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 bool do_destroy_container(struct lxc_conf
*conf
);
87 static int lxc_rmdir_onedev_wrapper(void *data
);
88 static void lxc_destroy_container_on_signal(struct lxc_handler
*handler
,
91 static void print_top_failing_dir(const char *path
)
93 size_t len
= strlen(path
);
94 char *copy
= alloca(len
+1), *p
, *e
, saved
;
100 while (p
< e
&& *p
== '/') p
++;
101 while (p
< e
&& *p
!= '/') p
++;
104 if (access(copy
, X_OK
)) {
105 SYSERROR("could not access %s. Please grant it 'x' " \
106 "access, or add an ACL for the container root.",
114 static void close_ns(int ns_fd
[LXC_NS_MAX
]) {
117 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
125 static int preserve_ns(int ns_fd
[LXC_NS_MAX
], int clone_flags
) {
127 char path
[MAXPATHLEN
];
129 for (i
= 0; i
< LXC_NS_MAX
; i
++)
132 if (access("/proc/self/ns", X_OK
)) {
133 WARN("Kernel does not support attach; preserve_ns ignored");
137 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
138 if ((clone_flags
& ns_info
[i
].clone_flag
) == 0)
140 snprintf(path
, MAXPATHLEN
, "/proc/self/ns/%s", ns_info
[i
].proc_name
);
141 ns_fd
[i
] = open(path
, O_RDONLY
| O_CLOEXEC
);
152 SYSERROR("failed to open '%s'", path
);
156 static int attach_ns(const int ns_fd
[LXC_NS_MAX
]) {
159 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
163 if (setns(ns_fd
[i
], 0) != 0)
169 SYSERROR("failed to set namespace '%s'", ns_info
[i
].proc_name
);
173 static int match_fd(int fd
)
175 return (fd
== 0 || fd
== 1 || fd
== 2);
179 * Check for any fds we need to close
180 * * if fd_to_ignore != -1, then if we find that fd open we will ignore it.
181 * * By default we warn about open fds we find.
182 * * If closeall is true, we will close open fds.
183 * * If lxc-start was passed "-C", then conf->close_all_fds will be true,
184 * in which case we also close all open fds.
185 * * A daemonized container will always pass closeall=true.
187 int lxc_check_inherited(struct lxc_conf
*conf
, bool closeall
, int fd_to_ignore
)
189 struct dirent dirent
, *direntp
;
193 if (conf
&& conf
->close_all_fds
)
197 dir
= opendir("/proc/self/fd");
199 WARN("failed to open directory: %m");
205 while (!readdir_r(dir
, &dirent
, &direntp
)) {
209 if (!strcmp(direntp
->d_name
, "."))
212 if (!strcmp(direntp
->d_name
, ".."))
215 fd
= atoi(direntp
->d_name
);
217 if (fd
== fddir
|| fd
== lxc_log_fd
|| fd
== fd_to_ignore
)
220 if (current_config
&& fd
== current_config
->logfd
)
229 INFO("closed inherited fd %d", fd
);
232 WARN("inherited fd %d", fd
);
235 closedir(dir
); /* cannot fail */
239 static int setup_signal_fd(sigset_t
*oldmask
)
244 /* Block everything except serious error signals */
245 if (sigfillset(&mask
) ||
246 sigdelset(&mask
, SIGILL
) ||
247 sigdelset(&mask
, SIGSEGV
) ||
248 sigdelset(&mask
, SIGBUS
) ||
249 sigdelset(&mask
, SIGWINCH
) ||
250 sigprocmask(SIG_BLOCK
, &mask
, oldmask
)) {
251 SYSERROR("failed to set signal mask");
255 fd
= signalfd(-1, &mask
, 0);
257 SYSERROR("failed to create the signal fd");
261 if (fcntl(fd
, F_SETFD
, FD_CLOEXEC
)) {
262 SYSERROR("failed to set sigfd to close-on-exec");
267 DEBUG("sigchild handler set");
272 static int signal_handler(int fd
, uint32_t events
, void *data
,
273 struct lxc_epoll_descr
*descr
)
275 struct signalfd_siginfo siginfo
;
279 bool init_died
= false;
281 ret
= read(fd
, &siginfo
, sizeof(siginfo
));
283 ERROR("failed to read signal info");
287 if (ret
!= sizeof(siginfo
)) {
288 ERROR("unexpected siginfo size");
292 // check whether init is running
294 ret
= waitid(P_PID
, *pid
, &info
, WEXITED
| WNOWAIT
| WNOHANG
);
295 if (ret
== 0 && info
.si_pid
== *pid
) {
299 if (siginfo
.ssi_signo
!= SIGCHLD
) {
300 kill(*pid
, siginfo
.ssi_signo
);
301 INFO("forwarded signal %d to pid %d", siginfo
.ssi_signo
, *pid
);
302 return init_died
? 1 : 0;
305 if (siginfo
.ssi_code
== CLD_STOPPED
||
306 siginfo
.ssi_code
== CLD_CONTINUED
) {
307 INFO("container init process was stopped/continued");
308 return init_died
? 1 : 0;
311 /* more robustness, protect ourself from a SIGCHLD sent
312 * by a process different from the container init
314 if (siginfo
.ssi_pid
!= *pid
) {
315 WARN("invalid pid for SIGCHLD");
316 return init_died
? 1 : 0;
319 DEBUG("container init process exited");
323 int lxc_set_state(const char *name
, struct lxc_handler
*handler
, lxc_state_t state
)
325 handler
->state
= state
;
326 lxc_monitor_send_state(name
, state
, handler
->lxcpath
);
330 int lxc_poll(const char *name
, struct lxc_handler
*handler
)
332 int sigfd
= handler
->sigfd
;
333 int pid
= handler
->pid
;
334 struct lxc_epoll_descr descr
;
336 if (lxc_mainloop_open(&descr
)) {
337 ERROR("failed to create mainloop");
341 if (lxc_mainloop_add_handler(&descr
, sigfd
, signal_handler
, &pid
)) {
342 ERROR("failed to add handler for the signal");
343 goto out_mainloop_open
;
346 if (lxc_console_mainloop_add(&descr
, handler
)) {
347 ERROR("failed to add console handler to mainloop");
348 goto out_mainloop_open
;
351 if (lxc_cmd_mainloop_add(name
, &descr
, handler
)) {
352 ERROR("failed to add command handler to mainloop");
353 goto out_mainloop_open
;
356 if (handler
->conf
->need_utmp_watch
) {
357 #if HAVE_SYS_CAPABILITY_H
358 if (lxc_utmp_mainloop_add(&descr
, handler
)) {
359 ERROR("failed to add utmp handler to mainloop");
360 goto out_mainloop_open
;
363 DEBUG("not starting utmp handler as cap_sys_boot cannot be dropped without capabilities support");
367 return lxc_mainloop(&descr
, -1);
370 lxc_mainloop_close(&descr
);
376 struct lxc_handler
*lxc_init(const char *name
, struct lxc_conf
*conf
, const char *lxcpath
)
378 struct lxc_handler
*handler
;
380 handler
= malloc(sizeof(*handler
));
384 memset(handler
, 0, sizeof(*handler
));
386 handler
->ttysock
[0] = handler
->ttysock
[1] = -1;
387 handler
->conf
= conf
;
388 handler
->lxcpath
= lxcpath
;
393 handler
->name
= strdup(name
);
394 if (!handler
->name
) {
395 ERROR("failed to allocate memory");
399 if (lxc_cmd_init(name
, handler
, lxcpath
))
402 if (lxc_read_seccomp_config(conf
) != 0) {
403 ERROR("failed loading seccomp policy");
404 goto out_close_maincmd_fd
;
407 /* Begin by setting the state to STARTING */
408 if (lxc_set_state(name
, handler
, STARTING
)) {
409 ERROR("failed to set state '%s'", lxc_state2str(STARTING
));
410 goto out_close_maincmd_fd
;
413 /* Start of environment variable setup for hooks */
414 if (name
&& setenv("LXC_NAME", name
, 1)) {
415 SYSERROR("failed to set environment variable for container name");
417 if (conf
->rcfile
&& setenv("LXC_CONFIG_FILE", conf
->rcfile
, 1)) {
418 SYSERROR("failed to set environment variable for config path");
420 if (conf
->rootfs
.mount
&& setenv("LXC_ROOTFS_MOUNT", conf
->rootfs
.mount
, 1)) {
421 SYSERROR("failed to set environment variable for rootfs mount");
423 if (conf
->rootfs
.path
&& setenv("LXC_ROOTFS_PATH", conf
->rootfs
.path
, 1)) {
424 SYSERROR("failed to set environment variable for rootfs mount");
426 if (conf
->console
.path
&& setenv("LXC_CONSOLE", conf
->console
.path
, 1)) {
427 SYSERROR("failed to set environment variable for console path");
429 if (conf
->console
.log_path
&& setenv("LXC_CONSOLE_LOGPATH", conf
->console
.log_path
, 1)) {
430 SYSERROR("failed to set environment variable for console log");
432 /* End of environment variable setup for hooks */
434 if (run_lxc_hooks(name
, "pre-start", conf
, handler
->lxcpath
, NULL
)) {
435 ERROR("failed to run pre-start hooks for container '%s'.", name
);
439 /* the signal fd has to be created before forking otherwise
440 * if the child process exits before we setup the signal fd,
441 * the event will be lost and the command will be stuck */
442 handler
->sigfd
= setup_signal_fd(&handler
->oldmask
);
443 if (handler
->sigfd
< 0) {
444 ERROR("failed to set sigchild fd handler");
448 /* do this after setting up signals since it might unblock SIGWINCH */
449 if (lxc_console_create(conf
)) {
450 ERROR("failed to create console");
451 goto out_restore_sigmask
;
454 if (ttys_shift_ids(conf
) < 0) {
455 ERROR("Failed to shift tty into container");
456 goto out_restore_sigmask
;
459 INFO("'%s' is initialized", name
);
463 sigprocmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
465 lxc_delete_tty(&conf
->tty_info
);
467 lxc_set_state(name
, handler
, ABORTING
);
468 out_close_maincmd_fd
:
469 close(conf
->maincmd_fd
);
470 conf
->maincmd_fd
= -1;
473 handler
->name
= NULL
;
479 void lxc_fini(const char *name
, struct lxc_handler
*handler
)
481 /* The STOPPING state is there for future cleanup code
482 * which can take awhile
484 lxc_set_state(name
, handler
, STOPPING
);
485 lxc_set_state(name
, handler
, STOPPED
);
487 if (run_lxc_hooks(name
, "post-stop", handler
->conf
, handler
->lxcpath
, NULL
))
488 ERROR("failed to run post-stop hooks for container '%s'.", name
);
490 /* reset mask set by setup_signal_fd */
491 if (sigprocmask(SIG_SETMASK
, &handler
->oldmask
, NULL
))
492 WARN("failed to restore sigprocmask");
494 lxc_console_delete(&handler
->conf
->console
);
495 lxc_delete_tty(&handler
->conf
->tty_info
);
496 close(handler
->conf
->maincmd_fd
);
497 handler
->conf
->maincmd_fd
= -1;
499 if (handler
->ttysock
[0] != -1) {
500 close(handler
->ttysock
[0]);
501 close(handler
->ttysock
[1]);
503 if (handler
->conf
->ephemeral
== 1) {
504 lxc_destroy_container_on_signal(handler
, name
);
506 cgroup_destroy(handler
);
510 void lxc_abort(const char *name
, struct lxc_handler
*handler
)
514 lxc_set_state(name
, handler
, ABORTING
);
515 if (handler
->pid
> 0)
516 kill(handler
->pid
, SIGKILL
);
517 while ((ret
= waitpid(-1, &status
, 0)) > 0) ;
520 #include <sys/reboot.h>
521 #include <linux/reboot.h>
524 * reboot(LINUX_REBOOT_CMD_CAD_ON) will return -EINVAL
525 * in a child pid namespace if container reboot support exists.
526 * Otherwise, it will either succeed or return -EPERM.
528 static int container_reboot_supported(void *arg
)
534 if (ret
== -1 && errno
== EINVAL
)
539 static int must_drop_cap_sys_boot(struct lxc_conf
*conf
)
542 int ret
, cmd
, v
, flags
;
543 long stack_size
= 4096;
544 void *stack
= alloca(stack_size
);
548 f
= fopen("/proc/sys/kernel/ctrl-alt-del", "r");
550 DEBUG("failed to open /proc/sys/kernel/ctrl-alt-del");
554 ret
= fscanf(f
, "%d", &v
);
557 DEBUG("Failed to read /proc/sys/kernel/ctrl-alt-del");
560 cmd
= v
? LINUX_REBOOT_CMD_CAD_ON
: LINUX_REBOOT_CMD_CAD_OFF
;
562 flags
= CLONE_NEWPID
| SIGCHLD
;
563 if (!lxc_list_empty(&conf
->id_map
))
564 flags
|= CLONE_NEWUSER
;
567 pid
= __clone2(container_reboot_supported
, stack
, stack_size
, flags
, &cmd
);
570 pid
= clone(container_reboot_supported
, stack
, flags
, &cmd
);
573 if (flags
& CLONE_NEWUSER
)
574 ERROR("failed to clone (%#x): %s (includes CLONE_NEWUSER)", flags
, strerror(errno
));
576 ERROR("failed to clone (%#x): %s", flags
, strerror(errno
));
579 if (wait(&status
) < 0) {
580 SYSERROR("unexpected wait error: %m");
584 if (WEXITSTATUS(status
) != 1)
591 * netpipe is used in the unprivileged case to transfer the ifindexes
592 * from parent to child
594 static int netpipe
= -1;
596 static inline int count_veths(struct lxc_list
*network
)
598 struct lxc_list
*iterator
;
599 struct lxc_netdev
*netdev
;
602 lxc_list_for_each(iterator
, network
) {
603 netdev
= iterator
->elem
;
604 if (netdev
->type
!= LXC_NET_VETH
)
611 static int read_unpriv_netifindex(struct lxc_list
*network
)
613 struct lxc_list
*iterator
;
614 struct lxc_netdev
*netdev
;
618 lxc_list_for_each(iterator
, network
) {
619 netdev
= iterator
->elem
;
620 if (netdev
->type
!= LXC_NET_VETH
)
622 if (!(netdev
->name
= malloc(IFNAMSIZ
))) {
623 ERROR("Out of memory");
627 if (read(netpipe
, netdev
->name
, IFNAMSIZ
) != IFNAMSIZ
) {
636 static int do_start(void *data
)
638 struct lxc_list
*iterator
;
639 struct lxc_handler
*handler
= data
;
641 if (sigprocmask(SIG_SETMASK
, &handler
->oldmask
, NULL
)) {
642 SYSERROR("failed to set sigprocmask");
646 /* This prctl must be before the synchro, so if the parent
647 * dies before we set the parent death signal, we will detect
648 * its death with the synchro right after, otherwise we have
649 * a window where the parent can exit before we set the pdeath
650 * signal leading to a unsupervized container.
652 if (prctl(PR_SET_PDEATHSIG
, SIGKILL
, 0, 0, 0)) {
653 SYSERROR("failed to set pdeath signal");
657 lxc_sync_fini_parent(handler
);
659 /* don't leak the pinfd to the container */
660 if (handler
->pinfd
>= 0) {
661 close(handler
->pinfd
);
664 /* Tell the parent task it can begin to configure the
665 * container and wait for it to finish
667 if (lxc_sync_barrier_parent(handler
, LXC_SYNC_CONFIGURE
))
670 if (read_unpriv_netifindex(&handler
->conf
->network
) < 0)
671 goto out_warn_father
;
674 * if we are in a new user namespace, become root there to have
675 * privilege over our namespace. When using lxc-execute we default to root,
676 * but this can be overriden using the lxc.init_uid and lxc.init_gid
677 * configuration options.
679 if (!lxc_list_empty(&handler
->conf
->id_map
)) {
681 if (handler
->conf
->is_execute
&& handler
->conf
->init_gid
)
682 new_gid
= handler
->conf
->init_gid
;
685 if (handler
->conf
->is_execute
&& handler
->conf
->init_uid
)
686 new_uid
= handler
->conf
->init_uid
;
688 NOTICE("switching to gid/uid %d/%d in new user namespace", new_gid
, new_uid
);
689 if (setgid(new_gid
)) {
691 goto out_warn_father
;
693 if (setuid(new_uid
)) {
695 goto out_warn_father
;
697 if (setgroups(0, NULL
)) {
698 SYSERROR("setgroups");
699 goto out_warn_father
;
703 if (access(handler
->lxcpath
, X_OK
)) {
704 print_top_failing_dir(handler
->lxcpath
);
705 goto out_warn_father
;
708 #if HAVE_SYS_CAPABILITY_H
709 if (handler
->conf
->need_utmp_watch
) {
710 if (prctl(PR_CAPBSET_DROP
, CAP_SYS_BOOT
, 0, 0, 0)) {
711 SYSERROR("failed to remove CAP_SYS_BOOT capability");
712 goto out_warn_father
;
714 DEBUG("Dropped cap_sys_boot");
718 /* Setup the container, ip, names, utsname, ... */
719 if (lxc_setup(handler
)) {
720 ERROR("failed to setup the container");
721 goto out_warn_father
;
724 /* ask father to setup cgroups and wait for him to finish */
725 if (lxc_sync_barrier_parent(handler
, LXC_SYNC_CGROUP
))
728 /* Set the label to change to when we exec(2) the container's init */
729 if (lsm_process_label_set(NULL
, handler
->conf
, 1, 1) < 0)
730 goto out_warn_father
;
732 /* Some init's such as busybox will set sane tty settings on stdin,
733 * stdout, stderr which it thinks is the console. We already set them
734 * the way we wanted on the real terminal, and we want init to do its
735 * setup on its console ie. the pty allocated in lxc_console_create()
736 * so make sure that that pty is stdin,stdout,stderr.
738 if (lxc_console_set_stdfds(handler
) < 0)
739 goto out_warn_father
;
741 /* If we mounted a temporary proc, then unmount it now */
742 tmp_proc_unmount(handler
->conf
);
744 if (lxc_seccomp_load(handler
->conf
) != 0)
745 goto out_warn_father
;
747 if (run_lxc_hooks(handler
->name
, "start", handler
->conf
, handler
->lxcpath
, NULL
)) {
748 ERROR("failed to run start hooks for container '%s'.", handler
->name
);
749 goto out_warn_father
;
752 /* The clearenv() and putenv() calls have been moved here
753 * to allow us to use environment variables passed to the various
754 * hooks, such as the start hook above. Not all of the
755 * variables like CONFIG_PATH or ROOTFS are valid in this
756 * context but others are. */
758 SYSERROR("failed to clear environment");
759 /* don't error out though */
762 lxc_list_for_each(iterator
, &handler
->conf
->environment
) {
763 if (putenv((char *)iterator
->elem
)) {
764 SYSERROR("failed to set environment variable '%s'", (char *)iterator
->elem
);
765 goto out_warn_father
;
769 if (putenv("container=lxc")) {
770 SYSERROR("failed to set environment variable 'container=lxc'");
771 goto out_warn_father
;
774 if (handler
->conf
->pty_names
) {
775 if (putenv(handler
->conf
->pty_names
)) {
776 SYSERROR("failed to set environment variable for container ptys");
777 goto out_warn_father
;
781 close(handler
->sigfd
);
783 if (handler
->backgrounded
&& null_stdfds() < 0)
784 goto out_warn_father
;
786 /* after this call, we are in error because this
787 * ops should not return as it execs */
788 handler
->ops
->start(handler
, handler
->data
);
791 /* we want the parent to know something went wrong, so any
792 * value other than what it expects is ok. */
793 lxc_sync_wake_parent(handler
, LXC_SYNC_POST_CONFIGURE
);
797 static int save_phys_nics(struct lxc_conf
*conf
)
799 struct lxc_list
*iterator
;
801 lxc_list_for_each(iterator
, &conf
->network
) {
802 struct lxc_netdev
*netdev
= iterator
->elem
;
804 if (netdev
->type
!= LXC_NET_PHYS
)
806 conf
->saved_nics
= realloc(conf
->saved_nics
,
807 (conf
->num_savednics
+1)*sizeof(struct saved_nic
));
808 if (!conf
->saved_nics
) {
809 SYSERROR("failed to allocate memory");
812 conf
->saved_nics
[conf
->num_savednics
].ifindex
= netdev
->ifindex
;
813 conf
->saved_nics
[conf
->num_savednics
].orig_name
= strdup(netdev
->link
);
814 if (!conf
->saved_nics
[conf
->num_savednics
].orig_name
) {
815 SYSERROR("failed to allocate memory");
818 INFO("stored saved_nic #%d idx %d name %s", conf
->num_savednics
,
819 conf
->saved_nics
[conf
->num_savednics
].ifindex
,
820 conf
->saved_nics
[conf
->num_savednics
].orig_name
);
821 conf
->num_savednics
++;
827 static int recv_fd(int sock
, int *fd
)
829 if (lxc_abstract_unix_recv_fd(sock
, fd
, NULL
, 0) < 0) {
830 SYSERROR("Error receiving tty fd from child");
838 static int recv_ttys_from_child(struct lxc_handler
*handler
)
840 struct lxc_conf
*conf
= handler
->conf
;
841 int i
, sock
= handler
->ttysock
[1];
842 struct lxc_tty_info
*tty_info
= &conf
->tty_info
;
847 tty_info
->pty_info
= malloc(sizeof(*tty_info
->pty_info
)*conf
->tty
);
848 if (!tty_info
->pty_info
) {
849 SYSERROR("failed to allocate pty_info");
853 for (i
= 0; i
< conf
->tty
; i
++) {
854 struct lxc_pty_info
*pty_info
= &tty_info
->pty_info
[i
];
856 if (recv_fd(sock
, &pty_info
->slave
) < 0 ||
857 recv_fd(sock
, &pty_info
->master
) < 0) {
858 ERROR("Error receiving tty info from child");
862 tty_info
->nbtty
= conf
->tty
;
867 void resolve_clone_flags(struct lxc_handler
*handler
)
869 handler
->clone_flags
= CLONE_NEWPID
| CLONE_NEWNS
;
871 if (!lxc_list_empty(&handler
->conf
->id_map
)) {
872 INFO("Cloning a new user namespace");
873 handler
->clone_flags
|= CLONE_NEWUSER
;
876 if (handler
->conf
->inherit_ns_fd
[LXC_NS_NET
] == -1) {
877 if (!lxc_requests_empty_network(handler
))
878 handler
->clone_flags
|= CLONE_NEWNET
;
880 INFO("Inheriting a net namespace");
883 if (handler
->conf
->inherit_ns_fd
[LXC_NS_IPC
] == -1) {
884 handler
->clone_flags
|= CLONE_NEWIPC
;
886 INFO("Inheriting an IPC namespace");
889 if (handler
->conf
->inherit_ns_fd
[LXC_NS_UTS
] == -1) {
890 handler
->clone_flags
|= CLONE_NEWUTS
;
892 INFO("Inheriting a UTS namespace");
896 static int lxc_spawn(struct lxc_handler
*handler
)
898 int failed_before_rename
= 0;
899 const char *name
= handler
->name
;
900 bool cgroups_connected
= false;
901 int saved_ns_fd
[LXC_NS_MAX
];
902 int preserve_mask
= 0, i
;
903 int netpipepair
[2], nveths
;
907 for (i
= 0; i
< LXC_NS_MAX
; i
++)
908 if (handler
->conf
->inherit_ns_fd
[i
] != -1)
909 preserve_mask
|= ns_info
[i
].clone_flag
;
911 if (lxc_sync_init(handler
))
914 if (socketpair(AF_UNIX
, SOCK_DGRAM
, 0, handler
->ttysock
) < 0) {
915 lxc_sync_fini(handler
);
919 resolve_clone_flags(handler
);
921 if (handler
->clone_flags
& CLONE_NEWNET
) {
922 if (!lxc_list_empty(&handler
->conf
->network
)) {
924 /* Find gateway addresses from the link device, which is
925 * no longer accessible inside the container. Do this
926 * before creating network interfaces, since goto
927 * out_delete_net does not work before lxc_clone. */
928 if (lxc_find_gateway_addresses(handler
)) {
929 ERROR("failed to find gateway addresses");
930 lxc_sync_fini(handler
);
934 /* that should be done before the clone because we will
935 * fill the netdev index and use them in the child
937 if (lxc_create_network(handler
)) {
938 ERROR("failed to create the network");
939 lxc_sync_fini(handler
);
944 if (save_phys_nics(handler
->conf
)) {
945 ERROR("failed to save physical nic info");
950 if (!cgroup_init(handler
)) {
951 ERROR("failed initializing cgroup support");
955 cgroups_connected
= true;
957 if (!cgroup_create(handler
)) {
958 ERROR("failed creating cgroups");
963 * if the rootfs is not a blockdev, prevent the container from
964 * marking it readonly.
966 * if the container is unprivileged then skip rootfs pinning
968 if (lxc_list_empty(&handler
->conf
->id_map
)) {
969 handler
->pinfd
= pin_rootfs(handler
->conf
->rootfs
.path
);
970 if (handler
->pinfd
== -1)
971 INFO("failed to pin the container's rootfs");
974 if (preserve_ns(saved_ns_fd
, preserve_mask
) < 0)
976 if (attach_ns(handler
->conf
->inherit_ns_fd
) < 0)
979 if (am_unpriv() && (nveths
= count_veths(&handler
->conf
->network
))) {
980 if (pipe(netpipepair
) < 0) {
981 SYSERROR("Error creating pipe");
984 /* store netpipe in the global var for do_start's use */
985 netpipe
= netpipepair
[0];
988 /* Create a process in a new set of namespaces */
989 handler
->pid
= lxc_clone(do_start
, handler
, handler
->clone_flags
);
990 if (handler
->pid
< 0) {
991 SYSERROR("failed to fork into a new namespace");
995 if (attach_ns(saved_ns_fd
))
996 WARN("failed to restore saved namespaces");
998 lxc_sync_fini_child(handler
);
1000 if (lxc_sync_wait_child(handler
, LXC_SYNC_CONFIGURE
))
1001 failed_before_rename
= 1;
1003 if (!cgroup_create_legacy(handler
)) {
1004 ERROR("failed to setup the legacy cgroups for %s", name
);
1005 goto out_delete_net
;
1007 if (!cgroup_setup_limits(handler
, false)) {
1008 ERROR("failed to setup the cgroup limits for '%s'", name
);
1009 goto out_delete_net
;
1012 if (!cgroup_enter(handler
))
1013 goto out_delete_net
;
1015 if (!cgroup_chown(handler
))
1016 goto out_delete_net
;
1018 if (failed_before_rename
)
1019 goto out_delete_net
;
1021 /* Create the network configuration */
1022 if (handler
->clone_flags
& CLONE_NEWNET
) {
1023 if (lxc_assign_network(&handler
->conf
->network
, handler
->pid
)) {
1024 ERROR("failed to create the configured network");
1025 goto out_delete_net
;
1029 if (netpipe
!= -1) {
1030 struct lxc_list
*iterator
;
1031 struct lxc_netdev
*netdev
;
1034 lxc_list_for_each(iterator
, &handler
->conf
->network
) {
1035 netdev
= iterator
->elem
;
1036 if (netdev
->type
!= LXC_NET_VETH
)
1038 if (write(netpipepair
[1], netdev
->name
, IFNAMSIZ
) != IFNAMSIZ
) {
1039 ERROR("Error writing veth name to container");
1040 goto out_delete_net
;
1043 close(netpipepair
[1]);
1046 /* map the container uids - the container became an invalid
1047 * userid the moment it was cloned with CLONE_NEWUSER - this
1048 * call doesn't change anything immediately, but allows the
1049 * container to setuid(0) (0 being mapped to something else on
1050 * the host) later to become a valid uid again */
1051 if (lxc_map_ids(&handler
->conf
->id_map
, handler
->pid
)) {
1052 ERROR("failed to set up id mapping");
1053 goto out_delete_net
;
1056 /* Tell the child to continue its initialization. we'll get
1057 * LXC_SYNC_CGROUP when it is ready for us to setup cgroups
1059 if (lxc_sync_barrier_child(handler
, LXC_SYNC_POST_CONFIGURE
))
1060 goto out_delete_net
;
1062 if (!cgroup_setup_limits(handler
, true)) {
1063 ERROR("failed to setup the devices cgroup for '%s'", name
);
1064 goto out_delete_net
;
1067 cgroup_disconnect();
1068 cgroups_connected
= false;
1070 /* read tty fds allocated by child */
1071 if (recv_ttys_from_child(handler
) < 0) {
1072 ERROR("failed to receive tty info from child");
1073 goto out_delete_net
;
1076 /* Tell the child to complete its initialization and wait for
1077 * it to exec or return an error. (the child will never
1078 * return LXC_SYNC_POST_CGROUP+1. It will either close the
1079 * sync pipe, causing lxc_sync_barrier_child to return
1080 * success, or return a different value, causing us to error
1083 if (lxc_sync_barrier_child(handler
, LXC_SYNC_POST_CGROUP
))
1086 if (detect_shared_rootfs())
1087 umount2(handler
->conf
->rootfs
.mount
, MNT_DETACH
);
1089 if (handler
->ops
->post_start(handler
, handler
->data
))
1092 if (lxc_set_state(name
, handler
, RUNNING
)) {
1093 ERROR("failed to set state to %s",
1094 lxc_state2str(RUNNING
));
1098 lxc_sync_fini(handler
);
1103 if (cgroups_connected
)
1104 cgroup_disconnect();
1105 if (handler
->clone_flags
& CLONE_NEWNET
)
1106 lxc_delete_network(handler
);
1108 lxc_abort(name
, handler
);
1109 lxc_sync_fini(handler
);
1110 if (handler
->pinfd
>= 0) {
1111 close(handler
->pinfd
);
1112 handler
->pinfd
= -1;
1118 int get_netns_fd(int pid
)
1120 char path
[MAXPATHLEN
];
1123 ret
= snprintf(path
, MAXPATHLEN
, "/proc/%d/ns/net", pid
);
1124 if (ret
< 0 || ret
>= MAXPATHLEN
) {
1125 WARN("Failed to pin netns file for pid %d", pid
);
1129 fd
= open(path
, O_RDONLY
);
1131 WARN("Failed to pin netns file %s for pid %d: %s",
1132 path
, pid
, strerror(errno
));
1138 int __lxc_start(const char *name
, struct lxc_conf
*conf
,
1139 struct lxc_operations
* ops
, void *data
, const char *lxcpath
,
1142 struct lxc_handler
*handler
;
1147 handler
= lxc_init(name
, conf
, lxcpath
);
1149 ERROR("failed to initialize the container");
1153 handler
->data
= data
;
1154 handler
->backgrounded
= backgrounded
;
1156 if (must_drop_cap_sys_boot(handler
->conf
)) {
1157 #if HAVE_SYS_CAPABILITY_H
1158 DEBUG("Dropping cap_sys_boot");
1160 DEBUG("Can't drop cap_sys_boot as capabilities aren't supported");
1163 DEBUG("Not dropping cap_sys_boot or watching utmp");
1164 handler
->conf
->need_utmp_watch
= 0;
1167 if (!attach_block_device(handler
->conf
)) {
1168 ERROR("Failure attaching block device");
1169 goto out_fini_nonet
;
1172 if (geteuid() == 0 && !lxc_list_empty(&conf
->id_map
)) {
1173 /* if the backing store is a device, mount it here and now */
1174 if (rootfs_is_blockdev(conf
)) {
1175 if (unshare(CLONE_NEWNS
) < 0) {
1176 ERROR("Error unsharing mounts");
1177 goto out_fini_nonet
;
1179 remount_all_slave();
1180 if (do_rootfs_setup(conf
, name
, lxcpath
) < 0) {
1181 ERROR("Error setting up rootfs mount as root before spawn");
1182 goto out_fini_nonet
;
1184 INFO("Set up container rootfs as host root");
1188 err
= lxc_spawn(handler
);
1190 ERROR("failed to spawn '%s'", name
);
1191 goto out_detach_blockdev
;
1194 handler
->conf
->reboot
= 0;
1196 netnsfd
= get_netns_fd(handler
->pid
);
1198 err
= lxc_poll(name
, handler
);
1200 ERROR("mainloop exited with an error");
1206 while (waitpid(handler
->pid
, &status
, 0) < 0 && errno
== EINTR
)
1210 * If the child process exited but was not signaled,
1211 * it didn't call reboot. This should mean it was an
1212 * lxc-execute which simply exited. In any case, treat
1215 if (WIFSIGNALED(status
)) {
1216 switch(WTERMSIG(status
)) {
1217 case SIGINT
: /* halt */
1218 DEBUG("Container halting");
1220 case SIGHUP
: /* reboot */
1221 DEBUG("Container rebooting");
1222 handler
->conf
->reboot
= 1;
1224 case SIGSYS
: /* seccomp */
1225 DEBUG("Container violated its seccomp policy");
1228 DEBUG("unknown exit status for init: %d", WTERMSIG(status
));
1233 DEBUG("Pushing physical nics back to host namespace");
1234 lxc_rename_phys_nics_on_shutdown(netnsfd
, handler
->conf
);
1236 DEBUG("Tearing down virtual network devices used by container");
1237 lxc_delete_network(handler
);
1242 if (handler
->pinfd
>= 0) {
1243 close(handler
->pinfd
);
1244 handler
->pinfd
= -1;
1247 lxc_monitor_send_exit_code(name
, status
, handler
->lxcpath
);
1248 err
= lxc_error_set_and_log(handler
->pid
, status
);
1250 lxc_delete_network(handler
);
1252 out_detach_blockdev
:
1253 detach_block_device(handler
->conf
);
1256 lxc_fini(name
, handler
);
1260 lxc_abort(name
, handler
);
1268 static int start(struct lxc_handler
*handler
, void* data
)
1270 struct start_args
*arg
= data
;
1272 NOTICE("exec'ing '%s'", arg
->argv
[0]);
1274 execvp(arg
->argv
[0], arg
->argv
);
1275 SYSERROR("failed to exec %s", arg
->argv
[0]);
1279 static int post_start(struct lxc_handler
*handler
, void* data
)
1281 struct start_args
*arg
= data
;
1283 NOTICE("'%s' started with pid '%d'", arg
->argv
[0], handler
->pid
);
1287 static struct lxc_operations start_ops
= {
1289 .post_start
= post_start
1292 int lxc_start(const char *name
, char *const argv
[], struct lxc_conf
*conf
,
1293 const char *lxcpath
, bool backgrounded
)
1295 struct start_args start_arg
= {
1299 conf
->need_utmp_watch
= 1;
1300 return __lxc_start(name
, conf
, &start_ops
, &start_arg
, lxcpath
, backgrounded
);
1303 static void lxc_destroy_container_on_signal(struct lxc_handler
*handler
,
1306 char destroy
[MAXPATHLEN
];
1309 if (handler
->conf
&& handler
->conf
->rootfs
.path
&& handler
->conf
->rootfs
.mount
) {
1310 bret
= do_destroy_container(handler
->conf
);
1312 ERROR("Error destroying rootfs for %s", name
);
1316 INFO("Destroyed rootfs for %s", name
);
1318 ret
= snprintf(destroy
, MAXPATHLEN
, "%s/%s", handler
->lxcpath
, name
);
1319 if (ret
< 0 || ret
>= MAXPATHLEN
) {
1320 ERROR("Error printing path for %s", name
);
1321 ERROR("Error destroying directory for %s", name
);
1326 ret
= userns_exec_1(handler
->conf
, lxc_rmdir_onedev_wrapper
, destroy
);
1328 ret
= lxc_rmdir_onedev(destroy
, NULL
);
1331 ERROR("Error destroying directory for %s", name
);
1334 INFO("Destroyed directory for %s", name
);
1337 static int lxc_rmdir_onedev_wrapper(void *data
)
1339 char *arg
= (char *) data
;
1340 return lxc_rmdir_onedev(arg
, NULL
);
1343 static bool do_destroy_container(struct lxc_conf
*conf
) {
1345 if (userns_exec_1(conf
, bdev_destroy_wrapper
, conf
) < 0)
1349 return bdev_destroy(conf
);