1 /* SPDX-License-Identifier: LGPL-2.1+ */
17 #include <sys/mount.h>
18 #include <sys/param.h>
19 #include <sys/prctl.h>
20 #include <sys/socket.h>
22 #include <sys/syscall.h>
23 #include <sys/types.h>
32 #include "commands_utils.h"
36 #include "confile_utils.h"
38 #include "file_utils.h"
42 #include "lxccontainer.h"
44 #include "lxcseccomp.h"
47 #include "memory_utils.h"
49 #include "namespace.h"
51 #include "process_utils.h"
53 #include "storage/storage.h"
54 #include "storage/storage_utils.h"
56 #include "syscall_wrappers.h"
61 #include <sys/capability.h>
65 #include "include/strlcpy.h"
68 lxc_log_define(start
, lxc
);
70 extern void mod_all_rdeps(struct lxc_container
*c
, bool inc
);
71 static bool do_destroy_container(struct lxc_handler
*handler
);
72 static int lxc_rmdir_onedev_wrapper(void *data
);
73 static void lxc_destroy_container_on_signal(struct lxc_handler
*handler
,
76 static void print_top_failing_dir(const char *path
)
78 __do_free
char *copy
= NULL
;
82 copy
= must_copy_string(path
);
84 e
= copy
+ strlen(path
);
87 while (p
< e
&& *p
== '/')
90 while (p
< e
&& *p
!= '/')
96 ret
= access(copy
, X_OK
);
98 SYSERROR("Could not access %s. Please grant it x access, or add an ACL for the container " "root", copy
);
105 static void lxc_put_nsfds(struct lxc_handler
*handler
)
107 for (int i
= 0; i
< LXC_NS_MAX
; i
++) {
108 if (handler
->nsfd
[i
] < 0)
111 close_prot_errno_disarm(handler
->nsfd
[i
]);
115 static int lxc_try_preserve_ns(const int pid
, const char *ns
)
119 fd
= lxc_preserve_ns(pid
, ns
);
122 return log_error_errno(-EINVAL
,
123 errno
, "Failed to preserve %s namespace",
126 return log_warn_errno(-EOPNOTSUPP
,
127 errno
, "Kernel does not support preserving %s namespaces",
134 /* lxc_try_preserve_namespaces: open /proc/@pid/ns/@ns for each namespace
135 * specified in ns_clone_flags.
136 * Return true on success, false on failure.
138 static bool lxc_try_preserve_namespaces(struct lxc_handler
*handler
,
139 int ns_clone_flags
, pid_t pid
)
143 for (i
= 0; i
< LXC_NS_MAX
; i
++)
144 handler
->nsfd
[i
] = -EBADF
;
146 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
149 if ((ns_clone_flags
& ns_info
[i
].clone_flag
) == 0)
152 fd
= lxc_try_preserve_ns(pid
, ns_info
[i
].proc_name
);
154 /* Do not fail to start container on kernels that do
155 * not support interacting with namespaces through
158 if (fd
== -EOPNOTSUPP
)
161 lxc_put_nsfds(handler
);
165 handler
->nsfd
[i
] = fd
;
166 DEBUG("Preserved %s namespace via fd %d", ns_info
[i
].proc_name
,
173 static inline bool match_stdfds(int fd
)
175 return (fd
== STDIN_FILENO
|| fd
== STDOUT_FILENO
|| fd
== STDERR_FILENO
);
179 static bool match_dlog_fds(struct dirent
*direntp
)
181 char path
[PATH_MAX
] = {0};
182 char link
[PATH_MAX
] = {0};
186 ret
= snprintf(path
, PATH_MAX
, "/proc/self/fd/%s", direntp
->d_name
);
187 if (ret
< 0 || ret
>= PATH_MAX
)
188 return log_error(false, "Failed to create file descriptor name");
190 linklen
= readlink(path
, link
, PATH_MAX
);
192 return log_error(false, "Failed to read link path - \"%s\"", path
);
193 else if (linklen
>= PATH_MAX
)
194 return log_error(false, "The name of link path is too long - \"%s\"", path
);
196 if (strcmp(link
, "/dev/log_main") == 0 ||
197 strcmp(link
, "/dev/log_system") == 0 ||
198 strcmp(link
, "/dev/log_radio") == 0)
205 int lxc_check_inherited(struct lxc_conf
*conf
, bool closeall
,
206 int *fds_to_ignore
, size_t len_fds
)
211 struct dirent
*direntp
;
213 if (conf
&& conf
->close_all_fds
)
217 * Disable syslog at this point to avoid the above logging
218 * function to open a new fd and make the check_inherited function
219 * enter an infinite loop.
221 lxc_log_syslog_disable();
224 dir
= opendir("/proc/self/fd");
226 return log_warn(-1, "Failed to open directory");
230 while ((direntp
= readdir(dir
))) {
232 struct lxc_list
*cur
;
233 bool matched
= false;
235 if (strcmp(direntp
->d_name
, ".") == 0)
238 if (strcmp(direntp
->d_name
, "..") == 0)
241 ret
= lxc_safe_int(direntp
->d_name
, &fd
);
243 INFO("Could not parse file descriptor for \"%s\"", direntp
->d_name
);
247 for (i
= 0; i
< len_fds
; i
++)
248 if (fds_to_ignore
[i
] == fd
)
251 if (fd
== fddir
|| fd
== lxc_log_fd
||
252 (i
< len_fds
&& fd
== fds_to_ignore
[i
]))
255 /* Keep state clients that wait on reboots. */
257 lxc_list_for_each(cur
, &conf
->state_clients
) {
258 struct lxc_state_client
*client
= cur
->elem
;
260 if (client
->clientfd
!= fd
)
271 if (current_config
&& fd
== current_config
->logfd
)
274 if (match_stdfds(fd
))
278 if (match_dlog_fds(direntp
))
284 SYSINFO("Closed inherited fd %d", fd
);
286 INFO("Closed inherited fd %d", fd
);
290 WARN("Inherited fd %d", fd
);
295 * Only enable syslog at this point to avoid the above logging
296 * function to open a new fd and make the check_inherited function
297 * enter an infinite loop.
299 lxc_log_syslog_enable();
304 static int setup_signal_fd(sigset_t
*oldmask
)
308 const int signals
[] = {SIGBUS
, SIGILL
, SIGSEGV
, SIGWINCH
};
310 /* Block everything except serious error signals. */
311 ret
= sigfillset(&mask
);
315 for (int sig
= 0; sig
< (sizeof(signals
) / sizeof(signals
[0])); sig
++) {
316 ret
= sigdelset(&mask
, signals
[sig
]);
321 ret
= pthread_sigmask(SIG_BLOCK
, &mask
, oldmask
);
323 return log_error_errno(-EBADF
, errno
,
324 "Failed to set signal mask");
326 ret
= signalfd(-1, &mask
, SFD_CLOEXEC
);
328 return log_error_errno(-EBADF
,
329 errno
, "Failed to create signal file descriptor");
331 TRACE("Created signal file descriptor %d", ret
);
336 static int signal_handler(int fd
, uint32_t events
, void *data
,
337 struct lxc_epoll_descr
*descr
)
341 struct signalfd_siginfo siginfo
;
342 struct lxc_handler
*hdlr
= data
;
344 ret
= lxc_read_nointr(fd
, &siginfo
, sizeof(siginfo
));
346 return log_error(LXC_MAINLOOP_ERROR
, "Failed to read signal info from signal file descriptor %d", fd
);
348 if (ret
!= sizeof(siginfo
))
349 return log_error(LXC_MAINLOOP_ERROR
, "Unexpected size for struct signalfd_siginfo");
351 /* Check whether init is running. */
353 ret
= waitid(P_PID
, hdlr
->pid
, &info
, WEXITED
| WNOWAIT
| WNOHANG
);
354 if (ret
== 0 && info
.si_pid
== hdlr
->pid
)
355 hdlr
->init_died
= true;
357 /* Try to figure out a reasonable exit status to report. */
358 if (hdlr
->init_died
) {
359 switch (info
.si_code
) {
361 hdlr
->exit_status
= info
.si_status
<< 8;
366 hdlr
->exit_status
= info
.si_status
<< 8 | 0x7f;
369 /* Huh? The waitid() told us it's dead *and* continued? */
370 WARN("Init %d dead and continued?", hdlr
->pid
);
371 hdlr
->exit_status
= 1;
374 ERROR("Unknown si_code: %d", info
.si_code
);
375 hdlr
->exit_status
= 1;
379 if (siginfo
.ssi_signo
== SIGHUP
) {
380 if (hdlr
->pidfd
>= 0)
381 lxc_raw_pidfd_send_signal(hdlr
->pidfd
, SIGTERM
, NULL
, 0);
383 kill(hdlr
->pid
, SIGTERM
);
384 INFO("Killing %d since terminal hung up", hdlr
->pid
);
385 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
386 : LXC_MAINLOOP_CONTINUE
;
389 if (siginfo
.ssi_signo
!= SIGCHLD
) {
390 if (hdlr
->pidfd
>= 0)
391 lxc_raw_pidfd_send_signal(hdlr
->pidfd
,
392 siginfo
.ssi_signo
, NULL
, 0);
394 kill(hdlr
->pid
, siginfo
.ssi_signo
);
395 INFO("Forwarded signal %d to pid %d", siginfo
.ssi_signo
, hdlr
->pid
);
396 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
397 : LXC_MAINLOOP_CONTINUE
;
400 /* More robustness, protect ourself from a SIGCHLD sent
401 * by a process different from the container init.
403 if (siginfo
.ssi_pid
!= hdlr
->pid
) {
404 NOTICE("Received %d from pid %d instead of container init %d",
405 siginfo
.ssi_signo
, siginfo
.ssi_pid
, hdlr
->pid
);
406 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
407 : LXC_MAINLOOP_CONTINUE
;
410 if (siginfo
.ssi_code
== CLD_STOPPED
) {
411 INFO("Container init process was stopped");
412 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
413 : LXC_MAINLOOP_CONTINUE
;
416 if (siginfo
.ssi_code
== CLD_CONTINUED
) {
417 INFO("Container init process was continued");
418 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
419 : LXC_MAINLOOP_CONTINUE
;
422 return log_debug(LXC_MAINLOOP_CLOSE
, "Container init process %d exited", hdlr
->pid
);
425 int lxc_serve_state_clients(const char *name
, struct lxc_handler
*handler
,
430 struct lxc_list
*cur
, *next
;
431 struct lxc_msg msg
= {.type
= lxc_msg_state
, .value
= state
};
434 handler
->state
= RUNNING
;
436 handler
->state
= state
;
438 TRACE("Set container state to %s", lxc_state2str(state
));
440 if (lxc_list_empty(&handler
->conf
->state_clients
))
441 return log_trace(0, "No state clients registered");
443 retlen
= strlcpy(msg
.name
, name
, sizeof(msg
.name
));
444 if (retlen
>= sizeof(msg
.name
))
447 lxc_list_for_each_safe(cur
, &handler
->conf
->state_clients
, next
) {
448 struct lxc_state_client
*client
= cur
->elem
;
450 if (client
->states
[state
] == 0) {
451 TRACE("State %s not registered for state client %d",
452 lxc_state2str(state
), client
->clientfd
);
456 TRACE("Sending state %s to state client %d",
457 lxc_state2str(state
), client
->clientfd
);
459 ret
= lxc_send_nointr(client
->clientfd
, &msg
, sizeof(msg
), MSG_NOSIGNAL
);
461 SYSERROR("Failed to send message to client");
463 /* kick client from list */
465 close(client
->clientfd
);
473 static int lxc_serve_state_socket_pair(const char *name
,
474 struct lxc_handler
*handler
,
479 if (!handler
->daemonize
||
480 handler
->state_socket_pair
[1] < 0 ||
484 /* Close read end of the socket pair. */
485 close_prot_errno_disarm(handler
->state_socket_pair
[0]);
488 ret
= lxc_abstract_unix_send_credential(handler
->state_socket_pair
[1],
489 &(int){state
}, sizeof(int));
491 SYSERROR("Failed to send state to %d", handler
->state_socket_pair
[1]);
499 if (ret
!= sizeof(int))
500 return log_error(-1, "Message too long : %d", handler
->state_socket_pair
[1]);
502 TRACE("Sent container state \"%s\" to %d", lxc_state2str(state
),
503 handler
->state_socket_pair
[1]);
505 /* Close write end of the socket pair. */
506 close_prot_errno_disarm(handler
->state_socket_pair
[1]);
511 int lxc_set_state(const char *name
, struct lxc_handler
*handler
,
516 ret
= lxc_serve_state_socket_pair(name
, handler
, state
);
518 return log_error(-1, "Failed to synchronize via anonymous pair of unix sockets");
520 ret
= lxc_serve_state_clients(name
, handler
, state
);
524 /* This function will try to connect to the legacy lxc-monitord state
525 * server and only exists for backwards compatibility.
527 lxc_monitor_send_state(name
, state
, handler
->lxcpath
);
532 int lxc_poll(const char *name
, struct lxc_handler
*handler
)
535 bool has_console
= true;
536 struct lxc_epoll_descr descr
, descr_console
;
538 if (handler
->conf
->console
.path
&&
539 strcmp(handler
->conf
->console
.path
, "none") == 0)
542 ret
= lxc_mainloop_open(&descr
);
544 ERROR("Failed to create mainloop");
549 ret
= lxc_mainloop_open(&descr_console
);
551 ERROR("Failed to create console mainloop");
556 ret
= lxc_mainloop_add_handler(&descr
, handler
->sigfd
, signal_handler
, handler
);
558 ERROR("Failed to add signal handler for %d to mainloop", handler
->sigfd
);
559 goto out_mainloop_console
;
562 ret
= lxc_seccomp_setup_proxy(&handler
->conf
->seccomp
, &descr
, handler
);
564 ERROR("Failed to setup seccomp proxy");
565 goto out_mainloop_console
;
569 struct lxc_terminal
*console
= &handler
->conf
->console
;
571 ret
= lxc_terminal_mainloop_add(&descr
, console
);
573 ERROR("Failed to add console handlers to mainloop");
574 goto out_mainloop_console
;
577 ret
= lxc_terminal_mainloop_add(&descr_console
, console
);
579 ERROR("Failed to add console handlers to console mainloop");
580 goto out_mainloop_console
;
583 handler
->conf
->console
.descr
= &descr
;
586 ret
= lxc_cmd_mainloop_add(name
, &descr
, handler
);
588 ERROR("Failed to add command handler to mainloop");
589 goto out_mainloop_console
;
592 TRACE("Mainloop is ready");
594 ret
= lxc_mainloop(&descr
, -1);
595 close_prot_errno_disarm(descr
.epfd
);
596 if (ret
< 0 || !handler
->init_died
)
597 goto out_mainloop_console
;
600 ret
= lxc_mainloop(&descr_console
, 0);
602 out_mainloop_console
:
604 lxc_mainloop_close(&descr_console
);
605 TRACE("Closed console mainloop");
609 lxc_mainloop_close(&descr
);
610 TRACE("Closed mainloop");
613 TRACE("Closed signal file descriptor %d", handler
->sigfd
);
614 close_prot_errno_disarm(handler
->sigfd
);
619 void lxc_put_handler(struct lxc_handler
*handler
)
621 close_prot_errno_disarm(handler
->pinfd
);
622 close_prot_errno_disarm(handler
->pidfd
);
623 close_prot_errno_disarm(handler
->sigfd
);
624 lxc_put_nsfds(handler
);
625 if (handler
->conf
&& handler
->conf
->reboot
== REBOOT_NONE
)
626 close_prot_errno_disarm(handler
->conf
->maincmd_fd
);
627 close_prot_errno_disarm(handler
->monitor_status_fd
);
628 close_prot_errno_disarm(handler
->state_socket_pair
[0]);
629 close_prot_errno_disarm(handler
->state_socket_pair
[1]);
630 cgroup_exit(handler
->cgroup_ops
);
631 if (handler
->conf
&& handler
->conf
->reboot
== REBOOT_NONE
)
632 free_disarm(handler
);
634 handler
->conf
= NULL
;
637 struct lxc_handler
*lxc_init_handler(struct lxc_handler
*old
,
638 const char *name
, struct lxc_conf
*conf
,
639 const char *lxcpath
, bool daemonize
)
643 struct lxc_handler
*handler
;
646 handler
= zalloc(sizeof(*handler
));
652 /* Note that am_guest_unpriv() checks the effective uid. We
653 * probably don't care if we are real root only if we are running
654 * as root so this should be fine.
656 handler
->am_root
= !am_guest_unpriv();
657 handler
->conf
= conf
;
658 handler
->lxcpath
= lxcpath
;
659 handler
->init_died
= false;
660 handler
->data_sock
[0] = -EBADF
;
661 handler
->data_sock
[1] = -EBADF
;
662 handler
->monitor_status_fd
= -EBADF
;
663 handler
->pinfd
= -EBADF
;
664 handler
->pidfd
= -EBADF
;
665 handler
->sigfd
= -EBADF
;
666 handler
->state_socket_pair
[0] = -EBADF
;
667 handler
->state_socket_pair
[1] = -EBADF
;
668 if (handler
->conf
->reboot
== REBOOT_NONE
)
669 lxc_list_init(&handler
->conf
->state_clients
);
671 for (int i
= 0; i
< LXC_NS_MAX
; i
++)
672 handler
->nsfd
[i
] = -EBADF
;
674 handler
->name
= name
;
676 handler
->transient_pid
= lxc_raw_getpid();
678 handler
->transient_pid
= -1;
680 if (daemonize
&& handler
->conf
->reboot
== REBOOT_NONE
) {
681 /* Create socketpair() to synchronize on daemonized startup.
682 * When the container reboots we don't need to synchronize
683 * again currently so don't open another socketpair().
685 ret
= socketpair(AF_UNIX
, SOCK_STREAM
| SOCK_CLOEXEC
, 0,
686 handler
->state_socket_pair
);
688 ERROR("Failed to create anonymous pair of unix sockets");
692 TRACE("Created anonymous pair {%d,%d} of unix sockets",
693 handler
->state_socket_pair
[0],
694 handler
->state_socket_pair
[1]);
695 handler
->keep_fds
[nr_keep_fds
++] = handler
->state_socket_pair
[0];
696 handler
->keep_fds
[nr_keep_fds
++] = handler
->state_socket_pair
[1];
699 if (handler
->conf
->reboot
== REBOOT_NONE
) {
700 handler
->conf
->maincmd_fd
= lxc_cmd_init(name
, lxcpath
, "command");
701 if (handler
->conf
->maincmd_fd
< 0) {
702 ERROR("Failed to set up command socket");
705 handler
->keep_fds
[nr_keep_fds
++] = handler
->conf
->maincmd_fd
;
708 TRACE("Unix domain socket %d for command server is ready",
709 handler
->conf
->maincmd_fd
);
714 lxc_put_handler(handler
);
719 int lxc_init(const char *name
, struct lxc_handler
*handler
)
721 __do_close
int status_fd
= -EBADF
;
723 const char *loglevel
;
724 struct lxc_conf
*conf
= handler
->conf
;
726 handler
->monitor_pid
= lxc_raw_getpid();
727 status_fd
= open("/proc/self/status", O_RDONLY
| O_CLOEXEC
);
729 return log_error_errno(-1, errno
, "Failed to open monitor status fd");
731 handler
->lsm_ops
= lsm_init();
732 TRACE("Initialized LSM");
734 /* Begin by setting the state to STARTING. */
735 ret
= lxc_set_state(name
, handler
, STARTING
);
737 return log_error(-1, "Failed to set state to \"%s\"", lxc_state2str(STARTING
));
738 TRACE("Set container state to \"STARTING\"");
740 /* Start of environment variable setup for hooks. */
741 ret
= setenv("LXC_NAME", name
, 1);
743 SYSERROR("Failed to set environment variable: LXC_NAME=%s", name
);
746 ret
= setenv("LXC_CONFIG_FILE", conf
->rcfile
, 1);
748 SYSERROR("Failed to set environment variable: LXC_CONFIG_FILE=%s", conf
->rcfile
);
751 if (conf
->rootfs
.mount
) {
752 ret
= setenv("LXC_ROOTFS_MOUNT", conf
->rootfs
.mount
, 1);
754 SYSERROR("Failed to set environment variable: LXC_ROOTFS_MOUNT=%s", conf
->rootfs
.mount
);
757 if (conf
->rootfs
.path
) {
758 ret
= setenv("LXC_ROOTFS_PATH", conf
->rootfs
.path
, 1);
760 SYSERROR("Failed to set environment variable: LXC_ROOTFS_PATH=%s", conf
->rootfs
.path
);
763 if (conf
->console
.path
) {
764 ret
= setenv("LXC_CONSOLE", conf
->console
.path
, 1);
766 SYSERROR("Failed to set environment variable: LXC_CONSOLE=%s", conf
->console
.path
);
769 if (conf
->console
.log_path
) {
770 ret
= setenv("LXC_CONSOLE_LOGPATH", conf
->console
.log_path
, 1);
772 SYSERROR("Failed to set environment variable: LXC_CONSOLE_LOGPATH=%s", conf
->console
.log_path
);
775 if (cgns_supported()) {
776 ret
= setenv("LXC_CGNS_AWARE", "1", 1);
778 SYSERROR("Failed to set environment variable LXC_CGNS_AWARE=1");
781 loglevel
= lxc_log_priority_to_string(lxc_log_get_level());
782 ret
= setenv("LXC_LOG_LEVEL", loglevel
, 1);
784 SYSERROR("Set environment variable LXC_LOG_LEVEL=%s", loglevel
);
786 if (conf
->hooks_version
== 0)
787 ret
= setenv("LXC_HOOK_VERSION", "0", 1);
789 ret
= setenv("LXC_HOOK_VERSION", "1", 1);
791 SYSERROR("Failed to set environment variable LXC_HOOK_VERSION=%u", conf
->hooks_version
);
792 /* End of environment variable setup for hooks. */
794 TRACE("Set environment variables");
796 ret
= run_lxc_hooks(name
, "pre-start", conf
, NULL
);
798 return log_error(-1, "Failed to run lxc.hook.pre-start for container \"%s\"", name
);
799 TRACE("Ran pre-start hooks");
801 /* The signal fd has to be created before forking otherwise if the child
802 * process exits before we setup the signal fd, the event will be lost
803 * and the command will be stuck.
805 handler
->sigfd
= setup_signal_fd(&handler
->oldmask
);
806 if (handler
->sigfd
< 0)
807 return log_error(-1, "Failed to setup SIGCHLD fd handler.");
808 TRACE("Set up signal fd");
810 /* Do this after setting up signals since it might unblock SIGWINCH. */
811 ret
= lxc_terminal_setup(conf
);
813 ERROR("Failed to create console");
814 goto out_restore_sigmask
;
816 TRACE("Created console");
818 handler
->cgroup_ops
= cgroup_init(handler
->conf
);
819 if (!handler
->cgroup_ops
) {
820 ERROR("Failed to initialize cgroup driver");
821 goto out_delete_terminal
;
823 TRACE("Initialized cgroup driver");
825 ret
= lxc_read_seccomp_config(conf
);
827 return log_error(-1, "Failed loading seccomp policy");
828 TRACE("Read seccomp policy");
830 ret
= handler
->lsm_ops
->prepare(handler
->lsm_ops
, conf
, handler
->lxcpath
);
832 ERROR("Failed to initialize LSM");
833 goto out_delete_terminal
;
835 TRACE("Initialized LSM");
837 INFO("Container \"%s\" is initialized", name
);
838 handler
->monitor_status_fd
= move_fd(status_fd
);
842 lxc_terminal_delete(&handler
->conf
->console
);
845 (void)pthread_sigmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
850 void lxc_end(struct lxc_handler
*handler
)
854 struct lxc_list
*cur
, *next
;
855 char *namespaces
[LXC_NS_MAX
+ 1];
856 size_t namespace_count
= 0;
857 const char *name
= handler
->name
;
858 struct cgroup_ops
*cgroup_ops
= handler
->cgroup_ops
;
860 /* The STOPPING state is there for future cleanup code which can take
863 lxc_set_state(name
, handler
, STOPPING
);
865 self
= lxc_raw_getpid();
866 for (int i
= 0; i
< LXC_NS_MAX
; i
++) {
867 if (handler
->nsfd
[i
] < 0)
870 if (handler
->conf
->hooks_version
== 0)
871 ret
= asprintf(&namespaces
[namespace_count
],
872 "%s:/proc/%d/fd/%d", ns_info
[i
].proc_name
,
873 self
, handler
->nsfd
[i
]);
875 ret
= asprintf(&namespaces
[namespace_count
],
876 "/proc/%d/fd/%d", self
, handler
->nsfd
[i
]);
878 SYSERROR("Failed to allocate memory");
882 if (handler
->conf
->hooks_version
== 0) {
887 ret
= setenv(ns_info
[i
].env_name
, namespaces
[namespace_count
], 1);
889 SYSERROR("Failed to set environment variable %s=%s",
890 ns_info
[i
].env_name
, namespaces
[namespace_count
]);
892 TRACE("Set environment variable %s=%s",
893 ns_info
[i
].env_name
, namespaces
[namespace_count
]);
897 namespaces
[namespace_count
] = NULL
;
899 if (handler
->conf
->reboot
> REBOOT_NONE
) {
900 ret
= setenv("LXC_TARGET", "reboot", 1);
902 SYSERROR("Failed to set environment variable: LXC_TARGET=reboot");
905 if (handler
->conf
->reboot
== REBOOT_NONE
) {
906 ret
= setenv("LXC_TARGET", "stop", 1);
908 SYSERROR("Failed to set environment variable: LXC_TARGET=stop");
911 if (handler
->conf
->hooks_version
== 0)
912 ret
= run_lxc_hooks(name
, "stop", handler
->conf
, namespaces
);
914 ret
= run_lxc_hooks(name
, "stop", handler
->conf
, NULL
);
916 ERROR("Failed to run \"lxc.hook.stop\" hook");
918 while (namespace_count
--)
919 free(namespaces
[namespace_count
]);
921 handler
->lsm_ops
->cleanup(handler
->lsm_ops
, handler
->conf
, handler
->lxcpath
);
924 cgroup_ops
->payload_destroy(cgroup_ops
, handler
);
925 cgroup_ops
->monitor_destroy(cgroup_ops
, handler
);
928 if (handler
->conf
->reboot
== REBOOT_NONE
) {
929 /* For all new state clients simply close the command socket.
930 * This will inform all state clients that the container is
931 * STOPPED and also prevents a race between a open()/close() on
932 * the command socket causing a new process to get ECONNREFUSED
933 * because we haven't yet closed the command socket.
935 close_prot_errno_disarm(handler
->conf
->maincmd_fd
);
936 TRACE("Closed command socket");
938 /* This function will try to connect to the legacy lxc-monitord
939 * state server and only exists for backwards compatibility.
941 lxc_monitor_send_state(name
, STOPPED
, handler
->lxcpath
);
943 /* The command socket is closed so no one can acces the command
944 * socket anymore so there's no need to lock it.
946 handler
->state
= STOPPED
;
947 TRACE("Set container state to \"STOPPED\"");
949 lxc_set_state(name
, handler
, STOPPED
);
950 TRACE("Set container state to \"STOPPED\"");
953 /* Avoid lingering namespace references. */
954 lxc_put_nsfds(handler
);
956 ret
= run_lxc_hooks(name
, "post-stop", handler
->conf
, NULL
);
958 ERROR("Failed to run lxc.hook.post-stop for container \"%s\"", name
);
959 if (handler
->conf
->reboot
> REBOOT_NONE
) {
960 WARN("Container will be stopped instead of rebooted");
961 handler
->conf
->reboot
= REBOOT_NONE
;
963 ret
= setenv("LXC_TARGET", "stop", 1);
965 WARN("Failed to set environment variable: LXC_TARGET=stop");
969 /* Reset mask set by setup_signal_fd. */
970 ret
= pthread_sigmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
972 SYSWARN("Failed to restore signal mask");
974 lxc_terminal_delete(&handler
->conf
->console
);
975 lxc_delete_tty(&handler
->conf
->ttys
);
976 close_prot_errno_disarm(handler
->conf
->devpts_fd
);
978 /* The command socket is now closed, no more state clients can register
979 * themselves from now on. So free the list of state clients.
981 lxc_list_for_each_safe(cur
, &handler
->conf
->state_clients
, next
) {
982 struct lxc_state_client
*client
= cur
->elem
;
984 /* Keep state clients that want to be notified about reboots. */
985 if ((handler
->conf
->reboot
> REBOOT_NONE
) &&
986 (client
->states
[RUNNING
] == 2))
989 /* close state client socket */
991 close(client
->clientfd
);
996 if (handler
->conf
->ephemeral
== 1 && handler
->conf
->reboot
!= REBOOT_REQ
)
997 lxc_destroy_container_on_signal(handler
, name
);
999 lxc_put_handler(handler
);
1002 void lxc_abort(struct lxc_handler
*handler
)
1007 lxc_set_state(handler
->name
, handler
, ABORTING
);
1009 if (handler
->pidfd
>= 0) {
1010 ret
= lxc_raw_pidfd_send_signal(handler
->pidfd
, SIGKILL
, NULL
, 0);
1012 SYSWARN("Failed to send SIGKILL via pidfd %d for process %d",
1013 handler
->pidfd
, handler
->pid
);
1016 if ((!ret
|| errno
!= ESRCH
) && handler
->pid
> 0)
1017 if (kill(handler
->pid
, SIGKILL
))
1018 SYSWARN("Failed to send SIGKILL to %d", handler
->pid
);
1021 ret
= waitpid(-1, &status
, 0);
1025 static int do_start(void *data
)
1027 struct lxc_handler
*handler
= data
;
1028 __lxc_unused __do_close
int data_sock0
= handler
->data_sock
[0],
1029 data_sock1
= handler
->data_sock
[1];
1030 __do_close
int devnull_fd
= -EBADF
, status_fd
= -EBADF
;
1034 struct lxc_list
*iterator
;
1038 lxc_sync_fini_parent(handler
);
1040 if (lxc_abstract_unix_recv_fds(data_sock1
, &status_fd
, 1, NULL
, 0) < 0) {
1041 ERROR("Failed to receive status file descriptor to child process");
1042 goto out_warn_father
;
1045 /* This prctl must be before the synchro, so if the parent dies before
1046 * we set the parent death signal, we will detect its death with the
1047 * synchro right after, otherwise we have a window where the parent can
1048 * exit before we set the pdeath signal leading to a unsupervized
1051 ret
= lxc_set_death_signal(SIGKILL
, handler
->monitor_pid
, status_fd
);
1053 SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL");
1054 goto out_warn_father
;
1057 ret
= lxc_ambient_caps_up();
1059 ERROR("Failed to raise ambient capabilities");
1060 goto out_warn_father
;
1063 ret
= pthread_sigmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
1065 SYSERROR("Failed to set signal mask");
1066 goto out_warn_father
;
1069 /* Don't leak the pinfd to the container. */
1070 close_prot_errno_disarm(handler
->pinfd
);
1072 ret
= lxc_sync_wait_parent(handler
, LXC_SYNC_STARTUP
);
1074 goto out_warn_father
;
1076 /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See
1077 * https://github.com/lxc/lxd/issues/1978.
1079 if (handler
->ns_unshare_flags
& CLONE_NEWNET
) {
1080 ret
= unshare(CLONE_NEWNET
);
1082 SYSERROR("Failed to unshare CLONE_NEWNET");
1083 goto out_warn_father
;
1085 INFO("Unshared CLONE_NEWNET");
1088 /* Tell the parent task it can begin to configure the container and wait
1091 ret
= lxc_sync_barrier_parent(handler
, LXC_SYNC_CONFIGURE
);
1095 if (handler
->ns_clone_flags
& CLONE_NEWNET
) {
1096 ret
= lxc_network_recv_from_parent(handler
);
1098 ERROR("Failed to receive veth names from parent");
1099 goto out_warn_father
;
1103 /* If we are in a new user namespace, become root there to have
1104 * privilege over our namespace.
1106 if (!lxc_list_empty(&handler
->conf
->id_map
)) {
1107 if (!handler
->conf
->root_nsuid_map
)
1108 nsuid
= handler
->conf
->init_uid
;
1110 if (!handler
->conf
->root_nsgid_map
)
1111 nsgid
= handler
->conf
->init_gid
;
1113 /* Drop groups only after we switched to a valid gid in the new
1116 if (!lxc_setgroups(0, NULL
) &&
1117 (handler
->am_root
|| errno
!= EPERM
))
1118 goto out_warn_father
;
1120 if (!lxc_switch_uid_gid(nsuid
, nsgid
))
1121 goto out_warn_father
;
1123 ret
= prctl(PR_SET_DUMPABLE
, prctl_arg(1), prctl_arg(0),
1124 prctl_arg(0), prctl_arg(0));
1126 goto out_warn_father
;
1128 /* set{g,u}id() clears deathsignal */
1129 ret
= lxc_set_death_signal(SIGKILL
, handler
->monitor_pid
, status_fd
);
1131 SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL");
1132 goto out_warn_father
;
1136 ret
= access(handler
->lxcpath
, X_OK
);
1138 print_top_failing_dir(handler
->lxcpath
);
1139 goto out_warn_father
;
1142 /* In order to checkpoint restore, we need to have everything in the
1143 * same mount namespace. However, some containers may not have a
1144 * reasonable /dev (in particular, they may not have /dev/null), so we
1145 * can't set init's std fds to /dev/null by opening it from inside the
1148 * If that's the case, fall back to using the host's /dev/null. This
1149 * means that migration won't work, but at least we won't spew output
1150 * where it isn't wanted.
1152 if (handler
->daemonize
&& !handler
->conf
->autodev
) {
1153 char path
[PATH_MAX
];
1155 ret
= snprintf(path
, sizeof(path
), "%s/dev/null",
1156 handler
->conf
->rootfs
.mount
);
1157 if (ret
< 0 || ret
>= sizeof(path
))
1158 goto out_warn_father
;
1160 ret
= access(path
, F_OK
);
1162 devnull_fd
= open_devnull();
1165 goto out_warn_father
;
1166 WARN("Using /dev/null from the host for container init's standard file descriptors. Migration will not work");
1170 /* Ask father to setup cgroups and wait for him to finish. */
1171 ret
= lxc_sync_barrier_parent(handler
, LXC_SYNC_CGROUP
);
1175 /* Unshare cgroup namespace after we have setup our cgroups. If we do it
1176 * earlier we end up with a wrong view of /proc/self/cgroup. For
1177 * example, assume we unshare(CLONE_NEWCGROUP) first, and then create
1178 * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then
1179 * /proc/self/cgroup would show us:
1183 * whereas it should actually show
1187 if (handler
->ns_unshare_flags
& CLONE_NEWCGROUP
) {
1188 ret
= unshare(CLONE_NEWCGROUP
);
1190 if (errno
!= EINVAL
) {
1191 SYSERROR("Failed to unshare CLONE_NEWCGROUP");
1192 goto out_warn_father
;
1195 handler
->ns_clone_flags
&= ~CLONE_NEWCGROUP
;
1196 SYSINFO("Kernel does not support CLONE_NEWCGROUP");
1198 INFO("Unshared CLONE_NEWCGROUP");
1202 if (handler
->ns_unshare_flags
& CLONE_NEWTIME
) {
1203 ret
= unshare(CLONE_NEWTIME
);
1205 if (errno
!= EINVAL
) {
1206 SYSERROR("Failed to unshare CLONE_NEWTIME");
1207 goto out_warn_father
;
1210 handler
->ns_clone_flags
&= ~CLONE_NEWTIME
;
1211 SYSINFO("Kernel does not support CLONE_NEWTIME");
1213 __do_close
int timens_fd
= -EBADF
;
1215 INFO("Unshared CLONE_NEWTIME");
1217 if (handler
->conf
->timens
.s_boot
)
1218 ret
= timens_offset_write(CLOCK_BOOTTIME
, handler
->conf
->timens
.s_boot
, 0);
1219 else if (handler
->conf
->timens
.ns_boot
)
1220 ret
= timens_offset_write(CLOCK_BOOTTIME
, 0, handler
->conf
->timens
.ns_boot
);
1222 SYSERROR("Failed to write CLONE_BOOTTIME offset");
1223 goto out_warn_father
;
1225 TRACE("Wrote CLOCK_BOOTTIME offset");
1227 if (handler
->conf
->timens
.s_monotonic
)
1228 ret
= timens_offset_write(CLOCK_MONOTONIC
, handler
->conf
->timens
.s_monotonic
, 0);
1229 else if (handler
->conf
->timens
.ns_monotonic
)
1230 ret
= timens_offset_write(CLOCK_MONOTONIC
, 0, handler
->conf
->timens
.ns_monotonic
);
1232 SYSERROR("Failed to write CLONE_MONOTONIC offset");
1233 goto out_warn_father
;
1235 TRACE("Wrote CLOCK_MONOTONIC offset");
1237 timens_fd
= open("/proc/self/ns/time_for_children", O_RDONLY
| O_CLOEXEC
);
1238 if (timens_fd
< 0) {
1239 SYSERROR("Failed to open \"/proc/self/ns/time_for_children\"");
1240 goto out_warn_father
;
1243 ret
= setns(timens_fd
, CLONE_NEWTIME
);
1245 SYSERROR("Failed to setns(%d(\"/proc/self/ns/time_for_children\"))", timens_fd
);
1246 goto out_warn_father
;
1251 /* Add the requested environment variables to the current environment to
1252 * allow them to be used by the various hooks, such as the start hook
1255 lxc_list_for_each(iterator
, &handler
->conf
->environment
) {
1256 ret
= putenv((char *)iterator
->elem
);
1258 SYSERROR("Failed to set environment variable: %s",
1259 (char *)iterator
->elem
);
1260 goto out_warn_father
;
1264 /* Setup the container, ip, names, utsname, ... */
1265 ret
= lxc_setup(handler
);
1267 ERROR("Failed to setup container \"%s\"", handler
->name
);
1268 goto out_warn_father
;
1271 /* Set the label to change to when we exec(2) the container's init. */
1272 ret
= handler
->lsm_ops
->process_label_set(handler
->lsm_ops
, NULL
, handler
->conf
, true);
1274 goto out_warn_father
;
1276 /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it
1277 * before we aren't allowed anymore.
1279 if (handler
->conf
->no_new_privs
) {
1280 ret
= prctl(PR_SET_NO_NEW_PRIVS
, prctl_arg(1), prctl_arg(0),
1281 prctl_arg(0), prctl_arg(0));
1283 SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges");
1284 goto out_warn_father
;
1286 DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges");
1289 /* Some init's such as busybox will set sane tty settings on stdin,
1290 * stdout, stderr which it thinks is the console. We already set them
1291 * the way we wanted on the real terminal, and we want init to do its
1292 * setup on its console ie. the pty allocated in lxc_terminal_setup() so
1293 * make sure that that pty is stdin,stdout,stderr.
1295 if (handler
->conf
->console
.pty
>= 0) {
1296 if (handler
->daemonize
|| !handler
->conf
->is_execute
)
1297 ret
= set_stdfds(handler
->conf
->console
.pty
);
1299 ret
= lxc_terminal_set_stdfds(handler
->conf
->console
.pty
);
1301 ERROR("Failed to redirect std{in,out,err} to pty file descriptor %d",
1302 handler
->conf
->console
.pty
);
1303 goto out_warn_father
;
1307 /* If we mounted a temporary proc, then unmount it now. */
1308 tmp_proc_unmount(handler
->conf
);
1310 ret
= lxc_seccomp_load(handler
->conf
);
1312 goto out_warn_father
;
1314 ret
= lxc_seccomp_send_notifier_fd(&handler
->conf
->seccomp
, data_sock0
);
1316 SYSERROR("Failed to send seccomp notify fd to parent");
1317 goto out_warn_father
;
1320 ret
= run_lxc_hooks(handler
->name
, "start", handler
->conf
, NULL
);
1322 ERROR("Failed to run lxc.hook.start for container \"%s\"",
1324 goto out_warn_father
;
1327 close_prot_errno_disarm(handler
->sigfd
);
1329 if (handler
->conf
->console
.pty
< 0 && handler
->daemonize
) {
1330 if (devnull_fd
< 0) {
1331 devnull_fd
= open_devnull();
1333 goto out_warn_father
;
1336 ret
= set_stdfds(devnull_fd
);
1338 ERROR("Failed to redirect std{in,out,err} to \"/dev/null\"");
1339 goto out_warn_father
;
1343 close_prot_errno_disarm(devnull_fd
);
1347 if (handler
->conf
->init_cwd
) {
1348 ret
= chdir(handler
->conf
->init_cwd
);
1350 SYSERROR("Could not change directory to \"%s\"",
1351 handler
->conf
->init_cwd
);
1352 goto out_warn_father
;
1356 ret
= lxc_sync_barrier_parent(handler
, LXC_SYNC_CGROUP_LIMITS
);
1358 goto out_warn_father
;
1360 /* Reset the environment variables the user requested in a clear
1364 /* Don't error out though. */
1366 SYSERROR("Failed to clear environment.");
1368 lxc_list_for_each(iterator
, &handler
->conf
->environment
) {
1369 ret
= putenv((char *)iterator
->elem
);
1371 SYSERROR("Failed to set environment variable: %s",
1372 (char *)iterator
->elem
);
1373 goto out_warn_father
;
1377 ret
= putenv("container=lxc");
1379 SYSERROR("Failed to set environment variable: container=lxc");
1380 goto out_warn_father
;
1383 if (handler
->conf
->ttys
.tty_names
) {
1384 ret
= putenv(handler
->conf
->ttys
.tty_names
);
1386 SYSERROR("Failed to set environment variable for container ptys");
1387 goto out_warn_father
;
1391 /* The container has been setup. We can now switch to an unprivileged
1394 new_uid
= handler
->conf
->init_uid
;
1395 new_gid
= handler
->conf
->init_gid
;
1397 /* Avoid unnecessary syscalls. */
1398 if (new_uid
== nsuid
)
1399 new_uid
= LXC_INVALID_UID
;
1401 if (new_gid
== nsgid
)
1402 new_gid
= LXC_INVALID_GID
;
1404 /* Make sure that the processes STDIO is correctly owned by the user that we are switching to */
1405 ret
= fix_stdio_permissions(new_uid
);
1407 WARN("Failed to ajust stdio permissions");
1409 /* If we are in a new user namespace we already dropped all groups when
1410 * we switched to root in the new user namespace further above. Only
1411 * drop groups if we can, so ensure that we have necessary privilege.
1413 if (lxc_list_empty(&handler
->conf
->id_map
))
1415 if (lxc_proc_cap_is_set(CAP_SETGID
, CAP_EFFECTIVE
))
1417 if (!lxc_setgroups(0, NULL
))
1418 goto out_warn_father
;
1420 if (!lxc_switch_uid_gid(new_uid
, new_gid
))
1421 goto out_warn_father
;
1423 ret
= lxc_ambient_caps_down();
1425 ERROR("Failed to clear ambient capabilities");
1426 goto out_warn_father
;
1429 if (handler
->conf
->monitor_signal_pdeath
!= SIGKILL
) {
1430 ret
= lxc_set_death_signal(handler
->conf
->monitor_signal_pdeath
,
1431 handler
->monitor_pid
, status_fd
);
1433 SYSERROR("Failed to set PR_SET_PDEATHSIG to %d",
1434 handler
->conf
->monitor_signal_pdeath
);
1435 goto out_warn_father
;
1440 * After this call, we are in error because this ops should not return
1443 handler
->ops
->start(handler
, handler
->data
);
1447 * We want the parent to know something went wrong, so we return a
1448 * special error code.
1450 lxc_sync_wake_parent(handler
, LXC_SYNC_ERROR
);
1456 static int lxc_recv_ttys_from_child(struct lxc_handler
*handler
)
1459 struct lxc_terminal_info
*tty
;
1461 int sock
= handler
->data_sock
[1];
1462 struct lxc_conf
*conf
= handler
->conf
;
1463 struct lxc_tty_info
*ttys
= &conf
->ttys
;
1465 if (!conf
->ttys
.max
)
1468 ttys
->tty
= malloc(sizeof(*ttys
->tty
) * ttys
->max
);
1472 for (i
= 0; i
< conf
->ttys
.max
; i
++) {
1475 ret
= lxc_abstract_unix_recv_fds(sock
, ttyfds
, 2, NULL
, 0);
1479 tty
= &ttys
->tty
[i
];
1481 tty
->ptx
= ttyfds
[0];
1482 tty
->pty
= ttyfds
[1];
1483 TRACE("Received pty with ptx fd %d and pty fd %d from child", tty
->ptx
, tty
->pty
);
1487 SYSERROR("Failed to receive %zu ttys from child", ttys
->max
);
1489 TRACE("Received %zu ttys from child", ttys
->max
);
1494 int resolve_clone_flags(struct lxc_handler
*handler
)
1497 struct lxc_conf
*conf
= handler
->conf
;
1498 bool wants_timens
= conf
->timens
.s_boot
|| conf
->timens
.ns_boot
||
1499 conf
->timens
.s_monotonic
|| conf
->timens
.ns_monotonic
;
1501 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1502 if (conf
->ns_keep
) {
1503 if (!(conf
->ns_keep
& ns_info
[i
].clone_flag
))
1504 handler
->ns_clone_flags
|= ns_info
[i
].clone_flag
;
1505 } else if (conf
->ns_clone
) {
1506 if ((conf
->ns_clone
& ns_info
[i
].clone_flag
))
1507 handler
->ns_clone_flags
|= ns_info
[i
].clone_flag
;
1509 if (i
== LXC_NS_USER
&& lxc_list_empty(&handler
->conf
->id_map
))
1512 if (i
== LXC_NS_NET
&& lxc_requests_empty_network(handler
))
1515 if (i
== LXC_NS_CGROUP
&& !cgns_supported())
1518 if (i
== LXC_NS_TIME
&& !wants_timens
)
1521 handler
->ns_clone_flags
|= ns_info
[i
].clone_flag
;
1524 if (!conf
->ns_share
[i
])
1527 handler
->ns_clone_flags
&= ~ns_info
[i
].clone_flag
;
1528 TRACE("Sharing %s namespace", ns_info
[i
].proc_name
);
1531 if (wants_timens
&& (conf
->ns_keep
& ns_info
[LXC_NS_TIME
].clone_flag
))
1532 return log_trace_errno(-1, EINVAL
, "Requested to keep time namespace while also specifying offsets");
1534 /* Deal with namespaces that are unshared. */
1535 if (handler
->ns_clone_flags
& CLONE_NEWTIME
)
1536 handler
->ns_unshare_flags
|= CLONE_NEWTIME
;
1538 if (!pure_unified_layout(handler
->cgroup_ops
) && handler
->ns_clone_flags
& CLONE_NEWCGROUP
)
1539 handler
->ns_unshare_flags
|= CLONE_NEWCGROUP
;
1541 if ((handler
->ns_clone_flags
& (CLONE_NEWNET
| CLONE_NEWUSER
)) ==
1542 (CLONE_NEWNET
| CLONE_NEWUSER
))
1543 handler
->ns_unshare_flags
|= CLONE_NEWNET
;
1545 /* Deal with namespaces that are spawned. */
1546 handler
->ns_on_clone_flags
= handler
->ns_clone_flags
& ~handler
->ns_unshare_flags
;
1548 handler
->clone_flags
= handler
->ns_on_clone_flags
| CLONE_PIDFD
;
1553 /* Note that this function is used with clone(CLONE_VM). Some glibc versions
1554 * used to reset the pid/tid to -1 when CLONE_VM was used without CLONE_THREAD.
1555 * But since the memory between parent and child is shared on CLONE_VM this
1556 * would invalidate the getpid() cache that glibc used to maintain and so
1557 * getpid() in the child would return the parent's pid. This is all fixed in
1558 * newer glibc versions where the getpid() cache is removed and the pid/tid is
1559 * not reset anymore.
1560 * However, if for whatever reason you - dear committer - somehow need to get the
1561 * pid of the dummy intermediate process for do_share_ns() you need to call
1562 * lxc_raw_getpid(). The next lxc_raw_clone() call does not employ CLONE_VM and
1565 static inline int do_share_ns(void *arg
)
1568 struct lxc_handler
*handler
= arg
;
1570 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1571 if (handler
->nsfd
[i
] < 0)
1574 ret
= setns(handler
->nsfd
[i
], 0);
1577 * Note that joining a user and/or mount namespace
1578 * requires the process is not multithreaded otherwise
1579 * setns() will fail here.
1581 SYSERROR("Failed to inherit %s namespace",
1582 ns_info
[i
].proc_name
);
1586 DEBUG("Inherited %s namespace", ns_info
[i
].proc_name
);
1589 flags
= handler
->ns_on_clone_flags
;
1590 flags
|= CLONE_PARENT
;
1591 handler
->pid
= lxc_raw_clone_cb(do_start
, handler
, CLONE_PIDFD
| flags
,
1593 if (handler
->pid
< 0)
1599 /* lxc_spawn() performs crucial setup tasks and clone()s the new process which
1600 * exec()s the requested container binary.
1601 * Note that lxc_spawn() runs in the parent namespaces. Any operations performed
1602 * right here should be double checked if they'd pose a security risk. (For
1603 * example, any {u}mount() operations performed here will be reflected on the
1606 static int lxc_spawn(struct lxc_handler
*handler
)
1608 __do_close
int data_sock0
= -EBADF
, data_sock1
= -EBADF
;
1611 bool wants_to_map_ids
;
1612 struct lxc_list
*id_map
;
1613 const char *name
= handler
->name
;
1614 const char *lxcpath
= handler
->lxcpath
;
1615 bool share_ns
= false;
1616 struct lxc_conf
*conf
= handler
->conf
;
1617 struct cgroup_ops
*cgroup_ops
= handler
->cgroup_ops
;
1619 id_map
= &conf
->id_map
;
1620 wants_to_map_ids
= !lxc_list_empty(id_map
);
1622 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1623 if (!conf
->ns_share
[i
])
1626 handler
->nsfd
[i
] = lxc_inherit_namespace(conf
->ns_share
[i
], lxcpath
, ns_info
[i
].proc_name
);
1627 if (handler
->nsfd
[i
] < 0)
1633 ret
= lxc_sync_init(handler
);
1637 ret
= socketpair(AF_UNIX
, SOCK_STREAM
| SOCK_CLOEXEC
, 0,
1638 handler
->data_sock
);
1641 data_sock0
= handler
->data_sock
[0];
1642 data_sock1
= handler
->data_sock
[1];
1644 ret
= resolve_clone_flags(handler
);
1648 if (handler
->ns_clone_flags
& CLONE_NEWNET
) {
1649 ret
= lxc_find_gateway_addresses(handler
);
1651 ERROR("Failed to find gateway addresses");
1656 if (!cgroup_ops
->payload_create(cgroup_ops
, handler
)) {
1657 ERROR("Failed creating cgroups");
1658 goto out_delete_net
;
1661 /* If the rootfs is not a blockdev, prevent the container from marking
1663 * If the container is unprivileged then skip rootfs pinning.
1665 if (!wants_to_map_ids
) {
1666 handler
->pinfd
= pin_rootfs(conf
->rootfs
.path
);
1667 if (handler
->pinfd
== -EBADF
)
1668 INFO("Failed to pin the rootfs for container \"%s\"", handler
->name
);
1671 /* Create a process in a new set of namespaces. */
1675 attacher_pid
= lxc_clone(do_share_ns
, handler
,
1676 CLONE_VFORK
| CLONE_VM
| CLONE_FILES
, NULL
);
1677 if (attacher_pid
< 0) {
1678 SYSERROR(LXC_CLONE_ERROR
);
1679 goto out_delete_net
;
1682 ret
= wait_for_pid(attacher_pid
);
1684 SYSERROR("Intermediate process failed");
1685 goto out_delete_net
;
1688 if (handler
->pid
< 0) {
1689 SYSERROR(LXC_CLONE_ERROR
);
1690 goto out_delete_net
;
1693 int cgroup_fd
= -EBADF
;
1695 struct lxc_clone_args clone_args
= {
1696 .flags
= handler
->clone_flags
,
1697 .pidfd
= ptr_to_u64(&handler
->pidfd
),
1698 .exit_signal
= SIGCHLD
,
1701 if (handler
->ns_clone_flags
& CLONE_NEWCGROUP
) {
1702 cgroup_fd
= cgroup_unified_fd(cgroup_ops
);
1703 if (cgroup_fd
>= 0) {
1704 handler
->clone_flags
|= CLONE_INTO_CGROUP
;
1705 clone_args
.flags
|= CLONE_INTO_CGROUP
;
1706 clone_args
.cgroup
= cgroup_fd
;
1710 /* Try to spawn directly into target cgroup. */
1711 handler
->pid
= lxc_clone3(&clone_args
, CLONE_ARGS_SIZE_VER2
);
1712 if (handler
->pid
< 0) {
1713 SYSTRACE("Failed to spawn container directly into target cgroup");
1715 /* Kernel might simply be too old for CLONE_INTO_CGROUP. */
1716 handler
->clone_flags
&= ~(CLONE_INTO_CGROUP
| CLONE_NEWCGROUP
);
1717 handler
->ns_on_clone_flags
&= ~CLONE_NEWCGROUP
;
1718 handler
->ns_unshare_flags
|= CLONE_NEWCGROUP
;
1720 clone_args
.flags
= handler
->clone_flags
;
1722 handler
->pid
= lxc_clone3(&clone_args
, CLONE_ARGS_SIZE_VER0
);
1723 } else if (cgroup_fd
>= 0) {
1724 TRACE("Spawned container directly into target cgroup via cgroup2 fd %d", cgroup_fd
);
1727 /* Kernel might be too old for clone3(). */
1728 if (handler
->pid
< 0) {
1729 SYSTRACE("Failed to spawn container via clone3()");
1730 handler
->pid
= lxc_raw_legacy_clone(handler
->clone_flags
, &handler
->pidfd
);
1733 if (handler
->pid
< 0) {
1734 SYSERROR(LXC_CLONE_ERROR
);
1735 goto out_delete_net
;
1738 if (handler
->pid
== 0) {
1739 (void)do_start(handler
);
1740 _exit(EXIT_FAILURE
);
1743 if (handler
->pidfd
< 0)
1744 handler
->clone_flags
&= ~CLONE_PIDFD
;
1745 TRACE("Cloned child process %d", handler
->pid
);
1747 /* Verify that we can actually make use of pidfds. */
1748 if (!lxc_can_use_pidfd(handler
->pidfd
))
1749 close_prot_errno_disarm(handler
->pidfd
);
1751 ret
= snprintf(pidstr
, 20, "%d", handler
->pid
);
1752 if (ret
< 0 || ret
>= 20)
1753 goto out_delete_net
;
1755 ret
= setenv("LXC_PID", pidstr
, 1);
1757 SYSERROR("Failed to set environment variable: LXC_PID=%s", pidstr
);
1759 for (i
= 0; i
< LXC_NS_MAX
; i
++)
1760 if (handler
->ns_on_clone_flags
& ns_info
[i
].clone_flag
)
1761 INFO("Cloned %s", ns_info
[i
].flag_name
);
1763 if (!lxc_try_preserve_namespaces(handler
, handler
->ns_on_clone_flags
, handler
->pid
)) {
1764 ERROR("Failed to preserve cloned namespaces for lxc.hook.stop");
1765 goto out_delete_net
;
1768 lxc_sync_fini_child(handler
);
1770 if (lxc_abstract_unix_send_fds(handler
->data_sock
[0], &handler
->monitor_status_fd
, 1, NULL
, 0) < 0) {
1771 ERROR("Failed to send status file descriptor to child process");
1772 goto out_delete_net
;
1774 close_prot_errno_disarm(handler
->monitor_status_fd
);
1776 /* Map the container uids. The container became an invalid userid the
1777 * moment it was cloned with CLONE_NEWUSER. This call doesn't change
1778 * anything immediately, but allows the container to setuid(0) (0 being
1779 * mapped to something else on the host.) later to become a valid uid
1782 if (wants_to_map_ids
) {
1783 if (!handler
->conf
->ns_share
[LXC_NS_USER
] &&
1784 (handler
->conf
->ns_keep
& CLONE_NEWUSER
) == 0) {
1785 ret
= lxc_map_ids(id_map
, handler
->pid
);
1787 ERROR("Failed to set up id mapping.");
1788 goto out_delete_net
;
1793 ret
= lxc_sync_wake_child(handler
, LXC_SYNC_STARTUP
);
1795 goto out_delete_net
;
1797 ret
= lxc_sync_wait_child(handler
, LXC_SYNC_CONFIGURE
);
1799 goto out_delete_net
;
1801 if (!cgroup_ops
->setup_limits_legacy(cgroup_ops
, handler
->conf
, false)) {
1802 ERROR("Failed to setup cgroup limits for container \"%s\"", name
);
1803 goto out_delete_net
;
1806 if (!cgroup_ops
->payload_enter(cgroup_ops
, handler
)) {
1807 ERROR("Failed to enter cgroups");
1808 goto out_delete_net
;
1811 if (!cgroup_ops
->payload_delegate_controllers(cgroup_ops
)) {
1812 ERROR("Failed to delegate controllers to payload cgroup");
1813 goto out_delete_net
;
1816 if (!cgroup_ops
->setup_limits(cgroup_ops
, handler
)) {
1817 ERROR("Failed to setup cgroup limits for container \"%s\"", name
);
1818 goto out_delete_net
;
1821 if (!cgroup_ops
->chown(cgroup_ops
, handler
->conf
))
1822 goto out_delete_net
;
1824 /* If not done yet, we're now ready to preserve the network namespace */
1825 if (handler
->nsfd
[LXC_NS_NET
] < 0) {
1826 ret
= lxc_try_preserve_ns(handler
->pid
, "net");
1828 if (ret
!= -EOPNOTSUPP
) {
1829 SYSERROR("Failed to preserve net namespace");
1830 goto out_delete_net
;
1833 handler
->nsfd
[LXC_NS_NET
] = ret
;
1834 DEBUG("Preserved net namespace via fd %d", ret
);
1837 ret
= lxc_netns_set_nsid(handler
->nsfd
[LXC_NS_NET
]);
1839 SYSWARN("Failed to allocate new network namespace id");
1841 TRACE("Allocated new network namespace id");
1843 /* Create the network configuration. */
1844 if (handler
->ns_clone_flags
& CLONE_NEWNET
) {
1845 ret
= lxc_create_network(handler
);
1847 ERROR("Failed to create the network");
1848 goto out_delete_net
;
1851 ret
= lxc_network_send_to_child(handler
);
1853 ERROR("Failed to send veth names to child");
1854 goto out_delete_net
;
1858 if (!lxc_list_empty(&conf
->procs
)) {
1859 ret
= setup_proc_filesystem(&conf
->procs
, handler
->pid
);
1861 goto out_delete_net
;
1864 /* Tell the child to continue its initialization. We'll get
1865 * LXC_SYNC_CGROUP when it is ready for us to setup cgroups.
1867 ret
= lxc_sync_barrier_child(handler
, LXC_SYNC_POST_CONFIGURE
);
1869 goto out_delete_net
;
1871 if (!lxc_list_empty(&conf
->limits
)) {
1872 ret
= setup_resource_limits(&conf
->limits
, handler
->pid
);
1874 ERROR("Failed to setup resource limits");
1875 goto out_delete_net
;
1879 ret
= lxc_sync_barrier_child(handler
, LXC_SYNC_CGROUP_UNSHARE
);
1881 goto out_delete_net
;
1884 * with isolation the limiting devices cgroup was already setup, so
1885 * only setup devices here if we have no namespace directory
1887 if (!handler
->conf
->cgroup_meta
.namespace_dir
&&
1888 !cgroup_ops
->setup_limits_legacy(cgroup_ops
, handler
->conf
, true)) {
1889 ERROR("Failed to setup legacy device cgroup controller limits");
1890 goto out_delete_net
;
1892 TRACE("Set up legacy device cgroup controller limits");
1894 if (!cgroup_ops
->devices_activate(cgroup_ops
, handler
)) {
1895 ERROR("Failed to setup cgroup2 device controller limits");
1896 goto out_delete_net
;
1898 TRACE("Set up cgroup2 device controller limits");
1900 if (handler
->ns_unshare_flags
& CLONE_NEWCGROUP
) {
1901 /* Now we're ready to preserve the cgroup namespace */
1902 ret
= lxc_try_preserve_ns(handler
->pid
, "cgroup");
1904 if (ret
!= -EOPNOTSUPP
) {
1905 SYSERROR("Failed to preserve cgroup namespace");
1906 goto out_delete_net
;
1909 handler
->nsfd
[LXC_NS_CGROUP
] = ret
;
1910 DEBUG("Preserved cgroup namespace via fd %d", ret
);
1914 cgroup_ops
->payload_finalize(cgroup_ops
);
1915 TRACE("Finished setting up cgroups");
1917 if (handler
->ns_unshare_flags
& CLONE_NEWTIME
) {
1918 /* Now we're ready to preserve the cgroup namespace */
1919 ret
= lxc_try_preserve_ns(handler
->pid
, "time");
1921 if (ret
!= -EOPNOTSUPP
) {
1922 SYSERROR("Failed to preserve time namespace");
1923 goto out_delete_net
;
1926 handler
->nsfd
[LXC_NS_TIME
] = ret
;
1927 DEBUG("Preserved time namespace via fd %d", ret
);
1931 /* Run any host-side start hooks */
1932 ret
= run_lxc_hooks(name
, "start-host", conf
, NULL
);
1934 ERROR("Failed to run lxc.hook.start-host");
1935 goto out_delete_net
;
1938 /* Tell the child to complete its initialization and wait for it to exec
1939 * or return an error. (The child will never return
1940 * LXC_SYNC_READY_START+1. It will either close the sync pipe, causing
1941 * lxc_sync_barrier_child to return success, or return a different
1942 * value, causing us to error out).
1944 ret
= lxc_sync_barrier_child(handler
, LXC_SYNC_READY_START
);
1946 goto out_delete_net
;
1948 if (handler
->ns_clone_flags
& CLONE_NEWNET
) {
1949 ret
= lxc_network_recv_name_and_ifindex_from_child(handler
);
1951 ERROR("Failed to receive names and ifindices for network devices from child");
1952 goto out_delete_net
;
1956 ret
= lxc_abstract_unix_recv_fds(data_sock1
, &handler
->conf
->devpts_fd
, 1, NULL
, 0);
1958 SYSERROR("Failed to receive devpts fd from child");
1959 goto out_delete_net
;
1962 handler
->conf
->devpts_fd
= -EBADF
;
1964 /* Now all networks are created, network devices are moved into place,
1965 * and the correct names and ifindices in the respective namespaces have
1966 * been recorded. The corresponding structs have now all been filled. So
1967 * log them for debugging purposes.
1969 lxc_log_configured_netdevs(conf
);
1971 /* Read tty fds allocated by child. */
1972 ret
= lxc_recv_ttys_from_child(handler
);
1974 ERROR("Failed to receive tty info from child process");
1975 goto out_delete_net
;
1978 ret
= lxc_seccomp_recv_notifier_fd(&handler
->conf
->seccomp
, data_sock1
);
1980 SYSERROR("Failed to receive seccomp notify fd from child");
1981 goto out_delete_net
;
1984 ret
= handler
->ops
->post_start(handler
, handler
->data
);
1988 ret
= lxc_set_state(name
, handler
, RUNNING
);
1990 ERROR("Failed to set state to \"%s\"", lxc_state2str(RUNNING
));
1994 lxc_sync_fini(handler
);
1999 if (handler
->ns_clone_flags
& CLONE_NEWNET
)
2000 lxc_delete_network(handler
);
2006 lxc_sync_fini(handler
);
2007 close_prot_errno_disarm(handler
->pinfd
);
2012 int __lxc_start(struct lxc_handler
*handler
, struct lxc_operations
*ops
,
2013 void *data
, const char *lxcpath
, bool daemonize
, int *error_num
)
2016 const char *name
= handler
->name
;
2017 struct lxc_conf
*conf
= handler
->conf
;
2018 struct cgroup_ops
*cgroup_ops
;
2020 ret
= lxc_init(name
, handler
);
2022 ERROR("Failed to initialize container \"%s\"", name
);
2026 handler
->data
= data
;
2027 handler
->daemonize
= daemonize
;
2028 cgroup_ops
= handler
->cgroup_ops
;
2030 if (!attach_block_device(handler
->conf
)) {
2031 ERROR("Failed to attach block device");
2036 if (!cgroup_ops
->monitor_create(cgroup_ops
, handler
)) {
2037 ERROR("Failed to create monitor cgroup");
2042 if (!cgroup_ops
->monitor_enter(cgroup_ops
, handler
)) {
2043 ERROR("Failed to enter monitor cgroup");
2048 if (!cgroup_ops
->monitor_delegate_controllers(cgroup_ops
)) {
2049 ERROR("Failed to delegate controllers to monitor cgroup");
2054 if (geteuid() == 0 && !lxc_list_empty(&conf
->id_map
)) {
2055 /* If the backing store is a device, mount it here and now. */
2056 if (rootfs_is_blockdev(conf
)) {
2057 ret
= unshare(CLONE_NEWNS
);
2059 ERROR("Failed to unshare CLONE_NEWNS");
2062 INFO("Unshared CLONE_NEWNS");
2064 turn_into_dependent_mounts();
2065 ret
= lxc_setup_rootfs_prepare_root(conf
, name
, lxcpath
);
2067 ERROR("Error setting up rootfs mount as root before spawn");
2070 INFO("Set up container rootfs as host root");
2074 ret
= lxc_spawn(handler
);
2076 ERROR("Failed to spawn container \"%s\"", name
);
2077 goto out_detach_blockdev
;
2080 handler
->conf
->reboot
= REBOOT_NONE
;
2082 ret
= lxc_poll(name
, handler
);
2084 ERROR("LXC mainloop exited with error: %d", ret
);
2085 goto out_delete_network
;
2088 if (!handler
->init_died
&& handler
->pid
> 0) {
2089 ERROR("Child process is not killed");
2091 goto out_delete_network
;
2094 status
= lxc_wait_for_pid_status(handler
->pid
);
2096 SYSERROR("Failed to retrieve status for %d", handler
->pid
);
2098 /* If the child process exited but was not signaled, it didn't call
2099 * reboot. This should mean it was an lxc-execute which simply exited.
2100 * In any case, treat it as a 'halt'.
2102 if (WIFSIGNALED(status
)) {
2103 switch(WTERMSIG(status
)) {
2104 case SIGINT
: /* halt */
2105 DEBUG("Container \"%s\" is halting", name
);
2107 case SIGHUP
: /* reboot */
2108 DEBUG("Container \"%s\" is rebooting", name
);
2109 handler
->conf
->reboot
= REBOOT_REQ
;
2111 case SIGSYS
: /* seccomp */
2112 DEBUG("Container \"%s\" violated its seccomp policy", name
);
2115 DEBUG("Unknown exit status for container \"%s\" init %d", name
, WTERMSIG(status
));
2120 ret
= lxc_restore_phys_nics_to_netns(handler
);
2122 ERROR("Failed to move physical network devices back to parent network namespace");
2124 close_prot_errno_disarm(handler
->pinfd
);
2126 lxc_monitor_send_exit_code(name
, status
, handler
->lxcpath
);
2127 lxc_error_set_and_log(handler
->pid
, status
);
2129 *error_num
= handler
->exit_status
;
2131 lxc_delete_network(handler
);
2132 detach_block_device(handler
->conf
);
2141 out_detach_blockdev
:
2143 detach_block_device(handler
->conf
);
2149 lxc_delete_network(handler
);
2150 detach_block_device(handler
->conf
);
2159 static int start(struct lxc_handler
*handler
, void* data
)
2161 struct start_args
*arg
= data
;
2163 NOTICE("Exec'ing \"%s\"", arg
->argv
[0]);
2165 execvp(arg
->argv
[0], arg
->argv
);
2166 SYSERROR("Failed to exec \"%s\"", arg
->argv
[0]);
2170 static int post_start(struct lxc_handler
*handler
, void* data
)
2172 struct start_args
*arg
= data
;
2174 NOTICE("Started \"%s\" with pid \"%d\"", arg
->argv
[0], handler
->pid
);
2178 static struct lxc_operations start_ops
= {
2180 .post_start
= post_start
2183 int lxc_start(char *const argv
[], struct lxc_handler
*handler
,
2184 const char *lxcpath
, bool daemonize
, int *error_num
)
2186 struct start_args start_arg
= {
2190 TRACE("Doing lxc_start");
2191 return __lxc_start(handler
, &start_ops
, &start_arg
, lxcpath
, daemonize
, error_num
);
2194 static void lxc_destroy_container_on_signal(struct lxc_handler
*handler
,
2197 char destroy
[PATH_MAX
];
2198 struct lxc_container
*c
;
2202 if (handler
->conf
->rootfs
.path
&& handler
->conf
->rootfs
.mount
) {
2203 bret
= do_destroy_container(handler
);
2205 ERROR("Error destroying rootfs for container \"%s\"", name
);
2209 INFO("Destroyed rootfs for container \"%s\"", name
);
2211 ret
= snprintf(destroy
, PATH_MAX
, "%s/%s", handler
->lxcpath
, name
);
2212 if (ret
< 0 || ret
>= PATH_MAX
) {
2213 ERROR("Error destroying directory for container \"%s\"", name
);
2217 c
= lxc_container_new(name
, handler
->lxcpath
);
2219 if (container_disk_lock(c
)) {
2220 INFO("Could not update lxc_snapshots file");
2221 lxc_container_put(c
);
2223 mod_all_rdeps(c
, false);
2224 container_disk_unlock(c
);
2225 lxc_container_put(c
);
2229 if (!handler
->am_root
)
2230 ret
= userns_exec_full(handler
->conf
, lxc_rmdir_onedev_wrapper
,
2231 destroy
, "lxc_rmdir_onedev_wrapper");
2233 ret
= lxc_rmdir_onedev(destroy
, NULL
);
2236 ERROR("Error destroying directory for container \"%s\"", name
);
2239 INFO("Destroyed directory for container \"%s\"", name
);
2242 static int lxc_rmdir_onedev_wrapper(void *data
)
2244 char *arg
= (char *) data
;
2245 return lxc_rmdir_onedev(arg
, NULL
);
2248 static bool do_destroy_container(struct lxc_handler
*handler
)
2252 if (!handler
->am_root
) {
2253 ret
= userns_exec_full(handler
->conf
, storage_destroy_wrapper
,
2254 handler
->conf
, "storage_destroy_wrapper");
2261 return storage_destroy(handler
->conf
);