src/shutdown/shutdown.c

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2 /***
   3   Copyright © 2010 ProFUSION embedded systems
   4 ***/
   5
   6 #include <errno.h>
   7 #include <getopt.h>
   8 #include <linux/reboot.h>
   9 #include <signal.h>
  10 #include <stdbool.h>
  11 #include <stdlib.h>
  12 #include <sys/mman.h>
  13 #include <sys/mount.h>
  14 #include <sys/reboot.h>
  15 #include <sys/stat.h>
  16 #include <unistd.h>
  17
  18 #include "alloc-util.h"
  19 #include "async.h"
  20 #include "cgroup-util.h"
  21 #include "def.h"
  22 #include "exec-util.h"
  23 #include "fd-util.h"
  24 #include "fileio.h"
  25 #include "killall.h"
  26 #include "log.h"
  27 #include "missing.h"
  28 #include "parse-util.h"
  29 #include "process-util.h"
  30 #include "reboot-util.h"
  31 #include "rlimit-util.h"
  32 #include "signal-util.h"
  33 #include "string-util.h"
  34 #include "switch-root.h"
  35 #include "sysctl-util.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: %m", 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: %m", 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: %m", 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: %m", 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: %m", 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: %m", 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                 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
 144                                        "Verb argument missing.");
 145
 146         return 0;
 147 }
 148
 149 static int switch_root_initramfs(void) {
 150         if (mount("/run/initramfs", "/run/initramfs", NULL, MS_BIND, NULL) < 0)
 151                 return log_error_errno(errno, "Failed to mount bind /run/initramfs on /run/initramfs: %m");
 152
 153         if (mount(NULL, "/run/initramfs", NULL, MS_PRIVATE, NULL) < 0)
 154                 return log_error_errno(errno, "Failed to make /run/initramfs private mount: %m");
 155
 156         /* switch_root with MS_BIND, because there might still be processes lurking around, which have open file descriptors.
 157          * /run/initramfs/shutdown will take care of these.
 158          * Also do not detach the old root, because /run/initramfs/shutdown needs to access it.
 159          */
 160         return switch_root("/run/initramfs", "/oldroot", false, MS_BIND);
 161 }
 162
 163 /* Read the following fields from /proc/meminfo:
 164  *
 165  *  NFS_Unstable
 166  *  Writeback
 167  *  Dirty
 168  *
 169  * Return true if the sum of these fields is greater than the previous
 170  * value input. For all other issues, report the failure and indicate that
 171  * the sync is not making progress.
 172  */
 173 static int sync_making_progress(unsigned long long *prev_dirty) {
 174         _cleanup_fclose_ FILE *f = NULL;
 175         unsigned long long val = 0;
 176         int ret;
 177
 178         f = fopen("/proc/meminfo", "re");
 179         if (!f)
 180                 return log_warning_errno(errno, "Failed to open /proc/meminfo: %m");
 181
 182         for (;;) {
 183                 _cleanup_free_ char *line = NULL;
 184                 unsigned long long ull = 0;
 185                 int q;
 186
 187                 q = read_line(f, LONG_LINE_MAX, &line);
 188                 if (q < 0)
 189                         return log_warning_errno(q, "Failed to parse /proc/meminfo: %m");
 190                 if (q == 0)
 191                         break;
 192
 193                 if (!first_word(line, "NFS_Unstable:") && !first_word(line, "Writeback:") && !first_word(line, "Dirty:"))
 194                         continue;
 195
 196                 errno = 0;
 197                 if (sscanf(line, "%*s %llu %*s", &ull) != 1) {
 198                         if (errno != 0)
 199                                 log_warning_errno(errno, "Failed to parse /proc/meminfo: %m");
 200                         else
 201                                 log_warning("Failed to parse /proc/meminfo");
 202
 203                         return false;
 204                 }
 205
 206                 val += ull;
 207         }
 208
 209         ret = *prev_dirty > val;
 210         *prev_dirty = val;
 211         return ret;
 212 }
 213
 214 static void sync_with_progress(void) {
 215         unsigned long long dirty = ULLONG_MAX;
 216         unsigned checks;
 217         pid_t pid;
 218         int r;
 219
 220         BLOCK_SIGNALS(SIGCHLD);
 221
 222         /* Due to the possibility of the sync operation hanging, we fork a child process and monitor the progress. If
 223          * the timeout lapses, the assumption is that that particular sync stalled. */
 224
 225         r = asynchronous_sync(&pid);
 226         if (r < 0) {
 227                 log_error_errno(r, "Failed to fork sync(): %m");
 228                 return;
 229         }
 230
 231         log_info("Syncing filesystems and block devices.");
 232
 233         /* Start monitoring the sync operation. If more than
 234          * SYNC_PROGRESS_ATTEMPTS lapse without progress being made,
 235          * we assume that the sync is stalled */
 236         for (checks = 0; checks < SYNC_PROGRESS_ATTEMPTS; checks++) {
 237                 r = wait_for_terminate_with_timeout(pid, SYNC_TIMEOUT_USEC);
 238                 if (r == 0)
 239                         /* Sync finished without error.
 240                          * (The sync itself does not return an error code) */
 241                         return;
 242                 else if (r == -ETIMEDOUT) {
 243                         /* Reset the check counter if the "Dirty" value is
 244                          * decreasing */
 245                         if (sync_making_progress(&dirty) > 0)
 246                                 checks = 0;
 247                 } else {
 248                         log_error_errno(r, "Failed to sync filesystems and block devices: %m");
 249                         return;
 250                 }
 251         }
 252
 253         /* Only reached in the event of a timeout. We should issue a kill
 254          * to the stray process. */
 255         log_error("Syncing filesystems and block devices - timed out, issuing SIGKILL to PID "PID_FMT".", pid);
 256         (void) kill(pid, SIGKILL);
 257 }
 258
 259 static int read_current_sysctl_printk_log_level(void) {
 260         _cleanup_free_ char *sysctl_printk_vals = NULL, *sysctl_printk_curr = NULL;
 261         unsigned current_lvl = 0;
 262         const char *p;
 263         int r;
 264
 265         r = sysctl_read("kernel/printk", &sysctl_printk_vals);
 266         if (r < 0)
 267                 return log_debug_errno(r, "Cannot read sysctl kernel.printk: %m");
 268
 269         p = sysctl_printk_vals;
 270         r = extract_first_word(&p, &sysctl_printk_curr, NULL, 0);
 271         if (r > 0)
 272                 r = safe_atou(sysctl_printk_curr, &current_lvl);
 273         else if (r == 0)
 274                 r = -EINVAL;
 275
 276         if (r < 0)
 277                 return log_debug_errno(r, "Unexpected sysctl kernel.printk content: %s", sysctl_printk_vals);
 278
 279         return current_lvl;
 280 }
 281
 282 static void bump_sysctl_printk_log_level(int min_level) {
 283         /* Set the logging level to be able to see messages with log level smaller or equal to min_level */
 284
 285         int current_lvl = read_current_sysctl_printk_log_level();
 286         if (current_lvl >= 0 && current_lvl <= min_level) {
 287                 char buf[DECIMAL_STR_MAX(int)];
 288                 xsprintf(buf, "%d", min_level + 1);
 289                 int r = sysctl_write("kernel/printk", buf);
 290                 if (r < 0)
 291                         log_debug_errno(r, "Failed to bump kernel.printk to %s: %m", buf);
 292         }
 293 }
 294
 295 int main(int argc, char *argv[]) {
 296         bool need_umount, need_swapoff, need_loop_detach, need_dm_detach;
 297         bool in_container, use_watchdog = false, can_initrd;
 298         _cleanup_free_ char *cgroup = NULL;
 299         char *arguments[3];
 300         int cmd, r, umount_log_level = LOG_INFO;
 301         static const char* const dirs[] = {SYSTEM_SHUTDOWN_PATH, NULL};
 302         char *watchdog_device;
 303
 304         /* The log target defaults to console, but the original systemd process will pass its log target in through a
 305          * command line argument, which will override this default. Also, ensure we'll never log to the journal or
 306          * syslog, as these logging daemons are either already dead or will die very soon. */
 307
 308         log_set_target(LOG_TARGET_CONSOLE);
 309         log_set_prohibit_ipc(true);
 310         log_parse_environment();
 311
 312         r = parse_argv(argc, argv);
 313         if (r < 0)
 314                 goto error;
 315
 316         log_open();
 317
 318         umask(0022);
 319
 320         if (getpid_cached() != 1) {
 321                 log_error("Not executed by init (PID 1).");
 322                 r = -EPERM;
 323                 goto error;
 324         }
 325
 326         if (streq(arg_verb, "reboot"))
 327                 cmd = RB_AUTOBOOT;
 328         else if (streq(arg_verb, "poweroff"))
 329                 cmd = RB_POWER_OFF;
 330         else if (streq(arg_verb, "halt"))
 331                 cmd = RB_HALT_SYSTEM;
 332         else if (streq(arg_verb, "kexec"))
 333                 cmd = LINUX_REBOOT_CMD_KEXEC;
 334         else if (streq(arg_verb, "exit"))
 335                 cmd = 0; /* ignored, just checking that arg_verb is valid */
 336         else {
 337                 log_error("Unknown action '%s'.", arg_verb);
 338                 r = -EINVAL;
 339                 goto error;
 340         }
 341
 342         (void) cg_get_root_path(&cgroup);
 343         in_container = detect_container() > 0;
 344
 345         /* If the logging messages are going to KMSG, and if we are not running from a container,
 346          * then try to update the sysctl kernel.printk current value in order to see "info" messages;
 347          * This current log level is not updated if already big enough.
 348          */
 349         if (!in_container && IN_SET(log_get_target(), LOG_TARGET_AUTO,
 350                                                       LOG_TARGET_JOURNAL_OR_KMSG,
 351                                                       LOG_TARGET_SYSLOG_OR_KMSG,
 352                                                       LOG_TARGET_KMSG))
 353                 bump_sysctl_printk_log_level(LOG_INFO);
 354
 355         use_watchdog = getenv("WATCHDOG_USEC");
 356         watchdog_device = getenv("WATCHDOG_DEVICE");
 357         if (watchdog_device) {
 358                 r = watchdog_set_device(watchdog_device);
 359                 if (r < 0)
 360                         log_warning_errno(r, "Failed to set watchdog device to %s, ignoring: %m",
 361                                           watchdog_device);
 362         }
 363
 364         /* Lock us into memory */
 365         (void) mlockall(MCL_CURRENT|MCL_FUTURE);
 366
 367         /* Synchronize everything that is not written to disk yet at this point already. This is a good idea so that
 368          * slow IO is processed here already and the final process killing spree is not impacted by processes
 369          * desperately trying to sync IO to disk within their timeout. Do not remove this sync, data corruption will
 370          * result. */
 371         if (!in_container)
 372                 sync_with_progress();
 373
 374         disable_coredumps();
 375
 376         log_info("Sending SIGTERM to remaining processes...");
 377         broadcast_signal(SIGTERM, true, true, arg_timeout);
 378
 379         log_info("Sending SIGKILL to remaining processes...");
 380         broadcast_signal(SIGKILL, true, false, arg_timeout);
 381
 382         need_umount = !in_container;
 383         need_swapoff = !in_container;
 384         need_loop_detach = !in_container;
 385         need_dm_detach = !in_container;
 386         can_initrd = !in_container && !in_initrd() && access("/run/initramfs/shutdown", X_OK) == 0;
 387
 388         /* Unmount all mountpoints, swaps, and loopback devices */
 389         for (;;) {
 390                 bool changed = false;
 391
 392                 if (use_watchdog)
 393                         (void) watchdog_ping();
 394
 395                 /* Let's trim the cgroup tree on each iteration so
 396                    that we leave an empty cgroup tree around, so that
 397                    container managers get a nice notify event when we
 398                    are down */
 399                 if (cgroup)
 400                         (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER, cgroup, false);
 401
 402                 if (need_umount) {
 403                         log_info("Unmounting file systems.");
 404                         r = umount_all(&changed, umount_log_level);
 405                         if (r == 0) {
 406                                 need_umount = false;
 407                                 log_info("All filesystems unmounted.");
 408                         } else if (r > 0)
 409                                 log_info("Not all file systems unmounted, %d left.", r);
 410                         else
 411                                 log_error_errno(r, "Failed to unmount file systems: %m");
 412                 }
 413
 414                 if (need_swapoff) {
 415                         log_info("Deactivating swaps.");
 416                         r = swapoff_all(&changed);
 417                         if (r == 0) {
 418                                 need_swapoff = false;
 419                                 log_info("All swaps deactivated.");
 420                         } else if (r > 0)
 421                                 log_info("Not all swaps deactivated, %d left.", r);
 422                         else
 423                                 log_error_errno(r, "Failed to deactivate swaps: %m");
 424                 }
 425
 426                 if (need_loop_detach) {
 427                         log_info("Detaching loop devices.");
 428                         r = loopback_detach_all(&changed, umount_log_level);
 429                         if (r == 0) {
 430                                 need_loop_detach = false;
 431                                 log_info("All loop devices detached.");
 432                         } else if (r > 0)
 433                                 log_info("Not all loop devices detached, %d left.", r);
 434                         else
 435                                 log_error_errno(r, "Failed to detach loop devices: %m");
 436                 }
 437
 438                 if (need_dm_detach) {
 439                         log_info("Detaching DM devices.");
 440                         r = dm_detach_all(&changed, umount_log_level);
 441                         if (r == 0) {
 442                                 need_dm_detach = false;
 443                                 log_info("All DM devices detached.");
 444                         } else if (r > 0)
 445                                 log_info("Not all DM devices detached, %d left.", r);
 446                         else
 447                                 log_error_errno(r, "Failed to detach DM devices: %m");
 448                 }
 449
 450                 if (!need_umount && !need_swapoff && !need_loop_detach && !need_dm_detach) {
 451                         log_info("All filesystems, swaps, loop devices and DM devices detached.");
 452                         /* Yay, done */
 453                         break;
 454                 }
 455
 456                 if (!changed && umount_log_level == LOG_INFO && !can_initrd) {
 457                         /* There are things we cannot get rid of. Loop one more time
 458                          * with LOG_ERR to inform the user. Note that we don't need
 459                          * to do this if there is a initrd to switch to, because that
 460                          * one is likely to get rid of the remounting mounts. If not,
 461                          * it will log about them. */
 462                         umount_log_level = LOG_ERR;
 463                         continue;
 464                 }
 465
 466                 /* If in this iteration we didn't manage to
 467                  * unmount/deactivate anything, we simply give up */
 468                 if (!changed) {
 469                         log_info("Cannot finalize remaining%s%s%s%s continuing.",
 470                                  need_umount ? " file systems," : "",
 471                                  need_swapoff ? " swap devices," : "",
 472                                  need_loop_detach ? " loop devices," : "",
 473                                  need_dm_detach ? " DM devices," : "");
 474                         break;
 475                 }
 476
 477                 log_debug("Couldn't finalize remaining %s%s%s%s trying again.",
 478                           need_umount ? " file systems," : "",
 479                           need_swapoff ? " swap devices," : "",
 480                           need_loop_detach ? " loop devices," : "",
 481                           need_dm_detach ? " DM devices," : "");
 482         }
 483
 484         /* We're done with the watchdog. */
 485         watchdog_free_device();
 486
 487         arguments[0] = NULL;
 488         arguments[1] = arg_verb;
 489         arguments[2] = NULL;
 490         (void) execute_directories(dirs, DEFAULT_TIMEOUT_USEC, NULL, NULL, arguments, NULL, EXEC_DIR_PARALLEL | EXEC_DIR_IGNORE_ERRORS);
 491
 492         (void) rlimit_nofile_safe();
 493
 494         if (can_initrd) {
 495                 r = switch_root_initramfs();
 496                 if (r >= 0) {
 497                         argv[0] = (char*) "/shutdown";
 498
 499                         (void) setsid();
 500                         (void) make_console_stdio();
 501
 502                         log_info("Successfully changed into root pivot.\n"
 503                                  "Returning to initrd...");
 504
 505                         execv("/shutdown", argv);
 506                         log_error_errno(errno, "Failed to execute shutdown binary: %m");
 507                 } else
 508                         log_error_errno(r, "Failed to switch root to \"/run/initramfs\": %m");
 509
 510         }
 511
 512         if (need_umount || need_swapoff || need_loop_detach || need_dm_detach)
 513                 log_error("Failed to finalize %s%s%s%s ignoring",
 514                           need_umount ? " file systems," : "",
 515                           need_swapoff ? " swap devices," : "",
 516                           need_loop_detach ? " loop devices," : "",
 517                           need_dm_detach ? " DM devices," : "");
 518
 519         /* The kernel will automatically flush ATA disks and suchlike on reboot(), but the file systems need to be
 520          * sync'ed explicitly in advance. So let's do this here, but not needlessly slow down containers. Note that we
 521          * sync'ed things already once above, but we did some more work since then which might have caused IO, hence
 522          * let's do it once more. Do not remove this sync, data corruption will result. */
 523         if (!in_container)
 524                 sync_with_progress();
 525
 526         if (streq(arg_verb, "exit")) {
 527                 if (in_container)
 528                         return arg_exit_code;
 529
 530                 cmd = RB_POWER_OFF; /* We cannot exit() on the host, fallback on another method. */
 531         }
 532
 533         switch (cmd) {
 534
 535         case LINUX_REBOOT_CMD_KEXEC:
 536
 537                 if (!in_container) {
 538                         /* We cheat and exec kexec to avoid doing all its work */
 539                         log_info("Rebooting with kexec.");
 540
 541                         r = safe_fork("(sd-kexec)", FORK_RESET_SIGNALS|FORK_CLOSE_ALL_FDS|FORK_LOG|FORK_WAIT, NULL);
 542                         if (r == 0) {
 543                                 const char * const args[] = {
 544                                         KEXEC, "-e", NULL
 545                                 };
 546
 547                                 /* Child */
 548
 549                                 execv(args[0], (char * const *) args);
 550                                 _exit(EXIT_FAILURE);
 551                         }
 552
 553                         /* If we are still running, then the kexec can't have worked, let's fall through */
 554                 }
 555
 556                 cmd = RB_AUTOBOOT;
 557                 _fallthrough_;
 558
 559         case RB_AUTOBOOT:
 560                 (void) reboot_with_parameter(REBOOT_LOG);
 561                 log_info("Rebooting.");
 562                 break;
 563
 564         case RB_POWER_OFF:
 565                 log_info("Powering off.");
 566                 break;
 567
 568         case RB_HALT_SYSTEM:
 569                 log_info("Halting system.");
 570                 break;
 571
 572         default:
 573                 assert_not_reached("Unknown magic");
 574         }
 575
 576         (void) reboot(cmd);
 577         if (errno == EPERM && in_container) {
 578                 /* If we are in a container, and we lacked
 579                  * CAP_SYS_BOOT just exit, this will kill our
 580                  * container for good. */
 581                 log_info("Exiting container.");
 582                 return EXIT_SUCCESS;
 583         }
 584
 585         r = log_error_errno(errno, "Failed to invoke reboot(): %m");
 586
 587   error:
 588         log_emergency_errno(r, "Critical error while doing system shutdown: %m");
 589         freeze();
 590 }