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], 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 #undef DECLARE_ARG
 360         execv(argv[0], argv);
 361 err:
 362         for (i = 0; argv[i]; i++)
 363                 free(argv[i]);
 364         free(argv);
 365 }
 366
 367 /*
 368  * Check to see if the criu version is recent enough for all the features we
 369  * use. This version allows either CRIU_VERSION or (CRIU_GITID_VERSION and
 370  * CRIU_GITID_PATCHLEVEL) to work, enabling users building from git to c/r
 371  * things potentially before a version is released with a particular feature.
 372  *
 373  * The intent is that when criu development slows down, we can drop this, but
 374  * for now we shouldn't attempt to c/r with versions that we know won't work.
 375  */
 376 static bool criu_version_ok()
 377 {
 378         int pipes[2];
 379         pid_t pid;
 380
 381         if (pipe(pipes) < 0) {
 382                 SYSERROR("pipe() failed");
 383                 return false;
 384         }
 385
 386         pid = fork();
 387         if (pid < 0) {
 388                 SYSERROR("fork() failed");
 389                 return false;
 390         }
 391
 392         if (pid == 0) {
 393                 char *args[] = { "criu", "--version", NULL };
 394                 char *path;
 395                 close(pipes[0]);
 396
 397                 close(STDERR_FILENO);
 398                 if (dup2(pipes[1], STDOUT_FILENO) < 0)
 399                         exit(1);
 400
 401                 path = on_path("criu", NULL);
 402                 if (!path)
 403                         exit(1);
 404
 405                 execv(path, args);
 406                 exit(1);
 407         } else {
 408                 FILE *f;
 409                 char version[1024];
 410                 int patch;
 411
 412                 close(pipes[1]);
 413                 if (wait_for_pid(pid) < 0) {
 414                         close(pipes[0]);
 415                         SYSERROR("execing criu failed, is it installed?");
 416                         return false;
 417                 }
 418
 419                 f = fdopen(pipes[0], "r");
 420                 if (!f) {
 421                         close(pipes[0]);
 422                         return false;
 423                 }
 424
 425                 if (fscanf(f, "Version: %1023[^\n]s", version) != 1)
 426                         goto version_error;
 427
 428                 if (fgetc(f) != '\n')
 429                         goto version_error;
 430
 431                 if (strcmp(version, CRIU_VERSION) >= 0)
 432                         goto version_match;
 433
 434                 if (fscanf(f, "GitID: v%1023[^-]s", version) != 1)
 435                         goto version_error;
 436
 437                 if (fgetc(f) != '-')
 438                         goto version_error;
 439
 440                 if (fscanf(f, "%d", &patch) != 1)
 441                         goto version_error;
 442
 443                 if (strcmp(version, CRIU_GITID_VERSION) < 0)
 444                         goto version_error;
 445
 446                 if (patch < CRIU_GITID_PATCHLEVEL)
 447                         goto version_error;
 448
 449 version_match:
 450                 fclose(f);
 451                 return true;
 452
 453 version_error:
 454                 fclose(f);
 455                 ERROR("must have criu " CRIU_VERSION " or greater to checkpoint/restore\n");
 456                 return false;
 457         }
 458 }
 459
 460 /* Check and make sure the container has a configuration that we know CRIU can
 461  * dump. */
 462 static bool criu_ok(struct lxc_container *c)
 463 {
 464         struct lxc_list *it;
 465
 466         if (!criu_version_ok())
 467                 return false;
 468
 469         if (geteuid()) {
 470                 ERROR("Must be root to checkpoint\n");
 471                 return false;
 472         }
 473
 474         /* We only know how to restore containers with veth networks. */
 475         lxc_list_for_each(it, &c->lxc_conf->network) {
 476                 struct lxc_netdev *n = it->elem;
 477                 switch(n->type) {
 478                 case LXC_NET_VETH:
 479                 case LXC_NET_NONE:
 480                 case LXC_NET_EMPTY:
 481                         break;
 482                 default:
 483                         ERROR("Found network that is not VETH or NONE\n");
 484                         return false;
 485                 }
 486         }
 487
 488         return true;
 489 }
 490
 491 static bool restore_net_info(struct lxc_container *c)
 492 {
 493         struct lxc_list *it;
 494         bool has_error = true;
 495
 496         if (container_mem_lock(c))
 497                 return false;
 498
 499         lxc_list_for_each(it, &c->lxc_conf->network) {
 500                 struct lxc_netdev *netdev = it->elem;
 501                 char template[IFNAMSIZ];
 502
 503                 if (netdev->type != LXC_NET_VETH)
 504                         continue;
 505
 506                 snprintf(template, sizeof(template), "vethXXXXXX");
 507
 508                 if (!netdev->priv.veth_attr.pair)
 509                         netdev->priv.veth_attr.pair = lxc_mkifname(template);
 510
 511                 if (!netdev->priv.veth_attr.pair)
 512                         goto out_unlock;
 513         }
 514
 515         has_error = false;
 516
 517 out_unlock:
 518         container_mem_unlock(c);
 519         return !has_error;
 520 }
 521
 522 // do_restore never returns, the calling process is used as the
 523 // monitor process. do_restore calls exit() if it fails.
 524 void do_restore(struct lxc_container *c, int pipe, char *directory, bool verbose)
 525 {
 526         pid_t pid;
 527         char pidfile[L_tmpnam];
 528         struct lxc_handler *handler;
 529         int status;
 530
 531         if (!tmpnam(pidfile))
 532                 goto out;
 533
 534         handler = lxc_init(c->name, c->lxc_conf, c->config_path);
 535         if (!handler)
 536                 goto out;
 537
 538         if (!cgroup_init(handler)) {
 539                 ERROR("failed initing cgroups");
 540                 goto out_fini_handler;
 541         }
 542
 543         if (!cgroup_create(handler)) {
 544                 ERROR("failed creating groups");
 545                 goto out_fini_handler;
 546         }
 547
 548         if (!restore_net_info(c)) {
 549                 ERROR("failed restoring network info");
 550                 goto out_fini_handler;
 551         }
 552
 553         resolve_clone_flags(handler);
 554
 555         pid = fork();
 556         if (pid < 0)
 557                 goto out_fini_handler;
 558
 559         if (pid == 0) {
 560                 struct criu_opts os;
 561                 struct lxc_rootfs *rootfs;
 562                 int flags;
 563
 564                 close(pipe);
 565                 pipe = -1;
 566
 567                 if (unshare(CLONE_NEWNS))
 568                         goto out_fini_handler;
 569
 570                 /* CRIU needs the lxc root bind mounted so that it is the root of some
 571                  * mount. */
 572                 rootfs = &c->lxc_conf->rootfs;
 573
 574                 if (rootfs_is_blockdev(c->lxc_conf)) {
 575                         if (do_rootfs_setup(c->lxc_conf, c->name, c->config_path) < 0)
 576                                 goto out_fini_handler;
 577                 } else {
 578                         if (mkdir(rootfs->mount, 0755) < 0 && errno != EEXIST)
 579                                 goto out_fini_handler;
 580
 581                         if (mount(NULL, "/", NULL, MS_SLAVE | MS_REC, NULL) < 0) {
 582                                 SYSERROR("remount / to private failed");
 583                                 goto out_fini_handler;
 584                         }
 585
 586                         if (mount(rootfs->path, rootfs->mount, NULL, MS_BIND, NULL) < 0) {
 587                                 rmdir(rootfs->mount);
 588                                 goto out_fini_handler;
 589                         }
 590                 }
 591
 592                 os.action = "restore";
 593                 os.directory = directory;
 594                 os.c = c;
 595                 os.pidfile = pidfile;
 596                 os.verbose = verbose;
 597                 os.cgroup_path = cgroup_canonical_path(handler);
 598                 os.console_fd = c->lxc_conf->console.slave;
 599
 600                 /* Twiddle the FD_CLOEXEC bit. We want to pass this FD to criu
 601                  * via --inherit-fd, so we don't want it to close.
 602                  */
 603                 flags = fcntl(os.console_fd, F_GETFD);
 604                 if (flags < 0) {
 605                         SYSERROR("F_GETFD failed");
 606                         goto out_fini_handler;
 607                 }
 608
 609                 flags &= ~FD_CLOEXEC;
 610
 611                 if (fcntl(os.console_fd, F_SETFD, flags) < 0) {
 612                         SYSERROR("F_SETFD failed");
 613                         goto out_fini_handler;
 614                 }
 615                 os.console_name = c->lxc_conf->console.name;
 616
 617                 /* exec_criu() returning is an error */
 618                 exec_criu(&os);
 619                 umount(rootfs->mount);
 620                 rmdir(rootfs->mount);
 621                 goto out_fini_handler;
 622         } else {
 623                 int ret;
 624                 char title[2048];
 625
 626                 pid_t w = waitpid(pid, &status, 0);
 627                 if (w == -1) {
 628                         SYSERROR("waitpid");
 629                         goto out_fini_handler;
 630                 }
 631
 632                 ret = write(pipe, &status, sizeof(status));
 633                 close(pipe);
 634                 pipe = -1;
 635
 636                 if (sizeof(status) != ret) {
 637                         SYSERROR("failed to write all of status");
 638                         goto out_fini_handler;
 639                 }
 640
 641                 if (WIFEXITED(status)) {
 642                         if (WEXITSTATUS(status)) {
 643                                 ERROR("criu process exited %d\n", WEXITSTATUS(status));
 644                                 goto out_fini_handler;
 645                         } else {
 646                                 int ret;
 647                                 FILE *f = fopen(pidfile, "r");
 648                                 if (!f) {
 649                                         SYSERROR("couldn't read restore's init pidfile %s\n", pidfile);
 650                                         goto out_fini_handler;
 651                                 }
 652
 653                                 ret = fscanf(f, "%d", (int*) &handler->pid);
 654                                 fclose(f);
 655                                 if (unlink(pidfile) < 0 && errno != ENOENT)
 656                                         SYSERROR("unlinking pidfile failed");
 657
 658                                 if (ret != 1) {
 659                                         ERROR("reading restore pid failed");
 660                                         goto out_fini_handler;
 661                                 }
 662
 663                                 if (lxc_set_state(c->name, handler, RUNNING)) {
 664                                         ERROR("error setting running state after restore");
 665                                         goto out_fini_handler;
 666                                 }
 667                         }
 668                 } else {
 669                         ERROR("CRIU was killed with signal %d\n", WTERMSIG(status));
 670                         goto out_fini_handler;
 671                 }
 672
 673                 /*
 674                  * See comment in lxcapi_start; we don't care if these
 675                  * fail because it's just a beauty thing. We just
 676                  * assign the return here to silence potential.
 677                  */
 678                 ret = snprintf(title, sizeof(title), "[lxc monitor] %s %s", c->config_path, c->name);
 679                 ret = setproctitle(title);
 680
 681                 ret = lxc_poll(c->name, handler);
 682                 if (ret)
 683                         lxc_abort(c->name, handler);
 684                 lxc_fini(c->name, handler);
 685                 exit(ret);
 686         }
 687
 688 out_fini_handler:
 689         lxc_fini(c->name, handler);
 690         if (unlink(pidfile) < 0 && errno != ENOENT)
 691                 SYSERROR("unlinking pidfile failed");
 692
 693 out:
 694         if (pipe >= 0) {
 695                 status = 1;
 696                 if (write(pipe, &status, sizeof(status)) != sizeof(status)) {
 697                         SYSERROR("writing status failed");
 698                 }
 699                 close(pipe);
 700         }
 701
 702         exit(1);
 703 }
 704
 705 static int save_tty_major_minor(char *directory, struct lxc_container *c, char *tty_id, int len)
 706 {
 707         FILE *f;
 708         char path[PATH_MAX];
 709         int ret;
 710         struct stat sb;
 711
 712         if (c->lxc_conf->console.path && !strcmp(c->lxc_conf->console.path, "none")) {
 713                 tty_id[0] = 0;
 714                 return 0;
 715         }
 716
 717         ret = snprintf(path, sizeof(path), "/proc/%d/root/dev/console", c->init_pid(c));
 718         if (ret < 0 || ret >= sizeof(path)) {
 719                 ERROR("snprintf'd too many chacters: %d", ret);
 720                 return -1;
 721         }
 722
 723         ret = stat(path, &sb);
 724         if (ret < 0) {
 725                 SYSERROR("stat of %s failed", path);
 726                 return -1;
 727         }
 728
 729         ret = snprintf(path, sizeof(path), "%s/tty.info", directory);
 730         if (ret < 0 || ret >= sizeof(path)) {
 731                 ERROR("snprintf'd too many characters: %d", ret);
 732                 return -1;
 733         }
 734
 735         ret = snprintf(tty_id, len, "tty[%llx:%llx]",
 736                                         (long long unsigned) sb.st_rdev,
 737                                         (long long unsigned) sb.st_dev);
 738         if (ret < 0 || ret >= sizeof(path)) {
 739                 ERROR("snprintf'd too many characters: %d", ret);
 740                 return -1;
 741         }
 742
 743         f = fopen(path, "w");
 744         if (!f) {
 745                 SYSERROR("failed to open %s", path);
 746                 return -1;
 747         }
 748
 749         ret = fprintf(f, "%s", tty_id);
 750         fclose(f);
 751         if (ret < 0)
 752                 SYSERROR("failed to write to %s", path);
 753         return ret;
 754 }
 755
 756 /* do one of either predump or a regular dump */
 757 static bool do_dump(struct lxc_container *c, char *mode, char *directory,
 758                     bool stop, bool verbose, char *predump_dir)
 759 {
 760         pid_t pid;
 761
 762         if (!criu_ok(c))
 763                 return false;
 764
 765         if (mkdir_p(directory, 0700) < 0)
 766                 return false;
 767
 768         pid = fork();
 769         if (pid < 0) {
 770                 SYSERROR("fork failed");
 771                 return false;
 772         }
 773
 774         if (pid == 0) {
 775                 struct criu_opts os;
 776
 777                 os.action = mode;
 778                 os.directory = directory;
 779                 os.c = c;
 780                 os.stop = stop;
 781                 os.verbose = verbose;
 782                 os.predump_dir = predump_dir;
 783                 os.console_name = c->lxc_conf->console.path;
 784
 785                 if (save_tty_major_minor(directory, c, os.tty_id, sizeof(os.tty_id)) < 0)
 786                         exit(1);
 787
 788                 /* exec_criu() returning is an error */
 789                 exec_criu(&os);
 790                 exit(1);
 791         } else {
 792                 int status;
 793                 pid_t w = waitpid(pid, &status, 0);
 794                 if (w == -1) {
 795                         SYSERROR("waitpid");
 796                         return false;
 797                 }
 798
 799                 if (WIFEXITED(status)) {
 800                         if (WEXITSTATUS(status)) {
 801                                 ERROR("dump failed with %d\n", WEXITSTATUS(status));
 802                                 return false;
 803                         }
 804
 805                         return true;
 806                 } else if (WIFSIGNALED(status)) {
 807                         ERROR("dump signaled with %d\n", WTERMSIG(status));
 808                         return false;
 809                 } else {
 810                         ERROR("unknown dump exit %d\n", status);
 811                         return false;
 812                 }
 813         }
 814 }
 815
 816 bool pre_dump(struct lxc_container *c, char *directory, bool verbose, char *predump_dir)
 817 {
 818         return do_dump(c, "pre-dump", directory, false, verbose, predump_dir);
 819 }
 820
 821 bool dump(struct lxc_container *c, char *directory, bool stop, bool verbose, char *predump_dir)
 822 {
 823         char path[PATH_MAX];
 824         int ret;
 825
 826         ret = snprintf(path, sizeof(path), "%s/inventory.img", directory);
 827         if (ret < 0 || ret >= sizeof(path))
 828                 return false;
 829
 830         if (access(path, F_OK) == 0) {
 831                 ERROR("please use a fresh directory for the dump directory\n");
 832                 return false;
 833         }
 834
 835         return do_dump(c, "dump", directory, stop, verbose, predump_dir);
 836 }
 837
 838 bool restore(struct lxc_container *c, char *directory, bool verbose)
 839 {
 840         pid_t pid;
 841         int status, nread;
 842         int pipefd[2];
 843
 844         if (!criu_ok(c))
 845                 return false;
 846
 847         if (geteuid()) {
 848                 ERROR("Must be root to restore\n");
 849                 return false;
 850         }
 851
 852         if (pipe(pipefd)) {
 853                 ERROR("failed to create pipe");
 854                 return false;
 855         }
 856
 857         pid = fork();
 858         if (pid < 0) {
 859                 close(pipefd[0]);
 860                 close(pipefd[1]);
 861                 return false;
 862         }
 863
 864         if (pid == 0) {
 865                 close(pipefd[0]);
 866                 // this never returns
 867                 do_restore(c, pipefd[1], directory, verbose);
 868         }
 869
 870         close(pipefd[1]);
 871
 872         nread = read(pipefd[0], &status, sizeof(status));
 873         close(pipefd[0]);
 874         if (sizeof(status) != nread) {
 875                 ERROR("reading status from pipe failed");
 876                 goto err_wait;
 877         }
 878
 879         // If the criu process was killed or exited nonzero, wait() for the
 880         // handler, since the restore process died. Otherwise, we don't need to
 881         // wait, since the child becomes the monitor process.
 882         if (!WIFEXITED(status) || WEXITSTATUS(status))
 883                 goto err_wait;
 884         return true;
 885
 886 err_wait:
 887         if (wait_for_pid(pid))
 888                 ERROR("restore process died");
 889         return false;
 890 }