]>
git.proxmox.com Git - mirror_lxc.git/blob - src/lxc/criu.c
2 * lxc: linux Container library
4 * Copyright © 2014-2015 Canonical Ltd.
7 * Tycho Andersen <tycho.andersen@canonical.com>
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.
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.
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
25 #include <linux/limits.h>
30 #include <sys/mount.h>
31 #include <sys/types.h>
49 #include <../include/lxcmntent.h>
54 #define CRIU_VERSION "2.0"
56 #define CRIU_GITID_VERSION "2.0"
57 #define CRIU_GITID_PATCHLEVEL 0
59 #define CRIU_IN_FLIGHT_SUPPORT "2.4"
60 #define CRIU_EXTERNAL_NOT_VETH "2.8"
62 lxc_log_define(lxc_criu
, lxc
);
65 /* the thing to hook to stdout and stderr for logging */
68 /* The type of criu invocation, one of "dump" or "restore" */
71 /* the user-provided migrate options relevant to this action */
72 struct migrate_opts
*user
;
74 /* The container to dump */
75 struct lxc_container
*c
;
77 /* dump: stop the container or not after dumping? */
78 char tty_id
[32]; /* the criu tty id for /dev/console, i.e. "tty[${rdev}:${dev}]" */
80 /* restore: the file to write the init process' pid into */
81 struct lxc_handler
*handler
;
83 /* The path that is bind mounted from /dev/console, if any. We don't
84 * want to use `--ext-mount-map auto`'s result here because the pts
85 * device may have a different path (e.g. if the pty number is
86 * different) on the target host. NULL if lxc.console.path = "none".
90 /* The detected version of criu */
94 static int load_tty_major_minor(char *directory
, char *output
, int len
)
100 ret
= snprintf(path
, sizeof(path
), "%s/tty.info", directory
);
101 if (ret
< 0 || ret
>= sizeof(path
)) {
102 ERROR("snprintf'd too many chacters: %d", ret
);
106 f
= fopen(path
, "r");
108 /* This means we're coming from a liblxc which didn't export
109 * the tty info. In this case they had to have lxc.console.path
110 * = * none, so there's no problem restoring.
115 SYSERROR("couldn't open %s", path
);
119 if (!fgets(output
, len
, f
)) {
121 SYSERROR("couldn't read %s", path
);
129 static int cmp_version(const char *v1
, const char *v2
)
132 int oct_v1
[3], oct_v2
[3];
134 memset(oct_v1
, -1, sizeof(oct_v1
));
135 memset(oct_v2
, -1, sizeof(oct_v2
));
137 ret
= sscanf(v1
, "%d.%d.%d", &oct_v1
[0], &oct_v1
[1], &oct_v1
[2]);
141 ret
= sscanf(v2
, "%d.%d.%d", &oct_v2
[0], &oct_v2
[1], &oct_v2
[2]);
145 /* Major version is greater. */
146 if (oct_v1
[0] > oct_v2
[0])
149 if (oct_v1
[0] < oct_v2
[0])
152 /* Minor number is greater.*/
153 if (oct_v1
[1] > oct_v2
[1])
156 if (oct_v1
[1] < oct_v2
[1])
159 /* Patch number is greater. */
160 if (oct_v1
[2] > oct_v2
[2])
163 /* Patch numbers are equal. */
164 if (oct_v1
[2] == oct_v2
[2])
170 static void exec_criu(struct criu_opts
*opts
)
172 char **argv
, log
[PATH_MAX
];
173 int static_args
= 23, argc
= 0, i
, ret
;
177 struct mntent mntent
;
179 char buf
[4096], tty_info
[32];
182 /* If we are currently in a cgroup /foo/bar, and the container is in a
183 * cgroup /lxc/foo, lxcfs will give us an ENOENT if some task in the
184 * container has an open fd that points to one of the cgroup files
185 * (systemd always opens its "root" cgroup). So, let's escape to the
186 * /actual/ root cgroup so that lxcfs thinks criu has enough rights to
189 if (!cgroup_escape()) {
190 ERROR("failed to escape cgroups");
194 /* The command line always looks like:
195 * criu $(action) --tcp-established --file-locks --link-remap \
196 * --manage-cgroups=full --action-script foo.sh -D $(directory) \
197 * -o $(directory)/$(action).log --ext-mount-map auto
198 * --enable-external-sharing --enable-external-masters
199 * --enable-fs hugetlbfs --enable-fs tracefs --ext-mount-map console:/dev/pts/n
200 * +1 for final NULL */
202 if (strcmp(opts
->action
, "dump") == 0 || strcmp(opts
->action
, "pre-dump") == 0) {
203 /* -t pid --freeze-cgroup /lxc/ct */
206 /* --prev-images-dir <path-to-directory-A-relative-to-B> */
207 if (opts
->user
->predump_dir
)
210 /* --page-server --address <address> --port <port> */
211 if (opts
->user
->pageserver_address
&& opts
->user
->pageserver_port
)
214 /* --leave-running (only for final dump) */
215 if (strcmp(opts
->action
, "dump") == 0 && !opts
->user
->stop
)
218 /* --external tty[88,4] */
223 if (!opts
->user
->preserves_inodes
)
226 /* --ghost-limit 1024 */
227 if (opts
->user
->ghost_limit
)
229 } else if (strcmp(opts
->action
, "restore") == 0) {
230 /* --root $(lxc_mount_point) --restore-detached
232 * --lsm-profile apparmor:whatever
237 if (load_tty_major_minor(opts
->user
->directory
, tty_info
, sizeof(tty_info
)))
240 /* --inherit-fd fd[%d]:tty[%s] */
247 if (cgroup_num_hierarchies() > 0)
248 static_args
+= 2 * cgroup_num_hierarchies();
250 if (opts
->user
->verbose
)
253 if (opts
->user
->action_script
)
256 static_args
+= 2 * lxc_list_len(&opts
->c
->lxc_conf
->mount_list
);
258 ret
= snprintf(log
, PATH_MAX
, "%s/%s.log", opts
->user
->directory
, opts
->action
);
259 if (ret
< 0 || ret
>= PATH_MAX
) {
260 ERROR("logfile name too long");
264 argv
= malloc(static_args
* sizeof(*argv
));
268 memset(argv
, 0, static_args
* sizeof(*argv
));
270 #define DECLARE_ARG(arg) \
273 ERROR("Got NULL argument for criu"); \
276 argv[argc++] = strdup(arg); \
281 argv
[argc
++] = on_path("criu", NULL
);
283 ERROR("Couldn't find criu binary");
287 DECLARE_ARG(opts
->action
);
288 DECLARE_ARG("--tcp-established");
289 DECLARE_ARG("--file-locks");
290 DECLARE_ARG("--link-remap");
291 DECLARE_ARG("--manage-cgroups=full");
292 DECLARE_ARG("--ext-mount-map");
294 DECLARE_ARG("--enable-external-sharing");
295 DECLARE_ARG("--enable-external-masters");
296 DECLARE_ARG("--enable-fs");
297 DECLARE_ARG("hugetlbfs");
298 DECLARE_ARG("--enable-fs");
299 DECLARE_ARG("tracefs");
301 DECLARE_ARG(opts
->user
->directory
);
305 for (i
= 0; i
< cgroup_num_hierarchies(); i
++) {
306 char **controllers
= NULL
, *fullname
;
309 if (!cgroup_get_hierarchies(i
, &controllers
)) {
310 ERROR("failed to get hierarchy %d", i
);
314 /* if we are in a dump, we have to ask the monitor process what
315 * the right cgroup is. if this is a restore, we can just use
316 * the handler the restore task created.
318 if (!strcmp(opts
->action
, "dump") || !strcmp(opts
->action
, "pre-dump")) {
319 path
= lxc_cmd_get_cgroup_path(opts
->c
->name
, opts
->c
->config_path
, controllers
[0]);
321 ERROR("failed to get cgroup path for %s", controllers
[0]);
327 p
= cgroup_get_cgroup(opts
->handler
, controllers
[0]);
329 ERROR("failed to get cgroup path for %s", controllers
[0]);
335 ERROR("strdup failed");
340 tmp
= lxc_deslashify(path
);
342 ERROR("Failed to remove extraneous slashes from \"%s\"",
350 fullname
= lxc_string_join(",", (const char **) controllers
, false);
352 ERROR("failed to join controllers");
357 ret
= sprintf(buf
, "%s:%s", fullname
, path
);
360 if (ret
< 0 || ret
>= sizeof(buf
)) {
361 ERROR("sprintf of cgroup root arg failed");
365 DECLARE_ARG("--cgroup-root");
369 if (opts
->user
->verbose
)
370 DECLARE_ARG("-vvvvvv");
372 if (opts
->user
->action_script
) {
373 DECLARE_ARG("--action-script");
374 DECLARE_ARG(opts
->user
->action_script
);
377 mnts
= make_anonymous_mount_file(&opts
->c
->lxc_conf
->mount_list
);
381 while (getmntent_r(mnts
, &mntent
, buf
, sizeof(buf
))) {
382 char *fmt
, *key
, *val
, *mntdata
;
383 char arg
[2 * PATH_MAX
+ 2];
386 if (parse_mntopts(mntent
.mnt_opts
, &flags
, &mntdata
) < 0)
391 /* only add --ext-mount-map for actual bind mounts */
392 if (!(flags
& MS_BIND
))
395 if (strcmp(opts
->action
, "dump") == 0) {
397 key
= mntent
.mnt_dir
;
398 val
= mntent
.mnt_dir
;
401 key
= mntent
.mnt_dir
;
402 val
= mntent
.mnt_fsname
;
405 ret
= snprintf(arg
, sizeof(arg
), fmt
, key
, val
);
406 if (ret
< 0 || ret
>= sizeof(arg
)) {
408 ERROR("snprintf failed");
412 DECLARE_ARG("--ext-mount-map");
417 if (strcmp(opts
->action
, "dump") == 0 || strcmp(opts
->action
, "pre-dump") == 0) {
418 char pid
[32], *freezer_relative
;
420 if (sprintf(pid
, "%d", opts
->c
->init_pid(opts
->c
)) < 0)
426 freezer_relative
= lxc_cmd_get_cgroup_path(opts
->c
->name
,
427 opts
->c
->config_path
,
429 if (!freezer_relative
) {
430 ERROR("failed getting freezer path");
434 ret
= snprintf(log
, sizeof(log
), "/sys/fs/cgroup/freezer/%s", freezer_relative
);
435 if (ret
< 0 || ret
>= sizeof(log
))
438 if (!opts
->user
->disable_skip_in_flight
&&
439 strcmp(opts
->criu_version
, CRIU_IN_FLIGHT_SUPPORT
) >= 0)
440 DECLARE_ARG("--skip-in-flight");
442 DECLARE_ARG("--freeze-cgroup");
445 if (opts
->tty_id
[0]) {
446 DECLARE_ARG("--ext-mount-map");
447 DECLARE_ARG("/dev/console:console");
449 DECLARE_ARG("--external");
450 DECLARE_ARG(opts
->tty_id
);
453 if (opts
->user
->predump_dir
) {
454 DECLARE_ARG("--prev-images-dir");
455 DECLARE_ARG(opts
->user
->predump_dir
);
456 DECLARE_ARG("--track-mem");
459 if (opts
->user
->pageserver_address
&& opts
->user
->pageserver_port
) {
460 DECLARE_ARG("--page-server");
461 DECLARE_ARG("--address");
462 DECLARE_ARG(opts
->user
->pageserver_address
);
463 DECLARE_ARG("--port");
464 DECLARE_ARG(opts
->user
->pageserver_port
);
467 if (!opts
->user
->preserves_inodes
)
468 DECLARE_ARG("--force-irmap");
470 if (opts
->user
->ghost_limit
) {
471 char ghost_limit
[32];
473 ret
= sprintf(ghost_limit
, "%"PRIu64
, opts
->user
->ghost_limit
);
474 if (ret
< 0 || ret
>= sizeof(ghost_limit
)) {
475 ERROR("failed to print ghost limit %"PRIu64
, opts
->user
->ghost_limit
);
479 DECLARE_ARG("--ghost-limit");
480 DECLARE_ARG(ghost_limit
);
483 /* only for final dump */
484 if (strcmp(opts
->action
, "dump") == 0 && !opts
->user
->stop
)
485 DECLARE_ARG("--leave-running");
486 } else if (strcmp(opts
->action
, "restore") == 0) {
489 struct lxc_conf
*lxc_conf
= opts
->c
->lxc_conf
;
491 DECLARE_ARG("--root");
492 DECLARE_ARG(opts
->c
->lxc_conf
->rootfs
.mount
);
493 DECLARE_ARG("--restore-detached");
494 DECLARE_ARG("--restore-sibling");
497 if (opts
->console_fd
< 0) {
498 ERROR("lxc.console.path configured on source host but not target");
502 ret
= snprintf(buf
, sizeof(buf
), "fd[%d]:%s", opts
->console_fd
, tty_info
);
503 if (ret
< 0 || ret
>= sizeof(buf
))
506 DECLARE_ARG("--inherit-fd");
509 if (opts
->console_name
) {
510 if (snprintf(buf
, sizeof(buf
), "console:%s", opts
->console_name
) < 0) {
511 SYSERROR("sprintf'd too many bytes");
513 DECLARE_ARG("--ext-mount-map");
517 if (lxc_conf
->lsm_aa_profile
|| lxc_conf
->lsm_se_context
) {
519 if (lxc_conf
->lsm_aa_profile
)
520 ret
= snprintf(buf
, sizeof(buf
), "apparmor:%s", lxc_conf
->lsm_aa_profile
);
522 ret
= snprintf(buf
, sizeof(buf
), "selinux:%s", lxc_conf
->lsm_se_context
);
524 if (ret
< 0 || ret
>= sizeof(buf
))
527 DECLARE_ARG("--lsm-profile");
531 additional
= lxc_list_len(&opts
->c
->lxc_conf
->network
) * 2;
533 m
= realloc(argv
, (argc
+ additional
+ 1) * sizeof(*argv
));
538 lxc_list_for_each(it
, &opts
->c
->lxc_conf
->network
) {
539 char eth
[128], *veth
;
541 struct lxc_netdev
*n
= it
->elem
;
542 bool external_not_veth
;
544 if (cmp_version(opts
->criu_version
, CRIU_EXTERNAL_NOT_VETH
) >= 0) {
545 /* Since criu version 2.8 the usage of --veth-pair
546 * has been deprecated:
547 * git tag --contains f2037e6d3445fc400
549 external_not_veth
= true;
551 external_not_veth
= false;
554 if (n
->name
[0] != '\0') {
555 if (strlen(n
->name
) >= sizeof(eth
))
557 strncpy(eth
, n
->name
, sizeof(eth
));
559 ret
= snprintf(eth
, sizeof(eth
), "eth%d", netnr
);
560 if (ret
< 0 || ret
>= sizeof(eth
))
566 veth
= n
->priv
.veth_attr
.pair
;
568 veth
= n
->priv
.veth_attr
.veth1
;
570 if (n
->link
[0] != '\0') {
571 if (external_not_veth
)
572 fmt
= "veth[%s]:%s@%s";
576 ret
= snprintf(buf
, sizeof(buf
), fmt
, eth
, veth
, n
->link
);
578 if (external_not_veth
)
583 ret
= snprintf(buf
, sizeof(buf
), fmt
, eth
, veth
);
585 if (ret
< 0 || ret
>= sizeof(buf
))
588 case LXC_NET_MACVLAN
:
589 if (n
->link
[0] == '\0') {
590 ERROR("no host interface for macvlan %s", n
->name
);
594 ret
= snprintf(buf
, sizeof(buf
), "macvlan[%s]:%s", eth
, n
->link
);
595 if (ret
< 0 || ret
>= sizeof(buf
))
602 /* we have screened for this earlier... */
603 ERROR("unexpected network type %d", n
->type
);
607 if (external_not_veth
)
608 DECLARE_ARG("--external");
610 DECLARE_ARG("--veth-pair");
622 for (i
= 0; argv
[i
]; i
++) {
623 ret
= snprintf(buf
+ pos
, sizeof(buf
) - pos
, "%s ", argv
[i
]);
624 if (ret
< 0 || ret
>= sizeof(buf
) - pos
)
630 INFO("execing: %s", buf
);
632 /* before criu inits its log, it sometimes prints things to stdout/err;
633 * let's be sure we capture that.
635 if (dup2(opts
->pipefd
, STDOUT_FILENO
) < 0) {
636 SYSERROR("dup2 stdout failed");
640 if (dup2(opts
->pipefd
, STDERR_FILENO
) < 0) {
641 SYSERROR("dup2 stderr failed");
648 execv(argv
[0], argv
);
650 for (i
= 0; argv
[i
]; i
++)
656 * Function to check if the checks activated in 'features_to_check' are
657 * available with the current architecture/kernel/criu combination.
659 * Parameter features_to_check is a bit mask of all features that should be
660 * checked (see feature check defines in lxc/lxccontainer.h).
662 * If the return value is true, all requested features are supported. If
663 * the return value is false the features_to_check parameter is updated
664 * to reflect which features are available. '0' means no feature but
665 * also that something went totally wrong.
667 * Some of the code flow of criu_version_ok() is duplicated and maybe it
668 * is a good candidate for refactoring.
670 bool __criu_check_feature(uint64_t *features_to_check
)
673 uint64_t current_bit
= 0;
675 int features
= *features_to_check
;
676 /* Feature checking is currently always like
677 * criu check --feature <feature-name>
679 char *args
[] = { "criu", "check", "--feature", NULL
, NULL
};
681 if ((features
& ~FEATURE_MEM_TRACK
& ~FEATURE_LAZY_PAGES
) != 0) {
682 /* There are feature bits activated we do not understand.
683 * Refusing to answer at all */
684 *features_to_check
= 0;
688 while (current_bit
< sizeof(uint64_t) * 8) {
689 /* only test requested features */
690 if (!(features
& (1ULL << current_bit
))) {
698 SYSERROR("fork() failed");
699 *features_to_check
= 0;
704 if ((1ULL << current_bit
) == FEATURE_MEM_TRACK
)
705 /* This is needed for pre-dump support, which
706 * enables pre-copy migration. */
707 args
[3] = "mem_dirty_track";
708 else if ((1ULL << current_bit
) == FEATURE_LAZY_PAGES
)
709 /* CRIU has two checks for userfaultfd support.
711 * The simpler check is only for 'uffd'. If the
712 * kernel supports userfaultfd without noncoop
713 * then only process can be lazily restored
714 * which do not fork. With 'uffd-noncoop'
715 * it is also possible to lazily restore processes
716 * which do fork. For a container runtime like
717 * LXC checking only for 'uffd' makes not much sense. */
718 args
[3] = "uffd-noncoop";
724 execvp("criu", args
);
725 SYSERROR("Failed to exec \"criu\"");
729 ret
= wait_for_pid(pid
);
732 /* It is not known why CRIU failed. Either
733 * CRIU is not available, the feature check
734 * does not exist or the feature is not
736 INFO("feature not supported");
737 /* Clear not supported feature bit */
738 features
&= ~(1ULL << current_bit
);
742 /* no more checks requested; exit check loop */
743 if (!(features
& ~((1ULL << current_bit
)-1)))
746 if (features
!= *features_to_check
) {
747 *features_to_check
= features
;
754 * Check to see if the criu version is recent enough for all the features we
755 * use. This version allows either CRIU_VERSION or (CRIU_GITID_VERSION and
756 * CRIU_GITID_PATCHLEVEL) to work, enabling users building from git to c/r
757 * things potentially before a version is released with a particular feature.
759 * The intent is that when criu development slows down, we can drop this, but
760 * for now we shouldn't attempt to c/r with versions that we know won't work.
762 * Note: If version != NULL criu_version() stores the detected criu version in
763 * version. Allocates memory for version which must be freed by caller.
765 static bool criu_version_ok(char **version
)
770 if (pipe(pipes
) < 0) {
771 SYSERROR("pipe() failed");
777 SYSERROR("fork() failed");
782 char *args
[] = { "criu", "--version", NULL
};
786 close(STDERR_FILENO
);
787 if (dup2(pipes
[1], STDOUT_FILENO
) < 0)
790 path
= on_path("criu", NULL
);
802 if (wait_for_pid(pid
) < 0) {
804 SYSERROR("execing criu failed, is it installed?");
808 f
= fdopen(pipes
[0], "r");
820 if (fscanf(f
, "Version: %1023[^\n]s", tmp
) != 1)
823 if (fgetc(f
) != '\n')
826 if (strcmp(tmp
, CRIU_VERSION
) >= 0)
829 if (fscanf(f
, "GitID: v%1023[^-]s", tmp
) != 1)
835 if (fscanf(f
, "%d", &patch
) != 1)
838 if (strcmp(tmp
, CRIU_GITID_VERSION
) < 0)
841 if (patch
< CRIU_GITID_PATCHLEVEL
)
855 ERROR("must have criu " CRIU_VERSION
" or greater to checkpoint/restore");
860 /* Check and make sure the container has a configuration that we know CRIU can
862 static bool criu_ok(struct lxc_container
*c
, char **criu_version
)
866 if (!criu_version_ok(criu_version
))
870 ERROR("Must be root to checkpoint");
874 /* We only know how to restore containers with veth networks. */
875 lxc_list_for_each(it
, &c
->lxc_conf
->network
) {
876 struct lxc_netdev
*n
= it
->elem
;
881 case LXC_NET_MACVLAN
:
884 ERROR("Found un-dumpable network: %s (%s)", lxc_net_type_to_str(n
->type
), n
->name
);
892 static bool restore_net_info(struct lxc_container
*c
)
895 bool has_error
= true;
897 if (container_mem_lock(c
))
900 lxc_list_for_each(it
, &c
->lxc_conf
->network
) {
901 struct lxc_netdev
*netdev
= it
->elem
;
902 char template[IFNAMSIZ
];
904 if (netdev
->type
!= LXC_NET_VETH
)
907 snprintf(template, sizeof(template), "vethXXXXXX");
909 if (netdev
->priv
.veth_attr
.pair
[0] == '\0' &&
910 netdev
->priv
.veth_attr
.veth1
[0] == '\0') {
911 if (!lxc_mkifname(template))
914 strcpy(netdev
->priv
.veth_attr
.veth1
, template);
921 container_mem_unlock(c
);
925 /* do_restore never returns, the calling process is used as the monitor process.
926 * do_restore calls exit() if it fails.
928 static void do_restore(struct lxc_container
*c
, int status_pipe
, struct migrate_opts
*opts
, char *criu_version
)
932 struct lxc_handler
*handler
;
934 int pipes
[2] = {-1, -1};
936 /* Try to detach from the current controlling tty if it exists.
937 * Othwerise, lxc_init (via lxc_console) will attach the container's
938 * console output to the current tty, which is probably not what any
939 * library user wants, and if they do, they can just manually configure
942 fd
= open("/dev/tty", O_RDWR
);
944 if (ioctl(fd
, TIOCNOTTY
, NULL
) < 0)
945 SYSERROR("couldn't detach from tty");
949 handler
= lxc_init_handler(c
->name
, c
->lxc_conf
, c
->config_path
, false);
953 if (lxc_init(c
->name
, handler
) < 0)
956 if (!cgroup_init(handler
)) {
957 ERROR("failed initing cgroups");
958 goto out_fini_handler
;
961 if (!cgroup_create(handler
)) {
962 ERROR("failed creating groups");
963 goto out_fini_handler
;
966 if (!restore_net_info(c
)) {
967 ERROR("failed restoring network info");
968 goto out_fini_handler
;
971 ret
= resolve_clone_flags(handler
);
973 ERROR("%s - Unsupported clone flag specified", strerror(errno
));
974 goto out_fini_handler
;
977 if (pipe(pipes
) < 0) {
978 SYSERROR("pipe() failed");
979 goto out_fini_handler
;
984 goto out_fini_handler
;
988 struct lxc_rootfs
*rootfs
;
997 if (unshare(CLONE_NEWNS
))
998 goto out_fini_handler
;
1000 /* CRIU needs the lxc root bind mounted so that it is the root of some
1002 rootfs
= &c
->lxc_conf
->rootfs
;
1004 if (rootfs_is_blockdev(c
->lxc_conf
)) {
1005 if (do_rootfs_setup(c
->lxc_conf
, c
->name
, c
->config_path
) < 0)
1006 goto out_fini_handler
;
1008 if (mkdir(rootfs
->mount
, 0755) < 0 && errno
!= EEXIST
)
1009 goto out_fini_handler
;
1011 if (mount(NULL
, "/", NULL
, MS_SLAVE
| MS_REC
, NULL
) < 0) {
1012 SYSERROR("remount / to private failed");
1013 goto out_fini_handler
;
1016 if (mount(rootfs
->path
, rootfs
->mount
, NULL
, MS_BIND
, NULL
) < 0) {
1017 rmdir(rootfs
->mount
);
1018 goto out_fini_handler
;
1022 os
.pipefd
= pipes
[1];
1023 os
.action
= "restore";
1026 os
.console_fd
= c
->lxc_conf
->console
.slave
;
1027 os
.criu_version
= criu_version
;
1028 os
.handler
= handler
;
1030 if (os
.console_fd
>= 0) {
1031 /* Twiddle the FD_CLOEXEC bit. We want to pass this FD to criu
1032 * via --inherit-fd, so we don't want it to close.
1034 flags
= fcntl(os
.console_fd
, F_GETFD
);
1036 SYSERROR("F_GETFD failed: %d", os
.console_fd
);
1037 goto out_fini_handler
;
1040 flags
&= ~FD_CLOEXEC
;
1042 if (fcntl(os
.console_fd
, F_SETFD
, flags
) < 0) {
1043 SYSERROR("F_SETFD failed");
1044 goto out_fini_handler
;
1047 os
.console_name
= c
->lxc_conf
->console
.name
;
1049 /* exec_criu() returning is an error */
1051 umount(rootfs
->mount
);
1052 rmdir(rootfs
->mount
);
1053 goto out_fini_handler
;
1061 pid_t w
= waitpid(pid
, &status
, 0);
1063 SYSERROR("waitpid");
1064 goto out_fini_handler
;
1067 if (WIFEXITED(status
)) {
1070 if (WEXITSTATUS(status
)) {
1073 n
= read(pipes
[0], buf
, sizeof(buf
));
1075 SYSERROR("failed reading from criu stderr");
1076 goto out_fini_handler
;
1079 if (n
== sizeof(buf
))
1083 ERROR("criu process exited %d, output:\n%s", WEXITSTATUS(status
), buf
);
1084 goto out_fini_handler
;
1086 ret
= snprintf(buf
, sizeof(buf
), "/proc/self/task/%lu/children", (unsigned long)syscall(__NR_gettid
));
1087 if (ret
< 0 || ret
>= sizeof(buf
)) {
1088 ERROR("snprintf'd too many characters: %d", ret
);
1089 goto out_fini_handler
;
1092 FILE *f
= fopen(buf
, "r");
1094 SYSERROR("couldn't read restore's children file %s", buf
);
1095 goto out_fini_handler
;
1098 ret
= fscanf(f
, "%d", (int*) &handler
->pid
);
1101 ERROR("reading restore pid failed");
1102 goto out_fini_handler
;
1105 if (lxc_set_state(c
->name
, handler
, RUNNING
)) {
1106 ERROR("error setting running state after restore");
1107 goto out_fini_handler
;
1111 ERROR("CRIU was killed with signal %d", WTERMSIG(status
));
1112 goto out_fini_handler
;
1117 ret
= write(status_pipe
, &status
, sizeof(status
));
1121 if (sizeof(status
) != ret
) {
1122 SYSERROR("failed to write all of status");
1123 goto out_fini_handler
;
1127 * See comment in lxcapi_start; we don't care if these
1128 * fail because it's just a beauty thing. We just
1129 * assign the return here to silence potential.
1131 ret
= snprintf(title
, sizeof(title
), "[lxc monitor] %s %s", c
->config_path
, c
->name
);
1132 if (ret
< 0 || (size_t)ret
>= sizeof(title
))
1133 INFO("Setting truncated process name");
1135 ret
= setproctitle(title
);
1137 INFO("Failed to set process name");
1139 ret
= lxc_poll(c
->name
, handler
);
1141 lxc_abort(c
->name
, handler
);
1142 lxc_fini(c
->name
, handler
);
1152 lxc_fini(c
->name
, handler
);
1155 if (status_pipe
>= 0) {
1156 /* ensure getting here was a failure, e.g. if we failed to
1157 * parse the child pid or something, even after a successful
1163 if (write(status_pipe
, &status
, sizeof(status
)) != sizeof(status
))
1164 SYSERROR("writing status failed");
1171 static int save_tty_major_minor(char *directory
, struct lxc_container
*c
, char *tty_id
, int len
)
1174 char path
[PATH_MAX
];
1178 if (c
->lxc_conf
->console
.path
&& !strcmp(c
->lxc_conf
->console
.path
, "none")) {
1183 ret
= snprintf(path
, sizeof(path
), "/proc/%d/root/dev/console", c
->init_pid(c
));
1184 if (ret
< 0 || ret
>= sizeof(path
)) {
1185 ERROR("snprintf'd too many chacters: %d", ret
);
1189 ret
= stat(path
, &sb
);
1191 SYSERROR("stat of %s failed", path
);
1195 ret
= snprintf(path
, sizeof(path
), "%s/tty.info", directory
);
1196 if (ret
< 0 || ret
>= sizeof(path
)) {
1197 ERROR("snprintf'd too many characters: %d", ret
);
1201 ret
= snprintf(tty_id
, len
, "tty[%llx:%llx]",
1202 (long long unsigned) sb
.st_rdev
,
1203 (long long unsigned) sb
.st_dev
);
1204 if (ret
< 0 || ret
>= sizeof(path
)) {
1205 ERROR("snprintf'd too many characters: %d", ret
);
1209 f
= fopen(path
, "w");
1211 SYSERROR("failed to open %s", path
);
1215 ret
= fprintf(f
, "%s", tty_id
);
1218 SYSERROR("failed to write to %s", path
);
1222 /* do one of either predump or a regular dump */
1223 static bool do_dump(struct lxc_container
*c
, char *mode
, struct migrate_opts
*opts
)
1228 char *criu_version
= NULL
;
1230 if (!criu_ok(c
, &criu_version
))
1233 ret
= pipe(criuout
);
1235 SYSERROR("pipe() failed");
1239 if (mkdir_p(opts
->directory
, 0700) < 0)
1244 SYSERROR("fork failed");
1249 struct criu_opts os
;
1250 struct lxc_handler h
;
1254 lxc_zero_handler(&h
);
1257 if (!cgroup_init(&h
)) {
1258 ERROR("failed to cgroup_init()");
1262 os
.pipefd
= criuout
[1];
1266 os
.console_name
= c
->lxc_conf
->console
.path
;
1267 os
.criu_version
= criu_version
;
1269 ret
= save_tty_major_minor(opts
->directory
, c
, os
.tty_id
, sizeof(os
.tty_id
));
1275 /* exec_criu() returning is an error */
1287 pid_t w
= waitpid(pid
, &status
, 0);
1289 SYSERROR("waitpid");
1294 n
= read(criuout
[0], buf
, sizeof(buf
));
1302 if (WIFEXITED(status
)) {
1303 if (WEXITSTATUS(status
)) {
1304 ERROR("dump failed with %d", WEXITSTATUS(status
));
1309 } else if (WIFSIGNALED(status
)) {
1310 ERROR("dump signaled with %d", WTERMSIG(status
));
1313 ERROR("unknown dump exit %d", status
);
1318 ERROR("criu output: %s", buf
);
1324 rmdir(opts
->directory
);
1329 bool __criu_pre_dump(struct lxc_container
*c
, struct migrate_opts
*opts
)
1331 return do_dump(c
, "pre-dump", opts
);
1334 bool __criu_dump(struct lxc_container
*c
, struct migrate_opts
*opts
)
1336 char path
[PATH_MAX
];
1339 ret
= snprintf(path
, sizeof(path
), "%s/inventory.img", opts
->directory
);
1340 if (ret
< 0 || ret
>= sizeof(path
))
1343 if (access(path
, F_OK
) == 0) {
1344 ERROR("please use a fresh directory for the dump directory");
1348 return do_dump(c
, "dump", opts
);
1351 bool __criu_restore(struct lxc_container
*c
, struct migrate_opts
*opts
)
1356 char *criu_version
= NULL
;
1358 if (!criu_ok(c
, &criu_version
))
1362 ERROR("Must be root to restore");
1367 ERROR("failed to create pipe");
1380 /* this never returns */
1381 do_restore(c
, pipefd
[1], opts
, criu_version
);
1386 nread
= read(pipefd
[0], &status
, sizeof(status
));
1388 if (sizeof(status
) != nread
) {
1389 ERROR("reading status from pipe failed");
1393 /* If the criu process was killed or exited nonzero, wait() for the
1394 * handler, since the restore process died. Otherwise, we don't need to
1395 * wait, since the child becomes the monitor process.
1397 if (!WIFEXITED(status
) || WEXITSTATUS(status
))
1402 if (wait_for_pid(pid
))
1403 ERROR("restore process died");