src/lxc/criu.c

   1 /*
   2  * lxc: linux Container library
   3  *
   4  * Copyright © 2014-2015 Canonical Ltd.
   5  *
   6  * Authors:
   7  * Tycho Andersen <tycho.andersen@canonical.com>
   8  *
   9  * This library is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public
  11  * License as published by the Free Software Foundation; either
  12  * version 2.1 of the License, or (at your option) any later version.
  13  *
  14  * This library is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with this library; if not, write to the Free Software
  21  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  22  */
  23 #define _GNU_SOURCE
  24 #include <assert.h>
  25 #include <linux/limits.h>
  26 #include <sched.h>
  27 #include <stdio.h>
  28 #include <stdlib.h>
  29 #include <string.h>
  30 #include <sys/mount.h>
  31 #include <sys/types.h>
  32 #include <sys/wait.h>
  33 #include <unistd.h>
  34
  35 #include "config.h"
  36
  37 #include "bdev/bdev.h"
  38 #include "cgroup.h"
  39 #include "conf.h"
  40 #include "commands.h"
  41 #include "criu.h"
  42 #include "log.h"
  43 #include "lxc.h"
  44 #include "lxclock.h"
  45 #include "network.h"
  46 #include "utils.h"
  47
  48 #define CRIU_VERSION            "2.0"
  49
  50 #define CRIU_GITID_VERSION      "2.0"
  51 #define CRIU_GITID_PATCHLEVEL   0
  52
  53 lxc_log_define(lxc_criu, lxc);
  54
  55 struct criu_opts {
  56         /* The type of criu invocation, one of "dump" or "restore" */
  57         char *action;
  58
  59         /* The directory to pass to criu */
  60         char *directory;
  61
  62         /* The container to dump */
  63         struct lxc_container *c;
  64
  65         /* Enable criu verbose mode? */
  66         bool verbose;
  67
  68         /* (pre-)dump: a directory for the previous dump's images */
  69         char *predump_dir;
  70
  71         /* dump: stop the container or not after dumping? */
  72         bool stop;
  73         char tty_id[32]; /* the criu tty id for /dev/console, i.e. "tty[${rdev}:${dev}]" */
  74
  75         /* restore: the file to write the init process' pid into */
  76         char *pidfile;
  77         const char *cgroup_path;
  78         int console_fd;
  79         /* The path that is bind mounted from /dev/console, if any. We don't
  80          * want to use `--ext-mount-map auto`'s result here because the pts
  81          * device may have a different path (e.g. if the pty number is
  82          * different) on the target host. NULL if lxc.console = "none".
  83          */
  84         char *console_name;
  85 };
  86
  87 static int load_tty_major_minor(char *directory, char *output, int len)
  88 {
  89         FILE *f;
  90         char path[PATH_MAX];
  91         int ret;
  92
  93         ret = snprintf(path, sizeof(path), "%s/tty.info", directory);
  94         if (ret < 0 || ret >= sizeof(path)) {
  95                 ERROR("snprintf'd too many chacters: %d", ret);
  96                 return -1;
  97         }
  98
  99         f = fopen(path, "r");
 100         if (!f) {
 101                 /* This means we're coming from a liblxc which didn't export
 102                  * the tty info. In this case they had to have lxc.console =
 103                  * none, so there's no problem restoring.
 104                  */
 105                 if (errno == ENOENT)
 106                         return 0;
 107
 108                 SYSERROR("couldn't open %s", path);
 109                 return -1;
 110         }
 111
 112         if (!fgets(output, len, f)) {
 113                 fclose(f);
 114                 SYSERROR("couldn't read %s", path);
 115                 return -1;
 116         }
 117
 118         fclose(f);
 119         return 0;
 120 }
 121
 122 static void exec_criu(struct criu_opts *opts)
 123 {
 124         char **argv, log[PATH_MAX];
 125         int static_args = 24, argc = 0, i, ret;
 126         int netnr = 0;
 127         struct lxc_list *it;
 128
 129         char buf[4096], *pos, tty_info[32];
 130
 131         /* If we are currently in a cgroup /foo/bar, and the container is in a
 132          * cgroup /lxc/foo, lxcfs will give us an ENOENT if some task in the
 133          * container has an open fd that points to one of the cgroup files
 134          * (systemd always opens its "root" cgroup). So, let's escape to the
 135          * /actual/ root cgroup so that lxcfs thinks criu has enough rights to
 136          * see all cgroups.
 137          */
 138         if (!cgroup_escape()) {
 139                 ERROR("failed to escape cgroups");
 140                 return;
 141         }
 142
 143         /* The command line always looks like:
 144          * criu $(action) --tcp-established --file-locks --link-remap --force-irmap \
 145          * --manage-cgroups action-script foo.sh -D $(directory) \
 146          * -o $(directory)/$(action).log --ext-mount-map auto
 147          * --enable-external-sharing --enable-external-masters
 148          * --enable-fs hugetlbfs --enable-fs tracefs --ext-mount-map console:/dev/pts/n
 149          * +1 for final NULL */
 150
 151         if (strcmp(opts->action, "dump") == 0 || strcmp(opts->action, "pre-dump") == 0) {
 152                 /* -t pid --freeze-cgroup /lxc/ct */
 153                 static_args += 4;
 154
 155                 /* --prev-images-dir <path-to-directory-A-relative-to-B> */
 156                 if (opts->predump_dir)
 157                         static_args += 2;
 158
 159                 /* --leave-running (only for final dump) */
 160                 if (strcmp(opts->action, "dump") == 0 && !opts->stop)
 161                         static_args++;
 162
 163                 /* --external tty[88,4] */
 164                 if (opts->tty_id[0])
 165                         static_args += 2;
 166         } else if (strcmp(opts->action, "restore") == 0) {
 167                 /* --root $(lxc_mount_point) --restore-detached
 168                  * --restore-sibling --pidfile $foo --cgroup-root $foo
 169                  * --lsm-profile apparmor:whatever
 170                  */
 171                 static_args += 10;
 172
 173                 tty_info[0] = 0;
 174                 if (load_tty_major_minor(opts->directory, tty_info, sizeof(tty_info)))
 175                         return;
 176
 177                 /* --inherit-fd fd[%d]:tty[%s] */
 178                 if (tty_info[0])
 179                         static_args += 2;
 180         } else {
 181                 return;
 182         }
 183
 184         if (opts->verbose)
 185                 static_args++;
 186
 187         ret = snprintf(log, PATH_MAX, "%s/%s.log", opts->directory, opts->action);
 188         if (ret < 0 || ret >= PATH_MAX) {
 189                 ERROR("logfile name too long\n");
 190                 return;
 191         }
 192
 193         argv = malloc(static_args * sizeof(*argv));
 194         if (!argv)
 195                 return;
 196
 197         memset(argv, 0, static_args * sizeof(*argv));
 198
 199 #define DECLARE_ARG(arg)                                        \
 200         do {                                                    \
 201                 if (arg == NULL) {                              \
 202                         ERROR("Got NULL argument for criu");    \
 203                         goto err;                               \
 204                 }                                               \
 205                 argv[argc++] = strdup(arg);                     \
 206                 if (!argv[argc-1])                              \
 207                         goto err;                               \
 208         } while (0)
 209
 210         argv[argc++] = on_path("criu", NULL);
 211         if (!argv[argc-1]) {
 212                 ERROR("Couldn't find criu binary\n");
 213                 goto err;
 214         }
 215
 216         DECLARE_ARG(opts->action);
 217         DECLARE_ARG("--tcp-established");
 218         DECLARE_ARG("--file-locks");
 219         DECLARE_ARG("--link-remap");
 220         DECLARE_ARG("--force-irmap");
 221         DECLARE_ARG("--manage-cgroups");
 222         DECLARE_ARG("--ext-mount-map");
 223         DECLARE_ARG("auto");
 224         DECLARE_ARG("--enable-external-sharing");
 225         DECLARE_ARG("--enable-external-masters");
 226         DECLARE_ARG("--enable-fs");
 227         DECLARE_ARG("hugetlbfs");
 228         DECLARE_ARG("--enable-fs");
 229         DECLARE_ARG("tracefs");
 230         DECLARE_ARG("-D");
 231         DECLARE_ARG(opts->directory);
 232         DECLARE_ARG("-o");
 233         DECLARE_ARG(log);
 234
 235         if (opts->verbose)
 236                 DECLARE_ARG("-vvvvvv");
 237
 238         if (strcmp(opts->action, "dump") == 0 || strcmp(opts->action, "pre-dump") == 0) {
 239                 char pid[32], *freezer_relative;
 240
 241                 if (sprintf(pid, "%d", opts->c->init_pid(opts->c)) < 0)
 242                         goto err;
 243
 244                 DECLARE_ARG("-t");
 245                 DECLARE_ARG(pid);
 246
 247                 freezer_relative = lxc_cmd_get_cgroup_path(opts->c->name,
 248                                                            opts->c->config_path,
 249                                                            "freezer");
 250                 if (!freezer_relative) {
 251                         ERROR("failed getting freezer path");
 252                         goto err;
 253                 }
 254
 255                 ret = snprintf(log, sizeof(log), "/sys/fs/cgroup/freezer/%s", freezer_relative);
 256                 if (ret < 0 || ret >= sizeof(log))
 257                         goto err;
 258
 259                 DECLARE_ARG("--freeze-cgroup");
 260                 DECLARE_ARG(log);
 261
 262                 DECLARE_ARG("--ext-mount-map");
 263                 DECLARE_ARG("/dev/console:console");
 264                 if (opts->tty_id[0]) {
 265                         DECLARE_ARG("--external");
 266                         DECLARE_ARG(opts->tty_id);
 267                 }
 268
 269                 if (opts->predump_dir) {
 270                         DECLARE_ARG("--prev-images-dir");
 271                         DECLARE_ARG(opts->predump_dir);
 272                 }
 273
 274                 /* only for final dump */
 275                 if (strcmp(opts->action, "dump") == 0 && !opts->stop)
 276                         DECLARE_ARG("--leave-running");
 277         } else if (strcmp(opts->action, "restore") == 0) {
 278                 void *m;
 279                 int additional;
 280                 struct lxc_conf *lxc_conf = opts->c->lxc_conf;
 281
 282                 DECLARE_ARG("--root");
 283                 DECLARE_ARG(opts->c->lxc_conf->rootfs.mount);
 284                 DECLARE_ARG("--restore-detached");
 285                 DECLARE_ARG("--restore-sibling");
 286                 DECLARE_ARG("--pidfile");
 287                 DECLARE_ARG(opts->pidfile);
 288                 DECLARE_ARG("--cgroup-root");
 289                 DECLARE_ARG(opts->cgroup_path);
 290
 291                 if (tty_info[0]) {
 292                         ret = snprintf(buf, sizeof(buf), "fd[%d]:%s", opts->console_fd, tty_info);
 293                         if (ret < 0 || ret >= sizeof(buf))
 294                                 goto err;
 295
 296                         DECLARE_ARG("--inherit-fd");
 297                         DECLARE_ARG(buf);
 298                 }
 299                 if (opts->console_name) {
 300                         if (snprintf(buf, sizeof(buf), "console:%s", opts->console_name) < 0) {
 301                                 SYSERROR("sprintf'd too many bytes");
 302                         }
 303                         DECLARE_ARG("--ext-mount-map");
 304                         DECLARE_ARG(buf);
 305                 }
 306
 307                 if (lxc_conf->lsm_aa_profile || lxc_conf->lsm_se_context) {
 308
 309                         if (lxc_conf->lsm_aa_profile)
 310                                 ret = snprintf(buf, sizeof(buf), "apparmor:%s", lxc_conf->lsm_aa_profile);
 311                         else
 312                                 ret = snprintf(buf, sizeof(buf), "selinux:%s", lxc_conf->lsm_se_context);
 313
 314                         if (ret < 0 || ret >= sizeof(buf))
 315                                 goto err;
 316
 317                         DECLARE_ARG("--lsm-profile");
 318                         DECLARE_ARG(buf);
 319                 }
 320
 321                 additional = lxc_list_len(&opts->c->lxc_conf->network) * 2;
 322
 323                 m = realloc(argv, (argc + additional + 1) * sizeof(*argv));
 324                 if (!m)
 325                         goto err;
 326                 argv = m;
 327
 328                 lxc_list_for_each(it, &opts->c->lxc_conf->network) {
 329                         char eth[128], *veth;
 330                         struct lxc_netdev *n = it->elem;
 331
 332                         if (n->type != LXC_NET_VETH)
 333                                 continue;
 334
 335                         if (n->name) {
 336                                 if (strlen(n->name) >= sizeof(eth))
 337                                         goto err;
 338                                 strncpy(eth, n->name, sizeof(eth));
 339                         } else
 340                                 sprintf(eth, "eth%d", netnr);
 341
 342                         veth = n->priv.veth_attr.pair;
 343
 344                         if (n->link)
 345                                 ret = snprintf(buf, sizeof(buf), "%s=%s@%s", eth, veth, n->link);
 346                         else
 347                                 ret = snprintf(buf, sizeof(buf), "%s=%s", eth, veth);
 348                         if (ret < 0 || ret >= sizeof(buf))
 349                                 goto err;
 350
 351                         DECLARE_ARG("--veth-pair");
 352                         DECLARE_ARG(buf);
 353                 }
 354
 355         }
 356
 357         argv[argc] = NULL;
 358
 359         buf[0] = 0;
 360         pos = buf;
 361         for (i = 0; argv[i]; i++) {
 362                 pos = strncat(buf, argv[i], buf + sizeof(buf) - pos);
 363                 pos = strncat(buf, " ", buf + sizeof(buf) - pos);
 364         }
 365
 366         INFO("execing: %s", buf);
 367
 368 #undef DECLARE_ARG
 369         execv(argv[0], argv);
 370 err:
 371         for (i = 0; argv[i]; i++)
 372                 free(argv[i]);
 373         free(argv);
 374 }
 375
 376 /*
 377  * Check to see if the criu version is recent enough for all the features we
 378  * use. This version allows either CRIU_VERSION or (CRIU_GITID_VERSION and
 379  * CRIU_GITID_PATCHLEVEL) to work, enabling users building from git to c/r
 380  * things potentially before a version is released with a particular feature.
 381  *
 382  * The intent is that when criu development slows down, we can drop this, but
 383  * for now we shouldn't attempt to c/r with versions that we know won't work.
 384  */
 385 static bool criu_version_ok()
 386 {
 387         int pipes[2];
 388         pid_t pid;
 389
 390         if (pipe(pipes) < 0) {
 391                 SYSERROR("pipe() failed");
 392                 return false;
 393         }
 394
 395         pid = fork();
 396         if (pid < 0) {
 397                 SYSERROR("fork() failed");
 398                 return false;
 399         }
 400
 401         if (pid == 0) {
 402                 char *args[] = { "criu", "--version", NULL };
 403                 char *path;
 404                 close(pipes[0]);
 405
 406                 close(STDERR_FILENO);
 407                 if (dup2(pipes[1], STDOUT_FILENO) < 0)
 408                         exit(1);
 409
 410                 path = on_path("criu", NULL);
 411                 if (!path)
 412                         exit(1);
 413
 414                 execv(path, args);
 415                 exit(1);
 416         } else {
 417                 FILE *f;
 418                 char version[1024];
 419                 int patch;
 420
 421                 close(pipes[1]);
 422                 if (wait_for_pid(pid) < 0) {
 423                         close(pipes[0]);
 424                         SYSERROR("execing criu failed, is it installed?");
 425                         return false;
 426                 }
 427
 428                 f = fdopen(pipes[0], "r");
 429                 if (!f) {
 430                         close(pipes[0]);
 431                         return false;
 432                 }
 433
 434                 if (fscanf(f, "Version: %1023[^\n]s", version) != 1)
 435                         goto version_error;
 436
 437                 if (fgetc(f) != '\n')
 438                         goto version_error;
 439
 440                 if (strcmp(version, CRIU_VERSION) >= 0)
 441                         goto version_match;
 442
 443                 if (fscanf(f, "GitID: v%1023[^-]s", version) != 1)
 444                         goto version_error;
 445
 446                 if (fgetc(f) != '-')
 447                         goto version_error;
 448
 449                 if (fscanf(f, "%d", &patch) != 1)
 450                         goto version_error;
 451
 452                 if (strcmp(version, CRIU_GITID_VERSION) < 0)
 453                         goto version_error;
 454
 455                 if (patch < CRIU_GITID_PATCHLEVEL)
 456                         goto version_error;
 457
 458 version_match:
 459                 fclose(f);
 460                 return true;
 461
 462 version_error:
 463                 fclose(f);
 464                 ERROR("must have criu " CRIU_VERSION " or greater to checkpoint/restore\n");
 465                 return false;
 466         }
 467 }
 468
 469 /* Check and make sure the container has a configuration that we know CRIU can
 470  * dump. */
 471 static bool criu_ok(struct lxc_container *c)
 472 {
 473         struct lxc_list *it;
 474
 475         if (!criu_version_ok())
 476                 return false;
 477
 478         if (geteuid()) {
 479                 ERROR("Must be root to checkpoint\n");
 480                 return false;
 481         }
 482
 483         /* We only know how to restore containers with veth networks. */
 484         lxc_list_for_each(it, &c->lxc_conf->network) {
 485                 struct lxc_netdev *n = it->elem;
 486                 switch(n->type) {
 487                 case LXC_NET_VETH:
 488                 case LXC_NET_NONE:
 489                 case LXC_NET_EMPTY:
 490                         break;
 491                 default:
 492                         ERROR("Found network that is not VETH or NONE\n");
 493                         return false;
 494                 }
 495         }
 496
 497         return true;
 498 }
 499
 500 static bool restore_net_info(struct lxc_container *c)
 501 {
 502         struct lxc_list *it;
 503         bool has_error = true;
 504
 505         if (container_mem_lock(c))
 506                 return false;
 507
 508         lxc_list_for_each(it, &c->lxc_conf->network) {
 509                 struct lxc_netdev *netdev = it->elem;
 510                 char template[IFNAMSIZ];
 511
 512                 if (netdev->type != LXC_NET_VETH)
 513                         continue;
 514
 515                 snprintf(template, sizeof(template), "vethXXXXXX");
 516
 517                 if (!netdev->priv.veth_attr.pair)
 518                         netdev->priv.veth_attr.pair = lxc_mkifname(template);
 519
 520                 if (!netdev->priv.veth_attr.pair)
 521                         goto out_unlock;
 522         }
 523
 524         has_error = false;
 525
 526 out_unlock:
 527         container_mem_unlock(c);
 528         return !has_error;
 529 }
 530
 531 // do_restore never returns, the calling process is used as the
 532 // monitor process. do_restore calls exit() if it fails.
 533 void do_restore(struct lxc_container *c, int status_pipe, char *directory, bool verbose)
 534 {
 535         pid_t pid;
 536         char pidfile[L_tmpnam];
 537         struct lxc_handler *handler;
 538         int status, pipes[2] = {-1, -1};
 539
 540         if (!tmpnam(pidfile))
 541                 goto out;
 542
 543         handler = lxc_init(c->name, c->lxc_conf, c->config_path);
 544         if (!handler)
 545                 goto out;
 546
 547         if (!cgroup_init(handler)) {
 548                 ERROR("failed initing cgroups");
 549                 goto out_fini_handler;
 550         }
 551
 552         if (!cgroup_create(handler)) {
 553                 ERROR("failed creating groups");
 554                 goto out_fini_handler;
 555         }
 556
 557         if (!restore_net_info(c)) {
 558                 ERROR("failed restoring network info");
 559                 goto out_fini_handler;
 560         }
 561
 562         resolve_clone_flags(handler);
 563
 564         if (pipe(pipes) < 0) {
 565                 SYSERROR("pipe() failed");
 566                 goto out_fini_handler;
 567         }
 568
 569         pid = fork();
 570         if (pid < 0)
 571                 goto out_fini_handler;
 572
 573         if (pid == 0) {
 574                 struct criu_opts os;
 575                 struct lxc_rootfs *rootfs;
 576                 int flags;
 577
 578                 close(status_pipe);
 579                 status_pipe = -1;
 580
 581                 close(pipes[0]);
 582                 pipes[0] = -1;
 583                 if (dup2(pipes[1], STDERR_FILENO) < 0) {
 584                         SYSERROR("dup2 failed");
 585                         goto out_fini_handler;
 586                 }
 587
 588                 if (dup2(pipes[1], STDOUT_FILENO) < 0) {
 589                         SYSERROR("dup2 failed");
 590                         goto out_fini_handler;
 591                 }
 592
 593                 if (unshare(CLONE_NEWNS))
 594                         goto out_fini_handler;
 595
 596                 /* CRIU needs the lxc root bind mounted so that it is the root of some
 597                  * mount. */
 598                 rootfs = &c->lxc_conf->rootfs;
 599
 600                 if (rootfs_is_blockdev(c->lxc_conf)) {
 601                         if (do_rootfs_setup(c->lxc_conf, c->name, c->config_path) < 0)
 602                                 goto out_fini_handler;
 603                 } else {
 604                         if (mkdir(rootfs->mount, 0755) < 0 && errno != EEXIST)
 605                                 goto out_fini_handler;
 606
 607                         if (mount(NULL, "/", NULL, MS_SLAVE | MS_REC, NULL) < 0) {
 608                                 SYSERROR("remount / to private failed");
 609                                 goto out_fini_handler;
 610                         }
 611
 612                         if (mount(rootfs->path, rootfs->mount, NULL, MS_BIND, NULL) < 0) {
 613                                 rmdir(rootfs->mount);
 614                                 goto out_fini_handler;
 615                         }
 616                 }
 617
 618                 os.action = "restore";
 619                 os.directory = directory;
 620                 os.c = c;
 621                 os.pidfile = pidfile;
 622                 os.verbose = verbose;
 623                 os.cgroup_path = cgroup_canonical_path(handler);
 624                 os.console_fd = c->lxc_conf->console.slave;
 625
 626                 /* Twiddle the FD_CLOEXEC bit. We want to pass this FD to criu
 627                  * via --inherit-fd, so we don't want it to close.
 628                  */
 629                 flags = fcntl(os.console_fd, F_GETFD);
 630                 if (flags < 0) {
 631                         SYSERROR("F_GETFD failed");
 632                         goto out_fini_handler;
 633                 }
 634
 635                 flags &= ~FD_CLOEXEC;
 636
 637                 if (fcntl(os.console_fd, F_SETFD, flags) < 0) {
 638                         SYSERROR("F_SETFD failed");
 639                         goto out_fini_handler;
 640                 }
 641                 os.console_name = c->lxc_conf->console.name;
 642
 643                 /* exec_criu() returning is an error */
 644                 exec_criu(&os);
 645                 umount(rootfs->mount);
 646                 rmdir(rootfs->mount);
 647                 goto out_fini_handler;
 648         } else {
 649                 int ret;
 650                 char title[2048];
 651
 652                 close(pipes[1]);
 653                 pipes[1] = -1;
 654
 655                 pid_t w = waitpid(pid, &status, 0);
 656                 if (w == -1) {
 657                         SYSERROR("waitpid");
 658                         goto out_fini_handler;
 659                 }
 660
 661                 ret = write(status_pipe, &status, sizeof(status));
 662                 close(status_pipe);
 663                 status_pipe = -1;
 664
 665                 if (sizeof(status) != ret) {
 666                         SYSERROR("failed to write all of status");
 667                         goto out_fini_handler;
 668                 }
 669
 670                 if (WIFEXITED(status)) {
 671                         if (WEXITSTATUS(status)) {
 672                                 char buf[4096];
 673                                 int n;
 674
 675                                 n = read(pipes[0], buf, sizeof(buf));
 676                                 if (n < 0) {
 677                                         SYSERROR("failed reading from criu stderr");
 678                                         goto out_fini_handler;
 679                                 }
 680
 681                                 buf[n] = 0;
 682
 683                                 ERROR("criu process exited %d, output:\n%s\n", WEXITSTATUS(status), buf);
 684                                 goto out_fini_handler;
 685                         } else {
 686                                 int ret;
 687                                 FILE *f = fopen(pidfile, "r");
 688                                 if (!f) {
 689                                         SYSERROR("couldn't read restore's init pidfile %s\n", pidfile);
 690                                         goto out_fini_handler;
 691                                 }
 692
 693                                 ret = fscanf(f, "%d", (int*) &handler->pid);
 694                                 fclose(f);
 695                                 if (unlink(pidfile) < 0 && errno != ENOENT)
 696                                         SYSERROR("unlinking pidfile failed");
 697
 698                                 if (ret != 1) {
 699                                         ERROR("reading restore pid failed");
 700                                         goto out_fini_handler;
 701                                 }
 702
 703                                 if (lxc_set_state(c->name, handler, RUNNING)) {
 704                                         ERROR("error setting running state after restore");
 705                                         goto out_fini_handler;
 706                                 }
 707                         }
 708                 } else {
 709                         ERROR("CRIU was killed with signal %d\n", WTERMSIG(status));
 710                         goto out_fini_handler;
 711                 }
 712
 713                 close(pipes[0]);
 714
 715                 /*
 716                  * See comment in lxcapi_start; we don't care if these
 717                  * fail because it's just a beauty thing. We just
 718                  * assign the return here to silence potential.
 719                  */
 720                 ret = snprintf(title, sizeof(title), "[lxc monitor] %s %s", c->config_path, c->name);
 721                 ret = setproctitle(title);
 722
 723                 ret = lxc_poll(c->name, handler);
 724                 if (ret)
 725                         lxc_abort(c->name, handler);
 726                 lxc_fini(c->name, handler);
 727                 exit(ret);
 728         }
 729
 730 out_fini_handler:
 731         if (pipes[0] >= 0)
 732                 close(pipes[0]);
 733         if (pipes[1] >= 0)
 734                 close(pipes[1]);
 735
 736         lxc_fini(c->name, handler);
 737         if (unlink(pidfile) < 0 && errno != ENOENT)
 738                 SYSERROR("unlinking pidfile failed");
 739
 740 out:
 741         if (status_pipe >= 0) {
 742                 status = 1;
 743                 if (write(status_pipe, &status, sizeof(status)) != sizeof(status)) {
 744                         SYSERROR("writing status failed");
 745                 }
 746                 close(status_pipe);
 747         }
 748
 749         exit(1);
 750 }
 751
 752 static int save_tty_major_minor(char *directory, struct lxc_container *c, char *tty_id, int len)
 753 {
 754         FILE *f;
 755         char path[PATH_MAX];
 756         int ret;
 757         struct stat sb;
 758
 759         if (c->lxc_conf->console.path && !strcmp(c->lxc_conf->console.path, "none")) {
 760                 tty_id[0] = 0;
 761                 return 0;
 762         }
 763
 764         ret = snprintf(path, sizeof(path), "/proc/%d/root/dev/console", c->init_pid(c));
 765         if (ret < 0 || ret >= sizeof(path)) {
 766                 ERROR("snprintf'd too many chacters: %d", ret);
 767                 return -1;
 768         }
 769
 770         ret = stat(path, &sb);
 771         if (ret < 0) {
 772                 SYSERROR("stat of %s failed", path);
 773                 return -1;
 774         }
 775
 776         ret = snprintf(path, sizeof(path), "%s/tty.info", directory);
 777         if (ret < 0 || ret >= sizeof(path)) {
 778                 ERROR("snprintf'd too many characters: %d", ret);
 779                 return -1;
 780         }
 781
 782         ret = snprintf(tty_id, len, "tty[%llx:%llx]",
 783                                         (long long unsigned) sb.st_rdev,
 784                                         (long long unsigned) sb.st_dev);
 785         if (ret < 0 || ret >= sizeof(path)) {
 786                 ERROR("snprintf'd too many characters: %d", ret);
 787                 return -1;
 788         }
 789
 790         f = fopen(path, "w");
 791         if (!f) {
 792                 SYSERROR("failed to open %s", path);
 793                 return -1;
 794         }
 795
 796         ret = fprintf(f, "%s", tty_id);
 797         fclose(f);
 798         if (ret < 0)
 799                 SYSERROR("failed to write to %s", path);
 800         return ret;
 801 }
 802
 803 /* do one of either predump or a regular dump */
 804 static bool do_dump(struct lxc_container *c, char *mode, char *directory,
 805                     bool stop, bool verbose, char *predump_dir)
 806 {
 807         pid_t pid;
 808
 809         if (!criu_ok(c))
 810                 return false;
 811
 812         if (mkdir_p(directory, 0700) < 0)
 813                 return false;
 814
 815         pid = fork();
 816         if (pid < 0) {
 817                 SYSERROR("fork failed");
 818                 return false;
 819         }
 820
 821         if (pid == 0) {
 822                 struct criu_opts os;
 823
 824                 os.action = mode;
 825                 os.directory = directory;
 826                 os.c = c;
 827                 os.stop = stop;
 828                 os.verbose = verbose;
 829                 os.predump_dir = predump_dir;
 830                 os.console_name = c->lxc_conf->console.path;
 831
 832                 if (save_tty_major_minor(directory, c, os.tty_id, sizeof(os.tty_id)) < 0)
 833                         exit(1);
 834
 835                 /* exec_criu() returning is an error */
 836                 exec_criu(&os);
 837                 exit(1);
 838         } else {
 839                 int status;
 840                 pid_t w = waitpid(pid, &status, 0);
 841                 if (w == -1) {
 842                         SYSERROR("waitpid");
 843                         return false;
 844                 }
 845
 846                 if (WIFEXITED(status)) {
 847                         if (WEXITSTATUS(status)) {
 848                                 ERROR("dump failed with %d\n", WEXITSTATUS(status));
 849                                 return false;
 850                         }
 851
 852                         return true;
 853                 } else if (WIFSIGNALED(status)) {
 854                         ERROR("dump signaled with %d\n", WTERMSIG(status));
 855                         return false;
 856                 } else {
 857                         ERROR("unknown dump exit %d\n", status);
 858                         return false;
 859                 }
 860         }
 861 }
 862
 863 bool pre_dump(struct lxc_container *c, char *directory, bool verbose, char *predump_dir)
 864 {
 865         return do_dump(c, "pre-dump", directory, false, verbose, predump_dir);
 866 }
 867
 868 bool dump(struct lxc_container *c, char *directory, bool stop, bool verbose, char *predump_dir)
 869 {
 870         char path[PATH_MAX];
 871         int ret;
 872
 873         ret = snprintf(path, sizeof(path), "%s/inventory.img", directory);
 874         if (ret < 0 || ret >= sizeof(path))
 875                 return false;
 876
 877         if (access(path, F_OK) == 0) {
 878                 ERROR("please use a fresh directory for the dump directory\n");
 879                 return false;
 880         }
 881
 882         return do_dump(c, "dump", directory, stop, verbose, predump_dir);
 883 }
 884
 885 bool restore(struct lxc_container *c, char *directory, bool verbose)
 886 {
 887         pid_t pid;
 888         int status, nread;
 889         int pipefd[2];
 890
 891         if (!criu_ok(c))
 892                 return false;
 893
 894         if (geteuid()) {
 895                 ERROR("Must be root to restore\n");
 896                 return false;
 897         }
 898
 899         if (pipe(pipefd)) {
 900                 ERROR("failed to create pipe");
 901                 return false;
 902         }
 903
 904         pid = fork();
 905         if (pid < 0) {
 906                 close(pipefd[0]);
 907                 close(pipefd[1]);
 908                 return false;
 909         }
 910
 911         if (pid == 0) {
 912                 close(pipefd[0]);
 913                 // this never returns
 914                 do_restore(c, pipefd[1], directory, verbose);
 915         }
 916
 917         close(pipefd[1]);
 918
 919         nread = read(pipefd[0], &status, sizeof(status));
 920         close(pipefd[0]);
 921         if (sizeof(status) != nread) {
 922                 ERROR("reading status from pipe failed");
 923                 goto err_wait;
 924         }
 925
 926         // If the criu process was killed or exited nonzero, wait() for the
 927         // handler, since the restore process died. Otherwise, we don't need to
 928         // wait, since the child becomes the monitor process.
 929         if (!WIFEXITED(status) || WEXITSTATUS(status))
 930                 goto err_wait;
 931         return true;
 932
 933 err_wait:
 934         if (wait_for_pid(pid))
 935                 ERROR("restore process died");
 936         return false;
 937 }