall(int sign) {
  94         DIR *dir;
  95         struct dirent *d;
  96         unsigned int n_processes = 0;
  97
  98         dir = opendir("/proc");
  99         if (!dir)
 100                 return -errno;
 101
 102         while ((d = readdir(dir))) {
 103                 pid_t pid;
 104
 105                 if (parse_pid(d->d_name, &pid) < 0)
 106                         continue;
 107
 108                 if (ignore_proc(pid))
 109                         continue;
 110
 111                 if (kill(pid, sign) == 0)
 112                         n_processes++;
 113                 else
 114                         log_warning("Could not kill %d: %m", pid);
 115         }
 116
 117         closedir(dir);
 118
 119         return n_processes;
 120 }
 121
 122 static void wait_for_children(int n_processes, sigset_t *mask) {
 123         usec_t until;
 124
 125         assert(mask);
 126
 127         until = now(CLOCK_MONOTONIC) + TIMEOUT_USEC;
 128         for (;;) {
 129                 struct timespec ts;
 130                 int k;
 131                 usec_t n;
 132
 133                 for (;;) {
 134                         pid_t pid = waitpid(-1, NULL, WNOHANG);
 135
 136                         if (pid == 0)
 137                                 break;
 138
 139                         if (pid < 0 && errno == ECHILD)
 140                                 return;
 141
 142                         if (n_processes > 0)
 143                                 if (--n_processes == 0)
 144                                         return;
 145                 }
 146
 147                 n = now(CLOCK_MONOTONIC);
 148                 if (n >= until)
 149                         return;
 150
 151                 timespec_store(&ts, until - n);
 152
 153                 if ((k = sigtimedwait(mask, NULL, &ts)) != SIGCHLD) {
 154
 155                         if (k < 0 && errno != EAGAIN) {
 156                                 log_error("sigtimedwait() failed: %m");
 157                                 return;
 158                         }
 159
 160                         if (k >= 0)
 161                                 log_warning("sigtimedwait() returned unexpected signal.");
 162                 }
 163         }
 164 }
 165
 166 static void send_signal(int sign) {
 167         sigset_t mask, oldmask;
 168         int n_processes;
 169
 170         assert_se(sigemptyset(&mask) == 0);
 171         assert_se(sigaddset(&mask, SIGCHLD) == 0);
 172         assert_se(sigprocmask(SIG_BLOCK, &mask, &oldmask) == 0);
 173
 174         if (kill(-1, SIGSTOP) < 0 && errno != ESRCH)
 175                 log_warning("kill(-1, SIGSTOP) failed: %m");
 176
 177         n_processes = killall(sign);
 178
 179         if (kill(-1, SIGCONT) < 0 && errno != ESRCH)
 180                 log_warning("kill(-1, SIGCONT) failed: %m");
 181
 182         if (n_processes <= 0)
 183                 goto finish;
 184
 185         wait_for_children(n_processes, &mask);
 186
 187 finish:
 188         sigprocmask(SIG_SETMASK, &oldmask, NULL);
 189 }
 190
 191 static void ultimate_send_signal(int sign) {
 192         sigset_t mask, oldmask;
 193         int r;
 194
 195         assert_se(sigemptyset(&mask) == 0);
 196         assert_se(sigaddset(&mask, SIGCHLD) == 0);
 197         assert_se(sigprocmask(SIG_BLOCK, &mask, &oldmask) == 0);
 198
 199         if (kill(-1, SIGSTOP) < 0 && errno != ESRCH)
 200                 log_warning("kill(-1, SIGSTOP) failed: %m");
 201
 202         r = kill(-1, sign);
 203         if (r < 0 && errno != ESRCH)
 204                 log_warning("kill(-1, %s) failed: %m", signal_to_string(sign));
 205
 206         if (kill(-1, SIGCONT) < 0 && errno != ESRCH)
 207                 log_warning("kill(-1, SIGCONT) failed: %m");
 208
 209         if (r < 0)
 210                 goto finish;
 211
 212         wait_for_children(0, &mask);
 213
 214 finish:
 215         sigprocmask(SIG_SETMASK, &oldmask, NULL);
 216 }
 217
 218 static int prepare_new_root(void) {
 219         static const char dirs[] =
 220                 "/run/initramfs/oldroot\0"
 221                 "/run/initramfs/proc\0"
 222                 "/run/initramfs/sys\0"
 223                 "/run/initramfs/dev\0"
 224                 "/run/initramfs/run\0";
 225
 226         const char *dir;
 227
 228         if (mount("/run/initramfs", "/run/initramfs", NULL, MS_BIND, NULL) < 0) {
 229                 log_error("Failed to mount bind /run/initramfs on /run/initramfs: %m");
 230                 return -errno;
 231         }
 232
 233         if (mount(NULL, "/run/initramfs", NULL, MS_PRIVATE, NULL) < 0) {
 234                 log_error("Failed to make /run/initramfs private mount: %m");
 235                 return -errno;
 236         }
 237
 238         NULSTR_FOREACH(dir, dirs)
 239                 if (mkdir_p(dir, 0755) < 0 && errno != EEXIST) {
 240                         log_error("Failed to mkdir %s: %m", dir);
 241                         return -errno;
 242                 }
 243
 244         if (mount("/sys", "/run/initramfs/sys", NULL, MS_BIND, NULL) < 0) {
 245                 log_error("Failed to mount bind /sys on /run/initramfs/sys: %m");
 246                 return -errno;
 247         }
 248
 249         if (mount("/proc", "/run/initramfs/proc", NULL, MS_BIND, NULL) < 0) {
 250                 log_error("Failed to mount bind /proc on /run/initramfs/proc: %m");
 251                 return -errno;
 252         }
 253
 254         if (mount("/dev", "/run/initramfs/dev", NULL, MS_BIND, NULL) < 0) {
 255                 log_error("Failed to mount bind /dev on /run/initramfs/dev: %m");
 256                 return -errno;
 257         }
 258
 259         if (mount("/run", "/run/initramfs/run", NULL, MS_BIND, NULL) < 0) {
 260                 log_error("Failed to mount bind /run on /run/initramfs/run: %m");
 261                 return -errno;
 262         }
 263
 264         return 0;
 265 }
 266
 267 static int pivot_to_new_root(void) {
 268         int fd;
 269
 270         chdir("/run/initramfs");
 271
 272         /*
 273           In case some evil process made "/" MS_SHARED
 274           It works for pivot_root, but the ref count for the root device
 275           is not decreasing :-/
 276         */
 277         if (mount(NULL, "/", NULL, MS_PRIVATE, NULL) < 0) {
 278                 log_error("Failed to make \"/\" private mount %m");
 279                 return -errno;
 280         }
 281
 282         if (pivot_root(".", "oldroot") < 0) {
 283                 log_error("pivot failed: %m");
 284                 /* only chroot if pivot root succeded */
 285                 return -errno;
 286         }
 287
 288         chroot(".");
 289         log_info("Successfully changed into root pivot.");
 290
 291         fd = open("/dev/console", O_RDWR);
 292         if (fd < 0)
 293                 log_error("Failed to open /dev/console: %m");
 294         else {
 295                 make_stdio(fd);
 296
 297                 /* Initialize the controlling terminal */
 298                 setsid();
 299                 ioctl(STDIN_FILENO, TIOCSCTTY, NULL);
 300         }
 301
 302         return 0;
 303 }
 304
 305 int main(int argc, char *argv[]) {
 306         int cmd, r;
 307         unsigned retries;
 308         bool need_umount = true, need_swapoff = true, need_loop_detach = true, need_dm_detach = true;
 309         bool killed_everbody = false, in_container;
 310
 311         log_parse_environment();
 312         log_set_target(LOG_TARGET_CONSOLE); /* syslog will die if not gone yet */
 313         log_open();
 314
 315         umask(0022);
 316
 317         if (getpid() != 1) {
 318                 log_error("Not executed by init (pid 1).");
 319                 r = -EPERM;
 320                 goto error;
 321         }
 322
 323         if (argc != 2) {
 324                 log_error("Invalid number of arguments.");
 325                 r = -EINVAL;
 326                 goto error;
 327         }
 328
 329         in_container = detect_container(NULL) > 0;
 330
 331         if (streq(argv[1], "reboot"))
 332                 cmd = RB_AUTOBOOT;
 333         else if (streq(argv[1], "poweroff"))
 334                 cmd = RB_POWER_OFF;
 335         else if (streq(argv[1], "halt"))
 336                 cmd = RB_HALT_SYSTEM;
 337         else if (streq(argv[1], "kexec"))
 338                 cmd = LINUX_REBOOT_CMD_KEXEC;
 339         else {
 340                 log_error("Unknown action '%s'.", argv[1]);
 341                 r = -EINVAL;
 342                 goto error;
 343         }
 344
 345         /* lock us into memory */
 346         if (mlockall(MCL_CURRENT|MCL_FUTURE) != 0)
 347                 log_warning("Cannot lock process memory: %m");
 348
 349         log_info("Sending SIGTERM to remaining processes...");
 350         send_signal(SIGTERM);
 351
 352         log_info("Sending SIGKILL to remaining processes...");
 353         send_signal(SIGKILL);
 354
 355         if (in_container)
 356                 need_swapoff = false;
 357
 358         /* Unmount all mountpoints, swaps, and loopback devices */
 359         for (retries = 0; retries < FINALIZE_ATTEMPTS; retries++) {
 360                 bool changed = false;
 361
 362                 if (need_umount) {
 363                         log_info("Unmounting file systems.");
 364                         r = umount_all(&changed);
 365                         if (r == 0)
 366                                 need_umount = false;
 367                         else if (r > 0)
 368                                 log_info("Not all file systems unmounted, %d left.", r);
 369                         else
 370                                 log_error("Failed to unmount file systems: %s", strerror(-r));
 371                 }
 372
 373                 if (need_swapoff) {
 374                         log_info("Disabling swaps.");
 375                         r = swapoff_all(&changed);
 376                         if (r == 0)
 377                                 need_swapoff = false;
 378                         else if (r > 0)
 379                                 log_info("Not all swaps are turned off, %d left.", r);
 380                         else
 381                                 log_error("Failed to turn off swaps: %s", strerror(-r));
 382                 }
 383
 384                 if (need_loop_detach) {
 385                         log_info("Detaching loop devices.");
 386                         r = loopback_detach_all(&changed);
 387                         if (r == 0)
 388                                 need_loop_detach = false;
 389                         else if (r > 0)
 390                                 log_info("Not all loop devices detached, %d left.", r);
 391                         else
 392                                 log_error("Failed to detach loop devices: %s", strerror(-r));
 393                 }
 394
 395                 if (need_dm_detach) {
 396                         log_info("Detaching DM devices.");
 397                         r = dm_detach_all(&changed);
 398                         if (r == 0)
 399                                 need_dm_detach = false;
 400                         else if (r > 0)
 401                                 log_warning("Not all DM devices detached, %d left.", r);
 402                         else
 403                                 log_error("Failed to detach DM devices: %s", strerror(-r));
 404                 }
 405
 406                 if (!need_umount && !need_swapoff && !need_loop_detach && !need_dm_detach) {
 407                         if (retries > 0)
 408                                 log_info("All filesystems, swaps, loop devices, DM devices detached.");
 409                         /* Yay, done */
 410                         break;
 411                 }
 412
 413                 /* If in this iteration we didn't manage to
 414                  * unmount/deactivate anything, we either kill more
 415                  * processes, or simply give up */
 416                 if (!changed) {
 417
 418                         if (killed_everbody) {
 419                                 /* Hmm, we already killed everybody,
 420                                  * let's just give up */
 421                                 log_error("Cannot finalize remaining file systems and devices, giving up.");
 422                                 break;
 423                         }
 424
 425                         log_warning("Cannot finalize remaining file systems and devices, trying to kill remaining processes.");
 426                         ultimate_send_signal(SIGTERM);
 427                         ultimate_send_signal(SIGKILL);
 428                         killed_everbody = true;
 429                 }
 430
 431                 log_debug("Couldn't finalize remaining file systems and devices after %u retries, trying again.", retries+1);
 432         }
 433
 434         if (retries >= FINALIZE_ATTEMPTS)
 435                 log_error("Too many iterations, giving up.");
 436
 437         execute_directory(SYSTEM_SHUTDOWN_PATH, NULL, NULL);
 438
 439         /* If we are in a container, just exit, this will kill our
 440          * container for good. */
 441         if (in_container) {
 442                 log_error("Exiting container.");
 443                 exit(0);
 444         }
 445
 446         if (access("/run/initramfs/shutdown", X_OK) == 0) {
 447
 448                 if (prepare_new_root() >= 0 &&
 449                     pivot_to_new_root() >= 0) {
 450                         execv("/shutdown", argv);
 451                         log_error("Failed to execute shutdown binary: %m");
 452                 }
 453         }
 454
 455         sync();
 456
 457         if (cmd == LINUX_REBOOT_CMD_KEXEC) {
 458                 /* We cheat and exec kexec to avoid doing all its work */
 459                 pid_t pid = fork();
 460
 461                 if (pid < 0)
 462                         log_error("Could not fork: %m. Falling back to normal reboot.");
 463                 else if (pid > 0) {
 464                         wait_for_terminate_and_warn("kexec", pid);
 465                         log_warning("kexec failed. Falling back to normal reboot.");
 466                 } else {
 467                         /* Child */
 468                         const char *args[3] = { "/sbin/kexec", "-e", NULL };
 469                         execv(args[0], (char * const *) args);
 470                         return EXIT_FAILURE;
 471                 }
 472
 473                 cmd = RB_AUTOBOOT;
 474         }
 475
 476         reboot(cmd);
 477         log_error("Failed to invoke reboot(): %m");
 478         r = -errno;
 479
 480   error:
 481         log_error("Critical error while doing system shutdown: %s", strerror(-r));
 482
 483         freeze();
 484         return EXIT_FAILURE;
 485 }