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 "exec-util.h"
  36 #include "fileio.h"
  37 #include "killall.h"
  38 #include "log.h"
  39 #include "missing.h"
  40 #include "parse-util.h"
  41 #include "process-util.h"
  42 #include "string-util.h"
  43 #include "switch-root.h"
  44 #include "terminal-util.h"
  45 #include "umount.h"
  46 #include "util.h"
  47 #include "virt.h"
  48 #include "watchdog.h"
  49
  50 #define FINALIZE_ATTEMPTS 50
  51
  52 static char* arg_verb;
  53 static uint8_t arg_exit_code;
  54
  55 static int parse_argv(int argc, char *argv[]) {
  56         enum {
  57                 ARG_LOG_LEVEL = 0x100,
  58                 ARG_LOG_TARGET,
  59                 ARG_LOG_COLOR,
  60                 ARG_LOG_LOCATION,
  61                 ARG_EXIT_CODE,
  62         };
  63
  64         static const struct option options[] = {
  65                 { "log-level",     required_argument, NULL, ARG_LOG_LEVEL    },
  66                 { "log-target",    required_argument, NULL, ARG_LOG_TARGET   },
  67                 { "log-color",     optional_argument, NULL, ARG_LOG_COLOR    },
  68                 { "log-location",  optional_argument, NULL, ARG_LOG_LOCATION },
  69                 { "exit-code",     required_argument, NULL, ARG_EXIT_CODE    },
  70                 {}
  71         };
  72
  73         int c, r;
  74
  75         assert(argc >= 1);
  76         assert(argv);
  77
  78         /* "-" prevents getopt from permuting argv[] and moving the verb away
  79          * from argv[1]. Our interface to initrd promises it'll be there. */
  80         while ((c = getopt_long(argc, argv, "-", options, NULL)) >= 0)
  81                 switch (c) {
  82
  83                 case ARG_LOG_LEVEL:
  84                         r = log_set_max_level_from_string(optarg);
  85                         if (r < 0)
  86                                 log_error("Failed to parse log level %s, ignoring.", optarg);
  87
  88                         break;
  89
  90                 case ARG_LOG_TARGET:
  91                         r = log_set_target_from_string(optarg);
  92                         if (r < 0)
  93                                 log_error("Failed to parse log target %s, ignoring", optarg);
  94
  95                         break;
  96
  97                 case ARG_LOG_COLOR:
  98
  99                         if (optarg) {
 100                                 r = log_show_color_from_string(optarg);
 101                                 if (r < 0)
 102                                         log_error("Failed to parse log color setting %s, ignoring", optarg);
 103                         } else
 104                                 log_show_color(true);
 105
 106                         break;
 107
 108                 case ARG_LOG_LOCATION:
 109                         if (optarg) {
 110                                 r = log_show_location_from_string(optarg);
 111                                 if (r < 0)
 112                                         log_error("Failed to parse log location setting %s, ignoring", optarg);
 113                         } else
 114                                 log_show_location(true);
 115
 116                         break;
 117
 118                 case ARG_EXIT_CODE:
 119                         r = safe_atou8(optarg, &arg_exit_code);
 120                         if (r < 0)
 121                                 log_error("Failed to parse exit code %s, ignoring", optarg);
 122
 123                         break;
 124
 125                 case '\001':
 126                         if (!arg_verb)
 127                                 arg_verb = optarg;
 128                         else
 129                                 log_error("Excess arguments, ignoring");
 130                         break;
 131
 132                 case '?':
 133                         return -EINVAL;
 134
 135                 default:
 136                         assert_not_reached("Unhandled option code.");
 137                 }
 138
 139         if (!arg_verb) {
 140                 log_error("Verb argument missing.");
 141                 return -EINVAL;
 142         }
 143
 144         return 0;
 145 }
 146
 147 static int switch_root_initramfs(void) {
 148         if (mount("/run/initramfs", "/run/initramfs", NULL, MS_BIND, NULL) < 0)
 149                 return log_error_errno(errno, "Failed to mount bind /run/initramfs on /run/initramfs: %m");
 150
 151         if (mount(NULL, "/run/initramfs", NULL, MS_PRIVATE, NULL) < 0)
 152                 return log_error_errno(errno, "Failed to make /run/initramfs private mount: %m");
 153
 154         /* switch_root with MS_BIND, because there might still be processes lurking around, which have open file descriptors.
 155          * /run/initramfs/shutdown will take care of these.
 156          * Also do not detach the old root, because /run/initramfs/shutdown needs to access it.
 157          */
 158         return switch_root("/run/initramfs", "/oldroot", false, MS_BIND);
 159 }
 160
 161 int main(int argc, char *argv[]) {
 162         bool need_umount, need_swapoff, need_loop_detach, need_dm_detach;
 163         bool in_container, use_watchdog = false;
 164         _cleanup_free_ char *cgroup = NULL;
 165         char *arguments[3];
 166         unsigned retries;
 167         int cmd, r;
 168         static const char* const dirs[] = {SYSTEM_SHUTDOWN_PATH, NULL};
 169
 170         log_parse_environment();
 171         r = parse_argv(argc, argv);
 172         if (r < 0)
 173                 goto error;
 174
 175         /* journald will die if not gone yet. The log target defaults
 176          * to console, but may have been changed by command line options. */
 177
 178         log_close_console(); /* force reopen of /dev/console */
 179         log_open();
 180
 181         umask(0022);
 182
 183         if (getpid() != 1) {
 184                 log_error("Not executed by init (PID 1).");
 185                 r = -EPERM;
 186                 goto error;
 187         }
 188
 189         if (streq(arg_verb, "reboot"))
 190                 cmd = RB_AUTOBOOT;
 191         else if (streq(arg_verb, "poweroff"))
 192                 cmd = RB_POWER_OFF;
 193         else if (streq(arg_verb, "halt"))
 194                 cmd = RB_HALT_SYSTEM;
 195         else if (streq(arg_verb, "kexec"))
 196                 cmd = LINUX_REBOOT_CMD_KEXEC;
 197         else if (streq(arg_verb, "exit"))
 198                 cmd = 0; /* ignored, just checking that arg_verb is valid */
 199         else {
 200                 r = -EINVAL;
 201                 log_error("Unknown action '%s'.", arg_verb);
 202                 goto error;
 203         }
 204
 205         (void) cg_get_root_path(&cgroup);
 206         in_container = detect_container() > 0;
 207
 208         use_watchdog = !!getenv("WATCHDOG_USEC");
 209
 210         /* Lock us into memory */
 211         mlockall(MCL_CURRENT|MCL_FUTURE);
 212
 213         /* Synchronize everything that is not written to disk yet at this point already. This is a good idea so that
 214          * slow IO is processed here already and the final process killing spree is not impacted by processes
 215          * desperately trying to sync IO to disk within their timeout. */
 216         if (!in_container)
 217                 sync();
 218
 219         log_info("Sending SIGTERM to remaining processes...");
 220         broadcast_signal(SIGTERM, true, true);
 221
 222         log_info("Sending SIGKILL to remaining processes...");
 223         broadcast_signal(SIGKILL, true, false);
 224
 225         need_umount = !in_container;
 226         need_swapoff = !in_container;
 227         need_loop_detach = !in_container;
 228         need_dm_detach = !in_container;
 229
 230         /* Unmount all mountpoints, swaps, and loopback devices */
 231         for (retries = 0; retries < FINALIZE_ATTEMPTS; retries++) {
 232                 bool changed = false;
 233
 234                 if (use_watchdog)
 235                         watchdog_ping();
 236
 237                 /* Let's trim the cgroup tree on each iteration so
 238                    that we leave an empty cgroup tree around, so that
 239                    container managers get a nice notify event when we
 240                    are down */
 241                 if (cgroup)
 242                         cg_trim(SYSTEMD_CGROUP_CONTROLLER, cgroup, false);
 243
 244                 if (need_umount) {
 245                         log_info("Unmounting file systems.");
 246                         r = umount_all(&changed);
 247                         if (r == 0) {
 248                                 need_umount = false;
 249                                 log_info("All filesystems unmounted.");
 250                         } else if (r > 0)
 251                                 log_info("Not all file systems unmounted, %d left.", r);
 252                         else
 253                                 log_error_errno(r, "Failed to unmount file systems: %m");
 254                 }
 255
 256                 if (need_swapoff) {
 257                         log_info("Deactivating swaps.");
 258                         r = swapoff_all(&changed);
 259                         if (r == 0) {
 260                                 need_swapoff = false;
 261                                 log_info("All swaps deactivated.");
 262                         } else if (r > 0)
 263                                 log_info("Not all swaps deactivated, %d left.", r);
 264                         else
 265                                 log_error_errno(r, "Failed to deactivate swaps: %m");
 266                 }
 267
 268                 if (need_loop_detach) {
 269                         log_info("Detaching loop devices.");
 270                         r = loopback_detach_all(&changed);
 271                         if (r == 0) {
 272                                 need_loop_detach = false;
 273                                 log_info("All loop devices detached.");
 274                         } else if (r > 0)
 275                                 log_info("Not all loop devices detached, %d left.", r);
 276                         else
 277                                 log_error_errno(r, "Failed to detach loop devices: %m");
 278                 }
 279
 280                 if (need_dm_detach) {
 281                         log_info("Detaching DM devices.");
 282                         r = dm_detach_all(&changed);
 283                         if (r == 0) {
 284                                 need_dm_detach = false;
 285                                 log_info("All DM devices detached.");
 286                         } else if (r > 0)
 287                                 log_info("Not all DM devices detached, %d left.", r);
 288                         else
 289                                 log_error_errno(r, "Failed to detach DM devices: %m");
 290                 }
 291
 292                 if (!need_umount && !need_swapoff && !need_loop_detach && !need_dm_detach) {
 293                         if (retries > 0)
 294                                 log_info("All filesystems, swaps, loop devices, DM devices detached.");
 295                         /* Yay, done */
 296                         goto initrd_jump;
 297                 }
 298
 299                 /* If in this iteration we didn't manage to
 300                  * unmount/deactivate anything, we simply give up */
 301                 if (!changed) {
 302                         log_info("Cannot finalize remaining%s%s%s%s continuing.",
 303                                  need_umount ? " file systems," : "",
 304                                  need_swapoff ? " swap devices," : "",
 305                                  need_loop_detach ? " loop devices," : "",
 306                                  need_dm_detach ? " DM devices," : "");
 307                         goto initrd_jump;
 308                 }
 309
 310                 log_debug("After %u retries, couldn't finalize remaining %s%s%s%s trying again.",
 311                           retries + 1,
 312                           need_umount ? " file systems," : "",
 313                           need_swapoff ? " swap devices," : "",
 314                           need_loop_detach ? " loop devices," : "",
 315                           need_dm_detach ? " DM devices," : "");
 316         }
 317
 318         log_error("Too many iterations, giving up.");
 319
 320  initrd_jump:
 321
 322         arguments[0] = NULL;
 323         arguments[1] = arg_verb;
 324         arguments[2] = NULL;
 325         execute_directories(dirs, DEFAULT_TIMEOUT_USEC, NULL, NULL, arguments);
 326
 327         if (!in_container && !in_initrd() &&
 328             access("/run/initramfs/shutdown", X_OK) == 0) {
 329                 r = switch_root_initramfs();
 330                 if (r >= 0) {
 331                         argv[0] = (char*) "/shutdown";
 332
 333                         setsid();
 334                         make_console_stdio();
 335
 336                         log_info("Successfully changed into root pivot.\n"
 337                                  "Returning to initrd...");
 338
 339                         execv("/shutdown", argv);
 340                         log_error_errno(errno, "Failed to execute shutdown binary: %m");
 341                 } else
 342                         log_error_errno(r, "Failed to switch root to \"/run/initramfs\": %m");
 343
 344         }
 345
 346         if (need_umount || need_swapoff || need_loop_detach || need_dm_detach)
 347                 log_error("Failed to finalize %s%s%s%s ignoring",
 348                           need_umount ? " file systems," : "",
 349                           need_swapoff ? " swap devices," : "",
 350                           need_loop_detach ? " loop devices," : "",
 351                           need_dm_detach ? " DM devices," : "");
 352
 353         /* The kernel will automatically flush ATA disks and suchlike on reboot(), but the file systems need to be
 354          * sync'ed explicitly in advance. So let's do this here, but not needlessly slow down containers. Note that we
 355          * sync'ed things already once above, but we did some more work since then which might have caused IO, hence
 356          * let's doit once more. */
 357         if (!in_container)
 358                 sync();
 359
 360         if (streq(arg_verb, "exit")) {
 361                 if (in_container)
 362                         exit(arg_exit_code);
 363                 else {
 364                         /* We cannot exit() on the host, fallback on another
 365                          * method. */
 366                         cmd = RB_POWER_OFF;
 367                 }
 368         }
 369
 370         switch (cmd) {
 371
 372         case LINUX_REBOOT_CMD_KEXEC:
 373
 374                 if (!in_container) {
 375                         /* We cheat and exec kexec to avoid doing all its work */
 376                         pid_t pid;
 377
 378                         log_info("Rebooting with kexec.");
 379
 380                         pid = fork();
 381                         if (pid < 0)
 382                                 log_error_errno(errno, "Failed to fork: %m");
 383                         else if (pid == 0) {
 384
 385                                 const char * const args[] = {
 386                                         KEXEC, "-e", NULL
 387                                 };
 388
 389                                 /* Child */
 390
 391                                 execv(args[0], (char * const *) args);
 392                                 _exit(EXIT_FAILURE);
 393                         } else
 394                                 wait_for_terminate_and_warn("kexec", pid, true);
 395                 }
 396
 397                 cmd = RB_AUTOBOOT;
 398                 /* Fall through */
 399
 400         case RB_AUTOBOOT:
 401
 402                 if (!in_container) {
 403                         _cleanup_free_ char *param = NULL;
 404
 405                         r = read_one_line_file("/run/systemd/reboot-param", &param);
 406                         if (r < 0)
 407                                 log_warning_errno(r, "Failed to read reboot parameter file: %m");
 408
 409                         if (!isempty(param)) {
 410                                 log_info("Rebooting with argument '%s'.", param);
 411                                 syscall(SYS_reboot, LINUX_REBOOT_MAGIC1, LINUX_REBOOT_MAGIC2, LINUX_REBOOT_CMD_RESTART2, param);
 412                                 log_warning_errno(errno, "Failed to reboot with parameter, retrying without: %m");
 413                         }
 414                 }
 415
 416                 log_info("Rebooting.");
 417                 break;
 418
 419         case RB_POWER_OFF:
 420                 log_info("Powering off.");
 421                 break;
 422
 423         case RB_HALT_SYSTEM:
 424                 log_info("Halting system.");
 425                 break;
 426
 427         default:
 428                 assert_not_reached("Unknown magic");
 429         }
 430
 431         reboot(cmd);
 432         if (errno == EPERM && in_container) {
 433                 /* If we are in a container, and we lacked
 434                  * CAP_SYS_BOOT just exit, this will kill our
 435                  * container for good. */
 436                 log_info("Exiting container.");
 437                 exit(0);
 438         }
 439
 440         r = log_error_errno(errno, "Failed to invoke reboot(): %m");
 441
 442   error:
 443         log_emergency_errno(r, "Critical error while doing system shutdown: %m");
 444         freeze();
 445 }