src/core/shutdown.c

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2 /***
   3   This file is part of systemd.
   4
   5   Copyright 2010 ProFUSION embedded systems
   6 ***/
   7
   8 #include <errno.h>
   9 #include <getopt.h>
  10 #include <linux/reboot.h>
  11 #include <signal.h>
  12 #include <stdbool.h>
  13 #include <stdlib.h>
  14 #include <sys/mman.h>
  15 #include <sys/mount.h>
  16 #include <sys/reboot.h>
  17 #include <sys/stat.h>
  18 #include <unistd.h>
  19
  20 #include "alloc-util.h"
  21 #include "async.h"
  22 #include "cgroup-util.h"
  23 #include "def.h"
  24 #include "exec-util.h"
  25 #include "fd-util.h"
  26 #include "fileio.h"
  27 #include "killall.h"
  28 #include "log.h"
  29 #include "missing.h"
  30 #include "parse-util.h"
  31 #include "process-util.h"
  32 #include "reboot-util.h"
  33 #include "signal-util.h"
  34 #include "string-util.h"
  35 #include "switch-root.h"
  36 #include "terminal-util.h"
  37 #include "umount.h"
  38 #include "util.h"
  39 #include "virt.h"
  40 #include "watchdog.h"
  41
  42 #define SYNC_PROGRESS_ATTEMPTS 3
  43 #define SYNC_TIMEOUT_USEC (10*USEC_PER_SEC)
  44
  45 static char* arg_verb;
  46 static uint8_t arg_exit_code;
  47 static usec_t arg_timeout = DEFAULT_TIMEOUT_USEC;
  48
  49 static int parse_argv(int argc, char *argv[]) {
  50         enum {
  51                 ARG_LOG_LEVEL = 0x100,
  52                 ARG_LOG_TARGET,
  53                 ARG_LOG_COLOR,
  54                 ARG_LOG_LOCATION,
  55                 ARG_EXIT_CODE,
  56                 ARG_TIMEOUT,
  57         };
  58
  59         static const struct option options[] = {
  60                 { "log-level",     required_argument, NULL, ARG_LOG_LEVEL    },
  61                 { "log-target",    required_argument, NULL, ARG_LOG_TARGET   },
  62                 { "log-color",     optional_argument, NULL, ARG_LOG_COLOR    },
  63                 { "log-location",  optional_argument, NULL, ARG_LOG_LOCATION },
  64                 { "exit-code",     required_argument, NULL, ARG_EXIT_CODE    },
  65                 { "timeout",       required_argument, NULL, ARG_TIMEOUT      },
  66                 {}
  67         };
  68
  69         int c, r;
  70
  71         assert(argc >= 1);
  72         assert(argv);
  73
  74         /* "-" prevents getopt from permuting argv[] and moving the verb away
  75          * from argv[1]. Our interface to initrd promises it'll be there. */
  76         while ((c = getopt_long(argc, argv, "-", options, NULL)) >= 0)
  77                 switch (c) {
  78
  79                 case ARG_LOG_LEVEL:
  80                         r = log_set_max_level_from_string(optarg);
  81                         if (r < 0)
  82                                 log_error_errno(r, "Failed to parse log level %s, ignoring.", optarg);
  83
  84                         break;
  85
  86                 case ARG_LOG_TARGET:
  87                         r = log_set_target_from_string(optarg);
  88                         if (r < 0)
  89                                 log_error_errno(r, "Failed to parse log target %s, ignoring", optarg);
  90
  91                         break;
  92
  93                 case ARG_LOG_COLOR:
  94
  95                         if (optarg) {
  96                                 r = log_show_color_from_string(optarg);
  97                                 if (r < 0)
  98                                         log_error_errno(r, "Failed to parse log color setting %s, ignoring", optarg);
  99                         } else
 100                                 log_show_color(true);
 101
 102                         break;
 103
 104                 case ARG_LOG_LOCATION:
 105                         if (optarg) {
 106                                 r = log_show_location_from_string(optarg);
 107                                 if (r < 0)
 108                                         log_error_errno(r, "Failed to parse log location setting %s, ignoring", optarg);
 109                         } else
 110                                 log_show_location(true);
 111
 112                         break;
 113
 114                 case ARG_EXIT_CODE:
 115                         r = safe_atou8(optarg, &arg_exit_code);
 116                         if (r < 0)
 117                                 log_error_errno(r, "Failed to parse exit code %s, ignoring", optarg);
 118
 119                         break;
 120
 121                 case ARG_TIMEOUT:
 122                         r = parse_sec(optarg, &arg_timeout);
 123                         if (r < 0)
 124                                 log_error_errno(r, "Failed to parse shutdown timeout %s, ignoring", optarg);
 125
 126                         break;
 127
 128                 case '\001':
 129                         if (!arg_verb)
 130                                 arg_verb = optarg;
 131                         else
 132                                 log_error("Excess arguments, ignoring");
 133                         break;
 134
 135                 case '?':
 136                         return -EINVAL;
 137
 138                 default:
 139                         assert_not_reached("Unhandled option code.");
 140                 }
 141
 142         if (!arg_verb) {
 143                 log_error("Verb argument missing.");
 144                 return -EINVAL;
 145         }
 146
 147         return 0;
 148 }
 149
 150 static int switch_root_initramfs(void) {
 151         if (mount("/run/initramfs", "/run/initramfs", NULL, MS_BIND, NULL) < 0)
 152                 return log_error_errno(errno, "Failed to mount bind /run/initramfs on /run/initramfs: %m");
 153
 154         if (mount(NULL, "/run/initramfs", NULL, MS_PRIVATE, NULL) < 0)
 155                 return log_error_errno(errno, "Failed to make /run/initramfs private mount: %m");
 156
 157         /* switch_root with MS_BIND, because there might still be processes lurking around, which have open file descriptors.
 158          * /run/initramfs/shutdown will take care of these.
 159          * Also do not detach the old root, because /run/initramfs/shutdown needs to access it.
 160          */
 161         return switch_root("/run/initramfs", "/oldroot", false, MS_BIND);
 162 }
 163
 164 /* Read the following fields from /proc/meminfo:
 165  *
 166  *  NFS_Unstable
 167  *  Writeback
 168  *  Dirty
 169  *
 170  * Return true if the sum of these fields is greater than the previous
 171  * value input. For all other issues, report the failure and indicate that
 172  * the sync is not making progress.
 173  */
 174 static bool sync_making_progress(unsigned long long *prev_dirty) {
 175         _cleanup_fclose_ FILE *f = NULL;
 176         char line[LINE_MAX];
 177         bool r = false;
 178         unsigned long long val = 0;
 179
 180         f = fopen("/proc/meminfo", "re");
 181         if (!f)
 182                 return log_warning_errno(errno, "Failed to open /proc/meminfo: %m");
 183
 184         FOREACH_LINE(line, f, log_warning_errno(errno, "Failed to parse /proc/meminfo: %m")) {
 185                 unsigned long long ull = 0;
 186
 187                 if (!first_word(line, "NFS_Unstable:") && !first_word(line, "Writeback:") && !first_word(line, "Dirty:"))
 188                         continue;
 189
 190                 errno = 0;
 191                 if (sscanf(line, "%*s %llu %*s", &ull) != 1) {
 192                         if (errno != 0)
 193                                 log_warning_errno(errno, "Failed to parse /proc/meminfo: %m");
 194                         else
 195                                 log_warning("Failed to parse /proc/meminfo");
 196
 197                         return false;
 198                 }
 199
 200                 val += ull;
 201         }
 202
 203         r = *prev_dirty > val;
 204
 205         *prev_dirty = val;
 206
 207         return r;
 208 }
 209
 210 static void sync_with_progress(void) {
 211         unsigned long long dirty = ULONG_LONG_MAX;
 212         unsigned checks;
 213         pid_t pid;
 214         int r;
 215
 216         BLOCK_SIGNALS(SIGCHLD);
 217
 218         /* Due to the possiblity of the sync operation hanging, we fork a child process and monitor the progress. If
 219          * the timeout lapses, the assumption is that that particular sync stalled. */
 220
 221         r = asynchronous_sync(&pid);
 222         if (r < 0) {
 223                 log_error_errno(r, "Failed to fork sync(): %m");
 224                 return;
 225         }
 226
 227         log_info("Syncing filesystems and block devices.");
 228
 229         /* Start monitoring the sync operation. If more than
 230          * SYNC_PROGRESS_ATTEMPTS lapse without progress being made,
 231          * we assume that the sync is stalled */
 232         for (checks = 0; checks < SYNC_PROGRESS_ATTEMPTS; checks++) {
 233                 r = wait_for_terminate_with_timeout(pid, SYNC_TIMEOUT_USEC);
 234                 if (r == 0)
 235                         /* Sync finished without error.
 236                          * (The sync itself does not return an error code) */
 237                         return;
 238                 else if (r == -ETIMEDOUT) {
 239                         /* Reset the check counter if the "Dirty" value is
 240                          * decreasing */
 241                         if (sync_making_progress(&dirty))
 242                                 checks = 0;
 243                 } else {
 244                         log_error_errno(r, "Failed to sync filesystems and block devices: %m");
 245                         return;
 246                 }
 247         }
 248
 249         /* Only reached in the event of a timeout. We should issue a kill
 250          * to the stray process. */
 251         log_error("Syncing filesystems and block devices - timed out, issuing SIGKILL to PID "PID_FMT".", pid);
 252         (void) kill(pid, SIGKILL);
 253 }
 254
 255 int main(int argc, char *argv[]) {
 256         bool need_umount, need_swapoff, need_loop_detach, need_dm_detach;
 257         bool in_container, use_watchdog = false, can_initrd;
 258         _cleanup_free_ char *cgroup = NULL;
 259         char *arguments[3];
 260         int cmd, r, umount_log_level = LOG_INFO;
 261         static const char* const dirs[] = {SYSTEM_SHUTDOWN_PATH, NULL};
 262         char *watchdog_device;
 263
 264         /* The log target defaults to console, but the original systemd process will pass its log target in through a
 265          * command line argument, which will override this default. Also, ensure we'll never log to the journal or
 266          * syslog, as these logging daemons are either already dead or will die very soon. */
 267
 268         log_set_target(LOG_TARGET_CONSOLE);
 269         log_set_prohibit_ipc(true);
 270         log_parse_environment();
 271
 272         r = parse_argv(argc, argv);
 273         if (r < 0)
 274                 goto error;
 275
 276         log_open();
 277
 278         umask(0022);
 279
 280         if (getpid_cached() != 1) {
 281                 log_error("Not executed by init (PID 1).");
 282                 r = -EPERM;
 283                 goto error;
 284         }
 285
 286         if (streq(arg_verb, "reboot"))
 287                 cmd = RB_AUTOBOOT;
 288         else if (streq(arg_verb, "poweroff"))
 289                 cmd = RB_POWER_OFF;
 290         else if (streq(arg_verb, "halt"))
 291                 cmd = RB_HALT_SYSTEM;
 292         else if (streq(arg_verb, "kexec"))
 293                 cmd = LINUX_REBOOT_CMD_KEXEC;
 294         else if (streq(arg_verb, "exit"))
 295                 cmd = 0; /* ignored, just checking that arg_verb is valid */
 296         else {
 297                 log_error("Unknown action '%s'.", arg_verb);
 298                 r = -EINVAL;
 299                 goto error;
 300         }
 301
 302         (void) cg_get_root_path(&cgroup);
 303         in_container = detect_container() > 0;
 304
 305         use_watchdog = !!getenv("WATCHDOG_USEC");
 306         watchdog_device = getenv("WATCHDOG_DEVICE");
 307         if (watchdog_device) {
 308                 r = watchdog_set_device(watchdog_device);
 309                 if (r < 0)
 310                         log_warning_errno(r, "Failed to set watchdog device to %s, ignoring: %m",
 311                                           watchdog_device);
 312         }
 313
 314         /* Lock us into memory */
 315         (void) mlockall(MCL_CURRENT|MCL_FUTURE);
 316
 317         /* Synchronize everything that is not written to disk yet at this point already. This is a good idea so that
 318          * slow IO is processed here already and the final process killing spree is not impacted by processes
 319          * desperately trying to sync IO to disk within their timeout. Do not remove this sync, data corruption will
 320          * result. */
 321         if (!in_container)
 322                 sync_with_progress();
 323
 324         disable_coredumps();
 325
 326         log_info("Sending SIGTERM to remaining processes...");
 327         broadcast_signal(SIGTERM, true, true, arg_timeout);
 328
 329         log_info("Sending SIGKILL to remaining processes...");
 330         broadcast_signal(SIGKILL, true, false, arg_timeout);
 331
 332         need_umount = !in_container;
 333         need_swapoff = !in_container;
 334         need_loop_detach = !in_container;
 335         need_dm_detach = !in_container;
 336         can_initrd = !in_container && !in_initrd() && access("/run/initramfs/shutdown", X_OK) == 0;
 337
 338         /* Unmount all mountpoints, swaps, and loopback devices */
 339         for (;;) {
 340                 bool changed = false;
 341
 342                 if (use_watchdog)
 343                         watchdog_ping();
 344
 345                 /* Let's trim the cgroup tree on each iteration so
 346                    that we leave an empty cgroup tree around, so that
 347                    container managers get a nice notify event when we
 348                    are down */
 349                 if (cgroup)
 350                         cg_trim(SYSTEMD_CGROUP_CONTROLLER, cgroup, false);
 351
 352                 if (need_umount) {
 353                         log_info("Unmounting file systems.");
 354                         r = umount_all(&changed, umount_log_level);
 355                         if (r == 0) {
 356                                 need_umount = false;
 357                                 log_info("All filesystems unmounted.");
 358                         } else if (r > 0)
 359                                 log_info("Not all file systems unmounted, %d left.", r);
 360                         else
 361                                 log_error_errno(r, "Failed to unmount file systems: %m");
 362                 }
 363
 364                 if (need_swapoff) {
 365                         log_info("Deactivating swaps.");
 366                         r = swapoff_all(&changed);
 367                         if (r == 0) {
 368                                 need_swapoff = false;
 369                                 log_info("All swaps deactivated.");
 370                         } else if (r > 0)
 371                                 log_info("Not all swaps deactivated, %d left.", r);
 372                         else
 373                                 log_error_errno(r, "Failed to deactivate swaps: %m");
 374                 }
 375
 376                 if (need_loop_detach) {
 377                         log_info("Detaching loop devices.");
 378                         r = loopback_detach_all(&changed, umount_log_level);
 379                         if (r == 0) {
 380                                 need_loop_detach = false;
 381                                 log_info("All loop devices detached.");
 382                         } else if (r > 0)
 383                                 log_info("Not all loop devices detached, %d left.", r);
 384                         else
 385                                 log_error_errno(r, "Failed to detach loop devices: %m");
 386                 }
 387
 388                 if (need_dm_detach) {
 389                         log_info("Detaching DM devices.");
 390                         r = dm_detach_all(&changed, umount_log_level);
 391                         if (r == 0) {
 392                                 need_dm_detach = false;
 393                                 log_info("All DM devices detached.");
 394                         } else if (r > 0)
 395                                 log_info("Not all DM devices detached, %d left.", r);
 396                         else
 397                                 log_error_errno(r, "Failed to detach DM devices: %m");
 398                 }
 399
 400                 if (!need_umount && !need_swapoff && !need_loop_detach && !need_dm_detach) {
 401                         log_info("All filesystems, swaps, loop devices and DM devices detached.");
 402                         /* Yay, done */
 403                         break;
 404                 }
 405
 406                 if (!changed && umount_log_level == LOG_INFO && !can_initrd) {
 407                         /* There are things we cannot get rid of. Loop one more time
 408                          * with LOG_ERR to inform the user. Note that we don't need
 409                          * to do this if there is a initrd to switch to, because that
 410                          * one is likely to get rid of the remounting mounts. If not,
 411                          * it will log about them. */
 412                         umount_log_level = LOG_ERR;
 413                         continue;
 414                 }
 415
 416                 /* If in this iteration we didn't manage to
 417                  * unmount/deactivate anything, we simply give up */
 418                 if (!changed) {
 419                         log_info("Cannot finalize remaining%s%s%s%s continuing.",
 420                                  need_umount ? " file systems," : "",
 421                                  need_swapoff ? " swap devices," : "",
 422                                  need_loop_detach ? " loop devices," : "",
 423                                  need_dm_detach ? " DM devices," : "");
 424                         break;
 425                 }
 426
 427                 log_debug("Couldn't finalize remaining %s%s%s%s trying again.",
 428                           need_umount ? " file systems," : "",
 429                           need_swapoff ? " swap devices," : "",
 430                           need_loop_detach ? " loop devices," : "",
 431                           need_dm_detach ? " DM devices," : "");
 432         }
 433
 434         /* We're done with the watchdog. */
 435         watchdog_free_device();
 436
 437         arguments[0] = NULL;
 438         arguments[1] = arg_verb;
 439         arguments[2] = NULL;
 440         execute_directories(dirs, DEFAULT_TIMEOUT_USEC, NULL, NULL, arguments);
 441
 442         if (can_initrd) {
 443                 r = switch_root_initramfs();
 444                 if (r >= 0) {
 445                         argv[0] = (char*) "/shutdown";
 446
 447                         setsid();
 448                         make_console_stdio();
 449
 450                         log_info("Successfully changed into root pivot.\n"
 451                                  "Returning to initrd...");
 452
 453                         execv("/shutdown", argv);
 454                         log_error_errno(errno, "Failed to execute shutdown binary: %m");
 455                 } else
 456                         log_error_errno(r, "Failed to switch root to \"/run/initramfs\": %m");
 457
 458         }
 459
 460         if (need_umount || need_swapoff || need_loop_detach || need_dm_detach)
 461                 log_error("Failed to finalize %s%s%s%s ignoring",
 462                           need_umount ? " file systems," : "",
 463                           need_swapoff ? " swap devices," : "",
 464                           need_loop_detach ? " loop devices," : "",
 465                           need_dm_detach ? " DM devices," : "");
 466
 467         /* The kernel will automatically flush ATA disks and suchlike on reboot(), but the file systems need to be
 468          * sync'ed explicitly in advance. So let's do this here, but not needlessly slow down containers. Note that we
 469          * sync'ed things already once above, but we did some more work since then which might have caused IO, hence
 470          * let's do it once more. Do not remove this sync, data corruption will result. */
 471         if (!in_container)
 472                 sync_with_progress();
 473
 474         if (streq(arg_verb, "exit")) {
 475                 if (in_container)
 476                         return arg_exit_code;
 477
 478                 cmd = RB_POWER_OFF; /* We cannot exit() on the host, fallback on another method. */
 479         }
 480
 481         switch (cmd) {
 482
 483         case LINUX_REBOOT_CMD_KEXEC:
 484
 485                 if (!in_container) {
 486                         /* We cheat and exec kexec to avoid doing all its work */
 487                         log_info("Rebooting with kexec.");
 488
 489                         r = safe_fork("(sd-kexec)", FORK_RESET_SIGNALS|FORK_CLOSE_ALL_FDS|FORK_LOG|FORK_WAIT, NULL);
 490                         if (r == 0) {
 491                                 const char * const args[] = {
 492                                         KEXEC, "-e", NULL
 493                                 };
 494
 495                                 /* Child */
 496
 497                                 execv(args[0], (char * const *) args);
 498                                 _exit(EXIT_FAILURE);
 499                         }
 500
 501                         /* If we are still running, then the kexec can't have worked, let's fall through */
 502                 }
 503
 504                 cmd = RB_AUTOBOOT;
 505                 _fallthrough_;
 506
 507         case RB_AUTOBOOT:
 508                 (void) reboot_with_parameter(REBOOT_LOG);
 509                 log_info("Rebooting.");
 510                 break;
 511
 512         case RB_POWER_OFF:
 513                 log_info("Powering off.");
 514                 break;
 515
 516         case RB_HALT_SYSTEM:
 517                 log_info("Halting system.");
 518                 break;
 519
 520         default:
 521                 assert_not_reached("Unknown magic");
 522         }
 523
 524         (void) reboot(cmd);
 525         if (errno == EPERM && in_container) {
 526                 /* If we are in a container, and we lacked
 527                  * CAP_SYS_BOOT just exit, this will kill our
 528                  * container for good. */
 529                 log_info("Exiting container.");
 530                 return EXIT_SUCCESS;
 531         }
 532
 533         r = log_error_errno(errno, "Failed to invoke reboot(): %m");
 534
 535   error:
 536         log_emergency_errno(r, "Critical error while doing system shutdown: %m");
 537         freeze();
 538 }