3 * Copyright © 2012 Serge Hallyn <serge.hallyn@ubuntu.com>.
4 * Copyright © 2012 Canonical Ltd.
6 * This library is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with this library; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
26 #include <sys/types.h>
28 #include <sys/mount.h>
34 #include <arpa/inet.h>
38 #include <sys/syscall.h>
40 #include <lxc/lxccontainer.h>
41 #include <lxc/version.h>
42 #include <lxc/network.h>
57 #include "namespace.h"
65 #include <../include/ifaddrs.h>
68 #define MAX_BUFFER 4096
70 #define NOT_SUPPORTED_ERROR "the requested function %s is not currently supported with unprivileged containers"
72 /* Define faccessat() if missing from the C library */
73 #ifndef HAVE_FACCESSAT
74 static int faccessat(int __fd
, const char *__file
, int __type
, int __flag
)
77 return syscall(__NR_faccessat
, __fd
, __file
, __type
, __flag
);
86 lxc_log_define(lxc_container
, lxc
);
88 static bool do_lxcapi_destroy(struct lxc_container
*c
);
89 static const char *lxcapi_get_config_path(struct lxc_container
*c
);
90 #define do_lxcapi_get_config_path(c) lxcapi_get_config_path(c)
91 static bool do_lxcapi_set_config_item(struct lxc_container
*c
, const char *key
, const char *v
);
92 static bool container_destroy(struct lxc_container
*c
);
93 static bool get_snappath_dir(struct lxc_container
*c
, char *snappath
);
94 static bool lxcapi_snapshot_destroy_all(struct lxc_container
*c
);
95 static bool do_lxcapi_save_config(struct lxc_container
*c
, const char *alt_file
);
97 static bool config_file_exists(const char *lxcpath
, const char *cname
)
99 /* $lxcpath + '/' + $cname + '/config' + \0 */
100 int ret
, len
= strlen(lxcpath
) + strlen(cname
) + 9;
101 char *fname
= alloca(len
);
103 ret
= snprintf(fname
, len
, "%s/%s/config", lxcpath
, cname
);
104 if (ret
< 0 || ret
>= len
)
107 return file_exists(fname
);
111 * A few functions to help detect when a container creation failed.
112 * If a container creation was killed partway through, then trying
113 * to actually start that container could harm the host. We detect
114 * this by creating a 'partial' file under the container directory,
115 * and keeping an advisory lock. When container creation completes,
116 * we remove that file. When we load or try to start a container, if
117 * we find that file, without a flock, we remove the container.
119 static int ongoing_create(struct lxc_container
*c
)
121 int len
= strlen(c
->config_path
) + strlen(c
->name
) + 10;
122 char *path
= alloca(len
);
126 ret
= snprintf(path
, len
, "%s/%s/partial", c
->config_path
, c
->name
);
127 if (ret
< 0 || ret
>= len
) {
128 ERROR("Error writing partial pathname");
132 if (!file_exists(path
))
134 fd
= open(path
, O_RDWR
);
136 // give benefit of the doubt
137 SYSERROR("Error opening partial file");
141 lk
.l_whence
= SEEK_SET
;
145 if (fcntl(fd
, F_GETLK
, &lk
) == 0 && lk
.l_pid
!= -1) {
146 // create is still ongoing
150 // create completed but partial is still there.
155 static int create_partial(struct lxc_container
*c
)
157 // $lxcpath + '/' + $name + '/partial' + \0
158 int len
= strlen(c
->config_path
) + strlen(c
->name
) + 10;
159 char *path
= alloca(len
);
163 ret
= snprintf(path
, len
, "%s/%s/partial", c
->config_path
, c
->name
);
164 if (ret
< 0 || ret
>= len
) {
165 ERROR("Error writing partial pathname");
168 if ((fd
=open(path
, O_RDWR
| O_CREAT
| O_EXCL
, 0755)) < 0) {
169 SYSERROR("Erorr creating partial file");
173 lk
.l_whence
= SEEK_SET
;
176 if (fcntl(fd
, F_SETLKW
, &lk
) < 0) {
177 SYSERROR("Error locking partial file %s", path
);
185 static void remove_partial(struct lxc_container
*c
, int fd
)
187 // $lxcpath + '/' + $name + '/partial' + \0
188 int len
= strlen(c
->config_path
) + strlen(c
->name
) + 10;
189 char *path
= alloca(len
);
193 ret
= snprintf(path
, len
, "%s/%s/partial", c
->config_path
, c
->name
);
194 if (ret
< 0 || ret
>= len
) {
195 ERROR("Error writing partial pathname");
198 if (unlink(path
) < 0)
199 SYSERROR("Error unlink partial file %s", path
);
203 * 1. container_mem_lock(c) protects the struct lxc_container from multiple threads.
204 * 2. container_disk_lock(c) protects the on-disk container data - in particular the
205 * container configuration file.
206 * The container_disk_lock also takes the container_mem_lock.
207 * 3. thread_mutex protects process data (ex: fd table) from multiple threads.
208 * NOTHING mutexes two independent programs with their own struct
209 * lxc_container for the same c->name, between API calls. For instance,
210 * c->config_read(); c->start(); Between those calls, data on disk
211 * could change (which shouldn't bother the caller unless for instance
212 * the rootfs get moved). c->config_read(); update; c->config_write();
213 * Two such updaters could race. The callers should therefore check their
214 * results. Trying to prevent that would necessarily expose us to deadlocks
215 * due to hung callers. So I prefer to keep the locks only within our own
216 * functions, not across functions.
218 * If you're going to clone while holding a lxccontainer, increment
219 * c->numthreads (under privlock) before forking. When deleting,
220 * decrement numthreads under privlock, then if it hits 0 you can delete.
221 * Do not ever use a lxccontainer whose numthreads you did not bump.
224 static void lxc_container_free(struct lxc_container
*c
)
230 c
->configfile
= NULL
;
231 free(c
->error_string
);
232 c
->error_string
= NULL
;
234 lxc_putlock(c
->slock
);
238 lxc_putlock(c
->privlock
);
244 lxc_conf_free(c
->lxc_conf
);
247 free(c
->config_path
);
248 c
->config_path
= NULL
;
254 * Consider the following case:
255 freer | racing get()er
256 ==================================================================
257 lxc_container_put() | lxc_container_get()
258 \ lxclock(c->privlock) | c->numthreads < 1? (no)
259 \ c->numthreads = 0 | \ lxclock(c->privlock) -> waits
261 \ lxc_container_free() | \ lxclock() returns
262 | \ c->numthreads < 1 -> return 0
264 \ \ sem_destroy(privlock) |
266 * When the get()er checks numthreads the first time, one of the following
268 * 1. freer has set numthreads = 0. get() returns 0
269 * 2. freer is between lxclock and setting numthreads to 0. get()er will
270 * sem_wait on privlock, get lxclock after freer() drops it, then see
271 * numthreads is 0 and exit without touching lxclock again..
272 * 3. freer has not yet locked privlock. If get()er runs first, then put()er
273 * will see --numthreads = 1 and not call lxc_container_free().
276 int lxc_container_get(struct lxc_container
*c
)
281 // if someone else has already started freeing the container, don't
282 // try to take the lock, which may be invalid
283 if (c
->numthreads
< 1)
286 if (container_mem_lock(c
))
288 if (c
->numthreads
< 1) {
289 // bail without trying to unlock, bc the privlock is now probably
294 container_mem_unlock(c
);
298 int lxc_container_put(struct lxc_container
*c
)
302 if (container_mem_lock(c
))
304 if (--c
->numthreads
< 1) {
305 container_mem_unlock(c
);
306 lxc_container_free(c
);
309 container_mem_unlock(c
);
313 static bool do_lxcapi_is_defined(struct lxc_container
*c
)
322 if (container_mem_lock(c
))
326 statret
= stat(c
->configfile
, &statbuf
);
332 container_mem_unlock(c
);
336 #define WRAP_API(rettype, fnname) \
337 static rettype fnname(struct lxc_container *c) \
340 struct lxc_conf *old = current_config; \
341 current_config = c ? c->lxc_conf : NULL; \
342 ret = do_##fnname(c); \
343 current_config = old; \
347 #define WRAP_API_1(rettype, fnname, t1) \
348 static rettype fnname(struct lxc_container *c, t1 a1) \
351 struct lxc_conf *old = current_config; \
352 current_config = c ? c->lxc_conf : NULL; \
353 ret = do_##fnname(c, a1); \
354 current_config = old; \
358 #define WRAP_API_2(rettype, fnname, t1, t2) \
359 static rettype fnname(struct lxc_container *c, t1 a1, t2 a2) \
362 struct lxc_conf *old = current_config; \
363 current_config = c ? c->lxc_conf : NULL; \
364 ret = do_##fnname(c, a1, a2); \
365 current_config = old; \
369 #define WRAP_API_3(rettype, fnname, t1, t2, t3) \
370 static rettype fnname(struct lxc_container *c, t1 a1, t2 a2, t3 a3) \
373 struct lxc_conf *old = current_config; \
374 current_config = c ? c->lxc_conf : NULL; \
375 ret = do_##fnname(c, a1, a2, a3); \
376 current_config = old; \
380 WRAP_API(bool, lxcapi_is_defined
)
382 static const char *do_lxcapi_state(struct lxc_container
*c
)
388 s
= lxc_getstate(c
->name
, c
->config_path
);
389 return lxc_state2str(s
);
392 WRAP_API(const char *, lxcapi_state
)
394 static bool is_stopped(struct lxc_container
*c
)
397 s
= lxc_getstate(c
->name
, c
->config_path
);
398 return (s
== STOPPED
);
401 static bool do_lxcapi_is_running(struct lxc_container
*c
)
407 s
= do_lxcapi_state(c
);
408 if (!s
|| strcmp(s
, "STOPPED") == 0)
413 WRAP_API(bool, lxcapi_is_running
)
415 static bool do_lxcapi_freeze(struct lxc_container
*c
)
421 ret
= lxc_freeze(c
->name
, c
->config_path
);
427 WRAP_API(bool, lxcapi_freeze
)
429 static bool do_lxcapi_unfreeze(struct lxc_container
*c
)
435 ret
= lxc_unfreeze(c
->name
, c
->config_path
);
441 WRAP_API(bool, lxcapi_unfreeze
)
443 static int do_lxcapi_console_getfd(struct lxc_container
*c
, int *ttynum
, int *masterfd
)
449 ttyfd
= lxc_console_getfd(c
, ttynum
, masterfd
);
453 WRAP_API_2(int, lxcapi_console_getfd
, int *, int *)
455 static int lxcapi_console(struct lxc_container
*c
, int ttynum
, int stdinfd
,
456 int stdoutfd
, int stderrfd
, int escape
)
459 struct lxc_conf
*old
= current_config
;
460 current_config
= c
? c
->lxc_conf
: NULL
;
461 ret
= lxc_console(c
, ttynum
, stdinfd
, stdoutfd
, stderrfd
, escape
);
462 current_config
= old
;
466 static pid_t
do_lxcapi_init_pid(struct lxc_container
*c
)
471 return lxc_cmd_get_init_pid(c
->name
, c
->config_path
);
474 WRAP_API(pid_t
, lxcapi_init_pid
)
476 static bool load_config_locked(struct lxc_container
*c
, const char *fname
)
479 c
->lxc_conf
= lxc_conf_init();
482 if (lxc_config_read(fname
, c
->lxc_conf
, false) != 0)
487 static bool do_lxcapi_load_config(struct lxc_container
*c
, const char *alt_file
)
489 bool ret
= false, need_disklock
= false;
495 fname
= c
->configfile
;
501 * If we're reading something other than the container's config,
502 * we only need to lock the in-memory container. If loading the
503 * container's config file, take the disk lock.
505 if (strcmp(fname
, c
->configfile
) == 0)
506 need_disklock
= true;
509 lret
= container_disk_lock(c
);
511 lret
= container_mem_lock(c
);
515 ret
= load_config_locked(c
, fname
);
518 container_disk_unlock(c
);
520 container_mem_unlock(c
);
524 WRAP_API_1(bool, lxcapi_load_config
, const char *)
526 static bool do_lxcapi_want_daemonize(struct lxc_container
*c
, bool state
)
528 if (!c
|| !c
->lxc_conf
)
530 if (container_mem_lock(c
)) {
531 ERROR("Error getting mem lock");
534 c
->daemonize
= state
;
535 container_mem_unlock(c
);
539 WRAP_API_1(bool, lxcapi_want_daemonize
, bool)
541 static bool do_lxcapi_want_close_all_fds(struct lxc_container
*c
, bool state
)
543 if (!c
|| !c
->lxc_conf
)
545 if (container_mem_lock(c
)) {
546 ERROR("Error getting mem lock");
549 c
->lxc_conf
->close_all_fds
= state
;
550 container_mem_unlock(c
);
554 WRAP_API_1(bool, lxcapi_want_close_all_fds
, bool)
556 static bool do_lxcapi_wait(struct lxc_container
*c
, const char *state
, int timeout
)
563 ret
= lxc_wait(c
->name
, state
, timeout
, c
->config_path
);
567 WRAP_API_2(bool, lxcapi_wait
, const char *, int)
569 static bool do_wait_on_daemonized_start(struct lxc_container
*c
, int pid
)
571 /* we'll probably want to make this timeout configurable? */
572 int timeout
= 5, ret
, status
;
575 * our child is going to fork again, then exit. reap the
578 ret
= waitpid(pid
, &status
, 0);
579 if (ret
== -1 || !WIFEXITED(status
) || WEXITSTATUS(status
) != 0)
580 DEBUG("failed waiting for first dual-fork child");
581 return do_lxcapi_wait(c
, "RUNNING", timeout
);
584 WRAP_API_1(bool, wait_on_daemonized_start
, int)
586 static bool am_single_threaded(void)
588 struct dirent dirent
, *direntp
;
592 dir
= opendir("/proc/self/task");
594 INFO("failed to open /proc/self/task");
598 while (!readdir_r(dir
, &dirent
, &direntp
)) {
602 if (!strcmp(direntp
->d_name
, "."))
605 if (!strcmp(direntp
->d_name
, ".."))
615 * I can't decide if it'd be more convenient for callers if we accept '...',
616 * or a null-terminated array (i.e. execl vs execv)
618 static bool do_lxcapi_start(struct lxc_container
*c
, int useinit
, char * const argv
[])
621 struct lxc_conf
*conf
;
622 bool daemonize
= false;
624 char *default_args
[] = {
630 /* container exists */
633 /* container has been setup */
637 if ((ret
= ongoing_create(c
)) < 0) {
638 ERROR("Error checking for incomplete creation");
642 ERROR("Error: %s creation was not completed", c
->name
);
643 do_lxcapi_destroy(c
);
645 } else if (ret
== 1) {
646 ERROR("Error: creation of %s is ongoing", c
->name
);
650 /* is this app meant to be run through lxcinit, as in lxc-execute? */
651 if (useinit
&& !argv
)
654 if (container_mem_lock(c
))
657 daemonize
= c
->daemonize
;
658 container_mem_unlock(c
);
661 ret
= lxc_execute(c
->name
, argv
, 1, conf
, c
->config_path
);
662 return ret
== 0 ? true : false;
666 if (conf
->init_cmd
) {
667 init_cmd
[0] = conf
->init_cmd
;
676 * say, I'm not sure - what locks do we want here? Any?
677 * Is liblxc's locking enough here to protect the on disk
678 * container? We don't want to exclude things like lxc_info
679 * while container is running...
683 lxc_monitord_spawn(c
->config_path
);
690 /* Set to NULL because we don't want father unlink
691 * the PID file, child will do the free and unlink.
694 return wait_on_daemonized_start(c
, pid
);
697 /* We don't really care if this doesn't print all the
698 * characters; all that it means is that the proctitle will be
699 * ugly. Similarly, we also don't care if setproctitle()
701 snprintf(title
, sizeof(title
), "[lxc monitor] %s %s", c
->config_path
, c
->name
);
702 INFO("Attempting to set proc title to %s", title
);
705 /* second fork to be reparented by init */
708 SYSERROR("Error doing dual-fork");
713 /* like daemon(), chdir to / and redirect 0,1,2 to /dev/null */
715 SYSERROR("Error chdir()ing to /.");
718 lxc_check_inherited(conf
, true, -1);
722 open("/dev/zero", O_RDONLY
);
723 open("/dev/null", O_RDWR
);
724 open("/dev/null", O_RDWR
);
727 if (!am_single_threaded()) {
728 ERROR("Cannot start non-daemonized container when threaded");
733 /* We need to write PID file after daeminize, so we always
734 * write the right PID.
737 pid_fp
= fopen(c
->pidfile
, "w");
738 if (pid_fp
== NULL
) {
739 SYSERROR("Failed to create pidfile '%s' for '%s'",
740 c
->pidfile
, c
->name
);
744 if (fprintf(pid_fp
, "%d\n", getpid()) < 0) {
745 SYSERROR("Failed to write '%s'", c
->pidfile
);
758 if (lxc_check_inherited(conf
, daemonize
, -1)) {
759 ERROR("Inherited fds found");
764 ret
= lxc_start(c
->name
, argv
, conf
, c
->config_path
);
768 INFO("container requested reboot");
781 exit (ret
== 0 ? true : false);
783 return (ret
== 0 ? true : false);
786 static bool lxcapi_start(struct lxc_container
*c
, int useinit
, char * const argv
[])
789 struct lxc_conf
*old
= current_config
;
790 current_config
= c
? c
->lxc_conf
: NULL
;
791 ret
= do_lxcapi_start(c
, useinit
, argv
);
792 current_config
= old
;
797 * note there MUST be an ending NULL
799 static bool lxcapi_startl(struct lxc_container
*c
, int useinit
, ...)
802 char **inargs
= NULL
;
805 /* container exists */
809 struct lxc_conf
*old
= current_config
;
810 current_config
= c
->lxc_conf
;
812 va_start(ap
, useinit
);
813 inargs
= lxc_va_arg_list_to_argv(ap
, 0, 1);
817 ERROR("Memory allocation error.");
821 /* pass NULL if no arguments were supplied */
822 bret
= do_lxcapi_start(c
, useinit
, *inargs
? inargs
: NULL
);
827 for (arg
= inargs
; *arg
; arg
++)
832 current_config
= old
;
836 static bool do_lxcapi_stop(struct lxc_container
*c
)
843 ret
= lxc_cmd_stop(c
->name
, c
->config_path
);
848 WRAP_API(bool, lxcapi_stop
)
850 static int do_create_container_dir(const char *path
, struct lxc_conf
*conf
)
852 int ret
= -1, lasterr
;
853 char *p
= alloca(strlen(path
)+1);
854 mode_t mask
= umask(0002);
855 ret
= mkdir(path
, 0770);
863 SYSERROR("failed to create container path %s", path
);
868 if (!lxc_list_empty(&conf
->id_map
) && chown_mapped_root(p
, conf
) != 0) {
869 ERROR("Failed to chown container dir");
876 * create the standard expected container dir
878 static bool create_container_dir(struct lxc_container
*c
)
883 len
= strlen(c
->config_path
) + strlen(c
->name
) + 2;
887 ret
= snprintf(s
, len
, "%s/%s", c
->config_path
, c
->name
);
888 if (ret
< 0 || ret
>= len
) {
892 ret
= do_create_container_dir(s
, c
->lxc_conf
);
898 * do_bdev_create: thin wrapper around bdev_create(). Like bdev_create(),
899 * it returns a mounted bdev on success, NULL on error.
901 static struct bdev
*do_bdev_create(struct lxc_container
*c
, const char *type
,
902 struct bdev_specs
*specs
)
909 /* rootfs.path or lxcpath/lxcname/rootfs */
910 if (c
->lxc_conf
->rootfs
.path
&& access(c
->lxc_conf
->rootfs
.path
, F_OK
) == 0) {
911 const char *rpath
= c
->lxc_conf
->rootfs
.path
;
912 len
= strlen(rpath
) + 1;
914 ret
= snprintf(dest
, len
, "%s", rpath
);
916 const char *lxcpath
= do_lxcapi_get_config_path(c
);
917 len
= strlen(c
->name
) + strlen(lxcpath
) + 9;
919 ret
= snprintf(dest
, len
, "%s/%s/rootfs", lxcpath
, c
->name
);
921 if (ret
< 0 || ret
>= len
)
924 bdev
= bdev_create(dest
, type
, c
->name
, specs
);
926 ERROR("Failed to create backing store type %s", type
);
930 do_lxcapi_set_config_item(c
, "lxc.rootfs", bdev
->src
);
932 /* if we are not root, chown the rootfs dir to root in the
935 if (geteuid() != 0 || (c
->lxc_conf
&& !lxc_list_empty(&c
->lxc_conf
->id_map
))) {
936 if (chown_mapped_root(bdev
->dest
, c
->lxc_conf
) < 0) {
937 ERROR("Error chowning %s to container root", bdev
->dest
);
938 suggest_default_idmap();
947 static char *lxcbasename(char *path
)
949 char *p
= path
+ strlen(path
) - 1;
950 while (*p
!= '/' && p
> path
)
955 static bool create_run_template(struct lxc_container
*c
, char *tpath
, bool quiet
,
965 SYSERROR("failed to fork task for container creation template");
969 if (pid
== 0) { // child
970 char *patharg
, *namearg
, *rootfsarg
, *src
;
971 struct bdev
*bdev
= NULL
;
973 int ret
, len
, nargs
= 0;
975 struct lxc_conf
*conf
= c
->lxc_conf
;
981 open("/dev/zero", O_RDONLY
);
982 open("/dev/null", O_RDWR
);
983 open("/dev/null", O_RDWR
);
986 src
= c
->lxc_conf
->rootfs
.path
;
988 * for an overlay create, what the user wants is the template to fill
989 * in what will become the readonly lower layer. So don't mount for
992 if (strncmp(src
, "overlayfs:", 10) == 0)
993 src
= overlay_getlower(src
+10);
994 if (strncmp(src
, "aufs:", 5) == 0)
995 src
= overlay_getlower(src
+5);
997 bdev
= bdev_init(c
->lxc_conf
, src
, c
->lxc_conf
->rootfs
.mount
, NULL
);
999 ERROR("Error opening rootfs");
1003 if (geteuid() == 0) {
1004 if (unshare(CLONE_NEWNS
) < 0) {
1005 ERROR("error unsharing mounts");
1008 if (detect_shared_rootfs()) {
1009 if (mount(NULL
, "/", NULL
, MS_SLAVE
|MS_REC
, NULL
)) {
1010 SYSERROR("Failed to make / rslave to run template");
1011 ERROR("Continuing...");
1015 if (strcmp(bdev
->type
, "dir") && strcmp(bdev
->type
, "btrfs")) {
1016 if (geteuid() != 0) {
1017 ERROR("non-root users can only create btrfs and directory-backed containers");
1020 if (bdev
->ops
->mount(bdev
) < 0) {
1021 ERROR("Error mounting rootfs");
1024 } else { // TODO come up with a better way here!
1026 bdev
->dest
= strdup(bdev
->src
);
1030 * create our new array, pre-pend the template name and
1034 for (nargs
= 0; argv
[nargs
]; nargs
++) ;
1035 nargs
+= 4; // template, path, rootfs and name args
1037 newargv
= malloc(nargs
* sizeof(*newargv
));
1040 newargv
[0] = lxcbasename(tpath
);
1042 len
= strlen(c
->config_path
) + strlen(c
->name
) + strlen("--path=") + 2;
1043 patharg
= malloc(len
);
1046 ret
= snprintf(patharg
, len
, "--path=%s/%s", c
->config_path
, c
->name
);
1047 if (ret
< 0 || ret
>= len
)
1049 newargv
[1] = patharg
;
1050 len
= strlen("--name=") + strlen(c
->name
) + 1;
1051 namearg
= malloc(len
);
1054 ret
= snprintf(namearg
, len
, "--name=%s", c
->name
);
1055 if (ret
< 0 || ret
>= len
)
1057 newargv
[2] = namearg
;
1059 len
= strlen("--rootfs=") + 1 + strlen(bdev
->dest
);
1060 rootfsarg
= malloc(len
);
1063 ret
= snprintf(rootfsarg
, len
, "--rootfs=%s", bdev
->dest
);
1064 if (ret
< 0 || ret
>= len
)
1066 newargv
[3] = rootfsarg
;
1068 /* add passed-in args */
1070 for (i
= 4; i
< nargs
; i
++)
1071 newargv
[i
] = argv
[i
-4];
1073 /* add trailing NULL */
1075 newargv
= realloc(newargv
, nargs
* sizeof(*newargv
));
1078 newargv
[nargs
- 1] = NULL
;
1081 * If we're running the template in a mapped userns, then
1082 * we prepend the template command with:
1083 * lxc-usernsexec <-m map1> ... <-m mapn> --
1084 * and we append "--mapped-uid x", where x is the mapped uid
1087 if (!lxc_list_empty(&conf
->id_map
)) {
1091 char **n2
= malloc(n2args
* sizeof(*n2
));
1092 struct lxc_list
*it
;
1096 SYSERROR("out of memory");
1100 tpath
= "lxc-usernsexec";
1101 n2
[0] = "lxc-usernsexec";
1102 lxc_list_for_each(it
, &conf
->id_map
) {
1105 n2
= realloc(n2
, n2args
* sizeof(char *));
1108 n2
[n2args
-2] = "-m";
1109 n2
[n2args
-1] = malloc(200);
1112 ret
= snprintf(n2
[n2args
-1], 200, "%c:%lu:%lu:%lu",
1113 map
->idtype
== ID_TYPE_UID
? 'u' : 'g',
1114 map
->nsid
, map
->hostid
, map
->range
);
1115 if (ret
< 0 || ret
>= 200)
1118 int hostid_mapped
= mapped_hostid(geteuid(), conf
, ID_TYPE_UID
);
1119 int extraargs
= hostid_mapped
>= 0 ? 1 : 3;
1120 n2
= realloc(n2
, (nargs
+ n2args
+ extraargs
) * sizeof(char *));
1123 if (hostid_mapped
< 0) {
1124 hostid_mapped
= find_unmapped_nsuid(conf
, ID_TYPE_UID
);
1125 n2
[n2args
++] = "-m";
1126 if (hostid_mapped
< 0) {
1127 ERROR("Could not find free uid to map");
1130 n2
[n2args
++] = malloc(200);
1131 if (!n2
[n2args
-1]) {
1132 SYSERROR("out of memory");
1135 ret
= snprintf(n2
[n2args
-1], 200, "u:%d:%d:1",
1136 hostid_mapped
, geteuid());
1137 if (ret
< 0 || ret
>= 200) {
1138 ERROR("string too long");
1142 int hostgid_mapped
= mapped_hostid(getegid(), conf
, ID_TYPE_GID
);
1143 extraargs
= hostgid_mapped
>= 0 ? 1 : 3;
1144 n2
= realloc(n2
, (nargs
+ n2args
+ extraargs
) * sizeof(char *));
1147 if (hostgid_mapped
< 0) {
1148 hostgid_mapped
= find_unmapped_nsuid(conf
, ID_TYPE_GID
);
1149 n2
[n2args
++] = "-m";
1150 if (hostgid_mapped
< 0) {
1151 ERROR("Could not find free uid to map");
1154 n2
[n2args
++] = malloc(200);
1155 if (!n2
[n2args
-1]) {
1156 SYSERROR("out of memory");
1159 ret
= snprintf(n2
[n2args
-1], 200, "g:%d:%d:1",
1160 hostgid_mapped
, getegid());
1161 if (ret
< 0 || ret
>= 200) {
1162 ERROR("string too long");
1166 n2
[n2args
++] = "--";
1167 for (i
= 0; i
< nargs
; i
++)
1168 n2
[i
+ n2args
] = newargv
[i
];
1170 // Finally add "--mapped-uid $uid" to tell template what to chown
1173 n2
= realloc(n2
, n2args
* sizeof(char *));
1175 SYSERROR("out of memory");
1178 // note n2[n2args-1] is NULL
1179 n2
[n2args
-5] = "--mapped-uid";
1180 snprintf(txtuid
, 20, "%d", hostid_mapped
);
1181 n2
[n2args
-4] = txtuid
;
1182 n2
[n2args
-3] = "--mapped-gid";
1183 snprintf(txtgid
, 20, "%d", hostgid_mapped
);
1184 n2
[n2args
-2] = txtgid
;
1185 n2
[n2args
-1] = NULL
;
1190 execvp(tpath
, newargv
);
1191 SYSERROR("failed to execute template %s", tpath
);
1195 if (wait_for_pid(pid
) != 0) {
1196 ERROR("container creation template for %s failed", c
->name
);
1203 static bool prepend_lxc_header(char *path
, const char *t
, char *const argv
[])
1211 unsigned char md_value
[SHA_DIGEST_LENGTH
];
1215 f
= fopen(path
, "r");
1219 if (fseek(f
, 0, SEEK_END
) < 0)
1221 if ((flen
= ftell(f
)) < 0)
1223 if (fseek(f
, 0, SEEK_SET
) < 0)
1225 if ((contents
= malloc(flen
+ 1)) == NULL
)
1227 if (fread(contents
, 1, flen
, f
) != flen
)
1228 goto out_free_contents
;
1230 contents
[flen
] = '\0';
1234 goto out_free_contents
;
1237 tpath
= get_template_path(t
);
1239 ERROR("bad template: %s", t
);
1240 goto out_free_contents
;
1243 ret
= sha1sum_file(tpath
, md_value
);
1245 ERROR("Error getting sha1sum of %s", tpath
);
1247 goto out_free_contents
;
1252 f
= fopen(path
, "w");
1254 SYSERROR("reopening config for writing");
1258 fprintf(f
, "# Template used to create this container: %s\n", t
);
1260 fprintf(f
, "# Parameters passed to the template:");
1262 fprintf(f
, " %s", *argv
);
1268 fprintf(f
, "# Template script checksum (SHA-1): ");
1269 for (i
=0; i
<SHA_DIGEST_LENGTH
; i
++)
1270 fprintf(f
, "%02x", md_value
[i
]);
1273 fprintf(f
, "# For additional config options, please look at lxc.container.conf(5)\n");
1274 if (fwrite(contents
, 1, flen
, f
) != flen
) {
1275 SYSERROR("Writing original contents");
1291 SYSERROR("Error prepending header");
1297 static void lxcapi_clear_config(struct lxc_container
*c
)
1301 lxc_conf_free(c
->lxc_conf
);
1307 #define do_lxcapi_clear_config(c) lxcapi_clear_config(c)
1311 * create a container with the given parameters.
1312 * @c: container to be created. It has the lxcpath, name, and a starting
1313 * configuration already set
1314 * @t: the template to execute to instantiate the root filesystem and
1315 * adjust the configuration.
1316 * @bdevtype: backing store type to use. If NULL, dir will be used.
1317 * @specs: additional parameters for the backing store, i.e. LVM vg to
1320 * @argv: the arguments to pass to the template, terminated by NULL. If no
1321 * arguments, you can just pass NULL.
1323 static bool do_lxcapi_create(struct lxc_container
*c
, const char *t
,
1324 const char *bdevtype
, struct bdev_specs
*specs
, int flags
,
1336 tpath
= get_template_path(t
);
1338 ERROR("bad template: %s", t
);
1344 * If a template is passed in, and the rootfs already is defined in
1345 * the container config and exists, then * caller is trying to create
1346 * an existing container. Return an error, but do NOT delete the
1349 if (do_lxcapi_is_defined(c
) && c
->lxc_conf
&& c
->lxc_conf
->rootfs
.path
&&
1350 access(c
->lxc_conf
->rootfs
.path
, F_OK
) == 0 && tpath
) {
1351 ERROR("Container %s:%s already exists", c
->config_path
, c
->name
);
1356 if (!do_lxcapi_load_config(c
, lxc_global_config_value("lxc.default_config"))) {
1357 ERROR("Error loading default configuration file %s", lxc_global_config_value("lxc.default_config"));
1362 if (!create_container_dir(c
))
1366 * either template or rootfs.path should be set.
1367 * if both template and rootfs.path are set, template is setup as rootfs.path.
1368 * container is already created if we have a config and rootfs.path is accessible
1370 if (!c
->lxc_conf
->rootfs
.path
&& !tpath
)
1371 /* no template passed in and rootfs does not exist: error */
1373 if (c
->lxc_conf
->rootfs
.path
&& access(c
->lxc_conf
->rootfs
.path
, F_OK
) != 0)
1374 /* rootfs passed into configuration, but does not exist: error */
1376 if (do_lxcapi_is_defined(c
) && c
->lxc_conf
->rootfs
.path
&& !tpath
) {
1377 /* Rootfs already existed, user just wanted to save the
1378 * loaded configuration */
1383 /* Mark that this container is being created */
1384 if ((partial_fd
= create_partial(c
)) < 0)
1387 /* no need to get disk lock bc we have the partial locked */
1390 * Create the backing store
1391 * Note we can't do this in the same task as we use to execute the
1392 * template because of the way zfs works.
1393 * After you 'zfs create', zfs mounts the fs only in the initial
1398 SYSERROR("failed to fork task for container creation template");
1402 if (pid
== 0) { // child
1403 struct bdev
*bdev
= NULL
;
1405 if (!(bdev
= do_bdev_create(c
, bdevtype
, specs
))) {
1406 ERROR("Error creating backing store type %s for %s",
1407 bdevtype
? bdevtype
: "(none)", c
->name
);
1411 /* save config file again to store the new rootfs location */
1412 if (!do_lxcapi_save_config(c
, NULL
)) {
1413 ERROR("failed to save starting configuration for %s", c
->name
);
1414 // parent task won't see bdev in config so we delete it
1415 bdev
->ops
->umount(bdev
);
1416 bdev
->ops
->destroy(bdev
);
1421 if (wait_for_pid(pid
) != 0)
1424 /* reload config to get the rootfs */
1425 lxc_conf_free(c
->lxc_conf
);
1427 if (!load_config_locked(c
, c
->configfile
))
1430 if (!create_run_template(c
, tpath
, !!(flags
& LXC_CREATE_QUIET
), argv
))
1433 // now clear out the lxc_conf we have, reload from the created
1435 do_lxcapi_clear_config(c
);
1438 if (!prepend_lxc_header(c
->configfile
, tpath
, argv
)) {
1439 ERROR("Error prepending header to configuration file");
1443 ret
= load_config_locked(c
, c
->configfile
);
1446 if (partial_fd
>= 0)
1447 remove_partial(c
, partial_fd
);
1450 container_destroy(c
);
1456 static bool lxcapi_create(struct lxc_container
*c
, const char *t
,
1457 const char *bdevtype
, struct bdev_specs
*specs
, int flags
,
1461 struct lxc_conf
*old
= current_config
;
1462 current_config
= c
? c
->lxc_conf
: NULL
;
1463 ret
= do_lxcapi_create(c
, t
, bdevtype
, specs
, flags
, argv
);
1464 current_config
= old
;
1468 static bool do_lxcapi_reboot(struct lxc_container
*c
)
1471 int rebootsignal
= SIGINT
;
1475 if (!do_lxcapi_is_running(c
))
1477 pid
= do_lxcapi_init_pid(c
);
1480 if (c
->lxc_conf
&& c
->lxc_conf
->rebootsignal
)
1481 rebootsignal
= c
->lxc_conf
->rebootsignal
;
1482 if (kill(pid
, rebootsignal
) < 0)
1488 WRAP_API(bool, lxcapi_reboot
)
1490 static bool do_lxcapi_shutdown(struct lxc_container
*c
, int timeout
)
1494 int haltsignal
= SIGPWR
;
1499 if (!do_lxcapi_is_running(c
))
1501 pid
= do_lxcapi_init_pid(c
);
1504 if (c
->lxc_conf
&& c
->lxc_conf
->haltsignal
)
1505 haltsignal
= c
->lxc_conf
->haltsignal
;
1506 kill(pid
, haltsignal
);
1507 retv
= do_lxcapi_wait(c
, "STOPPED", timeout
);
1511 WRAP_API_1(bool, lxcapi_shutdown
, int)
1513 static bool lxcapi_createl(struct lxc_container
*c
, const char *t
,
1514 const char *bdevtype
, struct bdev_specs
*specs
, int flags
, ...)
1523 struct lxc_conf
*old
= current_config
;
1524 current_config
= c
? c
->lxc_conf
: NULL
;
1527 * since we're going to wait for create to finish, I don't think we
1528 * need to get a copy of the arguments.
1530 va_start(ap
, flags
);
1531 args
= lxc_va_arg_list_to_argv(ap
, 0, 0);
1534 ERROR("Memory allocation error.");
1538 bret
= do_lxcapi_create(c
, t
, bdevtype
, specs
, flags
, args
);
1542 current_config
= old
;
1546 static void do_clear_unexp_config_line(struct lxc_conf
*conf
, const char *key
)
1548 if (strcmp(key
, "lxc.cgroup") == 0)
1549 clear_unexp_config_line(conf
, key
, true);
1550 else if (strcmp(key
, "lxc.network") == 0)
1551 clear_unexp_config_line(conf
, key
, true);
1552 else if (strcmp(key
, "lxc.hook") == 0)
1553 clear_unexp_config_line(conf
, key
, true);
1555 clear_unexp_config_line(conf
, key
, false);
1556 if (!do_append_unexp_config_line(conf
, key
, ""))
1557 WARN("Error clearing configuration for %s", key
);
1560 static bool do_lxcapi_clear_config_item(struct lxc_container
*c
, const char *key
)
1564 if (!c
|| !c
->lxc_conf
)
1566 if (container_mem_lock(c
))
1568 ret
= lxc_clear_config_item(c
->lxc_conf
, key
);
1570 do_clear_unexp_config_line(c
->lxc_conf
, key
);
1571 container_mem_unlock(c
);
1575 WRAP_API_1(bool, lxcapi_clear_config_item
, const char *)
1577 static inline bool enter_net_ns(struct lxc_container
*c
)
1579 pid_t pid
= do_lxcapi_init_pid(c
);
1581 if ((geteuid() != 0 || (c
->lxc_conf
&& !lxc_list_empty(&c
->lxc_conf
->id_map
))) && access("/proc/self/ns/user", F_OK
) == 0) {
1582 if (!switch_to_ns(pid
, "user"))
1585 return switch_to_ns(pid
, "net");
1588 // used by qsort and bsearch functions for comparing names
1589 static inline int string_cmp(char **first
, char **second
)
1591 return strcmp(*first
, *second
);
1594 // used by qsort and bsearch functions for comparing container names
1595 static inline int container_cmp(struct lxc_container
**first
, struct lxc_container
**second
)
1597 return strcmp((*first
)->name
, (*second
)->name
);
1600 static bool add_to_array(char ***names
, char *cname
, int pos
)
1602 char **newnames
= realloc(*names
, (pos
+1) * sizeof(char *));
1604 ERROR("Out of memory");
1609 newnames
[pos
] = strdup(cname
);
1613 // sort the arrray as we will use binary search on it
1614 qsort(newnames
, pos
+ 1, sizeof(char *), (int (*)(const void *,const void *))string_cmp
);
1619 static bool add_to_clist(struct lxc_container
***list
, struct lxc_container
*c
, int pos
, bool sort
)
1621 struct lxc_container
**newlist
= realloc(*list
, (pos
+1) * sizeof(struct lxc_container
*));
1623 ERROR("Out of memory");
1630 // sort the arrray as we will use binary search on it
1632 qsort(newlist
, pos
+ 1, sizeof(struct lxc_container
*), (int (*)(const void *,const void *))container_cmp
);
1637 static char** get_from_array(char ***names
, char *cname
, int size
)
1639 return (char **)bsearch(&cname
, *names
, size
, sizeof(char *), (int (*)(const void *, const void *))string_cmp
);
1643 static bool array_contains(char ***names
, char *cname
, int size
) {
1644 if(get_from_array(names
, cname
, size
) != NULL
)
1649 static bool remove_from_array(char ***names
, char *cname
, int size
)
1651 char **result
= get_from_array(names
, cname
, size
);
1652 if (result
!= NULL
) {
1659 static char ** do_lxcapi_get_interfaces(struct lxc_container
*c
)
1662 int i
, count
= 0, pipefd
[2];
1663 char **interfaces
= NULL
;
1664 char interface
[IFNAMSIZ
];
1666 if(pipe(pipefd
) < 0) {
1667 SYSERROR("pipe failed");
1673 SYSERROR("failed to fork task to get interfaces information");
1679 if (pid
== 0) { // child
1680 int ret
= 1, nbytes
;
1681 struct ifaddrs
*interfaceArray
= NULL
, *tempIfAddr
= NULL
;
1683 /* close the read-end of the pipe */
1686 if (!enter_net_ns(c
)) {
1687 SYSERROR("failed to enter namespace");
1691 /* Grab the list of interfaces */
1692 if (getifaddrs(&interfaceArray
)) {
1693 SYSERROR("failed to get interfaces list");
1697 /* Iterate through the interfaces */
1698 for (tempIfAddr
= interfaceArray
; tempIfAddr
!= NULL
; tempIfAddr
= tempIfAddr
->ifa_next
) {
1699 nbytes
= write(pipefd
[1], tempIfAddr
->ifa_name
, IFNAMSIZ
);
1701 ERROR("write failed");
1710 freeifaddrs(interfaceArray
);
1712 /* close the write-end of the pipe, thus sending EOF to the reader */
1717 /* close the write-end of the pipe */
1720 while (read(pipefd
[0], &interface
, IFNAMSIZ
) == IFNAMSIZ
) {
1721 if (array_contains(&interfaces
, interface
, count
))
1724 if(!add_to_array(&interfaces
, interface
, count
))
1725 ERROR("PARENT: add_to_array failed");
1729 if (wait_for_pid(pid
) != 0) {
1730 for(i
=0;i
<count
;i
++)
1731 free(interfaces
[i
]);
1736 /* close the read-end of the pipe */
1739 /* Append NULL to the array */
1741 interfaces
= (char **)lxc_append_null_to_array((void **)interfaces
, count
);
1746 WRAP_API(char **, lxcapi_get_interfaces
)
1748 static char** do_lxcapi_get_ips(struct lxc_container
*c
, const char* interface
, const char* family
, int scope
)
1751 int i
, count
= 0, pipefd
[2];
1752 char **addresses
= NULL
;
1753 char address
[INET6_ADDRSTRLEN
];
1755 if(pipe(pipefd
) < 0) {
1756 SYSERROR("pipe failed");
1762 SYSERROR("failed to fork task to get container ips");
1768 if (pid
== 0) { // child
1769 int ret
= 1, nbytes
;
1770 struct ifaddrs
*interfaceArray
= NULL
, *tempIfAddr
= NULL
;
1771 char addressOutputBuffer
[INET6_ADDRSTRLEN
];
1772 void *tempAddrPtr
= NULL
;
1773 char *address
= NULL
;
1775 /* close the read-end of the pipe */
1778 if (!enter_net_ns(c
)) {
1779 SYSERROR("failed to enter namespace");
1783 /* Grab the list of interfaces */
1784 if (getifaddrs(&interfaceArray
)) {
1785 SYSERROR("failed to get interfaces list");
1789 /* Iterate through the interfaces */
1790 for (tempIfAddr
= interfaceArray
; tempIfAddr
!= NULL
; tempIfAddr
= tempIfAddr
->ifa_next
) {
1791 if (tempIfAddr
->ifa_addr
== NULL
)
1794 if(tempIfAddr
->ifa_addr
->sa_family
== AF_INET
) {
1795 if (family
&& strcmp(family
, "inet"))
1797 tempAddrPtr
= &((struct sockaddr_in
*)tempIfAddr
->ifa_addr
)->sin_addr
;
1800 if (family
&& strcmp(family
, "inet6"))
1803 if (((struct sockaddr_in6
*)tempIfAddr
->ifa_addr
)->sin6_scope_id
!= scope
)
1806 tempAddrPtr
= &((struct sockaddr_in6
*)tempIfAddr
->ifa_addr
)->sin6_addr
;
1809 if (interface
&& strcmp(interface
, tempIfAddr
->ifa_name
))
1811 else if (!interface
&& strcmp("lo", tempIfAddr
->ifa_name
) == 0)
1814 address
= (char *)inet_ntop(tempIfAddr
->ifa_addr
->sa_family
,
1816 addressOutputBuffer
,
1817 sizeof(addressOutputBuffer
));
1821 nbytes
= write(pipefd
[1], address
, INET6_ADDRSTRLEN
);
1823 ERROR("write failed");
1832 freeifaddrs(interfaceArray
);
1834 /* close the write-end of the pipe, thus sending EOF to the reader */
1839 /* close the write-end of the pipe */
1842 while (read(pipefd
[0], &address
, INET6_ADDRSTRLEN
) == INET6_ADDRSTRLEN
) {
1843 if(!add_to_array(&addresses
, address
, count
))
1844 ERROR("PARENT: add_to_array failed");
1848 if (wait_for_pid(pid
) != 0) {
1849 for(i
=0;i
<count
;i
++)
1855 /* close the read-end of the pipe */
1858 /* Append NULL to the array */
1860 addresses
= (char **)lxc_append_null_to_array((void **)addresses
, count
);
1865 WRAP_API_3(char **, lxcapi_get_ips
, const char *, const char *, int)
1867 static int do_lxcapi_get_config_item(struct lxc_container
*c
, const char *key
, char *retv
, int inlen
)
1871 if (!c
|| !c
->lxc_conf
)
1873 if (container_mem_lock(c
))
1875 ret
= lxc_get_config_item(c
->lxc_conf
, key
, retv
, inlen
);
1876 container_mem_unlock(c
);
1880 WRAP_API_3(int, lxcapi_get_config_item
, const char *, char *, int)
1882 static char* do_lxcapi_get_running_config_item(struct lxc_container
*c
, const char *key
)
1886 if (!c
|| !c
->lxc_conf
)
1888 if (container_mem_lock(c
))
1890 ret
= lxc_cmd_get_config_item(c
->name
, key
, do_lxcapi_get_config_path(c
));
1891 container_mem_unlock(c
);
1895 WRAP_API_1(char *, lxcapi_get_running_config_item
, const char *)
1897 static int do_lxcapi_get_keys(struct lxc_container
*c
, const char *key
, char *retv
, int inlen
)
1900 return lxc_listconfigs(retv
, inlen
);
1902 * Support 'lxc.network.<idx>', i.e. 'lxc.network.0'
1903 * This is an intelligent result to show which keys are valid given
1904 * the type of nic it is
1906 if (!c
|| !c
->lxc_conf
)
1908 if (container_mem_lock(c
))
1911 if (strncmp(key
, "lxc.network.", 12) == 0)
1912 ret
= lxc_list_nicconfigs(c
->lxc_conf
, key
, retv
, inlen
);
1913 container_mem_unlock(c
);
1917 WRAP_API_3(int, lxcapi_get_keys
, const char *, char *, int)
1919 static bool do_lxcapi_save_config(struct lxc_container
*c
, const char *alt_file
)
1922 bool ret
= false, need_disklock
= false;
1926 alt_file
= c
->configfile
;
1928 return false; // should we write to stdout if no file is specified?
1930 // If we haven't yet loaded a config, load the stock config
1932 if (!do_lxcapi_load_config(c
, lxc_global_config_value("lxc.default_config"))) {
1933 ERROR("Error loading default configuration file %s while saving %s", lxc_global_config_value("lxc.default_config"), c
->name
);
1938 if (!create_container_dir(c
))
1942 * If we're writing to the container's config file, take the
1943 * disk lock. Otherwise just take the memlock to protect the
1944 * struct lxc_container while we're traversing it.
1946 if (strcmp(c
->configfile
, alt_file
) == 0)
1947 need_disklock
= true;
1950 lret
= container_disk_lock(c
);
1952 lret
= container_mem_lock(c
);
1957 fout
= fopen(alt_file
, "w");
1960 write_config(fout
, c
->lxc_conf
);
1966 container_disk_unlock(c
);
1968 container_mem_unlock(c
);
1972 WRAP_API_1(bool, lxcapi_save_config
, const char *)
1974 static bool mod_rdep(struct lxc_container
*c
, bool inc
)
1976 char path
[MAXPATHLEN
];
1981 if (container_disk_lock(c
))
1983 ret
= snprintf(path
, MAXPATHLEN
, "%s/%s/lxc_snapshots", c
->config_path
,
1985 if (ret
< 0 || ret
> MAXPATHLEN
)
1987 f
= fopen(path
, "r");
1989 ret
= fscanf(f
, "%d", &v
);
1992 ERROR("Corrupted file %s", path
);
1997 f
= fopen(path
, "w");
2000 if (fprintf(f
, "%d\n", v
) < 0) {
2001 ERROR("Error writing new snapshots value");
2007 SYSERROR("Error writing to or closing snapshots file");
2014 container_disk_unlock(c
);
2018 static void strip_newline(char *p
)
2020 size_t len
= strlen(p
);
2023 if (p
[len
-1] == '\n')
2027 static void mod_all_rdeps(struct lxc_container
*c
, bool inc
)
2029 struct lxc_container
*p
;
2030 char *lxcpath
= NULL
, *lxcname
= NULL
, path
[MAXPATHLEN
];
2031 size_t pathlen
= 0, namelen
= 0;
2035 ret
= snprintf(path
, MAXPATHLEN
, "%s/%s/lxc_rdepends",
2036 c
->config_path
, c
->name
);
2037 if (ret
< 0 || ret
>= MAXPATHLEN
) {
2038 ERROR("Path name too long");
2041 f
= fopen(path
, "r");
2044 while (getline(&lxcpath
, &pathlen
, f
) != -1) {
2045 if (getline(&lxcname
, &namelen
, f
) == -1) {
2046 ERROR("badly formatted file %s", path
);
2049 strip_newline(lxcpath
);
2050 strip_newline(lxcname
);
2051 if ((p
= lxc_container_new(lxcname
, lxcpath
)) == NULL
) {
2052 ERROR("Unable to find dependent container %s:%s",
2056 if (!mod_rdep(p
, inc
))
2057 ERROR("Failed to increase numsnapshots for %s:%s",
2059 lxc_container_put(p
);
2067 static bool has_fs_snapshots(struct lxc_container
*c
)
2069 char path
[MAXPATHLEN
];
2074 ret
= snprintf(path
, MAXPATHLEN
, "%s/%s/lxc_snapshots", c
->config_path
,
2076 if (ret
< 0 || ret
> MAXPATHLEN
)
2078 f
= fopen(path
, "r");
2081 ret
= fscanf(f
, "%d", &v
);
2091 static bool has_snapshots(struct lxc_container
*c
)
2093 char path
[MAXPATHLEN
];
2094 struct dirent dirent
, *direntp
;
2098 if (!get_snappath_dir(c
, path
))
2100 dir
= opendir(path
);
2103 while (!readdir_r(dir
, &dirent
, &direntp
)) {
2107 if (!strcmp(direntp
->d_name
, "."))
2110 if (!strcmp(direntp
->d_name
, ".."))
2119 static int lxc_rmdir_onedev_wrapper(void *data
)
2121 char *arg
= (char *) data
;
2122 return lxc_rmdir_onedev(arg
, "snaps");
2125 static int do_bdev_destroy(struct lxc_conf
*conf
)
2130 r
= bdev_init(conf
, conf
->rootfs
.path
, conf
->rootfs
.mount
, NULL
);
2134 if (r
->ops
->destroy(r
) < 0)
2140 static int bdev_destroy_wrapper(void *data
)
2142 struct lxc_conf
*conf
= data
;
2144 if (setgid(0) < 0) {
2145 ERROR("Failed to setgid to 0");
2148 if (setgroups(0, NULL
) < 0)
2149 WARN("Failed to clear groups");
2150 if (setuid(0) < 0) {
2151 ERROR("Failed to setuid to 0");
2154 return do_bdev_destroy(conf
);
2157 static bool container_destroy(struct lxc_container
*c
)
2162 if (!c
|| !do_lxcapi_is_defined(c
))
2165 if (container_disk_lock(c
))
2168 if (!is_stopped(c
)) {
2169 // we should queue some sort of error - in c->error_string?
2170 ERROR("container %s is not stopped", c
->name
);
2174 if (c
->lxc_conf
&& c
->lxc_conf
->rootfs
.path
&& c
->lxc_conf
->rootfs
.mount
) {
2176 ret
= userns_exec_1(c
->lxc_conf
, bdev_destroy_wrapper
, c
->lxc_conf
);
2178 ret
= do_bdev_destroy(c
->lxc_conf
);
2180 ERROR("Error destroying rootfs for %s", c
->name
);
2185 mod_all_rdeps(c
, false);
2187 const char *p1
= do_lxcapi_get_config_path(c
);
2188 char *path
= alloca(strlen(p1
) + strlen(c
->name
) + 2);
2189 sprintf(path
, "%s/%s", p1
, c
->name
);
2191 ret
= userns_exec_1(c
->lxc_conf
, lxc_rmdir_onedev_wrapper
, path
);
2193 ret
= lxc_rmdir_onedev(path
, "snaps");
2195 ERROR("Error destroying container directory for %s", c
->name
);
2201 container_disk_unlock(c
);
2205 static bool do_lxcapi_destroy(struct lxc_container
*c
)
2207 if (!c
|| !lxcapi_is_defined(c
))
2209 if (has_snapshots(c
)) {
2210 ERROR("Container %s has snapshots; not removing", c
->name
);
2214 if (has_fs_snapshots(c
)) {
2215 ERROR("container %s has snapshots on its rootfs", c
->name
);
2219 return container_destroy(c
);
2222 WRAP_API(bool, lxcapi_destroy
)
2224 static bool do_lxcapi_destroy_with_snapshots(struct lxc_container
*c
)
2226 if (!c
|| !lxcapi_is_defined(c
))
2228 if (!lxcapi_snapshot_destroy_all(c
)) {
2229 ERROR("Error deleting all snapshots");
2232 return lxcapi_destroy(c
);
2235 WRAP_API(bool, lxcapi_destroy_with_snapshots
)
2237 static bool set_config_item_locked(struct lxc_container
*c
, const char *key
, const char *v
)
2239 struct lxc_config_t
*config
;
2242 c
->lxc_conf
= lxc_conf_init();
2245 config
= lxc_getconfig(key
);
2248 if (config
->cb(key
, v
, c
->lxc_conf
) != 0)
2250 return do_append_unexp_config_line(c
->lxc_conf
, key
, v
);
2253 static bool do_lxcapi_set_config_item(struct lxc_container
*c
, const char *key
, const char *v
)
2260 if (container_mem_lock(c
))
2263 b
= set_config_item_locked(c
, key
, v
);
2265 container_mem_unlock(c
);
2269 WRAP_API_2(bool, lxcapi_set_config_item
, const char *, const char *)
2271 static char *lxcapi_config_file_name(struct lxc_container
*c
)
2273 if (!c
|| !c
->configfile
)
2275 return strdup(c
->configfile
);
2278 static const char *lxcapi_get_config_path(struct lxc_container
*c
)
2280 if (!c
|| !c
->config_path
)
2282 return (const char *)(c
->config_path
);
2287 * Just recalculate the c->configfile based on the
2288 * c->config_path, which must be set.
2289 * The lxc_container must be locked or not yet public.
2291 static bool set_config_filename(struct lxc_container
*c
)
2296 if (!c
->config_path
)
2299 /* $lxc_path + "/" + c->name + "/" + "config" + '\0' */
2300 len
= strlen(c
->config_path
) + strlen(c
->name
) + strlen("config") + 3;
2301 newpath
= malloc(len
);
2305 ret
= snprintf(newpath
, len
, "%s/%s/config", c
->config_path
, c
->name
);
2306 if (ret
< 0 || ret
>= len
) {
2307 fprintf(stderr
, "Error printing out config file name\n");
2312 free(c
->configfile
);
2313 c
->configfile
= newpath
;
2318 static bool do_lxcapi_set_config_path(struct lxc_container
*c
, const char *path
)
2322 char *oldpath
= NULL
;
2327 if (container_mem_lock(c
))
2332 ERROR("Out of memory setting new lxc path");
2338 oldpath
= c
->config_path
;
2341 /* Since we've changed the config path, we have to change the
2342 * config file name too */
2343 if (!set_config_filename(c
)) {
2344 ERROR("Out of memory setting new config filename");
2346 free(c
->config_path
);
2347 c
->config_path
= oldpath
;
2352 container_mem_unlock(c
);
2356 WRAP_API_1(bool, lxcapi_set_config_path
, const char *)
2358 static bool do_lxcapi_set_cgroup_item(struct lxc_container
*c
, const char *subsys
, const char *value
)
2368 if (container_disk_lock(c
))
2371 ret
= lxc_cgroup_set(subsys
, value
, c
->name
, c
->config_path
);
2373 container_disk_unlock(c
);
2377 WRAP_API_2(bool, lxcapi_set_cgroup_item
, const char *, const char *)
2379 static int do_lxcapi_get_cgroup_item(struct lxc_container
*c
, const char *subsys
, char *retv
, int inlen
)
2389 if (container_disk_lock(c
))
2392 ret
= lxc_cgroup_get(subsys
, retv
, inlen
, c
->name
, c
->config_path
);
2394 container_disk_unlock(c
);
2398 WRAP_API_3(int, lxcapi_get_cgroup_item
, const char *, char *, int)
2400 const char *lxc_get_global_config_item(const char *key
)
2402 return lxc_global_config_value(key
);
2405 const char *lxc_get_version(void)
2410 static int copy_file(const char *old
, const char *new)
2417 if (file_exists(new)) {
2418 ERROR("copy destination %s exists", new);
2421 ret
= stat(old
, &sbuf
);
2423 INFO("Error stat'ing %s", old
);
2427 in
= open(old
, O_RDONLY
);
2429 SYSERROR("Error opening original file %s", old
);
2432 out
= open(new, O_CREAT
| O_EXCL
| O_WRONLY
, 0644);
2434 SYSERROR("Error opening new file %s", new);
2440 len
= read(in
, buf
, 8096);
2442 SYSERROR("Error reading old file %s", old
);
2447 ret
= write(out
, buf
, len
);
2448 if (ret
< len
) { // should we retry?
2449 SYSERROR("Error: write to new file %s was interrupted", new);
2456 // we set mode, but not owner/group
2457 ret
= chmod(new, sbuf
.st_mode
);
2459 SYSERROR("Error setting mode on %s", new);
2471 static int copyhooks(struct lxc_container
*oldc
, struct lxc_container
*c
)
2474 struct lxc_list
*it
;
2477 len
= strlen(oldc
->config_path
) + strlen(oldc
->name
) + 3;
2478 cpath
= alloca(len
);
2479 ret
= snprintf(cpath
, len
, "%s/%s/", oldc
->config_path
, oldc
->name
);
2480 if (ret
< 0 || ret
>= len
)
2483 for (i
=0; i
<NUM_LXC_HOOKS
; i
++) {
2484 lxc_list_for_each(it
, &c
->lxc_conf
->hooks
[i
]) {
2485 char *hookname
= it
->elem
;
2486 char *fname
= strrchr(hookname
, '/');
2487 char tmppath
[MAXPATHLEN
];
2488 if (!fname
) // relative path - we don't support, but maybe we should
2490 if (strncmp(hookname
, cpath
, len
- 1) != 0) {
2491 // this hook is public - ignore
2494 // copy the script, and change the entry in confile
2495 ret
= snprintf(tmppath
, MAXPATHLEN
, "%s/%s/%s",
2496 c
->config_path
, c
->name
, fname
+1);
2497 if (ret
< 0 || ret
>= MAXPATHLEN
)
2499 ret
= copy_file(it
->elem
, tmppath
);
2503 it
->elem
= strdup(tmppath
);
2505 ERROR("out of memory copying hook path");
2511 if (!clone_update_unexp_hooks(c
->lxc_conf
, oldc
->config_path
,
2512 c
->config_path
, oldc
->name
, c
->name
)) {
2513 ERROR("Error saving new hooks in clone");
2516 do_lxcapi_save_config(c
, NULL
);
2521 static int copy_fstab(struct lxc_container
*oldc
, struct lxc_container
*c
)
2523 char newpath
[MAXPATHLEN
];
2524 char *oldpath
= oldc
->lxc_conf
->fstab
;
2530 clear_unexp_config_line(c
->lxc_conf
, "lxc.mount", false);
2532 char *p
= strrchr(oldpath
, '/');
2535 ret
= snprintf(newpath
, MAXPATHLEN
, "%s/%s%s",
2536 c
->config_path
, c
->name
, p
);
2537 if (ret
< 0 || ret
>= MAXPATHLEN
) {
2538 ERROR("error printing new path for %s", oldpath
);
2541 if (file_exists(newpath
)) {
2542 ERROR("error: fstab file %s exists", newpath
);
2546 if (copy_file(oldpath
, newpath
) < 0) {
2547 ERROR("error: copying %s to %s", oldpath
, newpath
);
2550 free(c
->lxc_conf
->fstab
);
2551 c
->lxc_conf
->fstab
= strdup(newpath
);
2552 if (!c
->lxc_conf
->fstab
) {
2553 ERROR("error: allocating pathname");
2556 if (!do_append_unexp_config_line(c
->lxc_conf
, "lxc.mount", newpath
)) {
2557 ERROR("error saving new lxctab");
2564 static void copy_rdepends(struct lxc_container
*c
, struct lxc_container
*c0
)
2566 char path0
[MAXPATHLEN
], path1
[MAXPATHLEN
];
2569 ret
= snprintf(path0
, MAXPATHLEN
, "%s/%s/lxc_rdepends", c0
->config_path
,
2571 if (ret
< 0 || ret
>= MAXPATHLEN
) {
2572 WARN("Error copying reverse dependencies");
2575 ret
= snprintf(path1
, MAXPATHLEN
, "%s/%s/lxc_rdepends", c
->config_path
,
2577 if (ret
< 0 || ret
>= MAXPATHLEN
) {
2578 WARN("Error copying reverse dependencies");
2581 if (copy_file(path0
, path1
) < 0) {
2582 INFO("Error copying reverse dependencies");
2587 static bool add_rdepends(struct lxc_container
*c
, struct lxc_container
*c0
)
2590 char path
[MAXPATHLEN
];
2594 ret
= snprintf(path
, MAXPATHLEN
, "%s/%s/lxc_rdepends", c
->config_path
,
2596 if (ret
< 0 || ret
>= MAXPATHLEN
)
2598 f
= fopen(path
, "a");
2602 // if anything goes wrong, just return an error
2603 if (fprintf(f
, "%s\n%s\n", c0
->config_path
, c0
->name
) < 0)
2610 static int copy_storage(struct lxc_container
*c0
, struct lxc_container
*c
,
2611 const char *newtype
, int flags
, const char *bdevdata
, uint64_t newsize
)
2616 bdev
= bdev_copy(c0
, c
->name
, c
->config_path
, newtype
, flags
,
2617 bdevdata
, newsize
, &need_rdep
);
2619 ERROR("Error copying storage");
2622 free(c
->lxc_conf
->rootfs
.path
);
2623 c
->lxc_conf
->rootfs
.path
= strdup(bdev
->src
);
2625 if (!c
->lxc_conf
->rootfs
.path
) {
2626 ERROR("Out of memory while setting storage path");
2629 // We will simply append a new lxc.rootfs entry to the unexpanded config
2630 clear_unexp_config_line(c
->lxc_conf
, "lxc.rootfs", false);
2631 if (!do_append_unexp_config_line(c
->lxc_conf
, "lxc.rootfs", c
->lxc_conf
->rootfs
.path
)) {
2632 ERROR("Error saving new rootfs to cloend config");
2635 if (flags
& LXC_CLONE_SNAPSHOT
)
2636 copy_rdepends(c
, c0
);
2638 if (!add_rdepends(c
, c0
))
2639 WARN("Error adding reverse dependency from %s to %s",
2643 mod_all_rdeps(c
, true);
2648 struct clone_update_data
{
2649 struct lxc_container
*c0
;
2650 struct lxc_container
*c1
;
2655 static int clone_update_rootfs(struct clone_update_data
*data
)
2657 struct lxc_container
*c0
= data
->c0
;
2658 struct lxc_container
*c
= data
->c1
;
2659 int flags
= data
->flags
;
2660 char **hookargs
= data
->hookargs
;
2662 char path
[MAXPATHLEN
];
2665 struct lxc_conf
*conf
= c
->lxc_conf
;
2667 /* update hostname in rootfs */
2668 /* we're going to mount, so run in a clean namespace to simplify cleanup */
2670 if (setgid(0) < 0) {
2671 ERROR("Failed to setgid to 0");
2674 if (setuid(0) < 0) {
2675 ERROR("Failed to setuid to 0");
2678 if (setgroups(0, NULL
) < 0)
2679 WARN("Failed to clear groups");
2681 if (unshare(CLONE_NEWNS
) < 0)
2683 bdev
= bdev_init(c
->lxc_conf
, c
->lxc_conf
->rootfs
.path
, c
->lxc_conf
->rootfs
.mount
, NULL
);
2686 if (strcmp(bdev
->type
, "dir") != 0) {
2687 if (unshare(CLONE_NEWNS
) < 0) {
2688 ERROR("error unsharing mounts");
2692 if (detect_shared_rootfs()) {
2693 if (mount(NULL
, "/", NULL
, MS_SLAVE
|MS_REC
, NULL
)) {
2694 SYSERROR("Failed to make / rslave");
2695 ERROR("Continuing...");
2698 if (bdev
->ops
->mount(bdev
) < 0) {
2702 } else { // TODO come up with a better way
2704 bdev
->dest
= strdup(bdev
->src
);
2707 if (!lxc_list_empty(&conf
->hooks
[LXCHOOK_CLONE
])) {
2708 /* Start of environment variable setup for hooks */
2709 if (setenv("LXC_SRC_NAME", c0
->name
, 1)) {
2710 SYSERROR("failed to set environment variable for source container name");
2712 if (setenv("LXC_NAME", c
->name
, 1)) {
2713 SYSERROR("failed to set environment variable for container name");
2715 if (setenv("LXC_CONFIG_FILE", conf
->rcfile
, 1)) {
2716 SYSERROR("failed to set environment variable for config path");
2718 if (setenv("LXC_ROOTFS_MOUNT", bdev
->dest
, 1)) {
2719 SYSERROR("failed to set environment variable for rootfs mount");
2721 if (setenv("LXC_ROOTFS_PATH", conf
->rootfs
.path
, 1)) {
2722 SYSERROR("failed to set environment variable for rootfs mount");
2725 if (run_lxc_hooks(c
->name
, "clone", conf
, c
->get_config_path(c
), hookargs
)) {
2726 ERROR("Error executing clone hook for %s", c
->name
);
2732 if (!(flags
& LXC_CLONE_KEEPNAME
)) {
2733 ret
= snprintf(path
, MAXPATHLEN
, "%s/etc/hostname", bdev
->dest
);
2736 if (ret
< 0 || ret
>= MAXPATHLEN
)
2738 if (!file_exists(path
))
2740 if (!(fout
= fopen(path
, "w"))) {
2741 SYSERROR("unable to open %s: ignoring", path
);
2744 if (fprintf(fout
, "%s", c
->name
) < 0) {
2748 if (fclose(fout
) < 0)
2757 static int clone_update_rootfs_wrapper(void *data
)
2759 struct clone_update_data
*arg
= (struct clone_update_data
*) data
;
2760 return clone_update_rootfs(arg
);
2764 * We want to support:
2765 sudo lxc-clone -o o1 -n n1 -s -L|-fssize fssize -v|--vgname vgname \
2766 -p|--lvprefix lvprefix -t|--fstype fstype -B backingstore
2768 -s [ implies overlayfs]
2772 only rootfs gets converted (copied/snapshotted) on clone.
2775 static int create_file_dirname(char *path
, struct lxc_conf
*conf
)
2777 char *p
= strrchr(path
, '/');
2783 ret
= do_create_container_dir(path
, conf
);
2788 static struct lxc_container
*do_lxcapi_clone(struct lxc_container
*c
, const char *newname
,
2789 const char *lxcpath
, int flags
,
2790 const char *bdevtype
, const char *bdevdata
, uint64_t newsize
,
2793 struct lxc_container
*c2
= NULL
;
2794 char newpath
[MAXPATHLEN
];
2795 int ret
, storage_copied
= 0;
2796 char *origroot
= NULL
;
2797 struct clone_update_data data
;
2801 if (!c
|| !do_lxcapi_is_defined(c
))
2804 if (container_mem_lock(c
))
2807 if (!is_stopped(c
)) {
2808 ERROR("error: Original container (%s) is running", c
->name
);
2812 // Make sure the container doesn't yet exist.
2816 lxcpath
= do_lxcapi_get_config_path(c
);
2817 ret
= snprintf(newpath
, MAXPATHLEN
, "%s/%s/config", lxcpath
, newname
);
2818 if (ret
< 0 || ret
>= MAXPATHLEN
) {
2819 SYSERROR("clone: failed making config pathname");
2822 if (file_exists(newpath
)) {
2823 ERROR("error: clone: %s exists", newpath
);
2827 ret
= create_file_dirname(newpath
, c
->lxc_conf
);
2828 if (ret
< 0 && errno
!= EEXIST
) {
2829 ERROR("Error creating container dir for %s", newpath
);
2833 // copy the configuration, tweak it as needed,
2834 if (c
->lxc_conf
->rootfs
.path
) {
2835 origroot
= c
->lxc_conf
->rootfs
.path
;
2836 c
->lxc_conf
->rootfs
.path
= NULL
;
2838 fout
= fopen(newpath
, "w");
2840 SYSERROR("open %s", newpath
);
2843 write_config(fout
, c
->lxc_conf
);
2845 c
->lxc_conf
->rootfs
.path
= origroot
;
2847 sprintf(newpath
, "%s/%s/rootfs", lxcpath
, newname
);
2848 if (mkdir(newpath
, 0755) < 0) {
2849 SYSERROR("error creating %s", newpath
);
2854 if (chown_mapped_root(newpath
, c
->lxc_conf
) < 0) {
2855 ERROR("Error chowning %s to container root", newpath
);
2860 c2
= lxc_container_new(newname
, lxcpath
);
2862 ERROR("clone: failed to create new container (%s %s)", newname
,
2867 // copy/snapshot rootfs's
2868 ret
= copy_storage(c
, c2
, bdevtype
, flags
, bdevdata
, newsize
);
2872 clear_unexp_config_line(c2
->lxc_conf
, "lxc.utsname", false);
2875 if (!set_config_item_locked(c2
, "lxc.utsname", newname
)) {
2876 ERROR("Error setting new hostname");
2881 ret
= copyhooks(c
, c2
);
2883 ERROR("error copying hooks");
2887 if (copy_fstab(c
, c2
) < 0) {
2888 ERROR("error copying fstab");
2893 if (!(flags
& LXC_CLONE_KEEPMACADDR
)) {
2894 if (!network_new_hwaddrs(c2
->lxc_conf
)) {
2895 ERROR("Error updating mac addresses");
2900 // We've now successfully created c2's storage, so clear it out if we
2904 if (!c2
->save_config(c2
, NULL
))
2907 if ((pid
= fork()) < 0) {
2912 ret
= wait_for_pid(pid
);
2915 container_mem_unlock(c
);
2921 data
.hookargs
= hookargs
;
2923 ret
= userns_exec_1(c
->lxc_conf
, clone_update_rootfs_wrapper
,
2926 ret
= clone_update_rootfs(&data
);
2930 container_mem_unlock(c
);
2934 container_mem_unlock(c
);
2936 if (!storage_copied
)
2937 c2
->lxc_conf
->rootfs
.path
= NULL
;
2939 lxc_container_put(c2
);
2945 static struct lxc_container
*lxcapi_clone(struct lxc_container
*c
, const char *newname
,
2946 const char *lxcpath
, int flags
,
2947 const char *bdevtype
, const char *bdevdata
, uint64_t newsize
,
2950 struct lxc_container
* ret
;
2951 struct lxc_conf
*old
= current_config
;
2952 current_config
= c
? c
->lxc_conf
: NULL
;
2953 ret
= do_lxcapi_clone(c
, newname
, lxcpath
, flags
, bdevtype
, bdevdata
, newsize
, hookargs
);
2954 current_config
= old
;
2958 static bool do_lxcapi_rename(struct lxc_container
*c
, const char *newname
)
2961 struct lxc_container
*newc
;
2963 if (!c
|| !c
->name
|| !c
->config_path
|| !c
->lxc_conf
)
2966 if (has_fs_snapshots(c
) || has_snapshots(c
)) {
2967 ERROR("Renaming a container with snapshots is not supported");
2970 bdev
= bdev_init(c
->lxc_conf
, c
->lxc_conf
->rootfs
.path
, c
->lxc_conf
->rootfs
.mount
, NULL
);
2972 ERROR("Failed to find original backing store type");
2976 newc
= lxcapi_clone(c
, newname
, c
->config_path
, LXC_CLONE_KEEPMACADDR
, NULL
, bdev
->type
, 0, NULL
);
2979 lxc_container_put(newc
);
2983 if (newc
&& lxcapi_is_defined(newc
))
2984 lxc_container_put(newc
);
2986 if (!container_destroy(c
)) {
2987 ERROR("Could not destroy existing container %s", c
->name
);
2993 WRAP_API_1(bool, lxcapi_rename
, const char *)
2995 static int lxcapi_attach(struct lxc_container
*c
, lxc_attach_exec_t exec_function
, void *exec_payload
, lxc_attach_options_t
*options
, pid_t
*attached_process
)
2997 struct lxc_conf
*old
= current_config
;
3003 current_config
= c
->lxc_conf
;
3005 ret
= lxc_attach(c
->name
, c
->config_path
, exec_function
, exec_payload
, options
, attached_process
);
3006 current_config
= old
;
3010 static int do_lxcapi_attach_run_wait(struct lxc_container
*c
, lxc_attach_options_t
*options
, const char *program
, const char * const argv
[])
3012 lxc_attach_command_t command
;
3019 command
.program
= (char*)program
;
3020 command
.argv
= (char**)argv
;
3021 r
= lxc_attach(c
->name
, c
->config_path
, lxc_attach_run_command
, &command
, options
, &pid
);
3026 return lxc_wait_for_pid_status(pid
);
3029 static int lxcapi_attach_run_wait(struct lxc_container
*c
, lxc_attach_options_t
*options
, const char *program
, const char * const argv
[])
3032 struct lxc_conf
*old
= current_config
;
3033 current_config
= c
? c
->lxc_conf
: NULL
;
3034 ret
= do_lxcapi_attach_run_wait(c
, options
, program
, argv
);
3035 current_config
= old
;
3039 static int get_next_index(const char *lxcpath
, char *cname
)
3045 fname
= alloca(strlen(lxcpath
) + 20);
3047 sprintf(fname
, "%s/snap%d", lxcpath
, i
);
3048 ret
= stat(fname
, &sb
);
3055 static bool get_snappath_dir(struct lxc_container
*c
, char *snappath
)
3059 * If the old style snapshot path exists, use it
3060 * /var/lib/lxc -> /var/lib/lxcsnaps
3062 ret
= snprintf(snappath
, MAXPATHLEN
, "%ssnaps", c
->config_path
);
3063 if (ret
< 0 || ret
>= MAXPATHLEN
)
3065 if (dir_exists(snappath
)) {
3066 ret
= snprintf(snappath
, MAXPATHLEN
, "%ssnaps/%s", c
->config_path
, c
->name
);
3067 if (ret
< 0 || ret
>= MAXPATHLEN
)
3073 * Use the new style path
3074 * /var/lib/lxc -> /var/lib/lxc + c->name + /snaps + \0
3076 ret
= snprintf(snappath
, MAXPATHLEN
, "%s/%s/snaps", c
->config_path
, c
->name
);
3077 if (ret
< 0 || ret
>= MAXPATHLEN
)
3082 static int do_lxcapi_snapshot(struct lxc_container
*c
, const char *commentfile
)
3085 struct lxc_container
*c2
;
3086 char snappath
[MAXPATHLEN
], newname
[20];
3088 if (!c
|| !lxcapi_is_defined(c
))
3091 if (!bdev_can_backup(c
->lxc_conf
)) {
3092 ERROR("%s's backing store cannot be backed up.", c
->name
);
3093 ERROR("Your container must use another backing store type.");
3097 if (!get_snappath_dir(c
, snappath
))
3100 i
= get_next_index(snappath
, c
->name
);
3102 if (mkdir_p(snappath
, 0755) < 0) {
3103 ERROR("Failed to create snapshot directory %s", snappath
);
3107 ret
= snprintf(newname
, 20, "snap%d", i
);
3108 if (ret
< 0 || ret
>= 20)
3112 * We pass LXC_CLONE_SNAPSHOT to make sure that a rdepends file entry is
3113 * created in the original container
3115 flags
= LXC_CLONE_SNAPSHOT
| LXC_CLONE_KEEPMACADDR
| LXC_CLONE_KEEPNAME
|
3116 LXC_CLONE_KEEPBDEVTYPE
| LXC_CLONE_MAYBE_SNAPSHOT
;
3117 if (bdev_is_dir(c
->lxc_conf
, c
->lxc_conf
->rootfs
.path
)) {
3118 ERROR("Snapshot of directory-backed container requested.");
3119 ERROR("Making a copy-clone. If you do want snapshots, then");
3120 ERROR("please create an aufs or overlayfs clone first, snapshot that");
3121 ERROR("and keep the original container pristine.");
3122 flags
&= ~LXC_CLONE_SNAPSHOT
| LXC_CLONE_MAYBE_SNAPSHOT
;
3124 c2
= do_lxcapi_clone(c
, newname
, snappath
, flags
, NULL
, NULL
, 0, NULL
);
3126 ERROR("clone of %s:%s failed", c
->config_path
, c
->name
);
3130 lxc_container_put(c2
);
3132 // Now write down the creation time
3139 tm_info
= localtime(&timer
);
3141 strftime(buffer
, 25, "%Y:%m:%d %H:%M:%S", tm_info
);
3143 char *dfnam
= alloca(strlen(snappath
) + strlen(newname
) + 5);
3144 sprintf(dfnam
, "%s/%s/ts", snappath
, newname
);
3145 f
= fopen(dfnam
, "w");
3147 ERROR("Failed to open %s", dfnam
);
3150 if (fprintf(f
, "%s", buffer
) < 0) {
3151 SYSERROR("Writing timestamp");
3157 SYSERROR("Writing timestamp");
3162 // $p / $name / comment \0
3163 int len
= strlen(snappath
) + strlen(newname
) + 10;
3164 char *path
= alloca(len
);
3165 sprintf(path
, "%s/%s/comment", snappath
, newname
);
3166 return copy_file(commentfile
, path
) < 0 ? -1 : i
;
3172 WRAP_API_1(int, lxcapi_snapshot
, const char *)
3174 static void lxcsnap_free(struct lxc_snapshot
*s
)
3177 free(s
->comment_pathname
);
3182 static char *get_snapcomment_path(char* snappath
, char *name
)
3184 // $snappath/$name/comment
3185 int ret
, len
= strlen(snappath
) + strlen(name
) + 10;
3186 char *s
= malloc(len
);
3189 ret
= snprintf(s
, len
, "%s/%s/comment", snappath
, name
);
3190 if (ret
< 0 || ret
>= len
) {
3198 static char *get_timestamp(char* snappath
, char *name
)
3200 char path
[MAXPATHLEN
], *s
= NULL
;
3204 ret
= snprintf(path
, MAXPATHLEN
, "%s/%s/ts", snappath
, name
);
3205 if (ret
< 0 || ret
>= MAXPATHLEN
)
3207 fin
= fopen(path
, "r");
3210 (void) fseek(fin
, 0, SEEK_END
);
3212 (void) fseek(fin
, 0, SEEK_SET
);
3217 if (fread(s
, 1, len
, fin
) != len
) {
3218 SYSERROR("reading timestamp");
3228 static int do_lxcapi_snapshot_list(struct lxc_container
*c
, struct lxc_snapshot
**ret_snaps
)
3230 char snappath
[MAXPATHLEN
], path2
[MAXPATHLEN
];
3232 struct dirent dirent
, *direntp
;
3233 struct lxc_snapshot
*snaps
=NULL
, *nsnaps
;
3236 if (!c
|| !lxcapi_is_defined(c
))
3239 if (!get_snappath_dir(c
, snappath
)) {
3240 ERROR("path name too long");
3243 dir
= opendir(snappath
);
3245 INFO("failed to open %s - assuming no snapshots", snappath
);
3249 while (!readdir_r(dir
, &dirent
, &direntp
)) {
3253 if (!strcmp(direntp
->d_name
, "."))
3256 if (!strcmp(direntp
->d_name
, ".."))
3259 ret
= snprintf(path2
, MAXPATHLEN
, "%s/%s/config", snappath
, direntp
->d_name
);
3260 if (ret
< 0 || ret
>= MAXPATHLEN
) {
3261 ERROR("pathname too long");
3264 if (!file_exists(path2
))
3266 nsnaps
= realloc(snaps
, (count
+ 1)*sizeof(*snaps
));
3268 SYSERROR("Out of memory");
3272 snaps
[count
].free
= lxcsnap_free
;
3273 snaps
[count
].name
= strdup(direntp
->d_name
);
3274 if (!snaps
[count
].name
)
3276 snaps
[count
].lxcpath
= strdup(snappath
);
3277 if (!snaps
[count
].lxcpath
) {
3278 free(snaps
[count
].name
);
3281 snaps
[count
].comment_pathname
= get_snapcomment_path(snappath
, direntp
->d_name
);
3282 snaps
[count
].timestamp
= get_timestamp(snappath
, direntp
->d_name
);
3287 WARN("failed to close directory");
3295 for (i
=0; i
<count
; i
++)
3296 lxcsnap_free(&snaps
[i
]);
3300 WARN("failed to close directory");
3304 WRAP_API_1(int, lxcapi_snapshot_list
, struct lxc_snapshot
**)
3306 static bool do_lxcapi_snapshot_restore(struct lxc_container
*c
, const char *snapname
, const char *newname
)
3308 char clonelxcpath
[MAXPATHLEN
];
3310 struct lxc_container
*snap
, *rest
;
3314 if (!c
|| !c
->name
|| !c
->config_path
)
3317 if (has_fs_snapshots(c
)) {
3318 ERROR("container rootfs has dependent snapshots");
3322 bdev
= bdev_init(c
->lxc_conf
, c
->lxc_conf
->rootfs
.path
, c
->lxc_conf
->rootfs
.mount
, NULL
);
3324 ERROR("Failed to find original backing store type");
3331 if (!get_snappath_dir(c
, clonelxcpath
)) {
3335 // how should we lock this?
3337 snap
= lxc_container_new(snapname
, clonelxcpath
);
3338 if (!snap
|| !lxcapi_is_defined(snap
)) {
3339 ERROR("Could not open snapshot %s", snapname
);
3340 if (snap
) lxc_container_put(snap
);
3345 if (strcmp(c
->name
, newname
) == 0) {
3346 if (!container_destroy(c
)) {
3347 ERROR("Could not destroy existing container %s", newname
);
3348 lxc_container_put(snap
);
3354 if (strcmp(bdev
->type
, "dir") != 0 && strcmp(bdev
->type
, "loop") != 0)
3355 flags
= LXC_CLONE_SNAPSHOT
| LXC_CLONE_MAYBE_SNAPSHOT
;
3356 rest
= lxcapi_clone(snap
, newname
, c
->config_path
, flags
,
3357 bdev
->type
, NULL
, 0, NULL
);
3359 if (rest
&& lxcapi_is_defined(rest
))
3362 lxc_container_put(rest
);
3363 lxc_container_put(snap
);
3367 WRAP_API_2(bool, lxcapi_snapshot_restore
, const char *, const char *)
3369 static bool do_snapshot_destroy(const char *snapname
, const char *clonelxcpath
)
3371 struct lxc_container
*snap
= NULL
;
3374 snap
= lxc_container_new(snapname
, clonelxcpath
);
3376 ERROR("Could not find snapshot %s", snapname
);
3380 if (!do_lxcapi_destroy(snap
)) {
3381 ERROR("Could not destroy snapshot %s", snapname
);
3388 lxc_container_put(snap
);
3392 static bool remove_all_snapshots(const char *path
)
3395 struct dirent dirent
, *direntp
;
3398 dir
= opendir(path
);
3400 SYSERROR("opendir on snapshot path %s", path
);
3403 while (!readdir_r(dir
, &dirent
, &direntp
)) {
3406 if (!strcmp(direntp
->d_name
, "."))
3408 if (!strcmp(direntp
->d_name
, ".."))
3410 if (!do_snapshot_destroy(direntp
->d_name
, path
)) {
3419 SYSERROR("Error removing directory %s", path
);
3424 static bool do_lxcapi_snapshot_destroy(struct lxc_container
*c
, const char *snapname
)
3426 char clonelxcpath
[MAXPATHLEN
];
3428 if (!c
|| !c
->name
|| !c
->config_path
|| !snapname
)
3431 if (!get_snappath_dir(c
, clonelxcpath
))
3434 return do_snapshot_destroy(snapname
, clonelxcpath
);
3437 WRAP_API_1(bool, lxcapi_snapshot_destroy
, const char *)
3439 static bool do_lxcapi_snapshot_destroy_all(struct lxc_container
*c
)
3441 char clonelxcpath
[MAXPATHLEN
];
3443 if (!c
|| !c
->name
|| !c
->config_path
)
3446 if (!get_snappath_dir(c
, clonelxcpath
))
3449 return remove_all_snapshots(clonelxcpath
);
3452 WRAP_API(bool, lxcapi_snapshot_destroy_all
)
3454 static bool do_lxcapi_may_control(struct lxc_container
*c
)
3456 return lxc_try_cmd(c
->name
, c
->config_path
) == 0;
3459 WRAP_API(bool, lxcapi_may_control
)
3461 static bool do_add_remove_node(pid_t init_pid
, const char *path
, bool add
,
3464 char chrootpath
[MAXPATHLEN
];
3465 char *directory_path
= NULL
;
3469 if ((pid
= fork()) < 0) {
3470 SYSERROR("failed to fork a child helper");
3474 if (wait_for_pid(pid
) != 0) {
3475 ERROR("Failed to create note in guest");
3481 /* prepare the path */
3482 ret
= snprintf(chrootpath
, MAXPATHLEN
, "/proc/%d/root", init_pid
);
3483 if (ret
< 0 || ret
>= MAXPATHLEN
)
3486 if (chroot(chrootpath
) < 0)
3490 /* remove path if it exists */
3491 if(faccessat(AT_FDCWD
, path
, F_OK
, AT_SYMLINK_NOFOLLOW
) == 0) {
3492 if (unlink(path
) < 0) {
3493 ERROR("unlink failed");
3500 /* create any missing directories */
3501 directory_path
= dirname(strdup(path
));
3502 if (mkdir_p(directory_path
, 0755) < 0 && errno
!= EEXIST
) {
3503 ERROR("failed to create directory");
3507 /* create the device node */
3508 if (mknod(path
, st
->st_mode
, st
->st_rdev
) < 0) {
3509 ERROR("mknod failed");
3516 static bool add_remove_device_node(struct lxc_container
*c
, const char *src_path
, const char *dest_path
, bool add
)
3520 char value
[MAX_BUFFER
];
3523 /* make sure container is running */
3524 if (!do_lxcapi_is_running(c
)) {
3525 ERROR("container is not running");
3529 /* use src_path if dest_path is NULL otherwise use dest_path */
3530 p
= dest_path
? dest_path
: src_path
;
3532 /* make sure we can access p */
3533 if(access(p
, F_OK
) < 0 || stat(p
, &st
) < 0)
3536 /* continue if path is character device or block device */
3537 if (S_ISCHR(st
.st_mode
))
3538 ret
= snprintf(value
, MAX_BUFFER
, "c %d:%d rwm", major(st
.st_rdev
), minor(st
.st_rdev
));
3539 else if (S_ISBLK(st
.st_mode
))
3540 ret
= snprintf(value
, MAX_BUFFER
, "b %d:%d rwm", major(st
.st_rdev
), minor(st
.st_rdev
));
3544 /* check snprintf return code */
3545 if (ret
< 0 || ret
>= MAX_BUFFER
)
3548 if (!do_add_remove_node(do_lxcapi_init_pid(c
), p
, add
, &st
))
3551 /* add or remove device to/from cgroup access list */
3553 if (!do_lxcapi_set_cgroup_item(c
, "devices.allow", value
)) {
3554 ERROR("set_cgroup_item failed while adding the device node");
3558 if (!do_lxcapi_set_cgroup_item(c
, "devices.deny", value
)) {
3559 ERROR("set_cgroup_item failed while removing the device node");
3567 static bool do_lxcapi_add_device_node(struct lxc_container
*c
, const char *src_path
, const char *dest_path
)
3570 ERROR(NOT_SUPPORTED_ERROR
, __FUNCTION__
);
3573 return add_remove_device_node(c
, src_path
, dest_path
, true);
3576 WRAP_API_2(bool, lxcapi_add_device_node
, const char *, const char *)
3578 static bool do_lxcapi_remove_device_node(struct lxc_container
*c
, const char *src_path
, const char *dest_path
)
3581 ERROR(NOT_SUPPORTED_ERROR
, __FUNCTION__
);
3584 return add_remove_device_node(c
, src_path
, dest_path
, false);
3587 WRAP_API_2(bool, lxcapi_remove_device_node
, const char *, const char *)
3589 static bool do_lxcapi_attach_interface(struct lxc_container
*c
, const char *ifname
,
3590 const char *dst_ifname
)
3594 ERROR(NOT_SUPPORTED_ERROR
, __FUNCTION__
);
3599 ERROR("No source interface name given");
3603 ret
= lxc_netdev_isup(ifname
);
3606 /* netdev of ifname is up. */
3607 ret
= lxc_netdev_down(ifname
);
3612 ret
= lxc_netdev_move_by_name(ifname
, do_lxcapi_init_pid(c
), dst_ifname
);
3622 WRAP_API_2(bool, lxcapi_attach_interface
, const char *, const char *)
3624 static bool do_lxcapi_detach_interface(struct lxc_container
*c
, const char *ifname
,
3625 const char *dst_ifname
)
3627 pid_t pid
, pid_outside
;
3630 ERROR(NOT_SUPPORTED_ERROR
, __FUNCTION__
);
3635 ERROR("No source interface name given");
3639 pid_outside
= getpid();
3642 ERROR("failed to fork task to get interfaces information");
3646 if (pid
== 0) { // child
3648 if (!enter_net_ns(c
)) {
3649 ERROR("failed to enter namespace");
3653 ret
= lxc_netdev_isup(ifname
);
3657 /* netdev of ifname is up. */
3659 ret
= lxc_netdev_down(ifname
);
3664 ret
= lxc_netdev_move_by_name(ifname
, pid_outside
, dst_ifname
);
3666 /* -EINVAL means there is no netdev named as ifanme. */
3667 if (ret
== -EINVAL
) {
3668 ERROR("No network device named as %s.", ifname
);
3673 if (wait_for_pid(pid
) != 0)
3679 WRAP_API_2(bool, lxcapi_detach_interface
, const char *, const char *)
3682 /* The type of criu invocation, one of "dump" or "restore" */
3685 /* The directory to pass to criu */
3688 /* The container to dump */
3689 struct lxc_container
*c
;
3691 /* Enable criu verbose mode? */
3694 /* dump: stop the container or not after dumping? */
3697 /* restore: the file to write the init process' pid into */
3699 const char *cgroup_path
;
3702 static void exec_criu(struct criu_opts
*opts
)
3704 char **argv
, log
[PATH_MAX
], buf
[257];
3705 int static_args
= 14, argc
= 0, i
, ret
;
3707 struct lxc_list
*it
;
3709 /* The command line always looks like:
3710 * criu $(action) --tcp-established --file-locks --link-remap --force-irmap \
3711 * --manage-cgroups action-script foo.sh -D $(directory) \
3712 * -o $(directory)/$(action).log
3713 * +1 for final NULL */
3715 if (strcmp(opts
->action
, "dump") == 0) {
3719 /* --leave-running */
3722 } else if (strcmp(opts
->action
, "restore") == 0) {
3723 /* --root $(lxc_mount_point) --restore-detached
3724 * --restore-sibling --pidfile $foo --cgroup-root $foo */
3733 ret
= snprintf(log
, PATH_MAX
, "%s/%s.log", opts
->directory
, opts
->action
);
3734 if (ret
< 0 || ret
>= PATH_MAX
) {
3735 ERROR("logfile name too long\n");
3739 // We need to tell criu where cgmanager's socket is bind mounted from
3740 // if it exists since it's external.
3741 if (cgroup_driver() == CGMANAGER
)
3744 argv
= malloc(static_args
* sizeof(*argv
));
3748 memset(argv
, 0, static_args
* sizeof(*argv
));
3750 #define DECLARE_ARG(arg) \
3752 if (arg == NULL) { \
3753 ERROR("Got NULL argument for criu"); \
3756 argv[argc++] = strdup(arg); \
3757 if (!argv[argc-1]) \
3761 argv
[argc
++] = on_path("criu", NULL
);
3762 if (!argv
[argc
-1]) {
3763 ERROR("Couldn't find criu binary\n");
3767 DECLARE_ARG(opts
->action
);
3768 DECLARE_ARG("--tcp-established");
3769 DECLARE_ARG("--file-locks");
3770 DECLARE_ARG("--link-remap");
3771 DECLARE_ARG("--force-irmap");
3772 DECLARE_ARG("--manage-cgroups");
3773 DECLARE_ARG("--action-script");
3774 DECLARE_ARG(DATADIR
"/lxc/lxc-restore-net");
3776 DECLARE_ARG(opts
->directory
);
3781 DECLARE_ARG("-vvvvvv");
3783 if (strcmp(opts
->action
, "dump") == 0) {
3786 if (cgroup_driver() == CGMANAGER
) {
3787 DECLARE_ARG("--ext-mount-map");
3788 DECLARE_ARG("/sys/fs/cgroup/cgmanager:cgmanager");
3791 if (sprintf(pid
, "%d", lxcapi_init_pid(opts
->c
)) < 0)
3797 DECLARE_ARG("--leave-running");
3798 } else if (strcmp(opts
->action
, "restore") == 0) {
3800 if (cgroup_driver() == CGMANAGER
) {
3801 DECLARE_ARG("--ext-mount-map");
3802 DECLARE_ARG("cgmanager:/sys/fs/cgroup/cgmanager");
3805 DECLARE_ARG("--root");
3806 DECLARE_ARG(opts
->c
->lxc_conf
->rootfs
.mount
);
3807 DECLARE_ARG("--restore-detached");
3808 DECLARE_ARG("--restore-sibling");
3809 DECLARE_ARG("--pidfile");
3810 DECLARE_ARG(opts
->pidfile
);
3811 DECLARE_ARG("--cgroup-root");
3812 DECLARE_ARG(opts
->cgroup_path
);
3814 lxc_list_for_each(it
, &opts
->c
->lxc_conf
->network
) {
3815 char eth
[128], *veth
;
3817 struct lxc_netdev
*n
= it
->elem
;
3820 if (strlen(n
->name
) >= sizeof(eth
))
3822 strncpy(eth
, n
->name
, sizeof(eth
));
3824 sprintf(eth
, "eth%d", netnr
);
3826 veth
= n
->priv
.veth_attr
.pair
;
3828 ret
= snprintf(buf
, sizeof(buf
), "%s=%s", eth
, veth
);
3829 if (ret
< 0 || ret
>= sizeof(buf
))
3832 /* final NULL and --veth-pair eth0=vethASDF */
3833 m
= realloc(argv
, (argc
+ 1 + 2) * sizeof(*argv
));
3838 DECLARE_ARG("--veth-pair");
3846 lxc_list_for_each(it
, &opts
->c
->lxc_conf
->network
) {
3847 struct lxc_netdev
*n
= it
->elem
;
3851 * Here, we set some parameters that lxc-restore-net
3852 * will examine to figure out the right network to
3855 snprintf(buf
, sizeof(buf
), "LXC_CRIU_BRIDGE%d", netnr
);
3856 if (setenv(buf
, n
->link
, 1))
3859 if (strcmp("restore", opts
->action
) == 0)
3860 strncpy(veth
, n
->priv
.veth_attr
.pair
, sizeof(veth
));
3863 ret
= snprintf(buf
, sizeof(buf
), "lxc.network.%d.veth.pair", netnr
);
3864 if (ret
< 0 || ret
>= sizeof(buf
))
3866 tmp
= lxcapi_get_running_config_item(opts
->c
, buf
);
3867 strncpy(veth
, tmp
, sizeof(veth
));
3871 snprintf(buf
, sizeof(buf
), "LXC_CRIU_VETH%d", netnr
);
3872 if (setenv(buf
, veth
, 1))
3879 execv(argv
[0], argv
);
3881 for (i
= 0; argv
[i
]; i
++)
3886 /* Check and make sure the container has a configuration that we know CRIU can
3888 static bool criu_ok(struct lxc_container
*c
)
3890 struct lxc_list
*it
;
3891 bool found_deny_rule
= false;
3894 ERROR("Must be root to checkpoint\n");
3898 /* We only know how to restore containers with veth networks. */
3899 lxc_list_for_each(it
, &c
->lxc_conf
->network
) {
3900 struct lxc_netdev
*n
= it
->elem
;
3901 if (n
->type
!= LXC_NET_VETH
&& n
->type
!= LXC_NET_NONE
) {
3902 ERROR("Found network that is not VETH or NONE\n");
3907 // These requirements come from http://criu.org/LXC
3908 if (c
->lxc_conf
->console
.path
&&
3909 strcmp(c
->lxc_conf
->console
.path
, "none") != 0) {
3910 ERROR("lxc.console must be none\n");
3914 if (c
->lxc_conf
->tty
!= 0) {
3915 ERROR("lxc.tty must be 0\n");
3919 lxc_list_for_each(it
, &c
->lxc_conf
->cgroup
) {
3920 struct lxc_cgroup
*cg
= it
->elem
;
3921 if (strcmp(cg
->subsystem
, "devices.deny") == 0 &&
3922 strcmp(cg
->value
, "c 5:1 rwm") == 0) {
3924 found_deny_rule
= true;
3929 if (!found_deny_rule
) {
3930 ERROR("couldn't find devices.deny = c 5:1 rwm");
3937 static bool dump_net_info(struct lxc_container
*c
, char *directory
)
3940 struct lxc_list
*it
;
3943 lxc_list_for_each(it
, &c
->lxc_conf
->network
) {
3944 char *veth
= NULL
, *bridge
= NULL
, veth_path
[PATH_MAX
], eth
[128];
3945 struct lxc_netdev
*n
= it
->elem
;
3946 bool has_error
= true;
3949 pret
= snprintf(veth_path
, PATH_MAX
, "lxc.network.%d.veth.pair", netnr
);
3950 if (pret
< 0 || pret
>= PATH_MAX
)
3953 veth
= lxcapi_get_running_config_item(c
, veth_path
);
3955 /* criu_ok() checks that all interfaces are
3956 * LXC_NET{VETH,NONE}, and VETHs should have this
3958 assert(n
->type
== LXC_NET_NONE
);
3962 bridge
= lxcapi_get_running_config_item(c
, veth_path
);
3966 pret
= snprintf(veth_path
, PATH_MAX
, "%s/veth%d", directory
, netnr
);
3967 if (pret
< 0 || pret
>= PATH_MAX
|| print_to_file(veth_path
, veth
) < 0)
3971 if (strlen(n
->name
) >= 128)
3973 strncpy(eth
, n
->name
, 128);
3975 sprintf(eth
, "eth%d", netnr
);
3988 static bool do_lxcapi_checkpoint(struct lxc_container
*c
, char *directory
, bool stop
, bool verbose
)
3996 if (mkdir(directory
, 0700) < 0 && errno
!= EEXIST
)
3999 if (!dump_net_info(c
, directory
))
4007 struct criu_opts os
;
4010 os
.directory
= directory
;
4013 os
.verbose
= verbose
;
4015 /* exec_criu() returning is an error */
4019 pid_t w
= waitpid(pid
, &status
, 0);
4025 if (WIFEXITED(status
)) {
4026 return !WEXITSTATUS(status
);
4033 WRAP_API_3(bool, lxcapi_checkpoint
, char *, bool, bool)
4035 static bool restore_net_info(struct lxc_container
*c
)
4037 struct lxc_list
*it
;
4038 bool has_error
= true;
4040 if (container_mem_lock(c
))
4043 lxc_list_for_each(it
, &c
->lxc_conf
->network
) {
4044 struct lxc_netdev
*netdev
= it
->elem
;
4045 char template[IFNAMSIZ
];
4046 snprintf(template, sizeof(template), "vethXXXXXX");
4048 if (!netdev
->priv
.veth_attr
.pair
)
4049 netdev
->priv
.veth_attr
.pair
= lxc_mkifname(template);
4051 if (!netdev
->priv
.veth_attr
.pair
)
4058 container_mem_unlock(c
);
4062 // do_restore never returns, the calling process is used as the
4063 // monitor process. do_restore calls exit() if it fails.
4064 static void do_restore(struct lxc_container
*c
, int pipe
, char *directory
, bool verbose
)
4067 char pidfile
[L_tmpnam
];
4068 struct lxc_handler
*handler
;
4071 if (!tmpnam(pidfile
))
4074 handler
= lxc_init(c
->name
, c
->lxc_conf
, c
->config_path
);
4078 if (!cgroup_init(handler
)) {
4079 ERROR("failed initing cgroups");
4080 goto out_fini_handler
;
4083 if (!cgroup_create(handler
)) {
4084 ERROR("failed creating groups");
4085 goto out_fini_handler
;
4088 if (!restore_net_info(c
)) {
4089 ERROR("failed restoring network info");
4090 goto out_fini_handler
;
4095 goto out_fini_handler
;
4098 struct criu_opts os
;
4099 struct lxc_rootfs
*rootfs
;
4104 if (unshare(CLONE_NEWNS
))
4105 goto out_fini_handler
;
4107 /* CRIU needs the lxc root bind mounted so that it is the root of some
4109 rootfs
= &c
->lxc_conf
->rootfs
;
4111 if (rootfs_is_blockdev(c
->lxc_conf
)) {
4112 if (do_rootfs_setup(c
->lxc_conf
, c
->name
, c
->config_path
) < 0)
4113 goto out_fini_handler
;
4115 if (mkdir(rootfs
->mount
, 0755) < 0 && errno
!= EEXIST
)
4116 goto out_fini_handler
;
4118 if (mount(rootfs
->path
, rootfs
->mount
, NULL
, MS_BIND
, NULL
) < 0) {
4119 rmdir(rootfs
->mount
);
4120 goto out_fini_handler
;
4124 os
.action
= "restore";
4125 os
.directory
= directory
;
4127 os
.pidfile
= pidfile
;
4128 os
.verbose
= verbose
;
4129 os
.cgroup_path
= cgroup_canonical_path(handler
);
4131 /* exec_criu() returning is an error */
4133 umount(rootfs
->mount
);
4134 rmdir(rootfs
->mount
);
4135 goto out_fini_handler
;
4140 pid_t w
= waitpid(pid
, &status
, 0);
4143 goto out_fini_handler
;
4146 ret
= write(pipe
, &status
, sizeof(status
));
4150 if (sizeof(status
) != ret
) {
4152 ERROR("failed to write all of status");
4153 goto out_fini_handler
;
4156 if (WIFEXITED(status
)) {
4157 if (WEXITSTATUS(status
)) {
4158 goto out_fini_handler
;
4161 FILE *f
= fopen(pidfile
, "r");
4163 perror("reading pidfile");
4164 ERROR("couldn't read restore's init pidfile %s\n", pidfile
);
4165 goto out_fini_handler
;
4168 ret
= fscanf(f
, "%d", (int*) &handler
->pid
);
4171 ERROR("reading restore pid failed");
4172 goto out_fini_handler
;
4175 if (lxc_set_state(c
->name
, handler
, RUNNING
))
4176 goto out_fini_handler
;
4179 ERROR("CRIU was killed with signal %d\n", WTERMSIG(status
));
4180 goto out_fini_handler
;
4184 * See comment in lxcapi_start; we don't care if these
4185 * fail because it's just a beauty thing. We just
4186 * assign the return here to silence potential.
4188 ret
= snprintf(title
, sizeof(title
), "[lxc monitor] %s %s", c
->config_path
, c
->name
);
4189 ret
= setproctitle(title
);
4191 ret
= lxc_poll(c
->name
, handler
);
4193 lxc_abort(c
->name
, handler
);
4194 lxc_fini(c
->name
, handler
);
4199 lxc_fini(c
->name
, handler
);
4204 if (write(pipe
, &status
, sizeof(status
)) != sizeof(status
)) {
4205 SYSERROR("writing status failed");
4213 static bool do_lxcapi_restore(struct lxc_container
*c
, char *directory
, bool verbose
)
4223 ERROR("Must be root to restore\n");
4228 ERROR("failed to create pipe");
4241 // this never returns
4242 do_restore(c
, pipefd
[1], directory
, verbose
);
4247 nread
= read(pipefd
[0], &status
, sizeof(status
));
4249 if (sizeof(status
) != nread
) {
4250 ERROR("reading status from pipe failed");
4254 // If the criu process was killed or exited nonzero, wait() for the
4255 // handler, since the restore process died. Otherwise, we don't need to
4256 // wait, since the child becomes the monitor process.
4257 if (!WIFEXITED(status
) || WEXITSTATUS(status
))
4262 if (wait_for_pid(pid
))
4263 ERROR("restore process died");
4267 WRAP_API_2(bool, lxcapi_restore
, char *, bool)
4269 static int lxcapi_attach_run_waitl(struct lxc_container
*c
, lxc_attach_options_t
*options
, const char *program
, const char *arg
, ...)
4278 struct lxc_conf
*old
= current_config
;
4279 current_config
= c
->lxc_conf
;
4282 argv
= lxc_va_arg_list_to_argv_const(ap
, 1);
4286 ERROR("Memory allocation error.");
4292 ret
= do_lxcapi_attach_run_wait(c
, options
, program
, (const char * const *)argv
);
4295 current_config
= old
;
4299 struct lxc_container
*lxc_container_new(const char *name
, const char *configpath
)
4301 struct lxc_container
*c
;
4306 c
= malloc(sizeof(*c
));
4308 fprintf(stderr
, "failed to malloc lxc_container\n");
4311 memset(c
, 0, sizeof(*c
));
4314 c
->config_path
= strdup(configpath
);
4316 c
->config_path
= strdup(lxc_global_config_value("lxc.lxcpath"));
4318 if (!c
->config_path
) {
4319 fprintf(stderr
, "Out of memory\n");
4323 remove_trailing_slashes(c
->config_path
);
4324 c
->name
= malloc(strlen(name
)+1);
4326 fprintf(stderr
, "Error allocating lxc_container name\n");
4329 strcpy(c
->name
, name
);
4332 if (!(c
->slock
= lxc_newlock(c
->config_path
, name
))) {
4333 fprintf(stderr
, "failed to create lock\n");
4337 if (!(c
->privlock
= lxc_newlock(NULL
, NULL
))) {
4338 fprintf(stderr
, "failed to alloc privlock\n");
4342 if (!set_config_filename(c
)) {
4343 fprintf(stderr
, "Error allocating config file pathname\n");
4347 if (file_exists(c
->configfile
) && !lxcapi_load_config(c
, NULL
))
4350 if (ongoing_create(c
) == 2) {
4351 ERROR("Error: %s creation was not completed", c
->name
);
4352 container_destroy(c
);
4353 lxcapi_clear_config(c
);
4355 c
->daemonize
= true;
4358 // assign the member functions
4359 c
->is_defined
= lxcapi_is_defined
;
4360 c
->state
= lxcapi_state
;
4361 c
->is_running
= lxcapi_is_running
;
4362 c
->freeze
= lxcapi_freeze
;
4363 c
->unfreeze
= lxcapi_unfreeze
;
4364 c
->console
= lxcapi_console
;
4365 c
->console_getfd
= lxcapi_console_getfd
;
4366 c
->init_pid
= lxcapi_init_pid
;
4367 c
->load_config
= lxcapi_load_config
;
4368 c
->want_daemonize
= lxcapi_want_daemonize
;
4369 c
->want_close_all_fds
= lxcapi_want_close_all_fds
;
4370 c
->start
= lxcapi_start
;
4371 c
->startl
= lxcapi_startl
;
4372 c
->stop
= lxcapi_stop
;
4373 c
->config_file_name
= lxcapi_config_file_name
;
4374 c
->wait
= lxcapi_wait
;
4375 c
->set_config_item
= lxcapi_set_config_item
;
4376 c
->destroy
= lxcapi_destroy
;
4377 c
->destroy_with_snapshots
= lxcapi_destroy_with_snapshots
;
4378 c
->rename
= lxcapi_rename
;
4379 c
->save_config
= lxcapi_save_config
;
4380 c
->get_keys
= lxcapi_get_keys
;
4381 c
->create
= lxcapi_create
;
4382 c
->createl
= lxcapi_createl
;
4383 c
->shutdown
= lxcapi_shutdown
;
4384 c
->reboot
= lxcapi_reboot
;
4385 c
->clear_config
= lxcapi_clear_config
;
4386 c
->clear_config_item
= lxcapi_clear_config_item
;
4387 c
->get_config_item
= lxcapi_get_config_item
;
4388 c
->get_running_config_item
= lxcapi_get_running_config_item
;
4389 c
->get_cgroup_item
= lxcapi_get_cgroup_item
;
4390 c
->set_cgroup_item
= lxcapi_set_cgroup_item
;
4391 c
->get_config_path
= lxcapi_get_config_path
;
4392 c
->set_config_path
= lxcapi_set_config_path
;
4393 c
->clone
= lxcapi_clone
;
4394 c
->get_interfaces
= lxcapi_get_interfaces
;
4395 c
->get_ips
= lxcapi_get_ips
;
4396 c
->attach
= lxcapi_attach
;
4397 c
->attach_run_wait
= lxcapi_attach_run_wait
;
4398 c
->attach_run_waitl
= lxcapi_attach_run_waitl
;
4399 c
->snapshot
= lxcapi_snapshot
;
4400 c
->snapshot_list
= lxcapi_snapshot_list
;
4401 c
->snapshot_restore
= lxcapi_snapshot_restore
;
4402 c
->snapshot_destroy
= lxcapi_snapshot_destroy
;
4403 c
->snapshot_destroy_all
= lxcapi_snapshot_destroy_all
;
4404 c
->may_control
= lxcapi_may_control
;
4405 c
->add_device_node
= lxcapi_add_device_node
;
4406 c
->remove_device_node
= lxcapi_remove_device_node
;
4407 c
->attach_interface
= lxcapi_attach_interface
;
4408 c
->detach_interface
= lxcapi_detach_interface
;
4409 c
->checkpoint
= lxcapi_checkpoint
;
4410 c
->restore
= lxcapi_restore
;
4415 lxc_container_free(c
);
4419 int lxc_get_wait_states(const char **states
)
4424 for (i
=0; i
<MAX_STATE
; i
++)
4425 states
[i
] = lxc_state2str(i
);
4430 * These next two could probably be done smarter with reusing a common function
4431 * with different iterators and tests...
4433 int list_defined_containers(const char *lxcpath
, char ***names
, struct lxc_container
***cret
)
4436 int i
, cfound
= 0, nfound
= 0;
4437 struct dirent dirent
, *direntp
;
4438 struct lxc_container
*c
;
4441 lxcpath
= lxc_global_config_value("lxc.lxcpath");
4443 dir
= opendir(lxcpath
);
4445 SYSERROR("opendir on lxcpath");
4454 while (!readdir_r(dir
, &dirent
, &direntp
)) {
4457 if (!strcmp(direntp
->d_name
, "."))
4459 if (!strcmp(direntp
->d_name
, ".."))
4462 if (!config_file_exists(lxcpath
, direntp
->d_name
))
4466 if (!add_to_array(names
, direntp
->d_name
, cfound
))
4476 c
= lxc_container_new(direntp
->d_name
, lxcpath
);
4478 INFO("Container %s:%s has a config but could not be loaded",
4479 lxcpath
, direntp
->d_name
);
4481 if(!remove_from_array(names
, direntp
->d_name
, cfound
--))
4485 if (!do_lxcapi_is_defined(c
)) {
4486 INFO("Container %s:%s has a config but is not defined",
4487 lxcpath
, direntp
->d_name
);
4489 if(!remove_from_array(names
, direntp
->d_name
, cfound
--))
4491 lxc_container_put(c
);
4495 if (!add_to_clist(cret
, c
, nfound
, true)) {
4496 lxc_container_put(c
);
4506 if (names
&& *names
) {
4507 for (i
=0; i
<cfound
; i
++)
4511 if (cret
&& *cret
) {
4512 for (i
=0; i
<nfound
; i
++)
4513 lxc_container_put((*cret
)[i
]);
4520 int list_active_containers(const char *lxcpath
, char ***nret
,
4521 struct lxc_container
***cret
)
4523 int i
, ret
= -1, cret_cnt
= 0, ct_name_cnt
= 0;
4526 char **ct_name
= NULL
;
4528 struct lxc_container
*c
;
4532 lxcpath
= lxc_global_config_value("lxc.lxcpath");
4533 lxcpath_len
= strlen(lxcpath
);
4540 FILE *f
= fopen("/proc/net/unix", "r");
4544 while (getline(&line
, &len
, f
) != -1) {
4546 char *p
= strrchr(line
, ' '), *p2
;
4555 if (strncmp(p
, lxcpath
, lxcpath_len
) == 0) {
4557 } else if (strncmp(p
, "lxc/", 4) == 0) {
4567 // Now p is the start of lxc_name
4568 p2
= strchr(p
, '/');
4569 if (!p2
|| strncmp(p2
, "/command", 8) != 0)
4574 if (strncmp(lxcpath
, lxc_cmd_get_lxcpath(p
), lxcpath_len
) != 0)
4576 p
= lxc_cmd_get_name(p
);
4579 if (array_contains(&ct_name
, p
, ct_name_cnt
))
4582 if (!add_to_array(&ct_name
, p
, ct_name_cnt
))
4583 goto free_cret_list
;
4590 c
= lxc_container_new(p
, lxcpath
);
4592 INFO("Container %s:%s is running but could not be loaded",
4594 remove_from_array(&ct_name
, p
, ct_name_cnt
--);
4599 * If this is an anonymous container, then is_defined *can*
4600 * return false. So we don't do that check. Count on the
4601 * fact that the command socket exists.
4604 if (!add_to_clist(cret
, c
, cret_cnt
, true)) {
4605 lxc_container_put(c
);
4606 goto free_cret_list
;
4611 assert(!nret
|| !cret
|| cret_cnt
== ct_name_cnt
);
4620 if (cret
&& *cret
) {
4621 for (i
= 0; i
< cret_cnt
; i
++)
4622 lxc_container_put((*cret
)[i
]);
4628 for (i
= 0; i
< ct_name_cnt
; i
++)
4640 int list_all_containers(const char *lxcpath
, char ***nret
,
4641 struct lxc_container
***cret
)
4643 int i
, ret
, active_cnt
, ct_cnt
, ct_list_cnt
;
4646 struct lxc_container
**ct_list
= NULL
;
4648 ct_cnt
= list_defined_containers(lxcpath
, &ct_name
, NULL
);
4652 active_cnt
= list_active_containers(lxcpath
, &active_name
, NULL
);
4653 if (active_cnt
< 0) {
4658 for (i
= 0; i
< active_cnt
; i
++) {
4659 if (!array_contains(&ct_name
, active_name
[i
], ct_cnt
)) {
4660 if (!add_to_array(&ct_name
, active_name
[i
], ct_cnt
)) {
4662 goto free_active_name
;
4666 free(active_name
[i
]);
4667 active_name
[i
] = NULL
;
4673 for (i
= 0, ct_list_cnt
= 0; i
< ct_cnt
&& cret
; i
++) {
4674 struct lxc_container
*c
;
4676 c
= lxc_container_new(ct_name
[i
], lxcpath
);
4678 WARN("Container %s:%s could not be loaded", lxcpath
, ct_name
[i
]);
4679 remove_from_array(&ct_name
, ct_name
[i
], ct_cnt
--);
4683 if (!add_to_clist(&ct_list
, c
, ct_list_cnt
, false)) {
4684 lxc_container_put(c
);
4703 for (i
= 0; i
< ct_list_cnt
; i
++) {
4704 lxc_container_put(ct_list
[i
]);
4709 for (i
= 0; i
< active_cnt
; i
++) {
4710 free(active_name
[i
]);
4715 for (i
= 0; i
< ct_cnt
; i
++) {