src/core/shutdown.c

   1 /***
   2   This file is part of systemd.
   3
   4   Copyright 2010 ProFUSION embedded systems
   5
   6   systemd is free software; you can redistribute it and/or modify it
   7   under the terms of the GNU Lesser General Public License as published by
   8   the Free Software Foundation; either version 2.1 of the License, or
   9   (at your option) any later version.
  10
  11   systemd is distributed in the hope that it will be useful, but
  12   WITHOUT ANY WARRANTY; without even the implied warranty of
  13   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  14   Lesser General Public License for more details.
  15
  16   You should have received a copy of the GNU Lesser General Public License
  17   along with systemd; If not, see <http://www.gnu.org/licenses/>.
  18 ***/
  19
  20 #include <errno.h>
  21 #include <getopt.h>
  22 #include <linux/reboot.h>
  23 #include <signal.h>
  24 #include <stdbool.h>
  25 #include <stdlib.h>
  26 #include <sys/mman.h>
  27 #include <sys/mount.h>
  28 #include <sys/reboot.h>
  29 #include <sys/stat.h>
  30 #include <unistd.h>
  31
  32 #include "alloc-util.h"
  33 #include "cgroup-util.h"
  34 #include "def.h"
  35 #include "fileio.h"
  36 #include "killall.h"
  37 #include "log.h"
  38 #include "missing.h"
  39 #include "parse-util.h"
  40 #include "process-util.h"
  41 #include "string-util.h"
  42 #include "switch-root.h"
  43 #include "terminal-util.h"
  44 #include "umount.h"
  45 #include "util.h"
  46 #include "virt.h"
  47 #include "watchdog.h"
  48
  49 #define FINALIZE_ATTEMPTS 50
  50
  51 static char* arg_verb;
  52 static uint8_t arg_exit_code;
  53
  54 static int parse_argv(int argc, char *argv[]) {
  55         enum {
  56                 ARG_LOG_LEVEL = 0x100,
  57                 ARG_LOG_TARGET,
  58                 ARG_LOG_COLOR,
  59                 ARG_LOG_LOCATION,
  60                 ARG_EXIT_CODE,
  61         };
  62
  63         static const struct option options[] = {
  64                 { "log-level",     required_argument, NULL, ARG_LOG_LEVEL    },
  65                 { "log-target",    required_argument, NULL, ARG_LOG_TARGET   },
  66                 { "log-color",     optional_argument, NULL, ARG_LOG_COLOR    },
  67                 { "log-location",  optional_argument, NULL, ARG_LOG_LOCATION },
  68                 { "exit-code",     required_argument, NULL, ARG_EXIT_CODE    },
  69                 {}
  70         };
  71
  72         int c, r;
  73
  74         assert(argc >= 1);
  75         assert(argv);
  76
  77         /* "-" prevents getopt from permuting argv[] and moving the verb away
  78          * from argv[1]. Our interface to initrd promises it'll be there. */
  79         while ((c = getopt_long(argc, argv, "-", options, NULL)) >= 0)
  80                 switch (c) {
  81
  82                 case ARG_LOG_LEVEL:
  83                         r = log_set_max_level_from_string(optarg);
  84                         if (r < 0)
  85                                 log_error("Failed to parse log level %s, ignoring.", optarg);
  86
  87                         break;
  88
  89                 case ARG_LOG_TARGET:
  90                         r = log_set_target_from_string(optarg);
  91                         if (r < 0)
  92                                 log_error("Failed to parse log target %s, ignoring", optarg);
  93
  94                         break;
  95
  96                 case ARG_LOG_COLOR:
  97
  98                         if (optarg) {
  99                                 r = log_show_color_from_string(optarg);
 100                                 if (r < 0)
 101                                         log_error("Failed to parse log color setting %s, ignoring", optarg);
 102                         } else
 103                                 log_show_color(true);
 104
 105                         break;
 106
 107                 case ARG_LOG_LOCATION:
 108                         if (optarg) {
 109                                 r = log_show_location_from_string(optarg);
 110                                 if (r < 0)
 111                                         log_error("Failed to parse log location setting %s, ignoring", optarg);
 112                         } else
 113                                 log_show_location(true);
 114
 115                         break;
 116
 117                 case ARG_EXIT_CODE:
 118                         r = safe_atou8(optarg, &arg_exit_code);
 119                         if (r < 0)
 120                                 log_error("Failed to parse exit code %s, ignoring", optarg);
 121
 122                         break;
 123
 124                 case '\001':
 125                         if (!arg_verb)
 126                                 arg_verb = optarg;
 127                         else
 128                                 log_error("Excess arguments, ignoring");
 129                         break;
 130
 131                 case '?':
 132                         return -EINVAL;
 133
 134                 default:
 135                         assert_not_reached("Unhandled option code.");
 136                 }
 137
 138         if (!arg_verb) {
 139                 log_error("Verb argument missing.");
 140                 return -EINVAL;
 141         }
 142
 143         return 0;
 144 }
 145
 146 static int switch_root_initramfs(void) {
 147         if (mount("/run/initramfs", "/run/initramfs", NULL, MS_BIND, NULL) < 0)
 148                 return log_error_errno(errno, "Failed to mount bind /run/initramfs on /run/initramfs: %m");
 149
 150         if (mount(NULL, "/run/initramfs", NULL, MS_PRIVATE, NULL) < 0)
 151                 return log_error_errno(errno, "Failed to make /run/initramfs private mount: %m");
 152
 153         /* switch_root with MS_BIND, because there might still be processes lurking around, which have open file descriptors.
 154          * /run/initramfs/shutdown will take care of these.
 155          * Also do not detach the old root, because /run/initramfs/shutdown needs to access it.
 156          */
 157         return switch_root("/run/initramfs", "/oldroot", false, MS_BIND);
 158 }
 159
 160 int main(int argc, char *argv[]) {
 161         bool need_umount, need_swapoff, need_loop_detach, need_dm_detach;
 162         bool in_container, use_watchdog = false;
 163         _cleanup_free_ char *cgroup = NULL;
 164         char *arguments[3];
 165         unsigned retries;
 166         int cmd, r;
 167         static const char* const dirs[] = {SYSTEM_SHUTDOWN_PATH, NULL};
 168
 169         log_parse_environment();
 170         r = parse_argv(argc, argv);
 171         if (r < 0)
 172                 goto error;
 173
 174         /* journald will die if not gone yet. The log target defaults
 175          * to console, but may have been changed by command line options. */
 176
 177         log_close_console(); /* force reopen of /dev/console */
 178         log_open();
 179
 180         umask(0022);
 181
 182         if (getpid() != 1) {
 183                 log_error("Not executed by init (PID 1).");
 184                 r = -EPERM;
 185                 goto error;
 186         }
 187
 188         if (streq(arg_verb, "reboot"))
 189                 cmd = RB_AUTOBOOT;
 190         else if (streq(arg_verb, "poweroff"))
 191                 cmd = RB_POWER_OFF;
 192         else if (streq(arg_verb, "halt"))
 193                 cmd = RB_HALT_SYSTEM;
 194         else if (streq(arg_verb, "kexec"))
 195                 cmd = LINUX_REBOOT_CMD_KEXEC;
 196         else if (streq(arg_verb, "exit"))
 197                 cmd = 0; /* ignored, just checking that arg_verb is valid */
 198         else {
 199                 r = -EINVAL;
 200                 log_error("Unknown action '%s'.", arg_verb);
 201                 goto error;
 202         }
 203
 204         (void) cg_get_root_path(&cgroup);
 205         in_container = detect_container() > 0;
 206
 207         use_watchdog = !!getenv("WATCHDOG_USEC");
 208
 209         /* Lock us into memory */
 210         mlockall(MCL_CURRENT|MCL_FUTURE);
 211
 212         /* Synchronize everything that is not written to disk yet at this point already. This is a good idea so that
 213          * slow IO is processed here already and the final process killing spree is not impacted by processes
 214          * desperately trying to sync IO to disk within their timeout. */
 215         if (!in_container)
 216                 sync();
 217
 218         log_info("Sending SIGTERM to remaining processes...");
 219         broadcast_signal(SIGTERM, true, true);
 220
 221         log_info("Sending SIGKILL to remaining processes...");
 222         broadcast_signal(SIGKILL, true, false);
 223
 224         need_umount = !in_container;
 225         need_swapoff = !in_container;
 226         need_loop_detach = !in_container;
 227         need_dm_detach = !in_container;
 228
 229         /* Unmount all mountpoints, swaps, and loopback devices */
 230         for (retries = 0; retries < FINALIZE_ATTEMPTS; retries++) {
 231                 bool changed = false;
 232
 233                 if (use_watchdog)
 234                         watchdog_ping();
 235
 236                 /* Let's trim the cgroup tree on each iteration so
 237                    that we leave an empty cgroup tree around, so that
 238                    container managers get a nice notify event when we
 239                    are down */
 240                 if (cgroup)
 241                         cg_trim(SYSTEMD_CGROUP_CONTROLLER, cgroup, false);
 242
 243                 if (need_umount) {
 244                         log_info("Unmounting file systems.");
 245                         r = umount_all(&changed);
 246                         if (r == 0) {
 247                                 need_umount = false;
 248                                 log_info("All filesystems unmounted.");
 249                         } else if (r > 0)
 250                                 log_info("Not all file systems unmounted, %d left.", r);
 251                         else
 252                                 log_error_errno(r, "Failed to unmount file systems: %m");
 253                 }
 254
 255                 if (need_swapoff) {
 256                         log_info("Deactivating swaps.");
 257                         r = swapoff_all(&changed);
 258                         if (r == 0) {
 259                                 need_swapoff = false;
 260                                 log_info("All swaps deactivated.");
 261                         } else if (r > 0)
 262                                 log_info("Not all swaps deactivated, %d left.", r);
 263                         else
 264                                 log_error_errno(r, "Failed to deactivate swaps: %m");
 265                 }
 266
 267                 if (need_loop_detach) {
 268                         log_info("Detaching loop devices.");
 269                         r = loopback_detach_all(&changed);
 270                         if (r == 0) {
 271                                 need_loop_detach = false;
 272                                 log_info("All loop devices detached.");
 273                         } else if (r > 0)
 274                                 log_info("Not all loop devices detached, %d left.", r);
 275                         else
 276                                 log_error_errno(r, "Failed to detach loop devices: %m");
 277                 }
 278
 279                 if (need_dm_detach) {
 280                         log_info("Detaching DM devices.");
 281                         r = dm_detach_all(&changed);
 282                         if (r == 0) {
 283                                 need_dm_detach = false;
 284                                 log_info("All DM devices detached.");
 285                         } else if (r > 0)
 286                                 log_info("Not all DM devices detached, %d left.", r);
 287                         else
 288                                 log_error_errno(r, "Failed to detach DM devices: %m");
 289                 }
 290
 291                 if (!need_umount && !need_swapoff && !need_loop_detach && !need_dm_detach) {
 292                         if (retries > 0)
 293                                 log_info("All filesystems, swaps, loop devices, DM devices detached.");
 294                         /* Yay, done */
 295                         goto initrd_jump;
 296                 }
 297
 298                 /* If in this iteration we didn't manage to
 299                  * unmount/deactivate anything, we simply give up */
 300                 if (!changed) {
 301                         log_info("Cannot finalize remaining%s%s%s%s continuing.",
 302                                  need_umount ? " file systems," : "",
 303                                  need_swapoff ? " swap devices," : "",
 304                                  need_loop_detach ? " loop devices," : "",
 305                                  need_dm_detach ? " DM devices," : "");
 306                         goto initrd_jump;
 307                 }
 308
 309                 log_debug("After %u retries, couldn't finalize remaining %s%s%s%s trying again.",
 310                           retries + 1,
 311                           need_umount ? " file systems," : "",
 312                           need_swapoff ? " swap devices," : "",
 313                           need_loop_detach ? " loop devices," : "",
 314                           need_dm_detach ? " DM devices," : "");
 315         }
 316
 317         log_error("Too many iterations, giving up.");
 318
 319  initrd_jump:
 320
 321         arguments[0] = NULL;
 322         arguments[1] = arg_verb;
 323         arguments[2] = NULL;
 324         execute_directories(dirs, DEFAULT_TIMEOUT_USEC, arguments);
 325
 326         if (!in_container && !in_initrd() &&
 327             access("/run/initramfs/shutdown", X_OK) == 0) {
 328                 r = switch_root_initramfs();
 329                 if (r >= 0) {
 330                         argv[0] = (char*) "/shutdown";
 331
 332                         setsid();
 333                         make_console_stdio();
 334
 335                         log_info("Successfully changed into root pivot.\n"
 336                                  "Returning to initrd...");
 337
 338                         execv("/shutdown", argv);
 339                         log_error_errno(errno, "Failed to execute shutdown binary: %m");
 340                 } else
 341                         log_error_errno(r, "Failed to switch root to \"/run/initramfs\": %m");
 342
 343         }
 344
 345         if (need_umount || need_swapoff || need_loop_detach || need_dm_detach)
 346                 log_error("Failed to finalize %s%s%s%s ignoring",
 347                           need_umount ? " file systems," : "",
 348                           need_swapoff ? " swap devices," : "",
 349                           need_loop_detach ? " loop devices," : "",
 350                           need_dm_detach ? " DM devices," : "");
 351
 352         /* The kernel will automatically flush ATA disks and suchlike on reboot(), but the file systems need to be
 353          * sync'ed explicitly in advance. So let's do this here, but not needlessly slow down containers. Note that we
 354          * sync'ed things already once above, but we did some more work since then which might have caused IO, hence
 355          * let's doit once more. */
 356         if (!in_container)
 357                 sync();
 358
 359         if (streq(arg_verb, "exit")) {
 360                 if (in_container)
 361                         exit(arg_exit_code);
 362                 else {
 363                         /* We cannot exit() on the host, fallback on another
 364                          * method. */
 365                         cmd = RB_POWER_OFF;
 366                 }
 367         }
 368
 369         switch (cmd) {
 370
 371         case LINUX_REBOOT_CMD_KEXEC:
 372
 373                 if (!in_container) {
 374                         /* We cheat and exec kexec to avoid doing all its work */
 375                         pid_t pid;
 376
 377                         log_info("Rebooting with kexec.");
 378
 379                         pid = fork();
 380                         if (pid < 0)
 381                                 log_error_errno(errno, "Failed to fork: %m");
 382                         else if (pid == 0) {
 383
 384                                 const char * const args[] = {
 385                                         KEXEC, "-e", NULL
 386                                 };
 387
 388                                 /* Child */
 389
 390                                 execv(args[0], (char * const *) args);
 391                                 _exit(EXIT_FAILURE);
 392                         } else
 393                                 wait_for_terminate_and_warn("kexec", pid, true);
 394                 }
 395
 396                 cmd = RB_AUTOBOOT;
 397                 /* Fall through */
 398
 399         case RB_AUTOBOOT:
 400
 401                 if (!in_container) {
 402                         _cleanup_free_ char *param = NULL;
 403
 404                         r = read_one_line_file("/run/systemd/reboot-param", &param);
 405                         if (r < 0)
 406                                 log_warning_errno(r, "Failed to read reboot parameter file: %m");
 407
 408                         if (!isempty(param)) {
 409                                 log_info("Rebooting with argument '%s'.", param);
 410                                 syscall(SYS_reboot, LINUX_REBOOT_MAGIC1, LINUX_REBOOT_MAGIC2, LINUX_REBOOT_CMD_RESTART2, param);
 411                                 log_warning_errno(errno, "Failed to reboot with parameter, retrying without: %m");
 412                         }
 413                 }
 414
 415                 log_info("Rebooting.");
 416                 break;
 417
 418         case RB_POWER_OFF:
 419                 log_info("Powering off.");
 420                 break;
 421
 422         case RB_HALT_SYSTEM:
 423                 log_info("Halting system.");
 424                 break;
 425
 426         default:
 427                 assert_not_reached("Unknown magic");
 428         }
 429
 430         reboot(cmd);
 431         if (errno == EPERM && in_container) {
 432                 /* If we are in a container, and we lacked
 433                  * CAP_SYS_BOOT just exit, this will kill our
 434                  * container for good. */
 435                 log_info("Exiting container.");
 436                 exit(0);
 437         }
 438
 439         r = log_error_errno(errno, "Failed to invoke reboot(): %m");
 440
 441   error:
 442         log_emergency_errno(r, "Critical error while doing system shutdown: %m");
 443         freeze();
 444 }