]>
git.proxmox.com Git - mirror_lxc.git/blob - src/lxc/start.c
2485472e257de670e64a0ac08c59d70766adbc4a
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
), MSG_NOSIGNAL
);
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));
494 SYSERROR("Failed to send state to %d", handler
->state_socket_pair
[1]);
502 if (ret
!= sizeof(int)) {
503 ERROR("Message too long : %d", handler
->state_socket_pair
[1]);
507 TRACE("Sent container state \"%s\" to %d", lxc_state2str(state
),
508 handler
->state_socket_pair
[1]);
510 /* Close write end of the socket pair. */
511 close(handler
->state_socket_pair
[1]);
512 handler
->state_socket_pair
[1] = -1;
517 int lxc_set_state(const char *name
, struct lxc_handler
*handler
,
522 ret
= lxc_serve_state_socket_pair(name
, handler
, state
);
524 ERROR("Failed to synchronize via anonymous pair of unix sockets");
528 ret
= lxc_serve_state_clients(name
, handler
, state
);
532 /* This function will try to connect to the legacy lxc-monitord state
533 * server and only exists for backwards compatibility.
535 lxc_monitor_send_state(name
, state
, handler
->lxcpath
);
540 int lxc_poll(const char *name
, struct lxc_handler
*handler
)
543 bool has_console
= true;
544 struct lxc_epoll_descr descr
, descr_console
;
546 if (handler
->conf
->console
.path
&&
547 strcmp(handler
->conf
->console
.path
, "none") == 0)
550 ret
= lxc_mainloop_open(&descr
);
552 ERROR("Failed to create mainloop");
557 ret
= lxc_mainloop_open(&descr_console
);
559 ERROR("Failed to create console mainloop");
564 ret
= lxc_mainloop_add_handler(&descr
, handler
->sigfd
, signal_handler
, handler
);
566 ERROR("Failed to add signal handler for %d to mainloop", handler
->sigfd
);
567 goto out_mainloop_console
;
571 struct lxc_terminal
*console
= &handler
->conf
->console
;
573 ret
= lxc_terminal_mainloop_add(&descr
, console
);
575 ERROR("Failed to add console handlers to mainloop");
576 goto out_mainloop_console
;
579 ret
= lxc_terminal_mainloop_add(&descr_console
, console
);
581 ERROR("Failed to add console handlers to console mainloop");
582 goto out_mainloop_console
;
585 handler
->conf
->console
.descr
= &descr
;
588 ret
= lxc_cmd_mainloop_add(name
, &descr
, handler
);
590 ERROR("Failed to add command handler to mainloop");
591 goto out_mainloop_console
;
594 TRACE("Mainloop is ready");
596 ret
= lxc_mainloop(&descr
, -1);
599 if (ret
< 0 || !handler
->init_died
)
600 goto out_mainloop_console
;
603 ret
= lxc_mainloop(&descr_console
, 0);
605 out_mainloop_console
:
607 lxc_mainloop_close(&descr_console
);
608 TRACE("Closed console mainloop");
612 lxc_mainloop_close(&descr
);
613 TRACE("Closed mainloop");
616 close(handler
->sigfd
);
617 TRACE("Closed signal file descriptor %d", handler
->sigfd
);
618 handler
->sigfd
= -EBADF
;
623 void lxc_zero_handler(struct lxc_handler
*handler
)
627 memset(handler
, 0, sizeof(struct lxc_handler
));
633 for (i
= 0; i
< LXC_NS_MAX
; i
++)
634 handler
->nsfd
[i
] = -1;
636 handler
->data_sock
[0] = -1;
637 handler
->data_sock
[1] = -1;
639 handler
->state_socket_pair
[0] = -1;
640 handler
->state_socket_pair
[1] = -1;
642 handler
->sync_sock
[0] = -1;
643 handler
->sync_sock
[1] = -1;
646 void lxc_free_handler(struct lxc_handler
*handler
)
648 if (handler
->pinfd
>= 0)
649 close(handler
->pinfd
);
651 if (handler
->sigfd
>= 0)
652 close(handler
->sigfd
);
654 lxc_put_nsfds(handler
);
656 if (handler
->conf
&& handler
->conf
->reboot
== REBOOT_NONE
)
657 if (handler
->conf
->maincmd_fd
>= 0)
658 lxc_abstract_unix_close(handler
->conf
->maincmd_fd
);
660 if (handler
->state_socket_pair
[0] >= 0)
661 close(handler
->state_socket_pair
[0]);
663 if (handler
->state_socket_pair
[1] >= 0)
664 close(handler
->state_socket_pair
[1]);
666 handler
->conf
= NULL
;
671 struct lxc_handler
*lxc_init_handler(const char *name
, struct lxc_conf
*conf
,
672 const char *lxcpath
, bool daemonize
)
675 struct lxc_handler
*handler
;
677 handler
= malloc(sizeof(*handler
));
681 memset(handler
, 0, sizeof(*handler
));
683 /* Note that am_guest_unpriv() checks the effective uid. We
684 * probably don't care if we are real root only if we are running
685 * as root so this should be fine.
687 handler
->am_root
= !am_guest_unpriv();
688 handler
->data_sock
[0] = handler
->data_sock
[1] = -1;
689 handler
->conf
= conf
;
690 handler
->lxcpath
= lxcpath
;
692 handler
->sigfd
= -EBADF
;
693 handler
->init_died
= false;
694 handler
->state_socket_pair
[0] = handler
->state_socket_pair
[1] = -1;
695 if (handler
->conf
->reboot
== REBOOT_NONE
)
696 lxc_list_init(&handler
->conf
->state_clients
);
698 for (i
= 0; i
< LXC_NS_MAX
; i
++)
699 handler
->nsfd
[i
] = -1;
701 handler
->name
= name
;
703 if (daemonize
&& handler
->conf
->reboot
== REBOOT_NONE
) {
704 /* Create socketpair() to synchronize on daemonized startup.
705 * When the container reboots we don't need to synchronize
706 * again currently so don't open another socketpair().
708 ret
= socketpair(AF_UNIX
, SOCK_STREAM
| SOCK_CLOEXEC
, 0,
709 handler
->state_socket_pair
);
711 ERROR("Failed to create anonymous pair of unix sockets");
715 TRACE("Created anonymous pair {%d,%d} of unix sockets",
716 handler
->state_socket_pair
[0],
717 handler
->state_socket_pair
[1]);
720 if (handler
->conf
->reboot
== REBOOT_NONE
) {
721 handler
->conf
->maincmd_fd
= lxc_cmd_init(name
, lxcpath
, "command");
722 if (handler
->conf
->maincmd_fd
< 0) {
723 ERROR("Failed to set up command socket");
728 TRACE("Unix domain socket %d for command server is ready",
729 handler
->conf
->maincmd_fd
);
734 lxc_free_handler(handler
);
739 int lxc_init(const char *name
, struct lxc_handler
*handler
)
742 const char *loglevel
;
743 struct lxc_conf
*conf
= handler
->conf
;
746 TRACE("Initialized LSM");
748 ret
= lxc_read_seccomp_config(conf
);
750 ERROR("Failed loading seccomp policy");
751 goto out_close_maincmd_fd
;
753 TRACE("Read seccomp policy");
755 /* Begin by setting the state to STARTING. */
756 ret
= lxc_set_state(name
, handler
, STARTING
);
758 ERROR("Failed to set state to \"%s\"", lxc_state2str(STARTING
));
759 goto out_close_maincmd_fd
;
761 TRACE("Set container state to \"STARTING\"");
763 /* Start of environment variable setup for hooks. */
764 ret
= setenv("LXC_NAME", name
, 1);
766 SYSERROR("Failed to set environment variable: LXC_NAME=%s", name
);
769 ret
= setenv("LXC_CONFIG_FILE", conf
->rcfile
, 1);
771 SYSERROR("Failed to set environment variable: "
772 "LXC_CONFIG_FILE=%s", conf
->rcfile
);
775 if (conf
->rootfs
.mount
) {
776 ret
= setenv("LXC_ROOTFS_MOUNT", conf
->rootfs
.mount
, 1);
778 SYSERROR("Failed to set environment variable: "
779 "LXC_ROOTFS_MOUNT=%s", conf
->rootfs
.mount
);
782 if (conf
->rootfs
.path
) {
783 ret
= setenv("LXC_ROOTFS_PATH", conf
->rootfs
.path
, 1);
785 SYSERROR("Failed to set environment variable: "
786 "LXC_ROOTFS_PATH=%s", conf
->rootfs
.path
);
789 if (conf
->console
.path
) {
790 ret
= setenv("LXC_CONSOLE", conf
->console
.path
, 1);
792 SYSERROR("Failed to set environment variable: "
793 "LXC_CONSOLE=%s", conf
->console
.path
);
796 if (conf
->console
.log_path
) {
797 ret
= setenv("LXC_CONSOLE_LOGPATH", conf
->console
.log_path
, 1);
799 SYSERROR("Failed to set environment variable: "
800 "LXC_CONSOLE_LOGPATH=%s", conf
->console
.log_path
);
803 if (cgns_supported()) {
804 ret
= setenv("LXC_CGNS_AWARE", "1", 1);
806 SYSERROR("Failed to set environment variable "
810 loglevel
= lxc_log_priority_to_string(lxc_log_get_level());
811 ret
= setenv("LXC_LOG_LEVEL", loglevel
, 1);
813 SYSERROR("Set environment variable LXC_LOG_LEVEL=%s",
816 if (conf
->hooks_version
== 0)
817 ret
= setenv("LXC_HOOK_VERSION", "0", 1);
819 ret
= setenv("LXC_HOOK_VERSION", "1", 1);
821 SYSERROR("Failed to set environment variable LXC_HOOK_VERSION=%u", conf
->hooks_version
);
822 /* End of environment variable setup for hooks. */
824 TRACE("Set environment variables");
826 ret
= run_lxc_hooks(name
, "pre-start", conf
, NULL
);
828 ERROR("Failed to run lxc.hook.pre-start for container \"%s\"", name
);
831 TRACE("Ran pre-start hooks");
833 /* The signal fd has to be created before forking otherwise if the child
834 * process exits before we setup the signal fd, the event will be lost
835 * and the command will be stuck.
837 handler
->sigfd
= setup_signal_fd(&handler
->oldmask
);
838 if (handler
->sigfd
< 0) {
839 ERROR("Failed to setup SIGCHLD fd handler.");
842 TRACE("Set up signal fd");
844 /* Do this after setting up signals since it might unblock SIGWINCH. */
845 ret
= lxc_terminal_setup(conf
);
847 ERROR("Failed to create console");
848 goto out_restore_sigmask
;
850 TRACE("Created console");
852 ret
= lxc_terminal_map_ids(conf
, &conf
->console
);
854 ERROR("Failed to chown console");
855 goto out_delete_terminal
;
857 TRACE("Chowned console");
859 handler
->cgroup_ops
= cgroup_init(handler
);
860 if (!handler
->cgroup_ops
) {
861 ERROR("Failed to initialize cgroup driver");
862 goto out_delete_terminal
;
864 TRACE("Initialized cgroup driver");
866 ret
= lsm_process_prepare(conf
, handler
->lxcpath
);
868 ERROR("Failed to initialize LSM");
869 goto out_destroy_cgroups
;
871 TRACE("Initialized LSM");
873 INFO("Container \"%s\" is initialized", name
);
877 handler
->cgroup_ops
->destroy(handler
->cgroup_ops
, handler
);
880 lxc_terminal_delete(&handler
->conf
->console
);
883 (void)pthread_sigmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
886 lxc_delete_tty(&conf
->ttys
);
889 (void)lxc_set_state(name
, handler
, ABORTING
);
891 out_close_maincmd_fd
:
892 lxc_abstract_unix_close(conf
->maincmd_fd
);
893 conf
->maincmd_fd
= -1;
897 void lxc_fini(const char *name
, struct lxc_handler
*handler
)
901 struct lxc_list
*cur
, *next
;
902 char *namespaces
[LXC_NS_MAX
+ 1];
903 size_t namespace_count
= 0;
904 struct cgroup_ops
*cgroup_ops
= handler
->cgroup_ops
;
906 /* The STOPPING state is there for future cleanup code which can take
909 lxc_set_state(name
, handler
, STOPPING
);
911 self
= lxc_raw_getpid();
912 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
913 if (handler
->nsfd
[i
] < 0)
916 if (handler
->conf
->hooks_version
== 0)
917 ret
= asprintf(&namespaces
[namespace_count
],
918 "%s:/proc/%d/fd/%d", ns_info
[i
].proc_name
,
919 self
, handler
->nsfd
[i
]);
921 ret
= asprintf(&namespaces
[namespace_count
],
922 "/proc/%d/fd/%d", self
, handler
->nsfd
[i
]);
924 SYSERROR("Failed to allocate memory");
928 if (handler
->conf
->hooks_version
== 0) {
933 ret
= setenv(ns_info
[i
].env_name
, namespaces
[namespace_count
], 1);
935 SYSERROR("Failed to set environment variable %s=%s",
936 ns_info
[i
].env_name
, namespaces
[namespace_count
]);
938 TRACE("Set environment variable %s=%s",
939 ns_info
[i
].env_name
, namespaces
[namespace_count
]);
943 namespaces
[namespace_count
] = NULL
;
945 if (handler
->conf
->reboot
> REBOOT_NONE
) {
946 ret
= setenv("LXC_TARGET", "reboot", 1);
948 SYSERROR("Failed to set environment variable: "
949 "LXC_TARGET=reboot");
952 if (handler
->conf
->reboot
== REBOOT_NONE
) {
953 ret
= setenv("LXC_TARGET", "stop", 1);
955 SYSERROR("Failed to set environment variable: "
959 if (handler
->conf
->hooks_version
== 0)
960 ret
= run_lxc_hooks(name
, "stop", handler
->conf
, namespaces
);
962 ret
= run_lxc_hooks(name
, "stop", handler
->conf
, NULL
);
964 ERROR("Failed to run \"lxc.hook.stop\" hook");
966 while (namespace_count
--)
967 free(namespaces
[namespace_count
]);
969 lsm_process_cleanup(handler
->conf
, handler
->lxcpath
);
971 cgroup_ops
->destroy(cgroup_ops
, handler
);
972 cgroup_exit(cgroup_ops
);
974 if (handler
->conf
->reboot
== REBOOT_NONE
) {
975 /* For all new state clients simply close the command socket.
976 * This will inform all state clients that the container is
977 * STOPPED and also prevents a race between a open()/close() on
978 * the command socket causing a new process to get ECONNREFUSED
979 * because we haven't yet closed the command socket.
981 lxc_abstract_unix_close(handler
->conf
->maincmd_fd
);
982 handler
->conf
->maincmd_fd
= -1;
983 TRACE("Closed command socket");
985 /* This function will try to connect to the legacy lxc-monitord
986 * state server and only exists for backwards compatibility.
988 lxc_monitor_send_state(name
, STOPPED
, handler
->lxcpath
);
990 /* The command socket is closed so no one can acces the command
991 * socket anymore so there's no need to lock it.
993 handler
->state
= STOPPED
;
994 TRACE("Set container state to \"STOPPED\"");
996 lxc_set_state(name
, handler
, STOPPED
);
999 ret
= run_lxc_hooks(name
, "post-stop", handler
->conf
, NULL
);
1001 ERROR("Failed to run lxc.hook.post-stop for container \"%s\"", name
);
1002 if (handler
->conf
->reboot
> REBOOT_NONE
) {
1003 WARN("Container will be stopped instead of rebooted");
1004 handler
->conf
->reboot
= REBOOT_NONE
;
1006 ret
= setenv("LXC_TARGET", "stop", 1);
1008 WARN("Failed to set environment variable: "
1013 /* Reset mask set by setup_signal_fd. */
1014 ret
= pthread_sigmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
1016 SYSWARN("Failed to restore signal mask");
1018 lxc_terminal_delete(&handler
->conf
->console
);
1019 lxc_delete_tty(&handler
->conf
->ttys
);
1021 /* The command socket is now closed, no more state clients can register
1022 * themselves from now on. So free the list of state clients.
1024 lxc_list_for_each_safe(cur
, &handler
->conf
->state_clients
, next
) {
1025 struct lxc_state_client
*client
= cur
->elem
;
1027 /* Keep state clients that want to be notified about reboots. */
1028 if ((handler
->conf
->reboot
> REBOOT_NONE
) &&
1029 (client
->states
[RUNNING
] == 2))
1032 /* close state client socket */
1034 close(client
->clientfd
);
1039 if (handler
->conf
->ephemeral
== 1 && handler
->conf
->reboot
!= REBOOT_REQ
)
1040 lxc_destroy_container_on_signal(handler
, name
);
1042 lxc_free_handler(handler
);
1045 void lxc_abort(const char *name
, struct lxc_handler
*handler
)
1049 lxc_set_state(name
, handler
, ABORTING
);
1051 if (handler
->pid
> 0) {
1052 ret
= kill(handler
->pid
, SIGKILL
);
1054 SYSERROR("Failed to send SIGKILL to %d", handler
->pid
);
1057 while ((ret
= waitpid(-1, &status
, 0)) > 0) {
1062 static int do_start(void *data
)
1065 char path
[PATH_MAX
];
1066 bool have_cap_setgid
;
1069 struct lxc_list
*iterator
;
1070 int devnull_fd
= -1;
1071 struct lxc_handler
*handler
= data
;
1073 lxc_sync_fini_parent(handler
);
1075 /* This prctl must be before the synchro, so if the parent dies before
1076 * we set the parent death signal, we will detect its death with the
1077 * synchro right after, otherwise we have a window where the parent can
1078 * exit before we set the pdeath signal leading to a unsupervized
1081 ret
= lxc_set_death_signal(SIGKILL
);
1083 SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL");
1084 goto out_warn_father
;
1087 ret
= lxc_ambient_caps_up();
1089 ERROR("Failed to raise ambient capabilities");
1090 goto out_warn_father
;
1093 ret
= pthread_sigmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
1095 SYSERROR("Failed to set signal mask");
1096 goto out_warn_father
;
1099 /* Don't leak the pinfd to the container. */
1100 if (handler
->pinfd
>= 0)
1101 close(handler
->pinfd
);
1103 ret
= lxc_sync_wait_parent(handler
, LXC_SYNC_STARTUP
);
1105 goto out_warn_father
;
1107 /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See
1108 * https://github.com/lxc/lxd/issues/1978.
1110 if ((handler
->ns_clone_flags
& (CLONE_NEWNET
| CLONE_NEWUSER
)) ==
1111 (CLONE_NEWNET
| CLONE_NEWUSER
)) {
1112 ret
= unshare(CLONE_NEWNET
);
1114 SYSERROR("Failed to unshare CLONE_NEWNET");
1115 goto out_warn_father
;
1117 INFO("Unshared CLONE_NEWNET");
1120 /* Tell the parent task it can begin to configure the container and wait
1123 ret
= lxc_sync_barrier_parent(handler
, LXC_SYNC_CONFIGURE
);
1127 ret
= lxc_network_recv_veth_names_from_parent(handler
);
1129 ERROR("Failed to receive veth names from parent");
1130 goto out_warn_father
;
1133 /* If we are in a new user namespace, become root there to have
1134 * privilege over our namespace.
1136 if (!lxc_list_empty(&handler
->conf
->id_map
)) {
1137 uid_t nsuid
= (handler
->conf
->root_nsuid_map
!= NULL
)
1139 : handler
->conf
->init_uid
;
1140 gid_t nsgid
= (handler
->conf
->root_nsgid_map
!= NULL
)
1142 : handler
->conf
->init_gid
;
1144 ret
= lxc_switch_uid_gid(nsuid
, nsgid
);
1146 goto out_warn_father
;
1148 /* Drop groups only after we switched to a valid gid in the new
1151 ret
= lxc_setgroups(0, NULL
);
1152 if (ret
< 0 && (handler
->am_root
|| errno
!= EPERM
))
1153 goto out_warn_father
;
1155 ret
= prctl(PR_SET_DUMPABLE
, prctl_arg(1), prctl_arg(0),
1156 prctl_arg(0), prctl_arg(0));
1158 goto out_warn_father
;
1160 /* set{g,u}id() clears deathsignal */
1161 ret
= lxc_set_death_signal(SIGKILL
);
1163 SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL");
1164 goto out_warn_father
;
1168 ret
= access(handler
->lxcpath
, X_OK
);
1170 print_top_failing_dir(handler
->lxcpath
);
1171 goto out_warn_father
;
1174 ret
= snprintf(path
, sizeof(path
), "%s/dev/null",
1175 handler
->conf
->rootfs
.mount
);
1176 if (ret
< 0 || ret
>= sizeof(path
))
1177 goto out_warn_father
;
1179 /* In order to checkpoint restore, we need to have everything in the
1180 * same mount namespace. However, some containers may not have a
1181 * reasonable /dev (in particular, they may not have /dev/null), so we
1182 * can't set init's std fds to /dev/null by opening it from inside the
1185 * If that's the case, fall back to using the host's /dev/null. This
1186 * means that migration won't work, but at least we won't spew output
1187 * where it isn't wanted.
1189 if (handler
->backgrounded
&& !handler
->conf
->autodev
) {
1190 ret
= access(path
, F_OK
);
1192 devnull_fd
= open_devnull();
1195 goto out_warn_father
;
1196 WARN("Using /dev/null from the host for container "
1197 "init's standard file descriptors. Migration will "
1202 /* Ask father to setup cgroups and wait for him to finish. */
1203 ret
= lxc_sync_barrier_parent(handler
, LXC_SYNC_CGROUP
);
1207 /* Unshare cgroup namespace after we have setup our cgroups. If we do it
1208 * earlier we end up with a wrong view of /proc/self/cgroup. For
1209 * example, assume we unshare(CLONE_NEWCGROUP) first, and then create
1210 * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then
1211 * /proc/self/cgroup would show us:
1215 * whereas it should actually show
1219 if (handler
->ns_clone_flags
& CLONE_NEWCGROUP
) {
1220 ret
= unshare(CLONE_NEWCGROUP
);
1222 INFO("Failed to unshare CLONE_NEWCGROUP");
1223 goto out_warn_father
;
1225 INFO("Unshared CLONE_NEWCGROUP");
1228 /* Add the requested environment variables to the current environment to
1229 * allow them to be used by the various hooks, such as the start hook
1232 lxc_list_for_each(iterator
, &handler
->conf
->environment
) {
1233 ret
= putenv((char *)iterator
->elem
);
1235 SYSERROR("Failed to set environment variable: %s",
1236 (char *)iterator
->elem
);
1237 goto out_warn_father
;
1241 /* Setup the container, ip, names, utsname, ... */
1242 ret
= lxc_setup(handler
);
1243 close(handler
->data_sock
[1]);
1244 close(handler
->data_sock
[0]);
1246 ERROR("Failed to setup container \"%s\"", handler
->name
);
1247 goto out_warn_father
;
1250 /* Set the label to change to when we exec(2) the container's init. */
1251 ret
= lsm_process_label_set(NULL
, handler
->conf
, true);
1253 goto out_warn_father
;
1255 /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it
1256 * before we aren't allowed anymore.
1258 if (handler
->conf
->no_new_privs
) {
1259 ret
= prctl(PR_SET_NO_NEW_PRIVS
, prctl_arg(1), prctl_arg(0),
1260 prctl_arg(0), prctl_arg(0));
1262 SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block "
1263 "execve() gainable privileges");
1264 goto out_warn_father
;
1266 DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable "
1270 /* Some init's such as busybox will set sane tty settings on stdin,
1271 * stdout, stderr which it thinks is the console. We already set them
1272 * the way we wanted on the real terminal, and we want init to do its
1273 * setup on its console ie. the pty allocated in lxc_terminal_setup() so
1274 * make sure that that pty is stdin,stdout,stderr.
1276 if (handler
->conf
->console
.slave
>= 0) {
1277 if (handler
->backgrounded
|| !handler
->conf
->is_execute
)
1278 ret
= set_stdfds(handler
->conf
->console
.slave
);
1280 ret
= lxc_terminal_set_stdfds(handler
->conf
->console
.slave
);
1282 ERROR("Failed to redirect std{in,out,err} to pty file "
1283 "descriptor %d", handler
->conf
->console
.slave
);
1284 goto out_warn_father
;
1288 /* If we mounted a temporary proc, then unmount it now. */
1289 tmp_proc_unmount(handler
->conf
);
1291 ret
= lxc_seccomp_load(handler
->conf
);
1293 goto out_warn_father
;
1295 ret
= run_lxc_hooks(handler
->name
, "start", handler
->conf
, NULL
);
1297 ERROR("Failed to run lxc.hook.start for container \"%s\"",
1299 goto out_warn_father
;
1302 close(handler
->sigfd
);
1304 if (handler
->conf
->console
.slave
< 0 && handler
->backgrounded
) {
1305 if (devnull_fd
< 0) {
1306 devnull_fd
= open_devnull();
1308 goto out_warn_father
;
1311 ret
= set_stdfds(devnull_fd
);
1313 ERROR("Failed to redirect std{in,out,err} to \"/dev/null\"");
1314 goto out_warn_father
;
1318 if (devnull_fd
>= 0) {
1325 if (handler
->conf
->init_cwd
) {
1326 ret
= chdir(handler
->conf
->init_cwd
);
1328 SYSERROR("Could not change directory to \"%s\"",
1329 handler
->conf
->init_cwd
);
1330 goto out_warn_father
;
1334 ret
= lxc_sync_barrier_parent(handler
, LXC_SYNC_CGROUP_LIMITS
);
1336 goto out_warn_father
;
1338 /* Reset the environment variables the user requested in a clear
1342 /* Don't error out though. */
1344 SYSERROR("Failed to clear environment.");
1346 lxc_list_for_each(iterator
, &handler
->conf
->environment
) {
1347 ret
= putenv((char *)iterator
->elem
);
1349 SYSERROR("Failed to set environment variable: %s",
1350 (char *)iterator
->elem
);
1351 goto out_warn_father
;
1355 ret
= putenv("container=lxc");
1357 SYSERROR("Failed to set environment variable: container=lxc");
1358 goto out_warn_father
;
1361 if (handler
->conf
->ttys
.tty_names
) {
1362 ret
= putenv(handler
->conf
->ttys
.tty_names
);
1364 SYSERROR("Failed to set environment variable for container ptys");
1365 goto out_warn_father
;
1369 /* The container has been setup. We can now switch to an unprivileged
1372 new_uid
= handler
->conf
->init_uid
;
1373 new_gid
= handler
->conf
->init_gid
;
1375 /* If we are in a new user namespace we already dropped all groups when
1376 * we switched to root in the new user namespace further above. Only
1377 * drop groups if we can, so ensure that we have necessary privilege.
1380 have_cap_setgid
= lxc_proc_cap_is_set(CAP_SETGID
, CAP_EFFECTIVE
);
1382 have_cap_setgid
= false;
1384 if (lxc_list_empty(&handler
->conf
->id_map
) && have_cap_setgid
) {
1385 ret
= lxc_setgroups(0, NULL
);
1387 goto out_warn_father
;
1390 ret
= lxc_switch_uid_gid(new_uid
, new_gid
);
1392 goto out_warn_father
;
1394 ret
= lxc_ambient_caps_down();
1396 ERROR("Failed to clear ambient capabilities");
1397 goto out_warn_father
;
1400 if (handler
->conf
->monitor_signal_pdeath
!= SIGKILL
) {
1401 ret
= lxc_set_death_signal(handler
->conf
->monitor_signal_pdeath
);
1403 SYSERROR("Failed to set PR_SET_PDEATHSIG to %d",
1404 handler
->conf
->monitor_signal_pdeath
);
1405 goto out_warn_father
;
1409 /* After this call, we are in error because this ops should not return
1412 handler
->ops
->start(handler
, handler
->data
);
1415 /* We want the parent to know something went wrong, so we return a
1416 * special error code.
1418 lxc_sync_wake_parent(handler
, LXC_SYNC_ERROR
);
1421 if (devnull_fd
>= 0)
1427 static int lxc_recv_ttys_from_child(struct lxc_handler
*handler
)
1430 struct lxc_terminal_info
*tty
;
1432 int sock
= handler
->data_sock
[1];
1433 struct lxc_conf
*conf
= handler
->conf
;
1434 struct lxc_tty_info
*ttys
= &conf
->ttys
;
1436 if (!conf
->ttys
.max
)
1439 ttys
->tty
= malloc(sizeof(*ttys
->tty
) * ttys
->max
);
1443 for (i
= 0; i
< conf
->ttys
.max
; i
++) {
1446 ret
= lxc_abstract_unix_recv_fds(sock
, ttyfds
, 2, NULL
, 0);
1450 tty
= &ttys
->tty
[i
];
1452 tty
->master
= ttyfds
[0];
1453 tty
->slave
= ttyfds
[1];
1454 TRACE("Received pty with master fd %d and slave fd %d from "
1455 "parent", tty
->master
, tty
->slave
);
1459 SYSERROR("Failed to receive %zu ttys from child", ttys
->max
);
1461 TRACE("Received %zu ttys from child", ttys
->max
);
1466 int resolve_clone_flags(struct lxc_handler
*handler
)
1469 struct lxc_conf
*conf
= handler
->conf
;
1471 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1472 if (conf
->ns_keep
!= 0) {
1473 if ((conf
->ns_keep
& ns_info
[i
].clone_flag
) == 0)
1474 handler
->ns_clone_flags
|= ns_info
[i
].clone_flag
;
1475 } else if (conf
->ns_clone
!= 0) {
1476 if ((conf
->ns_clone
& ns_info
[i
].clone_flag
) > 0)
1477 handler
->ns_clone_flags
|= ns_info
[i
].clone_flag
;
1479 if (i
== LXC_NS_USER
&& lxc_list_empty(&handler
->conf
->id_map
))
1482 if (i
== LXC_NS_NET
&& lxc_requests_empty_network(handler
))
1485 if (i
== LXC_NS_CGROUP
&& !cgns_supported())
1488 handler
->ns_clone_flags
|= ns_info
[i
].clone_flag
;
1491 if (!conf
->ns_share
[i
])
1494 handler
->ns_clone_flags
&= ~ns_info
[i
].clone_flag
;
1495 TRACE("Sharing %s namespace", ns_info
[i
].proc_name
);
1501 /* Note that this function is used with clone(CLONE_VM). Some glibc versions
1502 * used to reset the pid/tid to -1 when CLONE_VM was used without CLONE_THREAD.
1503 * But since the memory between parent and child is shared on CLONE_VM this
1504 * would invalidate the getpid() cache that glibc used to maintain and so
1505 * getpid() in the child would return the parent's pid. This is all fixed in
1506 * newer glibc versions where the getpid() cache is removed and the pid/tid is
1507 * not reset anymore.
1508 * However, if for whatever reason you - dear commiter - somehow need to get the
1509 * pid of the dummy intermediate process for do_share_ns() you need to call
1510 * lxc_raw_getpid(). The next lxc_raw_clone() call does not employ CLONE_VM and
1513 static inline int do_share_ns(void *arg
)
1516 struct lxc_handler
*handler
= arg
;
1518 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1519 if (handler
->nsfd
[i
] < 0)
1522 ret
= setns(handler
->nsfd
[i
], 0);
1525 * Note that joining a user and/or mount namespace
1526 * requires the process is not multithreaded otherwise
1527 * setns() will fail here.
1529 SYSERROR("Failed to inherit %s namespace",
1530 ns_info
[i
].proc_name
);
1534 DEBUG("Inherited %s namespace", ns_info
[i
].proc_name
);
1537 flags
= handler
->ns_on_clone_flags
;
1538 flags
|= CLONE_PARENT
;
1539 handler
->pid
= lxc_raw_clone_cb(do_start
, handler
, flags
);
1540 if (handler
->pid
< 0)
1546 static int lxc_setup_shmount(struct lxc_conf
*conf
)
1549 char *full_cont_path
;
1552 /* Construct the shmount path under the container root. */
1553 len_cont
= strlen(conf
->rootfs
.mount
) + 1 + strlen(conf
->shmount
.path_cont
);
1554 /* +1 for the terminating '\0' */
1555 full_cont_path
= malloc(len_cont
+ 1);
1556 if (!full_cont_path
) {
1557 SYSERROR("Not enough memory");
1561 ret
= snprintf(full_cont_path
, len_cont
+ 1, "%s/%s",
1562 conf
->rootfs
.mount
, conf
->shmount
.path_cont
);
1563 if (ret
< 0 || ret
>= len_cont
+ 1) {
1564 SYSERROR("Failed to create filename");
1565 free(full_cont_path
);
1569 /* Check if shmount point is already set up. */
1570 if (is_shared_mountpoint(conf
->shmount
.path_host
)) {
1571 INFO("Path \"%s\" is already MS_SHARED. Reusing",
1572 conf
->shmount
.path_host
);
1573 free(full_cont_path
);
1577 /* Create host and cont mount paths */
1578 ret
= mkdir_p(conf
->shmount
.path_host
, 0711);
1579 if (ret
< 0 && errno
!= EEXIST
) {
1580 SYSERROR("Failed to create directory \"%s\"",
1581 conf
->shmount
.path_host
);
1582 free(full_cont_path
);
1586 ret
= mkdir_p(full_cont_path
, 0711);
1587 if (ret
< 0 && errno
!= EEXIST
) {
1588 SYSERROR("Failed to create directory \"%s\"", full_cont_path
);
1589 free(full_cont_path
);
1593 /* Prepare host mountpoint */
1594 ret
= mount("tmpfs", conf
->shmount
.path_host
, "tmpfs", 0,
1595 "size=100k,mode=0711");
1597 SYSERROR("Failed to mount \"%s\"", conf
->shmount
.path_host
);
1598 free(full_cont_path
);
1602 ret
= mount(conf
->shmount
.path_host
, conf
->shmount
.path_host
, "none",
1603 MS_REC
| MS_SHARED
, "");
1605 SYSERROR("Failed to make shared \"%s\"", conf
->shmount
.path_host
);
1606 free(full_cont_path
);
1610 INFO("Setup shared mount point \"%s\"", conf
->shmount
.path_host
);
1611 free(full_cont_path
);
1615 /* lxc_spawn() performs crucial setup tasks and clone()s the new process which
1616 * exec()s the requested container binary.
1617 * Note that lxc_spawn() runs in the parent namespaces. Any operations performed
1618 * right here should be double checked if they'd pose a security risk. (For
1619 * example, any {u}mount() operations performed here will be reflected on the
1622 static int lxc_spawn(struct lxc_handler
*handler
)
1626 bool wants_to_map_ids
;
1627 struct lxc_list
*id_map
;
1628 const char *name
= handler
->name
;
1629 const char *lxcpath
= handler
->lxcpath
;
1630 bool share_ns
= false;
1631 struct lxc_conf
*conf
= handler
->conf
;
1632 struct cgroup_ops
*cgroup_ops
= handler
->cgroup_ops
;
1634 id_map
= &conf
->id_map
;
1635 wants_to_map_ids
= !lxc_list_empty(id_map
);
1637 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1638 if (!conf
->ns_share
[i
])
1641 handler
->nsfd
[i
] = lxc_inherit_namespace(conf
->ns_share
[i
], lxcpath
, ns_info
[i
].proc_name
);
1642 if (handler
->nsfd
[i
] < 0)
1648 ret
= lxc_sync_init(handler
);
1652 ret
= socketpair(AF_UNIX
, SOCK_STREAM
| SOCK_CLOEXEC
, 0,
1653 handler
->data_sock
);
1655 lxc_sync_fini(handler
);
1659 ret
= resolve_clone_flags(handler
);
1661 lxc_sync_fini(handler
);
1665 if (handler
->ns_clone_flags
& CLONE_NEWNET
) {
1666 if (!lxc_list_empty(&conf
->network
)) {
1668 /* Find gateway addresses from the link device, which is
1669 * no longer accessible inside the container. Do this
1670 * before creating network interfaces, since goto
1671 * out_delete_net does not work before lxc_clone.
1673 ret
= lxc_find_gateway_addresses(handler
);
1675 ERROR("Failed to find gateway addresses");
1676 lxc_sync_fini(handler
);
1680 /* That should be done before the clone because we will
1681 * fill the netdev index and use them in the child.
1683 ret
= lxc_create_network_priv(handler
);
1685 ERROR("Failed to create the network");
1686 lxc_sync_fini(handler
);
1692 if (conf
->shmount
.path_host
) {
1693 if (!conf
->shmount
.path_cont
) {
1694 lxc_sync_fini(handler
);
1698 ret
= lxc_setup_shmount(conf
);
1700 ERROR("Failed to setup shared mount point");
1701 lxc_sync_fini(handler
);
1706 if (!cgroup_init(handler
)) {
1707 ERROR("Failed initializing cgroup support");
1708 goto out_delete_net
;
1711 if (!cgroup_ops
->create(cgroup_ops
, handler
)) {
1712 ERROR("Failed creating cgroups");
1713 goto out_delete_net
;
1716 /* If the rootfs is not a blockdev, prevent the container from marking
1718 * If the container is unprivileged then skip rootfs pinning.
1720 if (!wants_to_map_ids
) {
1721 handler
->pinfd
= pin_rootfs(conf
->rootfs
.path
);
1722 if (handler
->pinfd
== -1)
1723 INFO("Failed to pin the rootfs for container \"%s\"", handler
->name
);
1726 /* Create a process in a new set of namespaces. */
1727 handler
->ns_on_clone_flags
= handler
->ns_clone_flags
;
1728 if (handler
->ns_clone_flags
& CLONE_NEWUSER
) {
1729 /* If CLONE_NEWUSER and CLONE_NEWNET was requested, we need to
1730 * clone a new user namespace first and only later unshare our
1731 * network namespace to ensure that network devices ownership is
1734 handler
->ns_on_clone_flags
&= ~CLONE_NEWNET
;
1736 /* The cgroup namespace gets unshare()ed not clone()ed. */
1737 handler
->ns_on_clone_flags
&= ~CLONE_NEWCGROUP
;
1742 attacher_pid
= lxc_clone(do_share_ns
, handler
,
1743 CLONE_VFORK
| CLONE_VM
| CLONE_FILES
);
1744 if (attacher_pid
< 0) {
1745 SYSERROR(LXC_CLONE_ERROR
);
1746 goto out_delete_net
;
1749 ret
= wait_for_pid(attacher_pid
);
1751 SYSERROR("Intermediate process failed");
1752 goto out_delete_net
;
1755 handler
->pid
= lxc_raw_clone_cb(do_start
, handler
,
1756 handler
->ns_on_clone_flags
);
1758 if (handler
->pid
< 0) {
1759 SYSERROR(LXC_CLONE_ERROR
);
1760 goto out_delete_net
;
1762 TRACE("Cloned child process %d", handler
->pid
);
1764 for (i
= 0; i
< LXC_NS_MAX
; i
++)
1765 if (handler
->ns_on_clone_flags
& ns_info
[i
].clone_flag
)
1766 INFO("Cloned %s", ns_info
[i
].flag_name
);
1768 if (!lxc_try_preserve_namespaces(handler
, handler
->ns_on_clone_flags
, handler
->pid
)) {
1769 ERROR("Failed to preserve cloned namespaces for lxc.hook.stop");
1770 goto out_delete_net
;
1773 lxc_sync_fini_child(handler
);
1775 /* Map the container uids. The container became an invalid userid the
1776 * moment it was cloned with CLONE_NEWUSER. This call doesn't change
1777 * anything immediately, but allows the container to setuid(0) (0 being
1778 * mapped to something else on the host.) later to become a valid uid
1781 if (wants_to_map_ids
) {
1782 if (!handler
->conf
->ns_share
[LXC_NS_USER
] &&
1783 (handler
->conf
->ns_keep
& CLONE_NEWUSER
) == 0) {
1784 ret
= lxc_map_ids(id_map
, handler
->pid
);
1786 ERROR("Failed to set up id mapping.");
1787 goto out_delete_net
;
1792 ret
= lxc_sync_wake_child(handler
, LXC_SYNC_STARTUP
);
1794 goto out_delete_net
;
1796 ret
= lxc_sync_wait_child(handler
, LXC_SYNC_CONFIGURE
);
1798 goto out_delete_net
;
1800 if (!cgroup_ops
->setup_limits(cgroup_ops
, handler
->conf
, false)) {
1801 ERROR("Failed to setup cgroup limits for container \"%s\"", name
);
1802 goto out_delete_net
;
1805 if (!cgroup_ops
->enter(cgroup_ops
, handler
->pid
))
1806 goto out_delete_net
;
1808 if (!cgroup_ops
->chown(cgroup_ops
, handler
->conf
))
1809 goto out_delete_net
;
1811 /* Now we're ready to preserve the network namespace */
1812 ret
= lxc_try_preserve_ns(handler
->pid
, "net");
1814 if (ret
!= -EOPNOTSUPP
) {
1815 SYSERROR("Failed to preserve net namespace");
1816 goto out_delete_net
;
1819 handler
->nsfd
[LXC_NS_NET
] = ret
;
1820 DEBUG("Preserved net namespace via fd %d", ret
);
1822 ret
= lxc_netns_set_nsid(handler
->nsfd
[LXC_NS_NET
]);
1824 ERROR("Failed to allocate new network namespace id: %d", ret
);
1825 goto out_delete_net
;
1827 TRACE("Allocated new network namespace id");
1830 /* Create the network configuration. */
1831 if (handler
->ns_clone_flags
& CLONE_NEWNET
) {
1832 ret
= lxc_network_move_created_netdev_priv(handler
->lxcpath
,
1837 ERROR("Failed to create the configured network");
1838 goto out_delete_net
;
1841 ret
= lxc_create_network_unpriv(handler
->lxcpath
, handler
->name
,
1842 &conf
->network
, handler
->pid
, conf
->hooks_version
);
1844 ERROR("Failed to create the configured network");
1845 goto out_delete_net
;
1849 ret
= lxc_network_send_veth_names_to_child(handler
);
1851 ERROR("Failed to send veth names to child");
1852 goto out_delete_net
;
1855 if (!lxc_list_empty(&conf
->procs
)) {
1856 ret
= setup_proc_filesystem(&conf
->procs
, handler
->pid
);
1858 goto out_delete_net
;
1861 /* Tell the child to continue its initialization. We'll get
1862 * LXC_SYNC_CGROUP when it is ready for us to setup cgroups.
1864 ret
= lxc_sync_barrier_child(handler
, LXC_SYNC_POST_CONFIGURE
);
1866 goto out_delete_net
;
1868 if (!lxc_list_empty(&conf
->limits
)) {
1869 ret
= setup_resource_limits(&conf
->limits
, handler
->pid
);
1871 ERROR("Failed to setup resource limits");
1872 goto out_delete_net
;
1876 ret
= lxc_sync_barrier_child(handler
, LXC_SYNC_CGROUP_UNSHARE
);
1878 goto out_delete_net
;
1880 if (!cgroup_ops
->setup_limits(cgroup_ops
, handler
->conf
, true)) {
1881 ERROR("Failed to setup legacy device cgroup controller limits");
1882 goto out_delete_net
;
1884 TRACE("Set up legacy device cgroup controller limits");
1886 if (handler
->ns_clone_flags
& CLONE_NEWCGROUP
) {
1887 /* Now we're ready to preserve the cgroup namespace */
1888 ret
= lxc_try_preserve_ns(handler
->pid
, "cgroup");
1890 if (ret
!= -EOPNOTSUPP
) {
1891 SYSERROR("Failed to preserve cgroup namespace");
1892 goto out_delete_net
;
1895 handler
->nsfd
[LXC_NS_CGROUP
] = ret
;
1896 DEBUG("Preserved cgroup namespace via fd %d", ret
);
1900 ret
= snprintf(pidstr
, 20, "%d", handler
->pid
);
1901 if (ret
< 0 || ret
>= 20)
1902 goto out_delete_net
;
1904 ret
= setenv("LXC_PID", pidstr
, 1);
1906 SYSERROR("Failed to set environment variable: LXC_PID=%s", pidstr
);
1908 /* Run any host-side start hooks */
1909 ret
= run_lxc_hooks(name
, "start-host", conf
, NULL
);
1911 ERROR("Failed to run lxc.hook.start-host");
1912 goto out_delete_net
;
1915 /* Tell the child to complete its initialization and wait for it to exec
1916 * or return an error. (The child will never return
1917 * LXC_SYNC_READY_START+1. It will either close the sync pipe, causing
1918 * lxc_sync_barrier_child to return success, or return a different
1919 * value, causing us to error out).
1921 ret
= lxc_sync_barrier_child(handler
, LXC_SYNC_READY_START
);
1923 goto out_delete_net
;
1925 ret
= lxc_network_recv_name_and_ifindex_from_child(handler
);
1927 ERROR("Failed to receive names and ifindices for network "
1928 "devices from child");
1929 goto out_delete_net
;
1932 /* Now all networks are created, network devices are moved into place,
1933 * and the correct names and ifindeces in the respective namespaces have
1934 * been recorded. The corresponding structs have now all been filled. So
1935 * log them for debugging purposes.
1937 lxc_log_configured_netdevs(conf
);
1939 /* Read tty fds allocated by child. */
1940 ret
= lxc_recv_ttys_from_child(handler
);
1942 ERROR("Failed to receive tty info from child process");
1943 goto out_delete_net
;
1946 ret
= handler
->ops
->post_start(handler
, handler
->data
);
1950 ret
= lxc_set_state(name
, handler
, RUNNING
);
1952 ERROR("Failed to set state to \"%s\"", lxc_state2str(RUNNING
));
1956 lxc_sync_fini(handler
);
1961 if (handler
->ns_clone_flags
& CLONE_NEWNET
)
1962 lxc_delete_network(handler
);
1965 lxc_abort(name
, handler
);
1966 lxc_sync_fini(handler
);
1967 if (handler
->pinfd
>= 0) {
1968 close(handler
->pinfd
);
1969 handler
->pinfd
= -1;
1975 int __lxc_start(const char *name
, struct lxc_handler
*handler
,
1976 struct lxc_operations
* ops
, void *data
, const char *lxcpath
,
1977 bool backgrounded
, int *error_num
)
1980 struct lxc_conf
*conf
= handler
->conf
;
1982 ret
= lxc_init(name
, handler
);
1984 ERROR("Failed to initialize container \"%s\"", name
);
1988 handler
->data
= data
;
1989 handler
->backgrounded
= backgrounded
;
1991 if (!attach_block_device(handler
->conf
)) {
1992 ERROR("Failed to attach block device");
1993 goto out_fini_nonet
;
1996 if (geteuid() == 0 && !lxc_list_empty(&conf
->id_map
)) {
1997 /* If the backing store is a device, mount it here and now. */
1998 if (rootfs_is_blockdev(conf
)) {
1999 ret
= unshare(CLONE_NEWNS
);
2001 ERROR("Failed to unshare CLONE_NEWNS");
2002 goto out_fini_nonet
;
2004 INFO("Unshared CLONE_NEWNS");
2006 remount_all_slave();
2007 ret
= lxc_setup_rootfs_prepare_root(conf
, name
, lxcpath
);
2009 ERROR("Error setting up rootfs mount as root before spawn");
2010 goto out_fini_nonet
;
2012 INFO("Set up container rootfs as host root");
2016 ret
= lxc_spawn(handler
);
2018 ERROR("Failed to spawn container \"%s\"", name
);
2019 goto out_detach_blockdev
;
2021 /* close parent side of data socket */
2022 close(handler
->data_sock
[0]);
2023 handler
->data_sock
[0] = -1;
2024 close(handler
->data_sock
[1]);
2025 handler
->data_sock
[1] = -1;
2027 handler
->conf
->reboot
= REBOOT_NONE
;
2029 ret
= lxc_poll(name
, handler
);
2031 ERROR("LXC mainloop exited with error: %d", ret
);
2035 if (!handler
->init_died
&& handler
->pid
> 0) {
2036 ERROR("Child process is not killed");
2040 status
= lxc_wait_for_pid_status(handler
->pid
);
2042 SYSERROR("Failed to retrieve status for %d", handler
->pid
);
2044 /* If the child process exited but was not signaled, it didn't call
2045 * reboot. This should mean it was an lxc-execute which simply exited.
2046 * In any case, treat it as a 'halt'.
2048 if (WIFSIGNALED(status
)) {
2049 switch(WTERMSIG(status
)) {
2050 case SIGINT
: /* halt */
2051 DEBUG("Container \"%s\" is halting", name
);
2053 case SIGHUP
: /* reboot */
2054 DEBUG("Container \"%s\" is rebooting", name
);
2055 handler
->conf
->reboot
= REBOOT_REQ
;
2057 case SIGSYS
: /* seccomp */
2058 DEBUG("Container \"%s\" violated its seccomp policy", name
);
2061 DEBUG("Unknown exit status for container \"%s\" init %d", name
, WTERMSIG(status
));
2066 ret
= lxc_restore_phys_nics_to_netns(handler
);
2068 ERROR("Failed to move physical network devices back to parent "
2069 "network namespace");
2071 if (handler
->pinfd
>= 0) {
2072 close(handler
->pinfd
);
2073 handler
->pinfd
= -1;
2076 lxc_monitor_send_exit_code(name
, status
, handler
->lxcpath
);
2077 lxc_error_set_and_log(handler
->pid
, status
);
2079 *error_num
= handler
->exit_status
;
2082 lxc_delete_network(handler
);
2084 out_detach_blockdev
:
2085 detach_block_device(handler
->conf
);
2088 lxc_fini(name
, handler
);
2092 lxc_abort(name
, handler
);
2100 static int start(struct lxc_handler
*handler
, void* data
)
2102 struct start_args
*arg
= data
;
2104 NOTICE("Exec'ing \"%s\"", arg
->argv
[0]);
2106 execvp(arg
->argv
[0], arg
->argv
);
2107 SYSERROR("Failed to exec \"%s\"", arg
->argv
[0]);
2111 static int post_start(struct lxc_handler
*handler
, void* data
)
2113 struct start_args
*arg
= data
;
2115 NOTICE("Started \"%s\" with pid \"%d\"", arg
->argv
[0], handler
->pid
);
2119 static struct lxc_operations start_ops
= {
2121 .post_start
= post_start
2124 int lxc_start(const char *name
, char *const argv
[], struct lxc_handler
*handler
,
2125 const char *lxcpath
, bool backgrounded
, int *error_num
)
2127 struct start_args start_arg
= {
2131 TRACE("Doing lxc_start");
2132 return __lxc_start(name
, handler
, &start_ops
, &start_arg
, lxcpath
, backgrounded
, error_num
);
2135 static void lxc_destroy_container_on_signal(struct lxc_handler
*handler
,
2138 char destroy
[MAXPATHLEN
];
2139 struct lxc_container
*c
;
2143 if (handler
->conf
->rootfs
.path
&& handler
->conf
->rootfs
.mount
) {
2144 bret
= do_destroy_container(handler
);
2146 ERROR("Error destroying rootfs for container \"%s\"", name
);
2150 INFO("Destroyed rootfs for container \"%s\"", name
);
2152 ret
= snprintf(destroy
, MAXPATHLEN
, "%s/%s", handler
->lxcpath
, name
);
2153 if (ret
< 0 || ret
>= MAXPATHLEN
) {
2154 ERROR("Error destroying directory for container \"%s\"", name
);
2158 c
= lxc_container_new(name
, handler
->lxcpath
);
2160 if (container_disk_lock(c
)) {
2161 INFO("Could not update lxc_snapshots file");
2162 lxc_container_put(c
);
2164 mod_all_rdeps(c
, false);
2165 container_disk_unlock(c
);
2166 lxc_container_put(c
);
2170 if (!handler
->am_root
)
2171 ret
= userns_exec_full(handler
->conf
, lxc_rmdir_onedev_wrapper
,
2172 destroy
, "lxc_rmdir_onedev_wrapper");
2174 ret
= lxc_rmdir_onedev(destroy
, NULL
);
2177 ERROR("Error destroying directory for container \"%s\"", name
);
2180 INFO("Destroyed directory for container \"%s\"", name
);
2183 static int lxc_rmdir_onedev_wrapper(void *data
)
2185 char *arg
= (char *) data
;
2186 return lxc_rmdir_onedev(arg
, NULL
);
2189 static bool do_destroy_container(struct lxc_handler
*handler
)
2193 if (!handler
->am_root
) {
2194 ret
= userns_exec_full(handler
->conf
, storage_destroy_wrapper
,
2195 handler
->conf
, "storage_destroy_wrapper");
2202 return storage_destroy(handler
->conf
);