]>
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"
78 #include "syscall_wrappers.h"
83 #include <sys/capability.h>
87 #include "include/strlcpy.h"
90 lxc_log_define(start
, lxc
);
92 extern void mod_all_rdeps(struct lxc_container
*c
, bool inc
);
93 static bool do_destroy_container(struct lxc_handler
*handler
);
94 static int lxc_rmdir_onedev_wrapper(void *data
);
95 static void lxc_destroy_container_on_signal(struct lxc_handler
*handler
,
98 static void print_top_failing_dir(const char *path
)
102 char *copy
, *e
, *p
, saved
;
105 copy
= alloca(len
+ 1);
106 (void)strlcpy(copy
, path
, len
+ 1);
112 while (p
< e
&& *p
== '/')
115 while (p
< e
&& *p
!= '/')
121 ret
= access(copy
, X_OK
);
123 SYSERROR("Could not access %s. Please grant it x "
124 "access, or add an ACL for the container "
132 static void lxc_put_nsfds(struct lxc_handler
*handler
)
136 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
137 if (handler
->nsfd
[i
] < 0)
140 close(handler
->nsfd
[i
]);
141 handler
->nsfd
[i
] = -EBADF
;
145 static int lxc_try_preserve_ns(const int pid
, const char *ns
)
149 fd
= lxc_preserve_ns(pid
, ns
);
151 if (errno
!= ENOENT
) {
152 SYSERROR("Failed to preserve %s namespace", ns
);
156 SYSWARN("Kernel does not support preserving %s namespaces", ns
);
163 /* lxc_try_preserve_namespaces: open /proc/@pid/ns/@ns for each namespace
164 * specified in ns_clone_flags.
165 * Return true on success, false on failure.
167 static bool lxc_try_preserve_namespaces(struct lxc_handler
*handler
,
168 int ns_clone_flags
, pid_t pid
)
172 for (i
= 0; i
< LXC_NS_MAX
; i
++)
173 handler
->nsfd
[i
] = -EBADF
;
175 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
178 if ((ns_clone_flags
& ns_info
[i
].clone_flag
) == 0)
181 fd
= lxc_try_preserve_ns(pid
, ns_info
[i
].proc_name
);
183 /* Do not fail to start container on kernels that do
184 * not support interacting with namespaces through
187 if (fd
== -EOPNOTSUPP
)
190 lxc_put_nsfds(handler
);
194 handler
->nsfd
[i
] = fd
;
195 DEBUG("Preserved %s namespace via fd %d", ns_info
[i
].proc_name
,
202 static inline bool match_stdfds(int fd
)
204 return (fd
== STDIN_FILENO
|| fd
== STDOUT_FILENO
|| fd
== STDERR_FILENO
);
207 int lxc_check_inherited(struct lxc_conf
*conf
, bool closeall
,
208 int *fds_to_ignore
, size_t len_fds
)
213 struct dirent
*direntp
;
215 if (conf
&& conf
->close_all_fds
)
219 dir
= opendir("/proc/self/fd");
221 SYSWARN("Failed to open directory");
227 while ((direntp
= readdir(dir
))) {
229 struct lxc_list
*cur
;
230 bool matched
= false;
232 if (strcmp(direntp
->d_name
, ".") == 0)
235 if (strcmp(direntp
->d_name
, "..") == 0)
238 ret
= lxc_safe_int(direntp
->d_name
, &fd
);
240 INFO("Could not parse file descriptor for \"%s\"", direntp
->d_name
);
244 for (i
= 0; i
< len_fds
; i
++)
245 if (fds_to_ignore
[i
] == fd
)
248 if (fd
== fddir
|| fd
== lxc_log_fd
||
249 (i
< len_fds
&& fd
== fds_to_ignore
[i
]))
252 /* Keep state clients that wait on reboots. */
254 lxc_list_for_each(cur
, &conf
->state_clients
) {
255 struct lxc_state_client
*client
= cur
->elem
;
257 if (client
->clientfd
!= fd
)
268 if (current_config
&& fd
== current_config
->logfd
)
271 if (match_stdfds(fd
))
277 INFO("Closed inherited fd %d", fd
);
280 WARN("Inherited fd %d", fd
);
283 /* Only enable syslog at this point to avoid the above logging function
284 * to open a new fd and make the check_inherited function enter an
287 lxc_log_enable_syslog();
289 closedir(dir
); /* cannot fail */
293 static int setup_signal_fd(sigset_t
*oldmask
)
297 const int signals
[] = {SIGBUS
, SIGILL
, SIGSEGV
, SIGWINCH
};
299 /* Block everything except serious error signals. */
300 ret
= sigfillset(&mask
);
304 for (int sig
= 0; sig
< (sizeof(signals
) / sizeof(signals
[0])); sig
++) {
305 ret
= sigdelset(&mask
, signals
[sig
]);
310 ret
= pthread_sigmask(SIG_BLOCK
, &mask
, oldmask
);
312 SYSERROR("Failed to set signal mask");
316 ret
= signalfd(-1, &mask
, SFD_CLOEXEC
);
318 SYSERROR("Failed to create signal file descriptor");
322 TRACE("Created signal file descriptor %d", ret
);
327 static int signal_handler(int fd
, uint32_t events
, void *data
,
328 struct lxc_epoll_descr
*descr
)
332 struct signalfd_siginfo siginfo
;
333 struct lxc_handler
*hdlr
= data
;
335 ret
= lxc_read_nointr(fd
, &siginfo
, sizeof(siginfo
));
337 ERROR("Failed to read signal info from signal file descriptor %d", fd
);
338 return LXC_MAINLOOP_ERROR
;
341 if (ret
!= sizeof(siginfo
)) {
342 ERROR("Unexpected size for struct signalfd_siginfo");
346 /* Check whether init is running. */
348 ret
= waitid(P_PID
, hdlr
->pid
, &info
, WEXITED
| WNOWAIT
| WNOHANG
);
349 if (ret
== 0 && info
.si_pid
== hdlr
->pid
)
350 hdlr
->init_died
= true;
352 /* Try to figure out a reasonable exit status to report. */
353 if (hdlr
->init_died
) {
354 switch (info
.si_code
) {
356 hdlr
->exit_status
= info
.si_status
<< 8;
361 hdlr
->exit_status
= info
.si_status
<< 8 | 0x7f;
364 /* Huh? The waitid() told us it's dead *and* continued? */
365 WARN("Init %d dead and continued?", hdlr
->pid
);
366 hdlr
->exit_status
= 1;
369 ERROR("Unknown si_code: %d", info
.si_code
);
370 hdlr
->exit_status
= 1;
374 if (siginfo
.ssi_signo
== SIGHUP
) {
375 kill(hdlr
->pid
, SIGTERM
);
376 INFO("Killing %d since terminal hung up", hdlr
->pid
);
377 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
378 : LXC_MAINLOOP_CONTINUE
;
381 if (siginfo
.ssi_signo
!= SIGCHLD
) {
382 kill(hdlr
->pid
, siginfo
.ssi_signo
);
383 INFO("Forwarded signal %d to pid %d", siginfo
.ssi_signo
, hdlr
->pid
);
384 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
385 : 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
395 : LXC_MAINLOOP_CONTINUE
;
398 if (siginfo
.ssi_code
== CLD_STOPPED
) {
399 INFO("Container init process was stopped");
400 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
401 : LXC_MAINLOOP_CONTINUE
;
404 if (siginfo
.ssi_code
== CLD_CONTINUED
) {
405 INFO("Container init process was continued");
406 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
407 : LXC_MAINLOOP_CONTINUE
;
410 DEBUG("Container init process %d exited", hdlr
->pid
);
412 return LXC_MAINLOOP_CLOSE
;
415 int lxc_serve_state_clients(const char *name
, struct lxc_handler
*handler
,
420 struct lxc_list
*cur
, *next
;
421 struct lxc_state_client
*client
;
422 struct lxc_msg msg
= {.type
= lxc_msg_state
, .value
= state
};
425 handler
->state
= RUNNING
;
427 handler
->state
= state
;
429 TRACE("Set container state to %s", lxc_state2str(state
));
431 if (lxc_list_empty(&handler
->conf
->state_clients
)) {
432 TRACE("No state clients registered");
436 retlen
= strlcpy(msg
.name
, name
, sizeof(msg
.name
));
437 if (retlen
>= sizeof(msg
.name
))
440 lxc_list_for_each_safe(cur
, &handler
->conf
->state_clients
, next
) {
443 if (client
->states
[state
] == 0) {
444 TRACE("State %s not registered for state client %d",
445 lxc_state2str(state
), client
->clientfd
);
449 TRACE("Sending state %s to state client %d",
450 lxc_state2str(state
), client
->clientfd
);
452 ret
= lxc_send_nointr(client
->clientfd
, &msg
, sizeof(msg
), MSG_NOSIGNAL
);
454 SYSERROR("Failed to send message to client");
456 /* kick client from list */
458 close(client
->clientfd
);
466 static int lxc_serve_state_socket_pair(const char *name
,
467 struct lxc_handler
*handler
,
472 if (!handler
->daemonize
||
473 handler
->state_socket_pair
[1] < 0 ||
477 /* Close read end of the socket pair. */
478 close(handler
->state_socket_pair
[0]);
479 handler
->state_socket_pair
[0] = -1;
482 ret
= lxc_abstract_unix_send_credential(handler
->state_socket_pair
[1],
483 &(int){state
}, sizeof(int));
485 SYSERROR("Failed to send state to %d", handler
->state_socket_pair
[1]);
493 if (ret
!= sizeof(int)) {
494 ERROR("Message too long : %d", handler
->state_socket_pair
[1]);
498 TRACE("Sent container state \"%s\" to %d", lxc_state2str(state
),
499 handler
->state_socket_pair
[1]);
501 /* Close write end of the socket pair. */
502 close(handler
->state_socket_pair
[1]);
503 handler
->state_socket_pair
[1] = -1;
508 int lxc_set_state(const char *name
, struct lxc_handler
*handler
,
513 ret
= lxc_serve_state_socket_pair(name
, handler
, state
);
515 ERROR("Failed to synchronize via anonymous pair of unix sockets");
519 ret
= lxc_serve_state_clients(name
, handler
, state
);
523 /* This function will try to connect to the legacy lxc-monitord state
524 * server and only exists for backwards compatibility.
526 lxc_monitor_send_state(name
, state
, handler
->lxcpath
);
531 int lxc_poll(const char *name
, struct lxc_handler
*handler
)
534 bool has_console
= true;
535 struct lxc_epoll_descr descr
, descr_console
;
537 if (handler
->conf
->console
.path
&&
538 strcmp(handler
->conf
->console
.path
, "none") == 0)
541 ret
= lxc_mainloop_open(&descr
);
543 ERROR("Failed to create mainloop");
548 ret
= lxc_mainloop_open(&descr_console
);
550 ERROR("Failed to create console mainloop");
555 ret
= lxc_mainloop_add_handler(&descr
, handler
->sigfd
, signal_handler
, handler
);
557 ERROR("Failed to add signal handler for %d to mainloop", handler
->sigfd
);
558 goto out_mainloop_console
;
562 struct lxc_terminal
*console
= &handler
->conf
->console
;
564 ret
= lxc_terminal_mainloop_add(&descr
, console
);
566 ERROR("Failed to add console handlers to mainloop");
567 goto out_mainloop_console
;
570 ret
= lxc_terminal_mainloop_add(&descr_console
, console
);
572 ERROR("Failed to add console handlers to console mainloop");
573 goto out_mainloop_console
;
576 handler
->conf
->console
.descr
= &descr
;
579 ret
= lxc_cmd_mainloop_add(name
, &descr
, handler
);
581 ERROR("Failed to add command handler to mainloop");
582 goto out_mainloop_console
;
585 TRACE("Mainloop is ready");
587 ret
= lxc_mainloop(&descr
, -1);
590 if (ret
< 0 || !handler
->init_died
)
591 goto out_mainloop_console
;
594 ret
= lxc_mainloop(&descr_console
, 0);
596 out_mainloop_console
:
598 lxc_mainloop_close(&descr_console
);
599 TRACE("Closed console mainloop");
603 lxc_mainloop_close(&descr
);
604 TRACE("Closed mainloop");
607 close(handler
->sigfd
);
608 TRACE("Closed signal file descriptor %d", handler
->sigfd
);
609 handler
->sigfd
= -EBADF
;
614 void lxc_zero_handler(struct lxc_handler
*handler
)
618 memset(handler
, 0, sizeof(struct lxc_handler
));
624 for (i
= 0; i
< LXC_NS_MAX
; i
++)
625 handler
->nsfd
[i
] = -1;
627 handler
->data_sock
[0] = -1;
628 handler
->data_sock
[1] = -1;
630 handler
->state_socket_pair
[0] = -1;
631 handler
->state_socket_pair
[1] = -1;
633 handler
->sync_sock
[0] = -1;
634 handler
->sync_sock
[1] = -1;
637 void lxc_free_handler(struct lxc_handler
*handler
)
639 if (handler
->pinfd
>= 0)
640 close(handler
->pinfd
);
642 if (handler
->sigfd
>= 0)
643 close(handler
->sigfd
);
645 lxc_put_nsfds(handler
);
647 if (handler
->conf
&& handler
->conf
->reboot
== REBOOT_NONE
)
648 if (handler
->conf
->maincmd_fd
>= 0)
649 lxc_abstract_unix_close(handler
->conf
->maincmd_fd
);
651 if (handler
->state_socket_pair
[0] >= 0)
652 close(handler
->state_socket_pair
[0]);
654 if (handler
->state_socket_pair
[1] >= 0)
655 close(handler
->state_socket_pair
[1]);
657 if (handler
->cgroup_ops
)
658 cgroup_exit(handler
->cgroup_ops
);
660 handler
->conf
= NULL
;
665 struct lxc_handler
*lxc_init_handler(const char *name
, struct lxc_conf
*conf
,
666 const char *lxcpath
, bool daemonize
)
669 struct lxc_handler
*handler
;
671 handler
= malloc(sizeof(*handler
));
675 memset(handler
, 0, sizeof(*handler
));
677 /* Note that am_guest_unpriv() checks the effective uid. We
678 * probably don't care if we are real root only if we are running
679 * as root so this should be fine.
681 handler
->am_root
= !am_guest_unpriv();
682 handler
->data_sock
[0] = handler
->data_sock
[1] = -1;
683 handler
->conf
= conf
;
684 handler
->lxcpath
= lxcpath
;
686 handler
->sigfd
= -EBADF
;
687 handler
->init_died
= false;
688 handler
->state_socket_pair
[0] = handler
->state_socket_pair
[1] = -1;
689 if (handler
->conf
->reboot
== REBOOT_NONE
)
690 lxc_list_init(&handler
->conf
->state_clients
);
692 for (i
= 0; i
< LXC_NS_MAX
; i
++)
693 handler
->nsfd
[i
] = -1;
695 handler
->name
= name
;
697 if (daemonize
&& handler
->conf
->reboot
== REBOOT_NONE
) {
698 /* Create socketpair() to synchronize on daemonized startup.
699 * When the container reboots we don't need to synchronize
700 * again currently so don't open another socketpair().
702 ret
= socketpair(AF_UNIX
, SOCK_STREAM
| SOCK_CLOEXEC
, 0,
703 handler
->state_socket_pair
);
705 ERROR("Failed to create anonymous pair of unix sockets");
709 TRACE("Created anonymous pair {%d,%d} of unix sockets",
710 handler
->state_socket_pair
[0],
711 handler
->state_socket_pair
[1]);
714 if (handler
->conf
->reboot
== REBOOT_NONE
) {
715 handler
->conf
->maincmd_fd
= lxc_cmd_init(name
, lxcpath
, "command");
716 if (handler
->conf
->maincmd_fd
< 0) {
717 ERROR("Failed to set up command socket");
722 TRACE("Unix domain socket %d for command server is ready",
723 handler
->conf
->maincmd_fd
);
728 lxc_free_handler(handler
);
733 int lxc_init(const char *name
, struct lxc_handler
*handler
)
736 const char *loglevel
;
737 struct lxc_conf
*conf
= handler
->conf
;
739 handler
->monitor_pid
= lxc_raw_getpid();
742 TRACE("Initialized LSM");
744 ret
= lxc_read_seccomp_config(conf
);
746 ERROR("Failed loading seccomp policy");
747 goto out_close_maincmd_fd
;
749 TRACE("Read seccomp policy");
751 /* Begin by setting the state to STARTING. */
752 ret
= lxc_set_state(name
, handler
, STARTING
);
754 ERROR("Failed to set state to \"%s\"", lxc_state2str(STARTING
));
755 goto out_close_maincmd_fd
;
757 TRACE("Set container state to \"STARTING\"");
759 /* Start of environment variable setup for hooks. */
760 ret
= setenv("LXC_NAME", name
, 1);
762 SYSERROR("Failed to set environment variable: LXC_NAME=%s", name
);
765 ret
= setenv("LXC_CONFIG_FILE", conf
->rcfile
, 1);
767 SYSERROR("Failed to set environment variable: "
768 "LXC_CONFIG_FILE=%s", conf
->rcfile
);
771 if (conf
->rootfs
.mount
) {
772 ret
= setenv("LXC_ROOTFS_MOUNT", conf
->rootfs
.mount
, 1);
774 SYSERROR("Failed to set environment variable: "
775 "LXC_ROOTFS_MOUNT=%s", conf
->rootfs
.mount
);
778 if (conf
->rootfs
.path
) {
779 ret
= setenv("LXC_ROOTFS_PATH", conf
->rootfs
.path
, 1);
781 SYSERROR("Failed to set environment variable: "
782 "LXC_ROOTFS_PATH=%s", conf
->rootfs
.path
);
785 if (conf
->console
.path
) {
786 ret
= setenv("LXC_CONSOLE", conf
->console
.path
, 1);
788 SYSERROR("Failed to set environment variable: "
789 "LXC_CONSOLE=%s", conf
->console
.path
);
792 if (conf
->console
.log_path
) {
793 ret
= setenv("LXC_CONSOLE_LOGPATH", conf
->console
.log_path
, 1);
795 SYSERROR("Failed to set environment variable: "
796 "LXC_CONSOLE_LOGPATH=%s", conf
->console
.log_path
);
799 if (cgns_supported()) {
800 ret
= setenv("LXC_CGNS_AWARE", "1", 1);
802 SYSERROR("Failed to set environment variable "
806 loglevel
= lxc_log_priority_to_string(lxc_log_get_level());
807 ret
= setenv("LXC_LOG_LEVEL", loglevel
, 1);
809 SYSERROR("Set environment variable LXC_LOG_LEVEL=%s",
812 if (conf
->hooks_version
== 0)
813 ret
= setenv("LXC_HOOK_VERSION", "0", 1);
815 ret
= setenv("LXC_HOOK_VERSION", "1", 1);
817 SYSERROR("Failed to set environment variable LXC_HOOK_VERSION=%u", conf
->hooks_version
);
818 /* End of environment variable setup for hooks. */
820 TRACE("Set environment variables");
822 ret
= run_lxc_hooks(name
, "pre-start", conf
, NULL
);
824 ERROR("Failed to run lxc.hook.pre-start for container \"%s\"", name
);
827 TRACE("Ran pre-start hooks");
829 /* The signal fd has to be created before forking otherwise if the child
830 * process exits before we setup the signal fd, the event will be lost
831 * and the command will be stuck.
833 handler
->sigfd
= setup_signal_fd(&handler
->oldmask
);
834 if (handler
->sigfd
< 0) {
835 ERROR("Failed to setup SIGCHLD fd handler.");
838 TRACE("Set up signal fd");
840 /* Do this after setting up signals since it might unblock SIGWINCH. */
841 ret
= lxc_terminal_setup(conf
);
843 ERROR("Failed to create console");
844 goto out_restore_sigmask
;
846 TRACE("Created console");
848 ret
= lxc_terminal_map_ids(conf
, &conf
->console
);
850 ERROR("Failed to chown console");
851 goto out_delete_terminal
;
853 TRACE("Chowned console");
855 handler
->cgroup_ops
= cgroup_init(handler
->conf
);
856 if (!handler
->cgroup_ops
) {
857 ERROR("Failed to initialize cgroup driver");
858 goto out_delete_terminal
;
860 TRACE("Initialized cgroup driver");
862 ret
= lsm_process_prepare(conf
, handler
->lxcpath
);
864 ERROR("Failed to initialize LSM");
865 goto out_destroy_cgroups
;
867 TRACE("Initialized LSM");
869 INFO("Container \"%s\" is initialized", name
);
873 handler
->cgroup_ops
->payload_destroy(handler
->cgroup_ops
, handler
);
874 handler
->cgroup_ops
->monitor_destroy(handler
->cgroup_ops
, handler
);
877 lxc_terminal_delete(&handler
->conf
->console
);
880 (void)pthread_sigmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
883 lxc_delete_tty(&conf
->ttys
);
886 (void)lxc_set_state(name
, handler
, ABORTING
);
888 out_close_maincmd_fd
:
889 lxc_abstract_unix_close(conf
->maincmd_fd
);
890 conf
->maincmd_fd
= -1;
894 void lxc_fini(const char *name
, struct lxc_handler
*handler
)
898 struct lxc_list
*cur
, *next
;
899 char *namespaces
[LXC_NS_MAX
+ 1];
900 size_t namespace_count
= 0;
901 struct cgroup_ops
*cgroup_ops
= handler
->cgroup_ops
;
903 /* The STOPPING state is there for future cleanup code which can take
906 lxc_set_state(name
, handler
, STOPPING
);
908 self
= lxc_raw_getpid();
909 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
910 if (handler
->nsfd
[i
] < 0)
913 if (handler
->conf
->hooks_version
== 0)
914 ret
= asprintf(&namespaces
[namespace_count
],
915 "%s:/proc/%d/fd/%d", ns_info
[i
].proc_name
,
916 self
, handler
->nsfd
[i
]);
918 ret
= asprintf(&namespaces
[namespace_count
],
919 "/proc/%d/fd/%d", self
, handler
->nsfd
[i
]);
921 SYSERROR("Failed to allocate memory");
925 if (handler
->conf
->hooks_version
== 0) {
930 ret
= setenv(ns_info
[i
].env_name
, namespaces
[namespace_count
], 1);
932 SYSERROR("Failed to set environment variable %s=%s",
933 ns_info
[i
].env_name
, namespaces
[namespace_count
]);
935 TRACE("Set environment variable %s=%s",
936 ns_info
[i
].env_name
, namespaces
[namespace_count
]);
940 namespaces
[namespace_count
] = NULL
;
942 if (handler
->conf
->reboot
> REBOOT_NONE
) {
943 ret
= setenv("LXC_TARGET", "reboot", 1);
945 SYSERROR("Failed to set environment variable: "
946 "LXC_TARGET=reboot");
949 if (handler
->conf
->reboot
== REBOOT_NONE
) {
950 ret
= setenv("LXC_TARGET", "stop", 1);
952 SYSERROR("Failed to set environment variable: "
956 if (handler
->conf
->hooks_version
== 0)
957 ret
= run_lxc_hooks(name
, "stop", handler
->conf
, namespaces
);
959 ret
= run_lxc_hooks(name
, "stop", handler
->conf
, NULL
);
961 ERROR("Failed to run \"lxc.hook.stop\" hook");
963 while (namespace_count
--)
964 free(namespaces
[namespace_count
]);
966 lsm_process_cleanup(handler
->conf
, handler
->lxcpath
);
968 cgroup_ops
->payload_destroy(cgroup_ops
, handler
);
969 cgroup_ops
->monitor_destroy(cgroup_ops
, handler
);
971 if (handler
->conf
->reboot
== REBOOT_NONE
) {
972 /* For all new state clients simply close the command socket.
973 * This will inform all state clients that the container is
974 * STOPPED and also prevents a race between a open()/close() on
975 * the command socket causing a new process to get ECONNREFUSED
976 * because we haven't yet closed the command socket.
978 lxc_abstract_unix_close(handler
->conf
->maincmd_fd
);
979 handler
->conf
->maincmd_fd
= -1;
980 TRACE("Closed command socket");
982 /* This function will try to connect to the legacy lxc-monitord
983 * state server and only exists for backwards compatibility.
985 lxc_monitor_send_state(name
, STOPPED
, handler
->lxcpath
);
987 /* The command socket is closed so no one can acces the command
988 * socket anymore so there's no need to lock it.
990 handler
->state
= STOPPED
;
991 TRACE("Set container state to \"STOPPED\"");
993 lxc_set_state(name
, handler
, STOPPED
);
996 ret
= run_lxc_hooks(name
, "post-stop", handler
->conf
, NULL
);
998 ERROR("Failed to run lxc.hook.post-stop for container \"%s\"", name
);
999 if (handler
->conf
->reboot
> REBOOT_NONE
) {
1000 WARN("Container will be stopped instead of rebooted");
1001 handler
->conf
->reboot
= REBOOT_NONE
;
1003 ret
= setenv("LXC_TARGET", "stop", 1);
1005 WARN("Failed to set environment variable: "
1010 /* Reset mask set by setup_signal_fd. */
1011 ret
= pthread_sigmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
1013 SYSWARN("Failed to restore signal mask");
1015 lxc_terminal_delete(&handler
->conf
->console
);
1016 lxc_delete_tty(&handler
->conf
->ttys
);
1018 /* The command socket is now closed, no more state clients can register
1019 * themselves from now on. So free the list of state clients.
1021 lxc_list_for_each_safe(cur
, &handler
->conf
->state_clients
, next
) {
1022 struct lxc_state_client
*client
= cur
->elem
;
1024 /* Keep state clients that want to be notified about reboots. */
1025 if ((handler
->conf
->reboot
> REBOOT_NONE
) &&
1026 (client
->states
[RUNNING
] == 2))
1029 /* close state client socket */
1031 close(client
->clientfd
);
1036 if (handler
->conf
->ephemeral
== 1 && handler
->conf
->reboot
!= REBOOT_REQ
)
1037 lxc_destroy_container_on_signal(handler
, name
);
1039 lxc_free_handler(handler
);
1042 void lxc_abort(const char *name
, struct lxc_handler
*handler
)
1046 lxc_set_state(name
, handler
, ABORTING
);
1048 if (handler
->pid
> 0) {
1049 ret
= kill(handler
->pid
, SIGKILL
);
1051 SYSERROR("Failed to send SIGKILL to %d", handler
->pid
);
1054 while ((ret
= waitpid(-1, &status
, 0)) > 0) {
1059 static int do_start(void *data
)
1062 char path
[PATH_MAX
];
1065 struct lxc_list
*iterator
;
1068 int devnull_fd
= -1;
1069 struct lxc_handler
*handler
= data
;
1071 lxc_sync_fini_parent(handler
);
1073 /* This prctl must be before the synchro, so if the parent dies before
1074 * we set the parent death signal, we will detect its death with the
1075 * synchro right after, otherwise we have a window where the parent can
1076 * exit before we set the pdeath signal leading to a unsupervized
1079 ret
= lxc_set_death_signal(SIGKILL
, 0);
1081 SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL");
1082 goto out_warn_father
;
1085 ret
= lxc_ambient_caps_up();
1087 ERROR("Failed to raise ambient capabilities");
1088 goto out_warn_father
;
1091 ret
= pthread_sigmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
1093 SYSERROR("Failed to set signal mask");
1094 goto out_warn_father
;
1097 /* Don't leak the pinfd to the container. */
1098 if (handler
->pinfd
>= 0)
1099 close(handler
->pinfd
);
1101 ret
= lxc_sync_wait_parent(handler
, LXC_SYNC_STARTUP
);
1103 goto out_warn_father
;
1105 /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See
1106 * https://github.com/lxc/lxd/issues/1978.
1108 if ((handler
->ns_clone_flags
& (CLONE_NEWNET
| CLONE_NEWUSER
)) ==
1109 (CLONE_NEWNET
| CLONE_NEWUSER
)) {
1110 ret
= unshare(CLONE_NEWNET
);
1112 SYSERROR("Failed to unshare CLONE_NEWNET");
1113 goto out_warn_father
;
1115 INFO("Unshared CLONE_NEWNET");
1118 /* Tell the parent task it can begin to configure the container and wait
1121 ret
= lxc_sync_barrier_parent(handler
, LXC_SYNC_CONFIGURE
);
1125 ret
= lxc_network_recv_veth_names_from_parent(handler
);
1127 ERROR("Failed to receive veth names from parent");
1128 goto out_warn_father
;
1131 /* If we are in a new user namespace, become root there to have
1132 * privilege over our namespace.
1134 if (!lxc_list_empty(&handler
->conf
->id_map
)) {
1135 if (!handler
->conf
->root_nsuid_map
)
1136 nsuid
= handler
->conf
->init_uid
;
1138 if (!handler
->conf
->root_nsgid_map
)
1139 nsgid
= handler
->conf
->init_gid
;
1141 if (!lxc_switch_uid_gid(nsuid
, nsgid
))
1142 goto out_warn_father
;
1144 /* Drop groups only after we switched to a valid gid in the new
1147 if (!lxc_setgroups(0, NULL
) &&
1148 (handler
->am_root
|| errno
!= EPERM
))
1149 goto out_warn_father
;
1151 ret
= prctl(PR_SET_DUMPABLE
, prctl_arg(1), prctl_arg(0),
1152 prctl_arg(0), prctl_arg(0));
1154 goto out_warn_father
;
1156 /* set{g,u}id() clears deathsignal */
1157 ret
= lxc_set_death_signal(SIGKILL
, 0);
1159 SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL");
1160 goto out_warn_father
;
1164 ret
= access(handler
->lxcpath
, X_OK
);
1166 print_top_failing_dir(handler
->lxcpath
);
1167 goto out_warn_father
;
1170 ret
= snprintf(path
, sizeof(path
), "%s/dev/null",
1171 handler
->conf
->rootfs
.mount
);
1172 if (ret
< 0 || ret
>= sizeof(path
))
1173 goto out_warn_father
;
1175 /* In order to checkpoint restore, we need to have everything in the
1176 * same mount namespace. However, some containers may not have a
1177 * reasonable /dev (in particular, they may not have /dev/null), so we
1178 * can't set init's std fds to /dev/null by opening it from inside the
1181 * If that's the case, fall back to using the host's /dev/null. This
1182 * means that migration won't work, but at least we won't spew output
1183 * where it isn't wanted.
1185 if (handler
->daemonize
&& !handler
->conf
->autodev
) {
1186 ret
= access(path
, F_OK
);
1188 devnull_fd
= open_devnull();
1191 goto out_warn_father
;
1192 WARN("Using /dev/null from the host for container "
1193 "init's standard file descriptors. Migration will "
1198 /* Ask father to setup cgroups and wait for him to finish. */
1199 ret
= lxc_sync_barrier_parent(handler
, LXC_SYNC_CGROUP
);
1203 /* Unshare cgroup namespace after we have setup our cgroups. If we do it
1204 * earlier we end up with a wrong view of /proc/self/cgroup. For
1205 * example, assume we unshare(CLONE_NEWCGROUP) first, and then create
1206 * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then
1207 * /proc/self/cgroup would show us:
1211 * whereas it should actually show
1215 if (handler
->ns_clone_flags
& CLONE_NEWCGROUP
) {
1216 ret
= unshare(CLONE_NEWCGROUP
);
1218 INFO("Failed to unshare CLONE_NEWCGROUP");
1219 goto out_warn_father
;
1221 INFO("Unshared CLONE_NEWCGROUP");
1224 /* Add the requested environment variables to the current environment to
1225 * allow them to be used by the various hooks, such as the start hook
1228 lxc_list_for_each(iterator
, &handler
->conf
->environment
) {
1229 ret
= putenv((char *)iterator
->elem
);
1231 SYSERROR("Failed to set environment variable: %s",
1232 (char *)iterator
->elem
);
1233 goto out_warn_father
;
1237 /* Setup the container, ip, names, utsname, ... */
1238 ret
= lxc_setup(handler
);
1239 close(handler
->data_sock
[1]);
1240 close(handler
->data_sock
[0]);
1242 ERROR("Failed to setup container \"%s\"", handler
->name
);
1243 goto out_warn_father
;
1246 /* Set the label to change to when we exec(2) the container's init. */
1247 ret
= lsm_process_label_set(NULL
, handler
->conf
, true);
1249 goto out_warn_father
;
1251 /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it
1252 * before we aren't allowed anymore.
1254 if (handler
->conf
->no_new_privs
) {
1255 ret
= prctl(PR_SET_NO_NEW_PRIVS
, prctl_arg(1), prctl_arg(0),
1256 prctl_arg(0), prctl_arg(0));
1258 SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block "
1259 "execve() gainable privileges");
1260 goto out_warn_father
;
1262 DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable "
1266 /* Some init's such as busybox will set sane tty settings on stdin,
1267 * stdout, stderr which it thinks is the console. We already set them
1268 * the way we wanted on the real terminal, and we want init to do its
1269 * setup on its console ie. the pty allocated in lxc_terminal_setup() so
1270 * make sure that that pty is stdin,stdout,stderr.
1272 if (handler
->conf
->console
.slave
>= 0) {
1273 if (handler
->daemonize
|| !handler
->conf
->is_execute
)
1274 ret
= set_stdfds(handler
->conf
->console
.slave
);
1276 ret
= lxc_terminal_set_stdfds(handler
->conf
->console
.slave
);
1278 ERROR("Failed to redirect std{in,out,err} to pty file "
1279 "descriptor %d", handler
->conf
->console
.slave
);
1280 goto out_warn_father
;
1284 /* If we mounted a temporary proc, then unmount it now. */
1285 tmp_proc_unmount(handler
->conf
);
1287 ret
= lxc_seccomp_load(handler
->conf
);
1289 goto out_warn_father
;
1291 ret
= run_lxc_hooks(handler
->name
, "start", handler
->conf
, NULL
);
1293 ERROR("Failed to run lxc.hook.start for container \"%s\"",
1295 goto out_warn_father
;
1298 close(handler
->sigfd
);
1300 if (handler
->conf
->console
.slave
< 0 && handler
->daemonize
) {
1301 if (devnull_fd
< 0) {
1302 devnull_fd
= open_devnull();
1304 goto out_warn_father
;
1307 ret
= set_stdfds(devnull_fd
);
1309 ERROR("Failed to redirect std{in,out,err} to \"/dev/null\"");
1310 goto out_warn_father
;
1314 if (devnull_fd
>= 0) {
1321 if (handler
->conf
->init_cwd
) {
1322 ret
= chdir(handler
->conf
->init_cwd
);
1324 SYSERROR("Could not change directory to \"%s\"",
1325 handler
->conf
->init_cwd
);
1326 goto out_warn_father
;
1330 ret
= lxc_sync_barrier_parent(handler
, LXC_SYNC_CGROUP_LIMITS
);
1332 goto out_warn_father
;
1334 /* Reset the environment variables the user requested in a clear
1338 /* Don't error out though. */
1340 SYSERROR("Failed to clear environment.");
1342 lxc_list_for_each(iterator
, &handler
->conf
->environment
) {
1343 ret
= putenv((char *)iterator
->elem
);
1345 SYSERROR("Failed to set environment variable: %s",
1346 (char *)iterator
->elem
);
1347 goto out_warn_father
;
1351 ret
= putenv("container=lxc");
1353 SYSERROR("Failed to set environment variable: container=lxc");
1354 goto out_warn_father
;
1357 if (handler
->conf
->ttys
.tty_names
) {
1358 ret
= putenv(handler
->conf
->ttys
.tty_names
);
1360 SYSERROR("Failed to set environment variable for container ptys");
1361 goto out_warn_father
;
1365 /* The container has been setup. We can now switch to an unprivileged
1368 new_uid
= handler
->conf
->init_uid
;
1369 new_gid
= handler
->conf
->init_gid
;
1371 /* Avoid unnecessary syscalls. */
1372 if (new_uid
== nsuid
)
1373 new_uid
= LXC_INVALID_UID
;
1375 if (new_gid
== nsgid
)
1376 new_gid
= LXC_INVALID_GID
;
1378 if (!lxc_switch_uid_gid(new_uid
, new_gid
))
1379 goto out_warn_father
;
1381 /* If we are in a new user namespace we already dropped all groups when
1382 * we switched to root in the new user namespace further above. Only
1383 * drop groups if we can, so ensure that we have necessary privilege.
1385 if (lxc_list_empty(&handler
->conf
->id_map
))
1387 if (lxc_proc_cap_is_set(CAP_SETGID
, CAP_EFFECTIVE
))
1389 if (!lxc_setgroups(0, NULL
))
1390 goto out_warn_father
;
1392 ret
= lxc_ambient_caps_down();
1394 ERROR("Failed to clear ambient capabilities");
1395 goto out_warn_father
;
1398 if (handler
->conf
->monitor_signal_pdeath
!= SIGKILL
) {
1399 ret
= lxc_set_death_signal(handler
->conf
->monitor_signal_pdeath
, 0);
1401 SYSERROR("Failed to set PR_SET_PDEATHSIG to %d",
1402 handler
->conf
->monitor_signal_pdeath
);
1403 goto out_warn_father
;
1407 /* After this call, we are in error because this ops should not return
1410 handler
->ops
->start(handler
, handler
->data
);
1413 /* We want the parent to know something went wrong, so we return a
1414 * special error code.
1416 lxc_sync_wake_parent(handler
, LXC_SYNC_ERROR
);
1419 if (devnull_fd
>= 0)
1425 static int lxc_recv_ttys_from_child(struct lxc_handler
*handler
)
1428 struct lxc_terminal_info
*tty
;
1430 int sock
= handler
->data_sock
[1];
1431 struct lxc_conf
*conf
= handler
->conf
;
1432 struct lxc_tty_info
*ttys
= &conf
->ttys
;
1434 if (!conf
->ttys
.max
)
1437 ttys
->tty
= malloc(sizeof(*ttys
->tty
) * ttys
->max
);
1441 for (i
= 0; i
< conf
->ttys
.max
; i
++) {
1444 ret
= lxc_abstract_unix_recv_fds(sock
, ttyfds
, 2, NULL
, 0);
1448 tty
= &ttys
->tty
[i
];
1450 tty
->master
= ttyfds
[0];
1451 tty
->slave
= ttyfds
[1];
1452 TRACE("Received pty with master fd %d and slave fd %d from "
1453 "parent", tty
->master
, tty
->slave
);
1457 SYSERROR("Failed to receive %zu ttys from child", ttys
->max
);
1459 TRACE("Received %zu ttys from child", ttys
->max
);
1464 int resolve_clone_flags(struct lxc_handler
*handler
)
1467 struct lxc_conf
*conf
= handler
->conf
;
1469 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1470 if (conf
->ns_keep
> 0) {
1471 if ((conf
->ns_keep
& ns_info
[i
].clone_flag
) == 0)
1472 handler
->ns_clone_flags
|= ns_info
[i
].clone_flag
;
1473 } else if (conf
->ns_clone
> 0) {
1474 if ((conf
->ns_clone
& ns_info
[i
].clone_flag
) > 0)
1475 handler
->ns_clone_flags
|= ns_info
[i
].clone_flag
;
1477 if (i
== LXC_NS_USER
&& lxc_list_empty(&handler
->conf
->id_map
))
1480 if (i
== LXC_NS_NET
&& lxc_requests_empty_network(handler
))
1483 if (i
== LXC_NS_CGROUP
&& !cgns_supported())
1486 handler
->ns_clone_flags
|= ns_info
[i
].clone_flag
;
1489 if (!conf
->ns_share
[i
])
1492 handler
->ns_clone_flags
&= ~ns_info
[i
].clone_flag
;
1493 TRACE("Sharing %s namespace", ns_info
[i
].proc_name
);
1499 /* Note that this function is used with clone(CLONE_VM). Some glibc versions
1500 * used to reset the pid/tid to -1 when CLONE_VM was used without CLONE_THREAD.
1501 * But since the memory between parent and child is shared on CLONE_VM this
1502 * would invalidate the getpid() cache that glibc used to maintain and so
1503 * getpid() in the child would return the parent's pid. This is all fixed in
1504 * newer glibc versions where the getpid() cache is removed and the pid/tid is
1505 * not reset anymore.
1506 * However, if for whatever reason you - dear committer - somehow need to get the
1507 * pid of the dummy intermediate process for do_share_ns() you need to call
1508 * lxc_raw_getpid(). The next lxc_raw_clone() call does not employ CLONE_VM and
1511 static inline int do_share_ns(void *arg
)
1514 struct lxc_handler
*handler
= arg
;
1516 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1517 if (handler
->nsfd
[i
] < 0)
1520 ret
= setns(handler
->nsfd
[i
], 0);
1523 * Note that joining a user and/or mount namespace
1524 * requires the process is not multithreaded otherwise
1525 * setns() will fail here.
1527 SYSERROR("Failed to inherit %s namespace",
1528 ns_info
[i
].proc_name
);
1532 DEBUG("Inherited %s namespace", ns_info
[i
].proc_name
);
1535 flags
= handler
->ns_on_clone_flags
;
1536 flags
|= CLONE_PARENT
;
1537 handler
->pid
= lxc_raw_clone_cb(do_start
, handler
, flags
);
1538 if (handler
->pid
< 0)
1544 static int lxc_setup_shmount(struct lxc_conf
*conf
)
1547 char *full_cont_path
;
1550 /* Construct the shmount path under the container root. */
1551 len_cont
= strlen(conf
->rootfs
.mount
) + 1 + strlen(conf
->shmount
.path_cont
);
1552 /* +1 for the terminating '\0' */
1553 full_cont_path
= malloc(len_cont
+ 1);
1554 if (!full_cont_path
) {
1555 SYSERROR("Not enough memory");
1559 ret
= snprintf(full_cont_path
, len_cont
+ 1, "%s/%s",
1560 conf
->rootfs
.mount
, conf
->shmount
.path_cont
);
1561 if (ret
< 0 || ret
>= len_cont
+ 1) {
1562 SYSERROR("Failed to create filename");
1563 free(full_cont_path
);
1567 /* Check if shmount point is already set up. */
1568 if (is_shared_mountpoint(conf
->shmount
.path_host
)) {
1569 INFO("Path \"%s\" is already MS_SHARED. Reusing",
1570 conf
->shmount
.path_host
);
1571 free(full_cont_path
);
1575 /* Create host and cont mount paths */
1576 ret
= mkdir_p(conf
->shmount
.path_host
, 0711);
1577 if (ret
< 0 && errno
!= EEXIST
) {
1578 SYSERROR("Failed to create directory \"%s\"",
1579 conf
->shmount
.path_host
);
1580 free(full_cont_path
);
1584 ret
= mkdir_p(full_cont_path
, 0711);
1585 if (ret
< 0 && errno
!= EEXIST
) {
1586 SYSERROR("Failed to create directory \"%s\"", full_cont_path
);
1587 free(full_cont_path
);
1591 /* Prepare host mountpoint */
1592 ret
= mount("tmpfs", conf
->shmount
.path_host
, "tmpfs", 0,
1593 "size=100k,mode=0711");
1595 SYSERROR("Failed to mount \"%s\"", conf
->shmount
.path_host
);
1596 free(full_cont_path
);
1600 ret
= mount(conf
->shmount
.path_host
, conf
->shmount
.path_host
, "none",
1601 MS_REC
| MS_SHARED
, "");
1603 SYSERROR("Failed to make shared \"%s\"", conf
->shmount
.path_host
);
1604 free(full_cont_path
);
1608 INFO("Setup shared mount point \"%s\"", conf
->shmount
.path_host
);
1609 free(full_cont_path
);
1613 /* lxc_spawn() performs crucial setup tasks and clone()s the new process which
1614 * exec()s the requested container binary.
1615 * Note that lxc_spawn() runs in the parent namespaces. Any operations performed
1616 * right here should be double checked if they'd pose a security risk. (For
1617 * example, any {u}mount() operations performed here will be reflected on the
1620 static int lxc_spawn(struct lxc_handler
*handler
)
1624 bool wants_to_map_ids
;
1625 struct lxc_list
*id_map
;
1626 const char *name
= handler
->name
;
1627 const char *lxcpath
= handler
->lxcpath
;
1628 bool share_ns
= false;
1629 struct lxc_conf
*conf
= handler
->conf
;
1630 struct cgroup_ops
*cgroup_ops
= handler
->cgroup_ops
;
1632 id_map
= &conf
->id_map
;
1633 wants_to_map_ids
= !lxc_list_empty(id_map
);
1635 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1636 if (!conf
->ns_share
[i
])
1639 handler
->nsfd
[i
] = lxc_inherit_namespace(conf
->ns_share
[i
], lxcpath
, ns_info
[i
].proc_name
);
1640 if (handler
->nsfd
[i
] < 0)
1646 ret
= lxc_sync_init(handler
);
1650 ret
= socketpair(AF_UNIX
, SOCK_STREAM
| SOCK_CLOEXEC
, 0,
1651 handler
->data_sock
);
1655 ret
= resolve_clone_flags(handler
);
1659 if (conf
->shmount
.path_host
) {
1660 if (!conf
->shmount
.path_cont
)
1663 ret
= lxc_setup_shmount(conf
);
1665 ERROR("Failed to setup shared mount point");
1670 if (handler
->ns_clone_flags
& CLONE_NEWNET
) {
1671 if (!lxc_list_empty(&conf
->network
)) {
1673 /* Find gateway addresses from the link device, which is
1674 * no longer accessible inside the container. Do this
1675 * before creating network interfaces, since goto
1676 * out_delete_net does not work before lxc_clone.
1678 ret
= lxc_find_gateway_addresses(handler
);
1680 ERROR("Failed to find gateway addresses");
1684 /* That should be done before the clone because we will
1685 * fill the netdev index and use them in the child.
1687 ret
= lxc_create_network_priv(handler
);
1689 ERROR("Failed to create the network");
1690 goto out_delete_net
;
1695 if (!cgroup_ops
->payload_create(cgroup_ops
, handler
)) {
1696 ERROR("Failed creating cgroups");
1697 goto out_delete_net
;
1700 /* If the rootfs is not a blockdev, prevent the container from marking
1702 * If the container is unprivileged then skip rootfs pinning.
1704 if (!wants_to_map_ids
) {
1705 handler
->pinfd
= pin_rootfs(conf
->rootfs
.path
);
1706 if (handler
->pinfd
== -1)
1707 INFO("Failed to pin the rootfs for container \"%s\"", handler
->name
);
1710 /* Create a process in a new set of namespaces. */
1711 handler
->ns_on_clone_flags
= handler
->ns_clone_flags
;
1712 if (handler
->ns_clone_flags
& CLONE_NEWUSER
) {
1713 /* If CLONE_NEWUSER and CLONE_NEWNET was requested, we need to
1714 * clone a new user namespace first and only later unshare our
1715 * network namespace to ensure that network devices ownership is
1718 handler
->ns_on_clone_flags
&= ~CLONE_NEWNET
;
1720 /* The cgroup namespace gets unshare()ed not clone()ed. */
1721 handler
->ns_on_clone_flags
&= ~CLONE_NEWCGROUP
;
1726 attacher_pid
= lxc_clone(do_share_ns
, handler
,
1727 CLONE_VFORK
| CLONE_VM
| CLONE_FILES
);
1728 if (attacher_pid
< 0) {
1729 SYSERROR(LXC_CLONE_ERROR
);
1730 goto out_delete_net
;
1733 ret
= wait_for_pid(attacher_pid
);
1735 SYSERROR("Intermediate process failed");
1736 goto out_delete_net
;
1739 handler
->pid
= lxc_raw_clone_cb(do_start
, handler
,
1740 handler
->ns_on_clone_flags
);
1742 if (handler
->pid
< 0) {
1743 SYSERROR(LXC_CLONE_ERROR
);
1744 goto out_delete_net
;
1746 TRACE("Cloned child process %d", handler
->pid
);
1748 for (i
= 0; i
< LXC_NS_MAX
; i
++)
1749 if (handler
->ns_on_clone_flags
& ns_info
[i
].clone_flag
)
1750 INFO("Cloned %s", ns_info
[i
].flag_name
);
1752 if (!lxc_try_preserve_namespaces(handler
, handler
->ns_on_clone_flags
, handler
->pid
)) {
1753 ERROR("Failed to preserve cloned namespaces for lxc.hook.stop");
1754 goto out_delete_net
;
1757 lxc_sync_fini_child(handler
);
1759 /* Map the container uids. The container became an invalid userid the
1760 * moment it was cloned with CLONE_NEWUSER. This call doesn't change
1761 * anything immediately, but allows the container to setuid(0) (0 being
1762 * mapped to something else on the host.) later to become a valid uid
1765 if (wants_to_map_ids
) {
1766 if (!handler
->conf
->ns_share
[LXC_NS_USER
] &&
1767 (handler
->conf
->ns_keep
& CLONE_NEWUSER
) == 0) {
1768 ret
= lxc_map_ids(id_map
, handler
->pid
);
1770 ERROR("Failed to set up id mapping.");
1771 goto out_delete_net
;
1776 ret
= lxc_sync_wake_child(handler
, LXC_SYNC_STARTUP
);
1778 goto out_delete_net
;
1780 ret
= lxc_sync_wait_child(handler
, LXC_SYNC_CONFIGURE
);
1782 goto out_delete_net
;
1784 if (!cgroup_ops
->setup_limits(cgroup_ops
, handler
->conf
, false)) {
1785 ERROR("Failed to setup cgroup limits for container \"%s\"", name
);
1786 goto out_delete_net
;
1789 if (!cgroup_ops
->payload_enter(cgroup_ops
, handler
->pid
))
1790 goto out_delete_net
;
1792 if (!cgroup_ops
->chown(cgroup_ops
, handler
->conf
))
1793 goto out_delete_net
;
1795 /* Now we're ready to preserve the network namespace */
1796 ret
= lxc_try_preserve_ns(handler
->pid
, "net");
1798 if (ret
!= -EOPNOTSUPP
) {
1799 SYSERROR("Failed to preserve net namespace");
1800 goto out_delete_net
;
1803 handler
->nsfd
[LXC_NS_NET
] = ret
;
1804 DEBUG("Preserved net namespace via fd %d", ret
);
1806 ret
= lxc_netns_set_nsid(handler
->nsfd
[LXC_NS_NET
]);
1808 SYSWARN("Failed to allocate new network namespace id");
1810 TRACE("Allocated new network namespace id");
1813 /* Create the network configuration. */
1814 if (handler
->ns_clone_flags
& CLONE_NEWNET
) {
1815 ret
= lxc_network_move_created_netdev_priv(handler
->lxcpath
,
1820 ERROR("Failed to create the configured network");
1821 goto out_delete_net
;
1824 ret
= lxc_create_network_unpriv(handler
->lxcpath
, handler
->name
,
1825 &conf
->network
, handler
->pid
, conf
->hooks_version
);
1827 ERROR("Failed to create the configured network");
1828 goto out_delete_net
;
1832 ret
= lxc_network_send_veth_names_to_child(handler
);
1834 ERROR("Failed to send veth names to child");
1835 goto out_delete_net
;
1838 if (!lxc_list_empty(&conf
->procs
)) {
1839 ret
= setup_proc_filesystem(&conf
->procs
, handler
->pid
);
1841 goto out_delete_net
;
1844 /* Tell the child to continue its initialization. We'll get
1845 * LXC_SYNC_CGROUP when it is ready for us to setup cgroups.
1847 ret
= lxc_sync_barrier_child(handler
, LXC_SYNC_POST_CONFIGURE
);
1849 goto out_delete_net
;
1851 if (!lxc_list_empty(&conf
->limits
)) {
1852 ret
= setup_resource_limits(&conf
->limits
, handler
->pid
);
1854 ERROR("Failed to setup resource limits");
1855 goto out_delete_net
;
1859 ret
= lxc_sync_barrier_child(handler
, LXC_SYNC_CGROUP_UNSHARE
);
1861 goto out_delete_net
;
1863 if (!cgroup_ops
->setup_limits(cgroup_ops
, handler
->conf
, true)) {
1864 ERROR("Failed to setup legacy device cgroup controller limits");
1865 goto out_delete_net
;
1867 TRACE("Set up legacy device cgroup controller limits");
1869 if (handler
->ns_clone_flags
& CLONE_NEWCGROUP
) {
1870 /* Now we're ready to preserve the cgroup namespace */
1871 ret
= lxc_try_preserve_ns(handler
->pid
, "cgroup");
1873 if (ret
!= -EOPNOTSUPP
) {
1874 SYSERROR("Failed to preserve cgroup namespace");
1875 goto out_delete_net
;
1878 handler
->nsfd
[LXC_NS_CGROUP
] = ret
;
1879 DEBUG("Preserved cgroup namespace via fd %d", ret
);
1883 ret
= snprintf(pidstr
, 20, "%d", handler
->pid
);
1884 if (ret
< 0 || ret
>= 20)
1885 goto out_delete_net
;
1887 ret
= setenv("LXC_PID", pidstr
, 1);
1889 SYSERROR("Failed to set environment variable: LXC_PID=%s", pidstr
);
1891 /* Run any host-side start hooks */
1892 ret
= run_lxc_hooks(name
, "start-host", conf
, NULL
);
1894 ERROR("Failed to run lxc.hook.start-host");
1895 goto out_delete_net
;
1898 /* Tell the child to complete its initialization and wait for it to exec
1899 * or return an error. (The child will never return
1900 * LXC_SYNC_READY_START+1. It will either close the sync pipe, causing
1901 * lxc_sync_barrier_child to return success, or return a different
1902 * value, causing us to error out).
1904 ret
= lxc_sync_barrier_child(handler
, LXC_SYNC_READY_START
);
1906 goto out_delete_net
;
1908 ret
= lxc_network_recv_name_and_ifindex_from_child(handler
);
1910 ERROR("Failed to receive names and ifindices for network "
1911 "devices from child");
1912 goto out_delete_net
;
1915 /* Now all networks are created, network devices are moved into place,
1916 * and the correct names and ifindeces in the respective namespaces have
1917 * been recorded. The corresponding structs have now all been filled. So
1918 * log them for debugging purposes.
1920 lxc_log_configured_netdevs(conf
);
1922 /* Read tty fds allocated by child. */
1923 ret
= lxc_recv_ttys_from_child(handler
);
1925 ERROR("Failed to receive tty info from child process");
1926 goto out_delete_net
;
1929 ret
= handler
->ops
->post_start(handler
, handler
->data
);
1933 ret
= lxc_set_state(name
, handler
, RUNNING
);
1935 ERROR("Failed to set state to \"%s\"", lxc_state2str(RUNNING
));
1939 lxc_sync_fini(handler
);
1944 if (handler
->ns_clone_flags
& CLONE_NEWNET
)
1945 lxc_delete_network(handler
);
1948 lxc_abort(name
, handler
);
1951 lxc_sync_fini(handler
);
1952 if (handler
->pinfd
>= 0) {
1953 close(handler
->pinfd
);
1954 handler
->pinfd
= -1;
1960 int __lxc_start(const char *name
, struct lxc_handler
*handler
,
1961 struct lxc_operations
* ops
, void *data
, const char *lxcpath
,
1962 bool daemonize
, int *error_num
)
1965 struct lxc_conf
*conf
= handler
->conf
;
1966 struct cgroup_ops
*cgroup_ops
;
1968 ret
= lxc_init(name
, handler
);
1970 ERROR("Failed to initialize container \"%s\"", name
);
1974 handler
->data
= data
;
1975 handler
->daemonize
= daemonize
;
1976 cgroup_ops
= handler
->cgroup_ops
;
1978 if (!attach_block_device(handler
->conf
)) {
1979 ERROR("Failed to attach block device");
1980 goto out_fini_nonet
;
1983 if (!cgroup_ops
->monitor_create(cgroup_ops
, handler
)) {
1984 ERROR("Failed to create monitor cgroup");
1985 goto out_fini_nonet
;
1988 if (!cgroup_ops
->monitor_enter(cgroup_ops
, handler
->monitor_pid
)) {
1989 ERROR("Failed to enter monitor cgroup");
1990 goto out_fini_nonet
;
1993 if (geteuid() == 0 && !lxc_list_empty(&conf
->id_map
)) {
1994 /* If the backing store is a device, mount it here and now. */
1995 if (rootfs_is_blockdev(conf
)) {
1996 ret
= unshare(CLONE_NEWNS
);
1998 ERROR("Failed to unshare CLONE_NEWNS");
1999 goto out_fini_nonet
;
2001 INFO("Unshared CLONE_NEWNS");
2003 remount_all_slave();
2004 ret
= lxc_setup_rootfs_prepare_root(conf
, name
, lxcpath
);
2006 ERROR("Error setting up rootfs mount as root before spawn");
2007 goto out_fini_nonet
;
2009 INFO("Set up container rootfs as host root");
2013 ret
= lxc_spawn(handler
);
2015 ERROR("Failed to spawn container \"%s\"", name
);
2016 goto out_detach_blockdev
;
2018 /* close parent side of data socket */
2019 close(handler
->data_sock
[0]);
2020 handler
->data_sock
[0] = -1;
2021 close(handler
->data_sock
[1]);
2022 handler
->data_sock
[1] = -1;
2024 handler
->conf
->reboot
= REBOOT_NONE
;
2026 ret
= lxc_poll(name
, handler
);
2028 ERROR("LXC mainloop exited with error: %d", ret
);
2032 if (!handler
->init_died
&& handler
->pid
> 0) {
2033 ERROR("Child process is not killed");
2037 status
= lxc_wait_for_pid_status(handler
->pid
);
2039 SYSERROR("Failed to retrieve status for %d", handler
->pid
);
2041 /* If the child process exited but was not signaled, it didn't call
2042 * reboot. This should mean it was an lxc-execute which simply exited.
2043 * In any case, treat it as a 'halt'.
2045 if (WIFSIGNALED(status
)) {
2046 switch(WTERMSIG(status
)) {
2047 case SIGINT
: /* halt */
2048 DEBUG("Container \"%s\" is halting", name
);
2050 case SIGHUP
: /* reboot */
2051 DEBUG("Container \"%s\" is rebooting", name
);
2052 handler
->conf
->reboot
= REBOOT_REQ
;
2054 case SIGSYS
: /* seccomp */
2055 DEBUG("Container \"%s\" violated its seccomp policy", name
);
2058 DEBUG("Unknown exit status for container \"%s\" init %d", name
, WTERMSIG(status
));
2063 ret
= lxc_restore_phys_nics_to_netns(handler
);
2065 ERROR("Failed to move physical network devices back to parent "
2066 "network namespace");
2068 if (handler
->pinfd
>= 0) {
2069 close(handler
->pinfd
);
2070 handler
->pinfd
= -1;
2073 lxc_monitor_send_exit_code(name
, status
, handler
->lxcpath
);
2074 lxc_error_set_and_log(handler
->pid
, status
);
2076 *error_num
= handler
->exit_status
;
2079 lxc_delete_network(handler
);
2081 out_detach_blockdev
:
2082 detach_block_device(handler
->conf
);
2085 lxc_fini(name
, handler
);
2089 lxc_abort(name
, handler
);
2097 static int start(struct lxc_handler
*handler
, void* data
)
2099 struct start_args
*arg
= data
;
2101 NOTICE("Exec'ing \"%s\"", arg
->argv
[0]);
2103 execvp(arg
->argv
[0], arg
->argv
);
2104 SYSERROR("Failed to exec \"%s\"", arg
->argv
[0]);
2108 static int post_start(struct lxc_handler
*handler
, void* data
)
2110 struct start_args
*arg
= data
;
2112 NOTICE("Started \"%s\" with pid \"%d\"", arg
->argv
[0], handler
->pid
);
2116 static struct lxc_operations start_ops
= {
2118 .post_start
= post_start
2121 int lxc_start(const char *name
, char *const argv
[], struct lxc_handler
*handler
,
2122 const char *lxcpath
, bool daemonize
, int *error_num
)
2124 struct start_args start_arg
= {
2128 TRACE("Doing lxc_start");
2129 return __lxc_start(name
, handler
, &start_ops
, &start_arg
, lxcpath
, daemonize
, error_num
);
2132 static void lxc_destroy_container_on_signal(struct lxc_handler
*handler
,
2135 char destroy
[PATH_MAX
];
2136 struct lxc_container
*c
;
2140 if (handler
->conf
->rootfs
.path
&& handler
->conf
->rootfs
.mount
) {
2141 bret
= do_destroy_container(handler
);
2143 ERROR("Error destroying rootfs for container \"%s\"", name
);
2147 INFO("Destroyed rootfs for container \"%s\"", name
);
2149 ret
= snprintf(destroy
, PATH_MAX
, "%s/%s", handler
->lxcpath
, name
);
2150 if (ret
< 0 || ret
>= PATH_MAX
) {
2151 ERROR("Error destroying directory for container \"%s\"", name
);
2155 c
= lxc_container_new(name
, handler
->lxcpath
);
2157 if (container_disk_lock(c
)) {
2158 INFO("Could not update lxc_snapshots file");
2159 lxc_container_put(c
);
2161 mod_all_rdeps(c
, false);
2162 container_disk_unlock(c
);
2163 lxc_container_put(c
);
2167 if (!handler
->am_root
)
2168 ret
= userns_exec_full(handler
->conf
, lxc_rmdir_onedev_wrapper
,
2169 destroy
, "lxc_rmdir_onedev_wrapper");
2171 ret
= lxc_rmdir_onedev(destroy
, NULL
);
2174 ERROR("Error destroying directory for container \"%s\"", name
);
2177 INFO("Destroyed directory for container \"%s\"", name
);
2180 static int lxc_rmdir_onedev_wrapper(void *data
)
2182 char *arg
= (char *) data
;
2183 return lxc_rmdir_onedev(arg
, NULL
);
2186 static bool do_destroy_container(struct lxc_handler
*handler
)
2190 if (!handler
->am_root
) {
2191 ret
= userns_exec_full(handler
->conf
, storage_destroy_wrapper
,
2192 handler
->conf
, "storage_destroy_wrapper");
2199 return storage_destroy(handler
->conf
);