2 * lxc: linux Container library
4 * (C) Copyright IBM Corp. 2007, 2008
7 * Daniel Lezcano <daniel.lezcano at free.fr>
9 * This library is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; either
12 * version 2.1 of the License, or (at your option) any later version.
14 * This library is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with this library; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
36 #include <sys/param.h>
38 #include <sys/mount.h>
40 #include <sys/types.h>
41 #include <sys/socket.h>
42 #include <sys/prctl.h>
43 #include <sys/types.h>
46 #include <sys/syscall.h>
48 #if HAVE_SYS_CAPABILITY_H
49 #include <sys/capability.h>
52 #if !HAVE_DECL_PR_CAPBSET_DROP
53 #define PR_CAPBSET_DROP 24
69 #include "namespace.h"
70 #include "lxcseccomp.h"
75 lxc_log_define(lxc_start
, lxc
);
77 const struct ns_info ns_info
[LXC_NS_MAX
] = {
78 [LXC_NS_MNT
] = {"mnt", CLONE_NEWNS
},
79 [LXC_NS_PID
] = {"pid", CLONE_NEWPID
},
80 [LXC_NS_UTS
] = {"uts", CLONE_NEWUTS
},
81 [LXC_NS_IPC
] = {"ipc", CLONE_NEWIPC
},
82 [LXC_NS_USER
] = {"user", CLONE_NEWUSER
},
83 [LXC_NS_NET
] = {"net", CLONE_NEWNET
}
86 static void print_top_failing_dir(const char *path
)
88 size_t len
= strlen(path
);
89 char *copy
= alloca(len
+1), *p
, *e
, saved
;
95 while (p
< e
&& *p
== '/') p
++;
96 while (p
< e
&& *p
!= '/') p
++;
99 if (access(copy
, X_OK
)) {
100 SYSERROR("could not access %s. Please grant it 'x' " \
101 "access, or add an ACL for the container root.",
109 static void close_ns(int ns_fd
[LXC_NS_MAX
]) {
112 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
120 static int preserve_ns(int ns_fd
[LXC_NS_MAX
], int clone_flags
) {
122 char path
[MAXPATHLEN
];
124 for (i
= 0; i
< LXC_NS_MAX
; i
++)
127 if (access("/proc/self/ns", X_OK
)) {
128 WARN("Kernel does not support attach; preserve_ns ignored");
132 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
133 if ((clone_flags
& ns_info
[i
].clone_flag
) == 0)
135 snprintf(path
, MAXPATHLEN
, "/proc/self/ns/%s", ns_info
[i
].proc_name
);
136 ns_fd
[i
] = open(path
, O_RDONLY
| O_CLOEXEC
);
147 SYSERROR("failed to open '%s'", path
);
151 static int attach_ns(const int ns_fd
[LXC_NS_MAX
]) {
154 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
158 if (setns(ns_fd
[i
], 0) != 0)
164 SYSERROR("failed to set namespace '%s'", ns_info
[i
].proc_name
);
168 static int match_fd(int fd
)
170 return (fd
== 0 || fd
== 1 || fd
== 2);
173 int lxc_check_inherited(struct lxc_conf
*conf
, int fd_to_ignore
)
175 struct dirent dirent
, *direntp
;
180 dir
= opendir("/proc/self/fd");
182 WARN("failed to open directory: %m");
188 while (!readdir_r(dir
, &dirent
, &direntp
)) {
192 if (!strcmp(direntp
->d_name
, "."))
195 if (!strcmp(direntp
->d_name
, ".."))
198 fd
= atoi(direntp
->d_name
);
200 if (fd
== fddir
|| fd
== lxc_log_fd
|| fd
== fd_to_ignore
)
206 if (conf
== NULL
|| conf
->close_all_fds
) {
209 INFO("closed inherited fd %d", fd
);
212 WARN("inherited fd %d", fd
);
215 closedir(dir
); /* cannot fail */
219 static int setup_signal_fd(sigset_t
*oldmask
)
224 /* Block everything except serious error signals */
225 if (sigfillset(&mask
) ||
226 sigdelset(&mask
, SIGILL
) ||
227 sigdelset(&mask
, SIGSEGV
) ||
228 sigdelset(&mask
, SIGBUS
) ||
229 sigdelset(&mask
, SIGWINCH
) ||
230 sigprocmask(SIG_BLOCK
, &mask
, oldmask
)) {
231 SYSERROR("failed to set signal mask");
235 fd
= signalfd(-1, &mask
, 0);
237 SYSERROR("failed to create the signal fd");
241 if (fcntl(fd
, F_SETFD
, FD_CLOEXEC
)) {
242 SYSERROR("failed to set sigfd to close-on-exec");
247 DEBUG("sigchild handler set");
252 static int signal_handler(int fd
, uint32_t events
, void *data
,
253 struct lxc_epoll_descr
*descr
)
255 struct signalfd_siginfo siginfo
;
259 bool init_died
= false;
261 ret
= read(fd
, &siginfo
, sizeof(siginfo
));
263 ERROR("failed to read signal info");
267 if (ret
!= sizeof(siginfo
)) {
268 ERROR("unexpected siginfo size");
272 // check whether init is running
274 ret
= waitid(P_PID
, *pid
, &info
, WEXITED
| WNOWAIT
| WNOHANG
);
275 if (ret
== 0 && info
.si_pid
== *pid
) {
279 if (siginfo
.ssi_signo
!= SIGCHLD
) {
280 kill(*pid
, siginfo
.ssi_signo
);
281 INFO("forwarded signal %d to pid %d", siginfo
.ssi_signo
, *pid
);
282 return init_died
? 1 : 0;
285 if (siginfo
.ssi_code
== CLD_STOPPED
||
286 siginfo
.ssi_code
== CLD_CONTINUED
) {
287 INFO("container init process was stopped/continued");
288 return init_died
? 1 : 0;
291 /* more robustness, protect ourself from a SIGCHLD sent
292 * by a process different from the container init
294 if (siginfo
.ssi_pid
!= *pid
) {
295 WARN("invalid pid for SIGCHLD");
296 return init_died
? 1 : 0;
299 DEBUG("container init process exited");
303 int lxc_set_state(const char *name
, struct lxc_handler
*handler
, lxc_state_t state
)
305 handler
->state
= state
;
306 lxc_monitor_send_state(name
, state
, handler
->lxcpath
);
310 int lxc_poll(const char *name
, struct lxc_handler
*handler
)
312 int sigfd
= handler
->sigfd
;
313 int pid
= handler
->pid
;
314 struct lxc_epoll_descr descr
;
316 if (lxc_mainloop_open(&descr
)) {
317 ERROR("failed to create mainloop");
321 if (lxc_mainloop_add_handler(&descr
, sigfd
, signal_handler
, &pid
)) {
322 ERROR("failed to add handler for the signal");
323 goto out_mainloop_open
;
326 if (lxc_console_mainloop_add(&descr
, handler
)) {
327 ERROR("failed to add console handler to mainloop");
328 goto out_mainloop_open
;
331 if (lxc_cmd_mainloop_add(name
, &descr
, handler
)) {
332 ERROR("failed to add command handler to mainloop");
333 goto out_mainloop_open
;
336 if (handler
->conf
->need_utmp_watch
) {
337 #if HAVE_SYS_CAPABILITY_H
338 if (lxc_utmp_mainloop_add(&descr
, handler
)) {
339 ERROR("failed to add utmp handler to mainloop");
340 goto out_mainloop_open
;
343 DEBUG("not starting utmp handler as cap_sys_boot cannot be dropped without capabilities support");
347 return lxc_mainloop(&descr
, -1);
350 lxc_mainloop_close(&descr
);
356 struct lxc_handler
*lxc_init(const char *name
, struct lxc_conf
*conf
, const char *lxcpath
)
358 struct lxc_handler
*handler
;
360 handler
= malloc(sizeof(*handler
));
364 memset(handler
, 0, sizeof(*handler
));
366 handler
->conf
= conf
;
367 handler
->lxcpath
= lxcpath
;
372 handler
->name
= strdup(name
);
373 if (!handler
->name
) {
374 ERROR("failed to allocate memory");
378 if (lxc_cmd_init(name
, handler
, lxcpath
))
381 if (lxc_read_seccomp_config(conf
) != 0) {
382 ERROR("failed loading seccomp policy");
383 goto out_close_maincmd_fd
;
386 /* Begin by setting the state to STARTING */
387 if (lxc_set_state(name
, handler
, STARTING
)) {
388 ERROR("failed to set state '%s'", lxc_state2str(STARTING
));
389 goto out_close_maincmd_fd
;
392 /* Start of environment variable setup for hooks */
393 if (setenv("LXC_NAME", name
, 1)) {
394 SYSERROR("failed to set environment variable for container name");
396 if (setenv("LXC_CONFIG_FILE", conf
->rcfile
, 1)) {
397 SYSERROR("failed to set environment variable for config path");
399 if (setenv("LXC_ROOTFS_MOUNT", conf
->rootfs
.mount
, 1)) {
400 SYSERROR("failed to set environment variable for rootfs mount");
402 if (setenv("LXC_ROOTFS_PATH", conf
->rootfs
.path
, 1)) {
403 SYSERROR("failed to set environment variable for rootfs mount");
405 if (conf
->console
.path
&& setenv("LXC_CONSOLE", conf
->console
.path
, 1)) {
406 SYSERROR("failed to set environment variable for console path");
408 if (conf
->console
.log_path
&& setenv("LXC_CONSOLE_LOGPATH", conf
->console
.log_path
, 1)) {
409 SYSERROR("failed to set environment variable for console log");
411 /* End of environment variable setup for hooks */
413 if (run_lxc_hooks(name
, "pre-start", conf
, handler
->lxcpath
, NULL
)) {
414 ERROR("failed to run pre-start hooks for container '%s'.", name
);
418 if (lxc_create_tty(name
, conf
)) {
419 ERROR("failed to create the ttys");
423 /* the signal fd has to be created before forking otherwise
424 * if the child process exits before we setup the signal fd,
425 * the event will be lost and the command will be stuck */
426 handler
->sigfd
= setup_signal_fd(&handler
->oldmask
);
427 if (handler
->sigfd
< 0) {
428 ERROR("failed to set sigchild fd handler");
432 /* do this after setting up signals since it might unblock SIGWINCH */
433 if (lxc_console_create(conf
)) {
434 ERROR("failed to create console");
435 goto out_restore_sigmask
;
438 if (ttys_shift_ids(conf
) < 0) {
439 ERROR("Failed to shift tty into container");
440 goto out_restore_sigmask
;
443 INFO("'%s' is initialized", name
);
447 sigprocmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
449 lxc_delete_tty(&conf
->tty_info
);
451 lxc_set_state(name
, handler
, ABORTING
);
452 out_close_maincmd_fd
:
453 close(conf
->maincmd_fd
);
454 conf
->maincmd_fd
= -1;
457 handler
->name
= NULL
;
463 void lxc_fini(const char *name
, struct lxc_handler
*handler
)
465 /* The STOPPING state is there for future cleanup code
466 * which can take awhile
468 lxc_set_state(name
, handler
, STOPPING
);
469 lxc_set_state(name
, handler
, STOPPED
);
471 if (run_lxc_hooks(name
, "post-stop", handler
->conf
, handler
->lxcpath
, NULL
))
472 ERROR("failed to run post-stop hooks for container '%s'.", name
);
474 /* reset mask set by setup_signal_fd */
475 if (sigprocmask(SIG_SETMASK
, &handler
->oldmask
, NULL
))
476 WARN("failed to restore sigprocmask");
478 lxc_console_delete(&handler
->conf
->console
);
479 lxc_delete_tty(&handler
->conf
->tty_info
);
480 lxc_delete_autodev(handler
);
481 close(handler
->conf
->maincmd_fd
);
482 handler
->conf
->maincmd_fd
= -1;
484 cgroup_destroy(handler
);
488 void lxc_abort(const char *name
, struct lxc_handler
*handler
)
492 lxc_set_state(name
, handler
, ABORTING
);
493 if (handler
->pid
> 0)
494 kill(handler
->pid
, SIGKILL
);
495 while ((ret
= waitpid(-1, &status
, 0)) > 0) ;
498 #include <sys/reboot.h>
499 #include <linux/reboot.h>
502 * reboot(LINUX_REBOOT_CMD_CAD_ON) will return -EINVAL
503 * in a child pid namespace if container reboot support exists.
504 * Otherwise, it will either succeed or return -EPERM.
506 static int container_reboot_supported(void *arg
)
512 if (ret
== -1 && errno
== EINVAL
)
517 static int must_drop_cap_sys_boot(struct lxc_conf
*conf
)
520 int ret
, cmd
, v
, flags
;
521 long stack_size
= 4096;
522 void *stack
= alloca(stack_size
);
526 f
= fopen("/proc/sys/kernel/ctrl-alt-del", "r");
528 DEBUG("failed to open /proc/sys/kernel/ctrl-alt-del");
532 ret
= fscanf(f
, "%d", &v
);
535 DEBUG("Failed to read /proc/sys/kernel/ctrl-alt-del");
538 cmd
= v
? LINUX_REBOOT_CMD_CAD_ON
: LINUX_REBOOT_CMD_CAD_OFF
;
540 flags
= CLONE_NEWPID
| SIGCHLD
;
541 if (!lxc_list_empty(&conf
->id_map
))
542 flags
|= CLONE_NEWUSER
;
545 pid
= __clone2(container_reboot_supported
, stack
, stack_size
, flags
, &cmd
);
548 pid
= clone(container_reboot_supported
, stack
, flags
, &cmd
);
551 if (flags
& CLONE_NEWUSER
)
552 ERROR("failed to clone (%#x): %s (includes CLONE_NEWUSER)", flags
, strerror(errno
));
554 ERROR("failed to clone (%#x): %s", flags
, strerror(errno
));
557 if (wait(&status
) < 0) {
558 SYSERROR("unexpected wait error: %m");
562 if (WEXITSTATUS(status
) != 1)
569 * netpipe is used in the unprivileged case to transfer the ifindexes
570 * from parent to child
572 static int netpipe
= -1;
574 static inline int count_veths(struct lxc_list
*network
)
576 struct lxc_list
*iterator
;
577 struct lxc_netdev
*netdev
;
580 lxc_list_for_each(iterator
, network
) {
581 netdev
= iterator
->elem
;
582 if (netdev
->type
!= LXC_NET_VETH
)
589 static int read_unpriv_netifindex(struct lxc_list
*network
)
591 struct lxc_list
*iterator
;
592 struct lxc_netdev
*netdev
;
596 lxc_list_for_each(iterator
, network
) {
597 netdev
= iterator
->elem
;
598 if (netdev
->type
!= LXC_NET_VETH
)
600 if (!(netdev
->name
= malloc(IFNAMSIZ
))) {
601 ERROR("Out of memory");
605 if (read(netpipe
, netdev
->name
, IFNAMSIZ
) != IFNAMSIZ
) {
614 static int do_start(void *data
)
616 struct lxc_list
*iterator
;
617 struct lxc_handler
*handler
= data
;
618 const char *lsm_label
= NULL
;
620 if (sigprocmask(SIG_SETMASK
, &handler
->oldmask
, NULL
)) {
621 SYSERROR("failed to set sigprocmask");
625 /* This prctl must be before the synchro, so if the parent
626 * dies before we set the parent death signal, we will detect
627 * its death with the synchro right after, otherwise we have
628 * a window where the parent can exit before we set the pdeath
629 * signal leading to a unsupervized container.
631 if (prctl(PR_SET_PDEATHSIG
, SIGKILL
, 0, 0, 0)) {
632 SYSERROR("failed to set pdeath signal");
636 lxc_sync_fini_parent(handler
);
638 /* don't leak the pinfd to the container */
639 if (handler
->pinfd
>= 0) {
640 close(handler
->pinfd
);
643 /* Tell the parent task it can begin to configure the
644 * container and wait for it to finish
646 if (lxc_sync_barrier_parent(handler
, LXC_SYNC_CONFIGURE
))
649 if (read_unpriv_netifindex(&handler
->conf
->network
) < 0)
650 goto out_warn_father
;
653 * if we are in a new user namespace, become root there to have
654 * privilege over our namespace
656 if (!lxc_list_empty(&handler
->conf
->id_map
)) {
657 NOTICE("switching to gid/uid 0 in new user namespace");
660 goto out_warn_father
;
664 goto out_warn_father
;
666 if (setgroups(0, NULL
)) {
667 SYSERROR("setgroups");
668 goto out_warn_father
;
672 if (access(handler
->lxcpath
, X_OK
)) {
673 print_top_failing_dir(handler
->lxcpath
);
674 goto out_warn_father
;
677 #if HAVE_SYS_CAPABILITY_H
678 if (handler
->conf
->need_utmp_watch
) {
679 if (prctl(PR_CAPBSET_DROP
, CAP_SYS_BOOT
, 0, 0, 0)) {
680 SYSERROR("failed to remove CAP_SYS_BOOT capability");
681 goto out_warn_father
;
683 DEBUG("Dropped cap_sys_boot");
687 /* Setup the container, ip, names, utsname, ... */
688 if (lxc_setup(handler
)) {
689 ERROR("failed to setup the container");
690 goto out_warn_father
;
693 /* ask father to setup cgroups and wait for him to finish */
694 if (lxc_sync_barrier_parent(handler
, LXC_SYNC_CGROUP
))
697 /* Set the label to change to when we exec(2) the container's init */
698 if (!strcmp(lsm_name(), "AppArmor"))
699 lsm_label
= handler
->conf
->lsm_aa_profile
;
700 else if (!strcmp(lsm_name(), "SELinux"))
701 lsm_label
= handler
->conf
->lsm_se_context
;
702 if (lsm_process_label_set(lsm_label
, 1, 1) < 0)
703 goto out_warn_father
;
705 /* Some init's such as busybox will set sane tty settings on stdin,
706 * stdout, stderr which it thinks is the console. We already set them
707 * the way we wanted on the real terminal, and we want init to do its
708 * setup on its console ie. the pty allocated in lxc_console_create()
709 * so make sure that that pty is stdin,stdout,stderr.
711 if (lxc_console_set_stdfds(handler
) < 0)
712 goto out_warn_father
;
714 /* If we mounted a temporary proc, then unmount it now */
715 tmp_proc_unmount(handler
->conf
);
717 if (lxc_seccomp_load(handler
->conf
) != 0)
718 goto out_warn_father
;
720 if (run_lxc_hooks(handler
->name
, "start", handler
->conf
, handler
->lxcpath
, NULL
)) {
721 ERROR("failed to run start hooks for container '%s'.", handler
->name
);
722 goto out_warn_father
;
725 /* The clearenv() and putenv() calls have been moved here
726 * to allow us to use enviroment variables passed to the various
727 * hooks, such as the start hook above. Not all of the
728 * variables like CONFIG_PATH or ROOTFS are valid in this
729 * context but others are. */
731 SYSERROR("failed to clear environment");
732 /* don't error out though */
735 lxc_list_for_each(iterator
, &handler
->conf
->environment
) {
736 if (putenv((char *)iterator
->elem
)) {
737 SYSERROR("failed to set environment variable '%s'", (char *)iterator
->elem
);
738 goto out_warn_father
;
742 if (putenv("container=lxc")) {
743 SYSERROR("failed to set environment variable 'container=lxc'");
744 goto out_warn_father
;
747 close(handler
->sigfd
);
749 /* after this call, we are in error because this
750 * ops should not return as it execs */
751 handler
->ops
->start(handler
, handler
->data
);
754 /* we want the parent to know something went wrong, so any
755 * value other than what it expects is ok. */
756 lxc_sync_wake_parent(handler
, LXC_SYNC_POST_CONFIGURE
);
760 static int save_phys_nics(struct lxc_conf
*conf
)
762 struct lxc_list
*iterator
;
764 lxc_list_for_each(iterator
, &conf
->network
) {
765 struct lxc_netdev
*netdev
= iterator
->elem
;
767 if (netdev
->type
!= LXC_NET_PHYS
)
769 conf
->saved_nics
= realloc(conf
->saved_nics
,
770 (conf
->num_savednics
+1)*sizeof(struct saved_nic
));
771 if (!conf
->saved_nics
) {
772 SYSERROR("failed to allocate memory");
775 conf
->saved_nics
[conf
->num_savednics
].ifindex
= netdev
->ifindex
;
776 conf
->saved_nics
[conf
->num_savednics
].orig_name
= strdup(netdev
->link
);
777 if (!conf
->saved_nics
[conf
->num_savednics
].orig_name
) {
778 SYSERROR("failed to allocate memory");
781 INFO("stored saved_nic #%d idx %d name %s", conf
->num_savednics
,
782 conf
->saved_nics
[conf
->num_savednics
].ifindex
,
783 conf
->saved_nics
[conf
->num_savednics
].orig_name
);
784 conf
->num_savednics
++;
790 static int lxc_spawn(struct lxc_handler
*handler
)
792 int failed_before_rename
= 0;
793 const char *name
= handler
->name
;
794 bool cgroups_connected
= false;
795 int saved_ns_fd
[LXC_NS_MAX
];
796 int preserve_mask
= 0, i
;
797 int netpipepair
[2], nveths
;
801 for (i
= 0; i
< LXC_NS_MAX
; i
++)
802 if (handler
->conf
->inherit_ns_fd
[i
] != -1)
803 preserve_mask
|= ns_info
[i
].clone_flag
;
805 if (lxc_sync_init(handler
))
808 handler
->clone_flags
= CLONE_NEWPID
|CLONE_NEWNS
;
809 if (!lxc_list_empty(&handler
->conf
->id_map
)) {
810 INFO("Cloning a new user namespace");
811 handler
->clone_flags
|= CLONE_NEWUSER
;
814 if (handler
->conf
->inherit_ns_fd
[LXC_NS_NET
] == -1) {
815 if (!lxc_requests_empty_network(handler
))
816 handler
->clone_flags
|= CLONE_NEWNET
;
818 if (!lxc_list_empty(&handler
->conf
->network
)) {
820 /* Find gateway addresses from the link device, which is
821 * no longer accessible inside the container. Do this
822 * before creating network interfaces, since goto
823 * out_delete_net does not work before lxc_clone. */
824 if (lxc_find_gateway_addresses(handler
)) {
825 ERROR("failed to find gateway addresses");
826 lxc_sync_fini(handler
);
830 /* that should be done before the clone because we will
831 * fill the netdev index and use them in the child
833 if (lxc_create_network(handler
)) {
834 ERROR("failed to create the network");
835 lxc_sync_fini(handler
);
840 if (save_phys_nics(handler
->conf
)) {
841 ERROR("failed to save physical nic info");
845 INFO("Inheriting a net namespace");
848 if (handler
->conf
->inherit_ns_fd
[LXC_NS_IPC
] == -1) {
849 handler
->clone_flags
|= CLONE_NEWIPC
;
851 INFO("Inheriting an IPC namespace");
854 if (handler
->conf
->inherit_ns_fd
[LXC_NS_UTS
] == -1) {
855 handler
->clone_flags
|= CLONE_NEWUTS
;
857 INFO("Inheriting a UTS namespace");
861 if (!cgroup_init(handler
)) {
862 ERROR("failed initializing cgroup support");
866 cgroups_connected
= true;
868 if (!cgroup_create(handler
)) {
869 ERROR("failed creating cgroups");
874 * if the rootfs is not a blockdev, prevent the container from
875 * marking it readonly.
877 * if the container is unprivileged then skip rootfs pinning
879 if (lxc_list_empty(&handler
->conf
->id_map
)) {
880 handler
->pinfd
= pin_rootfs(handler
->conf
->rootfs
.path
);
881 if (handler
->pinfd
== -1)
882 INFO("failed to pin the container's rootfs");
885 if (preserve_ns(saved_ns_fd
, preserve_mask
) < 0)
887 if (attach_ns(handler
->conf
->inherit_ns_fd
) < 0)
890 if (am_unpriv() && (nveths
= count_veths(&handler
->conf
->network
))) {
891 if (pipe(netpipepair
) < 0) {
892 SYSERROR("Error creating pipe");
895 /* store netpipe in the global var for do_start's use */
896 netpipe
= netpipepair
[0];
899 /* Create a process in a new set of namespaces */
900 handler
->pid
= lxc_clone(do_start
, handler
, handler
->clone_flags
);
901 if (handler
->pid
< 0) {
902 SYSERROR("failed to fork into a new namespace");
906 if (attach_ns(saved_ns_fd
))
907 WARN("failed to restore saved namespaces");
909 lxc_sync_fini_child(handler
);
911 if (lxc_sync_wait_child(handler
, LXC_SYNC_CONFIGURE
))
912 failed_before_rename
= 1;
914 if (!cgroup_create_legacy(handler
)) {
915 ERROR("failed to setup the legacy cgroups for %s", name
);
918 if (!cgroup_setup_limits(handler
, false)) {
919 ERROR("failed to setup the cgroup limits for '%s'", name
);
923 if (!cgroup_enter(handler
))
926 if (!cgroup_chown(handler
))
929 if (failed_before_rename
)
932 /* Create the network configuration */
933 if (handler
->clone_flags
& CLONE_NEWNET
) {
934 if (lxc_assign_network(&handler
->conf
->network
, handler
->pid
)) {
935 ERROR("failed to create the configured network");
941 struct lxc_list
*iterator
;
942 struct lxc_netdev
*netdev
;
945 lxc_list_for_each(iterator
, &handler
->conf
->network
) {
946 netdev
= iterator
->elem
;
947 if (netdev
->type
!= LXC_NET_VETH
)
949 if (write(netpipepair
[1], netdev
->name
, IFNAMSIZ
) != IFNAMSIZ
) {
950 ERROR("Error writing veth name to container");
954 close(netpipepair
[1]);
957 /* map the container uids - the container became an invalid
958 * userid the moment it was cloned with CLONE_NEWUSER - this
959 * call doesn't change anything immediately, but allows the
960 * container to setuid(0) (0 being mapped to something else on
961 * the host) later to become a valid uid again */
962 if (lxc_map_ids(&handler
->conf
->id_map
, handler
->pid
)) {
963 ERROR("failed to set up id mapping");
967 /* Tell the child to continue its initialization. we'll get
968 * LXC_SYNC_CGROUP when it is ready for us to setup cgroups
970 if (lxc_sync_barrier_child(handler
, LXC_SYNC_POST_CONFIGURE
))
973 if (!cgroup_setup_limits(handler
, true)) {
974 ERROR("failed to setup the devices cgroup for '%s'", name
);
979 cgroups_connected
= false;
981 /* Tell the child to complete its initialization and wait for
982 * it to exec or return an error. (the child will never
983 * return LXC_SYNC_POST_CGROUP+1. It will either close the
984 * sync pipe, causing lxc_sync_barrier_child to return
985 * success, or return a different value, causing us to error
988 if (lxc_sync_barrier_child(handler
, LXC_SYNC_POST_CGROUP
))
991 if (detect_shared_rootfs())
992 umount2(handler
->conf
->rootfs
.mount
, MNT_DETACH
);
994 if (handler
->ops
->post_start(handler
, handler
->data
))
997 if (lxc_set_state(name
, handler
, RUNNING
)) {
998 ERROR("failed to set state to %s",
999 lxc_state2str(RUNNING
));
1003 lxc_sync_fini(handler
);
1008 if (cgroups_connected
)
1009 cgroup_disconnect();
1010 if (handler
->clone_flags
& CLONE_NEWNET
)
1011 lxc_delete_network(handler
);
1013 lxc_abort(name
, handler
);
1014 lxc_sync_fini(handler
);
1015 if (handler
->pinfd
>= 0) {
1016 close(handler
->pinfd
);
1017 handler
->pinfd
= -1;
1023 int get_netns_fd(int pid
)
1025 char path
[MAXPATHLEN
];
1028 ret
= snprintf(path
, MAXPATHLEN
, "/proc/%d/ns/net", pid
);
1029 if (ret
< 0 || ret
>= MAXPATHLEN
) {
1030 WARN("Failed to pin netns file for pid %d", pid
);
1034 fd
= open(path
, O_RDONLY
);
1036 WARN("Failed to pin netns file %s for pid %d: %s",
1037 path
, pid
, strerror(errno
));
1043 int __lxc_start(const char *name
, struct lxc_conf
*conf
,
1044 struct lxc_operations
* ops
, void *data
, const char *lxcpath
)
1046 struct lxc_handler
*handler
;
1051 handler
= lxc_init(name
, conf
, lxcpath
);
1053 ERROR("failed to initialize the container");
1057 handler
->data
= data
;
1059 if (must_drop_cap_sys_boot(handler
->conf
)) {
1060 #if HAVE_SYS_CAPABILITY_H
1061 DEBUG("Dropping cap_sys_boot");
1063 DEBUG("Can't drop cap_sys_boot as capabilities aren't supported");
1066 DEBUG("Not dropping cap_sys_boot or watching utmp");
1067 handler
->conf
->need_utmp_watch
= 0;
1070 if (!attach_block_device(handler
->conf
)) {
1071 ERROR("Failure attaching block device");
1072 goto out_fini_nonet
;
1075 if (geteuid() == 0 && !lxc_list_empty(&conf
->id_map
)) {
1076 /* if the backing store is a device, mount it here and now */
1077 if (rootfs_is_blockdev(conf
)) {
1078 if (unshare(CLONE_NEWNS
) < 0) {
1079 ERROR("Error unsharing mounts");
1080 goto out_fini_nonet
;
1082 if (do_rootfs_setup(conf
, name
, lxcpath
) < 0) {
1083 ERROR("Error setting up rootfs mount as root before spawn");
1084 goto out_fini_nonet
;
1086 INFO("Set up container rootfs as host root");
1090 err
= lxc_spawn(handler
);
1092 ERROR("failed to spawn '%s'", name
);
1093 goto out_detach_blockdev
;
1096 netnsfd
= get_netns_fd(handler
->pid
);
1098 err
= lxc_poll(name
, handler
);
1100 ERROR("mainloop exited with an error");
1106 while (waitpid(handler
->pid
, &status
, 0) < 0 && errno
== EINTR
)
1110 * If the child process exited but was not signaled,
1111 * it didn't call reboot. This should mean it was an
1112 * lxc-execute which simply exited. In any case, treat
1115 if (WIFSIGNALED(status
)) {
1116 switch(WTERMSIG(status
)) {
1117 case SIGINT
: /* halt */
1118 DEBUG("Container halting");
1120 case SIGHUP
: /* reboot */
1121 DEBUG("Container rebooting");
1122 handler
->conf
->reboot
= 1;
1124 case SIGSYS
: /* seccomp */
1125 DEBUG("Container violated its seccomp policy");
1128 DEBUG("unknown exit status for init: %d", WTERMSIG(status
));
1133 lxc_rename_phys_nics_on_shutdown(netnsfd
, handler
->conf
);
1137 if (handler
->pinfd
>= 0) {
1138 close(handler
->pinfd
);
1139 handler
->pinfd
= -1;
1142 lxc_monitor_send_exit_code(name
, status
, handler
->lxcpath
);
1143 err
= lxc_error_set_and_log(handler
->pid
, status
);
1145 lxc_delete_network(handler
);
1147 out_detach_blockdev
:
1148 detach_block_device(handler
->conf
);
1151 lxc_fini(name
, handler
);
1155 lxc_abort(name
, handler
);
1163 static int start(struct lxc_handler
*handler
, void* data
)
1165 struct start_args
*arg
= data
;
1167 NOTICE("exec'ing '%s'", arg
->argv
[0]);
1169 execvp(arg
->argv
[0], arg
->argv
);
1170 SYSERROR("failed to exec %s", arg
->argv
[0]);
1174 static int post_start(struct lxc_handler
*handler
, void* data
)
1176 struct start_args
*arg
= data
;
1178 NOTICE("'%s' started with pid '%d'", arg
->argv
[0], handler
->pid
);
1182 static struct lxc_operations start_ops
= {
1184 .post_start
= post_start
1187 int lxc_start(const char *name
, char *const argv
[], struct lxc_conf
*conf
,
1188 const char *lxcpath
)
1190 struct start_args start_arg
= {
1194 if (lxc_check_inherited(conf
, -1))
1197 conf
->need_utmp_watch
= 1;
1198 return __lxc_start(name
, conf
, &start_ops
, &start_arg
, lxcpath
);