]>
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
41 #include <sys/mount.h>
42 #include <sys/param.h>
43 #include <sys/prctl.h>
44 #include <sys/socket.h>
46 #include <sys/syscall.h>
47 #include <sys/types.h>
56 #include "commands_utils.h"
59 #include "confile_utils.h"
61 #include "file_utils.h"
65 #include "lxccontainer.h"
67 #include "lxcseccomp.h"
71 #include "namespace.h"
73 #include "raw_syscalls.h"
75 #include "storage/storage.h"
76 #include "storage/storage_utils.h"
82 #include <sys/capability.h>
86 #include "include/strlcpy.h"
89 lxc_log_define(start
, lxc
);
91 extern void mod_all_rdeps(struct lxc_container
*c
, bool inc
);
92 static bool do_destroy_container(struct lxc_handler
*handler
);
93 static int lxc_rmdir_onedev_wrapper(void *data
);
94 static void lxc_destroy_container_on_signal(struct lxc_handler
*handler
,
97 static void print_top_failing_dir(const char *path
)
101 char *copy
, *e
, *p
, saved
;
104 copy
= alloca(len
+ 1);
105 (void)strlcpy(copy
, path
, len
+ 1);
111 while (p
< e
&& *p
== '/')
114 while (p
< e
&& *p
!= '/')
120 ret
= access(copy
, X_OK
);
122 SYSERROR("Could not access %s. Please grant it x "
123 "access, or add an ACL for the container "
131 static void lxc_put_nsfds(struct lxc_handler
*handler
)
135 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
136 if (handler
->nsfd
[i
] < 0)
139 close(handler
->nsfd
[i
]);
140 handler
->nsfd
[i
] = -EBADF
;
144 static int lxc_try_preserve_ns(const int pid
, const char *ns
)
148 fd
= lxc_preserve_ns(pid
, ns
);
150 if (errno
!= ENOENT
) {
151 SYSERROR("Failed to preserve %s namespace", ns
);
155 SYSWARN("Kernel does not support preserving %s namespaces", ns
);
162 /* lxc_try_preserve_namespaces: open /proc/@pid/ns/@ns for each namespace
163 * specified in ns_clone_flags.
164 * Return true on success, false on failure.
166 static bool lxc_try_preserve_namespaces(struct lxc_handler
*handler
,
167 int ns_clone_flags
, pid_t pid
)
171 for (i
= 0; i
< LXC_NS_MAX
; i
++)
172 handler
->nsfd
[i
] = -EBADF
;
174 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
177 if ((ns_clone_flags
& ns_info
[i
].clone_flag
) == 0)
180 fd
= lxc_try_preserve_ns(pid
, ns_info
[i
].proc_name
);
182 handler
->nsfd
[i
] = -EBADF
;
184 /* Do not fail to start container on kernels that do
185 * not support interacting with namespaces through
188 if (fd
== -EOPNOTSUPP
)
191 lxc_put_nsfds(handler
);
195 handler
->nsfd
[i
] = fd
;
196 DEBUG("Preserved %s namespace via fd %d", ns_info
[i
].proc_name
,
203 static inline bool match_stdfds(int fd
)
205 return (fd
== STDIN_FILENO
|| fd
== STDOUT_FILENO
|| fd
== STDERR_FILENO
);
208 int lxc_check_inherited(struct lxc_conf
*conf
, bool closeall
,
209 int *fds_to_ignore
, size_t len_fds
)
214 struct dirent
*direntp
;
216 if (conf
&& conf
->close_all_fds
)
220 dir
= opendir("/proc/self/fd");
222 SYSWARN("Failed to open directory");
228 while ((direntp
= readdir(dir
))) {
230 struct lxc_list
*cur
;
231 bool matched
= false;
233 if (strcmp(direntp
->d_name
, ".") == 0)
236 if (strcmp(direntp
->d_name
, "..") == 0)
239 ret
= lxc_safe_int(direntp
->d_name
, &fd
);
241 INFO("Could not parse file descriptor for \"%s\"", direntp
->d_name
);
245 for (i
= 0; i
< len_fds
; i
++)
246 if (fds_to_ignore
[i
] == fd
)
249 if (fd
== fddir
|| fd
== lxc_log_fd
||
250 (i
< len_fds
&& fd
== fds_to_ignore
[i
]))
253 /* Keep state clients that wait on reboots. */
255 lxc_list_for_each(cur
, &conf
->state_clients
) {
256 struct lxc_state_client
*client
= cur
->elem
;
258 if (client
->clientfd
!= fd
)
269 if (current_config
&& fd
== current_config
->logfd
)
272 if (match_stdfds(fd
))
278 INFO("Closed inherited fd %d", fd
);
281 WARN("Inherited fd %d", fd
);
284 /* Only enable syslog at this point to avoid the above logging function
285 * to open a new fd and make the check_inherited function enter an
288 lxc_log_enable_syslog();
290 closedir(dir
); /* cannot fail */
294 static int setup_signal_fd(sigset_t
*oldmask
)
299 const int signals
[] = {SIGBUS
, SIGILL
, SIGSEGV
, SIGWINCH
};
301 /* Block everything except serious error signals. */
302 ret
= sigfillset(&mask
);
306 for (sig
= 0; sig
< (sizeof(signals
) / sizeof(signals
[0])); sig
++) {
307 ret
= sigdelset(&mask
, signals
[sig
]);
312 ret
= pthread_sigmask(SIG_BLOCK
, &mask
, oldmask
);
314 SYSERROR("Failed to set signal mask");
318 ret
= signalfd(-1, &mask
, SFD_CLOEXEC
);
320 SYSERROR("Failed to create signal file descriptor");
324 TRACE("Created signal file descriptor %d", ret
);
329 static int signal_handler(int fd
, uint32_t events
, void *data
,
330 struct lxc_epoll_descr
*descr
)
334 struct signalfd_siginfo siginfo
;
335 struct lxc_handler
*hdlr
= data
;
337 ret
= lxc_read_nointr(fd
, &siginfo
, sizeof(siginfo
));
339 ERROR("Failed to read signal info from signal file descriptor %d", fd
);
340 return LXC_MAINLOOP_ERROR
;
343 if (ret
!= sizeof(siginfo
)) {
344 ERROR("Unexpected size for struct signalfd_siginfo");
348 /* Check whether init is running. */
350 ret
= waitid(P_PID
, hdlr
->pid
, &info
, WEXITED
| WNOWAIT
| WNOHANG
);
351 if (ret
== 0 && info
.si_pid
== hdlr
->pid
)
352 hdlr
->init_died
= true;
354 /* Try to figure out a reasonable exit status to report. */
355 if (hdlr
->init_died
) {
356 switch (info
.si_code
) {
358 hdlr
->exit_status
= info
.si_status
<< 8;
363 hdlr
->exit_status
= info
.si_status
<< 8 | 0x7f;
366 /* Huh? The waitid() told us it's dead *and* continued? */
367 WARN("Init %d dead and continued?", hdlr
->pid
);
368 hdlr
->exit_status
= 1;
371 ERROR("Unknown si_code: %d", info
.si_code
);
372 hdlr
->exit_status
= 1;
376 if (siginfo
.ssi_signo
== SIGHUP
) {
377 kill(hdlr
->pid
, SIGTERM
);
378 INFO("Killing %d since terminal hung up", hdlr
->pid
);
379 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
: LXC_MAINLOOP_CONTINUE
;
382 if (siginfo
.ssi_signo
!= SIGCHLD
) {
383 kill(hdlr
->pid
, siginfo
.ssi_signo
);
384 INFO("Forwarded signal %d to pid %d", siginfo
.ssi_signo
, hdlr
->pid
);
385 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
: LXC_MAINLOOP_CONTINUE
;
388 /* More robustness, protect ourself from a SIGCHLD sent
389 * by a process different from the container init.
391 if (siginfo
.ssi_pid
!= hdlr
->pid
) {
392 NOTICE("Received %d from pid %d instead of container init %d",
393 siginfo
.ssi_signo
, siginfo
.ssi_pid
, hdlr
->pid
);
394 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
: LXC_MAINLOOP_CONTINUE
;
397 if (siginfo
.ssi_code
== CLD_STOPPED
) {
398 INFO("Container init process was stopped");
399 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
: LXC_MAINLOOP_CONTINUE
;
400 } else if (siginfo
.ssi_code
== CLD_CONTINUED
) {
401 INFO("Container init process was continued");
402 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
: LXC_MAINLOOP_CONTINUE
;
405 DEBUG("Container init process %d exited", hdlr
->pid
);
406 return LXC_MAINLOOP_CLOSE
;
409 int lxc_serve_state_clients(const char *name
, struct lxc_handler
*handler
,
414 struct lxc_list
*cur
, *next
;
415 struct lxc_state_client
*client
;
416 struct lxc_msg msg
= {.type
= lxc_msg_state
, .value
= state
};
419 handler
->state
= RUNNING
;
421 handler
->state
= state
;
423 TRACE("Set container state to %s", lxc_state2str(state
));
425 if (lxc_list_empty(&handler
->conf
->state_clients
)) {
426 TRACE("No state clients registered");
430 retlen
= strlcpy(msg
.name
, name
, sizeof(msg
.name
));
431 if (retlen
>= sizeof(msg
.name
))
434 lxc_list_for_each_safe(cur
, &handler
->conf
->state_clients
, next
) {
437 if (client
->states
[state
] == 0) {
438 TRACE("State %s not registered for state client %d",
439 lxc_state2str(state
), client
->clientfd
);
443 TRACE("Sending state %s to state client %d",
444 lxc_state2str(state
), client
->clientfd
);
446 ret
= lxc_send_nointr(client
->clientfd
, &msg
, sizeof(msg
), MSG_NOSIGNAL
);
448 SYSERROR("Failed to send message to client");
450 /* kick client from list */
452 close(client
->clientfd
);
460 static int lxc_serve_state_socket_pair(const char *name
,
461 struct lxc_handler
*handler
,
466 if (!handler
->daemonize
||
467 handler
->state_socket_pair
[1] < 0 ||
471 /* Close read end of the socket pair. */
472 close(handler
->state_socket_pair
[0]);
473 handler
->state_socket_pair
[0] = -1;
476 ret
= lxc_abstract_unix_send_credential(handler
->state_socket_pair
[1],
477 &(int){state
}, sizeof(int));
479 SYSERROR("Failed to send state to %d", handler
->state_socket_pair
[1]);
487 if (ret
!= sizeof(int)) {
488 ERROR("Message too long : %d", handler
->state_socket_pair
[1]);
492 TRACE("Sent container state \"%s\" to %d", lxc_state2str(state
),
493 handler
->state_socket_pair
[1]);
495 /* Close write end of the socket pair. */
496 close(handler
->state_socket_pair
[1]);
497 handler
->state_socket_pair
[1] = -1;
502 int lxc_set_state(const char *name
, struct lxc_handler
*handler
,
507 ret
= lxc_serve_state_socket_pair(name
, handler
, state
);
509 ERROR("Failed to synchronize via anonymous pair of unix sockets");
513 ret
= lxc_serve_state_clients(name
, handler
, state
);
517 /* This function will try to connect to the legacy lxc-monitord state
518 * server and only exists for backwards compatibility.
520 lxc_monitor_send_state(name
, state
, handler
->lxcpath
);
525 int lxc_poll(const char *name
, struct lxc_handler
*handler
)
528 bool has_console
= true;
529 struct lxc_epoll_descr descr
, descr_console
;
531 if (handler
->conf
->console
.path
&&
532 strcmp(handler
->conf
->console
.path
, "none") == 0)
535 ret
= lxc_mainloop_open(&descr
);
537 ERROR("Failed to create mainloop");
542 ret
= lxc_mainloop_open(&descr_console
);
544 ERROR("Failed to create console mainloop");
549 ret
= lxc_mainloop_add_handler(&descr
, handler
->sigfd
, signal_handler
, handler
);
551 ERROR("Failed to add signal handler for %d to mainloop", handler
->sigfd
);
552 goto out_mainloop_console
;
556 struct lxc_terminal
*console
= &handler
->conf
->console
;
558 ret
= lxc_terminal_mainloop_add(&descr
, console
);
560 ERROR("Failed to add console handlers to mainloop");
561 goto out_mainloop_console
;
564 ret
= lxc_terminal_mainloop_add(&descr_console
, console
);
566 ERROR("Failed to add console handlers to console mainloop");
567 goto out_mainloop_console
;
570 handler
->conf
->console
.descr
= &descr
;
573 ret
= lxc_cmd_mainloop_add(name
, &descr
, handler
);
575 ERROR("Failed to add command handler to mainloop");
576 goto out_mainloop_console
;
579 TRACE("Mainloop is ready");
581 ret
= lxc_mainloop(&descr
, -1);
584 if (ret
< 0 || !handler
->init_died
)
585 goto out_mainloop_console
;
588 ret
= lxc_mainloop(&descr_console
, 0);
590 out_mainloop_console
:
592 lxc_mainloop_close(&descr_console
);
593 TRACE("Closed console mainloop");
597 lxc_mainloop_close(&descr
);
598 TRACE("Closed mainloop");
601 close(handler
->sigfd
);
602 TRACE("Closed signal file descriptor %d", handler
->sigfd
);
603 handler
->sigfd
= -EBADF
;
608 void lxc_zero_handler(struct lxc_handler
*handler
)
612 memset(handler
, 0, sizeof(struct lxc_handler
));
618 for (i
= 0; i
< LXC_NS_MAX
; i
++)
619 handler
->nsfd
[i
] = -1;
621 handler
->data_sock
[0] = -1;
622 handler
->data_sock
[1] = -1;
624 handler
->state_socket_pair
[0] = -1;
625 handler
->state_socket_pair
[1] = -1;
627 handler
->sync_sock
[0] = -1;
628 handler
->sync_sock
[1] = -1;
631 void lxc_free_handler(struct lxc_handler
*handler
)
633 if (handler
->pinfd
>= 0)
634 close(handler
->pinfd
);
636 if (handler
->sigfd
>= 0)
637 close(handler
->sigfd
);
639 lxc_put_nsfds(handler
);
641 if (handler
->conf
&& handler
->conf
->reboot
== REBOOT_NONE
)
642 if (handler
->conf
->maincmd_fd
>= 0)
643 lxc_abstract_unix_close(handler
->conf
->maincmd_fd
);
645 if (handler
->state_socket_pair
[0] >= 0)
646 close(handler
->state_socket_pair
[0]);
648 if (handler
->state_socket_pair
[1] >= 0)
649 close(handler
->state_socket_pair
[1]);
651 handler
->conf
= NULL
;
656 struct lxc_handler
*lxc_init_handler(const char *name
, struct lxc_conf
*conf
,
657 const char *lxcpath
, bool daemonize
)
660 struct lxc_handler
*handler
;
662 handler
= malloc(sizeof(*handler
));
666 memset(handler
, 0, sizeof(*handler
));
668 /* Note that am_guest_unpriv() checks the effective uid. We
669 * probably don't care if we are real root only if we are running
670 * as root so this should be fine.
672 handler
->am_root
= !am_guest_unpriv();
673 handler
->data_sock
[0] = handler
->data_sock
[1] = -1;
674 handler
->conf
= conf
;
675 handler
->lxcpath
= lxcpath
;
677 handler
->sigfd
= -EBADF
;
678 handler
->init_died
= false;
679 handler
->state_socket_pair
[0] = handler
->state_socket_pair
[1] = -1;
680 if (handler
->conf
->reboot
== REBOOT_NONE
)
681 lxc_list_init(&handler
->conf
->state_clients
);
683 for (i
= 0; i
< LXC_NS_MAX
; i
++)
684 handler
->nsfd
[i
] = -1;
686 handler
->name
= name
;
688 if (daemonize
&& handler
->conf
->reboot
== REBOOT_NONE
) {
689 /* Create socketpair() to synchronize on daemonized startup.
690 * When the container reboots we don't need to synchronize
691 * again currently so don't open another socketpair().
693 ret
= socketpair(AF_UNIX
, SOCK_STREAM
| SOCK_CLOEXEC
, 0,
694 handler
->state_socket_pair
);
696 ERROR("Failed to create anonymous pair of unix sockets");
700 TRACE("Created anonymous pair {%d,%d} of unix sockets",
701 handler
->state_socket_pair
[0],
702 handler
->state_socket_pair
[1]);
705 if (handler
->conf
->reboot
== REBOOT_NONE
) {
706 handler
->conf
->maincmd_fd
= lxc_cmd_init(name
, lxcpath
, "command");
707 if (handler
->conf
->maincmd_fd
< 0) {
708 ERROR("Failed to set up command socket");
713 TRACE("Unix domain socket %d for command server is ready",
714 handler
->conf
->maincmd_fd
);
719 lxc_free_handler(handler
);
724 int lxc_init(const char *name
, struct lxc_handler
*handler
)
727 const char *loglevel
;
728 struct lxc_conf
*conf
= handler
->conf
;
730 handler
->monitor_pid
= lxc_raw_getpid();
733 TRACE("Initialized LSM");
735 ret
= lxc_read_seccomp_config(conf
);
737 ERROR("Failed loading seccomp policy");
738 goto out_close_maincmd_fd
;
740 TRACE("Read seccomp policy");
742 /* Begin by setting the state to STARTING. */
743 ret
= lxc_set_state(name
, handler
, STARTING
);
745 ERROR("Failed to set state to \"%s\"", lxc_state2str(STARTING
));
746 goto out_close_maincmd_fd
;
748 TRACE("Set container state to \"STARTING\"");
750 /* Start of environment variable setup for hooks. */
751 ret
= setenv("LXC_NAME", name
, 1);
753 SYSERROR("Failed to set environment variable: LXC_NAME=%s", name
);
756 ret
= setenv("LXC_CONFIG_FILE", conf
->rcfile
, 1);
758 SYSERROR("Failed to set environment variable: "
759 "LXC_CONFIG_FILE=%s", conf
->rcfile
);
762 if (conf
->rootfs
.mount
) {
763 ret
= setenv("LXC_ROOTFS_MOUNT", conf
->rootfs
.mount
, 1);
765 SYSERROR("Failed to set environment variable: "
766 "LXC_ROOTFS_MOUNT=%s", conf
->rootfs
.mount
);
769 if (conf
->rootfs
.path
) {
770 ret
= setenv("LXC_ROOTFS_PATH", conf
->rootfs
.path
, 1);
772 SYSERROR("Failed to set environment variable: "
773 "LXC_ROOTFS_PATH=%s", conf
->rootfs
.path
);
776 if (conf
->console
.path
) {
777 ret
= setenv("LXC_CONSOLE", conf
->console
.path
, 1);
779 SYSERROR("Failed to set environment variable: "
780 "LXC_CONSOLE=%s", conf
->console
.path
);
783 if (conf
->console
.log_path
) {
784 ret
= setenv("LXC_CONSOLE_LOGPATH", conf
->console
.log_path
, 1);
786 SYSERROR("Failed to set environment variable: "
787 "LXC_CONSOLE_LOGPATH=%s", conf
->console
.log_path
);
790 if (cgns_supported()) {
791 ret
= setenv("LXC_CGNS_AWARE", "1", 1);
793 SYSERROR("Failed to set environment variable "
797 loglevel
= lxc_log_priority_to_string(lxc_log_get_level());
798 ret
= setenv("LXC_LOG_LEVEL", loglevel
, 1);
800 SYSERROR("Set environment variable LXC_LOG_LEVEL=%s",
803 if (conf
->hooks_version
== 0)
804 ret
= setenv("LXC_HOOK_VERSION", "0", 1);
806 ret
= setenv("LXC_HOOK_VERSION", "1", 1);
808 SYSERROR("Failed to set environment variable LXC_HOOK_VERSION=%u", conf
->hooks_version
);
809 /* End of environment variable setup for hooks. */
811 TRACE("Set environment variables");
813 ret
= run_lxc_hooks(name
, "pre-start", conf
, NULL
);
815 ERROR("Failed to run lxc.hook.pre-start for container \"%s\"", name
);
818 TRACE("Ran pre-start hooks");
820 /* The signal fd has to be created before forking otherwise if the child
821 * process exits before we setup the signal fd, the event will be lost
822 * and the command will be stuck.
824 handler
->sigfd
= setup_signal_fd(&handler
->oldmask
);
825 if (handler
->sigfd
< 0) {
826 ERROR("Failed to setup SIGCHLD fd handler.");
829 TRACE("Set up signal fd");
831 /* Do this after setting up signals since it might unblock SIGWINCH. */
832 ret
= lxc_terminal_setup(conf
);
834 ERROR("Failed to create console");
835 goto out_restore_sigmask
;
837 TRACE("Created console");
839 ret
= lxc_terminal_map_ids(conf
, &conf
->console
);
841 ERROR("Failed to chown console");
842 goto out_delete_terminal
;
844 TRACE("Chowned console");
846 handler
->cgroup_ops
= cgroup_init(handler
->conf
);
847 if (!handler
->cgroup_ops
) {
848 ERROR("Failed to initialize cgroup driver");
849 goto out_delete_terminal
;
851 TRACE("Initialized cgroup driver");
853 ret
= lsm_process_prepare(conf
, handler
->lxcpath
);
855 ERROR("Failed to initialize LSM");
856 goto out_destroy_cgroups
;
858 TRACE("Initialized LSM");
860 INFO("Container \"%s\" is initialized", name
);
864 handler
->cgroup_ops
->payload_destroy(handler
->cgroup_ops
, handler
);
865 handler
->cgroup_ops
->monitor_destroy(handler
->cgroup_ops
, handler
);
868 lxc_terminal_delete(&handler
->conf
->console
);
871 (void)pthread_sigmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
874 lxc_delete_tty(&conf
->ttys
);
877 (void)lxc_set_state(name
, handler
, ABORTING
);
879 out_close_maincmd_fd
:
880 lxc_abstract_unix_close(conf
->maincmd_fd
);
881 conf
->maincmd_fd
= -1;
885 void lxc_fini(const char *name
, struct lxc_handler
*handler
)
889 struct lxc_list
*cur
, *next
;
890 char *namespaces
[LXC_NS_MAX
+ 1];
891 size_t namespace_count
= 0;
892 struct cgroup_ops
*cgroup_ops
= handler
->cgroup_ops
;
894 /* The STOPPING state is there for future cleanup code which can take
897 lxc_set_state(name
, handler
, STOPPING
);
899 self
= lxc_raw_getpid();
900 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
901 if (handler
->nsfd
[i
] < 0)
904 if (handler
->conf
->hooks_version
== 0)
905 ret
= asprintf(&namespaces
[namespace_count
],
906 "%s:/proc/%d/fd/%d", ns_info
[i
].proc_name
,
907 self
, handler
->nsfd
[i
]);
909 ret
= asprintf(&namespaces
[namespace_count
],
910 "/proc/%d/fd/%d", self
, handler
->nsfd
[i
]);
912 SYSERROR("Failed to allocate memory");
916 if (handler
->conf
->hooks_version
== 0) {
921 ret
= setenv(ns_info
[i
].env_name
, namespaces
[namespace_count
], 1);
923 SYSERROR("Failed to set environment variable %s=%s",
924 ns_info
[i
].env_name
, namespaces
[namespace_count
]);
926 TRACE("Set environment variable %s=%s",
927 ns_info
[i
].env_name
, namespaces
[namespace_count
]);
931 namespaces
[namespace_count
] = NULL
;
933 if (handler
->conf
->reboot
> REBOOT_NONE
) {
934 ret
= setenv("LXC_TARGET", "reboot", 1);
936 SYSERROR("Failed to set environment variable: "
937 "LXC_TARGET=reboot");
940 if (handler
->conf
->reboot
== REBOOT_NONE
) {
941 ret
= setenv("LXC_TARGET", "stop", 1);
943 SYSERROR("Failed to set environment variable: "
947 if (handler
->conf
->hooks_version
== 0)
948 ret
= run_lxc_hooks(name
, "stop", handler
->conf
, namespaces
);
950 ret
= run_lxc_hooks(name
, "stop", handler
->conf
, NULL
);
952 ERROR("Failed to run \"lxc.hook.stop\" hook");
954 while (namespace_count
--)
955 free(namespaces
[namespace_count
]);
957 lsm_process_cleanup(handler
->conf
, handler
->lxcpath
);
959 cgroup_ops
->payload_destroy(cgroup_ops
, handler
);
960 cgroup_ops
->monitor_destroy(cgroup_ops
, handler
);
961 cgroup_exit(cgroup_ops
);
963 if (handler
->conf
->reboot
== REBOOT_NONE
) {
964 /* For all new state clients simply close the command socket.
965 * This will inform all state clients that the container is
966 * STOPPED and also prevents a race between a open()/close() on
967 * the command socket causing a new process to get ECONNREFUSED
968 * because we haven't yet closed the command socket.
970 lxc_abstract_unix_close(handler
->conf
->maincmd_fd
);
971 handler
->conf
->maincmd_fd
= -1;
972 TRACE("Closed command socket");
974 /* This function will try to connect to the legacy lxc-monitord
975 * state server and only exists for backwards compatibility.
977 lxc_monitor_send_state(name
, STOPPED
, handler
->lxcpath
);
979 /* The command socket is closed so no one can acces the command
980 * socket anymore so there's no need to lock it.
982 handler
->state
= STOPPED
;
983 TRACE("Set container state to \"STOPPED\"");
985 lxc_set_state(name
, handler
, STOPPED
);
988 ret
= run_lxc_hooks(name
, "post-stop", handler
->conf
, NULL
);
990 ERROR("Failed to run lxc.hook.post-stop for container \"%s\"", name
);
991 if (handler
->conf
->reboot
> REBOOT_NONE
) {
992 WARN("Container will be stopped instead of rebooted");
993 handler
->conf
->reboot
= REBOOT_NONE
;
995 ret
= setenv("LXC_TARGET", "stop", 1);
997 WARN("Failed to set environment variable: "
1002 /* Reset mask set by setup_signal_fd. */
1003 ret
= pthread_sigmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
1005 SYSWARN("Failed to restore signal mask");
1007 lxc_terminal_delete(&handler
->conf
->console
);
1008 lxc_delete_tty(&handler
->conf
->ttys
);
1010 /* The command socket is now closed, no more state clients can register
1011 * themselves from now on. So free the list of state clients.
1013 lxc_list_for_each_safe(cur
, &handler
->conf
->state_clients
, next
) {
1014 struct lxc_state_client
*client
= cur
->elem
;
1016 /* Keep state clients that want to be notified about reboots. */
1017 if ((handler
->conf
->reboot
> REBOOT_NONE
) &&
1018 (client
->states
[RUNNING
] == 2))
1021 /* close state client socket */
1023 close(client
->clientfd
);
1028 if (handler
->conf
->ephemeral
== 1 && handler
->conf
->reboot
!= REBOOT_REQ
)
1029 lxc_destroy_container_on_signal(handler
, name
);
1031 lxc_free_handler(handler
);
1034 void lxc_abort(const char *name
, struct lxc_handler
*handler
)
1038 lxc_set_state(name
, handler
, ABORTING
);
1040 if (handler
->pid
> 0) {
1041 ret
= kill(handler
->pid
, SIGKILL
);
1043 SYSERROR("Failed to send SIGKILL to %d", handler
->pid
);
1046 while ((ret
= waitpid(-1, &status
, 0)) > 0) {
1051 static int do_start(void *data
)
1054 char path
[PATH_MAX
];
1057 struct lxc_list
*iterator
;
1060 int devnull_fd
= -1;
1061 struct lxc_handler
*handler
= data
;
1063 lxc_sync_fini_parent(handler
);
1065 /* This prctl must be before the synchro, so if the parent dies before
1066 * we set the parent death signal, we will detect its death with the
1067 * synchro right after, otherwise we have a window where the parent can
1068 * exit before we set the pdeath signal leading to a unsupervized
1071 ret
= lxc_set_death_signal(SIGKILL
, 0);
1073 SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL");
1074 goto out_warn_father
;
1077 ret
= lxc_ambient_caps_up();
1079 ERROR("Failed to raise ambient capabilities");
1080 goto out_warn_father
;
1083 ret
= pthread_sigmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
1085 SYSERROR("Failed to set signal mask");
1086 goto out_warn_father
;
1089 /* Don't leak the pinfd to the container. */
1090 if (handler
->pinfd
>= 0)
1091 close(handler
->pinfd
);
1093 ret
= lxc_sync_wait_parent(handler
, LXC_SYNC_STARTUP
);
1095 goto out_warn_father
;
1097 /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See
1098 * https://github.com/lxc/lxd/issues/1978.
1100 if ((handler
->ns_clone_flags
& (CLONE_NEWNET
| CLONE_NEWUSER
)) ==
1101 (CLONE_NEWNET
| CLONE_NEWUSER
)) {
1102 ret
= unshare(CLONE_NEWNET
);
1104 SYSERROR("Failed to unshare CLONE_NEWNET");
1105 goto out_warn_father
;
1107 INFO("Unshared CLONE_NEWNET");
1110 /* Tell the parent task it can begin to configure the container and wait
1113 ret
= lxc_sync_barrier_parent(handler
, LXC_SYNC_CONFIGURE
);
1117 ret
= lxc_network_recv_veth_names_from_parent(handler
);
1119 ERROR("Failed to receive veth names from parent");
1120 goto out_warn_father
;
1123 /* If we are in a new user namespace, become root there to have
1124 * privilege over our namespace.
1126 if (!lxc_list_empty(&handler
->conf
->id_map
)) {
1127 if (!handler
->conf
->root_nsuid_map
)
1128 nsuid
= handler
->conf
->init_uid
;
1130 if (!handler
->conf
->root_nsgid_map
)
1131 nsgid
= handler
->conf
->init_gid
;
1133 if (!lxc_switch_uid_gid(nsuid
, nsgid
))
1134 goto out_warn_father
;
1136 /* Drop groups only after we switched to a valid gid in the new
1139 if (!lxc_setgroups(0, NULL
) &&
1140 (handler
->am_root
|| errno
!= EPERM
))
1141 goto out_warn_father
;
1143 ret
= prctl(PR_SET_DUMPABLE
, prctl_arg(1), prctl_arg(0),
1144 prctl_arg(0), prctl_arg(0));
1146 goto out_warn_father
;
1148 /* set{g,u}id() clears deathsignal */
1149 ret
= lxc_set_death_signal(SIGKILL
, 0);
1151 SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL");
1152 goto out_warn_father
;
1156 ret
= access(handler
->lxcpath
, X_OK
);
1158 print_top_failing_dir(handler
->lxcpath
);
1159 goto out_warn_father
;
1162 ret
= snprintf(path
, sizeof(path
), "%s/dev/null",
1163 handler
->conf
->rootfs
.mount
);
1164 if (ret
< 0 || ret
>= sizeof(path
))
1165 goto out_warn_father
;
1167 /* In order to checkpoint restore, we need to have everything in the
1168 * same mount namespace. However, some containers may not have a
1169 * reasonable /dev (in particular, they may not have /dev/null), so we
1170 * can't set init's std fds to /dev/null by opening it from inside the
1173 * If that's the case, fall back to using the host's /dev/null. This
1174 * means that migration won't work, but at least we won't spew output
1175 * where it isn't wanted.
1177 if (handler
->daemonize
&& !handler
->conf
->autodev
) {
1178 ret
= access(path
, F_OK
);
1180 devnull_fd
= open_devnull();
1183 goto out_warn_father
;
1184 WARN("Using /dev/null from the host for container "
1185 "init's standard file descriptors. Migration will "
1190 /* Ask father to setup cgroups and wait for him to finish. */
1191 ret
= lxc_sync_barrier_parent(handler
, LXC_SYNC_CGROUP
);
1195 /* Unshare cgroup namespace after we have setup our cgroups. If we do it
1196 * earlier we end up with a wrong view of /proc/self/cgroup. For
1197 * example, assume we unshare(CLONE_NEWCGROUP) first, and then create
1198 * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then
1199 * /proc/self/cgroup would show us:
1203 * whereas it should actually show
1207 if (handler
->ns_clone_flags
& CLONE_NEWCGROUP
) {
1208 ret
= unshare(CLONE_NEWCGROUP
);
1210 INFO("Failed to unshare CLONE_NEWCGROUP");
1211 goto out_warn_father
;
1213 INFO("Unshared CLONE_NEWCGROUP");
1216 /* Add the requested environment variables to the current environment to
1217 * allow them to be used by the various hooks, such as the start hook
1220 lxc_list_for_each(iterator
, &handler
->conf
->environment
) {
1221 ret
= putenv((char *)iterator
->elem
);
1223 SYSERROR("Failed to set environment variable: %s",
1224 (char *)iterator
->elem
);
1225 goto out_warn_father
;
1229 /* Setup the container, ip, names, utsname, ... */
1230 ret
= lxc_setup(handler
);
1231 close(handler
->data_sock
[1]);
1232 close(handler
->data_sock
[0]);
1234 ERROR("Failed to setup container \"%s\"", handler
->name
);
1235 goto out_warn_father
;
1238 /* Set the label to change to when we exec(2) the container's init. */
1239 ret
= lsm_process_label_set(NULL
, handler
->conf
, true);
1241 goto out_warn_father
;
1243 /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it
1244 * before we aren't allowed anymore.
1246 if (handler
->conf
->no_new_privs
) {
1247 ret
= prctl(PR_SET_NO_NEW_PRIVS
, prctl_arg(1), prctl_arg(0),
1248 prctl_arg(0), prctl_arg(0));
1250 SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block "
1251 "execve() gainable privileges");
1252 goto out_warn_father
;
1254 DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable "
1258 /* Some init's such as busybox will set sane tty settings on stdin,
1259 * stdout, stderr which it thinks is the console. We already set them
1260 * the way we wanted on the real terminal, and we want init to do its
1261 * setup on its console ie. the pty allocated in lxc_terminal_setup() so
1262 * make sure that that pty is stdin,stdout,stderr.
1264 if (handler
->conf
->console
.slave
>= 0) {
1265 if (handler
->daemonize
|| !handler
->conf
->is_execute
)
1266 ret
= set_stdfds(handler
->conf
->console
.slave
);
1268 ret
= lxc_terminal_set_stdfds(handler
->conf
->console
.slave
);
1270 ERROR("Failed to redirect std{in,out,err} to pty file "
1271 "descriptor %d", handler
->conf
->console
.slave
);
1272 goto out_warn_father
;
1276 /* If we mounted a temporary proc, then unmount it now. */
1277 tmp_proc_unmount(handler
->conf
);
1279 ret
= lxc_seccomp_load(handler
->conf
);
1281 goto out_warn_father
;
1283 ret
= run_lxc_hooks(handler
->name
, "start", handler
->conf
, NULL
);
1285 ERROR("Failed to run lxc.hook.start for container \"%s\"",
1287 goto out_warn_father
;
1290 close(handler
->sigfd
);
1292 if (handler
->conf
->console
.slave
< 0 && handler
->daemonize
) {
1293 if (devnull_fd
< 0) {
1294 devnull_fd
= open_devnull();
1296 goto out_warn_father
;
1299 ret
= set_stdfds(devnull_fd
);
1301 ERROR("Failed to redirect std{in,out,err} to \"/dev/null\"");
1302 goto out_warn_father
;
1306 if (devnull_fd
>= 0) {
1313 if (handler
->conf
->init_cwd
) {
1314 ret
= chdir(handler
->conf
->init_cwd
);
1316 SYSERROR("Could not change directory to \"%s\"",
1317 handler
->conf
->init_cwd
);
1318 goto out_warn_father
;
1322 ret
= lxc_sync_barrier_parent(handler
, LXC_SYNC_CGROUP_LIMITS
);
1324 goto out_warn_father
;
1326 /* Reset the environment variables the user requested in a clear
1330 /* Don't error out though. */
1332 SYSERROR("Failed to clear environment.");
1334 lxc_list_for_each(iterator
, &handler
->conf
->environment
) {
1335 ret
= putenv((char *)iterator
->elem
);
1337 SYSERROR("Failed to set environment variable: %s",
1338 (char *)iterator
->elem
);
1339 goto out_warn_father
;
1343 ret
= putenv("container=lxc");
1345 SYSERROR("Failed to set environment variable: container=lxc");
1346 goto out_warn_father
;
1349 if (handler
->conf
->ttys
.tty_names
) {
1350 ret
= putenv(handler
->conf
->ttys
.tty_names
);
1352 SYSERROR("Failed to set environment variable for container ptys");
1353 goto out_warn_father
;
1357 /* The container has been setup. We can now switch to an unprivileged
1360 new_uid
= handler
->conf
->init_uid
;
1361 new_gid
= handler
->conf
->init_gid
;
1363 /* Avoid unnecessary syscalls. */
1364 if (new_uid
== nsuid
)
1365 new_uid
= LXC_INVALID_UID
;
1367 if (new_gid
== nsgid
)
1368 new_gid
= LXC_INVALID_GID
;
1370 if (!lxc_switch_uid_gid(new_uid
, new_gid
))
1371 goto out_warn_father
;
1373 /* If we are in a new user namespace we already dropped all groups when
1374 * we switched to root in the new user namespace further above. Only
1375 * drop groups if we can, so ensure that we have necessary privilege.
1377 if (lxc_list_empty(&handler
->conf
->id_map
))
1379 if (lxc_proc_cap_is_set(CAP_SETGID
, CAP_EFFECTIVE
))
1381 if (!lxc_setgroups(0, NULL
))
1382 goto out_warn_father
;
1384 ret
= lxc_ambient_caps_down();
1386 ERROR("Failed to clear ambient capabilities");
1387 goto out_warn_father
;
1390 if (handler
->conf
->monitor_signal_pdeath
!= SIGKILL
) {
1391 ret
= lxc_set_death_signal(handler
->conf
->monitor_signal_pdeath
, 0);
1393 SYSERROR("Failed to set PR_SET_PDEATHSIG to %d",
1394 handler
->conf
->monitor_signal_pdeath
);
1395 goto out_warn_father
;
1399 /* After this call, we are in error because this ops should not return
1402 handler
->ops
->start(handler
, handler
->data
);
1405 /* We want the parent to know something went wrong, so we return a
1406 * special error code.
1408 lxc_sync_wake_parent(handler
, LXC_SYNC_ERROR
);
1411 if (devnull_fd
>= 0)
1417 static int lxc_recv_ttys_from_child(struct lxc_handler
*handler
)
1420 struct lxc_terminal_info
*tty
;
1422 int sock
= handler
->data_sock
[1];
1423 struct lxc_conf
*conf
= handler
->conf
;
1424 struct lxc_tty_info
*ttys
= &conf
->ttys
;
1426 if (!conf
->ttys
.max
)
1429 ttys
->tty
= malloc(sizeof(*ttys
->tty
) * ttys
->max
);
1433 for (i
= 0; i
< conf
->ttys
.max
; i
++) {
1436 ret
= lxc_abstract_unix_recv_fds(sock
, ttyfds
, 2, NULL
, 0);
1440 tty
= &ttys
->tty
[i
];
1442 tty
->master
= ttyfds
[0];
1443 tty
->slave
= ttyfds
[1];
1444 TRACE("Received pty with master fd %d and slave fd %d from "
1445 "parent", tty
->master
, tty
->slave
);
1449 SYSERROR("Failed to receive %zu ttys from child", ttys
->max
);
1451 TRACE("Received %zu ttys from child", ttys
->max
);
1456 int resolve_clone_flags(struct lxc_handler
*handler
)
1459 struct lxc_conf
*conf
= handler
->conf
;
1461 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1462 if (conf
->ns_keep
!= 0) {
1463 if ((conf
->ns_keep
& ns_info
[i
].clone_flag
) == 0)
1464 handler
->ns_clone_flags
|= ns_info
[i
].clone_flag
;
1465 } else if (conf
->ns_clone
!= 0) {
1466 if ((conf
->ns_clone
& ns_info
[i
].clone_flag
) > 0)
1467 handler
->ns_clone_flags
|= ns_info
[i
].clone_flag
;
1469 if (i
== LXC_NS_USER
&& lxc_list_empty(&handler
->conf
->id_map
))
1472 if (i
== LXC_NS_NET
&& lxc_requests_empty_network(handler
))
1475 if (i
== LXC_NS_CGROUP
&& !cgns_supported())
1478 handler
->ns_clone_flags
|= ns_info
[i
].clone_flag
;
1481 if (!conf
->ns_share
[i
])
1484 handler
->ns_clone_flags
&= ~ns_info
[i
].clone_flag
;
1485 TRACE("Sharing %s namespace", ns_info
[i
].proc_name
);
1491 /* Note that this function is used with clone(CLONE_VM). Some glibc versions
1492 * used to reset the pid/tid to -1 when CLONE_VM was used without CLONE_THREAD.
1493 * But since the memory between parent and child is shared on CLONE_VM this
1494 * would invalidate the getpid() cache that glibc used to maintain and so
1495 * getpid() in the child would return the parent's pid. This is all fixed in
1496 * newer glibc versions where the getpid() cache is removed and the pid/tid is
1497 * not reset anymore.
1498 * However, if for whatever reason you - dear commiter - somehow need to get the
1499 * pid of the dummy intermediate process for do_share_ns() you need to call
1500 * lxc_raw_getpid(). The next lxc_raw_clone() call does not employ CLONE_VM and
1503 static inline int do_share_ns(void *arg
)
1506 struct lxc_handler
*handler
= arg
;
1508 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1509 if (handler
->nsfd
[i
] < 0)
1512 ret
= setns(handler
->nsfd
[i
], 0);
1515 * Note that joining a user and/or mount namespace
1516 * requires the process is not multithreaded otherwise
1517 * setns() will fail here.
1519 SYSERROR("Failed to inherit %s namespace",
1520 ns_info
[i
].proc_name
);
1524 DEBUG("Inherited %s namespace", ns_info
[i
].proc_name
);
1527 flags
= handler
->ns_on_clone_flags
;
1528 flags
|= CLONE_PARENT
;
1529 handler
->pid
= lxc_raw_clone_cb(do_start
, handler
, flags
);
1530 if (handler
->pid
< 0)
1536 static int lxc_setup_shmount(struct lxc_conf
*conf
)
1539 char *full_cont_path
;
1542 /* Construct the shmount path under the container root. */
1543 len_cont
= strlen(conf
->rootfs
.mount
) + 1 + strlen(conf
->shmount
.path_cont
);
1544 /* +1 for the terminating '\0' */
1545 full_cont_path
= malloc(len_cont
+ 1);
1546 if (!full_cont_path
) {
1547 SYSERROR("Not enough memory");
1551 ret
= snprintf(full_cont_path
, len_cont
+ 1, "%s/%s",
1552 conf
->rootfs
.mount
, conf
->shmount
.path_cont
);
1553 if (ret
< 0 || ret
>= len_cont
+ 1) {
1554 SYSERROR("Failed to create filename");
1555 free(full_cont_path
);
1559 /* Check if shmount point is already set up. */
1560 if (is_shared_mountpoint(conf
->shmount
.path_host
)) {
1561 INFO("Path \"%s\" is already MS_SHARED. Reusing",
1562 conf
->shmount
.path_host
);
1563 free(full_cont_path
);
1567 /* Create host and cont mount paths */
1568 ret
= mkdir_p(conf
->shmount
.path_host
, 0711);
1569 if (ret
< 0 && errno
!= EEXIST
) {
1570 SYSERROR("Failed to create directory \"%s\"",
1571 conf
->shmount
.path_host
);
1572 free(full_cont_path
);
1576 ret
= mkdir_p(full_cont_path
, 0711);
1577 if (ret
< 0 && errno
!= EEXIST
) {
1578 SYSERROR("Failed to create directory \"%s\"", full_cont_path
);
1579 free(full_cont_path
);
1583 /* Prepare host mountpoint */
1584 ret
= mount("tmpfs", conf
->shmount
.path_host
, "tmpfs", 0,
1585 "size=100k,mode=0711");
1587 SYSERROR("Failed to mount \"%s\"", conf
->shmount
.path_host
);
1588 free(full_cont_path
);
1592 ret
= mount(conf
->shmount
.path_host
, conf
->shmount
.path_host
, "none",
1593 MS_REC
| MS_SHARED
, "");
1595 SYSERROR("Failed to make shared \"%s\"", conf
->shmount
.path_host
);
1596 free(full_cont_path
);
1600 INFO("Setup shared mount point \"%s\"", conf
->shmount
.path_host
);
1601 free(full_cont_path
);
1605 /* lxc_spawn() performs crucial setup tasks and clone()s the new process which
1606 * exec()s the requested container binary.
1607 * Note that lxc_spawn() runs in the parent namespaces. Any operations performed
1608 * right here should be double checked if they'd pose a security risk. (For
1609 * example, any {u}mount() operations performed here will be reflected on the
1612 static int lxc_spawn(struct lxc_handler
*handler
)
1616 bool wants_to_map_ids
;
1617 struct lxc_list
*id_map
;
1618 const char *name
= handler
->name
;
1619 const char *lxcpath
= handler
->lxcpath
;
1620 bool share_ns
= false;
1621 struct lxc_conf
*conf
= handler
->conf
;
1622 struct cgroup_ops
*cgroup_ops
= handler
->cgroup_ops
;
1624 id_map
= &conf
->id_map
;
1625 wants_to_map_ids
= !lxc_list_empty(id_map
);
1627 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1628 if (!conf
->ns_share
[i
])
1631 handler
->nsfd
[i
] = lxc_inherit_namespace(conf
->ns_share
[i
], lxcpath
, ns_info
[i
].proc_name
);
1632 if (handler
->nsfd
[i
] < 0)
1638 ret
= lxc_sync_init(handler
);
1642 ret
= socketpair(AF_UNIX
, SOCK_STREAM
| SOCK_CLOEXEC
, 0,
1643 handler
->data_sock
);
1647 ret
= resolve_clone_flags(handler
);
1651 if (conf
->shmount
.path_host
) {
1652 if (!conf
->shmount
.path_cont
)
1655 ret
= lxc_setup_shmount(conf
);
1657 ERROR("Failed to setup shared mount point");
1662 if (handler
->ns_clone_flags
& CLONE_NEWNET
) {
1663 if (!lxc_list_empty(&conf
->network
)) {
1665 /* Find gateway addresses from the link device, which is
1666 * no longer accessible inside the container. Do this
1667 * before creating network interfaces, since goto
1668 * out_delete_net does not work before lxc_clone.
1670 ret
= lxc_find_gateway_addresses(handler
);
1672 ERROR("Failed to find gateway addresses");
1676 /* That should be done before the clone because we will
1677 * fill the netdev index and use them in the child.
1679 ret
= lxc_create_network_priv(handler
);
1681 ERROR("Failed to create the network");
1682 goto out_delete_net
;
1687 if (!cgroup_ops
->payload_create(cgroup_ops
, handler
)) {
1688 ERROR("Failed creating cgroups");
1689 goto out_delete_net
;
1692 /* If the rootfs is not a blockdev, prevent the container from marking
1694 * If the container is unprivileged then skip rootfs pinning.
1696 if (!wants_to_map_ids
) {
1697 handler
->pinfd
= pin_rootfs(conf
->rootfs
.path
);
1698 if (handler
->pinfd
== -1)
1699 INFO("Failed to pin the rootfs for container \"%s\"", handler
->name
);
1702 /* Create a process in a new set of namespaces. */
1703 handler
->ns_on_clone_flags
= handler
->ns_clone_flags
;
1704 if (handler
->ns_clone_flags
& CLONE_NEWUSER
) {
1705 /* If CLONE_NEWUSER and CLONE_NEWNET was requested, we need to
1706 * clone a new user namespace first and only later unshare our
1707 * network namespace to ensure that network devices ownership is
1710 handler
->ns_on_clone_flags
&= ~CLONE_NEWNET
;
1712 /* The cgroup namespace gets unshare()ed not clone()ed. */
1713 handler
->ns_on_clone_flags
&= ~CLONE_NEWCGROUP
;
1718 attacher_pid
= lxc_clone(do_share_ns
, handler
,
1719 CLONE_VFORK
| CLONE_VM
| CLONE_FILES
);
1720 if (attacher_pid
< 0) {
1721 SYSERROR(LXC_CLONE_ERROR
);
1722 goto out_delete_net
;
1725 ret
= wait_for_pid(attacher_pid
);
1727 SYSERROR("Intermediate process failed");
1728 goto out_delete_net
;
1731 handler
->pid
= lxc_raw_clone_cb(do_start
, handler
,
1732 handler
->ns_on_clone_flags
);
1734 if (handler
->pid
< 0) {
1735 SYSERROR(LXC_CLONE_ERROR
);
1736 goto out_delete_net
;
1738 TRACE("Cloned child process %d", handler
->pid
);
1740 for (i
= 0; i
< LXC_NS_MAX
; i
++)
1741 if (handler
->ns_on_clone_flags
& ns_info
[i
].clone_flag
)
1742 INFO("Cloned %s", ns_info
[i
].flag_name
);
1744 if (!lxc_try_preserve_namespaces(handler
, handler
->ns_on_clone_flags
, handler
->pid
)) {
1745 ERROR("Failed to preserve cloned namespaces for lxc.hook.stop");
1746 goto out_delete_net
;
1749 lxc_sync_fini_child(handler
);
1751 /* Map the container uids. The container became an invalid userid the
1752 * moment it was cloned with CLONE_NEWUSER. This call doesn't change
1753 * anything immediately, but allows the container to setuid(0) (0 being
1754 * mapped to something else on the host.) later to become a valid uid
1757 if (wants_to_map_ids
) {
1758 if (!handler
->conf
->ns_share
[LXC_NS_USER
] &&
1759 (handler
->conf
->ns_keep
& CLONE_NEWUSER
) == 0) {
1760 ret
= lxc_map_ids(id_map
, handler
->pid
);
1762 ERROR("Failed to set up id mapping.");
1763 goto out_delete_net
;
1768 ret
= lxc_sync_wake_child(handler
, LXC_SYNC_STARTUP
);
1770 goto out_delete_net
;
1772 ret
= lxc_sync_wait_child(handler
, LXC_SYNC_CONFIGURE
);
1774 goto out_delete_net
;
1776 if (!cgroup_ops
->setup_limits(cgroup_ops
, handler
->conf
, false)) {
1777 ERROR("Failed to setup cgroup limits for container \"%s\"", name
);
1778 goto out_delete_net
;
1781 if (!cgroup_ops
->payload_enter(cgroup_ops
, handler
->pid
))
1782 goto out_delete_net
;
1784 if (!cgroup_ops
->chown(cgroup_ops
, handler
->conf
))
1785 goto out_delete_net
;
1787 /* Now we're ready to preserve the network namespace */
1788 ret
= lxc_try_preserve_ns(handler
->pid
, "net");
1790 if (ret
!= -EOPNOTSUPP
) {
1791 SYSERROR("Failed to preserve net namespace");
1792 goto out_delete_net
;
1795 handler
->nsfd
[LXC_NS_NET
] = ret
;
1796 DEBUG("Preserved net namespace via fd %d", ret
);
1798 ret
= lxc_netns_set_nsid(handler
->nsfd
[LXC_NS_NET
]);
1800 SYSERROR("Failed to allocate new network namespace id");
1802 TRACE("Allocated new network namespace id");
1805 /* Create the network configuration. */
1806 if (handler
->ns_clone_flags
& CLONE_NEWNET
) {
1807 ret
= lxc_network_move_created_netdev_priv(handler
->lxcpath
,
1812 ERROR("Failed to create the configured network");
1813 goto out_delete_net
;
1816 ret
= lxc_create_network_unpriv(handler
->lxcpath
, handler
->name
,
1817 &conf
->network
, handler
->pid
, conf
->hooks_version
);
1819 ERROR("Failed to create the configured network");
1820 goto out_delete_net
;
1824 ret
= lxc_network_send_veth_names_to_child(handler
);
1826 ERROR("Failed to send veth names to child");
1827 goto out_delete_net
;
1830 if (!lxc_list_empty(&conf
->procs
)) {
1831 ret
= setup_proc_filesystem(&conf
->procs
, handler
->pid
);
1833 goto out_delete_net
;
1836 /* Tell the child to continue its initialization. We'll get
1837 * LXC_SYNC_CGROUP when it is ready for us to setup cgroups.
1839 ret
= lxc_sync_barrier_child(handler
, LXC_SYNC_POST_CONFIGURE
);
1841 goto out_delete_net
;
1843 if (!lxc_list_empty(&conf
->limits
)) {
1844 ret
= setup_resource_limits(&conf
->limits
, handler
->pid
);
1846 ERROR("Failed to setup resource limits");
1847 goto out_delete_net
;
1851 ret
= lxc_sync_barrier_child(handler
, LXC_SYNC_CGROUP_UNSHARE
);
1853 goto out_delete_net
;
1855 if (!cgroup_ops
->setup_limits(cgroup_ops
, handler
->conf
, true)) {
1856 ERROR("Failed to setup legacy device cgroup controller limits");
1857 goto out_delete_net
;
1859 TRACE("Set up legacy device cgroup controller limits");
1861 if (handler
->ns_clone_flags
& CLONE_NEWCGROUP
) {
1862 /* Now we're ready to preserve the cgroup namespace */
1863 ret
= lxc_try_preserve_ns(handler
->pid
, "cgroup");
1865 if (ret
!= -EOPNOTSUPP
) {
1866 SYSERROR("Failed to preserve cgroup namespace");
1867 goto out_delete_net
;
1870 handler
->nsfd
[LXC_NS_CGROUP
] = ret
;
1871 DEBUG("Preserved cgroup namespace via fd %d", ret
);
1875 ret
= snprintf(pidstr
, 20, "%d", handler
->pid
);
1876 if (ret
< 0 || ret
>= 20)
1877 goto out_delete_net
;
1879 ret
= setenv("LXC_PID", pidstr
, 1);
1881 SYSERROR("Failed to set environment variable: LXC_PID=%s", pidstr
);
1883 /* Run any host-side start hooks */
1884 ret
= run_lxc_hooks(name
, "start-host", conf
, NULL
);
1886 ERROR("Failed to run lxc.hook.start-host");
1887 goto out_delete_net
;
1890 /* Tell the child to complete its initialization and wait for it to exec
1891 * or return an error. (The child will never return
1892 * LXC_SYNC_READY_START+1. It will either close the sync pipe, causing
1893 * lxc_sync_barrier_child to return success, or return a different
1894 * value, causing us to error out).
1896 ret
= lxc_sync_barrier_child(handler
, LXC_SYNC_READY_START
);
1898 goto out_delete_net
;
1900 ret
= lxc_network_recv_name_and_ifindex_from_child(handler
);
1902 ERROR("Failed to receive names and ifindices for network "
1903 "devices from child");
1904 goto out_delete_net
;
1907 /* Now all networks are created, network devices are moved into place,
1908 * and the correct names and ifindeces in the respective namespaces have
1909 * been recorded. The corresponding structs have now all been filled. So
1910 * log them for debugging purposes.
1912 lxc_log_configured_netdevs(conf
);
1914 /* Read tty fds allocated by child. */
1915 ret
= lxc_recv_ttys_from_child(handler
);
1917 ERROR("Failed to receive tty info from child process");
1918 goto out_delete_net
;
1921 ret
= handler
->ops
->post_start(handler
, handler
->data
);
1925 ret
= lxc_set_state(name
, handler
, RUNNING
);
1927 ERROR("Failed to set state to \"%s\"", lxc_state2str(RUNNING
));
1931 lxc_sync_fini(handler
);
1936 if (handler
->ns_clone_flags
& CLONE_NEWNET
)
1937 lxc_delete_network(handler
);
1940 lxc_abort(name
, handler
);
1943 lxc_sync_fini(handler
);
1944 if (handler
->pinfd
>= 0) {
1945 close(handler
->pinfd
);
1946 handler
->pinfd
= -1;
1952 int __lxc_start(const char *name
, struct lxc_handler
*handler
,
1953 struct lxc_operations
* ops
, void *data
, const char *lxcpath
,
1954 bool daemonize
, int *error_num
)
1957 struct lxc_conf
*conf
= handler
->conf
;
1958 struct cgroup_ops
*cgroup_ops
;
1960 ret
= lxc_init(name
, handler
);
1962 ERROR("Failed to initialize container \"%s\"", name
);
1966 handler
->data
= data
;
1967 handler
->daemonize
= daemonize
;
1968 cgroup_ops
= handler
->cgroup_ops
;
1970 if (!attach_block_device(handler
->conf
)) {
1971 ERROR("Failed to attach block device");
1972 goto out_fini_nonet
;
1975 if (!cgroup_ops
->monitor_create(cgroup_ops
, handler
)) {
1976 ERROR("Failed to create monitor cgroup");
1977 goto out_fini_nonet
;
1980 if (!cgroup_ops
->monitor_enter(cgroup_ops
, handler
->monitor_pid
)) {
1981 ERROR("Failed to enter monitor cgroup");
1982 goto out_fini_nonet
;
1985 if (geteuid() == 0 && !lxc_list_empty(&conf
->id_map
)) {
1986 /* If the backing store is a device, mount it here and now. */
1987 if (rootfs_is_blockdev(conf
)) {
1988 ret
= unshare(CLONE_NEWNS
);
1990 ERROR("Failed to unshare CLONE_NEWNS");
1991 goto out_fini_nonet
;
1993 INFO("Unshared CLONE_NEWNS");
1995 remount_all_slave();
1996 ret
= lxc_setup_rootfs_prepare_root(conf
, name
, lxcpath
);
1998 ERROR("Error setting up rootfs mount as root before spawn");
1999 goto out_fini_nonet
;
2001 INFO("Set up container rootfs as host root");
2005 ret
= lxc_spawn(handler
);
2007 ERROR("Failed to spawn container \"%s\"", name
);
2008 goto out_detach_blockdev
;
2010 /* close parent side of data socket */
2011 close(handler
->data_sock
[0]);
2012 handler
->data_sock
[0] = -1;
2013 close(handler
->data_sock
[1]);
2014 handler
->data_sock
[1] = -1;
2016 handler
->conf
->reboot
= REBOOT_NONE
;
2018 ret
= lxc_poll(name
, handler
);
2020 ERROR("LXC mainloop exited with error: %d", ret
);
2024 if (!handler
->init_died
&& handler
->pid
> 0) {
2025 ERROR("Child process is not killed");
2029 status
= lxc_wait_for_pid_status(handler
->pid
);
2031 SYSERROR("Failed to retrieve status for %d", handler
->pid
);
2033 /* If the child process exited but was not signaled, it didn't call
2034 * reboot. This should mean it was an lxc-execute which simply exited.
2035 * In any case, treat it as a 'halt'.
2037 if (WIFSIGNALED(status
)) {
2038 switch(WTERMSIG(status
)) {
2039 case SIGINT
: /* halt */
2040 DEBUG("Container \"%s\" is halting", name
);
2042 case SIGHUP
: /* reboot */
2043 DEBUG("Container \"%s\" is rebooting", name
);
2044 handler
->conf
->reboot
= REBOOT_REQ
;
2046 case SIGSYS
: /* seccomp */
2047 DEBUG("Container \"%s\" violated its seccomp policy", name
);
2050 DEBUG("Unknown exit status for container \"%s\" init %d", name
, WTERMSIG(status
));
2055 ret
= lxc_restore_phys_nics_to_netns(handler
);
2057 ERROR("Failed to move physical network devices back to parent "
2058 "network namespace");
2060 if (handler
->pinfd
>= 0) {
2061 close(handler
->pinfd
);
2062 handler
->pinfd
= -1;
2065 lxc_monitor_send_exit_code(name
, status
, handler
->lxcpath
);
2066 lxc_error_set_and_log(handler
->pid
, status
);
2068 *error_num
= handler
->exit_status
;
2071 lxc_delete_network(handler
);
2073 out_detach_blockdev
:
2074 detach_block_device(handler
->conf
);
2077 lxc_fini(name
, handler
);
2081 lxc_abort(name
, handler
);
2089 static int start(struct lxc_handler
*handler
, void* data
)
2091 struct start_args
*arg
= data
;
2093 NOTICE("Exec'ing \"%s\"", arg
->argv
[0]);
2095 execvp(arg
->argv
[0], arg
->argv
);
2096 SYSERROR("Failed to exec \"%s\"", arg
->argv
[0]);
2100 static int post_start(struct lxc_handler
*handler
, void* data
)
2102 struct start_args
*arg
= data
;
2104 NOTICE("Started \"%s\" with pid \"%d\"", arg
->argv
[0], handler
->pid
);
2108 static struct lxc_operations start_ops
= {
2110 .post_start
= post_start
2113 int lxc_start(const char *name
, char *const argv
[], struct lxc_handler
*handler
,
2114 const char *lxcpath
, bool daemonize
, int *error_num
)
2116 struct start_args start_arg
= {
2120 TRACE("Doing lxc_start");
2121 return __lxc_start(name
, handler
, &start_ops
, &start_arg
, lxcpath
, daemonize
, error_num
);
2124 static void lxc_destroy_container_on_signal(struct lxc_handler
*handler
,
2127 char destroy
[MAXPATHLEN
];
2128 struct lxc_container
*c
;
2132 if (handler
->conf
->rootfs
.path
&& handler
->conf
->rootfs
.mount
) {
2133 bret
= do_destroy_container(handler
);
2135 ERROR("Error destroying rootfs for container \"%s\"", name
);
2139 INFO("Destroyed rootfs for container \"%s\"", name
);
2141 ret
= snprintf(destroy
, MAXPATHLEN
, "%s/%s", handler
->lxcpath
, name
);
2142 if (ret
< 0 || ret
>= MAXPATHLEN
) {
2143 ERROR("Error destroying directory for container \"%s\"", name
);
2147 c
= lxc_container_new(name
, handler
->lxcpath
);
2149 if (container_disk_lock(c
)) {
2150 INFO("Could not update lxc_snapshots file");
2151 lxc_container_put(c
);
2153 mod_all_rdeps(c
, false);
2154 container_disk_unlock(c
);
2155 lxc_container_put(c
);
2159 if (!handler
->am_root
)
2160 ret
= userns_exec_full(handler
->conf
, lxc_rmdir_onedev_wrapper
,
2161 destroy
, "lxc_rmdir_onedev_wrapper");
2163 ret
= lxc_rmdir_onedev(destroy
, NULL
);
2166 ERROR("Error destroying directory for container \"%s\"", name
);
2169 INFO("Destroyed directory for container \"%s\"", name
);
2172 static int lxc_rmdir_onedev_wrapper(void *data
)
2174 char *arg
= (char *) data
;
2175 return lxc_rmdir_onedev(arg
, NULL
);
2178 static bool do_destroy_container(struct lxc_handler
*handler
)
2182 if (!handler
->am_root
) {
2183 ret
= userns_exec_full(handler
->conf
, storage_destroy_wrapper
,
2184 handler
->conf
, "storage_destroy_wrapper");
2191 return storage_destroy(handler
->conf
);