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
->pidfd
);
622 close_prot_errno_disarm(handler
->sigfd
);
623 lxc_put_nsfds(handler
);
624 if (handler
->conf
&& handler
->conf
->reboot
== REBOOT_NONE
)
625 close_prot_errno_disarm(handler
->conf
->maincmd_fd
);
626 close_prot_errno_disarm(handler
->monitor_status_fd
);
627 close_prot_errno_disarm(handler
->state_socket_pair
[0]);
628 close_prot_errno_disarm(handler
->state_socket_pair
[1]);
629 cgroup_exit(handler
->cgroup_ops
);
630 if (handler
->conf
&& handler
->conf
->reboot
== REBOOT_NONE
)
631 free_disarm(handler
);
633 handler
->conf
= NULL
;
636 struct lxc_handler
*lxc_init_handler(struct lxc_handler
*old
,
637 const char *name
, struct lxc_conf
*conf
,
638 const char *lxcpath
, bool daemonize
)
642 struct lxc_handler
*handler
;
645 handler
= zalloc(sizeof(*handler
));
651 /* Note that am_guest_unpriv() checks the effective uid. We
652 * probably don't care if we are real root only if we are running
653 * as root so this should be fine.
655 handler
->am_root
= !am_guest_unpriv();
656 handler
->conf
= conf
;
657 handler
->lxcpath
= lxcpath
;
658 handler
->init_died
= false;
659 handler
->data_sock
[0] = -EBADF
;
660 handler
->data_sock
[1] = -EBADF
;
661 handler
->monitor_status_fd
= -EBADF
;
662 handler
->pidfd
= -EBADF
;
663 handler
->sigfd
= -EBADF
;
664 handler
->state_socket_pair
[0] = -EBADF
;
665 handler
->state_socket_pair
[1] = -EBADF
;
666 if (handler
->conf
->reboot
== REBOOT_NONE
)
667 lxc_list_init(&handler
->conf
->state_clients
);
669 for (int i
= 0; i
< LXC_NS_MAX
; i
++)
670 handler
->nsfd
[i
] = -EBADF
;
672 handler
->name
= name
;
674 handler
->transient_pid
= lxc_raw_getpid();
676 handler
->transient_pid
= -1;
678 if (daemonize
&& handler
->conf
->reboot
== REBOOT_NONE
) {
679 /* Create socketpair() to synchronize on daemonized startup.
680 * When the container reboots we don't need to synchronize
681 * again currently so don't open another socketpair().
683 ret
= socketpair(AF_UNIX
, SOCK_STREAM
| SOCK_CLOEXEC
, 0,
684 handler
->state_socket_pair
);
686 ERROR("Failed to create anonymous pair of unix sockets");
690 TRACE("Created anonymous pair {%d,%d} of unix sockets",
691 handler
->state_socket_pair
[0],
692 handler
->state_socket_pair
[1]);
693 handler
->keep_fds
[nr_keep_fds
++] = handler
->state_socket_pair
[0];
694 handler
->keep_fds
[nr_keep_fds
++] = handler
->state_socket_pair
[1];
697 if (handler
->conf
->reboot
== REBOOT_NONE
) {
698 handler
->conf
->maincmd_fd
= lxc_cmd_init(name
, lxcpath
, "command");
699 if (handler
->conf
->maincmd_fd
< 0) {
700 ERROR("Failed to set up command socket");
703 handler
->keep_fds
[nr_keep_fds
++] = handler
->conf
->maincmd_fd
;
706 TRACE("Unix domain socket %d for command server is ready",
707 handler
->conf
->maincmd_fd
);
712 lxc_put_handler(handler
);
717 int lxc_init(const char *name
, struct lxc_handler
*handler
)
719 __do_close
int status_fd
= -EBADF
;
721 const char *loglevel
;
722 struct lxc_conf
*conf
= handler
->conf
;
724 handler
->monitor_pid
= lxc_raw_getpid();
725 status_fd
= open("/proc/self/status", O_RDONLY
| O_CLOEXEC
);
727 return log_error_errno(-1, errno
, "Failed to open monitor status fd");
729 handler
->lsm_ops
= lsm_init_static();
730 TRACE("Initialized LSM");
732 /* Begin by setting the state to STARTING. */
733 ret
= lxc_set_state(name
, handler
, STARTING
);
735 return log_error(-1, "Failed to set state to \"%s\"", lxc_state2str(STARTING
));
736 TRACE("Set container state to \"STARTING\"");
738 /* Start of environment variable setup for hooks. */
739 ret
= setenv("LXC_NAME", name
, 1);
741 SYSERROR("Failed to set environment variable: LXC_NAME=%s", name
);
744 ret
= setenv("LXC_CONFIG_FILE", conf
->rcfile
, 1);
746 SYSERROR("Failed to set environment variable: LXC_CONFIG_FILE=%s", conf
->rcfile
);
749 if (conf
->rootfs
.mount
) {
750 ret
= setenv("LXC_ROOTFS_MOUNT", conf
->rootfs
.mount
, 1);
752 SYSERROR("Failed to set environment variable: LXC_ROOTFS_MOUNT=%s", conf
->rootfs
.mount
);
755 if (conf
->rootfs
.path
) {
756 ret
= setenv("LXC_ROOTFS_PATH", conf
->rootfs
.path
, 1);
758 SYSERROR("Failed to set environment variable: LXC_ROOTFS_PATH=%s", conf
->rootfs
.path
);
761 if (conf
->console
.path
) {
762 ret
= setenv("LXC_CONSOLE", conf
->console
.path
, 1);
764 SYSERROR("Failed to set environment variable: LXC_CONSOLE=%s", conf
->console
.path
);
767 if (conf
->console
.log_path
) {
768 ret
= setenv("LXC_CONSOLE_LOGPATH", conf
->console
.log_path
, 1);
770 SYSERROR("Failed to set environment variable: LXC_CONSOLE_LOGPATH=%s", conf
->console
.log_path
);
773 if (cgns_supported()) {
774 ret
= setenv("LXC_CGNS_AWARE", "1", 1);
776 SYSERROR("Failed to set environment variable LXC_CGNS_AWARE=1");
779 loglevel
= lxc_log_priority_to_string(lxc_log_get_level());
780 ret
= setenv("LXC_LOG_LEVEL", loglevel
, 1);
782 SYSERROR("Set environment variable LXC_LOG_LEVEL=%s", loglevel
);
784 if (conf
->hooks_version
== 0)
785 ret
= setenv("LXC_HOOK_VERSION", "0", 1);
787 ret
= setenv("LXC_HOOK_VERSION", "1", 1);
789 SYSERROR("Failed to set environment variable LXC_HOOK_VERSION=%u", conf
->hooks_version
);
790 /* End of environment variable setup for hooks. */
792 TRACE("Set environment variables");
794 ret
= run_lxc_hooks(name
, "pre-start", conf
, NULL
);
796 return log_error(-1, "Failed to run lxc.hook.pre-start for container \"%s\"", name
);
797 TRACE("Ran pre-start hooks");
799 /* The signal fd has to be created before forking otherwise if the child
800 * process exits before we setup the signal fd, the event will be lost
801 * and the command will be stuck.
803 handler
->sigfd
= setup_signal_fd(&handler
->oldmask
);
804 if (handler
->sigfd
< 0)
805 return log_error(-1, "Failed to setup SIGCHLD fd handler.");
806 TRACE("Set up signal fd");
808 /* Do this after setting up signals since it might unblock SIGWINCH. */
809 ret
= lxc_terminal_setup(conf
);
811 ERROR("Failed to create console");
812 goto out_restore_sigmask
;
814 TRACE("Created console");
816 handler
->cgroup_ops
= cgroup_init(handler
->conf
);
817 if (!handler
->cgroup_ops
) {
818 ERROR("Failed to initialize cgroup driver");
819 goto out_delete_terminal
;
821 TRACE("Initialized cgroup driver");
823 ret
= lxc_read_seccomp_config(conf
);
825 return log_error(-1, "Failed loading seccomp policy");
826 TRACE("Read seccomp policy");
828 ret
= handler
->lsm_ops
->prepare(handler
->lsm_ops
, conf
, handler
->lxcpath
);
830 ERROR("Failed to initialize LSM");
831 goto out_delete_terminal
;
833 TRACE("Initialized LSM");
835 INFO("Container \"%s\" is initialized", name
);
836 handler
->monitor_status_fd
= move_fd(status_fd
);
840 lxc_terminal_delete(&handler
->conf
->console
);
843 (void)pthread_sigmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
848 void lxc_end(struct lxc_handler
*handler
)
852 struct lxc_list
*cur
, *next
;
853 char *namespaces
[LXC_NS_MAX
+ 1];
854 size_t namespace_count
= 0;
855 const char *name
= handler
->name
;
856 struct cgroup_ops
*cgroup_ops
= handler
->cgroup_ops
;
858 /* The STOPPING state is there for future cleanup code which can take
861 lxc_set_state(name
, handler
, STOPPING
);
863 self
= lxc_raw_getpid();
864 for (int i
= 0; i
< LXC_NS_MAX
; i
++) {
865 if (handler
->nsfd
[i
] < 0)
868 if (handler
->conf
->hooks_version
== 0)
869 ret
= asprintf(&namespaces
[namespace_count
],
870 "%s:/proc/%d/fd/%d", ns_info
[i
].proc_name
,
871 self
, handler
->nsfd
[i
]);
873 ret
= asprintf(&namespaces
[namespace_count
],
874 "/proc/%d/fd/%d", self
, handler
->nsfd
[i
]);
876 SYSERROR("Failed to allocate memory");
880 if (handler
->conf
->hooks_version
== 0) {
885 ret
= setenv(ns_info
[i
].env_name
, namespaces
[namespace_count
], 1);
887 SYSERROR("Failed to set environment variable %s=%s",
888 ns_info
[i
].env_name
, namespaces
[namespace_count
]);
890 TRACE("Set environment variable %s=%s",
891 ns_info
[i
].env_name
, namespaces
[namespace_count
]);
895 namespaces
[namespace_count
] = NULL
;
897 if (handler
->conf
->reboot
> REBOOT_NONE
) {
898 ret
= setenv("LXC_TARGET", "reboot", 1);
900 SYSERROR("Failed to set environment variable: LXC_TARGET=reboot");
903 if (handler
->conf
->reboot
== REBOOT_NONE
) {
904 ret
= setenv("LXC_TARGET", "stop", 1);
906 SYSERROR("Failed to set environment variable: LXC_TARGET=stop");
909 if (handler
->conf
->hooks_version
== 0)
910 ret
= run_lxc_hooks(name
, "stop", handler
->conf
, namespaces
);
912 ret
= run_lxc_hooks(name
, "stop", handler
->conf
, NULL
);
914 ERROR("Failed to run \"lxc.hook.stop\" hook");
916 while (namespace_count
--)
917 free(namespaces
[namespace_count
]);
919 handler
->lsm_ops
->cleanup(handler
->lsm_ops
, handler
->conf
, handler
->lxcpath
);
922 cgroup_ops
->payload_destroy(cgroup_ops
, handler
);
923 cgroup_ops
->monitor_destroy(cgroup_ops
, handler
);
926 put_lxc_rootfs(&handler
->conf
->rootfs
, true);
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 if (!lxc_sync_wait_parent(handler
, START_SYNC_STARTUP
))
1070 goto out_warn_father
;
1072 /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See
1073 * https://github.com/lxc/lxd/issues/1978.
1075 if (handler
->ns_unshare_flags
& CLONE_NEWNET
) {
1076 ret
= unshare(CLONE_NEWNET
);
1078 SYSERROR("Failed to unshare CLONE_NEWNET");
1079 goto out_warn_father
;
1081 INFO("Unshared CLONE_NEWNET");
1084 /* Tell the parent task it can begin to configure the container and wait
1087 if (!lxc_sync_barrier_parent(handler
, START_SYNC_CONFIGURE
))
1090 if (handler
->ns_clone_flags
& CLONE_NEWNET
) {
1091 ret
= lxc_network_recv_from_parent(handler
);
1093 ERROR("Failed to receive veth names from parent");
1094 goto out_warn_father
;
1098 /* If we are in a new user namespace, become root there to have
1099 * privilege over our namespace.
1101 if (!lxc_list_empty(&handler
->conf
->id_map
)) {
1102 if (!handler
->conf
->root_nsuid_map
)
1103 nsuid
= handler
->conf
->init_uid
;
1105 if (!handler
->conf
->root_nsgid_map
)
1106 nsgid
= handler
->conf
->init_gid
;
1108 /* Drop groups only after we switched to a valid gid in the new
1111 if (!lxc_drop_groups() &&
1112 (handler
->am_root
|| errno
!= EPERM
))
1113 goto out_warn_father
;
1115 if (!lxc_switch_uid_gid(nsuid
, nsgid
))
1116 goto out_warn_father
;
1118 ret
= prctl(PR_SET_DUMPABLE
, prctl_arg(1), prctl_arg(0),
1119 prctl_arg(0), prctl_arg(0));
1121 goto out_warn_father
;
1123 /* set{g,u}id() clears deathsignal */
1124 ret
= lxc_set_death_signal(SIGKILL
, handler
->monitor_pid
, status_fd
);
1126 SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL");
1127 goto out_warn_father
;
1131 ret
= access(handler
->lxcpath
, X_OK
);
1133 print_top_failing_dir(handler
->lxcpath
);
1134 goto out_warn_father
;
1137 /* In order to checkpoint restore, we need to have everything in the
1138 * same mount namespace. However, some containers may not have a
1139 * reasonable /dev (in particular, they may not have /dev/null), so we
1140 * can't set init's std fds to /dev/null by opening it from inside the
1143 * If that's the case, fall back to using the host's /dev/null. This
1144 * means that migration won't work, but at least we won't spew output
1145 * where it isn't wanted.
1147 if (handler
->daemonize
&& !handler
->conf
->autodev
) {
1148 char path
[PATH_MAX
];
1150 ret
= snprintf(path
, sizeof(path
), "%s/dev/null",
1151 handler
->conf
->rootfs
.mount
);
1152 if (ret
< 0 || ret
>= sizeof(path
))
1153 goto out_warn_father
;
1155 ret
= access(path
, F_OK
);
1157 devnull_fd
= open_devnull();
1160 goto out_warn_father
;
1161 WARN("Using /dev/null from the host for container init's standard file descriptors. Migration will not work");
1165 /* Ask father to setup cgroups and wait for him to finish. */
1166 if (!lxc_sync_barrier_parent(handler
, START_SYNC_CGROUP
))
1169 /* Unshare cgroup namespace after we have setup our cgroups. If we do it
1170 * earlier we end up with a wrong view of /proc/self/cgroup. For
1171 * example, assume we unshare(CLONE_NEWCGROUP) first, and then create
1172 * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then
1173 * /proc/self/cgroup would show us:
1177 * whereas it should actually show
1181 if (handler
->ns_unshare_flags
& CLONE_NEWCGROUP
) {
1182 ret
= unshare(CLONE_NEWCGROUP
);
1184 if (errno
!= EINVAL
) {
1185 SYSERROR("Failed to unshare CLONE_NEWCGROUP");
1186 goto out_warn_father
;
1189 handler
->ns_clone_flags
&= ~CLONE_NEWCGROUP
;
1190 SYSINFO("Kernel does not support CLONE_NEWCGROUP");
1192 INFO("Unshared CLONE_NEWCGROUP");
1196 if (handler
->ns_unshare_flags
& CLONE_NEWTIME
) {
1197 ret
= unshare(CLONE_NEWTIME
);
1199 if (errno
!= EINVAL
) {
1200 SYSERROR("Failed to unshare CLONE_NEWTIME");
1201 goto out_warn_father
;
1204 handler
->ns_clone_flags
&= ~CLONE_NEWTIME
;
1205 SYSINFO("Kernel does not support CLONE_NEWTIME");
1207 __do_close
int timens_fd
= -EBADF
;
1209 INFO("Unshared CLONE_NEWTIME");
1211 if (handler
->conf
->timens
.s_boot
)
1212 ret
= timens_offset_write(CLOCK_BOOTTIME
, handler
->conf
->timens
.s_boot
, 0);
1213 else if (handler
->conf
->timens
.ns_boot
)
1214 ret
= timens_offset_write(CLOCK_BOOTTIME
, 0, handler
->conf
->timens
.ns_boot
);
1216 SYSERROR("Failed to write CLONE_BOOTTIME offset");
1217 goto out_warn_father
;
1219 TRACE("Wrote CLOCK_BOOTTIME offset");
1221 if (handler
->conf
->timens
.s_monotonic
)
1222 ret
= timens_offset_write(CLOCK_MONOTONIC
, handler
->conf
->timens
.s_monotonic
, 0);
1223 else if (handler
->conf
->timens
.ns_monotonic
)
1224 ret
= timens_offset_write(CLOCK_MONOTONIC
, 0, handler
->conf
->timens
.ns_monotonic
);
1226 SYSERROR("Failed to write CLONE_MONOTONIC offset");
1227 goto out_warn_father
;
1229 TRACE("Wrote CLOCK_MONOTONIC offset");
1231 timens_fd
= open("/proc/self/ns/time_for_children", O_RDONLY
| O_CLOEXEC
);
1232 if (timens_fd
< 0) {
1233 SYSERROR("Failed to open \"/proc/self/ns/time_for_children\"");
1234 goto out_warn_father
;
1237 ret
= setns(timens_fd
, CLONE_NEWTIME
);
1239 SYSERROR("Failed to setns(%d(\"/proc/self/ns/time_for_children\"))", timens_fd
);
1240 goto out_warn_father
;
1245 /* Add the requested environment variables to the current environment to
1246 * allow them to be used by the various hooks, such as the start hook
1249 lxc_list_for_each(iterator
, &handler
->conf
->environment
) {
1250 ret
= putenv((char *)iterator
->elem
);
1252 SYSERROR("Failed to set environment variable: %s",
1253 (char *)iterator
->elem
);
1254 goto out_warn_father
;
1258 /* Setup the container, ip, names, utsname, ... */
1259 ret
= lxc_setup(handler
);
1261 ERROR("Failed to setup container \"%s\"", handler
->name
);
1262 goto out_warn_father
;
1265 /* Set the label to change to when we exec(2) the container's init. */
1266 ret
= handler
->lsm_ops
->process_label_set(handler
->lsm_ops
, NULL
, handler
->conf
, true);
1268 goto out_warn_father
;
1270 /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it
1271 * before we aren't allowed anymore.
1273 if (handler
->conf
->no_new_privs
) {
1274 ret
= prctl(PR_SET_NO_NEW_PRIVS
, prctl_arg(1), prctl_arg(0),
1275 prctl_arg(0), prctl_arg(0));
1277 SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges");
1278 goto out_warn_father
;
1280 DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges");
1283 /* Some init's such as busybox will set sane tty settings on stdin,
1284 * stdout, stderr which it thinks is the console. We already set them
1285 * the way we wanted on the real terminal, and we want init to do its
1286 * setup on its console ie. the pty allocated in lxc_terminal_setup() so
1287 * make sure that that pty is stdin,stdout,stderr.
1289 if (handler
->conf
->console
.pty
>= 0) {
1290 if (handler
->daemonize
|| !handler
->conf
->is_execute
)
1291 ret
= set_stdfds(handler
->conf
->console
.pty
);
1293 ret
= lxc_terminal_set_stdfds(handler
->conf
->console
.pty
);
1295 ERROR("Failed to redirect std{in,out,err} to pty file descriptor %d",
1296 handler
->conf
->console
.pty
);
1297 goto out_warn_father
;
1301 /* If we mounted a temporary proc, then unmount it now. */
1302 tmp_proc_unmount(handler
->conf
);
1304 ret
= lxc_seccomp_load(handler
->conf
);
1306 goto out_warn_father
;
1308 ret
= lxc_seccomp_send_notifier_fd(&handler
->conf
->seccomp
, data_sock0
);
1310 SYSERROR("Failed to send seccomp notify fd to parent");
1311 goto out_warn_father
;
1314 ret
= run_lxc_hooks(handler
->name
, "start", handler
->conf
, NULL
);
1316 ERROR("Failed to run lxc.hook.start for container \"%s\"",
1318 goto out_warn_father
;
1321 close_prot_errno_disarm(handler
->sigfd
);
1323 if (handler
->conf
->console
.pty
< 0 && handler
->daemonize
) {
1324 if (devnull_fd
< 0) {
1325 devnull_fd
= open_devnull();
1327 goto out_warn_father
;
1330 ret
= set_stdfds(devnull_fd
);
1332 ERROR("Failed to redirect std{in,out,err} to \"/dev/null\"");
1333 goto out_warn_father
;
1337 close_prot_errno_disarm(devnull_fd
);
1341 if (handler
->conf
->init_cwd
) {
1342 ret
= chdir(handler
->conf
->init_cwd
);
1344 SYSERROR("Could not change directory to \"%s\"",
1345 handler
->conf
->init_cwd
);
1346 goto out_warn_father
;
1350 if (!lxc_sync_barrier_parent(handler
, START_SYNC_CGROUP_LIMITS
))
1351 goto out_warn_father
;
1353 /* Reset the environment variables the user requested in a clear
1357 /* Don't error out though. */
1359 SYSERROR("Failed to clear environment.");
1361 lxc_list_for_each(iterator
, &handler
->conf
->environment
) {
1362 ret
= putenv((char *)iterator
->elem
);
1364 SYSERROR("Failed to set environment variable: %s",
1365 (char *)iterator
->elem
);
1366 goto out_warn_father
;
1370 ret
= putenv("container=lxc");
1372 SYSERROR("Failed to set environment variable: container=lxc");
1373 goto out_warn_father
;
1376 if (handler
->conf
->ttys
.tty_names
) {
1377 ret
= putenv(handler
->conf
->ttys
.tty_names
);
1379 SYSERROR("Failed to set environment variable for container ptys");
1380 goto out_warn_father
;
1384 /* The container has been setup. We can now switch to an unprivileged
1387 new_uid
= handler
->conf
->init_uid
;
1388 new_gid
= handler
->conf
->init_gid
;
1390 /* Avoid unnecessary syscalls. */
1391 if (new_uid
== nsuid
)
1392 new_uid
= LXC_INVALID_UID
;
1394 if (new_gid
== nsgid
)
1395 new_gid
= LXC_INVALID_GID
;
1397 /* Make sure that the processes STDIO is correctly owned by the user that we are switching to */
1398 ret
= fix_stdio_permissions(new_uid
);
1400 WARN("Failed to ajust stdio permissions");
1402 /* If we are in a new user namespace we already dropped all groups when
1403 * we switched to root in the new user namespace further above. Only
1404 * drop groups if we can, so ensure that we have necessary privilege.
1406 if (lxc_list_empty(&handler
->conf
->id_map
)) {
1408 if (lxc_proc_cap_is_set(CAP_SETGID
, CAP_EFFECTIVE
))
1411 if (handler
->conf
->init_groups
.size
> 0) {
1412 if (!lxc_setgroups(handler
->conf
->init_groups
.list
,
1413 handler
->conf
->init_groups
.size
))
1414 goto out_warn_father
;
1416 if (!lxc_drop_groups())
1417 goto out_warn_father
;
1422 if (!lxc_switch_uid_gid(new_uid
, new_gid
))
1423 goto out_warn_father
;
1425 ret
= lxc_ambient_caps_down();
1427 ERROR("Failed to clear ambient capabilities");
1428 goto out_warn_father
;
1431 if (handler
->conf
->monitor_signal_pdeath
!= SIGKILL
) {
1432 ret
= lxc_set_death_signal(handler
->conf
->monitor_signal_pdeath
,
1433 handler
->monitor_pid
, status_fd
);
1435 SYSERROR("Failed to set PR_SET_PDEATHSIG to %d",
1436 handler
->conf
->monitor_signal_pdeath
);
1437 goto out_warn_father
;
1442 * After this call, we are in error because this ops should not return
1445 handler
->ops
->start(handler
, handler
->data
);
1449 * We want the parent to know something went wrong, so we return a
1450 * special error code.
1452 lxc_sync_wake_parent(handler
, SYNC_ERROR
);
1458 static int lxc_recv_ttys_from_child(struct lxc_handler
*handler
)
1461 struct lxc_terminal_info
*tty
;
1463 int sock
= handler
->data_sock
[1];
1464 struct lxc_conf
*conf
= handler
->conf
;
1465 struct lxc_tty_info
*ttys
= &conf
->ttys
;
1467 if (!conf
->ttys
.max
)
1470 ttys
->tty
= malloc(sizeof(*ttys
->tty
) * ttys
->max
);
1474 for (i
= 0; i
< conf
->ttys
.max
; i
++) {
1477 ret
= lxc_abstract_unix_recv_fds(sock
, ttyfds
, 2, NULL
, 0);
1481 tty
= &ttys
->tty
[i
];
1483 tty
->ptx
= ttyfds
[0];
1484 tty
->pty
= ttyfds
[1];
1485 TRACE("Received pty with ptx fd %d and pty fd %d from child", tty
->ptx
, tty
->pty
);
1489 SYSERROR("Failed to receive %zu ttys from child", ttys
->max
);
1491 TRACE("Received %zu ttys from child", ttys
->max
);
1496 int resolve_clone_flags(struct lxc_handler
*handler
)
1499 struct lxc_conf
*conf
= handler
->conf
;
1500 bool wants_timens
= conf
->timens
.s_boot
|| conf
->timens
.ns_boot
||
1501 conf
->timens
.s_monotonic
|| conf
->timens
.ns_monotonic
;
1503 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1504 if (conf
->ns_keep
) {
1505 if (!(conf
->ns_keep
& ns_info
[i
].clone_flag
))
1506 handler
->ns_clone_flags
|= ns_info
[i
].clone_flag
;
1507 } else if (conf
->ns_clone
) {
1508 if ((conf
->ns_clone
& ns_info
[i
].clone_flag
))
1509 handler
->ns_clone_flags
|= ns_info
[i
].clone_flag
;
1511 if (i
== LXC_NS_USER
&& lxc_list_empty(&handler
->conf
->id_map
))
1514 if (i
== LXC_NS_NET
&& lxc_requests_empty_network(handler
))
1517 if (i
== LXC_NS_CGROUP
&& !cgns_supported())
1520 if (i
== LXC_NS_TIME
&& !wants_timens
)
1523 handler
->ns_clone_flags
|= ns_info
[i
].clone_flag
;
1526 if (!conf
->ns_share
[i
])
1529 handler
->ns_clone_flags
&= ~ns_info
[i
].clone_flag
;
1530 TRACE("Sharing %s namespace", ns_info
[i
].proc_name
);
1533 if (wants_timens
&& (conf
->ns_keep
& ns_info
[LXC_NS_TIME
].clone_flag
))
1534 return log_trace_errno(-1, EINVAL
, "Requested to keep time namespace while also specifying offsets");
1536 /* Deal with namespaces that are unshared. */
1537 if (handler
->ns_clone_flags
& CLONE_NEWTIME
)
1538 handler
->ns_unshare_flags
|= CLONE_NEWTIME
;
1540 if (!pure_unified_layout(handler
->cgroup_ops
) && handler
->ns_clone_flags
& CLONE_NEWCGROUP
)
1541 handler
->ns_unshare_flags
|= CLONE_NEWCGROUP
;
1543 if ((handler
->ns_clone_flags
& (CLONE_NEWNET
| CLONE_NEWUSER
)) ==
1544 (CLONE_NEWNET
| CLONE_NEWUSER
))
1545 handler
->ns_unshare_flags
|= CLONE_NEWNET
;
1547 /* Deal with namespaces that are spawned. */
1548 handler
->ns_on_clone_flags
= handler
->ns_clone_flags
& ~handler
->ns_unshare_flags
;
1550 handler
->clone_flags
= handler
->ns_on_clone_flags
| CLONE_PIDFD
;
1555 /* Note that this function is used with clone(CLONE_VM). Some glibc versions
1556 * used to reset the pid/tid to -1 when CLONE_VM was used without CLONE_THREAD.
1557 * But since the memory between parent and child is shared on CLONE_VM this
1558 * would invalidate the getpid() cache that glibc used to maintain and so
1559 * getpid() in the child would return the parent's pid. This is all fixed in
1560 * newer glibc versions where the getpid() cache is removed and the pid/tid is
1561 * not reset anymore.
1562 * However, if for whatever reason you - dear committer - somehow need to get the
1563 * pid of the dummy intermediate process for do_share_ns() you need to call
1564 * lxc_raw_getpid(). The next lxc_raw_clone() call does not employ CLONE_VM and
1567 static inline int do_share_ns(void *arg
)
1570 struct lxc_handler
*handler
= arg
;
1572 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1573 if (handler
->nsfd
[i
] < 0)
1576 ret
= setns(handler
->nsfd
[i
], 0);
1579 * Note that joining a user and/or mount namespace
1580 * requires the process is not multithreaded otherwise
1581 * setns() will fail here.
1583 SYSERROR("Failed to inherit %s namespace",
1584 ns_info
[i
].proc_name
);
1588 DEBUG("Inherited %s namespace", ns_info
[i
].proc_name
);
1591 flags
= handler
->ns_on_clone_flags
;
1592 flags
|= CLONE_PARENT
;
1593 handler
->pid
= lxc_raw_clone_cb(do_start
, handler
, CLONE_PIDFD
| flags
,
1595 if (handler
->pid
< 0)
1601 /* lxc_spawn() performs crucial setup tasks and clone()s the new process which
1602 * exec()s the requested container binary.
1603 * Note that lxc_spawn() runs in the parent namespaces. Any operations performed
1604 * right here should be double checked if they'd pose a security risk. (For
1605 * example, any {u}mount() operations performed here will be reflected on the
1608 static int lxc_spawn(struct lxc_handler
*handler
)
1610 __do_close
int data_sock0
= -EBADF
, data_sock1
= -EBADF
;
1613 bool wants_to_map_ids
;
1614 struct lxc_list
*id_map
;
1615 const char *name
= handler
->name
;
1616 const char *lxcpath
= handler
->lxcpath
;
1617 bool share_ns
= false;
1618 struct lxc_conf
*conf
= handler
->conf
;
1619 struct cgroup_ops
*cgroup_ops
= handler
->cgroup_ops
;
1621 id_map
= &conf
->id_map
;
1622 wants_to_map_ids
= !lxc_list_empty(id_map
);
1624 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1625 if (!conf
->ns_share
[i
])
1628 handler
->nsfd
[i
] = lxc_inherit_namespace(conf
->ns_share
[i
], lxcpath
, ns_info
[i
].proc_name
);
1629 if (handler
->nsfd
[i
] < 0)
1635 if (!lxc_sync_init(handler
))
1638 ret
= socketpair(AF_UNIX
, SOCK_STREAM
| SOCK_CLOEXEC
, 0,
1639 handler
->data_sock
);
1642 data_sock0
= handler
->data_sock
[0];
1643 data_sock1
= handler
->data_sock
[1];
1645 ret
= resolve_clone_flags(handler
);
1649 if (handler
->ns_clone_flags
& CLONE_NEWNET
) {
1650 ret
= lxc_find_gateway_addresses(handler
);
1652 ERROR("Failed to find gateway addresses");
1657 if (!cgroup_ops
->payload_create(cgroup_ops
, handler
)) {
1658 ERROR("Failed creating cgroups");
1659 goto out_delete_net
;
1662 /* If the rootfs is not a blockdev, prevent the container from marking
1664 * If the container is unprivileged then skip rootfs pinning.
1666 ret
= lxc_rootfs_prepare(&conf
->rootfs
, wants_to_map_ids
);
1668 ERROR("Failed to handle rootfs pinning for container \"%s\"", handler
->name
);
1669 goto out_delete_net
;
1672 /* Create a process in a new set of namespaces. */
1676 attacher_pid
= lxc_clone(do_share_ns
, handler
,
1677 CLONE_VFORK
| CLONE_VM
| CLONE_FILES
, NULL
);
1678 if (attacher_pid
< 0) {
1679 SYSERROR(LXC_CLONE_ERROR
);
1680 goto out_delete_net
;
1683 ret
= wait_for_pid(attacher_pid
);
1685 SYSERROR("Intermediate process failed");
1686 goto out_delete_net
;
1689 if (handler
->pid
< 0) {
1690 SYSERROR(LXC_CLONE_ERROR
);
1691 goto out_delete_net
;
1694 int cgroup_fd
= -EBADF
;
1696 struct lxc_clone_args clone_args
= {
1697 .flags
= handler
->clone_flags
,
1698 .pidfd
= ptr_to_u64(&handler
->pidfd
),
1699 .exit_signal
= SIGCHLD
,
1702 if (handler
->ns_clone_flags
& CLONE_NEWCGROUP
) {
1703 cgroup_fd
= cgroup_unified_fd(cgroup_ops
);
1704 if (cgroup_fd
>= 0) {
1705 handler
->clone_flags
|= CLONE_INTO_CGROUP
;
1706 clone_args
.flags
|= CLONE_INTO_CGROUP
;
1707 clone_args
.cgroup
= cgroup_fd
;
1711 /* Try to spawn directly into target cgroup. */
1712 handler
->pid
= lxc_clone3(&clone_args
, CLONE_ARGS_SIZE_VER2
);
1713 if (handler
->pid
< 0) {
1714 SYSTRACE("Failed to spawn container directly into target cgroup");
1716 /* Kernel might simply be too old for CLONE_INTO_CGROUP. */
1717 handler
->clone_flags
&= ~(CLONE_INTO_CGROUP
| CLONE_NEWCGROUP
);
1718 handler
->ns_on_clone_flags
&= ~CLONE_NEWCGROUP
;
1719 handler
->ns_unshare_flags
|= CLONE_NEWCGROUP
;
1721 clone_args
.flags
= handler
->clone_flags
;
1723 handler
->pid
= lxc_clone3(&clone_args
, CLONE_ARGS_SIZE_VER0
);
1724 } else if (cgroup_fd
>= 0) {
1725 TRACE("Spawned container directly into target cgroup via cgroup2 fd %d", cgroup_fd
);
1728 /* Kernel might be too old for clone3(). */
1729 if (handler
->pid
< 0) {
1730 SYSTRACE("Failed to spawn container via clone3()");
1731 handler
->pid
= lxc_raw_legacy_clone(handler
->clone_flags
, &handler
->pidfd
);
1734 if (handler
->pid
< 0) {
1735 SYSERROR(LXC_CLONE_ERROR
);
1736 goto out_delete_net
;
1739 if (handler
->pid
== 0) {
1740 (void)do_start(handler
);
1741 _exit(EXIT_FAILURE
);
1744 if (handler
->pidfd
< 0)
1745 handler
->clone_flags
&= ~CLONE_PIDFD
;
1746 TRACE("Cloned child process %d", handler
->pid
);
1748 /* Verify that we can actually make use of pidfds. */
1749 if (!lxc_can_use_pidfd(handler
->pidfd
))
1750 close_prot_errno_disarm(handler
->pidfd
);
1752 ret
= snprintf(pidstr
, 20, "%d", handler
->pid
);
1753 if (ret
< 0 || ret
>= 20)
1754 goto out_delete_net
;
1756 ret
= setenv("LXC_PID", pidstr
, 1);
1758 SYSERROR("Failed to set environment variable: LXC_PID=%s", pidstr
);
1760 for (i
= 0; i
< LXC_NS_MAX
; i
++)
1761 if (handler
->ns_on_clone_flags
& ns_info
[i
].clone_flag
)
1762 INFO("Cloned %s", ns_info
[i
].flag_name
);
1764 if (!lxc_try_preserve_namespaces(handler
, handler
->ns_on_clone_flags
, handler
->pid
)) {
1765 ERROR("Failed to preserve cloned namespaces for lxc.hook.stop");
1766 goto out_delete_net
;
1769 lxc_sync_fini_child(handler
);
1771 if (lxc_abstract_unix_send_fds(handler
->data_sock
[0], &handler
->monitor_status_fd
, 1, NULL
, 0) < 0) {
1772 ERROR("Failed to send status file descriptor to child process");
1773 goto out_delete_net
;
1775 close_prot_errno_disarm(handler
->monitor_status_fd
);
1777 /* Map the container uids. The container became an invalid userid the
1778 * moment it was cloned with CLONE_NEWUSER. This call doesn't change
1779 * anything immediately, but allows the container to setuid(0) (0 being
1780 * mapped to something else on the host.) later to become a valid uid
1783 if (wants_to_map_ids
) {
1784 if (!handler
->conf
->ns_share
[LXC_NS_USER
] &&
1785 (handler
->conf
->ns_keep
& CLONE_NEWUSER
) == 0) {
1786 ret
= lxc_map_ids(id_map
, handler
->pid
);
1788 ERROR("Failed to set up id mapping.");
1789 goto out_delete_net
;
1794 if (!lxc_sync_wake_child(handler
, START_SYNC_STARTUP
))
1795 goto out_delete_net
;
1797 if (!lxc_sync_wait_child(handler
, START_SYNC_CONFIGURE
))
1798 goto out_delete_net
;
1800 if (!cgroup_ops
->setup_limits_legacy(cgroup_ops
, handler
->conf
, false)) {
1801 ERROR("Failed to setup cgroup limits for container \"%s\"", name
);
1802 goto out_delete_net
;
1805 if (!cgroup_ops
->payload_enter(cgroup_ops
, handler
)) {
1806 ERROR("Failed to enter cgroups");
1807 goto out_delete_net
;
1810 if (!cgroup_ops
->payload_delegate_controllers(cgroup_ops
)) {
1811 ERROR("Failed to delegate controllers to payload cgroup");
1812 goto out_delete_net
;
1815 if (!cgroup_ops
->setup_limits(cgroup_ops
, handler
)) {
1816 ERROR("Failed to setup cgroup limits for container \"%s\"", name
);
1817 goto out_delete_net
;
1820 if (!cgroup_ops
->chown(cgroup_ops
, handler
->conf
))
1821 goto out_delete_net
;
1823 /* If not done yet, we're now ready to preserve the network namespace */
1824 if (handler
->nsfd
[LXC_NS_NET
] < 0) {
1825 ret
= lxc_try_preserve_ns(handler
->pid
, "net");
1827 if (ret
!= -EOPNOTSUPP
) {
1828 SYSERROR("Failed to preserve net namespace");
1829 goto out_delete_net
;
1832 handler
->nsfd
[LXC_NS_NET
] = ret
;
1833 DEBUG("Preserved net namespace via fd %d", ret
);
1836 ret
= lxc_netns_set_nsid(handler
->nsfd
[LXC_NS_NET
]);
1838 SYSWARN("Failed to allocate new network namespace id");
1840 TRACE("Allocated new network namespace id");
1842 /* Create the network configuration. */
1843 if (handler
->ns_clone_flags
& CLONE_NEWNET
) {
1844 ret
= lxc_create_network(handler
);
1846 ERROR("Failed to create the network");
1847 goto out_delete_net
;
1850 ret
= lxc_network_send_to_child(handler
);
1852 ERROR("Failed to send veth names to child");
1853 goto out_delete_net
;
1857 if (!lxc_list_empty(&conf
->procs
)) {
1858 ret
= setup_proc_filesystem(&conf
->procs
, handler
->pid
);
1860 goto out_delete_net
;
1863 /* Tell the child to continue its initialization. We'll get
1864 * START_SYNC_CGROUP when it is ready for us to setup cgroups.
1866 if (!lxc_sync_barrier_child(handler
, START_SYNC_POST_CONFIGURE
))
1867 goto out_delete_net
;
1869 if (!lxc_list_empty(&conf
->limits
)) {
1870 ret
= setup_resource_limits(&conf
->limits
, handler
->pid
);
1872 ERROR("Failed to setup resource limits");
1873 goto out_delete_net
;
1877 if (!lxc_sync_barrier_child(handler
, START_SYNC_CGROUP_UNSHARE
))
1878 goto out_delete_net
;
1881 * with isolation the limiting devices cgroup was already setup, so
1882 * only setup devices here if we have no namespace directory
1884 if (!handler
->conf
->cgroup_meta
.namespace_dir
&&
1885 !cgroup_ops
->setup_limits_legacy(cgroup_ops
, handler
->conf
, true)) {
1886 ERROR("Failed to setup legacy device cgroup controller limits");
1887 goto out_delete_net
;
1889 TRACE("Set up legacy device cgroup controller limits");
1891 if (!cgroup_ops
->devices_activate(cgroup_ops
, handler
)) {
1892 ERROR("Failed to setup cgroup2 device controller limits");
1893 goto out_delete_net
;
1895 TRACE("Set up cgroup2 device controller limits");
1897 if (handler
->ns_unshare_flags
& CLONE_NEWCGROUP
) {
1898 /* Now we're ready to preserve the cgroup namespace */
1899 ret
= lxc_try_preserve_ns(handler
->pid
, "cgroup");
1901 if (ret
!= -EOPNOTSUPP
) {
1902 SYSERROR("Failed to preserve cgroup namespace");
1903 goto out_delete_net
;
1906 handler
->nsfd
[LXC_NS_CGROUP
] = ret
;
1907 DEBUG("Preserved cgroup namespace via fd %d", ret
);
1911 cgroup_ops
->payload_finalize(cgroup_ops
);
1912 TRACE("Finished setting up cgroups");
1914 if (handler
->ns_unshare_flags
& CLONE_NEWTIME
) {
1915 /* Now we're ready to preserve the cgroup namespace */
1916 ret
= lxc_try_preserve_ns(handler
->pid
, "time");
1918 if (ret
!= -EOPNOTSUPP
) {
1919 SYSERROR("Failed to preserve time namespace");
1920 goto out_delete_net
;
1923 handler
->nsfd
[LXC_NS_TIME
] = ret
;
1924 DEBUG("Preserved time namespace via fd %d", ret
);
1928 /* Run any host-side start hooks */
1929 ret
= run_lxc_hooks(name
, "start-host", conf
, NULL
);
1931 ERROR("Failed to run lxc.hook.start-host");
1932 goto out_delete_net
;
1935 /* Tell the child to complete its initialization and wait for it to exec
1936 * or return an error. (The child will never return
1937 * START_SYNC_READY_START+1. It will either close the sync pipe, causing
1938 * lxc_sync_barrier_child to return success, or return a different
1939 * value, causing us to error out).
1941 if (!lxc_sync_barrier_child(handler
, START_SYNC_READY_START
))
1942 goto out_delete_net
;
1944 if (handler
->ns_clone_flags
& CLONE_NEWNET
) {
1945 ret
= lxc_network_recv_name_and_ifindex_from_child(handler
);
1947 ERROR("Failed to receive names and ifindices for network devices from child");
1948 goto out_delete_net
;
1952 ret
= lxc_setup_devpts_parent(handler
);
1954 SYSERROR("Failed to receive devpts fd from child");
1955 goto out_delete_net
;
1958 /* Now all networks are created, network devices are moved into place,
1959 * and the correct names and ifindices in the respective namespaces have
1960 * been recorded. The corresponding structs have now all been filled. So
1961 * log them for debugging purposes.
1963 lxc_log_configured_netdevs(conf
);
1965 /* Read tty fds allocated by child. */
1966 ret
= lxc_recv_ttys_from_child(handler
);
1968 ERROR("Failed to receive tty info from child process");
1969 goto out_delete_net
;
1972 ret
= lxc_seccomp_recv_notifier_fd(&handler
->conf
->seccomp
, data_sock1
);
1974 SYSERROR("Failed to receive seccomp notify fd from child");
1975 goto out_delete_net
;
1978 ret
= handler
->ops
->post_start(handler
, handler
->data
);
1982 ret
= lxc_set_state(name
, handler
, RUNNING
);
1984 ERROR("Failed to set state to \"%s\"", lxc_state2str(RUNNING
));
1988 lxc_sync_fini(handler
);
1993 if (handler
->ns_clone_flags
& CLONE_NEWNET
)
1994 lxc_delete_network(handler
);
2000 lxc_sync_fini(handler
);
2005 int __lxc_start(struct lxc_handler
*handler
, struct lxc_operations
*ops
,
2006 void *data
, const char *lxcpath
, bool daemonize
, int *error_num
)
2009 const char *name
= handler
->name
;
2010 struct lxc_conf
*conf
= handler
->conf
;
2011 struct cgroup_ops
*cgroup_ops
;
2013 ret
= lxc_init(name
, handler
);
2015 ERROR("Failed to initialize container \"%s\"", name
);
2019 handler
->data
= data
;
2020 handler
->daemonize
= daemonize
;
2021 cgroup_ops
= handler
->cgroup_ops
;
2023 if (!attach_block_device(handler
->conf
)) {
2024 ERROR("Failed to attach block device");
2029 if (!cgroup_ops
->monitor_create(cgroup_ops
, handler
)) {
2030 ERROR("Failed to create monitor cgroup");
2035 if (!cgroup_ops
->monitor_enter(cgroup_ops
, handler
)) {
2036 ERROR("Failed to enter monitor cgroup");
2041 if (!cgroup_ops
->monitor_delegate_controllers(cgroup_ops
)) {
2042 ERROR("Failed to delegate controllers to monitor cgroup");
2047 if (geteuid() == 0 && !lxc_list_empty(&conf
->id_map
)) {
2048 /* If the backing store is a device, mount it here and now. */
2049 if (rootfs_is_blockdev(conf
)) {
2050 ret
= unshare(CLONE_NEWNS
);
2052 ERROR("Failed to unshare CLONE_NEWNS");
2055 INFO("Unshared CLONE_NEWNS");
2057 turn_into_dependent_mounts();
2058 ret
= lxc_setup_rootfs_prepare_root(conf
, name
, lxcpath
);
2060 ERROR("Error setting up rootfs mount as root before spawn");
2063 INFO("Set up container rootfs as host root");
2067 ret
= lxc_spawn(handler
);
2069 ERROR("Failed to spawn container \"%s\"", name
);
2070 goto out_detach_blockdev
;
2073 handler
->conf
->reboot
= REBOOT_NONE
;
2075 ret
= lxc_poll(name
, handler
);
2077 ERROR("LXC mainloop exited with error: %d", ret
);
2078 goto out_delete_network
;
2081 if (!handler
->init_died
&& handler
->pid
> 0) {
2082 ERROR("Child process is not killed");
2084 goto out_delete_network
;
2087 status
= lxc_wait_for_pid_status(handler
->pid
);
2089 SYSERROR("Failed to retrieve status for %d", handler
->pid
);
2091 /* If the child process exited but was not signaled, it didn't call
2092 * reboot. This should mean it was an lxc-execute which simply exited.
2093 * In any case, treat it as a 'halt'.
2095 if (WIFSIGNALED(status
)) {
2096 switch(WTERMSIG(status
)) {
2097 case SIGINT
: /* halt */
2098 DEBUG("Container \"%s\" is halting", name
);
2100 case SIGHUP
: /* reboot */
2101 DEBUG("Container \"%s\" is rebooting", name
);
2102 handler
->conf
->reboot
= REBOOT_REQ
;
2104 case SIGSYS
: /* seccomp */
2105 DEBUG("Container \"%s\" violated its seccomp policy", name
);
2108 DEBUG("Unknown exit status for container \"%s\" init %d", name
, WTERMSIG(status
));
2113 ret
= lxc_restore_phys_nics_to_netns(handler
);
2115 ERROR("Failed to move physical network devices back to parent network namespace");
2117 lxc_monitor_send_exit_code(name
, status
, handler
->lxcpath
);
2118 lxc_error_set_and_log(handler
->pid
, status
);
2120 *error_num
= handler
->exit_status
;
2122 lxc_delete_network(handler
);
2123 detach_block_device(handler
->conf
);
2132 out_detach_blockdev
:
2134 detach_block_device(handler
->conf
);
2140 lxc_restore_phys_nics_to_netns(handler
);
2141 lxc_delete_network(handler
);
2142 detach_block_device(handler
->conf
);
2151 static int start(struct lxc_handler
*handler
, void* data
)
2153 struct start_args
*arg
= data
;
2155 NOTICE("Exec'ing \"%s\"", arg
->argv
[0]);
2157 execvp(arg
->argv
[0], arg
->argv
);
2158 SYSERROR("Failed to exec \"%s\"", arg
->argv
[0]);
2162 static int post_start(struct lxc_handler
*handler
, void* data
)
2164 struct start_args
*arg
= data
;
2166 NOTICE("Started \"%s\" with pid \"%d\"", arg
->argv
[0], handler
->pid
);
2170 static struct lxc_operations start_ops
= {
2172 .post_start
= post_start
2175 int lxc_start(char *const argv
[], struct lxc_handler
*handler
,
2176 const char *lxcpath
, bool daemonize
, int *error_num
)
2178 struct start_args start_arg
= {
2182 TRACE("Doing lxc_start");
2183 return __lxc_start(handler
, &start_ops
, &start_arg
, lxcpath
, daemonize
, error_num
);
2186 static void lxc_destroy_container_on_signal(struct lxc_handler
*handler
,
2189 char destroy
[PATH_MAX
];
2190 struct lxc_container
*c
;
2194 if (handler
->conf
->rootfs
.path
&& handler
->conf
->rootfs
.mount
) {
2195 bret
= do_destroy_container(handler
);
2197 ERROR("Error destroying rootfs for container \"%s\"", name
);
2201 INFO("Destroyed rootfs for container \"%s\"", name
);
2203 ret
= snprintf(destroy
, PATH_MAX
, "%s/%s", handler
->lxcpath
, name
);
2204 if (ret
< 0 || ret
>= PATH_MAX
) {
2205 ERROR("Error destroying directory for container \"%s\"", name
);
2209 c
= lxc_container_new(name
, handler
->lxcpath
);
2211 if (container_disk_lock(c
)) {
2212 INFO("Could not update lxc_snapshots file");
2213 lxc_container_put(c
);
2215 mod_all_rdeps(c
, false);
2216 container_disk_unlock(c
);
2217 lxc_container_put(c
);
2221 if (!handler
->am_root
)
2222 ret
= userns_exec_full(handler
->conf
, lxc_rmdir_onedev_wrapper
,
2223 destroy
, "lxc_rmdir_onedev_wrapper");
2225 ret
= lxc_rmdir_onedev(destroy
, NULL
);
2228 ERROR("Error destroying directory for container \"%s\"", name
);
2231 INFO("Destroyed directory for container \"%s\"", name
);
2234 static int lxc_rmdir_onedev_wrapper(void *data
)
2236 char *arg
= (char *) data
;
2237 return lxc_rmdir_onedev(arg
, NULL
);
2240 static bool do_destroy_container(struct lxc_handler
*handler
)
2244 if (!handler
->am_root
) {
2245 ret
= userns_exec_full(handler
->conf
, storage_destroy_wrapper
,
2246 handler
->conf
, "storage_destroy_wrapper");
2253 return storage_destroy(handler
->conf
);