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
39 #include <sys/mount.h>
40 #include <sys/param.h>
41 #include <sys/prctl.h>
42 #include <sys/socket.h>
44 #include <sys/syscall.h>
45 #include <sys/types.h>
49 #if HAVE_SYS_CAPABILITY_H
50 #include <sys/capability.h>
53 #if !HAVE_DECL_PR_CAPBSET_DROP
54 #define PR_CAPBSET_DROP 24
66 #include "lxcseccomp.h"
70 #include "namespace.h"
74 #include "bdev/bdev.h"
77 lxc_log_define(lxc_start
, lxc
);
79 const struct ns_info ns_info
[LXC_NS_MAX
] = {
80 [LXC_NS_MNT
] = {"mnt", CLONE_NEWNS
},
81 [LXC_NS_PID
] = {"pid", CLONE_NEWPID
},
82 [LXC_NS_UTS
] = {"uts", CLONE_NEWUTS
},
83 [LXC_NS_IPC
] = {"ipc", CLONE_NEWIPC
},
84 [LXC_NS_USER
] = {"user", CLONE_NEWUSER
},
85 [LXC_NS_NET
] = {"net", CLONE_NEWNET
}
88 extern void mod_all_rdeps(struct lxc_container
*c
, bool inc
);
89 static bool do_destroy_container(struct lxc_conf
*conf
);
90 static int lxc_rmdir_onedev_wrapper(void *data
);
91 static void lxc_destroy_container_on_signal(struct lxc_handler
*handler
,
94 static void print_top_failing_dir(const char *path
)
96 size_t len
= strlen(path
);
97 char *copy
= alloca(len
+ 1), *p
, *e
, saved
;
103 while (p
< e
&& *p
== '/')
105 while (p
< e
&& *p
!= '/')
109 if (access(copy
, X_OK
)) {
110 SYSERROR("could not access %s. Please grant it 'x' " \
111 "access, or add an ACL for the container root.",
119 static void close_ns(int ns_fd
[LXC_NS_MAX
]) {
122 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
131 * preserve_ns: open /proc/@pid/ns/@ns for each namespace specified
133 * Return true on success, false on failure. On failure, leave an error
134 * message in *errmsg, which caller must free.
136 static bool preserve_ns(int ns_fd
[LXC_NS_MAX
], int clone_flags
, pid_t pid
,
140 char path
[MAXPATHLEN
];
142 for (i
= 0; i
< LXC_NS_MAX
; i
++)
145 snprintf(path
, MAXPATHLEN
, "/proc/%d/ns", pid
);
146 if (access(path
, X_OK
)) {
147 if (asprintf(errmsg
, "Kernel does not support setns.") == -1)
152 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
153 if ((clone_flags
& ns_info
[i
].clone_flag
) == 0)
155 snprintf(path
, MAXPATHLEN
, "/proc/%d/ns/%s", pid
,
156 ns_info
[i
].proc_name
);
157 ns_fd
[i
] = open(path
, O_RDONLY
| O_CLOEXEC
);
165 if (errno
== ENOENT
) {
166 ret
= asprintf(errmsg
, "Kernel does not support setns for %s",
167 ns_info
[i
].proc_name
);
169 ret
= asprintf(errmsg
, "Failed to open %s: %s",
170 path
, strerror(errno
));
178 static int attach_ns(const int ns_fd
[LXC_NS_MAX
]) {
181 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
185 if (setns(ns_fd
[i
], 0) != 0)
191 SYSERROR("failed to set namespace '%s'", ns_info
[i
].proc_name
);
195 static int match_fd(int fd
)
197 return (fd
== 0 || fd
== 1 || fd
== 2);
201 * Check for any fds we need to close
202 * * if fd_to_ignore != -1, then if we find that fd open we will ignore it.
203 * * By default we warn about open fds we find.
204 * * If closeall is true, we will close open fds.
205 * * If lxc-start was passed "-C", then conf->close_all_fds will be true,
206 * in which case we also close all open fds.
207 * * A daemonized container will always pass closeall=true.
209 int lxc_check_inherited(struct lxc_conf
*conf
, bool closeall
, int fd_to_ignore
)
211 struct dirent dirent
, *direntp
;
215 if (conf
&& conf
->close_all_fds
)
219 dir
= opendir("/proc/self/fd");
221 WARN("failed to open directory: %m");
227 while (!readdir_r(dir
, &dirent
, &direntp
)) {
231 if (!strcmp(direntp
->d_name
, "."))
234 if (!strcmp(direntp
->d_name
, ".."))
237 fd
= atoi(direntp
->d_name
);
239 if (fd
== fddir
|| fd
== lxc_log_fd
|| fd
== fd_to_ignore
)
242 if (current_config
&& fd
== current_config
->logfd
)
251 INFO("closed inherited fd %d", fd
);
254 WARN("inherited fd %d", fd
);
257 closedir(dir
); /* cannot fail */
261 static int setup_signal_fd(sigset_t
*oldmask
)
266 /* Block everything except serious error signals */
267 if (sigfillset(&mask
) ||
268 sigdelset(&mask
, SIGILL
) ||
269 sigdelset(&mask
, SIGSEGV
) ||
270 sigdelset(&mask
, SIGBUS
) ||
271 sigdelset(&mask
, SIGWINCH
) ||
272 sigprocmask(SIG_BLOCK
, &mask
, oldmask
)) {
273 SYSERROR("failed to set signal mask");
277 fd
= signalfd(-1, &mask
, 0);
279 SYSERROR("failed to create the signal fd");
283 if (fcntl(fd
, F_SETFD
, FD_CLOEXEC
)) {
284 SYSERROR("failed to set sigfd to close-on-exec");
289 DEBUG("sigchild handler set");
294 static int signal_handler(int fd
, uint32_t events
, void *data
,
295 struct lxc_epoll_descr
*descr
)
297 struct signalfd_siginfo siginfo
;
301 bool init_died
= false;
303 ret
= read(fd
, &siginfo
, sizeof(siginfo
));
305 ERROR("failed to read signal info");
309 if (ret
!= sizeof(siginfo
)) {
310 ERROR("unexpected siginfo size");
314 // check whether init is running
316 ret
= waitid(P_PID
, *pid
, &info
, WEXITED
| WNOWAIT
| WNOHANG
);
317 if (ret
== 0 && info
.si_pid
== *pid
) {
321 if (siginfo
.ssi_signo
!= SIGCHLD
) {
322 kill(*pid
, siginfo
.ssi_signo
);
323 INFO("forwarded signal %d to pid %d", siginfo
.ssi_signo
, *pid
);
324 return init_died
? 1 : 0;
327 if (siginfo
.ssi_code
== CLD_STOPPED
||
328 siginfo
.ssi_code
== CLD_CONTINUED
) {
329 INFO("container init process was stopped/continued");
330 return init_died
? 1 : 0;
333 /* more robustness, protect ourself from a SIGCHLD sent
334 * by a process different from the container init
336 if (siginfo
.ssi_pid
!= *pid
) {
337 WARN("invalid pid for SIGCHLD");
338 return init_died
? 1 : 0;
341 DEBUG("container init process exited");
345 int lxc_set_state(const char *name
, struct lxc_handler
*handler
, lxc_state_t state
)
347 handler
->state
= state
;
348 lxc_monitor_send_state(name
, state
, handler
->lxcpath
);
352 int lxc_poll(const char *name
, struct lxc_handler
*handler
)
354 int sigfd
= handler
->sigfd
;
355 int pid
= handler
->pid
;
356 struct lxc_epoll_descr descr
;
358 if (lxc_mainloop_open(&descr
)) {
359 ERROR("failed to create mainloop");
363 if (lxc_mainloop_add_handler(&descr
, sigfd
, signal_handler
, &pid
)) {
364 ERROR("failed to add handler for the signal");
365 goto out_mainloop_open
;
368 if (lxc_console_mainloop_add(&descr
, handler
->conf
)) {
369 ERROR("failed to add console handler to mainloop");
370 goto out_mainloop_open
;
373 if (lxc_cmd_mainloop_add(name
, &descr
, handler
)) {
374 ERROR("failed to add command handler to mainloop");
375 goto out_mainloop_open
;
378 if (handler
->conf
->need_utmp_watch
) {
379 #if HAVE_SYS_CAPABILITY_H
380 if (lxc_utmp_mainloop_add(&descr
, handler
)) {
381 ERROR("failed to add utmp handler to mainloop");
382 goto out_mainloop_open
;
385 DEBUG("not starting utmp handler as cap_sys_boot cannot be dropped without capabilities support");
389 return lxc_mainloop(&descr
, -1);
392 lxc_mainloop_close(&descr
);
398 struct lxc_handler
*lxc_init(const char *name
, struct lxc_conf
*conf
, const char *lxcpath
)
401 struct lxc_handler
*handler
;
403 handler
= malloc(sizeof(*handler
));
407 memset(handler
, 0, sizeof(*handler
));
409 handler
->ttysock
[0] = handler
->ttysock
[1] = -1;
410 handler
->conf
= conf
;
411 handler
->lxcpath
= lxcpath
;
414 for (i
= 0; i
< LXC_NS_MAX
; i
++)
415 handler
->nsfd
[i
] = -1;
419 handler
->name
= strdup(name
);
420 if (!handler
->name
) {
421 ERROR("failed to allocate memory");
425 if (lxc_cmd_init(name
, handler
, lxcpath
))
428 if (lxc_read_seccomp_config(conf
) != 0) {
429 ERROR("failed loading seccomp policy");
430 goto out_close_maincmd_fd
;
433 /* Begin by setting the state to STARTING */
434 if (lxc_set_state(name
, handler
, STARTING
)) {
435 ERROR("failed to set state '%s'", lxc_state2str(STARTING
));
436 goto out_close_maincmd_fd
;
439 /* Start of environment variable setup for hooks */
440 if (name
&& setenv("LXC_NAME", name
, 1)) {
441 SYSERROR("failed to set environment variable for container name");
443 if (conf
->rcfile
&& setenv("LXC_CONFIG_FILE", conf
->rcfile
, 1)) {
444 SYSERROR("failed to set environment variable for config path");
446 if (conf
->rootfs
.mount
&& setenv("LXC_ROOTFS_MOUNT", conf
->rootfs
.mount
, 1)) {
447 SYSERROR("failed to set environment variable for rootfs mount");
449 if (conf
->rootfs
.path
&& setenv("LXC_ROOTFS_PATH", conf
->rootfs
.path
, 1)) {
450 SYSERROR("failed to set environment variable for rootfs mount");
452 if (conf
->console
.path
&& setenv("LXC_CONSOLE", conf
->console
.path
, 1)) {
453 SYSERROR("failed to set environment variable for console path");
455 if (conf
->console
.log_path
&& setenv("LXC_CONSOLE_LOGPATH", conf
->console
.log_path
, 1)) {
456 SYSERROR("failed to set environment variable for console log");
458 if (setenv("LXC_CGNS_AWARE", "1", 1)) {
459 SYSERROR("failed to set LXC_CGNS_AWARE environment variable");
461 /* End of environment variable setup for hooks */
463 if (run_lxc_hooks(name
, "pre-start", conf
, handler
->lxcpath
, NULL
)) {
464 ERROR("failed to run pre-start hooks for container '%s'.", name
);
468 /* the signal fd has to be created before forking otherwise
469 * if the child process exits before we setup the signal fd,
470 * the event will be lost and the command will be stuck */
471 handler
->sigfd
= setup_signal_fd(&handler
->oldmask
);
472 if (handler
->sigfd
< 0) {
473 ERROR("failed to set sigchild fd handler");
477 /* do this after setting up signals since it might unblock SIGWINCH */
478 if (lxc_console_create(conf
)) {
479 ERROR("failed to create console");
480 goto out_restore_sigmask
;
483 if (ttys_shift_ids(conf
) < 0) {
484 ERROR("Failed to shift tty into container");
485 goto out_restore_sigmask
;
488 INFO("'%s' is initialized", name
);
492 sigprocmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
494 lxc_delete_tty(&conf
->tty_info
);
496 lxc_set_state(name
, handler
, ABORTING
);
497 out_close_maincmd_fd
:
498 close(conf
->maincmd_fd
);
499 conf
->maincmd_fd
= -1;
502 handler
->name
= NULL
;
508 void lxc_fini(const char *name
, struct lxc_handler
*handler
)
511 pid_t self
= getpid();
512 char *namespaces
[LXC_NS_MAX
+1];
513 size_t namespace_count
= 0;
515 /* The STOPPING state is there for future cleanup code
516 * which can take awhile
518 lxc_set_state(name
, handler
, STOPPING
);
520 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
521 if (handler
->nsfd
[i
] != -1) {
522 rc
= asprintf(&namespaces
[namespace_count
], "%s:/proc/%d/fd/%d",
523 ns_info
[i
].proc_name
, self
, handler
->nsfd
[i
]);
525 SYSERROR("failed to allocate memory");
531 namespaces
[namespace_count
] = NULL
;
533 if (handler
->conf
->reboot
&& setenv("LXC_TARGET", "reboot", 1)) {
534 SYSERROR("failed to set environment variable for stop target");
536 if (!handler
->conf
->reboot
&& setenv("LXC_TARGET", "stop", 1)) {
537 SYSERROR("failed to set environment variable for stop target");
540 if (run_lxc_hooks(name
, "stop", handler
->conf
, handler
->lxcpath
, namespaces
))
541 ERROR("failed to run stop hooks for container '%s'.", name
);
543 while (namespace_count
--)
544 free(namespaces
[namespace_count
]);
545 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
546 if (handler
->nsfd
[i
] != -1) {
547 close(handler
->nsfd
[i
]);
548 handler
->nsfd
[i
] = -1;
551 lxc_set_state(name
, handler
, STOPPED
);
553 if (run_lxc_hooks(name
, "post-stop", handler
->conf
, handler
->lxcpath
, NULL
)) {
554 ERROR("failed to run post-stop hooks for container '%s'.", name
);
555 if (handler
->conf
->reboot
) {
556 WARN("Container will be stopped instead of rebooted.");
557 handler
->conf
->reboot
= 0;
558 setenv("LXC_TARGET", "stop", 1);
562 /* reset mask set by setup_signal_fd */
563 if (sigprocmask(SIG_SETMASK
, &handler
->oldmask
, NULL
))
564 WARN("failed to restore sigprocmask");
566 lxc_console_delete(&handler
->conf
->console
);
567 lxc_delete_tty(&handler
->conf
->tty_info
);
568 close(handler
->conf
->maincmd_fd
);
569 handler
->conf
->maincmd_fd
= -1;
571 if (handler
->ttysock
[0] != -1) {
572 close(handler
->ttysock
[0]);
573 close(handler
->ttysock
[1]);
576 if (handler
->conf
->ephemeral
== 1 && handler
->conf
->reboot
!= 1)
577 lxc_destroy_container_on_signal(handler
, name
);
579 cgroup_destroy(handler
);
583 void lxc_abort(const char *name
, struct lxc_handler
*handler
)
587 lxc_set_state(name
, handler
, ABORTING
);
588 if (handler
->pid
> 0)
589 kill(handler
->pid
, SIGKILL
);
590 while ((ret
= waitpid(-1, &status
, 0)) > 0) ;
593 #include <sys/reboot.h>
594 #include <linux/reboot.h>
597 * reboot(LINUX_REBOOT_CMD_CAD_ON) will return -EINVAL
598 * in a child pid namespace if container reboot support exists.
599 * Otherwise, it will either succeed or return -EPERM.
601 static int container_reboot_supported(void *arg
)
607 if (ret
== -1 && errno
== EINVAL
)
612 static int must_drop_cap_sys_boot(struct lxc_conf
*conf
)
615 int ret
, cmd
, v
, flags
;
616 long stack_size
= 4096;
617 void *stack
= alloca(stack_size
);
621 f
= fopen("/proc/sys/kernel/ctrl-alt-del", "r");
623 DEBUG("failed to open /proc/sys/kernel/ctrl-alt-del");
627 ret
= fscanf(f
, "%d", &v
);
630 DEBUG("Failed to read /proc/sys/kernel/ctrl-alt-del");
633 cmd
= v
? LINUX_REBOOT_CMD_CAD_ON
: LINUX_REBOOT_CMD_CAD_OFF
;
635 flags
= CLONE_NEWPID
| SIGCHLD
;
636 if (!lxc_list_empty(&conf
->id_map
))
637 flags
|= CLONE_NEWUSER
;
640 pid
= __clone2(container_reboot_supported
, stack
, stack_size
, flags
, &cmd
);
643 pid
= clone(container_reboot_supported
, stack
, flags
, &cmd
);
646 if (flags
& CLONE_NEWUSER
)
647 ERROR("failed to clone (%#x): %s (includes CLONE_NEWUSER)", flags
, strerror(errno
));
649 ERROR("failed to clone (%#x): %s", flags
, strerror(errno
));
652 if (wait(&status
) < 0) {
653 SYSERROR("unexpected wait error: %m");
657 if (WEXITSTATUS(status
) != 1)
664 * netpipe is used in the unprivileged case to transfer the ifindexes
665 * from parent to child
667 static int netpipe
= -1;
669 static inline int count_veths(struct lxc_list
*network
)
671 struct lxc_list
*iterator
;
672 struct lxc_netdev
*netdev
;
675 lxc_list_for_each(iterator
, network
) {
676 netdev
= iterator
->elem
;
677 if (netdev
->type
!= LXC_NET_VETH
)
684 static int read_unpriv_netifindex(struct lxc_list
*network
)
686 struct lxc_list
*iterator
;
687 struct lxc_netdev
*netdev
;
691 lxc_list_for_each(iterator
, network
) {
692 netdev
= iterator
->elem
;
693 if (netdev
->type
!= LXC_NET_VETH
)
695 if (!(netdev
->name
= malloc(IFNAMSIZ
))) {
696 ERROR("Out of memory");
700 if (read(netpipe
, netdev
->name
, IFNAMSIZ
) != IFNAMSIZ
) {
709 static int do_start(void *data
)
711 struct lxc_list
*iterator
;
712 struct lxc_handler
*handler
= data
;
713 int devnull_fd
= -1, ret
;
716 if (sigprocmask(SIG_SETMASK
, &handler
->oldmask
, NULL
)) {
717 SYSERROR("failed to set sigprocmask");
721 /* This prctl must be before the synchro, so if the parent
722 * dies before we set the parent death signal, we will detect
723 * its death with the synchro right after, otherwise we have
724 * a window where the parent can exit before we set the pdeath
725 * signal leading to a unsupervized container.
727 if (prctl(PR_SET_PDEATHSIG
, SIGKILL
, 0, 0, 0)) {
728 SYSERROR("failed to set pdeath signal");
732 lxc_sync_fini_parent(handler
);
734 /* don't leak the pinfd to the container */
735 if (handler
->pinfd
>= 0) {
736 close(handler
->pinfd
);
739 /* Tell the parent task it can begin to configure the
740 * container and wait for it to finish
742 if (lxc_sync_barrier_parent(handler
, LXC_SYNC_CONFIGURE
))
745 if (read_unpriv_netifindex(&handler
->conf
->network
) < 0)
746 goto out_warn_father
;
749 * if we are in a new user namespace, become root there to have
750 * privilege over our namespace. When using lxc-execute we default to root,
751 * but this can be overriden using the lxc.init_uid and lxc.init_gid
752 * configuration options.
754 if (!lxc_list_empty(&handler
->conf
->id_map
)) {
756 if (handler
->conf
->is_execute
&& handler
->conf
->init_gid
)
757 new_gid
= handler
->conf
->init_gid
;
760 if (handler
->conf
->is_execute
&& handler
->conf
->init_uid
)
761 new_uid
= handler
->conf
->init_uid
;
763 NOTICE("switching to gid/uid %d/%d in new user namespace", new_gid
, new_uid
);
764 if (setgid(new_gid
)) {
766 goto out_warn_father
;
768 if (setuid(new_uid
)) {
770 goto out_warn_father
;
772 if (setgroups(0, NULL
)) {
773 SYSERROR("setgroups");
774 goto out_warn_father
;
778 if (access(handler
->lxcpath
, X_OK
)) {
779 print_top_failing_dir(handler
->lxcpath
);
780 goto out_warn_father
;
783 #if HAVE_SYS_CAPABILITY_H
784 if (handler
->conf
->need_utmp_watch
) {
785 if (prctl(PR_CAPBSET_DROP
, CAP_SYS_BOOT
, 0, 0, 0)) {
786 SYSERROR("failed to remove CAP_SYS_BOOT capability");
787 goto out_warn_father
;
789 DEBUG("Dropped cap_sys_boot");
793 ret
= snprintf(path
, sizeof(path
), "%s/dev/null", handler
->conf
->rootfs
.mount
);
794 if (ret
< 0 || ret
>= sizeof(path
)) {
795 SYSERROR("sprintf'd too many chars");
796 goto out_warn_father
;
799 /* In order to checkpoint restore, we need to have everything in the
800 * same mount namespace. However, some containers may not have a
801 * reasonable /dev (in particular, they may not have /dev/null), so we
802 * can't set init's std fds to /dev/null by opening it from inside the
805 * If that's the case, fall back to using the host's /dev/null. This
806 * means that migration won't work, but at least we won't spew output
807 * where it isn't wanted.
809 if (handler
->backgrounded
&& !handler
->conf
->autodev
&& access(path
, F_OK
) < 0) {
810 devnull_fd
= open_devnull();
813 goto out_warn_father
;
814 WARN("using host's /dev/null for container init's std fds, migraiton won't work");
817 /* Setup the container, ip, names, utsname, ... */
818 if (lxc_setup(handler
)) {
819 ERROR("failed to setup the container");
820 goto out_warn_father
;
823 /* ask father to setup cgroups and wait for him to finish */
824 if (lxc_sync_barrier_parent(handler
, LXC_SYNC_CGROUP
))
827 /* Set the label to change to when we exec(2) the container's init */
828 if (lsm_process_label_set(NULL
, handler
->conf
, 1, 1) < 0)
829 goto out_warn_father
;
831 /* Some init's such as busybox will set sane tty settings on stdin,
832 * stdout, stderr which it thinks is the console. We already set them
833 * the way we wanted on the real terminal, and we want init to do its
834 * setup on its console ie. the pty allocated in lxc_console_create()
835 * so make sure that that pty is stdin,stdout,stderr.
837 if (lxc_console_set_stdfds(handler
->conf
->console
.slave
) < 0)
838 goto out_warn_father
;
840 /* If we mounted a temporary proc, then unmount it now */
841 tmp_proc_unmount(handler
->conf
);
843 if (lxc_seccomp_load(handler
->conf
) != 0)
844 goto out_warn_father
;
846 if (run_lxc_hooks(handler
->name
, "start", handler
->conf
, handler
->lxcpath
, NULL
)) {
847 ERROR("failed to run start hooks for container '%s'.", handler
->name
);
848 goto out_warn_father
;
851 /* The clearenv() and putenv() calls have been moved here
852 * to allow us to use environment variables passed to the various
853 * hooks, such as the start hook above. Not all of the
854 * variables like CONFIG_PATH or ROOTFS are valid in this
855 * context but others are. */
857 SYSERROR("failed to clear environment");
858 /* don't error out though */
861 lxc_list_for_each(iterator
, &handler
->conf
->environment
) {
862 if (putenv((char *)iterator
->elem
)) {
863 SYSERROR("failed to set environment variable '%s'", (char *)iterator
->elem
);
864 goto out_warn_father
;
868 if (putenv("container=lxc")) {
869 SYSERROR("failed to set environment variable 'container=lxc'");
870 goto out_warn_father
;
873 if (handler
->conf
->pty_names
) {
874 if (putenv(handler
->conf
->pty_names
)) {
875 SYSERROR("failed to set environment variable for container ptys");
876 goto out_warn_father
;
880 close(handler
->sigfd
);
882 if (devnull_fd
< 0) {
883 devnull_fd
= open_devnull();
886 goto out_warn_father
;
889 if (handler
->backgrounded
&& set_stdfds(devnull_fd
))
890 goto out_warn_father
;
892 if (devnull_fd
>= 0) {
897 if (cgns_supported() && unshare(CLONE_NEWCGROUP
) != 0) {
898 SYSERROR("Failed to unshare cgroup namespace");
899 goto out_warn_father
;
904 /* after this call, we are in error because this
905 * ops should not return as it execs */
906 handler
->ops
->start(handler
, handler
->data
);
909 /* we want the parent to know something went wrong, so we return a special
911 lxc_sync_wake_parent(handler
, LXC_SYNC_ERROR
);
920 static int save_phys_nics(struct lxc_conf
*conf
)
922 struct lxc_list
*iterator
;
923 int am_root
= (getuid() == 0);
928 lxc_list_for_each(iterator
, &conf
->network
) {
929 struct lxc_netdev
*netdev
= iterator
->elem
;
931 if (netdev
->type
!= LXC_NET_PHYS
)
933 conf
->saved_nics
= realloc(conf
->saved_nics
,
934 (conf
->num_savednics
+1)*sizeof(struct saved_nic
));
935 if (!conf
->saved_nics
) {
936 SYSERROR("failed to allocate memory");
939 conf
->saved_nics
[conf
->num_savednics
].ifindex
= netdev
->ifindex
;
940 conf
->saved_nics
[conf
->num_savednics
].orig_name
= strdup(netdev
->link
);
941 if (!conf
->saved_nics
[conf
->num_savednics
].orig_name
) {
942 SYSERROR("failed to allocate memory");
945 INFO("stored saved_nic #%d idx %d name %s", conf
->num_savednics
,
946 conf
->saved_nics
[conf
->num_savednics
].ifindex
,
947 conf
->saved_nics
[conf
->num_savednics
].orig_name
);
948 conf
->num_savednics
++;
954 static int recv_fd(int sock
, int *fd
)
956 if (lxc_abstract_unix_recv_fd(sock
, fd
, NULL
, 0) < 0) {
957 SYSERROR("Error receiving tty fd from child");
965 static int recv_ttys_from_child(struct lxc_handler
*handler
)
967 struct lxc_conf
*conf
= handler
->conf
;
968 int i
, sock
= handler
->ttysock
[1];
969 struct lxc_tty_info
*tty_info
= &conf
->tty_info
;
974 tty_info
->pty_info
= malloc(sizeof(*tty_info
->pty_info
)*conf
->tty
);
975 if (!tty_info
->pty_info
) {
976 SYSERROR("failed to allocate pty_info");
980 for (i
= 0; i
< conf
->tty
; i
++) {
981 struct lxc_pty_info
*pty_info
= &tty_info
->pty_info
[i
];
983 if (recv_fd(sock
, &pty_info
->slave
) < 0 ||
984 recv_fd(sock
, &pty_info
->master
) < 0) {
985 ERROR("Error receiving tty info from child");
989 tty_info
->nbtty
= conf
->tty
;
994 void resolve_clone_flags(struct lxc_handler
*handler
)
996 handler
->clone_flags
= CLONE_NEWPID
| CLONE_NEWNS
;
998 if (!lxc_list_empty(&handler
->conf
->id_map
)) {
999 INFO("Cloning a new user namespace");
1000 handler
->clone_flags
|= CLONE_NEWUSER
;
1003 if (handler
->conf
->inherit_ns_fd
[LXC_NS_NET
] == -1) {
1004 if (!lxc_requests_empty_network(handler
))
1005 handler
->clone_flags
|= CLONE_NEWNET
;
1007 INFO("Inheriting a net namespace");
1010 if (handler
->conf
->inherit_ns_fd
[LXC_NS_IPC
] == -1) {
1011 handler
->clone_flags
|= CLONE_NEWIPC
;
1013 INFO("Inheriting an IPC namespace");
1016 if (handler
->conf
->inherit_ns_fd
[LXC_NS_UTS
] == -1) {
1017 handler
->clone_flags
|= CLONE_NEWUTS
;
1019 INFO("Inheriting a UTS namespace");
1023 static int lxc_spawn(struct lxc_handler
*handler
)
1025 int failed_before_rename
= 0;
1026 const char *name
= handler
->name
;
1027 char *errmsg
= NULL
;
1028 bool cgroups_connected
= false;
1029 int saved_ns_fd
[LXC_NS_MAX
];
1030 int preserve_mask
= 0, i
;
1031 int netpipepair
[2], nveths
;
1035 for (i
= 0; i
< LXC_NS_MAX
; i
++)
1036 if (handler
->conf
->inherit_ns_fd
[i
] != -1)
1037 preserve_mask
|= ns_info
[i
].clone_flag
;
1039 if (lxc_sync_init(handler
))
1042 if (socketpair(AF_UNIX
, SOCK_DGRAM
, 0, handler
->ttysock
) < 0) {
1043 lxc_sync_fini(handler
);
1047 resolve_clone_flags(handler
);
1049 if (handler
->clone_flags
& CLONE_NEWNET
) {
1050 if (!lxc_list_empty(&handler
->conf
->network
)) {
1052 /* Find gateway addresses from the link device, which is
1053 * no longer accessible inside the container. Do this
1054 * before creating network interfaces, since goto
1055 * out_delete_net does not work before lxc_clone. */
1056 if (lxc_find_gateway_addresses(handler
)) {
1057 ERROR("failed to find gateway addresses");
1058 lxc_sync_fini(handler
);
1062 /* that should be done before the clone because we will
1063 * fill the netdev index and use them in the child
1065 if (lxc_create_network(handler
)) {
1066 ERROR("failed to create the network");
1067 lxc_sync_fini(handler
);
1072 if (save_phys_nics(handler
->conf
)) {
1073 ERROR("failed to save physical nic info");
1078 if (!cgroup_init(handler
)) {
1079 ERROR("failed initializing cgroup support");
1080 goto out_delete_net
;
1083 cgroups_connected
= true;
1085 if (!cgroup_create(handler
)) {
1086 ERROR("failed creating cgroups");
1087 goto out_delete_net
;
1091 * if the rootfs is not a blockdev, prevent the container from
1092 * marking it readonly.
1094 * if the container is unprivileged then skip rootfs pinning
1096 if (lxc_list_empty(&handler
->conf
->id_map
)) {
1097 handler
->pinfd
= pin_rootfs(handler
->conf
->rootfs
.path
);
1098 if (handler
->pinfd
== -1)
1099 INFO("failed to pin the container's rootfs");
1102 if (!preserve_ns(saved_ns_fd
, preserve_mask
, getpid(), &errmsg
)) {
1103 SYSERROR("Failed to preserve requested namespaces: %s",
1104 errmsg
? errmsg
: "(Out of memory)");
1106 goto out_delete_net
;
1108 if (attach_ns(handler
->conf
->inherit_ns_fd
) < 0)
1109 goto out_delete_net
;
1111 if (am_unpriv() && (nveths
= count_veths(&handler
->conf
->network
))) {
1112 if (pipe(netpipepair
) < 0) {
1113 SYSERROR("Error creating pipe");
1114 goto out_delete_net
;
1116 /* store netpipe in the global var for do_start's use */
1117 netpipe
= netpipepair
[0];
1120 /* Create a process in a new set of namespaces */
1121 handler
->pid
= lxc_clone(do_start
, handler
, handler
->clone_flags
);
1122 if (handler
->pid
< 0) {
1123 SYSERROR("failed to fork into a new namespace");
1124 goto out_delete_net
;
1127 if (!preserve_ns(handler
->nsfd
, handler
->clone_flags
| preserve_mask
, handler
->pid
, &errmsg
)) {
1128 INFO("Failed to store namespace references for stop hook: %s",
1129 errmsg
? errmsg
: "(Out of memory)");
1133 if (attach_ns(saved_ns_fd
))
1134 WARN("failed to restore saved namespaces");
1136 lxc_sync_fini_child(handler
);
1138 if (lxc_sync_wait_child(handler
, LXC_SYNC_CONFIGURE
))
1139 failed_before_rename
= 1;
1141 if (!cgroup_create_legacy(handler
)) {
1142 ERROR("failed to setup the legacy cgroups for %s", name
);
1143 goto out_delete_net
;
1145 if (!cgroup_setup_limits(handler
, false)) {
1146 ERROR("failed to setup the cgroup limits for '%s'", name
);
1147 goto out_delete_net
;
1150 if (!cgroup_enter(handler
))
1151 goto out_delete_net
;
1153 if (!cgroup_chown(handler
))
1154 goto out_delete_net
;
1156 if (failed_before_rename
)
1157 goto out_delete_net
;
1159 /* Create the network configuration */
1160 if (handler
->clone_flags
& CLONE_NEWNET
) {
1161 if (lxc_assign_network(handler
->lxcpath
, handler
->name
,
1162 &handler
->conf
->network
, handler
->pid
)) {
1163 ERROR("failed to create the configured network");
1164 goto out_delete_net
;
1168 if (netpipe
!= -1) {
1169 struct lxc_list
*iterator
;
1170 struct lxc_netdev
*netdev
;
1173 lxc_list_for_each(iterator
, &handler
->conf
->network
) {
1174 netdev
= iterator
->elem
;
1175 if (netdev
->type
!= LXC_NET_VETH
)
1177 if (write(netpipepair
[1], netdev
->name
, IFNAMSIZ
) != IFNAMSIZ
) {
1178 ERROR("Error writing veth name to container");
1179 goto out_delete_net
;
1182 close(netpipepair
[1]);
1185 /* map the container uids - the container became an invalid
1186 * userid the moment it was cloned with CLONE_NEWUSER - this
1187 * call doesn't change anything immediately, but allows the
1188 * container to setuid(0) (0 being mapped to something else on
1189 * the host) later to become a valid uid again */
1190 if (lxc_map_ids(&handler
->conf
->id_map
, handler
->pid
)) {
1191 ERROR("failed to set up id mapping");
1192 goto out_delete_net
;
1195 /* Tell the child to continue its initialization. we'll get
1196 * LXC_SYNC_CGROUP when it is ready for us to setup cgroups
1198 if (lxc_sync_barrier_child(handler
, LXC_SYNC_POST_CONFIGURE
))
1199 goto out_delete_net
;
1201 if (!cgroup_setup_limits(handler
, true)) {
1202 ERROR("failed to setup the devices cgroup for '%s'", name
);
1203 goto out_delete_net
;
1206 cgroup_disconnect();
1207 cgroups_connected
= false;
1209 /* read tty fds allocated by child */
1210 if (recv_ttys_from_child(handler
) < 0) {
1211 ERROR("failed to receive tty info from child");
1212 goto out_delete_net
;
1215 /* Tell the child to complete its initialization and wait for
1216 * it to exec or return an error. (the child will never
1217 * return LXC_SYNC_POST_CGROUP+1. It will either close the
1218 * sync pipe, causing lxc_sync_barrier_child to return
1219 * success, or return a different value, causing us to error
1222 if (lxc_sync_barrier_child(handler
, LXC_SYNC_POST_CGROUP
))
1225 if (detect_shared_rootfs())
1226 umount2(handler
->conf
->rootfs
.mount
, MNT_DETACH
);
1228 if (handler
->ops
->post_start(handler
, handler
->data
))
1231 if (lxc_set_state(name
, handler
, RUNNING
)) {
1232 ERROR("failed to set state to %s",
1233 lxc_state2str(RUNNING
));
1237 lxc_sync_fini(handler
);
1242 if (cgroups_connected
)
1243 cgroup_disconnect();
1244 if (handler
->clone_flags
& CLONE_NEWNET
)
1245 lxc_delete_network(handler
);
1247 lxc_abort(name
, handler
);
1248 lxc_sync_fini(handler
);
1249 if (handler
->pinfd
>= 0) {
1250 close(handler
->pinfd
);
1251 handler
->pinfd
= -1;
1257 int get_netns_fd(int pid
)
1259 char path
[MAXPATHLEN
];
1262 ret
= snprintf(path
, MAXPATHLEN
, "/proc/%d/ns/net", pid
);
1263 if (ret
< 0 || ret
>= MAXPATHLEN
) {
1264 WARN("Failed to pin netns file for pid %d", pid
);
1268 fd
= open(path
, O_RDONLY
);
1270 WARN("Failed to pin netns file %s for pid %d: %s",
1271 path
, pid
, strerror(errno
));
1277 int __lxc_start(const char *name
, struct lxc_conf
*conf
,
1278 struct lxc_operations
* ops
, void *data
, const char *lxcpath
,
1281 struct lxc_handler
*handler
;
1286 handler
= lxc_init(name
, conf
, lxcpath
);
1288 ERROR("failed to initialize the container");
1292 handler
->data
= data
;
1293 handler
->backgrounded
= backgrounded
;
1295 if (must_drop_cap_sys_boot(handler
->conf
)) {
1296 #if HAVE_SYS_CAPABILITY_H
1297 DEBUG("Dropping cap_sys_boot");
1299 DEBUG("Can't drop cap_sys_boot as capabilities aren't supported");
1302 DEBUG("Not dropping cap_sys_boot or watching utmp");
1303 handler
->conf
->need_utmp_watch
= 0;
1306 if (!attach_block_device(handler
->conf
)) {
1307 ERROR("Failure attaching block device");
1308 goto out_fini_nonet
;
1311 if (geteuid() == 0 && !lxc_list_empty(&conf
->id_map
)) {
1312 /* if the backing store is a device, mount it here and now */
1313 if (rootfs_is_blockdev(conf
)) {
1314 if (unshare(CLONE_NEWNS
) < 0) {
1315 ERROR("Error unsharing mounts");
1316 goto out_fini_nonet
;
1318 remount_all_slave();
1319 if (do_rootfs_setup(conf
, name
, lxcpath
) < 0) {
1320 ERROR("Error setting up rootfs mount as root before spawn");
1321 goto out_fini_nonet
;
1323 INFO("Set up container rootfs as host root");
1327 err
= lxc_spawn(handler
);
1329 ERROR("failed to spawn '%s'", name
);
1330 goto out_detach_blockdev
;
1333 handler
->conf
->reboot
= 0;
1335 netnsfd
= get_netns_fd(handler
->pid
);
1337 err
= lxc_poll(name
, handler
);
1339 ERROR("mainloop exited with an error");
1345 while (waitpid(handler
->pid
, &status
, 0) < 0 && errno
== EINTR
)
1349 * If the child process exited but was not signaled,
1350 * it didn't call reboot. This should mean it was an
1351 * lxc-execute which simply exited. In any case, treat
1354 if (WIFSIGNALED(status
)) {
1355 switch(WTERMSIG(status
)) {
1356 case SIGINT
: /* halt */
1357 DEBUG("Container halting");
1359 case SIGHUP
: /* reboot */
1360 DEBUG("Container rebooting");
1361 handler
->conf
->reboot
= 1;
1363 case SIGSYS
: /* seccomp */
1364 DEBUG("Container violated its seccomp policy");
1367 DEBUG("unknown exit status for init: %d", WTERMSIG(status
));
1372 DEBUG("Pushing physical nics back to host namespace");
1373 lxc_rename_phys_nics_on_shutdown(netnsfd
, handler
->conf
);
1375 DEBUG("Tearing down virtual network devices used by container");
1376 lxc_delete_network(handler
);
1381 if (handler
->pinfd
>= 0) {
1382 close(handler
->pinfd
);
1383 handler
->pinfd
= -1;
1386 lxc_monitor_send_exit_code(name
, status
, handler
->lxcpath
);
1387 err
= lxc_error_set_and_log(handler
->pid
, status
);
1389 lxc_delete_network(handler
);
1391 out_detach_blockdev
:
1392 detach_block_device(handler
->conf
);
1395 lxc_fini(name
, handler
);
1399 lxc_abort(name
, handler
);
1407 static int start(struct lxc_handler
*handler
, void* data
)
1409 struct start_args
*arg
= data
;
1411 NOTICE("exec'ing '%s'", arg
->argv
[0]);
1413 execvp(arg
->argv
[0], arg
->argv
);
1414 SYSERROR("failed to exec %s", arg
->argv
[0]);
1418 static int post_start(struct lxc_handler
*handler
, void* data
)
1420 struct start_args
*arg
= data
;
1422 NOTICE("'%s' started with pid '%d'", arg
->argv
[0], handler
->pid
);
1426 static struct lxc_operations start_ops
= {
1428 .post_start
= post_start
1431 int lxc_start(const char *name
, char *const argv
[], struct lxc_conf
*conf
,
1432 const char *lxcpath
, bool backgrounded
)
1434 struct start_args start_arg
= {
1438 conf
->need_utmp_watch
= 1;
1439 return __lxc_start(name
, conf
, &start_ops
, &start_arg
, lxcpath
, backgrounded
);
1442 static void lxc_destroy_container_on_signal(struct lxc_handler
*handler
,
1445 char destroy
[MAXPATHLEN
];
1448 struct lxc_container
*c
;
1449 if (handler
->conf
->rootfs
.path
&& handler
->conf
->rootfs
.mount
) {
1450 bret
= do_destroy_container(handler
->conf
);
1452 ERROR("Error destroying rootfs for %s", name
);
1456 INFO("Destroyed rootfs for %s", name
);
1458 ret
= snprintf(destroy
, MAXPATHLEN
, "%s/%s", handler
->lxcpath
, name
);
1459 if (ret
< 0 || ret
>= MAXPATHLEN
) {
1460 ERROR("Error printing path for %s", name
);
1461 ERROR("Error destroying directory for %s", name
);
1465 c
= lxc_container_new(name
, handler
->lxcpath
);
1467 if (container_disk_lock(c
)) {
1468 INFO("Could not update lxc_snapshots file");
1469 lxc_container_put(c
);
1471 mod_all_rdeps(c
, false);
1472 container_disk_unlock(c
);
1473 lxc_container_put(c
);
1478 ret
= userns_exec_1(handler
->conf
, lxc_rmdir_onedev_wrapper
, destroy
);
1480 ret
= lxc_rmdir_onedev(destroy
, NULL
);
1483 ERROR("Error destroying directory for %s", name
);
1486 INFO("Destroyed directory for %s", name
);
1489 static int lxc_rmdir_onedev_wrapper(void *data
)
1491 char *arg
= (char *) data
;
1492 return lxc_rmdir_onedev(arg
, NULL
);
1495 static bool do_destroy_container(struct lxc_conf
*conf
) {
1497 if (userns_exec_1(conf
, bdev_destroy_wrapper
, conf
) < 0)
1501 return bdev_destroy(conf
);