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"
35 #include "confile_utils.h"
37 #include "file_utils.h"
41 #include "lxccontainer.h"
43 #include "lxcseccomp.h"
46 #include "memory_utils.h"
48 #include "namespace.h"
50 #include "raw_syscalls.h"
52 #include "storage/storage.h"
53 #include "storage/storage_utils.h"
55 #include "syscall_wrappers.h"
60 #include <sys/capability.h>
64 #include "include/strlcpy.h"
67 lxc_log_define(start
, lxc
);
69 extern void mod_all_rdeps(struct lxc_container
*c
, bool inc
);
70 static bool do_destroy_container(struct lxc_handler
*handler
);
71 static int lxc_rmdir_onedev_wrapper(void *data
);
72 static void lxc_destroy_container_on_signal(struct lxc_handler
*handler
,
75 static void print_top_failing_dir(const char *path
)
77 __do_free
char *copy
= NULL
;
81 copy
= must_copy_string(path
);
83 e
= copy
+ strlen(path
);
86 while (p
< e
&& *p
== '/')
89 while (p
< e
&& *p
!= '/')
95 ret
= access(copy
, X_OK
);
97 SYSERROR("Could not access %s. Please grant it x access, or add an ACL for the container " "root", copy
);
104 static void lxc_put_nsfds(struct lxc_handler
*handler
)
106 for (int i
= 0; i
< LXC_NS_MAX
; i
++) {
107 if (handler
->nsfd
[i
] < 0)
110 close_prot_errno_disarm(handler
->nsfd
[i
]);
114 static int lxc_try_preserve_ns(const int pid
, const char *ns
)
118 fd
= lxc_preserve_ns(pid
, ns
);
121 return log_error_errno(-EINVAL
,
122 errno
, "Failed to preserve %s namespace",
125 return log_warn_errno(-EOPNOTSUPP
,
126 errno
, "Kernel does not support preserving %s namespaces",
133 /* lxc_try_preserve_namespaces: open /proc/@pid/ns/@ns for each namespace
134 * specified in ns_clone_flags.
135 * Return true on success, false on failure.
137 static bool lxc_try_preserve_namespaces(struct lxc_handler
*handler
,
138 int ns_clone_flags
, pid_t pid
)
142 for (i
= 0; i
< LXC_NS_MAX
; i
++)
143 handler
->nsfd
[i
] = -EBADF
;
145 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
148 if ((ns_clone_flags
& ns_info
[i
].clone_flag
) == 0)
151 fd
= lxc_try_preserve_ns(pid
, ns_info
[i
].proc_name
);
153 /* Do not fail to start container on kernels that do
154 * not support interacting with namespaces through
157 if (fd
== -EOPNOTSUPP
)
160 lxc_put_nsfds(handler
);
164 handler
->nsfd
[i
] = fd
;
165 DEBUG("Preserved %s namespace via fd %d", ns_info
[i
].proc_name
,
172 static inline bool match_stdfds(int fd
)
174 return (fd
== STDIN_FILENO
|| fd
== STDOUT_FILENO
|| fd
== STDERR_FILENO
);
178 static bool match_dlog_fds(struct dirent
*direntp
)
180 char path
[PATH_MAX
] = {0};
181 char link
[PATH_MAX
] = {0};
185 ret
= snprintf(path
, PATH_MAX
, "/proc/self/fd/%s", direntp
->d_name
);
186 if (ret
< 0 || ret
>= PATH_MAX
)
187 return log_error(false, "Failed to create file descriptor name");
189 linklen
= readlink(path
, link
, PATH_MAX
);
191 return log_error(false, "Failed to read link path - \"%s\"", path
);
192 else if (linklen
>= PATH_MAX
)
193 return log_error(false, "The name of link path is too long - \"%s\"", path
);
195 if (strcmp(link
, "/dev/log_main") == 0 ||
196 strcmp(link
, "/dev/log_system") == 0 ||
197 strcmp(link
, "/dev/log_radio") == 0)
204 int lxc_check_inherited(struct lxc_conf
*conf
, bool closeall
,
205 int *fds_to_ignore
, size_t len_fds
)
210 struct dirent
*direntp
;
212 if (conf
&& conf
->close_all_fds
)
216 dir
= opendir("/proc/self/fd");
218 return log_warn(-1, "Failed to open directory");
222 while ((direntp
= readdir(dir
))) {
224 struct lxc_list
*cur
;
225 bool matched
= false;
227 if (strcmp(direntp
->d_name
, ".") == 0)
230 if (strcmp(direntp
->d_name
, "..") == 0)
233 ret
= lxc_safe_int(direntp
->d_name
, &fd
);
235 INFO("Could not parse file descriptor for \"%s\"", direntp
->d_name
);
239 for (i
= 0; i
< len_fds
; i
++)
240 if (fds_to_ignore
[i
] == fd
)
243 if (fd
== fddir
|| fd
== lxc_log_fd
||
244 (i
< len_fds
&& fd
== fds_to_ignore
[i
]))
247 /* Keep state clients that wait on reboots. */
249 lxc_list_for_each(cur
, &conf
->state_clients
) {
250 struct lxc_state_client
*client
= cur
->elem
;
252 if (client
->clientfd
!= fd
)
263 if (current_config
&& fd
== current_config
->logfd
)
266 if (match_stdfds(fd
))
270 if (match_dlog_fds(direntp
))
277 INFO("Closed inherited fd %d", fd
);
280 WARN("Inherited fd %d", fd
);
283 /* Only enable syslog at this point to avoid the above logging function
284 * to open a new fd and make the check_inherited function enter an
287 lxc_log_enable_syslog();
289 closedir(dir
); /* cannot fail */
293 static int setup_signal_fd(sigset_t
*oldmask
)
297 const int signals
[] = {SIGBUS
, SIGILL
, SIGSEGV
, SIGWINCH
};
299 /* Block everything except serious error signals. */
300 ret
= sigfillset(&mask
);
304 for (int sig
= 0; sig
< (sizeof(signals
) / sizeof(signals
[0])); sig
++) {
305 ret
= sigdelset(&mask
, signals
[sig
]);
310 ret
= pthread_sigmask(SIG_BLOCK
, &mask
, oldmask
);
312 return log_error_errno(-EBADF
, errno
,
313 "Failed to set signal mask");
315 ret
= signalfd(-1, &mask
, SFD_CLOEXEC
);
317 return log_error_errno(-EBADF
,
318 errno
, "Failed to create signal file descriptor");
320 TRACE("Created signal file descriptor %d", ret
);
325 static int signal_handler(int fd
, uint32_t events
, void *data
,
326 struct lxc_epoll_descr
*descr
)
330 struct signalfd_siginfo siginfo
;
331 struct lxc_handler
*hdlr
= data
;
333 ret
= lxc_read_nointr(fd
, &siginfo
, sizeof(siginfo
));
335 return log_error(LXC_MAINLOOP_ERROR
, "Failed to read signal info from signal file descriptor %d", fd
);
337 if (ret
!= sizeof(siginfo
))
338 return log_error(-EINVAL
, "Unexpected size for struct signalfd_siginfo");
340 /* Check whether init is running. */
342 ret
= waitid(P_PID
, hdlr
->pid
, &info
, WEXITED
| WNOWAIT
| WNOHANG
);
343 if (ret
== 0 && info
.si_pid
== hdlr
->pid
)
344 hdlr
->init_died
= true;
346 /* Try to figure out a reasonable exit status to report. */
347 if (hdlr
->init_died
) {
348 switch (info
.si_code
) {
350 hdlr
->exit_status
= info
.si_status
<< 8;
355 hdlr
->exit_status
= info
.si_status
<< 8 | 0x7f;
358 /* Huh? The waitid() told us it's dead *and* continued? */
359 WARN("Init %d dead and continued?", hdlr
->pid
);
360 hdlr
->exit_status
= 1;
363 ERROR("Unknown si_code: %d", info
.si_code
);
364 hdlr
->exit_status
= 1;
368 if (siginfo
.ssi_signo
== SIGHUP
) {
369 if (hdlr
->pidfd
>= 0)
370 lxc_raw_pidfd_send_signal(hdlr
->pidfd
, SIGTERM
, NULL
, 0);
372 kill(hdlr
->pid
, SIGTERM
);
373 INFO("Killing %d since terminal hung up", hdlr
->pid
);
374 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
375 : LXC_MAINLOOP_CONTINUE
;
378 if (siginfo
.ssi_signo
!= SIGCHLD
) {
379 if (hdlr
->pidfd
>= 0)
380 lxc_raw_pidfd_send_signal(hdlr
->pidfd
,
381 siginfo
.ssi_signo
, NULL
, 0);
383 kill(hdlr
->pid
, siginfo
.ssi_signo
);
384 INFO("Forwarded signal %d to pid %d", siginfo
.ssi_signo
, hdlr
->pid
);
385 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
386 : LXC_MAINLOOP_CONTINUE
;
389 /* More robustness, protect ourself from a SIGCHLD sent
390 * by a process different from the container init.
392 if (siginfo
.ssi_pid
!= hdlr
->pid
) {
393 NOTICE("Received %d from pid %d instead of container init %d",
394 siginfo
.ssi_signo
, siginfo
.ssi_pid
, hdlr
->pid
);
395 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
396 : LXC_MAINLOOP_CONTINUE
;
399 if (siginfo
.ssi_code
== CLD_STOPPED
) {
400 INFO("Container init process was stopped");
401 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
402 : LXC_MAINLOOP_CONTINUE
;
405 if (siginfo
.ssi_code
== CLD_CONTINUED
) {
406 INFO("Container init process was continued");
407 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
408 : LXC_MAINLOOP_CONTINUE
;
411 return log_debug(LXC_MAINLOOP_CLOSE
, "Container init process %d exited", hdlr
->pid
);
414 int lxc_serve_state_clients(const char *name
, struct lxc_handler
*handler
,
419 struct lxc_list
*cur
, *next
;
420 struct lxc_msg msg
= {.type
= lxc_msg_state
, .value
= state
};
423 handler
->state
= RUNNING
;
425 handler
->state
= state
;
427 TRACE("Set container state to %s", lxc_state2str(state
));
429 if (lxc_list_empty(&handler
->conf
->state_clients
))
430 return log_trace(0, "No state clients registered");
432 retlen
= strlcpy(msg
.name
, name
, sizeof(msg
.name
));
433 if (retlen
>= sizeof(msg
.name
))
436 lxc_list_for_each_safe(cur
, &handler
->conf
->state_clients
, next
) {
437 struct lxc_state_client
*client
= cur
->elem
;
439 if (client
->states
[state
] == 0) {
440 TRACE("State %s not registered for state client %d",
441 lxc_state2str(state
), client
->clientfd
);
445 TRACE("Sending state %s to state client %d",
446 lxc_state2str(state
), client
->clientfd
);
448 ret
= lxc_send_nointr(client
->clientfd
, &msg
, sizeof(msg
), MSG_NOSIGNAL
);
450 SYSERROR("Failed to send message to client");
452 /* kick client from list */
454 close(client
->clientfd
);
462 static int lxc_serve_state_socket_pair(const char *name
,
463 struct lxc_handler
*handler
,
468 if (!handler
->daemonize
||
469 handler
->state_socket_pair
[1] < 0 ||
473 /* Close read end of the socket pair. */
474 close_prot_errno_disarm(handler
->state_socket_pair
[0]);
477 ret
= lxc_abstract_unix_send_credential(handler
->state_socket_pair
[1],
478 &(int){state
}, sizeof(int));
480 SYSERROR("Failed to send state to %d", handler
->state_socket_pair
[1]);
488 if (ret
!= sizeof(int))
489 return log_error(-1, "Message too long : %d", handler
->state_socket_pair
[1]);
491 TRACE("Sent container state \"%s\" to %d", lxc_state2str(state
),
492 handler
->state_socket_pair
[1]);
494 /* Close write end of the socket pair. */
495 close_prot_errno_disarm(handler
->state_socket_pair
[1]);
500 int lxc_set_state(const char *name
, struct lxc_handler
*handler
,
505 ret
= lxc_serve_state_socket_pair(name
, handler
, state
);
507 return log_error(-1, "Failed to synchronize via anonymous pair of unix sockets");
509 ret
= lxc_serve_state_clients(name
, handler
, state
);
513 /* This function will try to connect to the legacy lxc-monitord state
514 * server and only exists for backwards compatibility.
516 lxc_monitor_send_state(name
, state
, handler
->lxcpath
);
521 int lxc_poll(const char *name
, struct lxc_handler
*handler
)
524 bool has_console
= true;
525 struct lxc_epoll_descr descr
, descr_console
;
527 if (handler
->conf
->console
.path
&&
528 strcmp(handler
->conf
->console
.path
, "none") == 0)
531 ret
= lxc_mainloop_open(&descr
);
533 ERROR("Failed to create mainloop");
538 ret
= lxc_mainloop_open(&descr_console
);
540 ERROR("Failed to create console mainloop");
545 ret
= lxc_mainloop_add_handler(&descr
, handler
->sigfd
, signal_handler
, handler
);
547 ERROR("Failed to add signal handler for %d to mainloop", handler
->sigfd
);
548 goto out_mainloop_console
;
551 ret
= lxc_seccomp_setup_proxy(&handler
->conf
->seccomp
, &descr
, handler
);
553 ERROR("Failed to setup seccomp proxy");
554 goto out_mainloop_console
;
558 struct lxc_terminal
*console
= &handler
->conf
->console
;
560 ret
= lxc_terminal_mainloop_add(&descr
, console
);
562 ERROR("Failed to add console handlers to mainloop");
563 goto out_mainloop_console
;
566 ret
= lxc_terminal_mainloop_add(&descr_console
, console
);
568 ERROR("Failed to add console handlers to console mainloop");
569 goto out_mainloop_console
;
572 handler
->conf
->console
.descr
= &descr
;
575 ret
= lxc_cmd_mainloop_add(name
, &descr
, handler
);
577 ERROR("Failed to add command handler to mainloop");
578 goto out_mainloop_console
;
581 TRACE("Mainloop is ready");
583 ret
= lxc_mainloop(&descr
, -1);
584 close_prot_errno_disarm(descr
.epfd
);
585 if (ret
< 0 || !handler
->init_died
)
586 goto out_mainloop_console
;
589 ret
= lxc_mainloop(&descr_console
, 0);
591 out_mainloop_console
:
593 lxc_mainloop_close(&descr_console
);
594 TRACE("Closed console mainloop");
598 lxc_mainloop_close(&descr
);
599 TRACE("Closed mainloop");
602 TRACE("Closed signal file descriptor %d", handler
->sigfd
);
603 close_prot_errno_disarm(handler
->sigfd
);
608 void lxc_zero_handler(struct lxc_handler
*handler
)
610 memset(handler
, 0, sizeof(struct lxc_handler
));
612 handler
->state
= STOPPED
;
614 handler
->pinfd
= -EBADF
;
616 handler
->pidfd
= -EBADF
;
618 handler
->sigfd
= -EBADF
;
620 for (int i
= 0; i
< LXC_NS_MAX
; i
++)
621 handler
->nsfd
[i
] = -EBADF
;
623 handler
->data_sock
[0] = -EBADF
;
624 handler
->data_sock
[1] = -EBADF
;
626 handler
->state_socket_pair
[0] = -EBADF
;
627 handler
->state_socket_pair
[1] = -EBADF
;
629 handler
->sync_sock
[0] = -EBADF
;
630 handler
->sync_sock
[1] = -EBADF
;
633 void lxc_free_handler(struct lxc_handler
*handler
)
635 close_prot_errno_disarm(handler
->pinfd
);
636 close_prot_errno_disarm(handler
->pidfd
);
637 close_prot_errno_disarm(handler
->sigfd
);
638 lxc_put_nsfds(handler
);
639 if (handler
->conf
&& handler
->conf
->reboot
== REBOOT_NONE
)
640 close_prot_errno_disarm(handler
->conf
->maincmd_fd
);
641 close_prot_errno_disarm(handler
->monitor_status_fd
);
642 close_prot_errno_disarm(handler
->state_socket_pair
[0]);
643 close_prot_errno_disarm(handler
->state_socket_pair
[1]);
644 cgroup_exit(handler
->cgroup_ops
);
645 handler
->conf
= NULL
;
646 free_disarm(handler
);
649 struct lxc_handler
*lxc_init_handler(const char *name
, struct lxc_conf
*conf
,
650 const char *lxcpath
, bool daemonize
)
653 struct lxc_handler
*handler
;
655 handler
= malloc(sizeof(*handler
));
659 memset(handler
, 0, sizeof(*handler
));
661 /* Note that am_guest_unpriv() checks the effective uid. We
662 * probably don't care if we are real root only if we are running
663 * as root so this should be fine.
665 handler
->am_root
= !am_guest_unpriv();
666 handler
->conf
= conf
;
667 handler
->lxcpath
= lxcpath
;
668 handler
->init_died
= false;
669 handler
->data_sock
[0] = -EBADF
;
670 handler
->data_sock
[1] = -EBADF
;
671 handler
->monitor_status_fd
= -EBADF
;
672 handler
->pinfd
= -EBADF
;
673 handler
->pidfd
= -EBADF
;
674 handler
->sigfd
= -EBADF
;
675 handler
->state_socket_pair
[0] = -EBADF
;
676 handler
->state_socket_pair
[1] = -EBADF
;
677 if (handler
->conf
->reboot
== REBOOT_NONE
)
678 lxc_list_init(&handler
->conf
->state_clients
);
680 for (int i
= 0; i
< LXC_NS_MAX
; i
++)
681 handler
->nsfd
[i
] = -EBADF
;
683 handler
->name
= name
;
685 handler
->transient_pid
= lxc_raw_getpid();
687 handler
->transient_pid
= -1;
689 if (daemonize
&& handler
->conf
->reboot
== REBOOT_NONE
) {
690 /* Create socketpair() to synchronize on daemonized startup.
691 * When the container reboots we don't need to synchronize
692 * again currently so don't open another socketpair().
694 ret
= socketpair(AF_UNIX
, SOCK_STREAM
| SOCK_CLOEXEC
, 0,
695 handler
->state_socket_pair
);
697 ERROR("Failed to create anonymous pair of unix sockets");
701 TRACE("Created anonymous pair {%d,%d} of unix sockets",
702 handler
->state_socket_pair
[0],
703 handler
->state_socket_pair
[1]);
706 if (handler
->conf
->reboot
== REBOOT_NONE
) {
707 handler
->conf
->maincmd_fd
= lxc_cmd_init(name
, lxcpath
, "command");
708 if (handler
->conf
->maincmd_fd
< 0) {
709 ERROR("Failed to set up command socket");
714 TRACE("Unix domain socket %d for command server is ready",
715 handler
->conf
->maincmd_fd
);
720 lxc_free_handler(handler
);
725 int lxc_init(const char *name
, struct lxc_handler
*handler
)
727 __do_close_prot_errno
int status_fd
= -EBADF
;
729 const char *loglevel
;
730 struct lxc_conf
*conf
= handler
->conf
;
732 handler
->monitor_pid
= lxc_raw_getpid();
733 status_fd
= open("/proc/self/status", O_RDONLY
| O_CLOEXEC
);
735 SYSERROR("Failed to open monitor status fd");
736 goto out_close_maincmd_fd
;
740 TRACE("Initialized LSM");
742 ret
= lxc_read_seccomp_config(conf
);
744 ERROR("Failed loading seccomp policy");
745 goto out_close_maincmd_fd
;
747 TRACE("Read seccomp policy");
749 /* Begin by setting the state to STARTING. */
750 ret
= lxc_set_state(name
, handler
, STARTING
);
752 ERROR("Failed to set state to \"%s\"", lxc_state2str(STARTING
));
753 goto out_close_maincmd_fd
;
755 TRACE("Set container state to \"STARTING\"");
757 /* Start of environment variable setup for hooks. */
758 ret
= setenv("LXC_NAME", name
, 1);
760 SYSERROR("Failed to set environment variable: LXC_NAME=%s", name
);
763 ret
= setenv("LXC_CONFIG_FILE", conf
->rcfile
, 1);
765 SYSERROR("Failed to set environment variable: LXC_CONFIG_FILE=%s", conf
->rcfile
);
768 if (conf
->rootfs
.mount
) {
769 ret
= setenv("LXC_ROOTFS_MOUNT", conf
->rootfs
.mount
, 1);
771 SYSERROR("Failed to set environment variable: LXC_ROOTFS_MOUNT=%s", conf
->rootfs
.mount
);
774 if (conf
->rootfs
.path
) {
775 ret
= setenv("LXC_ROOTFS_PATH", conf
->rootfs
.path
, 1);
777 SYSERROR("Failed to set environment variable: LXC_ROOTFS_PATH=%s", conf
->rootfs
.path
);
780 if (conf
->console
.path
) {
781 ret
= setenv("LXC_CONSOLE", conf
->console
.path
, 1);
783 SYSERROR("Failed to set environment variable: LXC_CONSOLE=%s", conf
->console
.path
);
786 if (conf
->console
.log_path
) {
787 ret
= setenv("LXC_CONSOLE_LOGPATH", conf
->console
.log_path
, 1);
789 SYSERROR("Failed to set environment variable: LXC_CONSOLE_LOGPATH=%s", conf
->console
.log_path
);
792 if (cgns_supported()) {
793 ret
= setenv("LXC_CGNS_AWARE", "1", 1);
795 SYSERROR("Failed to set environment variable LXC_CGNS_AWARE=1");
798 loglevel
= lxc_log_priority_to_string(lxc_log_get_level());
799 ret
= setenv("LXC_LOG_LEVEL", loglevel
, 1);
801 SYSERROR("Set environment variable LXC_LOG_LEVEL=%s", loglevel
);
803 if (conf
->hooks_version
== 0)
804 ret
= setenv("LXC_HOOK_VERSION", "0", 1);
806 ret
= setenv("LXC_HOOK_VERSION", "1", 1);
808 SYSERROR("Failed to set environment variable LXC_HOOK_VERSION=%u", conf
->hooks_version
);
809 /* End of environment variable setup for hooks. */
811 TRACE("Set environment variables");
813 ret
= run_lxc_hooks(name
, "pre-start", conf
, NULL
);
815 ERROR("Failed to run lxc.hook.pre-start for container \"%s\"", name
);
818 TRACE("Ran pre-start hooks");
820 /* The signal fd has to be created before forking otherwise if the child
821 * process exits before we setup the signal fd, the event will be lost
822 * and the command will be stuck.
824 handler
->sigfd
= setup_signal_fd(&handler
->oldmask
);
825 if (handler
->sigfd
< 0) {
826 ERROR("Failed to setup SIGCHLD fd handler.");
829 TRACE("Set up signal fd");
831 /* Do this after setting up signals since it might unblock SIGWINCH. */
832 ret
= lxc_terminal_setup(conf
);
834 ERROR("Failed to create console");
835 goto out_restore_sigmask
;
837 TRACE("Created console");
839 ret
= lxc_terminal_map_ids(conf
, &conf
->console
);
841 ERROR("Failed to chown console");
842 goto out_delete_terminal
;
844 TRACE("Chowned console");
846 handler
->cgroup_ops
= cgroup_init(handler
->conf
);
847 if (!handler
->cgroup_ops
) {
848 ERROR("Failed to initialize cgroup driver");
849 goto out_delete_terminal
;
851 TRACE("Initialized cgroup driver");
853 ret
= lsm_process_prepare(conf
, handler
->lxcpath
);
855 ERROR("Failed to initialize LSM");
856 goto out_delete_terminal
;
858 TRACE("Initialized LSM");
860 INFO("Container \"%s\" is initialized", name
);
861 handler
->monitor_status_fd
= move_fd(status_fd
);
865 lxc_terminal_delete(&handler
->conf
->console
);
868 (void)pthread_sigmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
871 lxc_delete_tty(&conf
->ttys
);
874 (void)lxc_set_state(name
, handler
, ABORTING
);
876 out_close_maincmd_fd
:
877 close_prot_errno_disarm(conf
->maincmd_fd
);
881 void lxc_fini(const char *name
, struct lxc_handler
*handler
)
885 struct lxc_list
*cur
, *next
;
886 char *namespaces
[LXC_NS_MAX
+ 1];
887 size_t namespace_count
= 0;
888 struct cgroup_ops
*cgroup_ops
= handler
->cgroup_ops
;
890 /* The STOPPING state is there for future cleanup code which can take
893 lxc_set_state(name
, handler
, STOPPING
);
895 self
= lxc_raw_getpid();
896 for (int i
= 0; i
< LXC_NS_MAX
; i
++) {
897 if (handler
->nsfd
[i
] < 0)
900 if (handler
->conf
->hooks_version
== 0)
901 ret
= asprintf(&namespaces
[namespace_count
],
902 "%s:/proc/%d/fd/%d", ns_info
[i
].proc_name
,
903 self
, handler
->nsfd
[i
]);
905 ret
= asprintf(&namespaces
[namespace_count
],
906 "/proc/%d/fd/%d", self
, handler
->nsfd
[i
]);
908 SYSERROR("Failed to allocate memory");
912 if (handler
->conf
->hooks_version
== 0) {
917 ret
= setenv(ns_info
[i
].env_name
, namespaces
[namespace_count
], 1);
919 SYSERROR("Failed to set environment variable %s=%s",
920 ns_info
[i
].env_name
, namespaces
[namespace_count
]);
922 TRACE("Set environment variable %s=%s",
923 ns_info
[i
].env_name
, namespaces
[namespace_count
]);
927 namespaces
[namespace_count
] = NULL
;
929 if (handler
->conf
->reboot
> REBOOT_NONE
) {
930 ret
= setenv("LXC_TARGET", "reboot", 1);
932 SYSERROR("Failed to set environment variable: LXC_TARGET=reboot");
935 if (handler
->conf
->reboot
== REBOOT_NONE
) {
936 ret
= setenv("LXC_TARGET", "stop", 1);
938 SYSERROR("Failed to set environment variable: LXC_TARGET=stop");
941 if (handler
->conf
->hooks_version
== 0)
942 ret
= run_lxc_hooks(name
, "stop", handler
->conf
, namespaces
);
944 ret
= run_lxc_hooks(name
, "stop", handler
->conf
, NULL
);
946 ERROR("Failed to run \"lxc.hook.stop\" hook");
948 while (namespace_count
--)
949 free(namespaces
[namespace_count
]);
951 lsm_process_cleanup(handler
->conf
, handler
->lxcpath
);
954 cgroup_ops
->payload_destroy(cgroup_ops
, handler
);
955 cgroup_ops
->monitor_destroy(cgroup_ops
, handler
);
958 if (handler
->conf
->reboot
== REBOOT_NONE
) {
959 /* For all new state clients simply close the command socket.
960 * This will inform all state clients that the container is
961 * STOPPED and also prevents a race between a open()/close() on
962 * the command socket causing a new process to get ECONNREFUSED
963 * because we haven't yet closed the command socket.
965 close_prot_errno_disarm(handler
->conf
->maincmd_fd
);
966 TRACE("Closed command socket");
968 /* This function will try to connect to the legacy lxc-monitord
969 * state server and only exists for backwards compatibility.
971 lxc_monitor_send_state(name
, STOPPED
, handler
->lxcpath
);
973 /* The command socket is closed so no one can acces the command
974 * socket anymore so there's no need to lock it.
976 handler
->state
= STOPPED
;
977 TRACE("Set container state to \"STOPPED\"");
979 lxc_set_state(name
, handler
, STOPPED
);
982 /* Avoid lingering namespace references. */
983 lxc_put_nsfds(handler
);
985 ret
= run_lxc_hooks(name
, "post-stop", handler
->conf
, NULL
);
987 ERROR("Failed to run lxc.hook.post-stop for container \"%s\"", name
);
988 if (handler
->conf
->reboot
> REBOOT_NONE
) {
989 WARN("Container will be stopped instead of rebooted");
990 handler
->conf
->reboot
= REBOOT_NONE
;
992 ret
= setenv("LXC_TARGET", "stop", 1);
994 WARN("Failed to set environment variable: LXC_TARGET=stop");
998 /* Reset mask set by setup_signal_fd. */
999 ret
= pthread_sigmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
1001 SYSWARN("Failed to restore signal mask");
1003 lxc_terminal_delete(&handler
->conf
->console
);
1004 lxc_delete_tty(&handler
->conf
->ttys
);
1006 /* The command socket is now closed, no more state clients can register
1007 * themselves from now on. So free the list of state clients.
1009 lxc_list_for_each_safe(cur
, &handler
->conf
->state_clients
, next
) {
1010 struct lxc_state_client
*client
= cur
->elem
;
1012 /* Keep state clients that want to be notified about reboots. */
1013 if ((handler
->conf
->reboot
> REBOOT_NONE
) &&
1014 (client
->states
[RUNNING
] == 2))
1017 /* close state client socket */
1019 close(client
->clientfd
);
1024 if (handler
->conf
->ephemeral
== 1 && handler
->conf
->reboot
!= REBOOT_REQ
)
1025 lxc_destroy_container_on_signal(handler
, name
);
1027 lxc_free_handler(handler
);
1030 void lxc_abort(const char *name
, struct lxc_handler
*handler
)
1035 lxc_set_state(name
, handler
, ABORTING
);
1037 if (handler
->pidfd
>= 0) {
1038 ret
= lxc_raw_pidfd_send_signal(handler
->pidfd
, SIGKILL
, NULL
, 0);
1040 SYSWARN("Failed to send SIGKILL via pidfd %d for process %d",
1041 handler
->pidfd
, handler
->pid
);
1044 if (!ret
|| errno
!= ESRCH
)
1045 if (kill(handler
->pid
, SIGKILL
))
1046 SYSWARN("Failed to send SIGKILL to %d", handler
->pid
);
1049 ret
= waitpid(-1, &status
, 0);
1053 static int do_start(void *data
)
1055 struct lxc_handler
*handler
= data
;
1056 __lxc_unused __do_close_prot_errno
int data_sock0
= handler
->data_sock
[0],
1057 data_sock1
= handler
->data_sock
[1];
1058 __do_close_prot_errno
int status_fd
= -EBADF
;
1062 struct lxc_list
*iterator
;
1065 int devnull_fd
= -1;
1067 lxc_sync_fini_parent(handler
);
1069 if (lxc_abstract_unix_recv_fds(data_sock1
, &status_fd
, 1, NULL
, 0) < 0) {
1070 ERROR("Failed to receive status file descriptor to child process");
1071 goto out_warn_father
;
1074 /* This prctl must be before the synchro, so if the parent dies before
1075 * we set the parent death signal, we will detect its death with the
1076 * synchro right after, otherwise we have a window where the parent can
1077 * exit before we set the pdeath signal leading to a unsupervized
1080 ret
= lxc_set_death_signal(SIGKILL
, handler
->monitor_pid
, status_fd
);
1082 SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL");
1083 goto out_warn_father
;
1086 ret
= lxc_ambient_caps_up();
1088 ERROR("Failed to raise ambient capabilities");
1089 goto out_warn_father
;
1092 ret
= pthread_sigmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
1094 SYSERROR("Failed to set signal mask");
1095 goto out_warn_father
;
1098 /* Don't leak the pinfd to the container. */
1099 close_prot_errno_disarm(handler
->pinfd
);
1101 ret
= lxc_sync_wait_parent(handler
, LXC_SYNC_STARTUP
);
1103 goto out_warn_father
;
1105 /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See
1106 * https://github.com/lxc/lxd/issues/1978.
1108 if ((handler
->ns_clone_flags
& (CLONE_NEWNET
| CLONE_NEWUSER
)) ==
1109 (CLONE_NEWNET
| CLONE_NEWUSER
)) {
1110 ret
= unshare(CLONE_NEWNET
);
1112 SYSERROR("Failed to unshare CLONE_NEWNET");
1113 goto out_warn_father
;
1115 INFO("Unshared CLONE_NEWNET");
1118 /* Tell the parent task it can begin to configure the container and wait
1121 ret
= lxc_sync_barrier_parent(handler
, LXC_SYNC_CONFIGURE
);
1125 if (handler
->ns_clone_flags
& CLONE_NEWNET
) {
1126 ret
= lxc_network_recv_from_parent(handler
);
1128 ERROR("Failed to receive veth names from parent");
1129 goto out_warn_father
;
1133 /* If we are in a new user namespace, become root there to have
1134 * privilege over our namespace.
1136 if (!lxc_list_empty(&handler
->conf
->id_map
)) {
1137 if (!handler
->conf
->root_nsuid_map
)
1138 nsuid
= handler
->conf
->init_uid
;
1140 if (!handler
->conf
->root_nsgid_map
)
1141 nsgid
= handler
->conf
->init_gid
;
1143 /* Drop groups only after we switched to a valid gid in the new
1146 if (!lxc_setgroups(0, NULL
) &&
1147 (handler
->am_root
|| errno
!= EPERM
))
1148 goto out_warn_father
;
1150 if (!lxc_switch_uid_gid(nsuid
, nsgid
))
1151 goto out_warn_father
;
1153 ret
= prctl(PR_SET_DUMPABLE
, prctl_arg(1), prctl_arg(0),
1154 prctl_arg(0), prctl_arg(0));
1156 goto out_warn_father
;
1158 /* set{g,u}id() clears deathsignal */
1159 ret
= lxc_set_death_signal(SIGKILL
, handler
->monitor_pid
, status_fd
);
1161 SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL");
1162 goto out_warn_father
;
1166 ret
= access(handler
->lxcpath
, X_OK
);
1168 print_top_failing_dir(handler
->lxcpath
);
1169 goto out_warn_father
;
1172 /* In order to checkpoint restore, we need to have everything in the
1173 * same mount namespace. However, some containers may not have a
1174 * reasonable /dev (in particular, they may not have /dev/null), so we
1175 * can't set init's std fds to /dev/null by opening it from inside the
1178 * If that's the case, fall back to using the host's /dev/null. This
1179 * means that migration won't work, but at least we won't spew output
1180 * where it isn't wanted.
1182 if (handler
->daemonize
&& !handler
->conf
->autodev
) {
1183 char path
[PATH_MAX
];
1185 ret
= snprintf(path
, sizeof(path
), "%s/dev/null",
1186 handler
->conf
->rootfs
.mount
);
1187 if (ret
< 0 || ret
>= sizeof(path
))
1188 goto out_warn_father
;
1190 ret
= access(path
, F_OK
);
1192 devnull_fd
= open_devnull();
1195 goto out_warn_father
;
1196 WARN("Using /dev/null from the host for container init's standard file descriptors. Migration will not work");
1200 /* Ask father to setup cgroups and wait for him to finish. */
1201 ret
= lxc_sync_barrier_parent(handler
, LXC_SYNC_CGROUP
);
1205 /* Unshare cgroup namespace after we have setup our cgroups. If we do it
1206 * earlier we end up with a wrong view of /proc/self/cgroup. For
1207 * example, assume we unshare(CLONE_NEWCGROUP) first, and then create
1208 * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then
1209 * /proc/self/cgroup would show us:
1213 * whereas it should actually show
1217 if (handler
->ns_clone_flags
& CLONE_NEWCGROUP
) {
1218 ret
= unshare(CLONE_NEWCGROUP
);
1220 if (errno
!= EINVAL
) {
1221 SYSERROR("Failed to unshare CLONE_NEWCGROUP");
1222 goto out_warn_father
;
1225 handler
->ns_clone_flags
&= ~CLONE_NEWCGROUP
;
1226 SYSINFO("Kernel does not support CLONE_NEWCGROUP");
1228 INFO("Unshared CLONE_NEWCGROUP");
1232 /* Add the requested environment variables to the current environment to
1233 * allow them to be used by the various hooks, such as the start hook
1236 lxc_list_for_each(iterator
, &handler
->conf
->environment
) {
1237 ret
= putenv((char *)iterator
->elem
);
1239 SYSERROR("Failed to set environment variable: %s",
1240 (char *)iterator
->elem
);
1241 goto out_warn_father
;
1245 /* Setup the container, ip, names, utsname, ... */
1246 ret
= lxc_setup(handler
);
1248 ERROR("Failed to setup container \"%s\"", handler
->name
);
1249 goto out_warn_father
;
1252 /* Set the label to change to when we exec(2) the container's init. */
1253 ret
= lsm_process_label_set(NULL
, handler
->conf
, true);
1255 goto out_warn_father
;
1257 /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it
1258 * before we aren't allowed anymore.
1260 if (handler
->conf
->no_new_privs
) {
1261 ret
= prctl(PR_SET_NO_NEW_PRIVS
, prctl_arg(1), prctl_arg(0),
1262 prctl_arg(0), prctl_arg(0));
1264 SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges");
1265 goto out_warn_father
;
1267 DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges");
1270 /* Some init's such as busybox will set sane tty settings on stdin,
1271 * stdout, stderr which it thinks is the console. We already set them
1272 * the way we wanted on the real terminal, and we want init to do its
1273 * setup on its console ie. the pty allocated in lxc_terminal_setup() so
1274 * make sure that that pty is stdin,stdout,stderr.
1276 if (handler
->conf
->console
.slave
>= 0) {
1277 if (handler
->daemonize
|| !handler
->conf
->is_execute
)
1278 ret
= set_stdfds(handler
->conf
->console
.slave
);
1280 ret
= lxc_terminal_set_stdfds(handler
->conf
->console
.slave
);
1282 ERROR("Failed to redirect std{in,out,err} to pty file descriptor %d",
1283 handler
->conf
->console
.slave
);
1284 goto out_warn_father
;
1288 /* If we mounted a temporary proc, then unmount it now. */
1289 tmp_proc_unmount(handler
->conf
);
1291 ret
= lxc_seccomp_load(handler
->conf
);
1293 goto out_warn_father
;
1295 ret
= lxc_seccomp_send_notifier_fd(&handler
->conf
->seccomp
, data_sock0
);
1297 SYSERROR("Failed to send seccomp notify fd to parent");
1298 goto out_warn_father
;
1301 ret
= run_lxc_hooks(handler
->name
, "start", handler
->conf
, NULL
);
1303 ERROR("Failed to run lxc.hook.start for container \"%s\"",
1305 goto out_warn_father
;
1308 close_prot_errno_disarm(handler
->sigfd
);
1310 if (handler
->conf
->console
.slave
< 0 && handler
->daemonize
) {
1311 if (devnull_fd
< 0) {
1312 devnull_fd
= open_devnull();
1314 goto out_warn_father
;
1317 ret
= set_stdfds(devnull_fd
);
1319 ERROR("Failed to redirect std{in,out,err} to \"/dev/null\"");
1320 goto out_warn_father
;
1324 close_prot_errno_disarm(devnull_fd
);
1328 if (handler
->conf
->init_cwd
) {
1329 ret
= chdir(handler
->conf
->init_cwd
);
1331 SYSERROR("Could not change directory to \"%s\"",
1332 handler
->conf
->init_cwd
);
1333 goto out_warn_father
;
1337 ret
= lxc_sync_barrier_parent(handler
, LXC_SYNC_CGROUP_LIMITS
);
1339 goto out_warn_father
;
1341 /* Reset the environment variables the user requested in a clear
1345 /* Don't error out though. */
1347 SYSERROR("Failed to clear environment.");
1349 lxc_list_for_each(iterator
, &handler
->conf
->environment
) {
1350 ret
= putenv((char *)iterator
->elem
);
1352 SYSERROR("Failed to set environment variable: %s",
1353 (char *)iterator
->elem
);
1354 goto out_warn_father
;
1358 ret
= putenv("container=lxc");
1360 SYSERROR("Failed to set environment variable: container=lxc");
1361 goto out_warn_father
;
1364 if (handler
->conf
->ttys
.tty_names
) {
1365 ret
= putenv(handler
->conf
->ttys
.tty_names
);
1367 SYSERROR("Failed to set environment variable for container ptys");
1368 goto out_warn_father
;
1372 /* The container has been setup. We can now switch to an unprivileged
1375 new_uid
= handler
->conf
->init_uid
;
1376 new_gid
= handler
->conf
->init_gid
;
1378 /* Avoid unnecessary syscalls. */
1379 if (new_uid
== nsuid
)
1380 new_uid
= LXC_INVALID_UID
;
1382 if (new_gid
== nsgid
)
1383 new_gid
= LXC_INVALID_GID
;
1385 /* If we are in a new user namespace we already dropped all groups when
1386 * we switched to root in the new user namespace further above. Only
1387 * drop groups if we can, so ensure that we have necessary privilege.
1389 if (lxc_list_empty(&handler
->conf
->id_map
))
1391 if (lxc_proc_cap_is_set(CAP_SETGID
, CAP_EFFECTIVE
))
1393 if (!lxc_setgroups(0, NULL
))
1394 goto out_warn_father
;
1396 if (!lxc_switch_uid_gid(new_uid
, new_gid
))
1397 goto out_warn_father
;
1399 ret
= lxc_ambient_caps_down();
1401 ERROR("Failed to clear ambient capabilities");
1402 goto out_warn_father
;
1405 if (handler
->conf
->monitor_signal_pdeath
!= SIGKILL
) {
1406 ret
= lxc_set_death_signal(handler
->conf
->monitor_signal_pdeath
,
1407 handler
->monitor_pid
, status_fd
);
1409 SYSERROR("Failed to set PR_SET_PDEATHSIG to %d",
1410 handler
->conf
->monitor_signal_pdeath
);
1411 goto out_warn_father
;
1415 /* After this call, we are in error because this ops should not return
1418 handler
->ops
->start(handler
, handler
->data
);
1421 /* We want the parent to know something went wrong, so we return a
1422 * special error code.
1424 lxc_sync_wake_parent(handler
, LXC_SYNC_ERROR
);
1427 close_prot_errno_disarm(devnull_fd
);
1432 static int lxc_recv_ttys_from_child(struct lxc_handler
*handler
)
1435 struct lxc_terminal_info
*tty
;
1437 int sock
= handler
->data_sock
[1];
1438 struct lxc_conf
*conf
= handler
->conf
;
1439 struct lxc_tty_info
*ttys
= &conf
->ttys
;
1441 if (!conf
->ttys
.max
)
1444 ttys
->tty
= malloc(sizeof(*ttys
->tty
) * ttys
->max
);
1448 for (i
= 0; i
< conf
->ttys
.max
; i
++) {
1451 ret
= lxc_abstract_unix_recv_fds(sock
, ttyfds
, 2, NULL
, 0);
1455 tty
= &ttys
->tty
[i
];
1457 tty
->master
= ttyfds
[0];
1458 tty
->slave
= ttyfds
[1];
1459 TRACE("Received pty with master fd %d and slave fd %d from child", tty
->master
, tty
->slave
);
1463 SYSERROR("Failed to receive %zu ttys from child", ttys
->max
);
1465 TRACE("Received %zu ttys from child", ttys
->max
);
1470 int resolve_clone_flags(struct lxc_handler
*handler
)
1473 struct lxc_conf
*conf
= handler
->conf
;
1475 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1476 if (conf
->ns_keep
) {
1477 if (!(conf
->ns_keep
& ns_info
[i
].clone_flag
))
1478 handler
->ns_clone_flags
|= ns_info
[i
].clone_flag
;
1479 } else if (conf
->ns_clone
) {
1480 if ((conf
->ns_clone
& ns_info
[i
].clone_flag
))
1481 handler
->ns_clone_flags
|= ns_info
[i
].clone_flag
;
1483 if (i
== LXC_NS_USER
&& lxc_list_empty(&handler
->conf
->id_map
))
1486 if (i
== LXC_NS_NET
&& lxc_requests_empty_network(handler
))
1489 if (i
== LXC_NS_CGROUP
&& !cgns_supported())
1492 handler
->ns_clone_flags
|= ns_info
[i
].clone_flag
;
1495 if (!conf
->ns_share
[i
])
1498 handler
->ns_clone_flags
&= ~ns_info
[i
].clone_flag
;
1499 TRACE("Sharing %s namespace", ns_info
[i
].proc_name
);
1505 /* Note that this function is used with clone(CLONE_VM). Some glibc versions
1506 * used to reset the pid/tid to -1 when CLONE_VM was used without CLONE_THREAD.
1507 * But since the memory between parent and child is shared on CLONE_VM this
1508 * would invalidate the getpid() cache that glibc used to maintain and so
1509 * getpid() in the child would return the parent's pid. This is all fixed in
1510 * newer glibc versions where the getpid() cache is removed and the pid/tid is
1511 * not reset anymore.
1512 * However, if for whatever reason you - dear committer - somehow need to get the
1513 * pid of the dummy intermediate process for do_share_ns() you need to call
1514 * lxc_raw_getpid(). The next lxc_raw_clone() call does not employ CLONE_VM and
1517 static inline int do_share_ns(void *arg
)
1520 struct lxc_handler
*handler
= arg
;
1522 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1523 if (handler
->nsfd
[i
] < 0)
1526 ret
= setns(handler
->nsfd
[i
], 0);
1529 * Note that joining a user and/or mount namespace
1530 * requires the process is not multithreaded otherwise
1531 * setns() will fail here.
1533 SYSERROR("Failed to inherit %s namespace",
1534 ns_info
[i
].proc_name
);
1538 DEBUG("Inherited %s namespace", ns_info
[i
].proc_name
);
1541 flags
= handler
->ns_on_clone_flags
;
1542 flags
|= CLONE_PARENT
;
1543 handler
->pid
= lxc_raw_clone_cb(do_start
, handler
, CLONE_PIDFD
| flags
,
1545 if (handler
->pid
< 0)
1551 /* lxc_spawn() performs crucial setup tasks and clone()s the new process which
1552 * exec()s the requested container binary.
1553 * Note that lxc_spawn() runs in the parent namespaces. Any operations performed
1554 * right here should be double checked if they'd pose a security risk. (For
1555 * example, any {u}mount() operations performed here will be reflected on the
1558 static int lxc_spawn(struct lxc_handler
*handler
)
1560 __do_close_prot_errno
int data_sock0
= -EBADF
, data_sock1
= -EBADF
;
1563 bool wants_to_map_ids
;
1564 struct lxc_list
*id_map
;
1565 const char *name
= handler
->name
;
1566 const char *lxcpath
= handler
->lxcpath
;
1567 bool share_ns
= false;
1568 struct lxc_conf
*conf
= handler
->conf
;
1569 struct cgroup_ops
*cgroup_ops
= handler
->cgroup_ops
;
1571 id_map
= &conf
->id_map
;
1572 wants_to_map_ids
= !lxc_list_empty(id_map
);
1574 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1575 if (!conf
->ns_share
[i
])
1578 handler
->nsfd
[i
] = lxc_inherit_namespace(conf
->ns_share
[i
], lxcpath
, ns_info
[i
].proc_name
);
1579 if (handler
->nsfd
[i
] < 0)
1585 ret
= lxc_sync_init(handler
);
1589 ret
= socketpair(AF_UNIX
, SOCK_STREAM
| SOCK_CLOEXEC
, 0,
1590 handler
->data_sock
);
1593 data_sock0
= handler
->data_sock
[0];
1594 data_sock1
= handler
->data_sock
[1];
1596 ret
= resolve_clone_flags(handler
);
1600 if (handler
->ns_clone_flags
& CLONE_NEWNET
) {
1601 ret
= lxc_find_gateway_addresses(handler
);
1603 ERROR("Failed to find gateway addresses");
1608 if (!cgroup_ops
->payload_create(cgroup_ops
, handler
)) {
1609 ERROR("Failed creating cgroups");
1610 goto out_delete_net
;
1613 /* If the rootfs is not a blockdev, prevent the container from marking
1615 * If the container is unprivileged then skip rootfs pinning.
1617 if (!wants_to_map_ids
) {
1618 handler
->pinfd
= pin_rootfs(conf
->rootfs
.path
);
1619 if (handler
->pinfd
== -EBADF
)
1620 INFO("Failed to pin the rootfs for container \"%s\"", handler
->name
);
1623 /* Create a process in a new set of namespaces. */
1624 handler
->ns_on_clone_flags
= handler
->ns_clone_flags
;
1625 if (handler
->ns_clone_flags
& CLONE_NEWUSER
) {
1626 /* If CLONE_NEWUSER and CLONE_NEWNET was requested, we need to
1627 * clone a new user namespace first and only later unshare our
1628 * network namespace to ensure that network devices ownership is
1631 handler
->ns_on_clone_flags
&= ~CLONE_NEWNET
;
1633 /* The cgroup namespace gets unshare()ed not clone()ed. */
1634 handler
->ns_on_clone_flags
&= ~CLONE_NEWCGROUP
;
1639 attacher_pid
= lxc_clone(do_share_ns
, handler
,
1640 CLONE_VFORK
| CLONE_VM
| CLONE_FILES
, NULL
);
1641 if (attacher_pid
< 0) {
1642 SYSERROR(LXC_CLONE_ERROR
);
1643 goto out_delete_net
;
1646 ret
= wait_for_pid(attacher_pid
);
1648 SYSERROR("Intermediate process failed");
1649 goto out_delete_net
;
1652 handler
->pid
= lxc_raw_clone_cb(do_start
, handler
,
1653 CLONE_PIDFD
| handler
->ns_on_clone_flags
,
1656 if (handler
->pid
< 0) {
1657 SYSERROR(LXC_CLONE_ERROR
);
1658 goto out_delete_net
;
1660 TRACE("Cloned child process %d", handler
->pid
);
1662 /* Verify that we can actually make use of pidfds. */
1663 if (!lxc_can_use_pidfd(handler
->pidfd
))
1664 close_prot_errno_disarm(handler
->pidfd
);
1666 ret
= snprintf(pidstr
, 20, "%d", handler
->pid
);
1667 if (ret
< 0 || ret
>= 20)
1668 goto out_delete_net
;
1670 ret
= setenv("LXC_PID", pidstr
, 1);
1672 SYSERROR("Failed to set environment variable: LXC_PID=%s", pidstr
);
1674 for (i
= 0; i
< LXC_NS_MAX
; i
++)
1675 if (handler
->ns_on_clone_flags
& ns_info
[i
].clone_flag
)
1676 INFO("Cloned %s", ns_info
[i
].flag_name
);
1678 if (!lxc_try_preserve_namespaces(handler
, handler
->ns_on_clone_flags
, handler
->pid
)) {
1679 ERROR("Failed to preserve cloned namespaces for lxc.hook.stop");
1680 goto out_delete_net
;
1683 lxc_sync_fini_child(handler
);
1685 if (lxc_abstract_unix_send_fds(handler
->data_sock
[0], &handler
->monitor_status_fd
, 1, NULL
, 0) < 0) {
1686 ERROR("Failed to send status file descriptor to child process");
1687 goto out_delete_net
;
1689 close_prot_errno_disarm(handler
->monitor_status_fd
);
1691 /* Map the container uids. The container became an invalid userid the
1692 * moment it was cloned with CLONE_NEWUSER. This call doesn't change
1693 * anything immediately, but allows the container to setuid(0) (0 being
1694 * mapped to something else on the host.) later to become a valid uid
1697 if (wants_to_map_ids
) {
1698 if (!handler
->conf
->ns_share
[LXC_NS_USER
] &&
1699 (handler
->conf
->ns_keep
& CLONE_NEWUSER
) == 0) {
1700 ret
= lxc_map_ids(id_map
, handler
->pid
);
1702 ERROR("Failed to set up id mapping.");
1703 goto out_delete_net
;
1708 ret
= lxc_sync_wake_child(handler
, LXC_SYNC_STARTUP
);
1710 goto out_delete_net
;
1712 ret
= lxc_sync_wait_child(handler
, LXC_SYNC_CONFIGURE
);
1714 goto out_delete_net
;
1716 if (!cgroup_ops
->setup_limits_legacy(cgroup_ops
, handler
->conf
, false)) {
1717 ERROR("Failed to setup cgroup limits for container \"%s\"", name
);
1718 goto out_delete_net
;
1721 if (!cgroup_ops
->payload_enter(cgroup_ops
, handler
)) {
1722 goto out_delete_net
;
1725 if (!cgroup_ops
->payload_delegate_controllers(cgroup_ops
)) {
1726 ERROR("Failed to delegate controllers to payload cgroup");
1727 goto out_delete_net
;
1730 if (!cgroup_ops
->setup_limits(cgroup_ops
, handler
)) {
1731 ERROR("Failed to setup cgroup limits for container \"%s\"", name
);
1732 goto out_delete_net
;
1735 if (!cgroup_ops
->chown(cgroup_ops
, handler
->conf
))
1736 goto out_delete_net
;
1738 /* If not done yet, we're now ready to preserve the network namespace */
1739 if (handler
->nsfd
[LXC_NS_NET
] < 0) {
1740 ret
= lxc_try_preserve_ns(handler
->pid
, "net");
1742 if (ret
!= -EOPNOTSUPP
) {
1743 SYSERROR("Failed to preserve net namespace");
1744 goto out_delete_net
;
1747 handler
->nsfd
[LXC_NS_NET
] = ret
;
1748 DEBUG("Preserved net namespace via fd %d", ret
);
1751 ret
= lxc_netns_set_nsid(handler
->nsfd
[LXC_NS_NET
]);
1753 SYSWARN("Failed to allocate new network namespace id");
1755 TRACE("Allocated new network namespace id");
1757 /* Create the network configuration. */
1758 if (handler
->ns_clone_flags
& CLONE_NEWNET
) {
1759 ret
= lxc_create_network(handler
);
1761 ERROR("Failed to create the network");
1762 goto out_delete_net
;
1765 ret
= lxc_network_send_to_child(handler
);
1767 ERROR("Failed to send veth names to child");
1768 goto out_delete_net
;
1772 if (!lxc_list_empty(&conf
->procs
)) {
1773 ret
= setup_proc_filesystem(&conf
->procs
, handler
->pid
);
1775 goto out_delete_net
;
1778 /* Tell the child to continue its initialization. We'll get
1779 * LXC_SYNC_CGROUP when it is ready for us to setup cgroups.
1781 ret
= lxc_sync_barrier_child(handler
, LXC_SYNC_POST_CONFIGURE
);
1783 goto out_delete_net
;
1785 if (!lxc_list_empty(&conf
->limits
)) {
1786 ret
= setup_resource_limits(&conf
->limits
, handler
->pid
);
1788 ERROR("Failed to setup resource limits");
1789 goto out_delete_net
;
1793 ret
= lxc_sync_barrier_child(handler
, LXC_SYNC_CGROUP_UNSHARE
);
1795 goto out_delete_net
;
1797 if (!cgroup_ops
->setup_limits_legacy(cgroup_ops
, handler
->conf
, true)) {
1798 ERROR("Failed to setup legacy device cgroup controller limits");
1799 goto out_delete_net
;
1801 TRACE("Set up legacy device cgroup controller limits");
1803 if (!cgroup_ops
->devices_activate(cgroup_ops
, handler
)) {
1804 ERROR("Failed to setup cgroup2 device controller limits");
1805 goto out_delete_net
;
1807 TRACE("Set up cgroup2 device controller limits");
1809 if (handler
->ns_clone_flags
& CLONE_NEWCGROUP
) {
1810 /* Now we're ready to preserve the cgroup namespace */
1811 ret
= lxc_try_preserve_ns(handler
->pid
, "cgroup");
1813 if (ret
!= -EOPNOTSUPP
) {
1814 SYSERROR("Failed to preserve cgroup namespace");
1815 goto out_delete_net
;
1818 handler
->nsfd
[LXC_NS_CGROUP
] = ret
;
1819 DEBUG("Preserved cgroup namespace via fd %d", ret
);
1823 cgroup_ops
->payload_finalize(cgroup_ops
);
1824 TRACE("Finished setting up cgroups");
1826 /* Run any host-side start hooks */
1827 ret
= run_lxc_hooks(name
, "start-host", conf
, NULL
);
1829 ERROR("Failed to run lxc.hook.start-host");
1830 goto out_delete_net
;
1833 /* Tell the child to complete its initialization and wait for it to exec
1834 * or return an error. (The child will never return
1835 * LXC_SYNC_READY_START+1. It will either close the sync pipe, causing
1836 * lxc_sync_barrier_child to return success, or return a different
1837 * value, causing us to error out).
1839 ret
= lxc_sync_barrier_child(handler
, LXC_SYNC_READY_START
);
1841 goto out_delete_net
;
1843 if (handler
->ns_clone_flags
& CLONE_NEWNET
) {
1844 ret
= lxc_network_recv_name_and_ifindex_from_child(handler
);
1846 ERROR("Failed to receive names and ifindices for network devices from child");
1847 goto out_delete_net
;
1851 /* Now all networks are created, network devices are moved into place,
1852 * and the correct names and ifindices in the respective namespaces have
1853 * been recorded. The corresponding structs have now all been filled. So
1854 * log them for debugging purposes.
1856 lxc_log_configured_netdevs(conf
);
1858 /* Read tty fds allocated by child. */
1859 ret
= lxc_recv_ttys_from_child(handler
);
1861 ERROR("Failed to receive tty info from child process");
1862 goto out_delete_net
;
1865 ret
= lxc_seccomp_recv_notifier_fd(&handler
->conf
->seccomp
, data_sock1
);
1867 SYSERROR("Failed to receive seccomp notify fd from child");
1868 goto out_delete_net
;
1871 ret
= handler
->ops
->post_start(handler
, handler
->data
);
1875 ret
= lxc_set_state(name
, handler
, RUNNING
);
1877 ERROR("Failed to set state to \"%s\"", lxc_state2str(RUNNING
));
1881 lxc_sync_fini(handler
);
1886 if (handler
->ns_clone_flags
& CLONE_NEWNET
)
1887 lxc_delete_network(handler
);
1890 lxc_abort(name
, handler
);
1893 lxc_sync_fini(handler
);
1894 close_prot_errno_disarm(handler
->pinfd
);
1899 int __lxc_start(const char *name
, struct lxc_handler
*handler
,
1900 struct lxc_operations
* ops
, void *data
, const char *lxcpath
,
1901 bool daemonize
, int *error_num
)
1904 struct lxc_conf
*conf
= handler
->conf
;
1905 struct cgroup_ops
*cgroup_ops
;
1907 ret
= lxc_init(name
, handler
);
1909 ERROR("Failed to initialize container \"%s\"", name
);
1910 goto out_fini_nonet
;
1913 handler
->data
= data
;
1914 handler
->daemonize
= daemonize
;
1915 cgroup_ops
= handler
->cgroup_ops
;
1917 if (!attach_block_device(handler
->conf
)) {
1918 ERROR("Failed to attach block device");
1920 goto out_fini_nonet
;
1923 if (!cgroup_ops
->monitor_create(cgroup_ops
, handler
)) {
1924 ERROR("Failed to create monitor cgroup");
1926 goto out_fini_nonet
;
1929 if (!cgroup_ops
->monitor_enter(cgroup_ops
, handler
)) {
1930 ERROR("Failed to enter monitor cgroup");
1932 goto out_fini_nonet
;
1935 if (!cgroup_ops
->monitor_delegate_controllers(cgroup_ops
)) {
1936 ERROR("Failed to delegate controllers to monitor cgroup");
1938 goto out_fini_nonet
;
1941 if (geteuid() == 0 && !lxc_list_empty(&conf
->id_map
)) {
1942 /* If the backing store is a device, mount it here and now. */
1943 if (rootfs_is_blockdev(conf
)) {
1944 ret
= unshare(CLONE_NEWNS
);
1946 ERROR("Failed to unshare CLONE_NEWNS");
1947 goto out_fini_nonet
;
1949 INFO("Unshared CLONE_NEWNS");
1951 remount_all_slave();
1952 ret
= lxc_setup_rootfs_prepare_root(conf
, name
, lxcpath
);
1954 ERROR("Error setting up rootfs mount as root before spawn");
1955 goto out_fini_nonet
;
1957 INFO("Set up container rootfs as host root");
1961 ret
= lxc_spawn(handler
);
1963 ERROR("Failed to spawn container \"%s\"", name
);
1964 goto out_detach_blockdev
;
1967 handler
->conf
->reboot
= REBOOT_NONE
;
1969 ret
= lxc_poll(name
, handler
);
1971 ERROR("LXC mainloop exited with error: %d", ret
);
1975 if (!handler
->init_died
&& handler
->pid
> 0) {
1976 ERROR("Child process is not killed");
1981 status
= lxc_wait_for_pid_status(handler
->pid
);
1983 SYSERROR("Failed to retrieve status for %d", handler
->pid
);
1985 /* If the child process exited but was not signaled, it didn't call
1986 * reboot. This should mean it was an lxc-execute which simply exited.
1987 * In any case, treat it as a 'halt'.
1989 if (WIFSIGNALED(status
)) {
1990 switch(WTERMSIG(status
)) {
1991 case SIGINT
: /* halt */
1992 DEBUG("Container \"%s\" is halting", name
);
1994 case SIGHUP
: /* reboot */
1995 DEBUG("Container \"%s\" is rebooting", name
);
1996 handler
->conf
->reboot
= REBOOT_REQ
;
1998 case SIGSYS
: /* seccomp */
1999 DEBUG("Container \"%s\" violated its seccomp policy", name
);
2002 DEBUG("Unknown exit status for container \"%s\" init %d", name
, WTERMSIG(status
));
2007 ret
= lxc_restore_phys_nics_to_netns(handler
);
2009 ERROR("Failed to move physical network devices back to parent network namespace");
2011 close_prot_errno_disarm(handler
->pinfd
);
2013 lxc_monitor_send_exit_code(name
, status
, handler
->lxcpath
);
2014 lxc_error_set_and_log(handler
->pid
, status
);
2016 *error_num
= handler
->exit_status
;
2019 lxc_delete_network(handler
);
2021 out_detach_blockdev
:
2022 detach_block_device(handler
->conf
);
2025 lxc_fini(name
, handler
);
2029 lxc_abort(name
, handler
);
2037 static int start(struct lxc_handler
*handler
, void* data
)
2039 struct start_args
*arg
= data
;
2041 NOTICE("Exec'ing \"%s\"", arg
->argv
[0]);
2043 execvp(arg
->argv
[0], arg
->argv
);
2044 SYSERROR("Failed to exec \"%s\"", arg
->argv
[0]);
2048 static int post_start(struct lxc_handler
*handler
, void* data
)
2050 struct start_args
*arg
= data
;
2052 NOTICE("Started \"%s\" with pid \"%d\"", arg
->argv
[0], handler
->pid
);
2056 static struct lxc_operations start_ops
= {
2058 .post_start
= post_start
2061 int lxc_start(const char *name
, char *const argv
[], struct lxc_handler
*handler
,
2062 const char *lxcpath
, bool daemonize
, int *error_num
)
2064 struct start_args start_arg
= {
2068 TRACE("Doing lxc_start");
2069 return __lxc_start(name
, handler
, &start_ops
, &start_arg
, lxcpath
, daemonize
, error_num
);
2072 static void lxc_destroy_container_on_signal(struct lxc_handler
*handler
,
2075 char destroy
[PATH_MAX
];
2076 struct lxc_container
*c
;
2080 if (handler
->conf
->rootfs
.path
&& handler
->conf
->rootfs
.mount
) {
2081 bret
= do_destroy_container(handler
);
2083 ERROR("Error destroying rootfs for container \"%s\"", name
);
2087 INFO("Destroyed rootfs for container \"%s\"", name
);
2089 ret
= snprintf(destroy
, PATH_MAX
, "%s/%s", handler
->lxcpath
, name
);
2090 if (ret
< 0 || ret
>= PATH_MAX
) {
2091 ERROR("Error destroying directory for container \"%s\"", name
);
2095 c
= lxc_container_new(name
, handler
->lxcpath
);
2097 if (container_disk_lock(c
)) {
2098 INFO("Could not update lxc_snapshots file");
2099 lxc_container_put(c
);
2101 mod_all_rdeps(c
, false);
2102 container_disk_unlock(c
);
2103 lxc_container_put(c
);
2107 if (!handler
->am_root
)
2108 ret
= userns_exec_full(handler
->conf
, lxc_rmdir_onedev_wrapper
,
2109 destroy
, "lxc_rmdir_onedev_wrapper");
2111 ret
= lxc_rmdir_onedev(destroy
, NULL
);
2114 ERROR("Error destroying directory for container \"%s\"", name
);
2117 INFO("Destroyed directory for container \"%s\"", name
);
2120 static int lxc_rmdir_onedev_wrapper(void *data
)
2122 char *arg
= (char *) data
;
2123 return lxc_rmdir_onedev(arg
, NULL
);
2126 static bool do_destroy_container(struct lxc_handler
*handler
)
2130 if (!handler
->am_root
) {
2131 ret
= userns_exec_full(handler
->conf
, storage_destroy_wrapper
,
2132 handler
->conf
, "storage_destroy_wrapper");
2139 return storage_destroy(handler
->conf
);