]>
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>
52 #include <sys/capability.h>
55 #if !HAVE_DECL_PR_CAPBSET_DROP
56 #define PR_CAPBSET_DROP 24
59 #if !HAVE_DECL_PR_SET_NO_NEW_PRIVS
60 #define PR_SET_NO_NEW_PRIVS 38
63 #if !HAVE_DECL_PR_GET_NO_NEW_PRIVS
64 #define PR_GET_NO_NEW_PRIVS 39
71 #include "commands_utils.h"
73 #include "confile_utils.h"
78 #include "lxccontainer.h"
80 #include "lxcseccomp.h"
83 #include "namespace.h"
87 #include "storage_utils.h"
92 lxc_log_define(lxc_start
, lxc
);
94 extern void mod_all_rdeps(struct lxc_container
*c
, bool inc
);
95 static bool do_destroy_container(struct lxc_handler
*handler
);
96 static int lxc_rmdir_onedev_wrapper(void *data
);
97 static void lxc_destroy_container_on_signal(struct lxc_handler
*handler
,
100 static void print_top_failing_dir(const char *path
)
102 size_t len
= strlen(path
);
103 char *copy
= alloca(len
+ 1), *p
, *e
, saved
;
109 while (p
< e
&& *p
== '/')
111 while (p
< e
&& *p
!= '/')
115 if (access(copy
, X_OK
)) {
116 SYSERROR("Could not access %s. Please grant it x "
117 "access, or add an ACL for the container "
125 static void close_ns(int ns_fd
[LXC_NS_MAX
])
129 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
137 /* preserve_ns: open /proc/@pid/ns/@ns for each namespace specified
139 * Return true on success, false on failure.
141 static bool preserve_ns(int ns_fd
[LXC_NS_MAX
], int clone_flags
, pid_t pid
)
145 for (i
= 0; i
< LXC_NS_MAX
; i
++)
148 ret
= lxc_preserve_ns(pid
, "");
150 SYSERROR("Kernel does not support attaching to namespaces.");
156 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
157 if ((clone_flags
& ns_info
[i
].clone_flag
) == 0)
160 ns_fd
[i
] = lxc_preserve_ns(pid
, ns_info
[i
].proc_name
);
164 DEBUG("Preserved %s namespace via fd %d", ns_info
[i
].proc_name
, ns_fd
[i
]);
171 SYSERROR("Kernel does not support attaching to %s namespaces.", ns_info
[i
].proc_name
);
173 SYSERROR("Failed to open file descriptor for %s namespace: %s.", ns_info
[i
].proc_name
, strerror(errno
));
178 static int match_fd(int fd
)
180 return (fd
== 0 || fd
== 1 || fd
== 2);
183 int lxc_check_inherited(struct lxc_conf
*conf
, bool closeall
,
184 int *fds_to_ignore
, size_t len_fds
)
186 struct dirent
*direntp
;
191 if (conf
&& conf
->close_all_fds
)
195 dir
= opendir("/proc/self/fd");
197 WARN("Failed to open directory: %s.", strerror(errno
));
203 while ((direntp
= readdir(dir
))) {
204 struct lxc_list
*cur
;
205 bool matched
= false;
207 if (!strcmp(direntp
->d_name
, "."))
210 if (!strcmp(direntp
->d_name
, ".."))
213 if (lxc_safe_int(direntp
->d_name
, &fd
) < 0) {
214 INFO("Could not parse file descriptor for: %s", direntp
->d_name
);
218 for (i
= 0; i
< len_fds
; i
++)
219 if (fds_to_ignore
[i
] == fd
)
222 if (fd
== fddir
|| fd
== lxc_log_fd
||
223 (i
< len_fds
&& fd
== fds_to_ignore
[i
]))
226 /* Keep state clients that wait on reboots. */
228 lxc_list_for_each(cur
, &conf
->state_clients
) {
229 struct lxc_state_client
*client
= cur
->elem
;
231 if (client
->clientfd
!= fd
)
242 if (current_config
&& fd
== current_config
->logfd
)
251 INFO("Closed inherited fd %d", fd
);
254 WARN("Inherited fd %d", fd
);
257 /* Only enable syslog at this point to avoid the above logging function
258 * to open a new fd and make the check_inherited function enter an
261 lxc_log_enable_syslog();
263 closedir(dir
); /* cannot fail */
267 static int setup_signal_fd(sigset_t
*oldmask
)
271 int signals
[] = {SIGBUS
, SIGILL
, SIGSEGV
, SIGWINCH
};
273 /* Block everything except serious error signals. */
274 ret
= sigfillset(&mask
);
278 for (sig
= 0; sig
< (sizeof(signals
) / sizeof(signals
[0])); sig
++) {
279 ret
= sigdelset(&mask
, signals
[sig
]);
284 ret
= sigprocmask(SIG_BLOCK
, &mask
, oldmask
);
286 SYSERROR("Failed to set signal mask");
290 ret
= signalfd(-1, &mask
, SFD_CLOEXEC
);
292 SYSERROR("Failed to create signal file descriptor");
296 TRACE("Created signal file descriptor %d", ret
);
301 static int signal_handler(int fd
, uint32_t events
, void *data
,
302 struct lxc_epoll_descr
*descr
)
306 struct signalfd_siginfo siginfo
;
307 struct lxc_handler
*hdlr
= data
;
309 ret
= read(fd
, &siginfo
, sizeof(siginfo
));
311 ERROR("Failed to read signal info from signal file descriptor: %d.", fd
);
315 if (ret
!= sizeof(siginfo
)) {
316 ERROR("Unexpected size for siginfo struct.");
320 /* Check whether init is running. */
322 ret
= waitid(P_PID
, hdlr
->pid
, &info
, WEXITED
| WNOWAIT
| WNOHANG
);
323 if (ret
== 0 && info
.si_pid
== hdlr
->pid
)
324 hdlr
->init_died
= true;
326 /* Try to figure out a reasonable exit status to report. */
327 if (hdlr
->init_died
) {
328 switch (info
.si_code
) {
330 hdlr
->exit_status
= info
.si_status
<< 8;
335 hdlr
->exit_status
= info
.si_status
<< 8 | 0x7f;
338 /* Huh? The waitid() told us it's dead *and* continued? */
339 WARN("Init %d dead and continued?", hdlr
->pid
);
340 hdlr
->exit_status
= 1;
343 ERROR("Unknown si_code: %d", hdlr
->init_died
);
344 hdlr
->exit_status
= 1;
348 /* More robustness, protect ourself from a SIGCHLD sent
349 * by a process different from the container init.
351 if (siginfo
.ssi_pid
!= hdlr
->pid
) {
352 NOTICE("Received %d from pid %d instead of container init %d.", siginfo
.ssi_signo
, siginfo
.ssi_pid
, hdlr
->pid
);
353 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
: 0;
356 if (siginfo
.ssi_signo
!= SIGCHLD
) {
357 kill(hdlr
->pid
, siginfo
.ssi_signo
);
358 INFO("Forwarded signal %d to pid %d.", siginfo
.ssi_signo
, hdlr
->pid
);
359 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
: 0;
362 if (siginfo
.ssi_code
== CLD_STOPPED
) {
363 INFO("Container init process was stopped.");
364 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
: 0;
365 } else if (siginfo
.ssi_code
== CLD_CONTINUED
) {
366 INFO("Container init process was continued.");
367 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
: 0;
370 DEBUG("Container init process %d exited.", hdlr
->pid
);
371 return LXC_MAINLOOP_CLOSE
;
374 int lxc_serve_state_clients(const char *name
, struct lxc_handler
*handler
,
378 struct lxc_list
*cur
, *next
;
379 struct lxc_state_client
*client
;
380 struct lxc_msg msg
= {.type
= lxc_msg_state
, .value
= state
};
384 handler
->state
= RUNNING
;
386 handler
->state
= state
;
388 TRACE("Set container state to %s", lxc_state2str(state
));
390 if (lxc_list_empty(&handler
->conf
->state_clients
)) {
391 TRACE("No state clients registered");
393 lxc_monitor_send_state(name
, state
, handler
->lxcpath
);
397 strncpy(msg
.name
, name
, sizeof(msg
.name
));
398 msg
.name
[sizeof(msg
.name
) - 1] = 0;
400 lxc_list_for_each_safe(cur
, &handler
->conf
->state_clients
, next
) {
403 if (client
->states
[state
] == 0) {
404 TRACE("State %s not registered for state client %d",
405 lxc_state2str(state
), client
->clientfd
);
409 TRACE("Sending state %s to state client %d",
410 lxc_state2str(state
), client
->clientfd
);
413 ret
= send(client
->clientfd
, &msg
, sizeof(msg
), 0);
415 if (errno
== EINTR
) {
416 TRACE("Caught EINTR; retrying");
420 ERROR("%s - Failed to send message to client",
424 /* kick client from list */
425 close(client
->clientfd
);
435 static int lxc_serve_state_socket_pair(const char *name
,
436 struct lxc_handler
*handler
,
441 if (!handler
->backgrounded
||
442 handler
->state_socket_pair
[1] < 0 ||
446 /* Close read end of the socket pair. */
447 close(handler
->state_socket_pair
[0]);
448 handler
->state_socket_pair
[0] = -1;
451 ret
= lxc_abstract_unix_send_credential(handler
->state_socket_pair
[1],
452 &(int){state
}, sizeof(int));
453 if (ret
!= sizeof(int)) {
456 SYSERROR("Failed to send state to %d",
457 handler
->state_socket_pair
[1]);
461 TRACE("Sent container state \"%s\" to %d", lxc_state2str(state
),
462 handler
->state_socket_pair
[1]);
464 /* Close write end of the socket pair. */
465 close(handler
->state_socket_pair
[1]);
466 handler
->state_socket_pair
[1] = -1;
471 int lxc_set_state(const char *name
, struct lxc_handler
*handler
,
476 ret
= lxc_serve_state_socket_pair(name
, handler
, state
);
478 ERROR("Failed to synchronize via anonymous pair of unix sockets");
482 ret
= lxc_serve_state_clients(name
, handler
, state
);
486 /* This function will try to connect to the legacy lxc-monitord state
487 * server and only exists for backwards compatibility.
489 lxc_monitor_send_state(name
, state
, handler
->lxcpath
);
494 int lxc_poll(const char *name
, struct lxc_handler
*handler
)
497 bool has_console
= true;
498 struct lxc_epoll_descr descr
, descr_console
;
500 if (handler
->conf
->console
.path
&& !strcmp(handler
->conf
->console
.path
, "none"))
503 ret
= lxc_mainloop_open(&descr
);
505 ERROR("Failed to create mainloop");
510 ret
= lxc_mainloop_open(&descr_console
);
512 ERROR("Failed to create console mainloop");
517 ret
= lxc_mainloop_add_handler(&descr
, handler
->sigfd
, signal_handler
, handler
);
519 ERROR("Failed to add signal handler for %d to mainloop", handler
->sigfd
);
520 goto out_mainloop_console
;
524 struct lxc_console
*console
= &handler
->conf
->console
;
526 ret
= lxc_console_mainloop_add(&descr
, console
);
528 ERROR("Failed to add console handlers to mainloop");
529 goto out_mainloop_console
;
532 ret
= lxc_console_mainloop_add(&descr_console
, console
);
534 ERROR("Failed to add console handlers to console mainloop");
535 goto out_mainloop_console
;
539 ret
= lxc_cmd_mainloop_add(name
, &descr
, handler
);
541 ERROR("Failed to add command handler to mainloop");
542 goto out_mainloop_console
;
545 TRACE("Mainloop is ready");
547 ret
= lxc_mainloop(&descr
, -1);
550 if (ret
< 0 || !handler
->init_died
)
554 ret
= lxc_mainloop(&descr_console
, 0);
558 lxc_mainloop_close(&descr
);
559 TRACE("Closed mainloop");
561 out_mainloop_console
:
563 lxc_mainloop_close(&descr_console
);
564 TRACE("Closed console mainloop");
568 close(handler
->sigfd
);
569 TRACE("Closed signal file descriptor %d", handler
->sigfd
);
570 handler
->sigfd
= -EBADF
;
575 void lxc_zero_handler(struct lxc_handler
*handler
)
579 memset(handler
, 0, sizeof(struct lxc_handler
));
581 handler
->clone_flags
= -1;
587 for (i
= 0; i
< LXC_NS_MAX
; i
++)
588 handler
->nsfd
[i
] = -1;
590 handler
->data_sock
[0] = -1;
591 handler
->data_sock
[1] = -1;
593 handler
->state_socket_pair
[0] = -1;
594 handler
->state_socket_pair
[1] = -1;
596 handler
->sync_sock
[0] = -1;
597 handler
->sync_sock
[1] = -1;
600 static void lxc_put_nsfds(struct lxc_handler
*handler
)
604 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
605 if (handler
->nsfd
[i
] < 0)
608 close(handler
->nsfd
[i
]);
609 handler
->nsfd
[i
] = -EBADF
;
613 void lxc_free_handler(struct lxc_handler
*handler
)
615 if (handler
->pinfd
>= 0)
616 close(handler
->pinfd
);
618 if (handler
->sigfd
>= 0)
619 close(handler
->sigfd
);
621 lxc_put_nsfds(handler
);
623 if (handler
->conf
&& handler
->conf
->reboot
== 0)
624 if (handler
->conf
->maincmd_fd
)
625 close(handler
->conf
->maincmd_fd
);
627 if (handler
->state_socket_pair
[0] >= 0)
628 close(handler
->state_socket_pair
[0]);
630 if (handler
->state_socket_pair
[1] >= 0)
631 close(handler
->state_socket_pair
[1]);
633 handler
->conf
= NULL
;
638 struct lxc_handler
*lxc_init_handler(const char *name
, struct lxc_conf
*conf
,
639 const char *lxcpath
, bool daemonize
)
642 struct lxc_handler
*handler
;
644 handler
= malloc(sizeof(*handler
));
646 ERROR("failed to allocate memory");
650 memset(handler
, 0, sizeof(*handler
));
652 /* Note that am_unpriv() checks the effective uid. We probably don't
653 * care if we are real root only if we are running as root so this
656 handler
->am_root
= !am_unpriv();
657 handler
->data_sock
[0] = handler
->data_sock
[1] = -1;
658 handler
->conf
= conf
;
659 handler
->lxcpath
= lxcpath
;
661 handler
->sigfd
= -EBADF
;
662 handler
->init_died
= false;
663 handler
->state_socket_pair
[0] = handler
->state_socket_pair
[1] = -1;
664 if (handler
->conf
->reboot
== 0)
665 lxc_list_init(&handler
->conf
->state_clients
);
667 for (i
= 0; i
< LXC_NS_MAX
; i
++)
668 handler
->nsfd
[i
] = -1;
670 handler
->name
= name
;
672 if (daemonize
&& !handler
->conf
->reboot
) {
673 /* Create socketpair() to synchronize on daemonized startup.
674 * When the container reboots we don't need to synchronize again
675 * currently so don't open another socketpair().
677 ret
= socketpair(AF_UNIX
, SOCK_STREAM
| SOCK_CLOEXEC
, 0,
678 handler
->state_socket_pair
);
680 ERROR("Failed to create anonymous pair of unix sockets");
683 TRACE("Created anonymous pair {%d,%d} of unix sockets",
684 handler
->state_socket_pair
[0],
685 handler
->state_socket_pair
[1]);
688 if (handler
->conf
->reboot
== 0) {
689 handler
->conf
->maincmd_fd
= lxc_cmd_init(name
, lxcpath
, "command");
690 if (handler
->conf
->maincmd_fd
< 0) {
691 ERROR("Failed to set up command socket");
695 TRACE("Unix domain socket %d for command server is ready",
696 handler
->conf
->maincmd_fd
);
701 lxc_free_handler(handler
);
706 int lxc_init(const char *name
, struct lxc_handler
*handler
)
709 const char *loglevel
;
710 struct lxc_conf
*conf
= handler
->conf
;
713 TRACE("initialized LSM");
715 if (lxc_read_seccomp_config(conf
) != 0) {
716 ERROR("Failed loading seccomp policy.");
717 goto out_close_maincmd_fd
;
719 TRACE("read seccomp policy");
721 /* Begin by setting the state to STARTING. */
722 if (lxc_set_state(name
, handler
, STARTING
)) {
723 ERROR("Failed to set state for container \"%s\" to \"%s\".", name
, lxc_state2str(STARTING
));
724 goto out_close_maincmd_fd
;
726 TRACE("set container state to \"STARTING\"");
728 /* Start of environment variable setup for hooks. */
729 if (name
&& setenv("LXC_NAME", name
, 1))
730 SYSERROR("Failed to set environment variable: LXC_NAME=%s.", name
);
732 if (conf
->rcfile
&& setenv("LXC_CONFIG_FILE", conf
->rcfile
, 1))
733 SYSERROR("Failed to set environment variable: LXC_CONFIG_FILE=%s.", conf
->rcfile
);
735 if (conf
->rootfs
.mount
&& setenv("LXC_ROOTFS_MOUNT", conf
->rootfs
.mount
, 1))
736 SYSERROR("Failed to set environment variable: LXC_ROOTFS_MOUNT=%s.", conf
->rootfs
.mount
);
738 if (conf
->rootfs
.path
&& setenv("LXC_ROOTFS_PATH", conf
->rootfs
.path
, 1))
739 SYSERROR("Failed to set environment variable: LXC_ROOTFS_PATH=%s.", conf
->rootfs
.path
);
741 if (conf
->console
.path
&& setenv("LXC_CONSOLE", conf
->console
.path
, 1))
742 SYSERROR("Failed to set environment variable: LXC_CONSOLE=%s.", conf
->console
.path
);
744 if (conf
->console
.log_path
&& setenv("LXC_CONSOLE_LOGPATH", conf
->console
.log_path
, 1))
745 SYSERROR("Failed to set environment variable: LXC_CONSOLE_LOGPATH=%s.", conf
->console
.log_path
);
747 if (setenv("LXC_CGNS_AWARE", "1", 1))
748 SYSERROR("Failed to set environment variable LXC_CGNS_AWARE=1.");
750 loglevel
= lxc_log_priority_to_string(lxc_log_get_level());
751 if (setenv("LXC_LOG_LEVEL", loglevel
, 1))
752 SYSERROR("Failed to set environment variable LXC_LOG_LEVEL=%s", loglevel
);
754 if (conf
->hooks_version
== 0)
755 ret
= setenv("LXC_HOOK_VERSION", "0", 1);
757 ret
= setenv("LXC_HOOK_VERSION", "1", 1);
759 SYSERROR("Failed to set environment variable LXC_HOOK_VERSION=%u", conf
->hooks_version
);
760 /* End of environment variable setup for hooks. */
762 TRACE("set environment variables");
764 if (run_lxc_hooks(name
, "pre-start", conf
, NULL
)) {
765 ERROR("Failed to run lxc.hook.pre-start for container \"%s\".", name
);
768 TRACE("ran pre-start hooks");
770 /* The signal fd has to be created before forking otherwise if the child
771 * process exits before we setup the signal fd, the event will be lost
772 * and the command will be stuck.
774 handler
->sigfd
= setup_signal_fd(&handler
->oldmask
);
775 if (handler
->sigfd
< 0) {
776 ERROR("Failed to setup SIGCHLD fd handler.");
779 TRACE("set up signal fd");
781 /* Do this after setting up signals since it might unblock SIGWINCH. */
782 ret
= lxc_console_create(conf
);
784 ERROR("Failed to create console");
785 goto out_restore_sigmask
;
787 TRACE("Created console");
789 ret
= lxc_pty_map_ids(conf
, &conf
->console
);
791 ERROR("Failed to shift tty into container.");
792 goto out_restore_sigmask
;
794 TRACE("shifted tty ids");
796 INFO("container \"%s\" is initialized", name
);
800 sigprocmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
802 lxc_delete_tty(&conf
->tty_info
);
804 lxc_set_state(name
, handler
, ABORTING
);
805 out_close_maincmd_fd
:
806 close(conf
->maincmd_fd
);
807 conf
->maincmd_fd
= -1;
811 void lxc_fini(const char *name
, struct lxc_handler
*handler
)
815 struct lxc_list
*cur
, *next
;
816 char *namespaces
[LXC_NS_MAX
+ 1];
817 size_t namespace_count
= 0;
819 /* The STOPPING state is there for future cleanup code which can take
822 lxc_set_state(name
, handler
, STOPPING
);
824 self
= lxc_raw_getpid();
825 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
826 if (handler
->nsfd
[i
] < 0)
829 if (handler
->conf
->hooks_version
== 0)
830 rc
= asprintf(&namespaces
[namespace_count
],
831 "%s:/proc/%d/fd/%d", ns_info
[i
].proc_name
,
832 self
, handler
->nsfd
[i
]);
834 rc
= asprintf(&namespaces
[namespace_count
],
835 "/proc/%d/fd/%d", self
, handler
->nsfd
[i
]);
837 SYSERROR("Failed to allocate memory.");
841 if (handler
->conf
->hooks_version
== 0) {
846 rc
= setenv(ns_info
[i
].env_name
, namespaces
[namespace_count
], 1);
848 SYSERROR("Failed to set environment variable %s=%s",
849 ns_info
[i
].env_name
, namespaces
[namespace_count
]);
851 TRACE("Set environment variable %s=%s",
852 ns_info
[i
].env_name
, namespaces
[namespace_count
]);
856 namespaces
[namespace_count
] = NULL
;
858 if (handler
->conf
->reboot
&& setenv("LXC_TARGET", "reboot", 1))
859 SYSERROR("Failed to set environment variable: LXC_TARGET=reboot.");
861 if (!handler
->conf
->reboot
&& setenv("LXC_TARGET", "stop", 1))
862 SYSERROR("Failed to set environment variable: LXC_TARGET=stop.");
864 if (handler
->conf
->hooks_version
== 0)
865 rc
= run_lxc_hooks(name
, "stop", handler
->conf
, namespaces
);
867 rc
= run_lxc_hooks(name
, "stop", handler
->conf
, NULL
);
869 while (namespace_count
--)
870 free(namespaces
[namespace_count
]);
872 cgroup_destroy(handler
);
874 if (handler
->conf
->reboot
== 0) {
875 /* For all new state clients simply close the command socket.
876 * This will inform all state clients that the container is
877 * STOPPED and also prevents a race between a open()/close() on
878 * the command socket causing a new process to get ECONNREFUSED
879 * because we haven't yet closed the command socket.
881 close(handler
->conf
->maincmd_fd
);
882 handler
->conf
->maincmd_fd
= -1;
883 TRACE("Closed command socket");
885 /* This function will try to connect to the legacy lxc-monitord
886 * state server and only exists for backwards compatibility.
888 lxc_monitor_send_state(name
, STOPPED
, handler
->lxcpath
);
890 /* The command socket is closed so no one can acces the command
891 * socket anymore so there's no need to lock it.
893 handler
->state
= STOPPED
;
894 TRACE("Set container state to \"STOPPED\"");
896 lxc_set_state(name
, handler
, STOPPED
);
899 if (run_lxc_hooks(name
, "post-stop", handler
->conf
, NULL
)) {
900 ERROR("Failed to run lxc.hook.post-stop for container \"%s\".", name
);
901 if (handler
->conf
->reboot
) {
902 WARN("Container will be stopped instead of rebooted.");
903 handler
->conf
->reboot
= 0;
904 if (setenv("LXC_TARGET", "stop", 1))
905 WARN("Failed to set environment variable: LXC_TARGET=stop.");
909 /* Reset mask set by setup_signal_fd. */
910 if (sigprocmask(SIG_SETMASK
, &handler
->oldmask
, NULL
))
911 WARN("Failed to restore signal mask.");
913 lxc_console_delete(&handler
->conf
->console
);
914 lxc_delete_tty(&handler
->conf
->tty_info
);
916 /* The command socket is now closed, no more state clients can register
917 * themselves from now on. So free the list of state clients.
919 lxc_list_for_each_safe(cur
, &handler
->conf
->state_clients
, next
) {
920 struct lxc_state_client
*client
= cur
->elem
;
922 /* Keep state clients that want to be notified about reboots. */
923 if ((handler
->conf
->reboot
> 0) && (client
->states
[RUNNING
] == 2))
926 /* close state client socket */
927 close(client
->clientfd
);
933 if (handler
->conf
->ephemeral
== 1 && handler
->conf
->reboot
!= 1)
934 lxc_destroy_container_on_signal(handler
, name
);
936 lxc_free_handler(handler
);
939 void lxc_abort(const char *name
, struct lxc_handler
*handler
)
943 lxc_set_state(name
, handler
, ABORTING
);
944 if (handler
->pid
> 0)
945 kill(handler
->pid
, SIGKILL
);
946 while ((ret
= waitpid(-1, &status
, 0)) > 0) {
951 static int lxc_set_death_signal(int signal
)
956 ret
= prctl(PR_SET_PDEATHSIG
, signal
, 0, 0, 0);
958 /* Check whether we have been orphaned. */
959 ppid
= (pid_t
)syscall(SYS_getppid
);
963 self
= lxc_raw_getpid();
964 ret
= kill(self
, SIGKILL
);
970 SYSERROR("Failed to set PR_SET_PDEATHSIG to %d", signal
);
977 static int do_start(void *data
)
980 struct lxc_list
*iterator
;
982 struct lxc_handler
*handler
= data
;
983 bool have_cap_setgid
;
988 lxc_sync_fini_parent(handler
);
990 /* This prctl must be before the synchro, so if the parent dies before
991 * we set the parent death signal, we will detect its death with the
992 * synchro right after, otherwise we have a window where the parent can
993 * exit before we set the pdeath signal leading to a unsupervized
996 ret
= lxc_set_death_signal(SIGKILL
);
998 SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL");
999 goto out_warn_father
;
1002 ret
= sigprocmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
1004 SYSERROR("Failed to set signal mask");
1005 goto out_warn_father
;
1008 /* Don't leak the pinfd to the container. */
1009 if (handler
->pinfd
>= 0)
1010 close(handler
->pinfd
);
1012 ret
= lxc_sync_wait_parent(handler
, LXC_SYNC_STARTUP
);
1014 goto out_warn_father
;
1016 /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See
1017 * https://github.com/lxc/lxd/issues/1978.
1019 if ((handler
->clone_flags
& (CLONE_NEWNET
| CLONE_NEWUSER
)) ==
1020 (CLONE_NEWNET
| CLONE_NEWUSER
)) {
1021 ret
= unshare(CLONE_NEWNET
);
1023 SYSERROR("Failed to unshare CLONE_NEWNET.");
1024 goto out_warn_father
;
1026 INFO("Unshared CLONE_NEWNET.");
1029 /* Tell the parent task it can begin to configure the container and wait
1032 if (lxc_sync_barrier_parent(handler
, LXC_SYNC_CONFIGURE
))
1035 if (lxc_network_recv_veth_names_from_parent(handler
) < 0) {
1036 ERROR("Failed to receive veth names from parent");
1037 goto out_warn_father
;
1040 /* If we are in a new user namespace, become root there to have
1041 * privilege over our namespace.
1043 if (!lxc_list_empty(&handler
->conf
->id_map
)) {
1044 uid_t nsuid
= (handler
->conf
->root_nsuid_map
!= NULL
)
1046 : handler
->conf
->init_uid
;
1047 gid_t nsgid
= (handler
->conf
->root_nsgid_map
!= NULL
)
1049 : handler
->conf
->init_gid
;
1051 ret
= lxc_switch_uid_gid(nsuid
, nsgid
);
1053 goto out_warn_father
;
1055 /* Drop groups only after we switched to a valid gid in the new
1058 ret
= lxc_setgroups(0, NULL
);
1059 if (ret
< 0 && (handler
->am_root
|| errno
!= EPERM
))
1060 goto out_warn_father
;
1062 if (!handler
->am_root
) {
1063 ret
= prctl(PR_SET_DUMPABLE
, 1, 0, 0, 0);
1065 goto out_warn_father
;
1068 /* set{g,u}id() clears deathsignal */
1069 ret
= lxc_set_death_signal(SIGKILL
);
1071 SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL");
1072 goto out_warn_father
;
1076 if (access(handler
->lxcpath
, X_OK
)) {
1077 print_top_failing_dir(handler
->lxcpath
);
1078 goto out_warn_father
;
1081 ret
= snprintf(path
, sizeof(path
), "%s/dev/null", handler
->conf
->rootfs
.mount
);
1082 if (ret
< 0 || ret
>= sizeof(path
))
1083 goto out_warn_father
;
1085 /* In order to checkpoint restore, we need to have everything in the
1086 * same mount namespace. However, some containers may not have a
1087 * reasonable /dev (in particular, they may not have /dev/null), so we
1088 * can't set init's std fds to /dev/null by opening it from inside the
1091 * If that's the case, fall back to using the host's /dev/null. This
1092 * means that migration won't work, but at least we won't spew output
1093 * where it isn't wanted.
1095 if (handler
->backgrounded
&& !handler
->conf
->autodev
&& access(path
, F_OK
) < 0) {
1096 devnull_fd
= open_devnull();
1099 goto out_warn_father
;
1100 WARN("Using /dev/null from the host for container init's "
1101 "standard file descriptors. Migration will not work.");
1104 /* Ask father to setup cgroups and wait for him to finish. */
1105 if (lxc_sync_barrier_parent(handler
, LXC_SYNC_CGROUP
))
1108 /* Unshare cgroup namespace after we have setup our cgroups. If we do it
1109 * earlier we end up with a wrong view of /proc/self/cgroup. For
1110 * example, assume we unshare(CLONE_NEWCGROUP) first, and then create
1111 * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then
1112 * /proc/self/cgroup would show us:
1116 * whereas it should actually show
1120 if (handler
->clone_flags
& CLONE_NEWCGROUP
) {
1121 if (unshare(CLONE_NEWCGROUP
) < 0) {
1122 INFO("Failed to unshare CLONE_NEWCGROUP.");
1123 goto out_warn_father
;
1125 INFO("Unshared CLONE_NEWCGROUP.");
1128 /* Add the requested environment variables to the current environment to
1129 * allow them to be used by the various hooks, such as the start hook
1132 lxc_list_for_each(iterator
, &handler
->conf
->environment
) {
1133 if (putenv((char *)iterator
->elem
)) {
1134 SYSERROR("Failed to set environment variable: %s.", (char *)iterator
->elem
);
1135 goto out_warn_father
;
1139 /* Setup the container, ip, names, utsname, ... */
1140 ret
= lxc_setup(handler
);
1141 close(handler
->data_sock
[1]);
1142 close(handler
->data_sock
[0]);
1144 ERROR("Failed to setup container \"%s\".", handler
->name
);
1145 goto out_warn_father
;
1148 /* Set the label to change to when we exec(2) the container's init. */
1149 if (lsm_process_label_set(NULL
, handler
->conf
, 1, 1) < 0)
1150 goto out_warn_father
;
1152 /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it
1153 * before we aren't allowed anymore.
1155 if (handler
->conf
->no_new_privs
) {
1156 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
1157 SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges.");
1158 goto out_warn_father
;
1160 DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges.");
1163 /* Some init's such as busybox will set sane tty settings on stdin,
1164 * stdout, stderr which it thinks is the console. We already set them
1165 * the way we wanted on the real terminal, and we want init to do its
1166 * setup on its console ie. the pty allocated in lxc_console_create() so
1167 * make sure that that pty is stdin,stdout,stderr.
1169 if (handler
->conf
->console
.slave
>= 0) {
1170 if (handler
->backgrounded
|| handler
->conf
->is_execute
== 0)
1171 ret
= set_stdfds(handler
->conf
->console
.slave
);
1173 ret
= lxc_console_set_stdfds(handler
->conf
->console
.slave
);
1175 ERROR("Failed to redirect std{in,out,err} to pty file "
1177 handler
->conf
->console
.slave
);
1178 goto out_warn_father
;
1182 /* If we mounted a temporary proc, then unmount it now. */
1183 tmp_proc_unmount(handler
->conf
);
1185 if (lxc_seccomp_load(handler
->conf
) != 0)
1186 goto out_warn_father
;
1188 if (run_lxc_hooks(handler
->name
, "start", handler
->conf
, NULL
)) {
1189 ERROR("Failed to run lxc.hook.start for container \"%s\".", handler
->name
);
1190 goto out_warn_father
;
1193 close(handler
->sigfd
);
1195 if (devnull_fd
< 0) {
1196 devnull_fd
= open_devnull();
1199 goto out_warn_father
;
1202 if (handler
->conf
->console
.slave
< 0 && handler
->backgrounded
)
1203 if (set_stdfds(devnull_fd
) < 0) {
1204 ERROR("Failed to redirect std{in,out,err} to "
1206 goto out_warn_father
;
1209 if (devnull_fd
>= 0) {
1216 if (handler
->conf
->init_cwd
&& chdir(handler
->conf
->init_cwd
)) {
1217 SYSERROR("Could not change directory to \"%s\"", handler
->conf
->init_cwd
);
1218 goto out_warn_father
;
1221 if (lxc_sync_barrier_parent(handler
, LXC_SYNC_CGROUP_LIMITS
))
1222 goto out_warn_father
;
1224 /* Reset the environment variables the user requested in a clear
1228 SYSERROR("Failed to clear environment.");
1229 /* Don't error out though. */
1232 lxc_list_for_each(iterator
, &handler
->conf
->environment
) {
1233 if (putenv((char *)iterator
->elem
)) {
1234 SYSERROR("Failed to set environment variable: %s.", (char *)iterator
->elem
);
1235 goto out_warn_father
;
1239 if (putenv("container=lxc")) {
1240 SYSERROR("Failed to set environment variable: container=lxc.");
1241 goto out_warn_father
;
1244 if (handler
->conf
->pty_names
) {
1245 if (putenv(handler
->conf
->pty_names
)) {
1246 SYSERROR("Failed to set environment variable for container ptys.");
1247 goto out_warn_father
;
1251 /* The container has been setup. We can now switch to an unprivileged
1254 new_uid
= handler
->conf
->init_uid
;
1255 new_gid
= handler
->conf
->init_gid
;
1257 /* If we are in a new user namespace we already dropped all
1258 * groups when we switched to root in the new user namespace
1259 * further above. Only drop groups if we can, so ensure that we
1260 * have necessary privilege.
1263 have_cap_setgid
= lxc_proc_cap_is_set(CAP_SETGID
, CAP_EFFECTIVE
);
1265 have_cap_setgid
= false;
1267 if (lxc_list_empty(&handler
->conf
->id_map
) && have_cap_setgid
) {
1268 if (lxc_setgroups(0, NULL
) < 0)
1269 goto out_warn_father
;
1272 if (lxc_switch_uid_gid(new_uid
, new_gid
) < 0)
1273 goto out_warn_father
;
1275 /* After this call, we are in error because this ops should not return
1278 handler
->ops
->start(handler
, handler
->data
);
1281 /* We want the parent to know something went wrong, so we return a
1282 * special error code.
1284 lxc_sync_wake_parent(handler
, LXC_SYNC_ERROR
);
1287 if (devnull_fd
>= 0)
1293 static int lxc_recv_ttys_from_child(struct lxc_handler
*handler
)
1296 struct lxc_pty_info
*pty_info
;
1298 int sock
= handler
->data_sock
[1];
1299 struct lxc_conf
*conf
= handler
->conf
;
1300 struct lxc_tty_info
*tty_info
= &conf
->tty_info
;
1305 tty_info
->pty_info
= malloc(sizeof(*tty_info
->pty_info
) * conf
->tty
);
1306 if (!tty_info
->pty_info
)
1309 for (i
= 0; i
< conf
->tty
; i
++) {
1312 ret
= lxc_abstract_unix_recv_fds(sock
, ttyfds
, 2, NULL
, 0);
1316 pty_info
= &tty_info
->pty_info
[i
];
1318 pty_info
->master
= ttyfds
[0];
1319 pty_info
->slave
= ttyfds
[1];
1320 TRACE("Received pty with master fd %d and slave fd %d from "
1321 "parent", pty_info
->master
, pty_info
->slave
);
1324 ERROR("Failed to receive %d ttys from child: %s", conf
->tty
,
1327 TRACE("Received %d ttys from child", conf
->tty
);
1329 tty_info
->nbtty
= conf
->tty
;
1334 int resolve_clone_flags(struct lxc_handler
*handler
)
1336 handler
->clone_flags
= CLONE_NEWNS
;
1338 if (!handler
->conf
->inherit_ns
[LXC_NS_USER
]) {
1339 if (!lxc_list_empty(&handler
->conf
->id_map
))
1340 handler
->clone_flags
|= CLONE_NEWUSER
;
1342 INFO("Inheriting user namespace");
1345 if (!handler
->conf
->inherit_ns
[LXC_NS_NET
]) {
1346 if (!lxc_requests_empty_network(handler
))
1347 handler
->clone_flags
|= CLONE_NEWNET
;
1349 INFO("Inheriting net namespace");
1352 if (!handler
->conf
->inherit_ns
[LXC_NS_IPC
])
1353 handler
->clone_flags
|= CLONE_NEWIPC
;
1355 INFO("Inheriting ipc namespace");
1357 if (!handler
->conf
->inherit_ns
[LXC_NS_UTS
])
1358 handler
->clone_flags
|= CLONE_NEWUTS
;
1360 INFO("Inheriting uts namespace");
1362 if (!handler
->conf
->inherit_ns
[LXC_NS_PID
])
1363 handler
->clone_flags
|= CLONE_NEWPID
;
1365 INFO("Inheriting pid namespace");
1367 if (cgns_supported()) {
1368 if (!handler
->conf
->inherit_ns
[LXC_NS_CGROUP
])
1369 handler
->clone_flags
|= CLONE_NEWCGROUP
;
1371 INFO("Inheriting cgroup namespace");
1372 } else if (handler
->conf
->inherit_ns
[LXC_NS_CGROUP
]) {
1379 /* Note that this function is used with clone(CLONE_VM). Some glibc versions
1380 * used to reset the pid/tid to -1 when CLONE_VM was used without CLONE_THREAD.
1381 * But since the memory between parent and child is shared on CLONE_VM this
1382 * would invalidate the getpid() cache that glibc used to maintain and so
1383 * getpid() in the child would return the parent's pid. This is all fixed in
1384 * newer glibc versions where the getpid() cache is removed and the pid/tid is
1385 * not reset anymore.
1386 * However, if for whatever reason you - dear commiter - somehow need to get the
1387 * pid of the dummy intermediate process for do_share_ns() you need to call
1388 * lxc_raw_getpid(). The next lxc_raw_clone() call does not employ CLONE_VM and
1391 static inline int do_share_ns(void *arg
)
1394 struct lxc_handler
*handler
= arg
;
1396 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1397 if (handler
->nsfd
[i
] < 0)
1400 ret
= setns(handler
->nsfd
[i
], 0);
1404 DEBUG("Inherited %s namespace", ns_info
[i
].proc_name
);
1407 flags
= handler
->on_clone_flags
;
1408 flags
|= CLONE_PARENT
;
1409 handler
->pid
= lxc_raw_clone_cb(do_start
, handler
, flags
);
1410 if (handler
->pid
< 0)
1416 /* lxc_spawn() performs crucial setup tasks and clone()s the new process which
1417 * exec()s the requested container binary.
1418 * Note that lxc_spawn() runs in the parent namespaces. Any operations performed
1419 * right here should be double checked if they'd pose a security risk. (For
1420 * example, any {u}mount() operations performed here will be reflected on the
1423 static int lxc_spawn(struct lxc_handler
*handler
)
1427 bool wants_to_map_ids
;
1428 struct lxc_list
*id_map
;
1429 const char *name
= handler
->name
;
1430 const char *lxcpath
= handler
->lxcpath
;
1431 bool cgroups_connected
= false, share_ns
= false;
1432 struct lxc_conf
*conf
= handler
->conf
;
1434 id_map
= &conf
->id_map
;
1435 wants_to_map_ids
= !lxc_list_empty(id_map
);
1437 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1438 if (!conf
->inherit_ns
[i
])
1441 handler
->nsfd
[i
] = lxc_inherit_namespace(conf
->inherit_ns
[i
], lxcpath
, ns_info
[i
].proc_name
);
1442 if (handler
->nsfd
[i
] < 0)
1448 if (lxc_sync_init(handler
))
1451 ret
= socketpair(AF_UNIX
, SOCK_STREAM
| SOCK_CLOEXEC
, 0,
1452 handler
->data_sock
);
1454 lxc_sync_fini(handler
);
1458 ret
= resolve_clone_flags(handler
);
1460 lxc_sync_fini(handler
);
1464 if (handler
->clone_flags
& CLONE_NEWNET
) {
1465 if (!lxc_list_empty(&conf
->network
)) {
1467 /* Find gateway addresses from the link device, which is
1468 * no longer accessible inside the container. Do this
1469 * before creating network interfaces, since goto
1470 * out_delete_net does not work before lxc_clone.
1472 if (lxc_find_gateway_addresses(handler
)) {
1473 ERROR("Failed to find gateway addresses.");
1474 lxc_sync_fini(handler
);
1478 /* That should be done before the clone because we will
1479 * fill the netdev index and use them in the child.
1481 if (lxc_create_network_priv(handler
)) {
1482 ERROR("Failed to create the network.");
1483 lxc_sync_fini(handler
);
1489 if (!cgroup_init(handler
)) {
1490 ERROR("Failed initializing cgroup support.");
1491 goto out_delete_net
;
1494 cgroups_connected
= true;
1496 if (!cgroup_create(handler
)) {
1497 ERROR("Failed creating cgroups.");
1498 goto out_delete_net
;
1501 /* If the rootfs is not a blockdev, prevent the container from marking
1503 * If the container is unprivileged then skip rootfs pinning.
1505 if (!wants_to_map_ids
) {
1506 handler
->pinfd
= pin_rootfs(conf
->rootfs
.path
);
1507 if (handler
->pinfd
== -1)
1508 INFO("Failed to pin the rootfs for container \"%s\".", handler
->name
);
1511 /* Create a process in a new set of namespaces. */
1512 handler
->on_clone_flags
= handler
->clone_flags
;
1513 if (handler
->clone_flags
& CLONE_NEWUSER
) {
1514 /* If CLONE_NEWUSER and CLONE_NEWNET was requested, we need to
1515 * clone a new user namespace first and only later unshare our
1516 * network namespace to ensure that network devices ownership is
1519 handler
->on_clone_flags
&= ~CLONE_NEWNET
;
1521 /* The cgroup namespace gets unshare()ed not clone()ed. */
1522 handler
->on_clone_flags
&= ~CLONE_NEWCGROUP
;
1527 attacher_pid
= lxc_clone(do_share_ns
, handler
, CLONE_VFORK
| CLONE_VM
| CLONE_FILES
);
1528 if (attacher_pid
< 0) {
1529 SYSERROR(LXC_CLONE_ERROR
);
1530 goto out_delete_net
;
1533 ret
= wait_for_pid(attacher_pid
);
1535 SYSERROR("Intermediate process failed");
1536 goto out_delete_net
;
1539 handler
->pid
= lxc_raw_clone_cb(do_start
, handler
, handler
->on_clone_flags
);
1541 if (handler
->pid
< 0) {
1542 SYSERROR(LXC_CLONE_ERROR
);
1543 goto out_delete_net
;
1545 TRACE("Cloned child process %d", handler
->pid
);
1547 for (i
= 0; i
< LXC_NS_MAX
; i
++)
1548 if (handler
->on_clone_flags
& ns_info
[i
].clone_flag
)
1549 INFO("Cloned %s", ns_info
[i
].flag_name
);
1551 if (!preserve_ns(handler
->nsfd
, handler
->on_clone_flags
, handler
->pid
)) {
1552 ERROR("Failed to preserve cloned namespaces for lxc.hook.stop");
1553 goto out_delete_net
;
1556 lxc_sync_fini_child(handler
);
1558 /* Map the container uids. The container became an invalid userid the
1559 * moment it was cloned with CLONE_NEWUSER. This call doesn't change
1560 * anything immediately, but allows the container to setuid(0) (0 being
1561 * mapped to something else on the host.) later to become a valid uid
1564 if (wants_to_map_ids
) {
1565 if (!handler
->conf
->inherit_ns
[LXC_NS_USER
]) {
1566 ret
= lxc_map_ids(id_map
, handler
->pid
);
1568 ERROR("Failed to set up id mapping.");
1569 goto out_delete_net
;
1574 if (lxc_sync_wake_child(handler
, LXC_SYNC_STARTUP
))
1575 goto out_delete_net
;
1577 if (lxc_sync_wait_child(handler
, LXC_SYNC_CONFIGURE
))
1578 goto out_delete_net
;
1580 if (!cgroup_create_legacy(handler
)) {
1581 ERROR("Failed to setup legacy cgroups for container \"%s\".", name
);
1582 goto out_delete_net
;
1584 if (!cgroup_setup_limits(handler
, false)) {
1585 ERROR("Failed to setup cgroup limits for container \"%s\".", name
);
1586 goto out_delete_net
;
1589 if (!cgroup_enter(handler
))
1590 goto out_delete_net
;
1592 if (!cgroup_chown(handler
))
1593 goto out_delete_net
;
1595 /* Now we're ready to preserve the network namespace */
1596 ret
= lxc_preserve_ns(handler
->pid
, "net");
1598 ERROR("%s - Failed to preserve net namespace", strerror(errno
));
1599 goto out_delete_net
;
1601 handler
->nsfd
[LXC_NS_NET
] = ret
;
1602 DEBUG("Preserved net namespace via fd %d", ret
);
1604 /* Create the network configuration. */
1605 if (handler
->clone_flags
& CLONE_NEWNET
) {
1606 if (lxc_network_move_created_netdev_priv(handler
->lxcpath
,
1610 ERROR("Failed to create the configured network.");
1611 goto out_delete_net
;
1614 if (lxc_create_network_unpriv(handler
->lxcpath
, handler
->name
,
1617 ERROR("Failed to create the configured network.");
1618 goto out_delete_net
;
1622 if (lxc_network_send_veth_names_to_child(handler
) < 0) {
1623 ERROR("Failed to send veth names to child");
1624 goto out_delete_net
;
1627 if (!lxc_list_empty(&conf
->procs
)) {
1628 ret
= setup_proc_filesystem(&conf
->procs
, handler
->pid
);
1630 goto out_delete_net
;
1633 /* Tell the child to continue its initialization. We'll get
1634 * LXC_SYNC_CGROUP when it is ready for us to setup cgroups.
1636 if (lxc_sync_barrier_child(handler
, LXC_SYNC_POST_CONFIGURE
))
1637 goto out_delete_net
;
1639 if (!lxc_list_empty(&conf
->limits
) && setup_resource_limits(&conf
->limits
, handler
->pid
)) {
1640 ERROR("failed to setup resource limits for '%s'", name
);
1641 goto out_delete_net
;
1644 if (lxc_sync_barrier_child(handler
, LXC_SYNC_CGROUP_UNSHARE
))
1645 goto out_delete_net
;
1647 if (!cgroup_setup_limits(handler
, true)) {
1648 ERROR("Failed to setup the devices cgroup for container \"%s\".", name
);
1649 goto out_delete_net
;
1651 TRACE("Set up cgroup device limits");
1653 cgroup_disconnect();
1654 cgroups_connected
= false;
1656 if (handler
->clone_flags
& CLONE_NEWCGROUP
) {
1657 /* Now we're ready to preserve the cgroup namespace */
1658 ret
= lxc_preserve_ns(handler
->pid
, "cgroup");
1660 ERROR("%s - Failed to preserve cgroup namespace", strerror(errno
));
1661 goto out_delete_net
;
1663 handler
->nsfd
[LXC_NS_CGROUP
] = ret
;
1664 DEBUG("Preserved cgroup namespace via fd %d", ret
);
1667 snprintf(pidstr
, 20, "%d", handler
->pid
);
1668 if (setenv("LXC_PID", pidstr
, 1))
1669 SYSERROR("Failed to set environment variable: LXC_PID=%s.", pidstr
);
1671 /* Run any host-side start hooks */
1672 if (run_lxc_hooks(name
, "start-host", conf
, NULL
)) {
1673 ERROR("Failed to run lxc.hook.start-host for container \"%s\".", name
);
1677 /* Tell the child to complete its initialization and wait for it to exec
1678 * or return an error. (The child will never return
1679 * LXC_SYNC_READY_START+1. It will either close the sync pipe, causing
1680 * lxc_sync_barrier_child to return success, or return a different
1681 * value, causing us to error out).
1683 if (lxc_sync_barrier_child(handler
, LXC_SYNC_READY_START
))
1686 if (lxc_network_recv_name_and_ifindex_from_child(handler
) < 0) {
1687 ERROR("Failed to receive names and ifindices for network "
1688 "devices from child");
1689 goto out_delete_net
;
1692 /* Now all networks are created, network devices are moved into place,
1693 * and the correct names and ifindeces in the respective namespaces have
1694 * been recorded. The corresponding structs have now all been filled. So
1695 * log them for debugging purposes.
1697 lxc_log_configured_netdevs(conf
);
1699 /* Read tty fds allocated by child. */
1700 if (lxc_recv_ttys_from_child(handler
) < 0) {
1701 ERROR("Failed to receive tty info from child process.");
1702 goto out_delete_net
;
1705 if (handler
->ops
->post_start(handler
, handler
->data
))
1708 if (lxc_set_state(name
, handler
, RUNNING
)) {
1709 ERROR("Failed to set state for container \"%s\" to \"%s\".", name
,
1710 lxc_state2str(RUNNING
));
1714 lxc_sync_fini(handler
);
1719 if (cgroups_connected
)
1720 cgroup_disconnect();
1722 if (handler
->clone_flags
& CLONE_NEWNET
)
1723 lxc_delete_network(handler
);
1726 lxc_abort(name
, handler
);
1727 lxc_sync_fini(handler
);
1728 if (handler
->pinfd
>= 0) {
1729 close(handler
->pinfd
);
1730 handler
->pinfd
= -1;
1736 int __lxc_start(const char *name
, struct lxc_handler
*handler
,
1737 struct lxc_operations
* ops
, void *data
, const char *lxcpath
,
1742 struct lxc_conf
*conf
= handler
->conf
;
1744 if (lxc_init(name
, handler
) < 0) {
1745 ERROR("Failed to initialize container \"%s\".", name
);
1749 handler
->data
= data
;
1750 handler
->backgrounded
= backgrounded
;
1752 if (!attach_block_device(handler
->conf
)) {
1753 ERROR("Failed to attach block device.");
1754 goto out_fini_nonet
;
1757 if (geteuid() == 0 && !lxc_list_empty(&conf
->id_map
)) {
1758 /* If the backing store is a device, mount it here and now. */
1759 if (rootfs_is_blockdev(conf
)) {
1760 if (unshare(CLONE_NEWNS
) < 0) {
1761 ERROR("Failed to unshare CLONE_NEWNS.");
1762 goto out_fini_nonet
;
1764 INFO("Unshared CLONE_NEWNS.");
1766 remount_all_slave();
1767 if (do_rootfs_setup(conf
, name
, lxcpath
) < 0) {
1768 ERROR("Error setting up rootfs mount as root before spawn.");
1769 goto out_fini_nonet
;
1771 INFO("Set up container rootfs as host root.");
1775 err
= lxc_spawn(handler
);
1777 ERROR("Failed to spawn container \"%s\".", name
);
1778 goto out_detach_blockdev
;
1780 /* close parent side of data socket */
1781 close(handler
->data_sock
[0]);
1782 handler
->data_sock
[0] = -1;
1783 close(handler
->data_sock
[1]);
1784 handler
->data_sock
[1] = -1;
1786 handler
->conf
->reboot
= 0;
1788 err
= lxc_poll(name
, handler
);
1790 ERROR("LXC mainloop exited with error: %d.", err
);
1794 while (waitpid(handler
->pid
, &status
, 0) < 0 && errno
== EINTR
)
1797 /* If the child process exited but was not signaled, it didn't call
1798 * reboot. This should mean it was an lxc-execute which simply exited.
1799 * In any case, treat it as a 'halt'.
1801 if (WIFSIGNALED(status
)) {
1802 switch(WTERMSIG(status
)) {
1803 case SIGINT
: /* halt */
1804 DEBUG("Container \"%s\" is halting.", name
);
1806 case SIGHUP
: /* reboot */
1807 DEBUG("Container \"%s\" is rebooting.", name
);
1808 handler
->conf
->reboot
= 1;
1810 case SIGSYS
: /* seccomp */
1811 DEBUG("Container \"%s\" violated its seccomp policy.", name
);
1814 DEBUG("Unknown exit status for container \"%s\" init %d.", name
, WTERMSIG(status
));
1819 err
= lxc_restore_phys_nics_to_netns(handler
);
1821 ERROR("Failed to move physical network devices back to parent "
1822 "network namespace");
1824 if (handler
->pinfd
>= 0) {
1825 close(handler
->pinfd
);
1826 handler
->pinfd
= -1;
1829 lxc_monitor_send_exit_code(name
, status
, handler
->lxcpath
);
1830 lxc_error_set_and_log(handler
->pid
, status
);
1833 lxc_delete_network(handler
);
1835 out_detach_blockdev
:
1836 detach_block_device(handler
->conf
);
1839 lxc_fini(name
, handler
);
1843 lxc_abort(name
, handler
);
1851 static int start(struct lxc_handler
*handler
, void* data
)
1853 struct start_args
*arg
= data
;
1855 NOTICE("Exec'ing \"%s\".", arg
->argv
[0]);
1857 execvp(arg
->argv
[0], arg
->argv
);
1858 SYSERROR("Failed to exec \"%s\".", arg
->argv
[0]);
1862 static int post_start(struct lxc_handler
*handler
, void* data
)
1864 struct start_args
*arg
= data
;
1866 NOTICE("Started \"%s\" with pid \"%d\".", arg
->argv
[0], handler
->pid
);
1870 static struct lxc_operations start_ops
= {
1872 .post_start
= post_start
1875 int lxc_start(const char *name
, char *const argv
[], struct lxc_handler
*handler
,
1876 const char *lxcpath
, bool backgrounded
)
1878 struct start_args start_arg
= {
1882 return __lxc_start(name
, handler
, &start_ops
, &start_arg
, lxcpath
, backgrounded
);
1885 static void lxc_destroy_container_on_signal(struct lxc_handler
*handler
,
1888 char destroy
[MAXPATHLEN
];
1891 struct lxc_container
*c
;
1892 if (handler
->conf
->rootfs
.path
&& handler
->conf
->rootfs
.mount
) {
1893 bret
= do_destroy_container(handler
);
1895 ERROR("Error destroying rootfs for container \"%s\".", name
);
1899 INFO("Destroyed rootfs for container \"%s\".", name
);
1901 ret
= snprintf(destroy
, MAXPATHLEN
, "%s/%s", handler
->lxcpath
, name
);
1902 if (ret
< 0 || ret
>= MAXPATHLEN
) {
1903 ERROR("Error destroying directory for container \"%s\".", name
);
1907 c
= lxc_container_new(name
, handler
->lxcpath
);
1909 if (container_disk_lock(c
)) {
1910 INFO("Could not update lxc_snapshots file.");
1911 lxc_container_put(c
);
1913 mod_all_rdeps(c
, false);
1914 container_disk_unlock(c
);
1915 lxc_container_put(c
);
1919 if (!handler
->am_root
)
1920 ret
= userns_exec_full(handler
->conf
, lxc_rmdir_onedev_wrapper
,
1921 destroy
, "lxc_rmdir_onedev_wrapper");
1923 ret
= lxc_rmdir_onedev(destroy
, NULL
);
1926 ERROR("Error destroying directory for container \"%s\".", name
);
1929 INFO("Destroyed directory for container \"%s\".", name
);
1932 static int lxc_rmdir_onedev_wrapper(void *data
)
1934 char *arg
= (char *) data
;
1935 return lxc_rmdir_onedev(arg
, NULL
);
1938 static bool do_destroy_container(struct lxc_handler
*handler
) {
1941 if (!handler
->am_root
) {
1942 ret
= userns_exec_full(handler
->conf
, storage_destroy_wrapper
,
1943 handler
->conf
, "storage_destroy_wrapper");
1950 return storage_destroy(handler
->conf
);