]>
git.proxmox.com Git - mirror_lxc.git/blob - src/lxc/start.c
2 * lxc: linux Container library
4 * (C) Copyright IBM Corp. 2007, 2008
7 * Daniel Lezcano <daniel.lezcano at free.fr>
8 * Serge Hallyn <serge@hallyn.com>
9 * Christian Brauner <christian.brauner@ubuntu.com>
11 * This library is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU Lesser General Public
13 * License as published by the Free Software Foundation; either
14 * version 2.1 of the License, or (at your option) any later version.
16 * This library is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * Lesser General Public License for more details.
21 * You should have received a copy of the GNU Lesser General Public
22 * License along with this library; if not, write to the Free Software
23 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
42 #include <sys/mount.h>
43 #include <sys/param.h>
44 #include <sys/prctl.h>
45 #include <sys/socket.h>
47 #include <sys/syscall.h>
48 #include <sys/types.h>
53 #include <sys/capability.h>
56 #if !HAVE_DECL_PR_CAPBSET_DROP
57 #define PR_CAPBSET_DROP 24
60 #if !HAVE_DECL_PR_SET_NO_NEW_PRIVS
61 #define PR_SET_NO_NEW_PRIVS 38
64 #if !HAVE_DECL_PR_GET_NO_NEW_PRIVS
65 #define PR_GET_NO_NEW_PRIVS 39
72 #include "commands_utils.h"
74 #include "confile_utils.h"
79 #include "lxccontainer.h"
81 #include "lxcseccomp.h"
84 #include "namespace.h"
87 #include "storage/storage.h"
88 #include "storage/storage_utils.h"
94 #include "include/strlcpy.h"
97 lxc_log_define(start
, lxc
);
99 extern void mod_all_rdeps(struct lxc_container
*c
, bool inc
);
100 static bool do_destroy_container(struct lxc_handler
*handler
);
101 static int lxc_rmdir_onedev_wrapper(void *data
);
102 static void lxc_destroy_container_on_signal(struct lxc_handler
*handler
,
105 static void print_top_failing_dir(const char *path
)
109 char *copy
, *e
, *p
, saved
;
112 copy
= alloca(len
+ 1);
113 (void)strlcpy(copy
, path
, len
+ 1);
119 while (p
< e
&& *p
== '/')
122 while (p
< e
&& *p
!= '/')
128 ret
= access(copy
, X_OK
);
130 SYSERROR("Could not access %s. Please grant it x "
131 "access, or add an ACL for the container "
139 static void lxc_put_nsfds(struct lxc_handler
*handler
)
143 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
144 if (handler
->nsfd
[i
] < 0)
147 close(handler
->nsfd
[i
]);
148 handler
->nsfd
[i
] = -EBADF
;
152 static int lxc_try_preserve_ns(const int pid
, const char *ns
)
156 fd
= lxc_preserve_ns(pid
, ns
);
158 if (errno
!= ENOENT
) {
159 SYSERROR("Failed to preserve %s namespace", ns
);
163 SYSWARN("Kernel does not support preserving %s namespaces", ns
);
170 /* lxc_try_preserve_namespaces: open /proc/@pid/ns/@ns for each namespace
171 * specified in ns_clone_flags.
172 * Return true on success, false on failure.
174 static bool lxc_try_preserve_namespaces(struct lxc_handler
*handler
,
175 int ns_clone_flags
, pid_t pid
)
179 for (i
= 0; i
< LXC_NS_MAX
; i
++)
180 handler
->nsfd
[i
] = -EBADF
;
182 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
185 if ((ns_clone_flags
& ns_info
[i
].clone_flag
) == 0)
188 fd
= lxc_try_preserve_ns(pid
, ns_info
[i
].proc_name
);
190 handler
->nsfd
[i
] = -EBADF
;
192 /* Do not fail to start container on kernels that do
193 * not support interacting with namespaces through
196 if (fd
== -EOPNOTSUPP
)
199 lxc_put_nsfds(handler
);
203 handler
->nsfd
[i
] = fd
;
204 DEBUG("Preserved %s namespace via fd %d", ns_info
[i
].proc_name
,
211 static inline bool match_stdfds(int fd
)
213 return (fd
== STDIN_FILENO
|| fd
== STDOUT_FILENO
|| fd
== STDERR_FILENO
);
216 int lxc_check_inherited(struct lxc_conf
*conf
, bool closeall
,
217 int *fds_to_ignore
, size_t len_fds
)
222 struct dirent
*direntp
;
224 if (conf
&& conf
->close_all_fds
)
228 dir
= opendir("/proc/self/fd");
230 SYSWARN("Failed to open directory");
236 while ((direntp
= readdir(dir
))) {
238 struct lxc_list
*cur
;
239 bool matched
= false;
241 if (strcmp(direntp
->d_name
, ".") == 0)
244 if (strcmp(direntp
->d_name
, "..") == 0)
247 ret
= lxc_safe_int(direntp
->d_name
, &fd
);
249 INFO("Could not parse file descriptor for \"%s\"", direntp
->d_name
);
253 for (i
= 0; i
< len_fds
; i
++)
254 if (fds_to_ignore
[i
] == fd
)
257 if (fd
== fddir
|| fd
== lxc_log_fd
||
258 (i
< len_fds
&& fd
== fds_to_ignore
[i
]))
261 /* Keep state clients that wait on reboots. */
263 lxc_list_for_each(cur
, &conf
->state_clients
) {
264 struct lxc_state_client
*client
= cur
->elem
;
266 if (client
->clientfd
!= fd
)
277 if (current_config
&& fd
== current_config
->logfd
)
280 if (match_stdfds(fd
))
286 INFO("Closed inherited fd %d", fd
);
289 WARN("Inherited fd %d", fd
);
292 /* Only enable syslog at this point to avoid the above logging function
293 * to open a new fd and make the check_inherited function enter an
296 lxc_log_enable_syslog();
298 closedir(dir
); /* cannot fail */
302 static int setup_signal_fd(sigset_t
*oldmask
)
307 const int signals
[] = {SIGBUS
, SIGILL
, SIGSEGV
, SIGWINCH
};
309 /* Block everything except serious error signals. */
310 ret
= sigfillset(&mask
);
314 for (sig
= 0; sig
< (sizeof(signals
) / sizeof(signals
[0])); sig
++) {
315 ret
= sigdelset(&mask
, signals
[sig
]);
320 ret
= pthread_sigmask(SIG_BLOCK
, &mask
, oldmask
);
322 SYSERROR("Failed to set signal mask");
326 ret
= signalfd(-1, &mask
, SFD_CLOEXEC
);
328 SYSERROR("Failed to create signal file descriptor");
332 TRACE("Created signal file descriptor %d", ret
);
337 static int signal_handler(int fd
, uint32_t events
, void *data
,
338 struct lxc_epoll_descr
*descr
)
342 struct signalfd_siginfo siginfo
;
343 struct lxc_handler
*hdlr
= data
;
345 ret
= lxc_read_nointr(fd
, &siginfo
, sizeof(siginfo
));
347 ERROR("Failed to read signal info from signal file descriptor %d", fd
);
348 return LXC_MAINLOOP_ERROR
;
351 if (ret
!= sizeof(siginfo
)) {
352 ERROR("Unexpected size for struct signalfd_siginfo");
356 /* Check whether init is running. */
358 ret
= waitid(P_PID
, hdlr
->pid
, &info
, WEXITED
| WNOWAIT
| WNOHANG
);
359 if (ret
== 0 && info
.si_pid
== hdlr
->pid
)
360 hdlr
->init_died
= true;
362 /* Try to figure out a reasonable exit status to report. */
363 if (hdlr
->init_died
) {
364 switch (info
.si_code
) {
366 hdlr
->exit_status
= info
.si_status
<< 8;
371 hdlr
->exit_status
= info
.si_status
<< 8 | 0x7f;
374 /* Huh? The waitid() told us it's dead *and* continued? */
375 WARN("Init %d dead and continued?", hdlr
->pid
);
376 hdlr
->exit_status
= 1;
379 ERROR("Unknown si_code: %d", info
.si_code
);
380 hdlr
->exit_status
= 1;
384 if (siginfo
.ssi_signo
== SIGHUP
) {
385 kill(hdlr
->pid
, SIGTERM
);
386 INFO("Killing %d since terminal hung up", hdlr
->pid
);
387 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
: LXC_MAINLOOP_CONTINUE
;
390 if (siginfo
.ssi_signo
!= SIGCHLD
) {
391 kill(hdlr
->pid
, siginfo
.ssi_signo
);
392 INFO("Forwarded signal %d to pid %d", siginfo
.ssi_signo
, hdlr
->pid
);
393 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
: LXC_MAINLOOP_CONTINUE
;
396 /* More robustness, protect ourself from a SIGCHLD sent
397 * by a process different from the container init.
399 if (siginfo
.ssi_pid
!= hdlr
->pid
) {
400 NOTICE("Received %d from pid %d instead of container init %d",
401 siginfo
.ssi_signo
, siginfo
.ssi_pid
, hdlr
->pid
);
402 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
: LXC_MAINLOOP_CONTINUE
;
405 if (siginfo
.ssi_code
== CLD_STOPPED
) {
406 INFO("Container init process was stopped");
407 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
: LXC_MAINLOOP_CONTINUE
;
408 } else if (siginfo
.ssi_code
== CLD_CONTINUED
) {
409 INFO("Container init process was continued");
410 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
: LXC_MAINLOOP_CONTINUE
;
413 DEBUG("Container init process %d exited", hdlr
->pid
);
414 return LXC_MAINLOOP_CLOSE
;
417 int lxc_serve_state_clients(const char *name
, struct lxc_handler
*handler
,
422 struct lxc_list
*cur
, *next
;
423 struct lxc_state_client
*client
;
424 struct lxc_msg msg
= {.type
= lxc_msg_state
, .value
= state
};
427 handler
->state
= RUNNING
;
429 handler
->state
= state
;
431 TRACE("Set container state to %s", lxc_state2str(state
));
433 if (lxc_list_empty(&handler
->conf
->state_clients
)) {
434 TRACE("No state clients registered");
438 retlen
= strlcpy(msg
.name
, name
, sizeof(msg
.name
));
439 if (retlen
>= sizeof(msg
.name
))
442 lxc_list_for_each_safe(cur
, &handler
->conf
->state_clients
, next
) {
445 if (client
->states
[state
] == 0) {
446 TRACE("State %s not registered for state client %d",
447 lxc_state2str(state
), client
->clientfd
);
451 TRACE("Sending state %s to state client %d",
452 lxc_state2str(state
), client
->clientfd
);
455 ret
= send(client
->clientfd
, &msg
, sizeof(msg
), 0);
457 if (errno
== EINTR
) {
458 TRACE("Caught EINTR; retrying");
462 SYSERROR("Failed to send message to client");
465 /* kick client from list */
467 close(client
->clientfd
);
475 static int lxc_serve_state_socket_pair(const char *name
,
476 struct lxc_handler
*handler
,
481 if (!handler
->backgrounded
||
482 handler
->state_socket_pair
[1] < 0 ||
486 /* Close read end of the socket pair. */
487 close(handler
->state_socket_pair
[0]);
488 handler
->state_socket_pair
[0] = -1;
491 ret
= lxc_abstract_unix_send_credential(handler
->state_socket_pair
[1],
492 &(int){state
}, sizeof(int));
493 if (ret
!= sizeof(int)) {
496 SYSERROR("Failed to send state to %d",
497 handler
->state_socket_pair
[1]);
501 TRACE("Sent container state \"%s\" to %d", lxc_state2str(state
),
502 handler
->state_socket_pair
[1]);
504 /* Close write end of the socket pair. */
505 close(handler
->state_socket_pair
[1]);
506 handler
->state_socket_pair
[1] = -1;
511 int lxc_set_state(const char *name
, struct lxc_handler
*handler
,
516 ret
= lxc_serve_state_socket_pair(name
, handler
, state
);
518 ERROR("Failed to synchronize via anonymous pair of unix sockets");
522 ret
= lxc_serve_state_clients(name
, handler
, state
);
526 /* This function will try to connect to the legacy lxc-monitord state
527 * server and only exists for backwards compatibility.
529 lxc_monitor_send_state(name
, state
, handler
->lxcpath
);
534 int lxc_poll(const char *name
, struct lxc_handler
*handler
)
537 bool has_console
= true;
538 struct lxc_epoll_descr descr
, descr_console
;
540 if (handler
->conf
->console
.path
&&
541 strcmp(handler
->conf
->console
.path
, "none") == 0)
544 ret
= lxc_mainloop_open(&descr
);
546 ERROR("Failed to create mainloop");
551 ret
= lxc_mainloop_open(&descr_console
);
553 ERROR("Failed to create console mainloop");
558 ret
= lxc_mainloop_add_handler(&descr
, handler
->sigfd
, signal_handler
, handler
);
560 ERROR("Failed to add signal handler for %d to mainloop", handler
->sigfd
);
561 goto out_mainloop_console
;
565 struct lxc_terminal
*console
= &handler
->conf
->console
;
567 ret
= lxc_terminal_mainloop_add(&descr
, console
);
569 ERROR("Failed to add console handlers to mainloop");
570 goto out_mainloop_console
;
573 ret
= lxc_terminal_mainloop_add(&descr_console
, console
);
575 ERROR("Failed to add console handlers to console mainloop");
576 goto out_mainloop_console
;
579 handler
->conf
->console
.descr
= &descr
;
582 ret
= lxc_cmd_mainloop_add(name
, &descr
, handler
);
584 ERROR("Failed to add command handler to mainloop");
585 goto out_mainloop_console
;
588 TRACE("Mainloop is ready");
590 ret
= lxc_mainloop(&descr
, -1);
593 if (ret
< 0 || !handler
->init_died
)
594 goto out_mainloop_console
;
597 ret
= lxc_mainloop(&descr_console
, 0);
599 out_mainloop_console
:
601 lxc_mainloop_close(&descr_console
);
602 TRACE("Closed console mainloop");
606 lxc_mainloop_close(&descr
);
607 TRACE("Closed mainloop");
610 close(handler
->sigfd
);
611 TRACE("Closed signal file descriptor %d", handler
->sigfd
);
612 handler
->sigfd
= -EBADF
;
617 void lxc_zero_handler(struct lxc_handler
*handler
)
621 memset(handler
, 0, sizeof(struct lxc_handler
));
627 for (i
= 0; i
< LXC_NS_MAX
; i
++)
628 handler
->nsfd
[i
] = -1;
630 handler
->data_sock
[0] = -1;
631 handler
->data_sock
[1] = -1;
633 handler
->state_socket_pair
[0] = -1;
634 handler
->state_socket_pair
[1] = -1;
636 handler
->sync_sock
[0] = -1;
637 handler
->sync_sock
[1] = -1;
640 void lxc_free_handler(struct lxc_handler
*handler
)
642 if (handler
->pinfd
>= 0)
643 close(handler
->pinfd
);
645 if (handler
->sigfd
>= 0)
646 close(handler
->sigfd
);
648 lxc_put_nsfds(handler
);
650 if (handler
->conf
&& handler
->conf
->reboot
== REBOOT_NONE
)
651 if (handler
->conf
->maincmd_fd
>= 0)
652 close(handler
->conf
->maincmd_fd
);
654 if (handler
->state_socket_pair
[0] >= 0)
655 close(handler
->state_socket_pair
[0]);
657 if (handler
->state_socket_pair
[1] >= 0)
658 close(handler
->state_socket_pair
[1]);
660 handler
->conf
= NULL
;
665 struct lxc_handler
*lxc_init_handler(const char *name
, struct lxc_conf
*conf
,
666 const char *lxcpath
, bool daemonize
)
669 struct lxc_handler
*handler
;
671 handler
= malloc(sizeof(*handler
));
674 memset(handler
, 0, sizeof(*handler
));
676 /* Note that am_guest_unpriv() checks the effective uid. We
677 * probably don't care if we are real root only if we are running
678 * as root so this should be fine.
680 handler
->am_root
= !am_guest_unpriv();
681 handler
->data_sock
[0] = handler
->data_sock
[1] = -1;
682 handler
->conf
= conf
;
683 handler
->lxcpath
= lxcpath
;
685 handler
->sigfd
= -EBADF
;
686 handler
->init_died
= false;
687 handler
->state_socket_pair
[0] = handler
->state_socket_pair
[1] = -1;
688 if (handler
->conf
->reboot
== REBOOT_NONE
)
689 lxc_list_init(&handler
->conf
->state_clients
);
691 for (i
= 0; i
< LXC_NS_MAX
; i
++)
692 handler
->nsfd
[i
] = -1;
694 handler
->name
= name
;
696 if (daemonize
&& handler
->conf
->reboot
== REBOOT_NONE
) {
697 /* Create socketpair() to synchronize on daemonized startup.
698 * When the container reboots we don't need to synchronize
699 * again currently so don't open another socketpair().
701 ret
= socketpair(AF_UNIX
, SOCK_STREAM
| SOCK_CLOEXEC
, 0,
702 handler
->state_socket_pair
);
704 ERROR("Failed to create anonymous pair of unix sockets");
707 TRACE("Created anonymous pair {%d,%d} of unix sockets",
708 handler
->state_socket_pair
[0],
709 handler
->state_socket_pair
[1]);
712 if (handler
->conf
->reboot
== REBOOT_NONE
) {
713 handler
->conf
->maincmd_fd
= lxc_cmd_init(name
, lxcpath
, "command");
714 if (handler
->conf
->maincmd_fd
< 0) {
715 ERROR("Failed to set up command socket");
719 TRACE("Unix domain socket %d for command server is ready",
720 handler
->conf
->maincmd_fd
);
725 lxc_free_handler(handler
);
730 int lxc_init(const char *name
, struct lxc_handler
*handler
)
733 const char *loglevel
;
734 struct lxc_conf
*conf
= handler
->conf
;
737 TRACE("Initialized LSM");
739 ret
= lxc_read_seccomp_config(conf
);
741 ERROR("Failed loading seccomp policy");
742 goto out_close_maincmd_fd
;
744 TRACE("Read seccomp policy");
746 /* Begin by setting the state to STARTING. */
747 ret
= lxc_set_state(name
, handler
, STARTING
);
749 ERROR("Failed to set state to \"%s\"", lxc_state2str(STARTING
));
750 goto out_close_maincmd_fd
;
752 TRACE("Set container state to \"STARTING\"");
754 /* Start of environment variable setup for hooks. */
755 ret
= setenv("LXC_NAME", name
, 1);
757 SYSERROR("Failed to set environment variable: LXC_NAME=%s", name
);
760 ret
= setenv("LXC_CONFIG_FILE", conf
->rcfile
, 1);
762 SYSERROR("Failed to set environment variable: "
763 "LXC_CONFIG_FILE=%s", conf
->rcfile
);
766 if (conf
->rootfs
.mount
) {
767 ret
= setenv("LXC_ROOTFS_MOUNT", conf
->rootfs
.mount
, 1);
769 SYSERROR("Failed to set environment variable: "
770 "LXC_ROOTFS_MOUNT=%s", conf
->rootfs
.mount
);
773 if (conf
->rootfs
.path
) {
774 ret
= setenv("LXC_ROOTFS_PATH", conf
->rootfs
.path
, 1);
776 SYSERROR("Failed to set environment variable: "
777 "LXC_ROOTFS_PATH=%s", conf
->rootfs
.path
);
780 if (conf
->console
.path
) {
781 ret
= setenv("LXC_CONSOLE", conf
->console
.path
, 1);
783 SYSERROR("Failed to set environment variable: "
784 "LXC_CONSOLE=%s", conf
->console
.path
);
787 if (conf
->console
.log_path
) {
788 ret
= setenv("LXC_CONSOLE_LOGPATH", conf
->console
.log_path
, 1);
790 SYSERROR("Failed to set environment variable: "
791 "LXC_CONSOLE_LOGPATH=%s", conf
->console
.log_path
);
794 if (cgns_supported()) {
795 ret
= setenv("LXC_CGNS_AWARE", "1", 1);
797 SYSERROR("Failed to set environment variable "
801 loglevel
= lxc_log_priority_to_string(lxc_log_get_level());
802 ret
= setenv("LXC_LOG_LEVEL", loglevel
, 1);
804 SYSERROR("Set environment variable LXC_LOG_LEVEL=%s",
807 if (conf
->hooks_version
== 0)
808 ret
= setenv("LXC_HOOK_VERSION", "0", 1);
810 ret
= setenv("LXC_HOOK_VERSION", "1", 1);
812 SYSERROR("Failed to set environment variable LXC_HOOK_VERSION=%u", conf
->hooks_version
);
813 /* End of environment variable setup for hooks. */
815 TRACE("Set environment variables");
817 ret
= run_lxc_hooks(name
, "pre-start", conf
, NULL
);
819 ERROR("Failed to run lxc.hook.pre-start for container \"%s\"", name
);
822 TRACE("Ran pre-start hooks");
824 /* The signal fd has to be created before forking otherwise if the child
825 * process exits before we setup the signal fd, the event will be lost
826 * and the command will be stuck.
828 handler
->sigfd
= setup_signal_fd(&handler
->oldmask
);
829 if (handler
->sigfd
< 0) {
830 ERROR("Failed to setup SIGCHLD fd handler.");
833 TRACE("Set up signal fd");
835 /* Do this after setting up signals since it might unblock SIGWINCH. */
836 ret
= lxc_terminal_setup(conf
);
838 ERROR("Failed to create console");
839 goto out_restore_sigmask
;
841 TRACE("Created console");
843 ret
= lxc_terminal_map_ids(conf
, &conf
->console
);
845 ERROR("Failed to chown console");
846 goto out_delete_terminal
;
848 TRACE("Chowned console");
850 handler
->cgroup_ops
= cgroup_init(handler
);
851 if (!handler
->cgroup_ops
) {
852 ERROR("Failed to initialize cgroup driver");
853 goto out_delete_terminal
;
855 TRACE("Initialized cgroup driver");
857 INFO("Container \"%s\" is initialized", name
);
861 lxc_terminal_delete(&handler
->conf
->console
);
863 (void)pthread_sigmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
865 lxc_delete_tty(&conf
->ttys
);
867 (void)lxc_set_state(name
, handler
, ABORTING
);
868 out_close_maincmd_fd
:
869 close(conf
->maincmd_fd
);
870 conf
->maincmd_fd
= -1;
874 void lxc_fini(const char *name
, struct lxc_handler
*handler
)
878 struct lxc_list
*cur
, *next
;
879 char *namespaces
[LXC_NS_MAX
+ 1];
880 size_t namespace_count
= 0;
881 struct cgroup_ops
*cgroup_ops
= handler
->cgroup_ops
;
883 /* The STOPPING state is there for future cleanup code which can take
886 lxc_set_state(name
, handler
, STOPPING
);
888 self
= lxc_raw_getpid();
889 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
890 if (handler
->nsfd
[i
] < 0)
893 if (handler
->conf
->hooks_version
== 0)
894 ret
= asprintf(&namespaces
[namespace_count
],
895 "%s:/proc/%d/fd/%d", ns_info
[i
].proc_name
,
896 self
, handler
->nsfd
[i
]);
898 ret
= asprintf(&namespaces
[namespace_count
],
899 "/proc/%d/fd/%d", self
, handler
->nsfd
[i
]);
901 SYSERROR("Failed to allocate memory");
905 if (handler
->conf
->hooks_version
== 0) {
910 ret
= setenv(ns_info
[i
].env_name
, namespaces
[namespace_count
], 1);
912 SYSERROR("Failed to set environment variable %s=%s",
913 ns_info
[i
].env_name
, namespaces
[namespace_count
]);
915 TRACE("Set environment variable %s=%s",
916 ns_info
[i
].env_name
, namespaces
[namespace_count
]);
920 namespaces
[namespace_count
] = NULL
;
922 if (handler
->conf
->reboot
> REBOOT_NONE
) {
923 ret
= setenv("LXC_TARGET", "reboot", 1);
925 SYSERROR("Failed to set environment variable: "
926 "LXC_TARGET=reboot");
929 if (handler
->conf
->reboot
== REBOOT_NONE
) {
930 ret
= setenv("LXC_TARGET", "stop", 1);
932 SYSERROR("Failed to set environment variable: "
936 if (handler
->conf
->hooks_version
== 0)
937 ret
= run_lxc_hooks(name
, "stop", handler
->conf
, namespaces
);
939 ret
= run_lxc_hooks(name
, "stop", handler
->conf
, NULL
);
941 ERROR("Failed to run \"lxc.hook.stop\" hook");
943 while (namespace_count
--)
944 free(namespaces
[namespace_count
]);
946 cgroup_ops
->destroy(cgroup_ops
, handler
);
947 cgroup_exit(cgroup_ops
);
949 if (handler
->conf
->reboot
== REBOOT_NONE
) {
950 /* For all new state clients simply close the command socket.
951 * This will inform all state clients that the container is
952 * STOPPED and also prevents a race between a open()/close() on
953 * the command socket causing a new process to get ECONNREFUSED
954 * because we haven't yet closed the command socket.
956 close(handler
->conf
->maincmd_fd
);
957 handler
->conf
->maincmd_fd
= -1;
958 TRACE("Closed command socket");
960 /* This function will try to connect to the legacy lxc-monitord
961 * state server and only exists for backwards compatibility.
963 lxc_monitor_send_state(name
, STOPPED
, handler
->lxcpath
);
965 /* The command socket is closed so no one can acces the command
966 * socket anymore so there's no need to lock it.
968 handler
->state
= STOPPED
;
969 TRACE("Set container state to \"STOPPED\"");
971 lxc_set_state(name
, handler
, STOPPED
);
974 ret
= run_lxc_hooks(name
, "post-stop", handler
->conf
, NULL
);
976 ERROR("Failed to run lxc.hook.post-stop for container \"%s\"", name
);
977 if (handler
->conf
->reboot
> REBOOT_NONE
) {
978 WARN("Container will be stopped instead of rebooted");
979 handler
->conf
->reboot
= REBOOT_NONE
;
981 ret
= setenv("LXC_TARGET", "stop", 1);
983 WARN("Failed to set environment variable: "
988 /* Reset mask set by setup_signal_fd. */
989 ret
= pthread_sigmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
991 SYSWARN("Failed to restore signal mask");
993 lxc_terminal_delete(&handler
->conf
->console
);
994 lxc_delete_tty(&handler
->conf
->ttys
);
996 /* The command socket is now closed, no more state clients can register
997 * themselves from now on. So free the list of state clients.
999 lxc_list_for_each_safe(cur
, &handler
->conf
->state_clients
, next
) {
1000 struct lxc_state_client
*client
= cur
->elem
;
1002 /* Keep state clients that want to be notified about reboots. */
1003 if ((handler
->conf
->reboot
> REBOOT_NONE
) &&
1004 (client
->states
[RUNNING
] == 2))
1007 /* close state client socket */
1009 close(client
->clientfd
);
1014 if (handler
->conf
->ephemeral
== 1 && handler
->conf
->reboot
!= REBOOT_REQ
)
1015 lxc_destroy_container_on_signal(handler
, name
);
1017 lxc_free_handler(handler
);
1020 void lxc_abort(const char *name
, struct lxc_handler
*handler
)
1024 lxc_set_state(name
, handler
, ABORTING
);
1026 if (handler
->pid
> 0) {
1027 ret
= kill(handler
->pid
, SIGKILL
);
1029 SYSERROR("Failed to send SIGKILL to %d", handler
->pid
);
1032 while ((ret
= waitpid(-1, &status
, 0)) > 0) {
1037 static int do_start(void *data
)
1040 char path
[PATH_MAX
];
1041 bool have_cap_setgid
;
1044 struct lxc_list
*iterator
;
1045 int devnull_fd
= -1;
1046 struct lxc_handler
*handler
= data
;
1048 lxc_sync_fini_parent(handler
);
1050 /* This prctl must be before the synchro, so if the parent dies before
1051 * we set the parent death signal, we will detect its death with the
1052 * synchro right after, otherwise we have a window where the parent can
1053 * exit before we set the pdeath signal leading to a unsupervized
1056 ret
= lxc_set_death_signal(SIGKILL
);
1058 SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL");
1059 goto out_warn_father
;
1062 ret
= lxc_ambient_caps_up();
1064 SYSERROR("Failed to raise ambient capabilities");
1065 goto out_warn_father
;
1068 ret
= pthread_sigmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
1070 SYSERROR("Failed to set signal mask");
1071 goto out_warn_father
;
1074 /* Don't leak the pinfd to the container. */
1075 if (handler
->pinfd
>= 0)
1076 close(handler
->pinfd
);
1078 ret
= lxc_sync_wait_parent(handler
, LXC_SYNC_STARTUP
);
1080 goto out_warn_father
;
1082 /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See
1083 * https://github.com/lxc/lxd/issues/1978.
1085 if ((handler
->ns_clone_flags
& (CLONE_NEWNET
| CLONE_NEWUSER
)) ==
1086 (CLONE_NEWNET
| CLONE_NEWUSER
)) {
1087 ret
= unshare(CLONE_NEWNET
);
1089 SYSERROR("Failed to unshare CLONE_NEWNET");
1090 goto out_warn_father
;
1092 INFO("Unshared CLONE_NEWNET");
1095 /* Tell the parent task it can begin to configure the container and wait
1098 ret
= lxc_sync_barrier_parent(handler
, LXC_SYNC_CONFIGURE
);
1102 ret
= lxc_network_recv_veth_names_from_parent(handler
);
1104 ERROR("Failed to receive veth names from parent");
1105 goto out_warn_father
;
1108 /* If we are in a new user namespace, become root there to have
1109 * privilege over our namespace.
1111 if (!lxc_list_empty(&handler
->conf
->id_map
)) {
1112 uid_t nsuid
= (handler
->conf
->root_nsuid_map
!= NULL
)
1114 : handler
->conf
->init_uid
;
1115 gid_t nsgid
= (handler
->conf
->root_nsgid_map
!= NULL
)
1117 : handler
->conf
->init_gid
;
1119 ret
= lxc_switch_uid_gid(nsuid
, nsgid
);
1121 goto out_warn_father
;
1123 /* Drop groups only after we switched to a valid gid in the new
1126 ret
= lxc_setgroups(0, NULL
);
1127 if (ret
< 0 && (handler
->am_root
|| errno
!= EPERM
))
1128 goto out_warn_father
;
1130 ret
= prctl(PR_SET_DUMPABLE
, 1, 0, 0, 0);
1132 goto out_warn_father
;
1134 /* set{g,u}id() clears deathsignal */
1135 ret
= lxc_set_death_signal(SIGKILL
);
1137 SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL");
1138 goto out_warn_father
;
1142 ret
= access(handler
->lxcpath
, X_OK
);
1144 print_top_failing_dir(handler
->lxcpath
);
1145 goto out_warn_father
;
1148 ret
= snprintf(path
, sizeof(path
), "%s/dev/null",
1149 handler
->conf
->rootfs
.mount
);
1150 if (ret
< 0 || ret
>= sizeof(path
))
1151 goto out_warn_father
;
1153 /* In order to checkpoint restore, we need to have everything in the
1154 * same mount namespace. However, some containers may not have a
1155 * reasonable /dev (in particular, they may not have /dev/null), so we
1156 * can't set init's std fds to /dev/null by opening it from inside the
1159 * If that's the case, fall back to using the host's /dev/null. This
1160 * means that migration won't work, but at least we won't spew output
1161 * where it isn't wanted.
1163 if (handler
->backgrounded
&& !handler
->conf
->autodev
) {
1164 ret
= access(path
, F_OK
);
1166 devnull_fd
= open_devnull();
1169 goto out_warn_father
;
1170 WARN("Using /dev/null from the host for container "
1171 "init's standard file descriptors. Migration will "
1176 /* Ask father to setup cgroups and wait for him to finish. */
1177 ret
= lxc_sync_barrier_parent(handler
, LXC_SYNC_CGROUP
);
1181 /* Unshare cgroup namespace after we have setup our cgroups. If we do it
1182 * earlier we end up with a wrong view of /proc/self/cgroup. For
1183 * example, assume we unshare(CLONE_NEWCGROUP) first, and then create
1184 * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then
1185 * /proc/self/cgroup would show us:
1189 * whereas it should actually show
1193 if (handler
->ns_clone_flags
& CLONE_NEWCGROUP
) {
1194 ret
= unshare(CLONE_NEWCGROUP
);
1196 INFO("Failed to unshare CLONE_NEWCGROUP");
1197 goto out_warn_father
;
1199 INFO("Unshared CLONE_NEWCGROUP");
1202 /* Add the requested environment variables to the current environment to
1203 * allow them to be used by the various hooks, such as the start hook
1206 lxc_list_for_each(iterator
, &handler
->conf
->environment
) {
1207 ret
= putenv((char *)iterator
->elem
);
1209 SYSERROR("Failed to set environment variable: %s",
1210 (char *)iterator
->elem
);
1211 goto out_warn_father
;
1215 /* Setup the container, ip, names, utsname, ... */
1216 ret
= lxc_setup(handler
);
1217 close(handler
->data_sock
[1]);
1218 close(handler
->data_sock
[0]);
1220 ERROR("Failed to setup container \"%s\"", handler
->name
);
1221 goto out_warn_father
;
1224 /* Set the label to change to when we exec(2) the container's init. */
1225 ret
= lsm_process_label_set(NULL
, handler
->conf
, 1, 1);
1227 goto out_warn_father
;
1229 /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it
1230 * before we aren't allowed anymore.
1232 if (handler
->conf
->no_new_privs
) {
1233 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0);
1235 SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block "
1236 "execve() gainable privileges");
1237 goto out_warn_father
;
1239 DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable "
1243 /* Some init's such as busybox will set sane tty settings on stdin,
1244 * stdout, stderr which it thinks is the console. We already set them
1245 * the way we wanted on the real terminal, and we want init to do its
1246 * setup on its console ie. the pty allocated in lxc_terminal_setup() so
1247 * make sure that that pty is stdin,stdout,stderr.
1249 if (handler
->conf
->console
.slave
>= 0) {
1250 if (handler
->backgrounded
|| !handler
->conf
->is_execute
)
1251 ret
= set_stdfds(handler
->conf
->console
.slave
);
1253 ret
= lxc_terminal_set_stdfds(handler
->conf
->console
.slave
);
1255 ERROR("Failed to redirect std{in,out,err} to pty file "
1256 "descriptor %d", handler
->conf
->console
.slave
);
1257 goto out_warn_father
;
1261 /* If we mounted a temporary proc, then unmount it now. */
1262 tmp_proc_unmount(handler
->conf
);
1264 ret
= lxc_seccomp_load(handler
->conf
);
1266 goto out_warn_father
;
1268 ret
= run_lxc_hooks(handler
->name
, "start", handler
->conf
, NULL
);
1270 ERROR("Failed to run lxc.hook.start for container \"%s\"",
1272 goto out_warn_father
;
1275 close(handler
->sigfd
);
1277 if (devnull_fd
< 0) {
1278 devnull_fd
= open_devnull();
1280 goto out_warn_father
;
1283 if (handler
->conf
->console
.slave
< 0 && handler
->backgrounded
) {
1284 ret
= set_stdfds(devnull_fd
);
1286 ERROR("Failed to redirect std{in,out,err} to \"/dev/null\"");
1287 goto out_warn_father
;
1291 if (devnull_fd
>= 0) {
1298 if (handler
->conf
->init_cwd
) {
1299 ret
= chdir(handler
->conf
->init_cwd
);
1301 SYSERROR("Could not change directory to \"%s\"",
1302 handler
->conf
->init_cwd
);
1303 goto out_warn_father
;
1307 ret
= lxc_sync_barrier_parent(handler
, LXC_SYNC_CGROUP_LIMITS
);
1309 goto out_warn_father
;
1311 /* Reset the environment variables the user requested in a clear
1315 /* Don't error out though. */
1317 SYSERROR("Failed to clear environment.");
1319 lxc_list_for_each(iterator
, &handler
->conf
->environment
) {
1320 ret
= putenv((char *)iterator
->elem
);
1322 SYSERROR("Failed to set environment variable: %s",
1323 (char *)iterator
->elem
);
1324 goto out_warn_father
;
1328 ret
= putenv("container=lxc");
1330 SYSERROR("Failed to set environment variable: container=lxc");
1331 goto out_warn_father
;
1334 if (handler
->conf
->ttys
.tty_names
) {
1335 ret
= putenv(handler
->conf
->ttys
.tty_names
);
1337 SYSERROR("Failed to set environment variable for container ptys");
1338 goto out_warn_father
;
1342 /* The container has been setup. We can now switch to an unprivileged
1345 new_uid
= handler
->conf
->init_uid
;
1346 new_gid
= handler
->conf
->init_gid
;
1348 /* If we are in a new user namespace we already dropped all groups when
1349 * we switched to root in the new user namespace further above. Only
1350 * drop groups if we can, so ensure that we have necessary privilege.
1353 have_cap_setgid
= lxc_proc_cap_is_set(CAP_SETGID
, CAP_EFFECTIVE
);
1355 have_cap_setgid
= false;
1357 if (lxc_list_empty(&handler
->conf
->id_map
) && have_cap_setgid
) {
1358 ret
= lxc_setgroups(0, NULL
);
1360 goto out_warn_father
;
1363 ret
= lxc_switch_uid_gid(new_uid
, new_gid
);
1365 goto out_warn_father
;
1367 ret
= lxc_ambient_caps_down();
1369 SYSERROR("Failed to clear ambient capabilities");
1370 goto out_warn_father
;
1373 /* After this call, we are in error because this ops should not return
1376 handler
->ops
->start(handler
, handler
->data
);
1379 /* We want the parent to know something went wrong, so we return a
1380 * special error code.
1382 lxc_sync_wake_parent(handler
, LXC_SYNC_ERROR
);
1385 if (devnull_fd
>= 0)
1391 static int lxc_recv_ttys_from_child(struct lxc_handler
*handler
)
1394 struct lxc_terminal_info
*tty
;
1396 int sock
= handler
->data_sock
[1];
1397 struct lxc_conf
*conf
= handler
->conf
;
1398 struct lxc_tty_info
*ttys
= &conf
->ttys
;
1400 if (!conf
->ttys
.max
)
1403 ttys
->tty
= malloc(sizeof(*ttys
->tty
) * ttys
->max
);
1407 for (i
= 0; i
< conf
->ttys
.max
; i
++) {
1410 ret
= lxc_abstract_unix_recv_fds(sock
, ttyfds
, 2, NULL
, 0);
1414 tty
= &ttys
->tty
[i
];
1416 tty
->master
= ttyfds
[0];
1417 tty
->slave
= ttyfds
[1];
1418 TRACE("Received pty with master fd %d and slave fd %d from "
1419 "parent", tty
->master
, tty
->slave
);
1423 SYSERROR("Failed to receive %zu ttys from child", ttys
->max
);
1425 TRACE("Received %zu ttys from child", ttys
->max
);
1430 int resolve_clone_flags(struct lxc_handler
*handler
)
1433 struct lxc_conf
*conf
= handler
->conf
;
1435 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1436 if (conf
->ns_keep
!= 0) {
1437 if ((conf
->ns_keep
& ns_info
[i
].clone_flag
) == 0)
1438 handler
->ns_clone_flags
|= ns_info
[i
].clone_flag
;
1439 } else if (conf
->ns_clone
!= 0) {
1440 if ((conf
->ns_clone
& ns_info
[i
].clone_flag
) > 0)
1441 handler
->ns_clone_flags
|= ns_info
[i
].clone_flag
;
1443 if (i
== LXC_NS_USER
&& lxc_list_empty(&handler
->conf
->id_map
))
1446 if (i
== LXC_NS_NET
&& lxc_requests_empty_network(handler
))
1449 if (i
== LXC_NS_CGROUP
&& !cgns_supported())
1452 handler
->ns_clone_flags
|= ns_info
[i
].clone_flag
;
1455 if (!conf
->ns_share
[i
])
1458 handler
->ns_clone_flags
&= ~ns_info
[i
].clone_flag
;
1459 TRACE("Sharing %s namespace", ns_info
[i
].proc_name
);
1465 /* Note that this function is used with clone(CLONE_VM). Some glibc versions
1466 * used to reset the pid/tid to -1 when CLONE_VM was used without CLONE_THREAD.
1467 * But since the memory between parent and child is shared on CLONE_VM this
1468 * would invalidate the getpid() cache that glibc used to maintain and so
1469 * getpid() in the child would return the parent's pid. This is all fixed in
1470 * newer glibc versions where the getpid() cache is removed and the pid/tid is
1471 * not reset anymore.
1472 * However, if for whatever reason you - dear commiter - somehow need to get the
1473 * pid of the dummy intermediate process for do_share_ns() you need to call
1474 * lxc_raw_getpid(). The next lxc_raw_clone() call does not employ CLONE_VM and
1477 static inline int do_share_ns(void *arg
)
1480 struct lxc_handler
*handler
= arg
;
1482 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1483 if (handler
->nsfd
[i
] < 0)
1486 ret
= setns(handler
->nsfd
[i
], 0);
1489 * Note that joining a user and/or mount namespace
1490 * requires the process is not multithreaded otherwise
1491 * setns() will fail here.
1493 SYSERROR("Failed to inherit %s namespace",
1494 ns_info
[i
].proc_name
);
1498 DEBUG("Inherited %s namespace", ns_info
[i
].proc_name
);
1501 flags
= handler
->ns_on_clone_flags
;
1502 flags
|= CLONE_PARENT
;
1503 handler
->pid
= lxc_raw_clone_cb(do_start
, handler
, flags
);
1504 if (handler
->pid
< 0)
1510 /* lxc_spawn() performs crucial setup tasks and clone()s the new process which
1511 * exec()s the requested container binary.
1512 * Note that lxc_spawn() runs in the parent namespaces. Any operations performed
1513 * right here should be double checked if they'd pose a security risk. (For
1514 * example, any {u}mount() operations performed here will be reflected on the
1517 static int lxc_spawn(struct lxc_handler
*handler
)
1521 bool wants_to_map_ids
;
1522 struct lxc_list
*id_map
;
1523 const char *name
= handler
->name
;
1524 const char *lxcpath
= handler
->lxcpath
;
1525 bool share_ns
= false;
1526 struct lxc_conf
*conf
= handler
->conf
;
1527 struct cgroup_ops
*cgroup_ops
= handler
->cgroup_ops
;
1529 id_map
= &conf
->id_map
;
1530 wants_to_map_ids
= !lxc_list_empty(id_map
);
1532 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1533 if (!conf
->ns_share
[i
])
1536 handler
->nsfd
[i
] = lxc_inherit_namespace(conf
->ns_share
[i
], lxcpath
, ns_info
[i
].proc_name
);
1537 if (handler
->nsfd
[i
] < 0)
1543 ret
= lxc_sync_init(handler
);
1547 ret
= socketpair(AF_UNIX
, SOCK_STREAM
| SOCK_CLOEXEC
, 0,
1548 handler
->data_sock
);
1550 lxc_sync_fini(handler
);
1554 ret
= resolve_clone_flags(handler
);
1556 lxc_sync_fini(handler
);
1560 if (handler
->ns_clone_flags
& CLONE_NEWNET
) {
1561 if (!lxc_list_empty(&conf
->network
)) {
1563 /* Find gateway addresses from the link device, which is
1564 * no longer accessible inside the container. Do this
1565 * before creating network interfaces, since goto
1566 * out_delete_net does not work before lxc_clone.
1568 ret
= lxc_find_gateway_addresses(handler
);
1570 ERROR("Failed to find gateway addresses");
1571 lxc_sync_fini(handler
);
1575 /* That should be done before the clone because we will
1576 * fill the netdev index and use them in the child.
1578 ret
= lxc_create_network_priv(handler
);
1580 ERROR("Failed to create the network");
1581 lxc_sync_fini(handler
);
1587 if (!cgroup_ops
->create(cgroup_ops
, handler
)) {
1588 ERROR("Failed creating cgroups");
1589 goto out_delete_net
;
1592 /* If the rootfs is not a blockdev, prevent the container from marking
1594 * If the container is unprivileged then skip rootfs pinning.
1596 if (!wants_to_map_ids
) {
1597 handler
->pinfd
= pin_rootfs(conf
->rootfs
.path
);
1598 if (handler
->pinfd
== -1)
1599 INFO("Failed to pin the rootfs for container \"%s\"", handler
->name
);
1602 /* Create a process in a new set of namespaces. */
1603 handler
->ns_on_clone_flags
= handler
->ns_clone_flags
;
1604 if (handler
->ns_clone_flags
& CLONE_NEWUSER
) {
1605 /* If CLONE_NEWUSER and CLONE_NEWNET was requested, we need to
1606 * clone a new user namespace first and only later unshare our
1607 * network namespace to ensure that network devices ownership is
1610 handler
->ns_on_clone_flags
&= ~CLONE_NEWNET
;
1612 /* The cgroup namespace gets unshare()ed not clone()ed. */
1613 handler
->ns_on_clone_flags
&= ~CLONE_NEWCGROUP
;
1618 attacher_pid
= lxc_clone(do_share_ns
, handler
,
1619 CLONE_VFORK
| CLONE_VM
| CLONE_FILES
);
1620 if (attacher_pid
< 0) {
1621 SYSERROR(LXC_CLONE_ERROR
);
1622 goto out_delete_net
;
1625 ret
= wait_for_pid(attacher_pid
);
1627 SYSERROR("Intermediate process failed");
1628 goto out_delete_net
;
1631 handler
->pid
= lxc_raw_clone_cb(do_start
, handler
,
1632 handler
->ns_on_clone_flags
);
1634 if (handler
->pid
< 0) {
1635 SYSERROR(LXC_CLONE_ERROR
);
1636 goto out_delete_net
;
1638 TRACE("Cloned child process %d", handler
->pid
);
1640 for (i
= 0; i
< LXC_NS_MAX
; i
++)
1641 if (handler
->ns_on_clone_flags
& ns_info
[i
].clone_flag
)
1642 INFO("Cloned %s", ns_info
[i
].flag_name
);
1644 if (!lxc_try_preserve_namespaces(handler
, handler
->ns_on_clone_flags
, handler
->pid
)) {
1645 ERROR("Failed to preserve cloned namespaces for lxc.hook.stop");
1646 goto out_delete_net
;
1649 lxc_sync_fini_child(handler
);
1651 /* Map the container uids. The container became an invalid userid the
1652 * moment it was cloned with CLONE_NEWUSER. This call doesn't change
1653 * anything immediately, but allows the container to setuid(0) (0 being
1654 * mapped to something else on the host.) later to become a valid uid
1657 if (wants_to_map_ids
) {
1658 if (!handler
->conf
->ns_share
[LXC_NS_USER
] &&
1659 (handler
->conf
->ns_keep
& CLONE_NEWUSER
) == 0) {
1660 ret
= lxc_map_ids(id_map
, handler
->pid
);
1662 ERROR("Failed to set up id mapping.");
1663 goto out_delete_net
;
1668 ret
= lxc_sync_wake_child(handler
, LXC_SYNC_STARTUP
);
1670 goto out_delete_net
;
1672 ret
= lxc_sync_wait_child(handler
, LXC_SYNC_CONFIGURE
);
1674 goto out_delete_net
;
1676 if (!cgroup_ops
->setup_limits(cgroup_ops
, handler
->conf
, false)) {
1677 ERROR("Failed to setup cgroup limits for container \"%s\"", name
);
1678 goto out_delete_net
;
1681 if (!cgroup_ops
->enter(cgroup_ops
, handler
->pid
))
1682 goto out_delete_net
;
1684 if (!cgroup_ops
->chown(cgroup_ops
, handler
->conf
))
1685 goto out_delete_net
;
1687 /* Now we're ready to preserve the network namespace */
1688 ret
= lxc_try_preserve_ns(handler
->pid
, "net");
1690 if (ret
!= -EOPNOTSUPP
) {
1691 SYSERROR("Failed to preserve net namespace");
1692 goto out_delete_net
;
1695 handler
->nsfd
[LXC_NS_NET
] = ret
;
1696 DEBUG("Preserved net namespace via fd %d", ret
);
1699 /* Create the network configuration. */
1700 if (handler
->ns_clone_flags
& CLONE_NEWNET
) {
1701 ret
= lxc_network_move_created_netdev_priv(handler
->lxcpath
,
1706 ERROR("Failed to create the configured network");
1707 goto out_delete_net
;
1710 ret
= lxc_create_network_unpriv(handler
->lxcpath
, handler
->name
,
1711 &conf
->network
, handler
->pid
, conf
->hooks_version
);
1713 ERROR("Failed to create the configured network");
1714 goto out_delete_net
;
1718 ret
= lxc_network_send_veth_names_to_child(handler
);
1720 ERROR("Failed to send veth names to child");
1721 goto out_delete_net
;
1724 if (!lxc_list_empty(&conf
->procs
)) {
1725 ret
= setup_proc_filesystem(&conf
->procs
, handler
->pid
);
1727 goto out_delete_net
;
1730 /* Tell the child to continue its initialization. We'll get
1731 * LXC_SYNC_CGROUP when it is ready for us to setup cgroups.
1733 ret
= lxc_sync_barrier_child(handler
, LXC_SYNC_POST_CONFIGURE
);
1735 goto out_delete_net
;
1737 if (!lxc_list_empty(&conf
->limits
)) {
1738 ret
= setup_resource_limits(&conf
->limits
, handler
->pid
);
1740 ERROR("Failed to setup resource limits");
1741 goto out_delete_net
;
1745 ret
= lxc_sync_barrier_child(handler
, LXC_SYNC_CGROUP_UNSHARE
);
1747 goto out_delete_net
;
1749 if (!cgroup_ops
->setup_limits(cgroup_ops
, handler
->conf
, true)) {
1750 ERROR("Failed to setup legacy device cgroup controller limits");
1751 goto out_delete_net
;
1753 TRACE("Set up legacy device cgroup controller limits");
1755 if (handler
->ns_clone_flags
& CLONE_NEWCGROUP
) {
1756 /* Now we're ready to preserve the cgroup namespace */
1757 ret
= lxc_try_preserve_ns(handler
->pid
, "cgroup");
1759 if (ret
!= -EOPNOTSUPP
) {
1760 SYSERROR("Failed to preserve cgroup namespace");
1761 goto out_delete_net
;
1764 handler
->nsfd
[LXC_NS_CGROUP
] = ret
;
1765 DEBUG("Preserved cgroup namespace via fd %d", ret
);
1769 ret
= snprintf(pidstr
, 20, "%d", handler
->pid
);
1770 if (ret
< 0 || ret
>= 20)
1771 goto out_delete_net
;
1773 ret
= setenv("LXC_PID", pidstr
, 1);
1775 SYSERROR("Failed to set environment variable: LXC_PID=%s", pidstr
);
1777 /* Run any host-side start hooks */
1778 ret
= run_lxc_hooks(name
, "start-host", conf
, NULL
);
1780 ERROR("Failed to run lxc.hook.start-host");
1781 goto out_delete_net
;
1784 /* Tell the child to complete its initialization and wait for it to exec
1785 * or return an error. (The child will never return
1786 * LXC_SYNC_READY_START+1. It will either close the sync pipe, causing
1787 * lxc_sync_barrier_child to return success, or return a different
1788 * value, causing us to error out).
1790 ret
= lxc_sync_barrier_child(handler
, LXC_SYNC_READY_START
);
1792 goto out_delete_net
;
1794 ret
= lxc_network_recv_name_and_ifindex_from_child(handler
);
1796 ERROR("Failed to receive names and ifindices for network "
1797 "devices from child");
1798 goto out_delete_net
;
1801 /* Now all networks are created, network devices are moved into place,
1802 * and the correct names and ifindeces in the respective namespaces have
1803 * been recorded. The corresponding structs have now all been filled. So
1804 * log them for debugging purposes.
1806 lxc_log_configured_netdevs(conf
);
1808 /* Read tty fds allocated by child. */
1809 ret
= lxc_recv_ttys_from_child(handler
);
1811 ERROR("Failed to receive tty info from child process");
1812 goto out_delete_net
;
1815 ret
= handler
->ops
->post_start(handler
, handler
->data
);
1819 ret
= lxc_set_state(name
, handler
, RUNNING
);
1821 ERROR("Failed to set state to \"%s\"", lxc_state2str(RUNNING
));
1825 lxc_sync_fini(handler
);
1830 if (handler
->ns_clone_flags
& CLONE_NEWNET
)
1831 lxc_delete_network(handler
);
1834 lxc_abort(name
, handler
);
1835 lxc_sync_fini(handler
);
1836 if (handler
->pinfd
>= 0) {
1837 close(handler
->pinfd
);
1838 handler
->pinfd
= -1;
1844 int __lxc_start(const char *name
, struct lxc_handler
*handler
,
1845 struct lxc_operations
* ops
, void *data
, const char *lxcpath
,
1846 bool backgrounded
, int *error_num
)
1849 struct lxc_conf
*conf
= handler
->conf
;
1851 ret
= lxc_init(name
, handler
);
1853 ERROR("Failed to initialize container \"%s\"", name
);
1857 handler
->data
= data
;
1858 handler
->backgrounded
= backgrounded
;
1860 if (!attach_block_device(handler
->conf
)) {
1861 ERROR("Failed to attach block device");
1862 goto out_fini_nonet
;
1865 if (geteuid() == 0 && !lxc_list_empty(&conf
->id_map
)) {
1866 /* If the backing store is a device, mount it here and now. */
1867 if (rootfs_is_blockdev(conf
)) {
1868 ret
= unshare(CLONE_NEWNS
);
1870 ERROR("Failed to unshare CLONE_NEWNS");
1871 goto out_fini_nonet
;
1873 INFO("Unshared CLONE_NEWNS");
1875 remount_all_slave();
1876 ret
= do_rootfs_setup(conf
, name
, lxcpath
);
1878 ERROR("Error setting up rootfs mount as root before spawn");
1879 goto out_fini_nonet
;
1881 INFO("Set up container rootfs as host root");
1885 ret
= lxc_spawn(handler
);
1887 ERROR("Failed to spawn container \"%s\"", name
);
1888 goto out_detach_blockdev
;
1890 /* close parent side of data socket */
1891 close(handler
->data_sock
[0]);
1892 handler
->data_sock
[0] = -1;
1893 close(handler
->data_sock
[1]);
1894 handler
->data_sock
[1] = -1;
1896 handler
->conf
->reboot
= REBOOT_NONE
;
1898 ret
= lxc_poll(name
, handler
);
1900 ERROR("LXC mainloop exited with error: %d", ret
);
1904 if (!handler
->init_died
&& handler
->pid
> 0) {
1905 ERROR("Child process is not killed");
1909 status
= lxc_wait_for_pid_status(handler
->pid
);
1911 SYSERROR("Failed to retrieve status for %d", handler
->pid
);
1913 /* If the child process exited but was not signaled, it didn't call
1914 * reboot. This should mean it was an lxc-execute which simply exited.
1915 * In any case, treat it as a 'halt'.
1917 if (WIFSIGNALED(status
)) {
1918 switch(WTERMSIG(status
)) {
1919 case SIGINT
: /* halt */
1920 DEBUG("Container \"%s\" is halting", name
);
1922 case SIGHUP
: /* reboot */
1923 DEBUG("Container \"%s\" is rebooting", name
);
1924 handler
->conf
->reboot
= REBOOT_REQ
;
1926 case SIGSYS
: /* seccomp */
1927 DEBUG("Container \"%s\" violated its seccomp policy", name
);
1930 DEBUG("Unknown exit status for container \"%s\" init %d", name
, WTERMSIG(status
));
1935 ret
= lxc_restore_phys_nics_to_netns(handler
);
1937 ERROR("Failed to move physical network devices back to parent "
1938 "network namespace");
1940 if (handler
->pinfd
>= 0) {
1941 close(handler
->pinfd
);
1942 handler
->pinfd
= -1;
1945 lxc_monitor_send_exit_code(name
, status
, handler
->lxcpath
);
1946 lxc_error_set_and_log(handler
->pid
, status
);
1948 *error_num
= handler
->exit_status
;
1951 lxc_delete_network(handler
);
1953 out_detach_blockdev
:
1954 detach_block_device(handler
->conf
);
1957 lxc_fini(name
, handler
);
1961 lxc_abort(name
, handler
);
1969 static int start(struct lxc_handler
*handler
, void* data
)
1971 struct start_args
*arg
= data
;
1973 NOTICE("Exec'ing \"%s\"", arg
->argv
[0]);
1975 execvp(arg
->argv
[0], arg
->argv
);
1976 SYSERROR("Failed to exec \"%s\"", arg
->argv
[0]);
1980 static int post_start(struct lxc_handler
*handler
, void* data
)
1982 struct start_args
*arg
= data
;
1984 NOTICE("Started \"%s\" with pid \"%d\"", arg
->argv
[0], handler
->pid
);
1988 static struct lxc_operations start_ops
= {
1990 .post_start
= post_start
1993 int lxc_start(const char *name
, char *const argv
[], struct lxc_handler
*handler
,
1994 const char *lxcpath
, bool backgrounded
, int *error_num
)
1996 struct start_args start_arg
= {
2000 TRACE("Doing lxc_start");
2001 return __lxc_start(name
, handler
, &start_ops
, &start_arg
, lxcpath
, backgrounded
, error_num
);
2004 static void lxc_destroy_container_on_signal(struct lxc_handler
*handler
,
2007 char destroy
[MAXPATHLEN
];
2008 struct lxc_container
*c
;
2012 if (handler
->conf
->rootfs
.path
&& handler
->conf
->rootfs
.mount
) {
2013 bret
= do_destroy_container(handler
);
2015 ERROR("Error destroying rootfs for container \"%s\"", name
);
2019 INFO("Destroyed rootfs for container \"%s\"", name
);
2021 ret
= snprintf(destroy
, MAXPATHLEN
, "%s/%s", handler
->lxcpath
, name
);
2022 if (ret
< 0 || ret
>= MAXPATHLEN
) {
2023 ERROR("Error destroying directory for container \"%s\"", name
);
2027 c
= lxc_container_new(name
, handler
->lxcpath
);
2029 if (container_disk_lock(c
)) {
2030 INFO("Could not update lxc_snapshots file");
2031 lxc_container_put(c
);
2033 mod_all_rdeps(c
, false);
2034 container_disk_unlock(c
);
2035 lxc_container_put(c
);
2039 if (!handler
->am_root
)
2040 ret
= userns_exec_full(handler
->conf
, lxc_rmdir_onedev_wrapper
,
2041 destroy
, "lxc_rmdir_onedev_wrapper");
2043 ret
= lxc_rmdir_onedev(destroy
, NULL
);
2046 ERROR("Error destroying directory for container \"%s\"", name
);
2049 INFO("Destroyed directory for container \"%s\"", name
);
2052 static int lxc_rmdir_onedev_wrapper(void *data
)
2054 char *arg
= (char *) data
;
2055 return lxc_rmdir_onedev(arg
, NULL
);
2058 static bool do_destroy_container(struct lxc_handler
*handler
)
2062 if (!handler
->am_root
) {
2063 ret
= userns_exec_full(handler
->conf
, storage_destroy_wrapper
,
2064 handler
->conf
, "storage_destroy_wrapper");
2071 return storage_destroy(handler
->conf
);