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>
50 #include <sys/capability.h>
53 #ifndef HAVE_DECL_PR_CAPBSET_DROP
54 #define PR_CAPBSET_DROP 24
57 #ifndef HAVE_DECL_PR_SET_NO_NEW_PRIVS
58 #define PR_SET_NO_NEW_PRIVS 38
61 #ifndef HAVE_DECL_PR_GET_NO_NEW_PRIVS
62 #define PR_GET_NO_NEW_PRIVS 39
75 #include "lxcseccomp.h"
79 #include "namespace.h"
85 lxc_log_define(lxc_start
, lxc
);
87 extern void mod_all_rdeps(struct lxc_container
*c
, bool inc
);
88 static bool do_destroy_container(struct lxc_conf
*conf
);
89 static int lxc_rmdir_onedev_wrapper(void *data
);
90 static void lxc_destroy_container_on_signal(struct lxc_handler
*handler
,
93 static void print_top_failing_dir(const char *path
)
95 size_t len
= strlen(path
);
96 char *copy
= alloca(len
+ 1), *p
, *e
, saved
;
102 while (p
< e
&& *p
== '/')
104 while (p
< e
&& *p
!= '/')
108 if (access(copy
, X_OK
)) {
109 SYSERROR("Could not access %s. Please grant it x "
110 "access, or add an ACL for the container "
118 static void close_ns(int ns_fd
[LXC_NS_MAX
])
122 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
130 /* preserve_ns: open /proc/@pid/ns/@ns for each namespace specified
132 * Return true on success, false on failure.
134 static bool preserve_ns(int ns_fd
[LXC_NS_MAX
], int clone_flags
, pid_t pid
)
138 for (i
= 0; i
< LXC_NS_MAX
; i
++)
141 ret
= lxc_preserve_ns(pid
, "");
143 SYSERROR("Kernel does not support attaching to namespaces.");
149 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
150 if ((clone_flags
& ns_info
[i
].clone_flag
) == 0)
152 ns_fd
[i
] = lxc_preserve_ns(pid
, ns_info
[i
].proc_name
);
161 SYSERROR("Kernel does not support attaching to %s namespaces.", ns_info
[i
].proc_name
);
163 SYSERROR("Failed to open file descriptor for %s namespace: %s.", ns_info
[i
].proc_name
, strerror(errno
));
168 static int attach_ns(const int ns_fd
[LXC_NS_MAX
]) {
171 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
175 if (setns(ns_fd
[i
], 0) != 0)
181 SYSERROR("Failed to attach %s namespace.", ns_info
[i
].proc_name
);
185 static int match_fd(int fd
)
187 return (fd
== 0 || fd
== 1 || fd
== 2);
190 /* Check for any fds we need to close.
191 * - If fd_to_ignore != -1, then if we find that fd open we will ignore it.
192 * - By default we warn about open fds we find.
193 * - If closeall is true, we will close open fds.
194 * - If lxc-start was passed "-C", then conf->close_all_fds will be true, in
195 * which case we also close all open fds.
196 * - A daemonized container will always pass closeall=true.
198 int lxc_check_inherited(struct lxc_conf
*conf
, bool closeall
, int fd_to_ignore
)
200 struct dirent
*direntp
;
204 if (conf
&& conf
->close_all_fds
)
208 dir
= opendir("/proc/self/fd");
210 WARN("Failed to open directory: %m.");
216 while ((direntp
= readdir(dir
))) {
220 if (!strcmp(direntp
->d_name
, "."))
223 if (!strcmp(direntp
->d_name
, ".."))
226 if (lxc_safe_int(direntp
->d_name
, &fd
) < 0) {
227 INFO("Could not parse file descriptor for: %s", direntp
->d_name
);
231 if (fd
== fddir
|| fd
== lxc_log_fd
|| fd
== fd_to_ignore
)
234 if (current_config
&& fd
== current_config
->logfd
)
243 INFO("Closed inherited fd: %d.", fd
);
246 WARN("Inherited fd: %d.", fd
);
249 /* Only enable syslog at this point to avoid the above logging function
250 * to open a new fd and make the check_inherited function enter an
253 lxc_log_enable_syslog();
255 closedir(dir
); /* cannot fail */
259 static int setup_signal_fd(sigset_t
*oldmask
)
264 /* Block everything except serious error signals. */
265 if (sigfillset(&mask
) ||
266 sigdelset(&mask
, SIGILL
) ||
267 sigdelset(&mask
, SIGSEGV
) ||
268 sigdelset(&mask
, SIGBUS
) ||
269 sigdelset(&mask
, SIGWINCH
) ||
270 sigprocmask(SIG_BLOCK
, &mask
, oldmask
)) {
271 SYSERROR("Failed to set signal mask.");
275 fd
= signalfd(-1, &mask
, 0);
277 SYSERROR("Failed to create signal file descriptor.");
281 if (fcntl(fd
, F_SETFD
, FD_CLOEXEC
)) {
282 SYSERROR("Failed to set FD_CLOEXEC on the signal file descriptor: %d.", fd
);
287 DEBUG("Set SIGCHLD handler with file descriptor: %d.", fd
);
292 static int signal_handler(int fd
, uint32_t events
, void *data
,
293 struct lxc_epoll_descr
*descr
)
295 struct signalfd_siginfo siginfo
;
299 bool init_died
= false;
301 ret
= read(fd
, &siginfo
, sizeof(siginfo
));
303 ERROR("Failed to read signal info from signal file descriptor: %d.", fd
);
307 if (ret
!= sizeof(siginfo
)) {
308 ERROR("Unexpected size for siginfo struct.");
312 /* Check whether init is running. */
314 ret
= waitid(P_PID
, *pid
, &info
, WEXITED
| WNOWAIT
| WNOHANG
);
315 if (ret
== 0 && info
.si_pid
== *pid
)
318 if (siginfo
.ssi_signo
!= SIGCHLD
) {
319 kill(*pid
, siginfo
.ssi_signo
);
320 INFO("Forwarded signal %d to pid %d.", siginfo
.ssi_signo
, *pid
);
321 return init_died
? 1 : 0;
324 if (siginfo
.ssi_code
== CLD_STOPPED
) {
325 INFO("Container init process was stopped.");
326 return init_died
? 1 : 0;
327 } else if (siginfo
.ssi_code
== CLD_CONTINUED
) {
328 INFO("Container init process was continued.");
329 return init_died
? 1 : 0;
332 /* More robustness, protect ourself from a SIGCHLD sent
333 * by a process different from the container init.
335 if (siginfo
.ssi_pid
!= *pid
) {
336 NOTICE("Received SIGCHLD from pid %d instead of container init %d.", siginfo
.ssi_pid
, *pid
);
337 return init_died
? 1 : 0;
340 DEBUG("Container init process %d exited.", *pid
);
344 int lxc_set_state(const char *name
, struct lxc_handler
*handler
, lxc_state_t state
)
346 handler
->state
= state
;
347 lxc_monitor_send_state(name
, state
, handler
->lxcpath
);
351 int lxc_poll(const char *name
, struct lxc_handler
*handler
)
353 int sigfd
= handler
->sigfd
;
354 int pid
= handler
->pid
;
355 struct lxc_epoll_descr descr
;
357 if (lxc_mainloop_open(&descr
)) {
358 ERROR("Failed to create LXC mainloop.");
362 if (lxc_mainloop_add_handler(&descr
, sigfd
, signal_handler
, &pid
)) {
363 ERROR("Failed to add signal handler with file descriptor %d to LXC mainloop.", sigfd
);
364 goto out_mainloop_open
;
367 if (lxc_console_mainloop_add(&descr
, handler
->conf
)) {
368 ERROR("Failed to add console handler to LXC mainloop.");
369 goto out_mainloop_open
;
372 if (lxc_cmd_mainloop_add(name
, &descr
, handler
)) {
373 ERROR("Failed to add command handler to LXC mainloop.");
374 goto out_mainloop_open
;
377 if (handler
->conf
->need_utmp_watch
) {
379 if (lxc_utmp_mainloop_add(&descr
, handler
)) {
380 ERROR("Failed to add utmp handler to LXC mainloop.");
381 goto out_mainloop_open
;
384 DEBUG("Not starting utmp handler as CAP_SYS_BOOT cannot be dropped without capabilities support.");
387 TRACE("lxc mainloop is ready");
389 return lxc_mainloop(&descr
, -1);
392 lxc_mainloop_close(&descr
);
400 struct lxc_handler
*lxc_init_handler(const char *name
, struct lxc_conf
*conf
,
404 struct lxc_handler
*handler
;
406 handler
= malloc(sizeof(*handler
));
408 ERROR("failed to allocate memory");
412 memset(handler
, 0, sizeof(*handler
));
414 handler
->ttysock
[0] = handler
->ttysock
[1] = -1;
415 handler
->conf
= conf
;
416 handler
->lxcpath
= lxcpath
;
419 for (i
= 0; i
< LXC_NS_MAX
; i
++)
420 handler
->nsfd
[i
] = -1;
422 handler
->name
= strdup(name
);
423 if (!handler
->name
) {
424 ERROR("failed to allocate memory");
425 goto do_partial_cleanup
;
428 if (lxc_cmd_init(name
, handler
, lxcpath
)) {
429 ERROR("failed to set up command socket");
430 goto do_full_cleanup
;
433 TRACE("unix domain socket %d for command server is ready",
434 handler
->conf
->maincmd_fd
);
447 int lxc_init(const char *name
, struct lxc_handler
*handler
)
449 struct lxc_conf
*conf
= handler
->conf
;
452 TRACE("initialized LSM");
454 if (lxc_read_seccomp_config(conf
) != 0) {
455 ERROR("Failed loading seccomp policy.");
456 goto out_close_maincmd_fd
;
459 /* Begin by setting the state to STARTING. */
460 if (lxc_set_state(name
, handler
, STARTING
)) {
461 ERROR("Failed to set state for container \"%s\" to \"%s\".", name
, lxc_state2str(STARTING
));
462 goto out_close_maincmd_fd
;
465 /* Start of environment variable setup for hooks. */
466 if (name
&& setenv("LXC_NAME", name
, 1))
467 SYSERROR("Failed to set environment variable: LXC_NAME=%s.", name
);
469 if (conf
->rcfile
&& setenv("LXC_CONFIG_FILE", conf
->rcfile
, 1))
470 SYSERROR("Failed to set environment variable: LXC_CONFIG_FILE=%s.", conf
->rcfile
);
472 if (conf
->rootfs
.mount
&& setenv("LXC_ROOTFS_MOUNT", conf
->rootfs
.mount
, 1))
473 SYSERROR("Failed to set environment variable: LXC_ROOTFS_MOUNT=%s.", conf
->rootfs
.mount
);
475 if (conf
->rootfs
.path
&& setenv("LXC_ROOTFS_PATH", conf
->rootfs
.path
, 1))
476 SYSERROR("Failed to set environment variable: LXC_ROOTFS_PATH=%s.", conf
->rootfs
.path
);
478 if (conf
->console
.path
&& setenv("LXC_CONSOLE", conf
->console
.path
, 1))
479 SYSERROR("Failed to set environment variable: LXC_CONSOLE=%s.", conf
->console
.path
);
481 if (conf
->console
.log_path
&& setenv("LXC_CONSOLE_LOGPATH", conf
->console
.log_path
, 1))
482 SYSERROR("Failed to set environment variable: LXC_CONSOLE_LOGPATH=%s.", conf
->console
.log_path
);
484 if (setenv("LXC_CGNS_AWARE", "1", 1))
485 SYSERROR("Failed to set environment variable LXC_CGNS_AWARE=1.");
486 /* End of environment variable setup for hooks. */
488 if (run_lxc_hooks(name
, "pre-start", conf
, handler
->lxcpath
, NULL
)) {
489 ERROR("Failed to run lxc.hook.pre-start for container \"%s\".", name
);
493 /* The signal fd has to be created before forking otherwise if the child
494 * process exits before we setup the signal fd, the event will be lost
495 * and the command will be stuck.
497 handler
->sigfd
= setup_signal_fd(&handler
->oldmask
);
498 if (handler
->sigfd
< 0) {
499 ERROR("Failed to setup SIGCHLD fd handler.");
503 /* Do this after setting up signals since it might unblock SIGWINCH. */
504 if (lxc_console_create(conf
)) {
505 ERROR("Failed to create console for container \"%s\".", name
);
506 goto out_restore_sigmask
;
509 if (lxc_ttys_shift_ids(conf
) < 0) {
510 ERROR("Failed to shift tty into container.");
511 goto out_restore_sigmask
;
514 INFO("Container \"%s\" is initialized.", name
);
518 sigprocmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
520 lxc_delete_tty(&conf
->tty_info
);
522 lxc_set_state(name
, handler
, ABORTING
);
523 out_close_maincmd_fd
:
524 close(conf
->maincmd_fd
);
525 conf
->maincmd_fd
= -1;
529 void lxc_fini(const char *name
, struct lxc_handler
*handler
)
532 pid_t self
= getpid();
533 char *namespaces
[LXC_NS_MAX
+1];
534 size_t namespace_count
= 0;
536 /* The STOPPING state is there for future cleanup code which can take
539 lxc_set_state(name
, handler
, STOPPING
);
541 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
542 if (handler
->nsfd
[i
] != -1) {
543 rc
= asprintf(&namespaces
[namespace_count
], "%s:/proc/%d/fd/%d",
544 ns_info
[i
].proc_name
, self
, handler
->nsfd
[i
]);
546 SYSERROR("Failed to allocate memory.");
552 namespaces
[namespace_count
] = NULL
;
554 if (handler
->conf
->reboot
&& setenv("LXC_TARGET", "reboot", 1))
555 SYSERROR("Failed to set environment variable: LXC_TARGET=reboot.");
557 if (!handler
->conf
->reboot
&& setenv("LXC_TARGET", "stop", 1))
558 SYSERROR("Failed to set environment variable: LXC_TARGET=stop.");
560 if (run_lxc_hooks(name
, "stop", handler
->conf
, handler
->lxcpath
, namespaces
))
561 ERROR("Failed to run lxc.hook.stop for container \"%s\".", name
);
563 while (namespace_count
--)
564 free(namespaces
[namespace_count
]);
565 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
566 if (handler
->nsfd
[i
] != -1) {
567 close(handler
->nsfd
[i
]);
568 handler
->nsfd
[i
] = -1;
572 if (handler
->netnsfd
>= 0) {
573 close(handler
->netnsfd
);
574 handler
->netnsfd
= -1;
577 lxc_set_state(name
, handler
, STOPPED
);
579 if (run_lxc_hooks(name
, "post-stop", handler
->conf
, handler
->lxcpath
, NULL
)) {
580 ERROR("Failed to run lxc.hook.post-stop for container \"%s\".", name
);
581 if (handler
->conf
->reboot
) {
582 WARN("Container will be stopped instead of rebooted.");
583 handler
->conf
->reboot
= 0;
584 if (setenv("LXC_TARGET", "stop", 1))
585 WARN("Failed to set environment variable: LXC_TARGET=stop.");
589 /* Reset mask set by setup_signal_fd. */
590 if (sigprocmask(SIG_SETMASK
, &handler
->oldmask
, NULL
))
591 WARN("Failed to restore signal mask.");
593 lxc_console_delete(&handler
->conf
->console
);
594 lxc_delete_tty(&handler
->conf
->tty_info
);
595 close(handler
->conf
->maincmd_fd
);
596 handler
->conf
->maincmd_fd
= -1;
598 if (handler
->ttysock
[0] != -1) {
599 close(handler
->ttysock
[0]);
600 close(handler
->ttysock
[1]);
603 if (handler
->conf
->ephemeral
== 1 && handler
->conf
->reboot
!= 1)
604 lxc_destroy_container_on_signal(handler
, name
);
606 cgroup_destroy(handler
);
610 void lxc_abort(const char *name
, struct lxc_handler
*handler
)
614 lxc_set_state(name
, handler
, ABORTING
);
615 if (handler
->pid
> 0)
616 kill(handler
->pid
, SIGKILL
);
617 while ((ret
= waitpid(-1, &status
, 0)) > 0) {
622 #include <sys/reboot.h>
623 #include <linux/reboot.h>
625 /* reboot(LINUX_REBOOT_CMD_CAD_ON) will return -EINVAL in a child pid namespace
626 * if container reboot support exists. Otherwise, it will either succeed or
629 static int container_reboot_supported(void *arg
)
635 if (ret
== -1 && errno
== EINVAL
)
640 static int must_drop_cap_sys_boot(struct lxc_conf
*conf
)
643 int ret
, cmd
, v
, flags
;
644 long stack_size
= 4096;
645 void *stack
= alloca(stack_size
);
649 f
= fopen("/proc/sys/kernel/ctrl-alt-del", "r");
651 DEBUG("failed to open /proc/sys/kernel/ctrl-alt-del");
655 ret
= fscanf(f
, "%d", &v
);
658 DEBUG("Failed to read /proc/sys/kernel/ctrl-alt-del.");
661 cmd
= v
? LINUX_REBOOT_CMD_CAD_ON
: LINUX_REBOOT_CMD_CAD_OFF
;
663 flags
= CLONE_NEWPID
| SIGCHLD
;
664 if (!lxc_list_empty(&conf
->id_map
))
665 flags
|= CLONE_NEWUSER
;
668 pid
= __clone2(container_reboot_supported
, stack
, stack_size
, flags
, &cmd
);
671 pid
= clone(container_reboot_supported
, stack
, flags
, &cmd
);
674 if (flags
& CLONE_NEWUSER
)
675 ERROR("Failed to clone (%#x): %s (includes CLONE_NEWUSER).", flags
, strerror(errno
));
677 ERROR("Failed to clone (%#x): %s.", flags
, strerror(errno
));
680 if (wait(&status
) < 0) {
681 SYSERROR("Unexpected wait error: %m.");
685 if (WEXITSTATUS(status
) != 1)
691 /* netpipe is used in the unprivileged case to transfer the ifindexes from
694 static int netpipe
= -1;
696 static inline int count_veths(struct lxc_list
*network
)
698 struct lxc_list
*iterator
;
699 struct lxc_netdev
*netdev
;
702 lxc_list_for_each(iterator
, network
) {
703 netdev
= iterator
->elem
;
704 if (netdev
->type
!= LXC_NET_VETH
)
711 static int read_unpriv_netifindex(struct lxc_list
*network
)
713 struct lxc_list
*iterator
;
714 struct lxc_netdev
*netdev
;
718 lxc_list_for_each(iterator
, network
) {
719 netdev
= iterator
->elem
;
720 if (netdev
->type
!= LXC_NET_VETH
)
722 if (!(netdev
->name
= malloc(IFNAMSIZ
))) {
723 ERROR("Out of memory.");
727 if (read(netpipe
, netdev
->name
, IFNAMSIZ
) != IFNAMSIZ
) {
736 static int do_start(void *data
)
738 struct lxc_list
*iterator
;
739 struct lxc_handler
*handler
= data
;
740 int devnull_fd
= -1, ret
;
743 if (sigprocmask(SIG_SETMASK
, &handler
->oldmask
, NULL
)) {
744 SYSERROR("Failed to set signal mask.");
748 /* This prctl must be before the synchro, so if the parent dies before
749 * we set the parent death signal, we will detect its death with the
750 * synchro right after, otherwise we have a window where the parent can
751 * exit before we set the pdeath signal leading to a unsupervized
754 if (prctl(PR_SET_PDEATHSIG
, SIGKILL
, 0, 0, 0)) {
755 SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL.");
759 lxc_sync_fini_parent(handler
);
761 /* Don't leak the pinfd to the container. */
762 if (handler
->pinfd
>= 0) {
763 close(handler
->pinfd
);
766 if (lxc_sync_wait_parent(handler
, LXC_SYNC_STARTUP
))
769 /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See
770 * https://github.com/lxc/lxd/issues/1978.
772 if ((handler
->clone_flags
& (CLONE_NEWNET
| CLONE_NEWUSER
)) ==
773 (CLONE_NEWNET
| CLONE_NEWUSER
)) {
774 ret
= unshare(CLONE_NEWNET
);
776 SYSERROR("Failed to unshare CLONE_NEWNET.");
777 goto out_warn_father
;
779 INFO("Unshared CLONE_NEWNET.");
782 /* Tell the parent task it can begin to configure the container and wait
785 if (lxc_sync_barrier_parent(handler
, LXC_SYNC_CONFIGURE
))
788 if (read_unpriv_netifindex(&handler
->conf
->network
) < 0)
789 goto out_warn_father
;
791 /* If we are in a new user namespace, become root there to have
792 * privilege over our namespace.
794 if (!lxc_list_empty(&handler
->conf
->id_map
)) {
795 if (lxc_switch_uid_gid(0, 0) < 0)
796 goto out_warn_father
;
798 /* Drop groups only after we switched to a valid gid in the new
801 if (lxc_setgroups(0, NULL
) < 0)
802 goto out_warn_father
;
805 if (access(handler
->lxcpath
, X_OK
)) {
806 print_top_failing_dir(handler
->lxcpath
);
807 goto out_warn_father
;
811 if (handler
->conf
->need_utmp_watch
) {
812 if (prctl(PR_CAPBSET_DROP
, CAP_SYS_BOOT
, 0, 0, 0)) {
813 SYSERROR("Failed to remove the CAP_SYS_BOOT capability.");
814 goto out_warn_father
;
816 DEBUG("Dropped the CAP_SYS_BOOT capability.");
820 ret
= snprintf(path
, sizeof(path
), "%s/dev/null", handler
->conf
->rootfs
.mount
);
821 if (ret
< 0 || ret
>= sizeof(path
))
822 goto out_warn_father
;
824 /* In order to checkpoint restore, we need to have everything in the
825 * same mount namespace. However, some containers may not have a
826 * reasonable /dev (in particular, they may not have /dev/null), so we
827 * can't set init's std fds to /dev/null by opening it from inside the
830 * If that's the case, fall back to using the host's /dev/null. This
831 * means that migration won't work, but at least we won't spew output
832 * where it isn't wanted.
834 if (handler
->backgrounded
&& !handler
->conf
->autodev
&& access(path
, F_OK
) < 0) {
835 devnull_fd
= open_devnull();
838 goto out_warn_father
;
839 WARN("Using /dev/null from the host for container init's "
840 "standard file descriptors. Migration will not work.");
843 /* Setup the container, ip, names, utsname, ... */
844 if (lxc_setup(handler
)) {
845 ERROR("Failed to setup container \"%s\".", handler
->name
);
846 goto out_warn_father
;
849 /* Ask father to setup cgroups and wait for him to finish. */
850 if (lxc_sync_barrier_parent(handler
, LXC_SYNC_CGROUP
))
853 /* Unshare cgroup namespace after we have setup our cgroups. If we do it
854 * earlier we end up with a wrong view of /proc/self/cgroup. For
855 * example, assume we unshare(CLONE_NEWCGROUP) first, and then create
856 * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then
857 * /proc/self/cgroup would show us:
861 * whereas it should actually show
865 if (cgns_supported()) {
866 if (unshare(CLONE_NEWCGROUP
) < 0) {
867 INFO("Failed to unshare CLONE_NEWCGROUP.");
868 goto out_warn_father
;
870 INFO("Unshared CLONE_NEWCGROUP.");
873 /* Set the label to change to when we exec(2) the container's init. */
874 if (lsm_process_label_set(NULL
, handler
->conf
, 1, 1) < 0)
875 goto out_warn_father
;
877 /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it
878 * before we aren't allowed anymore.
880 if (handler
->conf
->no_new_privs
) {
881 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
882 SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges.");
883 goto out_warn_father
;
885 DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges.");
888 /* Some init's such as busybox will set sane tty settings on stdin,
889 * stdout, stderr which it thinks is the console. We already set them
890 * the way we wanted on the real terminal, and we want init to do its
891 * setup on its console ie. the pty allocated in lxc_console_create() so
892 * make sure that that pty is stdin,stdout,stderr.
894 if (lxc_console_set_stdfds(handler
->conf
->console
.slave
) < 0)
895 goto out_warn_father
;
897 /* If we mounted a temporary proc, then unmount it now. */
898 tmp_proc_unmount(handler
->conf
);
900 if (lxc_seccomp_load(handler
->conf
) != 0)
901 goto out_warn_father
;
903 if (run_lxc_hooks(handler
->name
, "start", handler
->conf
, handler
->lxcpath
, NULL
)) {
904 ERROR("Failed to run lxc.hook.start for container \"%s\".", handler
->name
);
905 goto out_warn_father
;
908 /* The container has been setup. We can now switch to an unprivileged
911 if (handler
->conf
->is_execute
) {
912 bool have_cap_setgid
;
913 uid_t new_uid
= handler
->conf
->init_uid
;
914 gid_t new_gid
= handler
->conf
->init_gid
;
916 /* If we are in a new user namespace we already dropped all
917 * groups when we switched to root in the new user namespace
918 * further above. Only drop groups if we can, so ensure that we
919 * have necessary privilege.
922 have_cap_setgid
= lxc_proc_cap_is_set(CAP_SETGID
, CAP_EFFECTIVE
);
924 have_cap_setgid
= false;
926 if (lxc_list_empty(&handler
->conf
->id_map
) && have_cap_setgid
) {
927 if (lxc_setgroups(0, NULL
) < 0)
928 goto out_warn_father
;
931 if (lxc_switch_uid_gid(new_uid
, new_gid
) < 0)
932 goto out_warn_father
;
935 /* The clearenv() and putenv() calls have been moved here to allow us to
936 * use environment variables passed to the various hooks, such as the
937 * start hook above. Not all of the variables like CONFIG_PATH or ROOTFS
938 * are valid in this context but others are.
941 SYSERROR("Failed to clear environment.");
942 /* Don't error out though. */
945 lxc_list_for_each(iterator
, &handler
->conf
->environment
) {
946 if (putenv((char *)iterator
->elem
)) {
947 SYSERROR("Failed to set environment variable: %s.", (char *)iterator
->elem
);
948 goto out_warn_father
;
952 if (putenv("container=lxc")) {
953 SYSERROR("Failed to set environment variable: container=lxc.");
954 goto out_warn_father
;
957 if (handler
->conf
->pty_names
) {
958 if (putenv(handler
->conf
->pty_names
)) {
959 SYSERROR("Failed to set environment variable for container ptys.");
960 goto out_warn_father
;
964 close(handler
->sigfd
);
966 if (devnull_fd
< 0) {
967 devnull_fd
= open_devnull();
970 goto out_warn_father
;
973 if (handler
->backgrounded
&& set_stdfds(devnull_fd
))
974 goto out_warn_father
;
976 if (devnull_fd
>= 0) {
983 /* After this call, we are in error because this ops should not return
986 handler
->ops
->start(handler
, handler
->data
);
989 /* We want the parent to know something went wrong, so we return a
990 * special error code.
992 lxc_sync_wake_parent(handler
, LXC_SYNC_ERROR
);
1001 static int save_phys_nics(struct lxc_conf
*conf
)
1003 struct lxc_list
*iterator
;
1004 int am_root
= (getuid() == 0);
1009 lxc_list_for_each(iterator
, &conf
->network
) {
1010 struct lxc_netdev
*netdev
= iterator
->elem
;
1012 if (netdev
->type
!= LXC_NET_PHYS
)
1014 conf
->saved_nics
= realloc(conf
->saved_nics
,
1015 (conf
->num_savednics
+1)*sizeof(struct saved_nic
));
1016 if (!conf
->saved_nics
)
1018 conf
->saved_nics
[conf
->num_savednics
].ifindex
= netdev
->ifindex
;
1019 conf
->saved_nics
[conf
->num_savednics
].orig_name
= strdup(netdev
->link
);
1020 if (!conf
->saved_nics
[conf
->num_savednics
].orig_name
)
1022 INFO("Stored saved_nic #%d idx %d name %s.", conf
->num_savednics
,
1023 conf
->saved_nics
[conf
->num_savednics
].ifindex
,
1024 conf
->saved_nics
[conf
->num_savednics
].orig_name
);
1025 conf
->num_savednics
++;
1031 static int lxc_recv_ttys_from_child(struct lxc_handler
*handler
)
1035 struct lxc_pty_info
*pty_info
;
1037 int sock
= handler
->ttysock
[1];
1038 struct lxc_conf
*conf
= handler
->conf
;
1039 struct lxc_tty_info
*tty_info
= &conf
->tty_info
;
1040 size_t num_ttyfds
= (2 * conf
->tty
);
1045 tty_info
->pty_info
= malloc(sizeof(*tty_info
->pty_info
) * conf
->tty
);
1046 if (!tty_info
->pty_info
)
1049 ttyfds
= malloc(num_ttyfds
* sizeof(int));
1053 ret
= lxc_abstract_unix_recv_fds(sock
, ttyfds
, num_ttyfds
, NULL
, 0);
1054 for (i
= 0; (ret
>= 0 && *ttyfds
!= -1) && (i
< num_ttyfds
); i
++) {
1055 pty_info
= &tty_info
->pty_info
[i
/ 2];
1057 pty_info
->slave
= ttyfds
[i
++];
1058 pty_info
->master
= ttyfds
[i
];
1059 TRACE("received pty with master fd %d and slave fd %d from "
1060 "parent", pty_info
->master
, pty_info
->slave
);
1063 tty_info
->nbtty
= conf
->tty
;
1068 ERROR("failed to receive %d ttys from child: %s", conf
->tty
,
1071 TRACE("received %d ttys from child", conf
->tty
);
1076 void resolve_clone_flags(struct lxc_handler
*handler
)
1078 handler
->clone_flags
= CLONE_NEWPID
| CLONE_NEWNS
;
1080 if (!lxc_list_empty(&handler
->conf
->id_map
))
1081 handler
->clone_flags
|= CLONE_NEWUSER
;
1083 if (handler
->conf
->inherit_ns_fd
[LXC_NS_NET
] == -1) {
1084 if (!lxc_requests_empty_network(handler
))
1085 handler
->clone_flags
|= CLONE_NEWNET
;
1087 INFO("Inheriting a NET namespace.");
1090 if (handler
->conf
->inherit_ns_fd
[LXC_NS_IPC
] == -1)
1091 handler
->clone_flags
|= CLONE_NEWIPC
;
1093 INFO("Inheriting an IPC namespace.");
1095 if (handler
->conf
->inherit_ns_fd
[LXC_NS_UTS
] == -1)
1096 handler
->clone_flags
|= CLONE_NEWUTS
;
1098 INFO("Inheriting a UTS namespace.");
1101 /* lxc_spawn() performs crucial setup tasks and clone()s the new process which
1102 * exec()s the requested container binary.
1103 * Note that lxc_spawn() runs in the parent namespaces. Any operations performed
1104 * right here should be double checked if they'd pose a security risk. (For
1105 * example, any {u}mount() operations performed here will be reflected on the
1108 static int lxc_spawn(struct lxc_handler
*handler
)
1110 int failed_before_rename
= 0;
1111 const char *name
= handler
->name
;
1112 bool cgroups_connected
= false;
1113 int saved_ns_fd
[LXC_NS_MAX
];
1114 int preserve_mask
= 0, i
, flags
;
1115 int netpipepair
[2], nveths
;
1116 bool wants_to_map_ids
;
1117 struct lxc_list
*id_map
;
1120 id_map
= &handler
->conf
->id_map
;
1121 wants_to_map_ids
= !lxc_list_empty(id_map
);
1123 for (i
= 0; i
< LXC_NS_MAX
; i
++)
1124 if (handler
->conf
->inherit_ns_fd
[i
] != -1)
1125 preserve_mask
|= ns_info
[i
].clone_flag
;
1127 if (lxc_sync_init(handler
))
1130 if (socketpair(AF_UNIX
, SOCK_DGRAM
, 0, handler
->ttysock
) < 0) {
1131 lxc_sync_fini(handler
);
1135 resolve_clone_flags(handler
);
1137 if (handler
->clone_flags
& CLONE_NEWNET
) {
1138 if (!lxc_list_empty(&handler
->conf
->network
)) {
1140 /* Find gateway addresses from the link device, which is
1141 * no longer accessible inside the container. Do this
1142 * before creating network interfaces, since goto
1143 * out_delete_net does not work before lxc_clone.
1145 if (lxc_find_gateway_addresses(handler
)) {
1146 ERROR("Failed to find gateway addresses.");
1147 lxc_sync_fini(handler
);
1151 /* That should be done before the clone because we will
1152 * fill the netdev index and use them in the child.
1154 if (lxc_create_network(handler
)) {
1155 ERROR("Failed to create the network.");
1156 lxc_sync_fini(handler
);
1161 if (save_phys_nics(handler
->conf
)) {
1162 ERROR("Failed to save physical nic info.");
1167 if (!cgroup_init(handler
)) {
1168 ERROR("Failed initializing cgroup support.");
1169 goto out_delete_net
;
1172 cgroups_connected
= true;
1174 if (!cgroup_create(handler
)) {
1175 ERROR("Failed creating cgroups.");
1176 goto out_delete_net
;
1179 /* If the rootfs is not a blockdev, prevent the container from marking
1181 * If the container is unprivileged then skip rootfs pinning.
1183 if (!wants_to_map_ids
) {
1184 handler
->pinfd
= pin_rootfs(handler
->conf
->rootfs
.path
);
1185 if (handler
->pinfd
== -1)
1186 INFO("Failed to pin the rootfs for container \"%s\".", handler
->name
);
1189 if (!preserve_ns(saved_ns_fd
, preserve_mask
, getpid()))
1190 goto out_delete_net
;
1192 if (attach_ns(handler
->conf
->inherit_ns_fd
) < 0)
1193 goto out_delete_net
;
1195 if (am_unpriv() && (nveths
= count_veths(&handler
->conf
->network
))) {
1196 if (pipe(netpipepair
) < 0) {
1197 SYSERROR("Failed to create pipe.");
1198 goto out_delete_net
;
1200 /* Store netpipe in the global var for do_start's use. */
1201 netpipe
= netpipepair
[0];
1204 /* Create a process in a new set of namespaces. */
1205 flags
= handler
->clone_flags
;
1206 if (handler
->clone_flags
& CLONE_NEWUSER
) {
1207 /* If CLONE_NEWUSER and CLONE_NEWNET was requested, we need to
1208 * clone a new user namespace first and only later unshare our
1209 * network namespace to ensure that network devices ownership is
1212 flags
&= ~CLONE_NEWNET
;
1214 handler
->pid
= lxc_clone(do_start
, handler
, flags
);
1215 if (handler
->pid
< 0) {
1216 SYSERROR("Failed to clone a new set of namespaces.");
1217 goto out_delete_net
;
1219 for (i
= 0; i
< LXC_NS_MAX
; i
++)
1220 if (flags
& ns_info
[i
].clone_flag
)
1221 INFO("Cloned %s.", ns_info
[i
].flag_name
);
1223 if (!preserve_ns(handler
->nsfd
, handler
->clone_flags
| preserve_mask
, handler
->pid
))
1224 INFO("Failed to preserve namespace for lxc.hook.stop.");
1226 if (attach_ns(saved_ns_fd
))
1227 WARN("Failed to restore saved namespaces.");
1229 lxc_sync_fini_child(handler
);
1231 /* Map the container uids. The container became an invalid userid the
1232 * moment it was cloned with CLONE_NEWUSER. This call doesn't change
1233 * anything immediately, but allows the container to setuid(0) (0 being
1234 * mapped to something else on the host.) later to become a valid uid
1237 if (wants_to_map_ids
&& lxc_map_ids(id_map
, handler
->pid
)) {
1238 ERROR("Failed to set up id mapping.");
1239 goto out_delete_net
;
1242 if (lxc_sync_wake_child(handler
, LXC_SYNC_STARTUP
)) {
1243 failed_before_rename
= 1;
1244 goto out_delete_net
;
1247 if (lxc_sync_wait_child(handler
, LXC_SYNC_CONFIGURE
)) {
1248 failed_before_rename
= 1;
1249 goto out_delete_net
;
1252 if (!cgroup_create_legacy(handler
)) {
1253 ERROR("Failed to setup legacy cgroups for container \"%s\".", name
);
1254 goto out_delete_net
;
1256 if (!cgroup_setup_limits(handler
, false)) {
1257 ERROR("Failed to setup cgroup limits for container \"%s\".", name
);
1258 goto out_delete_net
;
1261 if (!cgroup_enter(handler
))
1262 goto out_delete_net
;
1264 if (!cgroup_chown(handler
))
1265 goto out_delete_net
;
1267 if (failed_before_rename
)
1268 goto out_delete_net
;
1270 /* Create the network configuration. */
1271 if (handler
->clone_flags
& CLONE_NEWNET
) {
1272 if (lxc_assign_network(handler
->lxcpath
, handler
->name
,
1273 &handler
->conf
->network
, handler
->pid
)) {
1274 ERROR("Failed to create the configured network.");
1275 goto out_delete_net
;
1279 if (netpipe
!= -1) {
1280 struct lxc_list
*iterator
;
1281 struct lxc_netdev
*netdev
;
1284 lxc_list_for_each(iterator
, &handler
->conf
->network
) {
1285 netdev
= iterator
->elem
;
1286 if (netdev
->type
!= LXC_NET_VETH
)
1288 if (write(netpipepair
[1], netdev
->name
, IFNAMSIZ
) != IFNAMSIZ
) {
1289 ERROR("Error writing veth name to container.");
1290 goto out_delete_net
;
1293 close(netpipepair
[1]);
1296 /* Tell the child to continue its initialization. We'll get
1297 * LXC_SYNC_CGROUP when it is ready for us to setup cgroups.
1299 if (lxc_sync_barrier_child(handler
, LXC_SYNC_POST_CONFIGURE
))
1300 goto out_delete_net
;
1302 if (!lxc_list_empty(&handler
->conf
->limits
) && setup_resource_limits(&handler
->conf
->limits
, handler
->pid
)) {
1303 ERROR("failed to setup resource limits for '%s'", name
);
1304 goto out_delete_net
;
1307 if (!cgroup_setup_limits(handler
, true)) {
1308 ERROR("Failed to setup the devices cgroup for container \"%s\".", name
);
1309 goto out_delete_net
;
1312 cgroup_disconnect();
1313 cgroups_connected
= false;
1315 /* Read tty fds allocated by child. */
1316 if (lxc_recv_ttys_from_child(handler
) < 0) {
1317 ERROR("Failed to receive tty info from child process.");
1318 goto out_delete_net
;
1321 /* Tell the child to complete its initialization and wait for it to exec
1322 * or return an error. (The child will never return
1323 * LXC_SYNC_POST_CGROUP+1. It will either close the sync pipe, causing
1324 * lxc_sync_barrier_child to return success, or return a different
1325 * value, causing us to error out).
1327 if (lxc_sync_barrier_child(handler
, LXC_SYNC_POST_CGROUP
))
1330 if (handler
->ops
->post_start(handler
, handler
->data
))
1333 if (lxc_set_state(name
, handler
, RUNNING
)) {
1334 ERROR("Failed to set state for container \"%s\" to \"%s\".", name
,
1335 lxc_state2str(RUNNING
));
1339 lxc_sync_fini(handler
);
1340 handler
->netnsfd
= lxc_preserve_ns(handler
->pid
, "net");
1345 if (cgroups_connected
)
1346 cgroup_disconnect();
1347 if (handler
->clone_flags
& CLONE_NEWNET
)
1348 lxc_delete_network(handler
);
1350 lxc_abort(name
, handler
);
1351 lxc_sync_fini(handler
);
1352 if (handler
->pinfd
>= 0) {
1353 close(handler
->pinfd
);
1354 handler
->pinfd
= -1;
1360 int __lxc_start(const char *name
, struct lxc_handler
*handler
,
1361 struct lxc_operations
* ops
, void *data
, const char *lxcpath
,
1366 bool removed_all_netdevs
= true;
1367 struct lxc_conf
*conf
= handler
->conf
;
1369 if (lxc_init(name
, handler
) < 0) {
1370 ERROR("Failed to initialize container \"%s\".", name
);
1374 handler
->data
= data
;
1375 handler
->backgrounded
= backgrounded
;
1376 handler
->netnsfd
= -1;
1378 if (must_drop_cap_sys_boot(handler
->conf
)) {
1380 DEBUG("Dropping CAP_SYS_BOOT capability.");
1382 DEBUG("Not dropping CAP_SYS_BOOT capability as capabilities aren't supported.");
1385 DEBUG("Not dropping CAP_SYS_BOOT or watching utmp.");
1386 handler
->conf
->need_utmp_watch
= 0;
1389 if (!attach_block_device(handler
->conf
)) {
1390 ERROR("Failed to attach block device.");
1391 goto out_fini_nonet
;
1394 if (geteuid() == 0 && !lxc_list_empty(&conf
->id_map
)) {
1395 /* If the backing store is a device, mount it here and now. */
1396 if (rootfs_is_blockdev(conf
)) {
1397 if (unshare(CLONE_NEWNS
) < 0) {
1398 ERROR("Failed to unshare CLONE_NEWNS.");
1399 goto out_fini_nonet
;
1401 INFO("Unshared CLONE_NEWNS.");
1403 remount_all_slave();
1404 if (do_rootfs_setup(conf
, name
, lxcpath
) < 0) {
1405 ERROR("Error setting up rootfs mount as root before spawn.");
1406 goto out_fini_nonet
;
1408 INFO("Set up container rootfs as host root.");
1412 err
= lxc_spawn(handler
);
1414 ERROR("Failed to spawn container \"%s\".", name
);
1415 goto out_detach_blockdev
;
1418 handler
->conf
->reboot
= 0;
1420 err
= lxc_poll(name
, handler
);
1422 ERROR("LXC mainloop exited with error: %d.", err
);
1423 if (handler
->netnsfd
>= 0) {
1424 close(handler
->netnsfd
);
1425 handler
->netnsfd
= -1;
1430 while (waitpid(handler
->pid
, &status
, 0) < 0 && errno
== EINTR
)
1433 /* If the child process exited but was not signaled, it didn't call
1434 * reboot. This should mean it was an lxc-execute which simply exited.
1435 * In any case, treat it as a 'halt'.
1437 if (WIFSIGNALED(status
)) {
1438 switch(WTERMSIG(status
)) {
1439 case SIGINT
: /* halt */
1440 DEBUG("Container \"%s\" is halting.", name
);
1442 case SIGHUP
: /* reboot */
1443 DEBUG("Container \"%s\" is rebooting.", name
);
1444 handler
->conf
->reboot
= 1;
1446 case SIGSYS
: /* seccomp */
1447 DEBUG("Container \"%s\" violated its seccomp policy.", name
);
1450 DEBUG("Unknown exit status for container \"%s\" init %d.", name
, WTERMSIG(status
));
1455 DEBUG("Pushing physical nics back to host namespace");
1456 lxc_restore_phys_nics_to_netns(handler
->netnsfd
, handler
->conf
);
1458 DEBUG("Tearing down virtual network devices used by container \"%s\".", name
);
1459 removed_all_netdevs
= lxc_delete_network(handler
);
1461 if (handler
->pinfd
>= 0) {
1462 close(handler
->pinfd
);
1463 handler
->pinfd
= -1;
1466 lxc_monitor_send_exit_code(name
, status
, handler
->lxcpath
);
1467 err
= lxc_error_set_and_log(handler
->pid
, status
);
1469 if (!removed_all_netdevs
) {
1470 DEBUG("Failed tearing down network devices used by container. Trying again!");
1471 removed_all_netdevs
= lxc_delete_network(handler
);
1472 if (!removed_all_netdevs
)
1473 DEBUG("Failed tearing down network devices used by container. Not trying again!");
1476 out_detach_blockdev
:
1477 detach_block_device(handler
->conf
);
1480 lxc_fini(name
, handler
);
1484 lxc_abort(name
, handler
);
1492 static int start(struct lxc_handler
*handler
, void* data
)
1494 struct start_args
*arg
= data
;
1496 NOTICE("Exec'ing \"%s\".", arg
->argv
[0]);
1498 execvp(arg
->argv
[0], arg
->argv
);
1499 SYSERROR("Failed to exec \"%s\".", arg
->argv
[0]);
1503 static int post_start(struct lxc_handler
*handler
, void* data
)
1505 struct start_args
*arg
= data
;
1507 NOTICE("Started \"%s\" with pid \"%d\".", arg
->argv
[0], handler
->pid
);
1511 static struct lxc_operations start_ops
= {
1513 .post_start
= post_start
1516 int lxc_start(const char *name
, char *const argv
[], struct lxc_handler
*handler
,
1517 const char *lxcpath
, bool backgrounded
)
1519 struct start_args start_arg
= {
1523 handler
->conf
->need_utmp_watch
= 1;
1524 return __lxc_start(name
, handler
, &start_ops
, &start_arg
, lxcpath
, backgrounded
);
1527 static void lxc_destroy_container_on_signal(struct lxc_handler
*handler
,
1530 char destroy
[MAXPATHLEN
];
1533 struct lxc_container
*c
;
1534 if (handler
->conf
->rootfs
.path
&& handler
->conf
->rootfs
.mount
) {
1535 bret
= do_destroy_container(handler
->conf
);
1537 ERROR("Error destroying rootfs for container \"%s\".", name
);
1541 INFO("Destroyed rootfs for container \"%s\".", name
);
1543 ret
= snprintf(destroy
, MAXPATHLEN
, "%s/%s", handler
->lxcpath
, name
);
1544 if (ret
< 0 || ret
>= MAXPATHLEN
) {
1545 ERROR("Error destroying directory for container \"%s\".", name
);
1549 c
= lxc_container_new(name
, handler
->lxcpath
);
1551 if (container_disk_lock(c
)) {
1552 INFO("Could not update lxc_snapshots file.");
1553 lxc_container_put(c
);
1555 mod_all_rdeps(c
, false);
1556 container_disk_unlock(c
);
1557 lxc_container_put(c
);
1562 ret
= userns_exec_1(handler
->conf
, lxc_rmdir_onedev_wrapper
,
1563 destroy
, "lxc_rmdir_onedev_wrapper");
1565 ret
= lxc_rmdir_onedev(destroy
, NULL
);
1568 ERROR("Error destroying directory for container \"%s\".", name
);
1571 INFO("Destroyed directory for container \"%s\".", name
);
1574 static int lxc_rmdir_onedev_wrapper(void *data
)
1576 char *arg
= (char *) data
;
1577 return lxc_rmdir_onedev(arg
, NULL
);
1580 static bool do_destroy_container(struct lxc_conf
*conf
) {
1582 if (userns_exec_1(conf
, bdev_destroy_wrapper
, conf
,
1583 "bdev_destroy_wrapper") < 0)
1587 return bdev_destroy(conf
);