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
40 #include <sys/mount.h>
41 #include <sys/param.h>
42 #include <sys/prctl.h>
43 #include <sys/socket.h>
45 #include <sys/syscall.h>
46 #include <sys/types.h>
55 #include "commands_utils.h"
58 #include "confile_utils.h"
60 #include "file_utils.h"
64 #include "lxccontainer.h"
66 #include "lxcseccomp.h"
69 #include "memory_utils.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
)
100 __do_free
char *copy
= NULL
;
104 copy
= must_copy_string(path
);
106 e
= copy
+ strlen(path
);
109 while (p
< e
&& *p
== '/')
112 while (p
< e
&& *p
!= '/')
118 ret
= access(copy
, X_OK
);
120 SYSERROR("Could not access %s. Please grant it x "
121 "access, or add an ACL for the container "
129 static void lxc_put_nsfds(struct lxc_handler
*handler
)
133 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
134 if (handler
->nsfd
[i
] < 0)
137 close(handler
->nsfd
[i
]);
138 handler
->nsfd
[i
] = -EBADF
;
142 static int lxc_try_preserve_ns(const int pid
, const char *ns
)
146 fd
= lxc_preserve_ns(pid
, ns
);
148 if (errno
!= ENOENT
) {
149 SYSERROR("Failed to preserve %s namespace", ns
);
153 SYSWARN("Kernel does not support preserving %s namespaces", ns
);
160 /* lxc_try_preserve_namespaces: open /proc/@pid/ns/@ns for each namespace
161 * specified in ns_clone_flags.
162 * Return true on success, false on failure.
164 static bool lxc_try_preserve_namespaces(struct lxc_handler
*handler
,
165 int ns_clone_flags
, pid_t pid
)
169 for (i
= 0; i
< LXC_NS_MAX
; i
++)
170 handler
->nsfd
[i
] = -EBADF
;
172 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
175 if ((ns_clone_flags
& ns_info
[i
].clone_flag
) == 0)
178 fd
= lxc_try_preserve_ns(pid
, ns_info
[i
].proc_name
);
180 /* Do not fail to start container on kernels that do
181 * not support interacting with namespaces through
184 if (fd
== -EOPNOTSUPP
)
187 lxc_put_nsfds(handler
);
191 handler
->nsfd
[i
] = fd
;
192 DEBUG("Preserved %s namespace via fd %d", ns_info
[i
].proc_name
,
199 static inline bool match_stdfds(int fd
)
201 return (fd
== STDIN_FILENO
|| fd
== STDOUT_FILENO
|| fd
== STDERR_FILENO
);
205 static bool match_dlog_fds(struct dirent
*direntp
)
207 char path
[PATH_MAX
] = {0};
208 char link
[PATH_MAX
] = {0};
212 ret
= snprintf(path
, PATH_MAX
, "/proc/self/fd/%s", direntp
->d_name
);
213 if (ret
< 0 || ret
>= PATH_MAX
) {
214 ERROR("Failed to create file descriptor name");
218 linklen
= readlink(path
, link
, PATH_MAX
);
220 SYSERROR("Failed to read link path - \"%s\"", path
);
222 } else if (linklen
>= PATH_MAX
) {
223 ERROR("The name of link path is too long - \"%s\"", path
);
227 if (strcmp(link
, "/dev/log_main") == 0 ||
228 strcmp(link
, "/dev/log_system") == 0 ||
229 strcmp(link
, "/dev/log_radio") == 0)
236 int lxc_check_inherited(struct lxc_conf
*conf
, bool closeall
,
237 int *fds_to_ignore
, size_t len_fds
)
242 struct dirent
*direntp
;
244 if (conf
&& conf
->close_all_fds
)
248 dir
= opendir("/proc/self/fd");
250 SYSWARN("Failed to open directory");
256 while ((direntp
= readdir(dir
))) {
258 struct lxc_list
*cur
;
259 bool matched
= false;
261 if (strcmp(direntp
->d_name
, ".") == 0)
264 if (strcmp(direntp
->d_name
, "..") == 0)
267 ret
= lxc_safe_int(direntp
->d_name
, &fd
);
269 INFO("Could not parse file descriptor for \"%s\"", direntp
->d_name
);
273 for (i
= 0; i
< len_fds
; i
++)
274 if (fds_to_ignore
[i
] == fd
)
277 if (fd
== fddir
|| fd
== lxc_log_fd
||
278 (i
< len_fds
&& fd
== fds_to_ignore
[i
]))
281 /* Keep state clients that wait on reboots. */
283 lxc_list_for_each(cur
, &conf
->state_clients
) {
284 struct lxc_state_client
*client
= cur
->elem
;
286 if (client
->clientfd
!= fd
)
297 if (current_config
&& fd
== current_config
->logfd
)
300 if (match_stdfds(fd
))
304 if (match_dlog_fds(direntp
))
311 INFO("Closed inherited fd %d", fd
);
314 WARN("Inherited fd %d", fd
);
317 /* Only enable syslog at this point to avoid the above logging function
318 * to open a new fd and make the check_inherited function enter an
321 lxc_log_enable_syslog();
323 closedir(dir
); /* cannot fail */
327 static int setup_signal_fd(sigset_t
*oldmask
)
331 const int signals
[] = {SIGBUS
, SIGILL
, SIGSEGV
, SIGWINCH
};
333 /* Block everything except serious error signals. */
334 ret
= sigfillset(&mask
);
338 for (int sig
= 0; sig
< (sizeof(signals
) / sizeof(signals
[0])); sig
++) {
339 ret
= sigdelset(&mask
, signals
[sig
]);
344 ret
= pthread_sigmask(SIG_BLOCK
, &mask
, oldmask
);
346 SYSERROR("Failed to set signal mask");
350 ret
= signalfd(-1, &mask
, SFD_CLOEXEC
);
352 SYSERROR("Failed to create signal file descriptor");
356 TRACE("Created signal file descriptor %d", ret
);
361 static int signal_handler(int fd
, uint32_t events
, void *data
,
362 struct lxc_epoll_descr
*descr
)
366 struct signalfd_siginfo siginfo
;
367 struct lxc_handler
*hdlr
= data
;
369 ret
= lxc_read_nointr(fd
, &siginfo
, sizeof(siginfo
));
371 ERROR("Failed to read signal info from signal file descriptor %d", fd
);
372 return LXC_MAINLOOP_ERROR
;
375 if (ret
!= sizeof(siginfo
)) {
376 ERROR("Unexpected size for struct signalfd_siginfo");
380 /* Check whether init is running. */
382 ret
= waitid(P_PID
, hdlr
->pid
, &info
, WEXITED
| WNOWAIT
| WNOHANG
);
383 if (ret
== 0 && info
.si_pid
== hdlr
->pid
)
384 hdlr
->init_died
= true;
386 /* Try to figure out a reasonable exit status to report. */
387 if (hdlr
->init_died
) {
388 switch (info
.si_code
) {
390 hdlr
->exit_status
= info
.si_status
<< 8;
395 hdlr
->exit_status
= info
.si_status
<< 8 | 0x7f;
398 /* Huh? The waitid() told us it's dead *and* continued? */
399 WARN("Init %d dead and continued?", hdlr
->pid
);
400 hdlr
->exit_status
= 1;
403 ERROR("Unknown si_code: %d", info
.si_code
);
404 hdlr
->exit_status
= 1;
408 if (siginfo
.ssi_signo
== SIGHUP
) {
409 kill(hdlr
->pid
, SIGTERM
);
410 INFO("Killing %d since terminal hung up", hdlr
->pid
);
411 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
412 : LXC_MAINLOOP_CONTINUE
;
415 if (siginfo
.ssi_signo
!= SIGCHLD
) {
416 kill(hdlr
->pid
, siginfo
.ssi_signo
);
417 INFO("Forwarded signal %d to pid %d", siginfo
.ssi_signo
, hdlr
->pid
);
418 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
419 : LXC_MAINLOOP_CONTINUE
;
422 /* More robustness, protect ourself from a SIGCHLD sent
423 * by a process different from the container init.
425 if (siginfo
.ssi_pid
!= hdlr
->pid
) {
426 NOTICE("Received %d from pid %d instead of container init %d",
427 siginfo
.ssi_signo
, siginfo
.ssi_pid
, hdlr
->pid
);
428 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
429 : LXC_MAINLOOP_CONTINUE
;
432 if (siginfo
.ssi_code
== CLD_STOPPED
) {
433 INFO("Container init process was stopped");
434 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
435 : LXC_MAINLOOP_CONTINUE
;
438 if (siginfo
.ssi_code
== CLD_CONTINUED
) {
439 INFO("Container init process was continued");
440 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
441 : LXC_MAINLOOP_CONTINUE
;
444 DEBUG("Container init process %d exited", hdlr
->pid
);
446 return LXC_MAINLOOP_CLOSE
;
449 int lxc_serve_state_clients(const char *name
, struct lxc_handler
*handler
,
454 struct lxc_list
*cur
, *next
;
455 struct lxc_msg msg
= {.type
= lxc_msg_state
, .value
= state
};
458 handler
->state
= RUNNING
;
460 handler
->state
= state
;
462 TRACE("Set container state to %s", lxc_state2str(state
));
464 if (lxc_list_empty(&handler
->conf
->state_clients
)) {
465 TRACE("No state clients registered");
469 retlen
= strlcpy(msg
.name
, name
, sizeof(msg
.name
));
470 if (retlen
>= sizeof(msg
.name
))
473 lxc_list_for_each_safe(cur
, &handler
->conf
->state_clients
, next
) {
474 struct lxc_state_client
*client
= cur
->elem
;
476 if (client
->states
[state
] == 0) {
477 TRACE("State %s not registered for state client %d",
478 lxc_state2str(state
), client
->clientfd
);
482 TRACE("Sending state %s to state client %d",
483 lxc_state2str(state
), client
->clientfd
);
485 ret
= lxc_send_nointr(client
->clientfd
, &msg
, sizeof(msg
), MSG_NOSIGNAL
);
487 SYSERROR("Failed to send message to client");
489 /* kick client from list */
491 close(client
->clientfd
);
499 static int lxc_serve_state_socket_pair(const char *name
,
500 struct lxc_handler
*handler
,
505 if (!handler
->daemonize
||
506 handler
->state_socket_pair
[1] < 0 ||
510 /* Close read end of the socket pair. */
511 close(handler
->state_socket_pair
[0]);
512 handler
->state_socket_pair
[0] = -1;
515 ret
= lxc_abstract_unix_send_credential(handler
->state_socket_pair
[1],
516 &(int){state
}, sizeof(int));
518 SYSERROR("Failed to send state to %d", handler
->state_socket_pair
[1]);
526 if (ret
!= sizeof(int)) {
527 ERROR("Message too long : %d", handler
->state_socket_pair
[1]);
531 TRACE("Sent container state \"%s\" to %d", lxc_state2str(state
),
532 handler
->state_socket_pair
[1]);
534 /* Close write end of the socket pair. */
535 close(handler
->state_socket_pair
[1]);
536 handler
->state_socket_pair
[1] = -1;
541 int lxc_set_state(const char *name
, struct lxc_handler
*handler
,
546 ret
= lxc_serve_state_socket_pair(name
, handler
, state
);
548 ERROR("Failed to synchronize via anonymous pair of unix sockets");
552 ret
= lxc_serve_state_clients(name
, handler
, state
);
556 /* This function will try to connect to the legacy lxc-monitord state
557 * server and only exists for backwards compatibility.
559 lxc_monitor_send_state(name
, state
, handler
->lxcpath
);
564 int lxc_poll(const char *name
, struct lxc_handler
*handler
)
567 bool has_console
= true;
568 struct lxc_epoll_descr descr
, descr_console
;
570 if (handler
->conf
->console
.path
&&
571 strcmp(handler
->conf
->console
.path
, "none") == 0)
574 ret
= lxc_mainloop_open(&descr
);
576 ERROR("Failed to create mainloop");
581 ret
= lxc_mainloop_open(&descr_console
);
583 ERROR("Failed to create console mainloop");
588 ret
= lxc_mainloop_add_handler(&descr
, handler
->sigfd
, signal_handler
, handler
);
590 ERROR("Failed to add signal handler for %d to mainloop", handler
->sigfd
);
591 goto out_mainloop_console
;
594 #if HAVE_DECL_SECCOMP_NOTIF_GET_FD
595 if (handler
->conf
->has_seccomp_notify
&&
596 handler
->conf
->seccomp_notify_proxy_addr
.sun_path
[1] != '\0') {
597 __do_close_prot_errno
int notify_fd
= -EBADF
;
599 notify_fd
= lxc_unix_connect(&handler
->conf
->seccomp_notify_proxy_addr
);
601 goto out_mainloop_console
;
603 /* 30 second timeout */
604 ret
= lxc_socket_set_timeout(notify_fd
, 30, 30);
606 goto out_mainloop_console
;
608 ret
= lxc_mainloop_add_handler(&descr
,
609 handler
->conf
->seccomp_notify_fd
,
610 seccomp_notify_handler
, handler
);
612 ERROR("Failed to add seccomp notify handler for %d to mainloop",
613 handler
->conf
->seccomp_notify_fd
);
614 goto out_mainloop_console
;
617 handler
->conf
->seccomp_notify_proxy_fd
= move_fd(notify_fd
);
622 struct lxc_terminal
*console
= &handler
->conf
->console
;
624 ret
= lxc_terminal_mainloop_add(&descr
, console
);
626 ERROR("Failed to add console handlers to mainloop");
627 goto out_mainloop_console
;
630 ret
= lxc_terminal_mainloop_add(&descr_console
, console
);
632 ERROR("Failed to add console handlers to console mainloop");
633 goto out_mainloop_console
;
636 handler
->conf
->console
.descr
= &descr
;
639 ret
= lxc_cmd_mainloop_add(name
, &descr
, handler
);
641 ERROR("Failed to add command handler to mainloop");
642 goto out_mainloop_console
;
645 TRACE("Mainloop is ready");
647 ret
= lxc_mainloop(&descr
, -1);
650 if (ret
< 0 || !handler
->init_died
)
651 goto out_mainloop_console
;
654 ret
= lxc_mainloop(&descr_console
, 0);
656 out_mainloop_console
:
658 lxc_mainloop_close(&descr_console
);
659 TRACE("Closed console mainloop");
663 lxc_mainloop_close(&descr
);
664 TRACE("Closed mainloop");
667 close(handler
->sigfd
);
668 TRACE("Closed signal file descriptor %d", handler
->sigfd
);
669 handler
->sigfd
= -EBADF
;
674 void lxc_zero_handler(struct lxc_handler
*handler
)
678 memset(handler
, 0, sizeof(struct lxc_handler
));
684 for (i
= 0; i
< LXC_NS_MAX
; i
++)
685 handler
->nsfd
[i
] = -1;
687 handler
->data_sock
[0] = -1;
688 handler
->data_sock
[1] = -1;
690 handler
->state_socket_pair
[0] = -1;
691 handler
->state_socket_pair
[1] = -1;
693 handler
->sync_sock
[0] = -1;
694 handler
->sync_sock
[1] = -1;
697 void lxc_free_handler(struct lxc_handler
*handler
)
699 if (handler
->pinfd
>= 0)
700 close(handler
->pinfd
);
702 if (handler
->sigfd
>= 0)
703 close(handler
->sigfd
);
705 lxc_put_nsfds(handler
);
707 if (handler
->conf
&& handler
->conf
->reboot
== REBOOT_NONE
)
708 if (handler
->conf
->maincmd_fd
>= 0)
709 lxc_abstract_unix_close(handler
->conf
->maincmd_fd
);
711 if (handler
->state_socket_pair
[0] >= 0)
712 close(handler
->state_socket_pair
[0]);
714 if (handler
->state_socket_pair
[1] >= 0)
715 close(handler
->state_socket_pair
[1]);
717 if (handler
->cgroup_ops
)
718 cgroup_exit(handler
->cgroup_ops
);
720 handler
->conf
= NULL
;
725 struct lxc_handler
*lxc_init_handler(const char *name
, struct lxc_conf
*conf
,
726 const char *lxcpath
, bool daemonize
)
729 struct lxc_handler
*handler
;
731 handler
= malloc(sizeof(*handler
));
735 memset(handler
, 0, sizeof(*handler
));
737 /* Note that am_guest_unpriv() checks the effective uid. We
738 * probably don't care if we are real root only if we are running
739 * as root so this should be fine.
741 handler
->am_root
= !am_guest_unpriv();
742 handler
->data_sock
[0] = handler
->data_sock
[1] = -1;
743 handler
->conf
= conf
;
744 handler
->lxcpath
= lxcpath
;
746 handler
->sigfd
= -EBADF
;
747 handler
->init_died
= false;
748 handler
->state_socket_pair
[0] = handler
->state_socket_pair
[1] = -1;
749 if (handler
->conf
->reboot
== REBOOT_NONE
)
750 lxc_list_init(&handler
->conf
->state_clients
);
752 for (i
= 0; i
< LXC_NS_MAX
; i
++)
753 handler
->nsfd
[i
] = -1;
755 handler
->name
= name
;
757 if (daemonize
&& handler
->conf
->reboot
== REBOOT_NONE
) {
758 /* Create socketpair() to synchronize on daemonized startup.
759 * When the container reboots we don't need to synchronize
760 * again currently so don't open another socketpair().
762 ret
= socketpair(AF_UNIX
, SOCK_STREAM
| SOCK_CLOEXEC
, 0,
763 handler
->state_socket_pair
);
765 ERROR("Failed to create anonymous pair of unix sockets");
769 TRACE("Created anonymous pair {%d,%d} of unix sockets",
770 handler
->state_socket_pair
[0],
771 handler
->state_socket_pair
[1]);
774 if (handler
->conf
->reboot
== REBOOT_NONE
) {
775 handler
->conf
->maincmd_fd
= lxc_cmd_init(name
, lxcpath
, "command");
776 if (handler
->conf
->maincmd_fd
< 0) {
777 ERROR("Failed to set up command socket");
782 TRACE("Unix domain socket %d for command server is ready",
783 handler
->conf
->maincmd_fd
);
788 lxc_free_handler(handler
);
793 int lxc_init(const char *name
, struct lxc_handler
*handler
)
796 const char *loglevel
;
797 struct lxc_conf
*conf
= handler
->conf
;
799 handler
->monitor_pid
= lxc_raw_getpid();
802 TRACE("Initialized LSM");
804 ret
= lxc_read_seccomp_config(conf
);
806 ERROR("Failed loading seccomp policy");
807 goto out_close_maincmd_fd
;
809 TRACE("Read seccomp policy");
811 /* Begin by setting the state to STARTING. */
812 ret
= lxc_set_state(name
, handler
, STARTING
);
814 ERROR("Failed to set state to \"%s\"", lxc_state2str(STARTING
));
815 goto out_close_maincmd_fd
;
817 TRACE("Set container state to \"STARTING\"");
819 /* Start of environment variable setup for hooks. */
820 ret
= setenv("LXC_NAME", name
, 1);
822 SYSERROR("Failed to set environment variable: LXC_NAME=%s", name
);
825 ret
= setenv("LXC_CONFIG_FILE", conf
->rcfile
, 1);
827 SYSERROR("Failed to set environment variable: "
828 "LXC_CONFIG_FILE=%s", conf
->rcfile
);
831 if (conf
->rootfs
.mount
) {
832 ret
= setenv("LXC_ROOTFS_MOUNT", conf
->rootfs
.mount
, 1);
834 SYSERROR("Failed to set environment variable: "
835 "LXC_ROOTFS_MOUNT=%s", conf
->rootfs
.mount
);
838 if (conf
->rootfs
.path
) {
839 ret
= setenv("LXC_ROOTFS_PATH", conf
->rootfs
.path
, 1);
841 SYSERROR("Failed to set environment variable: "
842 "LXC_ROOTFS_PATH=%s", conf
->rootfs
.path
);
845 if (conf
->console
.path
) {
846 ret
= setenv("LXC_CONSOLE", conf
->console
.path
, 1);
848 SYSERROR("Failed to set environment variable: "
849 "LXC_CONSOLE=%s", conf
->console
.path
);
852 if (conf
->console
.log_path
) {
853 ret
= setenv("LXC_CONSOLE_LOGPATH", conf
->console
.log_path
, 1);
855 SYSERROR("Failed to set environment variable: "
856 "LXC_CONSOLE_LOGPATH=%s", conf
->console
.log_path
);
859 if (cgns_supported()) {
860 ret
= setenv("LXC_CGNS_AWARE", "1", 1);
862 SYSERROR("Failed to set environment variable "
866 loglevel
= lxc_log_priority_to_string(lxc_log_get_level());
867 ret
= setenv("LXC_LOG_LEVEL", loglevel
, 1);
869 SYSERROR("Set environment variable LXC_LOG_LEVEL=%s",
872 if (conf
->hooks_version
== 0)
873 ret
= setenv("LXC_HOOK_VERSION", "0", 1);
875 ret
= setenv("LXC_HOOK_VERSION", "1", 1);
877 SYSERROR("Failed to set environment variable LXC_HOOK_VERSION=%u", conf
->hooks_version
);
878 /* End of environment variable setup for hooks. */
880 TRACE("Set environment variables");
882 ret
= run_lxc_hooks(name
, "pre-start", conf
, NULL
);
884 ERROR("Failed to run lxc.hook.pre-start for container \"%s\"", name
);
887 TRACE("Ran pre-start hooks");
889 /* The signal fd has to be created before forking otherwise if the child
890 * process exits before we setup the signal fd, the event will be lost
891 * and the command will be stuck.
893 handler
->sigfd
= setup_signal_fd(&handler
->oldmask
);
894 if (handler
->sigfd
< 0) {
895 ERROR("Failed to setup SIGCHLD fd handler.");
898 TRACE("Set up signal fd");
900 /* Do this after setting up signals since it might unblock SIGWINCH. */
901 ret
= lxc_terminal_setup(conf
);
903 ERROR("Failed to create console");
904 goto out_restore_sigmask
;
906 TRACE("Created console");
908 ret
= lxc_terminal_map_ids(conf
, &conf
->console
);
910 ERROR("Failed to chown console");
911 goto out_delete_terminal
;
913 TRACE("Chowned console");
915 handler
->cgroup_ops
= cgroup_init(handler
->conf
);
916 if (!handler
->cgroup_ops
) {
917 ERROR("Failed to initialize cgroup driver");
918 goto out_delete_terminal
;
920 TRACE("Initialized cgroup driver");
922 ret
= lsm_process_prepare(conf
, handler
->lxcpath
);
924 ERROR("Failed to initialize LSM");
925 goto out_destroy_cgroups
;
927 TRACE("Initialized LSM");
929 INFO("Container \"%s\" is initialized", name
);
933 handler
->cgroup_ops
->payload_destroy(handler
->cgroup_ops
, handler
);
934 handler
->cgroup_ops
->monitor_destroy(handler
->cgroup_ops
, handler
);
937 lxc_terminal_delete(&handler
->conf
->console
);
940 (void)pthread_sigmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
943 lxc_delete_tty(&conf
->ttys
);
946 (void)lxc_set_state(name
, handler
, ABORTING
);
948 out_close_maincmd_fd
:
949 lxc_abstract_unix_close(conf
->maincmd_fd
);
950 conf
->maincmd_fd
= -1;
954 void lxc_fini(const char *name
, struct lxc_handler
*handler
)
958 struct lxc_list
*cur
, *next
;
959 char *namespaces
[LXC_NS_MAX
+ 1];
960 size_t namespace_count
= 0;
961 struct cgroup_ops
*cgroup_ops
= handler
->cgroup_ops
;
963 /* The STOPPING state is there for future cleanup code which can take
966 lxc_set_state(name
, handler
, STOPPING
);
968 self
= lxc_raw_getpid();
969 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
970 if (handler
->nsfd
[i
] < 0)
973 if (handler
->conf
->hooks_version
== 0)
974 ret
= asprintf(&namespaces
[namespace_count
],
975 "%s:/proc/%d/fd/%d", ns_info
[i
].proc_name
,
976 self
, handler
->nsfd
[i
]);
978 ret
= asprintf(&namespaces
[namespace_count
],
979 "/proc/%d/fd/%d", self
, handler
->nsfd
[i
]);
981 SYSERROR("Failed to allocate memory");
985 if (handler
->conf
->hooks_version
== 0) {
990 ret
= setenv(ns_info
[i
].env_name
, namespaces
[namespace_count
], 1);
992 SYSERROR("Failed to set environment variable %s=%s",
993 ns_info
[i
].env_name
, namespaces
[namespace_count
]);
995 TRACE("Set environment variable %s=%s",
996 ns_info
[i
].env_name
, namespaces
[namespace_count
]);
1000 namespaces
[namespace_count
] = NULL
;
1002 if (handler
->conf
->reboot
> REBOOT_NONE
) {
1003 ret
= setenv("LXC_TARGET", "reboot", 1);
1005 SYSERROR("Failed to set environment variable: "
1006 "LXC_TARGET=reboot");
1009 if (handler
->conf
->reboot
== REBOOT_NONE
) {
1010 ret
= setenv("LXC_TARGET", "stop", 1);
1012 SYSERROR("Failed to set environment variable: "
1016 if (handler
->conf
->hooks_version
== 0)
1017 ret
= run_lxc_hooks(name
, "stop", handler
->conf
, namespaces
);
1019 ret
= run_lxc_hooks(name
, "stop", handler
->conf
, NULL
);
1021 ERROR("Failed to run \"lxc.hook.stop\" hook");
1023 while (namespace_count
--)
1024 free(namespaces
[namespace_count
]);
1026 lsm_process_cleanup(handler
->conf
, handler
->lxcpath
);
1028 cgroup_ops
->payload_destroy(cgroup_ops
, handler
);
1029 cgroup_ops
->monitor_destroy(cgroup_ops
, handler
);
1031 if (handler
->conf
->reboot
== REBOOT_NONE
) {
1032 /* For all new state clients simply close the command socket.
1033 * This will inform all state clients that the container is
1034 * STOPPED and also prevents a race between a open()/close() on
1035 * the command socket causing a new process to get ECONNREFUSED
1036 * because we haven't yet closed the command socket.
1038 lxc_abstract_unix_close(handler
->conf
->maincmd_fd
);
1039 handler
->conf
->maincmd_fd
= -1;
1040 TRACE("Closed command socket");
1042 /* This function will try to connect to the legacy lxc-monitord
1043 * state server and only exists for backwards compatibility.
1045 lxc_monitor_send_state(name
, STOPPED
, handler
->lxcpath
);
1047 /* The command socket is closed so no one can acces the command
1048 * socket anymore so there's no need to lock it.
1050 handler
->state
= STOPPED
;
1051 TRACE("Set container state to \"STOPPED\"");
1053 lxc_set_state(name
, handler
, STOPPED
);
1056 /* Avoid lingering namespace references. */
1057 lxc_put_nsfds(handler
);
1059 ret
= run_lxc_hooks(name
, "post-stop", handler
->conf
, NULL
);
1061 ERROR("Failed to run lxc.hook.post-stop for container \"%s\"", name
);
1062 if (handler
->conf
->reboot
> REBOOT_NONE
) {
1063 WARN("Container will be stopped instead of rebooted");
1064 handler
->conf
->reboot
= REBOOT_NONE
;
1066 ret
= setenv("LXC_TARGET", "stop", 1);
1068 WARN("Failed to set environment variable: "
1073 /* Reset mask set by setup_signal_fd. */
1074 ret
= pthread_sigmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
1076 SYSWARN("Failed to restore signal mask");
1078 lxc_terminal_delete(&handler
->conf
->console
);
1079 lxc_delete_tty(&handler
->conf
->ttys
);
1081 /* The command socket is now closed, no more state clients can register
1082 * themselves from now on. So free the list of state clients.
1084 lxc_list_for_each_safe(cur
, &handler
->conf
->state_clients
, next
) {
1085 struct lxc_state_client
*client
= cur
->elem
;
1087 /* Keep state clients that want to be notified about reboots. */
1088 if ((handler
->conf
->reboot
> REBOOT_NONE
) &&
1089 (client
->states
[RUNNING
] == 2))
1092 /* close state client socket */
1094 close(client
->clientfd
);
1099 if (handler
->conf
->ephemeral
== 1 && handler
->conf
->reboot
!= REBOOT_REQ
)
1100 lxc_destroy_container_on_signal(handler
, name
);
1102 lxc_free_handler(handler
);
1105 void lxc_abort(const char *name
, struct lxc_handler
*handler
)
1109 lxc_set_state(name
, handler
, ABORTING
);
1111 if (handler
->pid
> 0) {
1112 ret
= kill(handler
->pid
, SIGKILL
);
1114 SYSERROR("Failed to send SIGKILL to %d", handler
->pid
);
1117 while ((ret
= waitpid(-1, &status
, 0)) > 0) {
1122 static int do_start(void *data
)
1124 struct lxc_handler
*handler
= data
;
1125 ATTR_UNUSED __do_close_prot_errno
int data_sock0
= handler
->data_sock
[0],
1126 data_sock1
= handler
->data_sock
[1];
1128 char path
[PATH_MAX
];
1131 struct lxc_list
*iterator
;
1134 int devnull_fd
= -1;
1136 lxc_sync_fini_parent(handler
);
1138 /* This prctl must be before the synchro, so if the parent dies before
1139 * we set the parent death signal, we will detect its death with the
1140 * synchro right after, otherwise we have a window where the parent can
1141 * exit before we set the pdeath signal leading to a unsupervized
1144 ret
= lxc_set_death_signal(SIGKILL
, handler
->monitor_pid
);
1146 SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL");
1147 goto out_warn_father
;
1150 ret
= lxc_ambient_caps_up();
1152 ERROR("Failed to raise ambient capabilities");
1153 goto out_warn_father
;
1156 ret
= pthread_sigmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
1158 SYSERROR("Failed to set signal mask");
1159 goto out_warn_father
;
1162 /* Don't leak the pinfd to the container. */
1163 if (handler
->pinfd
>= 0)
1164 close(handler
->pinfd
);
1166 ret
= lxc_sync_wait_parent(handler
, LXC_SYNC_STARTUP
);
1168 goto out_warn_father
;
1170 /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See
1171 * https://github.com/lxc/lxd/issues/1978.
1173 if ((handler
->ns_clone_flags
& (CLONE_NEWNET
| CLONE_NEWUSER
)) ==
1174 (CLONE_NEWNET
| CLONE_NEWUSER
)) {
1175 ret
= unshare(CLONE_NEWNET
);
1177 SYSERROR("Failed to unshare CLONE_NEWNET");
1178 goto out_warn_father
;
1180 INFO("Unshared CLONE_NEWNET");
1183 /* Tell the parent task it can begin to configure the container and wait
1186 ret
= lxc_sync_barrier_parent(handler
, LXC_SYNC_CONFIGURE
);
1190 ret
= lxc_network_recv_veth_names_from_parent(handler
);
1192 ERROR("Failed to receive veth names from parent");
1193 goto out_warn_father
;
1196 /* If we are in a new user namespace, become root there to have
1197 * privilege over our namespace.
1199 if (!lxc_list_empty(&handler
->conf
->id_map
)) {
1200 if (!handler
->conf
->root_nsuid_map
)
1201 nsuid
= handler
->conf
->init_uid
;
1203 if (!handler
->conf
->root_nsgid_map
)
1204 nsgid
= handler
->conf
->init_gid
;
1206 if (!lxc_switch_uid_gid(nsuid
, nsgid
))
1207 goto out_warn_father
;
1209 /* Drop groups only after we switched to a valid gid in the new
1212 if (!lxc_setgroups(0, NULL
) &&
1213 (handler
->am_root
|| errno
!= EPERM
))
1214 goto out_warn_father
;
1216 ret
= prctl(PR_SET_DUMPABLE
, prctl_arg(1), prctl_arg(0),
1217 prctl_arg(0), prctl_arg(0));
1219 goto out_warn_father
;
1221 /* set{g,u}id() clears deathsignal */
1222 ret
= lxc_set_death_signal(SIGKILL
, handler
->monitor_pid
);
1224 SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL");
1225 goto out_warn_father
;
1229 ret
= access(handler
->lxcpath
, X_OK
);
1231 print_top_failing_dir(handler
->lxcpath
);
1232 goto out_warn_father
;
1235 ret
= snprintf(path
, sizeof(path
), "%s/dev/null",
1236 handler
->conf
->rootfs
.mount
);
1237 if (ret
< 0 || ret
>= sizeof(path
))
1238 goto out_warn_father
;
1240 /* In order to checkpoint restore, we need to have everything in the
1241 * same mount namespace. However, some containers may not have a
1242 * reasonable /dev (in particular, they may not have /dev/null), so we
1243 * can't set init's std fds to /dev/null by opening it from inside the
1246 * If that's the case, fall back to using the host's /dev/null. This
1247 * means that migration won't work, but at least we won't spew output
1248 * where it isn't wanted.
1250 if (handler
->daemonize
&& !handler
->conf
->autodev
) {
1251 ret
= access(path
, F_OK
);
1253 devnull_fd
= open_devnull();
1256 goto out_warn_father
;
1257 WARN("Using /dev/null from the host for container "
1258 "init's standard file descriptors. Migration will "
1263 /* Ask father to setup cgroups and wait for him to finish. */
1264 ret
= lxc_sync_barrier_parent(handler
, LXC_SYNC_CGROUP
);
1268 /* Unshare cgroup namespace after we have setup our cgroups. If we do it
1269 * earlier we end up with a wrong view of /proc/self/cgroup. For
1270 * example, assume we unshare(CLONE_NEWCGROUP) first, and then create
1271 * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then
1272 * /proc/self/cgroup would show us:
1276 * whereas it should actually show
1280 if (handler
->ns_clone_flags
& CLONE_NEWCGROUP
) {
1281 ret
= unshare(CLONE_NEWCGROUP
);
1283 if (errno
!= EINVAL
) {
1284 SYSERROR("Failed to unshare CLONE_NEWCGROUP");
1285 goto out_warn_father
;
1288 handler
->ns_clone_flags
&= ~CLONE_NEWCGROUP
;
1289 SYSINFO("Kernel does not support CLONE_NEWCGROUP");
1291 INFO("Unshared CLONE_NEWCGROUP");
1295 /* Add the requested environment variables to the current environment to
1296 * allow them to be used by the various hooks, such as the start hook
1299 lxc_list_for_each(iterator
, &handler
->conf
->environment
) {
1300 ret
= putenv((char *)iterator
->elem
);
1302 SYSERROR("Failed to set environment variable: %s",
1303 (char *)iterator
->elem
);
1304 goto out_warn_father
;
1308 /* Setup the container, ip, names, utsname, ... */
1309 ret
= lxc_setup(handler
);
1311 ERROR("Failed to setup container \"%s\"", handler
->name
);
1312 goto out_warn_father
;
1315 /* Set the label to change to when we exec(2) the container's init. */
1316 ret
= lsm_process_label_set(NULL
, handler
->conf
, true);
1318 goto out_warn_father
;
1320 /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it
1321 * before we aren't allowed anymore.
1323 if (handler
->conf
->no_new_privs
) {
1324 ret
= prctl(PR_SET_NO_NEW_PRIVS
, prctl_arg(1), prctl_arg(0),
1325 prctl_arg(0), prctl_arg(0));
1327 SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block "
1328 "execve() gainable privileges");
1329 goto out_warn_father
;
1331 DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable "
1335 /* Some init's such as busybox will set sane tty settings on stdin,
1336 * stdout, stderr which it thinks is the console. We already set them
1337 * the way we wanted on the real terminal, and we want init to do its
1338 * setup on its console ie. the pty allocated in lxc_terminal_setup() so
1339 * make sure that that pty is stdin,stdout,stderr.
1341 if (handler
->conf
->console
.slave
>= 0) {
1342 if (handler
->daemonize
|| !handler
->conf
->is_execute
)
1343 ret
= set_stdfds(handler
->conf
->console
.slave
);
1345 ret
= lxc_terminal_set_stdfds(handler
->conf
->console
.slave
);
1347 ERROR("Failed to redirect std{in,out,err} to pty file "
1348 "descriptor %d", handler
->conf
->console
.slave
);
1349 goto out_warn_father
;
1353 /* If we mounted a temporary proc, then unmount it now. */
1354 tmp_proc_unmount(handler
->conf
);
1356 ret
= lxc_seccomp_load(handler
->conf
);
1358 goto out_warn_father
;
1360 #if HAVE_DECL_SECCOMP_NOTIF_GET_FD
1361 if (handler
->conf
->has_seccomp_notify
) {
1362 ret
= lxc_abstract_unix_send_fds(data_sock0
,
1363 &handler
->conf
->seccomp_notify_fd
,
1366 SYSERROR("Failed to send seccomp notify fd to parent");
1367 goto out_warn_father
;
1369 close(handler
->conf
->seccomp_notify_fd
);
1370 handler
->conf
->seccomp_notify_fd
= -EBADF
;
1374 ret
= run_lxc_hooks(handler
->name
, "start", handler
->conf
, NULL
);
1376 ERROR("Failed to run lxc.hook.start for container \"%s\"",
1378 goto out_warn_father
;
1381 close(handler
->sigfd
);
1383 if (handler
->conf
->console
.slave
< 0 && handler
->daemonize
) {
1384 if (devnull_fd
< 0) {
1385 devnull_fd
= open_devnull();
1387 goto out_warn_father
;
1390 ret
= set_stdfds(devnull_fd
);
1392 ERROR("Failed to redirect std{in,out,err} to \"/dev/null\"");
1393 goto out_warn_father
;
1397 if (devnull_fd
>= 0) {
1404 if (handler
->conf
->init_cwd
) {
1405 ret
= chdir(handler
->conf
->init_cwd
);
1407 SYSERROR("Could not change directory to \"%s\"",
1408 handler
->conf
->init_cwd
);
1409 goto out_warn_father
;
1413 ret
= lxc_sync_barrier_parent(handler
, LXC_SYNC_CGROUP_LIMITS
);
1415 goto out_warn_father
;
1417 /* Reset the environment variables the user requested in a clear
1421 /* Don't error out though. */
1423 SYSERROR("Failed to clear environment.");
1425 lxc_list_for_each(iterator
, &handler
->conf
->environment
) {
1426 ret
= putenv((char *)iterator
->elem
);
1428 SYSERROR("Failed to set environment variable: %s",
1429 (char *)iterator
->elem
);
1430 goto out_warn_father
;
1434 ret
= putenv("container=lxc");
1436 SYSERROR("Failed to set environment variable: container=lxc");
1437 goto out_warn_father
;
1440 if (handler
->conf
->ttys
.tty_names
) {
1441 ret
= putenv(handler
->conf
->ttys
.tty_names
);
1443 SYSERROR("Failed to set environment variable for container ptys");
1444 goto out_warn_father
;
1448 /* The container has been setup. We can now switch to an unprivileged
1451 new_uid
= handler
->conf
->init_uid
;
1452 new_gid
= handler
->conf
->init_gid
;
1454 /* Avoid unnecessary syscalls. */
1455 if (new_uid
== nsuid
)
1456 new_uid
= LXC_INVALID_UID
;
1458 if (new_gid
== nsgid
)
1459 new_gid
= LXC_INVALID_GID
;
1461 if (!lxc_switch_uid_gid(new_uid
, new_gid
))
1462 goto out_warn_father
;
1464 /* If we are in a new user namespace we already dropped all groups when
1465 * we switched to root in the new user namespace further above. Only
1466 * drop groups if we can, so ensure that we have necessary privilege.
1468 if (lxc_list_empty(&handler
->conf
->id_map
))
1470 if (lxc_proc_cap_is_set(CAP_SETGID
, CAP_EFFECTIVE
))
1472 if (!lxc_setgroups(0, NULL
))
1473 goto out_warn_father
;
1475 ret
= lxc_ambient_caps_down();
1477 ERROR("Failed to clear ambient capabilities");
1478 goto out_warn_father
;
1481 if (handler
->conf
->monitor_signal_pdeath
!= SIGKILL
) {
1482 ret
= lxc_set_death_signal(handler
->conf
->monitor_signal_pdeath
, handler
->monitor_pid
);
1484 SYSERROR("Failed to set PR_SET_PDEATHSIG to %d",
1485 handler
->conf
->monitor_signal_pdeath
);
1486 goto out_warn_father
;
1490 /* After this call, we are in error because this ops should not return
1493 handler
->ops
->start(handler
, handler
->data
);
1496 /* We want the parent to know something went wrong, so we return a
1497 * special error code.
1499 lxc_sync_wake_parent(handler
, LXC_SYNC_ERROR
);
1502 if (devnull_fd
>= 0)
1508 static int lxc_recv_ttys_from_child(struct lxc_handler
*handler
)
1511 struct lxc_terminal_info
*tty
;
1513 int sock
= handler
->data_sock
[1];
1514 struct lxc_conf
*conf
= handler
->conf
;
1515 struct lxc_tty_info
*ttys
= &conf
->ttys
;
1517 if (!conf
->ttys
.max
)
1520 ttys
->tty
= malloc(sizeof(*ttys
->tty
) * ttys
->max
);
1524 for (i
= 0; i
< conf
->ttys
.max
; i
++) {
1527 ret
= lxc_abstract_unix_recv_fds(sock
, ttyfds
, 2, NULL
, 0);
1531 tty
= &ttys
->tty
[i
];
1533 tty
->master
= ttyfds
[0];
1534 tty
->slave
= ttyfds
[1];
1535 TRACE("Received pty with master fd %d and slave fd %d from "
1536 "parent", tty
->master
, tty
->slave
);
1540 SYSERROR("Failed to receive %zu ttys from child", ttys
->max
);
1542 TRACE("Received %zu ttys from child", ttys
->max
);
1547 int resolve_clone_flags(struct lxc_handler
*handler
)
1550 struct lxc_conf
*conf
= handler
->conf
;
1552 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1553 if (conf
->ns_keep
) {
1554 if (!(conf
->ns_keep
& ns_info
[i
].clone_flag
))
1555 handler
->ns_clone_flags
|= ns_info
[i
].clone_flag
;
1556 } else if (conf
->ns_clone
) {
1557 if ((conf
->ns_clone
& ns_info
[i
].clone_flag
))
1558 handler
->ns_clone_flags
|= ns_info
[i
].clone_flag
;
1560 if (i
== LXC_NS_USER
&& lxc_list_empty(&handler
->conf
->id_map
))
1563 if (i
== LXC_NS_NET
&& lxc_requests_empty_network(handler
))
1566 if (i
== LXC_NS_CGROUP
&& !cgns_supported())
1569 handler
->ns_clone_flags
|= ns_info
[i
].clone_flag
;
1572 if (!conf
->ns_share
[i
])
1575 handler
->ns_clone_flags
&= ~ns_info
[i
].clone_flag
;
1576 TRACE("Sharing %s namespace", ns_info
[i
].proc_name
);
1582 /* Note that this function is used with clone(CLONE_VM). Some glibc versions
1583 * used to reset the pid/tid to -1 when CLONE_VM was used without CLONE_THREAD.
1584 * But since the memory between parent and child is shared on CLONE_VM this
1585 * would invalidate the getpid() cache that glibc used to maintain and so
1586 * getpid() in the child would return the parent's pid. This is all fixed in
1587 * newer glibc versions where the getpid() cache is removed and the pid/tid is
1588 * not reset anymore.
1589 * However, if for whatever reason you - dear committer - somehow need to get the
1590 * pid of the dummy intermediate process for do_share_ns() you need to call
1591 * lxc_raw_getpid(). The next lxc_raw_clone() call does not employ CLONE_VM and
1594 static inline int do_share_ns(void *arg
)
1597 struct lxc_handler
*handler
= arg
;
1599 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1600 if (handler
->nsfd
[i
] < 0)
1603 ret
= setns(handler
->nsfd
[i
], 0);
1606 * Note that joining a user and/or mount namespace
1607 * requires the process is not multithreaded otherwise
1608 * setns() will fail here.
1610 SYSERROR("Failed to inherit %s namespace",
1611 ns_info
[i
].proc_name
);
1615 DEBUG("Inherited %s namespace", ns_info
[i
].proc_name
);
1618 flags
= handler
->ns_on_clone_flags
;
1619 flags
|= CLONE_PARENT
;
1620 handler
->pid
= lxc_raw_clone_cb(do_start
, handler
, flags
);
1621 if (handler
->pid
< 0)
1627 /* lxc_spawn() performs crucial setup tasks and clone()s the new process which
1628 * exec()s the requested container binary.
1629 * Note that lxc_spawn() runs in the parent namespaces. Any operations performed
1630 * right here should be double checked if they'd pose a security risk. (For
1631 * example, any {u}mount() operations performed here will be reflected on the
1634 static int lxc_spawn(struct lxc_handler
*handler
)
1636 __do_close_prot_errno
int data_sock0
= -EBADF
, data_sock1
= -EBADF
;
1639 bool wants_to_map_ids
;
1640 struct lxc_list
*id_map
;
1641 const char *name
= handler
->name
;
1642 const char *lxcpath
= handler
->lxcpath
;
1643 bool share_ns
= false;
1644 struct lxc_conf
*conf
= handler
->conf
;
1645 struct cgroup_ops
*cgroup_ops
= handler
->cgroup_ops
;
1647 id_map
= &conf
->id_map
;
1648 wants_to_map_ids
= !lxc_list_empty(id_map
);
1650 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1651 if (!conf
->ns_share
[i
])
1654 handler
->nsfd
[i
] = lxc_inherit_namespace(conf
->ns_share
[i
], lxcpath
, ns_info
[i
].proc_name
);
1655 if (handler
->nsfd
[i
] < 0)
1661 ret
= lxc_sync_init(handler
);
1665 ret
= socketpair(AF_UNIX
, SOCK_STREAM
| SOCK_CLOEXEC
, 0,
1666 handler
->data_sock
);
1669 data_sock0
= handler
->data_sock
[0];
1670 data_sock1
= handler
->data_sock
[1];
1672 ret
= resolve_clone_flags(handler
);
1676 if (handler
->ns_clone_flags
& CLONE_NEWNET
) {
1677 if (!lxc_list_empty(&conf
->network
)) {
1679 /* Find gateway addresses from the link device, which is
1680 * no longer accessible inside the container. Do this
1681 * before creating network interfaces, since goto
1682 * out_delete_net does not work before lxc_clone.
1684 ret
= lxc_find_gateway_addresses(handler
);
1686 ERROR("Failed to find gateway addresses");
1690 /* That should be done before the clone because we will
1691 * fill the netdev index and use them in the child.
1693 ret
= lxc_create_network_priv(handler
);
1695 ERROR("Failed to create the network");
1696 goto out_delete_net
;
1701 if (!cgroup_ops
->payload_create(cgroup_ops
, handler
)) {
1702 ERROR("Failed creating cgroups");
1703 goto out_delete_net
;
1706 /* If the rootfs is not a blockdev, prevent the container from marking
1708 * If the container is unprivileged then skip rootfs pinning.
1710 if (!wants_to_map_ids
) {
1711 handler
->pinfd
= pin_rootfs(conf
->rootfs
.path
);
1712 if (handler
->pinfd
== -1)
1713 INFO("Failed to pin the rootfs for container \"%s\"", handler
->name
);
1716 /* Create a process in a new set of namespaces. */
1717 handler
->ns_on_clone_flags
= handler
->ns_clone_flags
;
1718 if (handler
->ns_clone_flags
& CLONE_NEWUSER
) {
1719 /* If CLONE_NEWUSER and CLONE_NEWNET was requested, we need to
1720 * clone a new user namespace first and only later unshare our
1721 * network namespace to ensure that network devices ownership is
1724 handler
->ns_on_clone_flags
&= ~CLONE_NEWNET
;
1726 /* The cgroup namespace gets unshare()ed not clone()ed. */
1727 handler
->ns_on_clone_flags
&= ~CLONE_NEWCGROUP
;
1732 attacher_pid
= lxc_clone(do_share_ns
, handler
,
1733 CLONE_VFORK
| CLONE_VM
| CLONE_FILES
);
1734 if (attacher_pid
< 0) {
1735 SYSERROR(LXC_CLONE_ERROR
);
1736 goto out_delete_net
;
1739 ret
= wait_for_pid(attacher_pid
);
1741 SYSERROR("Intermediate process failed");
1742 goto out_delete_net
;
1745 handler
->pid
= lxc_raw_clone_cb(do_start
, handler
,
1746 handler
->ns_on_clone_flags
);
1748 if (handler
->pid
< 0) {
1749 SYSERROR(LXC_CLONE_ERROR
);
1750 goto out_delete_net
;
1752 TRACE("Cloned child process %d", handler
->pid
);
1754 for (i
= 0; i
< LXC_NS_MAX
; i
++)
1755 if (handler
->ns_on_clone_flags
& ns_info
[i
].clone_flag
)
1756 INFO("Cloned %s", ns_info
[i
].flag_name
);
1758 if (!lxc_try_preserve_namespaces(handler
, handler
->ns_on_clone_flags
, handler
->pid
)) {
1759 ERROR("Failed to preserve cloned namespaces for lxc.hook.stop");
1760 goto out_delete_net
;
1763 lxc_sync_fini_child(handler
);
1765 /* Map the container uids. The container became an invalid userid the
1766 * moment it was cloned with CLONE_NEWUSER. This call doesn't change
1767 * anything immediately, but allows the container to setuid(0) (0 being
1768 * mapped to something else on the host.) later to become a valid uid
1771 if (wants_to_map_ids
) {
1772 if (!handler
->conf
->ns_share
[LXC_NS_USER
] &&
1773 (handler
->conf
->ns_keep
& CLONE_NEWUSER
) == 0) {
1774 ret
= lxc_map_ids(id_map
, handler
->pid
);
1776 ERROR("Failed to set up id mapping.");
1777 goto out_delete_net
;
1782 ret
= lxc_sync_wake_child(handler
, LXC_SYNC_STARTUP
);
1784 goto out_delete_net
;
1786 ret
= lxc_sync_wait_child(handler
, LXC_SYNC_CONFIGURE
);
1788 goto out_delete_net
;
1790 if (!cgroup_ops
->setup_limits(cgroup_ops
, handler
->conf
, false)) {
1791 ERROR("Failed to setup cgroup limits for container \"%s\"", name
);
1792 goto out_delete_net
;
1795 if (!cgroup_ops
->payload_enter(cgroup_ops
, handler
->pid
))
1796 goto out_delete_net
;
1798 if (!cgroup_ops
->chown(cgroup_ops
, handler
->conf
))
1799 goto out_delete_net
;
1801 /* Now we're ready to preserve the network namespace */
1802 ret
= lxc_try_preserve_ns(handler
->pid
, "net");
1804 if (ret
!= -EOPNOTSUPP
) {
1805 SYSERROR("Failed to preserve net namespace");
1806 goto out_delete_net
;
1809 handler
->nsfd
[LXC_NS_NET
] = ret
;
1810 DEBUG("Preserved net namespace via fd %d", ret
);
1812 ret
= lxc_netns_set_nsid(handler
->nsfd
[LXC_NS_NET
]);
1814 SYSWARN("Failed to allocate new network namespace id");
1816 TRACE("Allocated new network namespace id");
1819 /* Create the network configuration. */
1820 if (handler
->ns_clone_flags
& CLONE_NEWNET
) {
1821 ret
= lxc_network_move_created_netdev_priv(handler
->lxcpath
,
1826 ERROR("Failed to create the configured network");
1827 goto out_delete_net
;
1830 ret
= lxc_create_network_unpriv(handler
->lxcpath
, handler
->name
,
1831 &conf
->network
, handler
->pid
, conf
->hooks_version
);
1833 ERROR("Failed to create the configured network");
1834 goto out_delete_net
;
1838 ret
= lxc_network_send_veth_names_to_child(handler
);
1840 ERROR("Failed to send veth names to child");
1841 goto out_delete_net
;
1844 if (!lxc_list_empty(&conf
->procs
)) {
1845 ret
= setup_proc_filesystem(&conf
->procs
, handler
->pid
);
1847 goto out_delete_net
;
1850 /* Tell the child to continue its initialization. We'll get
1851 * LXC_SYNC_CGROUP when it is ready for us to setup cgroups.
1853 ret
= lxc_sync_barrier_child(handler
, LXC_SYNC_POST_CONFIGURE
);
1855 goto out_delete_net
;
1857 if (!lxc_list_empty(&conf
->limits
)) {
1858 ret
= setup_resource_limits(&conf
->limits
, handler
->pid
);
1860 ERROR("Failed to setup resource limits");
1861 goto out_delete_net
;
1865 ret
= lxc_sync_barrier_child(handler
, LXC_SYNC_CGROUP_UNSHARE
);
1867 goto out_delete_net
;
1869 if (!cgroup_ops
->setup_limits(cgroup_ops
, handler
->conf
, true)) {
1870 ERROR("Failed to setup legacy device cgroup controller limits");
1871 goto out_delete_net
;
1873 TRACE("Set up legacy device cgroup controller limits");
1875 if (handler
->ns_clone_flags
& CLONE_NEWCGROUP
) {
1876 /* Now we're ready to preserve the cgroup namespace */
1877 ret
= lxc_try_preserve_ns(handler
->pid
, "cgroup");
1879 if (ret
!= -EOPNOTSUPP
) {
1880 SYSERROR("Failed to preserve cgroup namespace");
1881 goto out_delete_net
;
1884 handler
->nsfd
[LXC_NS_CGROUP
] = ret
;
1885 DEBUG("Preserved cgroup namespace via fd %d", ret
);
1889 ret
= snprintf(pidstr
, 20, "%d", handler
->pid
);
1890 if (ret
< 0 || ret
>= 20)
1891 goto out_delete_net
;
1893 ret
= setenv("LXC_PID", pidstr
, 1);
1895 SYSERROR("Failed to set environment variable: LXC_PID=%s", pidstr
);
1897 /* Run any host-side start hooks */
1898 ret
= run_lxc_hooks(name
, "start-host", conf
, NULL
);
1900 ERROR("Failed to run lxc.hook.start-host");
1901 goto out_delete_net
;
1904 /* Tell the child to complete its initialization and wait for it to exec
1905 * or return an error. (The child will never return
1906 * LXC_SYNC_READY_START+1. It will either close the sync pipe, causing
1907 * lxc_sync_barrier_child to return success, or return a different
1908 * value, causing us to error out).
1910 ret
= lxc_sync_barrier_child(handler
, LXC_SYNC_READY_START
);
1912 goto out_delete_net
;
1914 ret
= lxc_network_recv_name_and_ifindex_from_child(handler
);
1916 ERROR("Failed to receive names and ifindices for network "
1917 "devices from child");
1918 goto out_delete_net
;
1921 /* Now all networks are created, network devices are moved into place,
1922 * and the correct names and ifindices in the respective namespaces have
1923 * been recorded. The corresponding structs have now all been filled. So
1924 * log them for debugging purposes.
1926 lxc_log_configured_netdevs(conf
);
1928 /* Read tty fds allocated by child. */
1929 ret
= lxc_recv_ttys_from_child(handler
);
1931 ERROR("Failed to receive tty info from child process");
1932 goto out_delete_net
;
1935 #if HAVE_DECL_SECCOMP_NOTIF_GET_FD
1936 if (handler
->conf
->has_seccomp_notify
) {
1937 ret
= lxc_abstract_unix_recv_fds(handler
->data_sock
[1],
1938 &handler
->conf
->seccomp_notify_fd
,
1941 SYSERROR("Failed to receive seccomp notify fd from child");
1942 goto out_delete_net
;
1945 ret
= seccomp_notif_alloc(&handler
->conf
->seccomp_notify_req
,
1946 &handler
->conf
->seccomp_notify_resp
);
1950 goto out_delete_net
;
1955 ret
= handler
->ops
->post_start(handler
, handler
->data
);
1959 ret
= lxc_set_state(name
, handler
, RUNNING
);
1961 ERROR("Failed to set state to \"%s\"", lxc_state2str(RUNNING
));
1965 lxc_sync_fini(handler
);
1970 if (handler
->ns_clone_flags
& CLONE_NEWNET
)
1971 lxc_delete_network(handler
);
1974 lxc_abort(name
, handler
);
1977 lxc_sync_fini(handler
);
1978 if (handler
->pinfd
>= 0) {
1979 close(handler
->pinfd
);
1980 handler
->pinfd
= -1;
1986 int __lxc_start(const char *name
, struct lxc_handler
*handler
,
1987 struct lxc_operations
* ops
, void *data
, const char *lxcpath
,
1988 bool daemonize
, int *error_num
)
1991 struct lxc_conf
*conf
= handler
->conf
;
1992 struct cgroup_ops
*cgroup_ops
;
1994 ret
= lxc_init(name
, handler
);
1996 ERROR("Failed to initialize container \"%s\"", name
);
2000 handler
->data
= data
;
2001 handler
->daemonize
= daemonize
;
2002 cgroup_ops
= handler
->cgroup_ops
;
2004 if (!attach_block_device(handler
->conf
)) {
2005 ERROR("Failed to attach block device");
2007 goto out_fini_nonet
;
2010 if (!cgroup_ops
->monitor_create(cgroup_ops
, handler
)) {
2011 ERROR("Failed to create monitor cgroup");
2013 goto out_fini_nonet
;
2016 if (!cgroup_ops
->monitor_enter(cgroup_ops
, handler
->monitor_pid
)) {
2017 ERROR("Failed to enter monitor cgroup");
2019 goto out_fini_nonet
;
2022 if (geteuid() == 0 && !lxc_list_empty(&conf
->id_map
)) {
2023 /* If the backing store is a device, mount it here and now. */
2024 if (rootfs_is_blockdev(conf
)) {
2025 ret
= unshare(CLONE_NEWNS
);
2027 ERROR("Failed to unshare CLONE_NEWNS");
2028 goto out_fini_nonet
;
2030 INFO("Unshared CLONE_NEWNS");
2032 remount_all_slave();
2033 ret
= lxc_setup_rootfs_prepare_root(conf
, name
, lxcpath
);
2035 ERROR("Error setting up rootfs mount as root before spawn");
2036 goto out_fini_nonet
;
2038 INFO("Set up container rootfs as host root");
2042 ret
= lxc_spawn(handler
);
2044 ERROR("Failed to spawn container \"%s\"", name
);
2045 goto out_detach_blockdev
;
2048 handler
->conf
->reboot
= REBOOT_NONE
;
2050 ret
= lxc_poll(name
, handler
);
2052 ERROR("LXC mainloop exited with error: %d", ret
);
2056 if (!handler
->init_died
&& handler
->pid
> 0) {
2057 ERROR("Child process is not killed");
2062 status
= lxc_wait_for_pid_status(handler
->pid
);
2064 SYSERROR("Failed to retrieve status for %d", handler
->pid
);
2066 /* If the child process exited but was not signaled, it didn't call
2067 * reboot. This should mean it was an lxc-execute which simply exited.
2068 * In any case, treat it as a 'halt'.
2070 if (WIFSIGNALED(status
)) {
2071 switch(WTERMSIG(status
)) {
2072 case SIGINT
: /* halt */
2073 DEBUG("Container \"%s\" is halting", name
);
2075 case SIGHUP
: /* reboot */
2076 DEBUG("Container \"%s\" is rebooting", name
);
2077 handler
->conf
->reboot
= REBOOT_REQ
;
2079 case SIGSYS
: /* seccomp */
2080 DEBUG("Container \"%s\" violated its seccomp policy", name
);
2083 DEBUG("Unknown exit status for container \"%s\" init %d", name
, WTERMSIG(status
));
2088 ret
= lxc_restore_phys_nics_to_netns(handler
);
2090 ERROR("Failed to move physical network devices back to parent "
2091 "network namespace");
2093 if (handler
->pinfd
>= 0) {
2094 close(handler
->pinfd
);
2095 handler
->pinfd
= -1;
2098 lxc_monitor_send_exit_code(name
, status
, handler
->lxcpath
);
2099 lxc_error_set_and_log(handler
->pid
, status
);
2101 *error_num
= handler
->exit_status
;
2104 lxc_delete_network(handler
);
2106 out_detach_blockdev
:
2107 detach_block_device(handler
->conf
);
2110 lxc_fini(name
, handler
);
2114 lxc_abort(name
, handler
);
2122 static int start(struct lxc_handler
*handler
, void* data
)
2124 struct start_args
*arg
= data
;
2126 NOTICE("Exec'ing \"%s\"", arg
->argv
[0]);
2128 execvp(arg
->argv
[0], arg
->argv
);
2129 SYSERROR("Failed to exec \"%s\"", arg
->argv
[0]);
2133 static int post_start(struct lxc_handler
*handler
, void* data
)
2135 struct start_args
*arg
= data
;
2137 NOTICE("Started \"%s\" with pid \"%d\"", arg
->argv
[0], handler
->pid
);
2141 static struct lxc_operations start_ops
= {
2143 .post_start
= post_start
2146 int lxc_start(const char *name
, char *const argv
[], struct lxc_handler
*handler
,
2147 const char *lxcpath
, bool daemonize
, int *error_num
)
2149 struct start_args start_arg
= {
2153 TRACE("Doing lxc_start");
2154 return __lxc_start(name
, handler
, &start_ops
, &start_arg
, lxcpath
, daemonize
, error_num
);
2157 static void lxc_destroy_container_on_signal(struct lxc_handler
*handler
,
2160 char destroy
[PATH_MAX
];
2161 struct lxc_container
*c
;
2165 if (handler
->conf
->rootfs
.path
&& handler
->conf
->rootfs
.mount
) {
2166 bret
= do_destroy_container(handler
);
2168 ERROR("Error destroying rootfs for container \"%s\"", name
);
2172 INFO("Destroyed rootfs for container \"%s\"", name
);
2174 ret
= snprintf(destroy
, PATH_MAX
, "%s/%s", handler
->lxcpath
, name
);
2175 if (ret
< 0 || ret
>= PATH_MAX
) {
2176 ERROR("Error destroying directory for container \"%s\"", name
);
2180 c
= lxc_container_new(name
, handler
->lxcpath
);
2182 if (container_disk_lock(c
)) {
2183 INFO("Could not update lxc_snapshots file");
2184 lxc_container_put(c
);
2186 mod_all_rdeps(c
, false);
2187 container_disk_unlock(c
);
2188 lxc_container_put(c
);
2192 if (!handler
->am_root
)
2193 ret
= userns_exec_full(handler
->conf
, lxc_rmdir_onedev_wrapper
,
2194 destroy
, "lxc_rmdir_onedev_wrapper");
2196 ret
= lxc_rmdir_onedev(destroy
, NULL
);
2199 ERROR("Error destroying directory for container \"%s\"", name
);
2202 INFO("Destroyed directory for container \"%s\"", name
);
2205 static int lxc_rmdir_onedev_wrapper(void *data
)
2207 char *arg
= (char *) data
;
2208 return lxc_rmdir_onedev(arg
, NULL
);
2211 static bool do_destroy_container(struct lxc_handler
*handler
)
2215 if (!handler
->am_root
) {
2216 ret
= userns_exec_full(handler
->conf
, storage_destroy_wrapper
,
2217 handler
->conf
, "storage_destroy_wrapper");
2224 return storage_destroy(handler
->conf
);