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   systemd is free software; you can redistribute it and/or modify it
   8   under the terms of the GNU Lesser General Public License as published by
   9   the Free Software Foundation; either version 2.1 of the License, or
  10   (at your option) any later version.
  11
  12   systemd is distributed in the hope that it will be useful, but
  13   WITHOUT ANY WARRANTY; without even the implied warranty of
  14   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  15   Lesser General Public License for more details.
  16
  17   You should have received a copy of the GNU Lesser General Public License
  18   along with systemd; If not, see <http://www.gnu.org/licenses/>.
  19 ***/
  20
  21 #include <errno.h>
  22 #include <getopt.h>
  23 #include <linux/reboot.h>
  24 #include <signal.h>
  25 #include <stdbool.h>
  26 #include <stdlib.h>
  27 #include <sys/mman.h>
  28 #include <sys/mount.h>
  29 #include <sys/reboot.h>
  30 #include <sys/stat.h>
  31 #include <unistd.h>
  32
  33 #include "alloc-util.h"
  34 #include "async.h"
  35 #include "cgroup-util.h"
  36 #include "def.h"
  37 #include "exec-util.h"
  38 #include "fd-util.h"
  39 #include "fileio.h"
  40 #include "killall.h"
  41 #include "log.h"
  42 #include "missing.h"
  43 #include "parse-util.h"
  44 #include "process-util.h"
  45 #include "reboot-util.h"
  46 #include "signal-util.h"
  47 #include "string-util.h"
  48 #include "switch-root.h"
  49 #include "terminal-util.h"
  50 #include "umount.h"
  51 #include "util.h"
  52 #include "virt.h"
  53 #include "watchdog.h"
  54
  55 #define FINALIZE_ATTEMPTS 50
  56
  57 #define SYNC_PROGRESS_ATTEMPTS 3
  58 #define SYNC_TIMEOUT_USEC (10*USEC_PER_SEC)
  59
  60 static char* arg_verb;
  61 static uint8_t arg_exit_code;
  62 static usec_t arg_timeout = DEFAULT_TIMEOUT_USEC;
  63
  64 static int parse_argv(int argc, char *argv[]) {
  65         enum {
  66                 ARG_LOG_LEVEL = 0x100,
  67                 ARG_LOG_TARGET,
  68                 ARG_LOG_COLOR,
  69                 ARG_LOG_LOCATION,
  70                 ARG_EXIT_CODE,
  71                 ARG_TIMEOUT,
  72         };
  73
  74         static const struct option options[] = {
  75                 { "log-level",     required_argument, NULL, ARG_LOG_LEVEL    },
  76                 { "log-target",    required_argument, NULL, ARG_LOG_TARGET   },
  77                 { "log-color",     optional_argument, NULL, ARG_LOG_COLOR    },
  78                 { "log-location",  optional_argument, NULL, ARG_LOG_LOCATION },
  79                 { "exit-code",     required_argument, NULL, ARG_EXIT_CODE    },
  80                 { "timeout",       required_argument, NULL, ARG_TIMEOUT      },
  81                 {}
  82         };
  83
  84         int c, r;
  85
  86         assert(argc >= 1);
  87         assert(argv);
  88
  89         /* "-" prevents getopt from permuting argv[] and moving the verb away
  90          * from argv[1]. Our interface to initrd promises it'll be there. */
  91         while ((c = getopt_long(argc, argv, "-", options, NULL)) >= 0)
  92                 switch (c) {
  93
  94                 case ARG_LOG_LEVEL:
  95                         r = log_set_max_level_from_string(optarg);
  96                         if (r < 0)
  97                                 log_error_errno(r, "Failed to parse log level %s, ignoring.", optarg);
  98
  99                         break;
 100
 101                 case ARG_LOG_TARGET:
 102                         r = log_set_target_from_string(optarg);
 103                         if (r < 0)
 104                                 log_error_errno(r, "Failed to parse log target %s, ignoring", optarg);
 105
 106                         break;
 107
 108                 case ARG_LOG_COLOR:
 109
 110                         if (optarg) {
 111                                 r = log_show_color_from_string(optarg);
 112                                 if (r < 0)
 113                                         log_error_errno(r, "Failed to parse log color setting %s, ignoring", optarg);
 114                         } else
 115                                 log_show_color(true);
 116
 117                         break;
 118
 119                 case ARG_LOG_LOCATION:
 120                         if (optarg) {
 121                                 r = log_show_location_from_string(optarg);
 122                                 if (r < 0)
 123                                         log_error_errno(r, "Failed to parse log location setting %s, ignoring", optarg);
 124                         } else
 125                                 log_show_location(true);
 126
 127                         break;
 128
 129                 case ARG_EXIT_CODE:
 130                         r = safe_atou8(optarg, &arg_exit_code);
 131                         if (r < 0)
 132                                 log_error_errno(r, "Failed to parse exit code %s, ignoring", optarg);
 133
 134                         break;
 135
 136                 case ARG_TIMEOUT:
 137                         r = parse_sec(optarg, &arg_timeout);
 138                         if (r < 0)
 139                                 log_error_errno(r, "Failed to parse shutdown timeout %s, ignoring", optarg);
 140
 141                         break;
 142
 143                 case '\001':
 144                         if (!arg_verb)
 145                                 arg_verb = optarg;
 146                         else
 147                                 log_error("Excess arguments, ignoring");
 148                         break;
 149
 150                 case '?':
 151                         return -EINVAL;
 152
 153                 default:
 154                         assert_not_reached("Unhandled option code.");
 155                 }
 156
 157         if (!arg_verb) {
 158                 log_error("Verb argument missing.");
 159                 return -EINVAL;
 160         }
 161
 162         return 0;
 163 }
 164
 165 static int switch_root_initramfs(void) {
 166         if (mount("/run/initramfs", "/run/initramfs", NULL, MS_BIND, NULL) < 0)
 167                 return log_error_errno(errno, "Failed to mount bind /run/initramfs on /run/initramfs: %m");
 168
 169         if (mount(NULL, "/run/initramfs", NULL, MS_PRIVATE, NULL) < 0)
 170                 return log_error_errno(errno, "Failed to make /run/initramfs private mount: %m");
 171
 172         /* switch_root with MS_BIND, because there might still be processes lurking around, which have open file descriptors.
 173          * /run/initramfs/shutdown will take care of these.
 174          * Also do not detach the old root, because /run/initramfs/shutdown needs to access it.
 175          */
 176         return switch_root("/run/initramfs", "/oldroot", false, MS_BIND);
 177 }
 178
 179 /* Read the following fields from /proc/meminfo:
 180  *
 181  *  NFS_Unstable
 182  *  Writeback
 183  *  Dirty
 184  *
 185  * Return true if the sum of these fields is greater than the previous
 186  * value input. For all other issues, report the failure and indicate that
 187  * the sync is not making progress.
 188  */
 189 static bool sync_making_progress(unsigned long long *prev_dirty) {
 190         _cleanup_fclose_ FILE *f = NULL;
 191         char line[LINE_MAX];
 192         bool r = false;
 193         unsigned long long val = 0;
 194
 195         f = fopen("/proc/meminfo", "re");
 196         if (!f)
 197                 return log_warning_errno(errno, "Failed to open /proc/meminfo: %m");
 198
 199         FOREACH_LINE(line, f, log_warning_errno(errno, "Failed to parse /proc/meminfo: %m")) {
 200                 unsigned long long ull = 0;
 201
 202                 if (!first_word(line, "NFS_Unstable:") && !first_word(line, "Writeback:") && !first_word(line, "Dirty:"))
 203                         continue;
 204
 205                 errno = 0;
 206                 if (sscanf(line, "%*s %llu %*s", &ull) != 1) {
 207                         if (errno != 0)
 208                                 log_warning_errno(errno, "Failed to parse /proc/meminfo: %m");
 209                         else
 210                                 log_warning("Failed to parse /proc/meminfo");
 211
 212                         return false;
 213                 }
 214
 215                 val += ull;
 216         }
 217
 218         r = *prev_dirty > val;
 219
 220         *prev_dirty = val;
 221
 222         return r;
 223 }
 224
 225 static void sync_with_progress(void) {
 226         unsigned long long dirty = ULONG_LONG_MAX;
 227         unsigned checks;
 228         pid_t pid;
 229         int r;
 230
 231         BLOCK_SIGNALS(SIGCHLD);
 232
 233         /* Due to the possiblity of the sync operation hanging, we fork a child process and monitor the progress. If
 234          * the timeout lapses, the assumption is that that particular sync stalled. */
 235
 236         r = asynchronous_sync(&pid);
 237         if (r < 0) {
 238                 log_error_errno(r, "Failed to fork sync(): %m");
 239                 return;
 240         }
 241
 242         log_info("Syncing filesystems and block devices.");
 243
 244         /* Start monitoring the sync operation. If more than
 245          * SYNC_PROGRESS_ATTEMPTS lapse without progress being made,
 246          * we assume that the sync is stalled */
 247         for (checks = 0; checks < SYNC_PROGRESS_ATTEMPTS; checks++) {
 248                 r = wait_for_terminate_with_timeout(pid, SYNC_TIMEOUT_USEC);
 249                 if (r == 0)
 250                         /* Sync finished without error.
 251                          * (The sync itself does not return an error code) */
 252                         return;
 253                 else if (r == -ETIMEDOUT) {
 254                         /* Reset the check counter if the "Dirty" value is
 255                          * decreasing */
 256                         if (sync_making_progress(&dirty))
 257                                 checks = 0;
 258                 } else {
 259                         log_error_errno(r, "Failed to sync filesystems and block devices: %m");
 260                         return;
 261                 }
 262         }
 263
 264         /* Only reached in the event of a timeout. We should issue a kill
 265          * to the stray process. */
 266         log_error("Syncing filesystems and block devices - timed out, issuing SIGKILL to PID "PID_FMT".", pid);
 267         (void) kill(pid, SIGKILL);
 268 }
 269
 270 int main(int argc, char *argv[]) {
 271         bool need_umount, need_swapoff, need_loop_detach, need_dm_detach;
 272         bool in_container, use_watchdog = false;
 273         _cleanup_free_ char *cgroup = NULL;
 274         char *arguments[3];
 275         unsigned retries;
 276         int cmd, r;
 277         static const char* const dirs[] = {SYSTEM_SHUTDOWN_PATH, NULL};
 278         char *watchdog_device;
 279
 280         /* The log target defaults to console, but the original systemd process will pass its log target in through a
 281          * command line argument, which will override this default. Also, ensure we'll never log to the journal or
 282          * syslog, as these logging daemons are either already dead or will die very soon. */
 283
 284         log_set_target(LOG_TARGET_CONSOLE);
 285         log_set_prohibit_ipc(true);
 286         log_parse_environment();
 287
 288         r = parse_argv(argc, argv);
 289         if (r < 0)
 290                 goto error;
 291
 292         log_open();
 293
 294         umask(0022);
 295
 296         if (getpid_cached() != 1) {
 297                 log_error("Not executed by init (PID 1).");
 298                 r = -EPERM;
 299                 goto error;
 300         }
 301
 302         if (streq(arg_verb, "reboot"))
 303                 cmd = RB_AUTOBOOT;
 304         else if (streq(arg_verb, "poweroff"))
 305                 cmd = RB_POWER_OFF;
 306         else if (streq(arg_verb, "halt"))
 307                 cmd = RB_HALT_SYSTEM;
 308         else if (streq(arg_verb, "kexec"))
 309                 cmd = LINUX_REBOOT_CMD_KEXEC;
 310         else if (streq(arg_verb, "exit"))
 311                 cmd = 0; /* ignored, just checking that arg_verb is valid */
 312         else {
 313                 log_error("Unknown action '%s'.", arg_verb);
 314                 r = -EINVAL;
 315                 goto error;
 316         }
 317
 318         (void) cg_get_root_path(&cgroup);
 319         in_container = detect_container() > 0;
 320
 321         use_watchdog = !!getenv("WATCHDOG_USEC");
 322         watchdog_device = getenv("WATCHDOG_DEVICE");
 323         if (watchdog_device) {
 324                 r = watchdog_set_device(watchdog_device);
 325                 if (r < 0)
 326                         log_warning_errno(r, "Failed to set watchdog device to %s, ignoring: %m",
 327                                           watchdog_device);
 328         }
 329
 330         /* Lock us into memory */
 331         (void) mlockall(MCL_CURRENT|MCL_FUTURE);
 332
 333         /* Synchronize everything that is not written to disk yet at this point already. This is a good idea so that
 334          * slow IO is processed here already and the final process killing spree is not impacted by processes
 335          * desperately trying to sync IO to disk within their timeout. Do not remove this sync, data corruption will
 336          * result. */
 337         if (!in_container)
 338                 sync_with_progress();
 339
 340         disable_coredumps();
 341
 342         log_info("Sending SIGTERM to remaining processes...");
 343         broadcast_signal(SIGTERM, true, true, arg_timeout);
 344
 345         log_info("Sending SIGKILL to remaining processes...");
 346         broadcast_signal(SIGKILL, true, false, arg_timeout);
 347
 348         need_umount = !in_container;
 349         need_swapoff = !in_container;
 350         need_loop_detach = !in_container;
 351         need_dm_detach = !in_container;
 352
 353         /* Unmount all mountpoints, swaps, and loopback devices */
 354         for (retries = 0; retries < FINALIZE_ATTEMPTS; retries++) {
 355                 bool changed = false;
 356
 357                 if (use_watchdog)
 358                         watchdog_ping();
 359
 360                 /* Let's trim the cgroup tree on each iteration so
 361                    that we leave an empty cgroup tree around, so that
 362                    container managers get a nice notify event when we
 363                    are down */
 364                 if (cgroup)
 365                         cg_trim(SYSTEMD_CGROUP_CONTROLLER, cgroup, false);
 366
 367                 if (need_umount) {
 368                         log_info("Unmounting file systems.");
 369                         r = umount_all(&changed);
 370                         if (r == 0) {
 371                                 need_umount = false;
 372                                 log_info("All filesystems unmounted.");
 373                         } else if (r > 0)
 374                                 log_info("Not all file systems unmounted, %d left.", r);
 375                         else
 376                                 log_error_errno(r, "Failed to unmount file systems: %m");
 377                 }
 378
 379                 if (need_swapoff) {
 380                         log_info("Deactivating swaps.");
 381                         r = swapoff_all(&changed);
 382                         if (r == 0) {
 383                                 need_swapoff = false;
 384                                 log_info("All swaps deactivated.");
 385                         } else if (r > 0)
 386                                 log_info("Not all swaps deactivated, %d left.", r);
 387                         else
 388                                 log_error_errno(r, "Failed to deactivate swaps: %m");
 389                 }
 390
 391                 if (need_loop_detach) {
 392                         log_info("Detaching loop devices.");
 393                         r = loopback_detach_all(&changed);
 394                         if (r == 0) {
 395                                 need_loop_detach = false;
 396                                 log_info("All loop devices detached.");
 397                         } else if (r > 0)
 398                                 log_info("Not all loop devices detached, %d left.", r);
 399                         else
 400                                 log_error_errno(r, "Failed to detach loop devices: %m");
 401                 }
 402
 403                 if (need_dm_detach) {
 404                         log_info("Detaching DM devices.");
 405                         r = dm_detach_all(&changed);
 406                         if (r == 0) {
 407                                 need_dm_detach = false;
 408                                 log_info("All DM devices detached.");
 409                         } else if (r > 0)
 410                                 log_info("Not all DM devices detached, %d left.", r);
 411                         else
 412                                 log_error_errno(r, "Failed to detach DM devices: %m");
 413                 }
 414
 415                 if (!need_umount && !need_swapoff && !need_loop_detach && !need_dm_detach) {
 416                         if (retries > 0)
 417                                 log_info("All filesystems, swaps, loop devices, DM devices detached.");
 418                         /* Yay, done */
 419                         goto initrd_jump;
 420                 }
 421
 422                 /* If in this iteration we didn't manage to
 423                  * unmount/deactivate anything, we simply give up */
 424                 if (!changed) {
 425                         log_info("Cannot finalize remaining%s%s%s%s continuing.",
 426                                  need_umount ? " file systems," : "",
 427                                  need_swapoff ? " swap devices," : "",
 428                                  need_loop_detach ? " loop devices," : "",
 429                                  need_dm_detach ? " DM devices," : "");
 430                         goto initrd_jump;
 431                 }
 432
 433                 log_debug("After %u retries, couldn't finalize remaining %s%s%s%s trying again.",
 434                           retries + 1,
 435                           need_umount ? " file systems," : "",
 436                           need_swapoff ? " swap devices," : "",
 437                           need_loop_detach ? " loop devices," : "",
 438                           need_dm_detach ? " DM devices," : "");
 439         }
 440
 441         log_error("Too many iterations, giving up.");
 442
 443  initrd_jump:
 444
 445         /* We're done with the watchdog. */
 446         watchdog_free_device();
 447
 448         arguments[0] = NULL;
 449         arguments[1] = arg_verb;
 450         arguments[2] = NULL;
 451         execute_directories(dirs, DEFAULT_TIMEOUT_USEC, NULL, NULL, arguments);
 452
 453         if (!in_container && !in_initrd() &&
 454             access("/run/initramfs/shutdown", X_OK) == 0) {
 455                 r = switch_root_initramfs();
 456                 if (r >= 0) {
 457                         argv[0] = (char*) "/shutdown";
 458
 459                         setsid();
 460                         make_console_stdio();
 461
 462                         log_info("Successfully changed into root pivot.\n"
 463                                  "Returning to initrd...");
 464
 465                         execv("/shutdown", argv);
 466                         log_error_errno(errno, "Failed to execute shutdown binary: %m");
 467                 } else
 468                         log_error_errno(r, "Failed to switch root to \"/run/initramfs\": %m");
 469
 470         }
 471
 472         if (need_umount || need_swapoff || need_loop_detach || need_dm_detach)
 473                 log_error("Failed to finalize %s%s%s%s ignoring",
 474                           need_umount ? " file systems," : "",
 475                           need_swapoff ? " swap devices," : "",
 476                           need_loop_detach ? " loop devices," : "",
 477                           need_dm_detach ? " DM devices," : "");
 478
 479         /* The kernel will automatically flush ATA disks and suchlike on reboot(), but the file systems need to be
 480          * sync'ed explicitly in advance. So let's do this here, but not needlessly slow down containers. Note that we
 481          * sync'ed things already once above, but we did some more work since then which might have caused IO, hence
 482          * let's do it once more. Do not remove this sync, data corruption will result. */
 483         if (!in_container)
 484                 sync_with_progress();
 485
 486         if (streq(arg_verb, "exit")) {
 487                 if (in_container)
 488                         return arg_exit_code;
 489
 490                 cmd = RB_POWER_OFF; /* We cannot exit() on the host, fallback on another method. */
 491         }
 492
 493         switch (cmd) {
 494
 495         case LINUX_REBOOT_CMD_KEXEC:
 496
 497                 if (!in_container) {
 498                         /* We cheat and exec kexec to avoid doing all its work */
 499                         log_info("Rebooting with kexec.");
 500
 501                         r = safe_fork("(sd-kexec)", FORK_RESET_SIGNALS|FORK_CLOSE_ALL_FDS|FORK_LOG|FORK_WAIT, NULL);
 502                         if (r == 0) {
 503                                 const char * const args[] = {
 504                                         KEXEC, "-e", NULL
 505                                 };
 506
 507                                 /* Child */
 508
 509                                 execv(args[0], (char * const *) args);
 510                                 _exit(EXIT_FAILURE);
 511                         }
 512
 513                         /* If we are still running, then the kexec can't have worked, let's fall through */
 514                 }
 515
 516                 cmd = RB_AUTOBOOT;
 517                 _fallthrough_;
 518
 519         case RB_AUTOBOOT:
 520                 (void) reboot_with_parameter(REBOOT_LOG);
 521                 log_info("Rebooting.");
 522                 break;
 523
 524         case RB_POWER_OFF:
 525                 log_info("Powering off.");
 526                 break;
 527
 528         case RB_HALT_SYSTEM:
 529                 log_info("Halting system.");
 530                 break;
 531
 532         default:
 533                 assert_not_reached("Unknown magic");
 534         }
 535
 536         (void) reboot(cmd);
 537         if (errno == EPERM && in_container) {
 538                 /* If we are in a container, and we lacked
 539                  * CAP_SYS_BOOT just exit, this will kill our
 540                  * container for good. */
 541                 log_info("Exiting container.");
 542                 return EXIT_SUCCESS;
 543         }
 544
 545         r = log_error_errno(errno, "Failed to invoke reboot(): %m");
 546
 547   error:
 548         log_emergency_errno(r, "Critical error while doing system shutdown: %m");
 549         freeze();
 550 }