]>
git.proxmox.com Git - mirror_lxc.git/blob - src/lxc/start.c
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(LXC_MAINLOOP_ERROR
, "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
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 return log_error_errno(-1, errno
, "Failed to open monitor status fd");
738 TRACE("Initialized LSM");
740 /* Begin by setting the state to STARTING. */
741 ret
= lxc_set_state(name
, handler
, STARTING
);
743 return log_error(-1, "Failed to set state to \"%s\"", lxc_state2str(STARTING
));
744 TRACE("Set container state to \"STARTING\"");
746 /* Start of environment variable setup for hooks. */
747 ret
= setenv("LXC_NAME", name
, 1);
749 SYSERROR("Failed to set environment variable: LXC_NAME=%s", name
);
752 ret
= setenv("LXC_CONFIG_FILE", conf
->rcfile
, 1);
754 SYSERROR("Failed to set environment variable: LXC_CONFIG_FILE=%s", conf
->rcfile
);
757 if (conf
->rootfs
.mount
) {
758 ret
= setenv("LXC_ROOTFS_MOUNT", conf
->rootfs
.mount
, 1);
760 SYSERROR("Failed to set environment variable: LXC_ROOTFS_MOUNT=%s", conf
->rootfs
.mount
);
763 if (conf
->rootfs
.path
) {
764 ret
= setenv("LXC_ROOTFS_PATH", conf
->rootfs
.path
, 1);
766 SYSERROR("Failed to set environment variable: LXC_ROOTFS_PATH=%s", conf
->rootfs
.path
);
769 if (conf
->console
.path
) {
770 ret
= setenv("LXC_CONSOLE", conf
->console
.path
, 1);
772 SYSERROR("Failed to set environment variable: LXC_CONSOLE=%s", conf
->console
.path
);
775 if (conf
->console
.log_path
) {
776 ret
= setenv("LXC_CONSOLE_LOGPATH", conf
->console
.log_path
, 1);
778 SYSERROR("Failed to set environment variable: LXC_CONSOLE_LOGPATH=%s", conf
->console
.log_path
);
781 if (cgns_supported()) {
782 ret
= setenv("LXC_CGNS_AWARE", "1", 1);
784 SYSERROR("Failed to set environment variable LXC_CGNS_AWARE=1");
787 loglevel
= lxc_log_priority_to_string(lxc_log_get_level());
788 ret
= setenv("LXC_LOG_LEVEL", loglevel
, 1);
790 SYSERROR("Set environment variable LXC_LOG_LEVEL=%s", loglevel
);
792 if (conf
->hooks_version
== 0)
793 ret
= setenv("LXC_HOOK_VERSION", "0", 1);
795 ret
= setenv("LXC_HOOK_VERSION", "1", 1);
797 SYSERROR("Failed to set environment variable LXC_HOOK_VERSION=%u", conf
->hooks_version
);
798 /* End of environment variable setup for hooks. */
800 TRACE("Set environment variables");
802 ret
= run_lxc_hooks(name
, "pre-start", conf
, NULL
);
804 return log_error(-1, "Failed to run lxc.hook.pre-start for container \"%s\"", name
);
805 TRACE("Ran pre-start hooks");
807 /* The signal fd has to be created before forking otherwise if the child
808 * process exits before we setup the signal fd, the event will be lost
809 * and the command will be stuck.
811 handler
->sigfd
= setup_signal_fd(&handler
->oldmask
);
812 if (handler
->sigfd
< 0)
813 return log_error(-1, "Failed to setup SIGCHLD fd handler.");
814 TRACE("Set up signal fd");
816 /* Do this after setting up signals since it might unblock SIGWINCH. */
817 ret
= lxc_terminal_setup(conf
);
819 ERROR("Failed to create console");
820 goto out_restore_sigmask
;
822 TRACE("Created console");
824 ret
= lxc_terminal_map_ids(conf
, &conf
->console
);
826 ERROR("Failed to chown console");
827 goto out_delete_terminal
;
829 TRACE("Chowned console");
831 handler
->cgroup_ops
= cgroup_init(handler
->conf
);
832 if (!handler
->cgroup_ops
) {
833 ERROR("Failed to initialize cgroup driver");
834 goto out_delete_terminal
;
836 TRACE("Initialized cgroup driver");
838 ret
= lxc_read_seccomp_config(conf
);
840 return log_error(-1, "Failed loading seccomp policy");
841 TRACE("Read seccomp policy");
843 ret
= lsm_process_prepare(conf
, handler
->lxcpath
);
845 ERROR("Failed to initialize LSM");
846 goto out_delete_terminal
;
848 TRACE("Initialized LSM");
850 INFO("Container \"%s\" is initialized", name
);
851 handler
->monitor_status_fd
= move_fd(status_fd
);
855 lxc_terminal_delete(&handler
->conf
->console
);
858 (void)pthread_sigmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
863 void lxc_end(struct lxc_handler
*handler
)
867 struct lxc_list
*cur
, *next
;
868 char *namespaces
[LXC_NS_MAX
+ 1];
869 size_t namespace_count
= 0;
870 const char *name
= handler
->name
;
871 struct cgroup_ops
*cgroup_ops
= handler
->cgroup_ops
;
873 /* The STOPPING state is there for future cleanup code which can take
876 lxc_set_state(name
, handler
, STOPPING
);
878 self
= lxc_raw_getpid();
879 for (int i
= 0; i
< LXC_NS_MAX
; i
++) {
880 if (handler
->nsfd
[i
] < 0)
883 if (handler
->conf
->hooks_version
== 0)
884 ret
= asprintf(&namespaces
[namespace_count
],
885 "%s:/proc/%d/fd/%d", ns_info
[i
].proc_name
,
886 self
, handler
->nsfd
[i
]);
888 ret
= asprintf(&namespaces
[namespace_count
],
889 "/proc/%d/fd/%d", self
, handler
->nsfd
[i
]);
891 SYSERROR("Failed to allocate memory");
895 if (handler
->conf
->hooks_version
== 0) {
900 ret
= setenv(ns_info
[i
].env_name
, namespaces
[namespace_count
], 1);
902 SYSERROR("Failed to set environment variable %s=%s",
903 ns_info
[i
].env_name
, namespaces
[namespace_count
]);
905 TRACE("Set environment variable %s=%s",
906 ns_info
[i
].env_name
, namespaces
[namespace_count
]);
910 namespaces
[namespace_count
] = NULL
;
912 if (handler
->conf
->reboot
> REBOOT_NONE
) {
913 ret
= setenv("LXC_TARGET", "reboot", 1);
915 SYSERROR("Failed to set environment variable: LXC_TARGET=reboot");
918 if (handler
->conf
->reboot
== REBOOT_NONE
) {
919 ret
= setenv("LXC_TARGET", "stop", 1);
921 SYSERROR("Failed to set environment variable: LXC_TARGET=stop");
924 if (handler
->conf
->hooks_version
== 0)
925 ret
= run_lxc_hooks(name
, "stop", handler
->conf
, namespaces
);
927 ret
= run_lxc_hooks(name
, "stop", handler
->conf
, NULL
);
929 ERROR("Failed to run \"lxc.hook.stop\" hook");
931 while (namespace_count
--)
932 free(namespaces
[namespace_count
]);
934 lsm_process_cleanup(handler
->conf
, handler
->lxcpath
);
937 cgroup_ops
->payload_destroy(cgroup_ops
, handler
);
938 cgroup_ops
->monitor_destroy(cgroup_ops
, handler
);
941 if (handler
->conf
->reboot
== REBOOT_NONE
) {
942 /* For all new state clients simply close the command socket.
943 * This will inform all state clients that the container is
944 * STOPPED and also prevents a race between a open()/close() on
945 * the command socket causing a new process to get ECONNREFUSED
946 * because we haven't yet closed the command socket.
948 close_prot_errno_disarm(handler
->conf
->maincmd_fd
);
949 TRACE("Closed command socket");
951 /* This function will try to connect to the legacy lxc-monitord
952 * state server and only exists for backwards compatibility.
954 lxc_monitor_send_state(name
, STOPPED
, handler
->lxcpath
);
956 /* The command socket is closed so no one can acces the command
957 * socket anymore so there's no need to lock it.
959 handler
->state
= STOPPED
;
960 TRACE("Set container state to \"STOPPED\"");
962 lxc_set_state(name
, handler
, STOPPED
);
963 TRACE("Set container state to \"STOPPED\"");
966 /* Avoid lingering namespace references. */
967 lxc_put_nsfds(handler
);
969 ret
= run_lxc_hooks(name
, "post-stop", handler
->conf
, NULL
);
971 ERROR("Failed to run lxc.hook.post-stop for container \"%s\"", name
);
972 if (handler
->conf
->reboot
> REBOOT_NONE
) {
973 WARN("Container will be stopped instead of rebooted");
974 handler
->conf
->reboot
= REBOOT_NONE
;
976 ret
= setenv("LXC_TARGET", "stop", 1);
978 WARN("Failed to set environment variable: LXC_TARGET=stop");
982 /* Reset mask set by setup_signal_fd. */
983 ret
= pthread_sigmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
985 SYSWARN("Failed to restore signal mask");
987 lxc_terminal_delete(&handler
->conf
->console
);
988 lxc_delete_tty(&handler
->conf
->ttys
);
990 /* The command socket is now closed, no more state clients can register
991 * themselves from now on. So free the list of state clients.
993 lxc_list_for_each_safe(cur
, &handler
->conf
->state_clients
, next
) {
994 struct lxc_state_client
*client
= cur
->elem
;
996 /* Keep state clients that want to be notified about reboots. */
997 if ((handler
->conf
->reboot
> REBOOT_NONE
) &&
998 (client
->states
[RUNNING
] == 2))
1001 /* close state client socket */
1003 close(client
->clientfd
);
1008 if (handler
->conf
->ephemeral
== 1 && handler
->conf
->reboot
!= REBOOT_REQ
)
1009 lxc_destroy_container_on_signal(handler
, name
);
1011 lxc_free_handler(handler
);
1014 void lxc_abort(struct lxc_handler
*handler
)
1019 lxc_set_state(handler
->name
, handler
, ABORTING
);
1021 if (handler
->pidfd
>= 0) {
1022 ret
= lxc_raw_pidfd_send_signal(handler
->pidfd
, SIGKILL
, NULL
, 0);
1024 SYSWARN("Failed to send SIGKILL via pidfd %d for process %d",
1025 handler
->pidfd
, handler
->pid
);
1028 if ((!ret
|| errno
!= ESRCH
) && handler
->pid
> 0)
1029 if (kill(handler
->pid
, SIGKILL
))
1030 SYSWARN("Failed to send SIGKILL to %d", handler
->pid
);
1033 ret
= waitpid(-1, &status
, 0);
1037 static int do_start(void *data
)
1039 struct lxc_handler
*handler
= data
;
1040 __lxc_unused __do_close
int data_sock0
= handler
->data_sock
[0],
1041 data_sock1
= handler
->data_sock
[1];
1042 __do_close
int status_fd
= -EBADF
;
1046 struct lxc_list
*iterator
;
1049 int devnull_fd
= -1;
1051 lxc_sync_fini_parent(handler
);
1053 if (lxc_abstract_unix_recv_fds(data_sock1
, &status_fd
, 1, NULL
, 0) < 0) {
1054 ERROR("Failed to receive status file descriptor to child process");
1055 goto out_warn_father
;
1058 /* This prctl must be before the synchro, so if the parent dies before
1059 * we set the parent death signal, we will detect its death with the
1060 * synchro right after, otherwise we have a window where the parent can
1061 * exit before we set the pdeath signal leading to a unsupervized
1064 ret
= lxc_set_death_signal(SIGKILL
, handler
->monitor_pid
, status_fd
);
1066 SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL");
1067 goto out_warn_father
;
1070 ret
= lxc_ambient_caps_up();
1072 ERROR("Failed to raise ambient capabilities");
1073 goto out_warn_father
;
1076 ret
= pthread_sigmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
1078 SYSERROR("Failed to set signal mask");
1079 goto out_warn_father
;
1082 /* Don't leak the pinfd to the container. */
1083 close_prot_errno_disarm(handler
->pinfd
);
1085 ret
= lxc_sync_wait_parent(handler
, LXC_SYNC_STARTUP
);
1087 goto out_warn_father
;
1089 /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See
1090 * https://github.com/lxc/lxd/issues/1978.
1092 if ((handler
->ns_clone_flags
& (CLONE_NEWNET
| CLONE_NEWUSER
)) ==
1093 (CLONE_NEWNET
| CLONE_NEWUSER
)) {
1094 ret
= unshare(CLONE_NEWNET
);
1096 SYSERROR("Failed to unshare CLONE_NEWNET");
1097 goto out_warn_father
;
1099 INFO("Unshared CLONE_NEWNET");
1102 /* Tell the parent task it can begin to configure the container and wait
1105 ret
= lxc_sync_barrier_parent(handler
, LXC_SYNC_CONFIGURE
);
1109 if (handler
->ns_clone_flags
& CLONE_NEWNET
) {
1110 ret
= lxc_network_recv_from_parent(handler
);
1112 ERROR("Failed to receive veth names from parent");
1113 goto out_warn_father
;
1117 /* If we are in a new user namespace, become root there to have
1118 * privilege over our namespace.
1120 if (!lxc_list_empty(&handler
->conf
->id_map
)) {
1121 if (!handler
->conf
->root_nsuid_map
)
1122 nsuid
= handler
->conf
->init_uid
;
1124 if (!handler
->conf
->root_nsgid_map
)
1125 nsgid
= handler
->conf
->init_gid
;
1127 /* Drop groups only after we switched to a valid gid in the new
1130 if (!lxc_setgroups(0, NULL
) &&
1131 (handler
->am_root
|| errno
!= EPERM
))
1132 goto out_warn_father
;
1134 if (!lxc_switch_uid_gid(nsuid
, nsgid
))
1135 goto out_warn_father
;
1137 ret
= prctl(PR_SET_DUMPABLE
, prctl_arg(1), prctl_arg(0),
1138 prctl_arg(0), prctl_arg(0));
1140 goto out_warn_father
;
1142 /* set{g,u}id() clears deathsignal */
1143 ret
= lxc_set_death_signal(SIGKILL
, handler
->monitor_pid
, status_fd
);
1145 SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL");
1146 goto out_warn_father
;
1150 ret
= access(handler
->lxcpath
, X_OK
);
1152 print_top_failing_dir(handler
->lxcpath
);
1153 goto out_warn_father
;
1156 /* In order to checkpoint restore, we need to have everything in the
1157 * same mount namespace. However, some containers may not have a
1158 * reasonable /dev (in particular, they may not have /dev/null), so we
1159 * can't set init's std fds to /dev/null by opening it from inside the
1162 * If that's the case, fall back to using the host's /dev/null. This
1163 * means that migration won't work, but at least we won't spew output
1164 * where it isn't wanted.
1166 if (handler
->daemonize
&& !handler
->conf
->autodev
) {
1167 char path
[PATH_MAX
];
1169 ret
= snprintf(path
, sizeof(path
), "%s/dev/null",
1170 handler
->conf
->rootfs
.mount
);
1171 if (ret
< 0 || ret
>= sizeof(path
))
1172 goto out_warn_father
;
1174 ret
= access(path
, F_OK
);
1176 devnull_fd
= open_devnull();
1179 goto out_warn_father
;
1180 WARN("Using /dev/null from the host for container init's standard file descriptors. Migration will not work");
1184 /* Ask father to setup cgroups and wait for him to finish. */
1185 ret
= lxc_sync_barrier_parent(handler
, LXC_SYNC_CGROUP
);
1189 /* Unshare cgroup namespace after we have setup our cgroups. If we do it
1190 * earlier we end up with a wrong view of /proc/self/cgroup. For
1191 * example, assume we unshare(CLONE_NEWCGROUP) first, and then create
1192 * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then
1193 * /proc/self/cgroup would show us:
1197 * whereas it should actually show
1201 if (handler
->ns_clone_flags
& CLONE_NEWCGROUP
) {
1202 ret
= unshare(CLONE_NEWCGROUP
);
1204 if (errno
!= EINVAL
) {
1205 SYSERROR("Failed to unshare CLONE_NEWCGROUP");
1206 goto out_warn_father
;
1209 handler
->ns_clone_flags
&= ~CLONE_NEWCGROUP
;
1210 SYSINFO("Kernel does not support CLONE_NEWCGROUP");
1212 INFO("Unshared CLONE_NEWCGROUP");
1216 /* Add the requested environment variables to the current environment to
1217 * allow them to be used by the various hooks, such as the start hook
1220 lxc_list_for_each(iterator
, &handler
->conf
->environment
) {
1221 ret
= putenv((char *)iterator
->elem
);
1223 SYSERROR("Failed to set environment variable: %s",
1224 (char *)iterator
->elem
);
1225 goto out_warn_father
;
1229 /* Setup the container, ip, names, utsname, ... */
1230 ret
= lxc_setup(handler
);
1232 ERROR("Failed to setup container \"%s\"", handler
->name
);
1233 goto out_warn_father
;
1236 /* Set the label to change to when we exec(2) the container's init. */
1237 ret
= lsm_process_label_set(NULL
, handler
->conf
, true);
1239 goto out_warn_father
;
1241 /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it
1242 * before we aren't allowed anymore.
1244 if (handler
->conf
->no_new_privs
) {
1245 ret
= prctl(PR_SET_NO_NEW_PRIVS
, prctl_arg(1), prctl_arg(0),
1246 prctl_arg(0), prctl_arg(0));
1248 SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges");
1249 goto out_warn_father
;
1251 DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges");
1254 /* Some init's such as busybox will set sane tty settings on stdin,
1255 * stdout, stderr which it thinks is the console. We already set them
1256 * the way we wanted on the real terminal, and we want init to do its
1257 * setup on its console ie. the pty allocated in lxc_terminal_setup() so
1258 * make sure that that pty is stdin,stdout,stderr.
1260 if (handler
->conf
->console
.slave
>= 0) {
1261 if (handler
->daemonize
|| !handler
->conf
->is_execute
)
1262 ret
= set_stdfds(handler
->conf
->console
.slave
);
1264 ret
= lxc_terminal_set_stdfds(handler
->conf
->console
.slave
);
1266 ERROR("Failed to redirect std{in,out,err} to pty file descriptor %d",
1267 handler
->conf
->console
.slave
);
1268 goto out_warn_father
;
1272 /* If we mounted a temporary proc, then unmount it now. */
1273 tmp_proc_unmount(handler
->conf
);
1275 ret
= lxc_seccomp_load(handler
->conf
);
1277 goto out_warn_father
;
1279 ret
= lxc_seccomp_send_notifier_fd(&handler
->conf
->seccomp
, data_sock0
);
1281 SYSERROR("Failed to send seccomp notify fd to parent");
1282 goto out_warn_father
;
1285 ret
= run_lxc_hooks(handler
->name
, "start", handler
->conf
, NULL
);
1287 ERROR("Failed to run lxc.hook.start for container \"%s\"",
1289 goto out_warn_father
;
1292 close_prot_errno_disarm(handler
->sigfd
);
1294 if (handler
->conf
->console
.slave
< 0 && handler
->daemonize
) {
1295 if (devnull_fd
< 0) {
1296 devnull_fd
= open_devnull();
1298 goto out_warn_father
;
1301 ret
= set_stdfds(devnull_fd
);
1303 ERROR("Failed to redirect std{in,out,err} to \"/dev/null\"");
1304 goto out_warn_father
;
1308 close_prot_errno_disarm(devnull_fd
);
1312 if (handler
->conf
->init_cwd
) {
1313 ret
= chdir(handler
->conf
->init_cwd
);
1315 SYSERROR("Could not change directory to \"%s\"",
1316 handler
->conf
->init_cwd
);
1317 goto out_warn_father
;
1321 ret
= lxc_sync_barrier_parent(handler
, LXC_SYNC_CGROUP_LIMITS
);
1323 goto out_warn_father
;
1325 /* Reset the environment variables the user requested in a clear
1329 /* Don't error out though. */
1331 SYSERROR("Failed to clear environment.");
1333 lxc_list_for_each(iterator
, &handler
->conf
->environment
) {
1334 ret
= putenv((char *)iterator
->elem
);
1336 SYSERROR("Failed to set environment variable: %s",
1337 (char *)iterator
->elem
);
1338 goto out_warn_father
;
1342 ret
= putenv("container=lxc");
1344 SYSERROR("Failed to set environment variable: container=lxc");
1345 goto out_warn_father
;
1348 if (handler
->conf
->ttys
.tty_names
) {
1349 ret
= putenv(handler
->conf
->ttys
.tty_names
);
1351 SYSERROR("Failed to set environment variable for container ptys");
1352 goto out_warn_father
;
1356 /* The container has been setup. We can now switch to an unprivileged
1359 new_uid
= handler
->conf
->init_uid
;
1360 new_gid
= handler
->conf
->init_gid
;
1362 /* Avoid unnecessary syscalls. */
1363 if (new_uid
== nsuid
)
1364 new_uid
= LXC_INVALID_UID
;
1366 if (new_gid
== nsgid
)
1367 new_gid
= LXC_INVALID_GID
;
1369 /* Make sure that the processes STDIO is correctly owned by the user that we are switching to */
1370 fix_stdio_permissions(new_uid
);
1372 /* If we are in a new user namespace we already dropped all groups when
1373 * we switched to root in the new user namespace further above. Only
1374 * drop groups if we can, so ensure that we have necessary privilege.
1376 if (lxc_list_empty(&handler
->conf
->id_map
))
1378 if (lxc_proc_cap_is_set(CAP_SETGID
, CAP_EFFECTIVE
))
1380 if (!lxc_setgroups(0, NULL
))
1381 goto out_warn_father
;
1383 if (!lxc_switch_uid_gid(new_uid
, new_gid
))
1384 goto out_warn_father
;
1386 ret
= lxc_ambient_caps_down();
1388 ERROR("Failed to clear ambient capabilities");
1389 goto out_warn_father
;
1392 if (handler
->conf
->monitor_signal_pdeath
!= SIGKILL
) {
1393 ret
= lxc_set_death_signal(handler
->conf
->monitor_signal_pdeath
,
1394 handler
->monitor_pid
, status_fd
);
1396 SYSERROR("Failed to set PR_SET_PDEATHSIG to %d",
1397 handler
->conf
->monitor_signal_pdeath
);
1398 goto out_warn_father
;
1402 /* After this call, we are in error because this ops should not return
1405 handler
->ops
->start(handler
, handler
->data
);
1408 /* We want the parent to know something went wrong, so we return a
1409 * special error code.
1411 lxc_sync_wake_parent(handler
, LXC_SYNC_ERROR
);
1414 close_prot_errno_disarm(devnull_fd
);
1419 static int lxc_recv_ttys_from_child(struct lxc_handler
*handler
)
1422 struct lxc_terminal_info
*tty
;
1424 int sock
= handler
->data_sock
[1];
1425 struct lxc_conf
*conf
= handler
->conf
;
1426 struct lxc_tty_info
*ttys
= &conf
->ttys
;
1428 if (!conf
->ttys
.max
)
1431 ttys
->tty
= malloc(sizeof(*ttys
->tty
) * ttys
->max
);
1435 for (i
= 0; i
< conf
->ttys
.max
; i
++) {
1438 ret
= lxc_abstract_unix_recv_fds(sock
, ttyfds
, 2, NULL
, 0);
1442 tty
= &ttys
->tty
[i
];
1444 tty
->master
= ttyfds
[0];
1445 tty
->slave
= ttyfds
[1];
1446 TRACE("Received pty with master fd %d and slave fd %d from child", tty
->master
, tty
->slave
);
1450 SYSERROR("Failed to receive %zu ttys from child", ttys
->max
);
1452 TRACE("Received %zu ttys from child", ttys
->max
);
1457 int resolve_clone_flags(struct lxc_handler
*handler
)
1460 struct lxc_conf
*conf
= handler
->conf
;
1462 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1463 if (conf
->ns_keep
) {
1464 if (!(conf
->ns_keep
& ns_info
[i
].clone_flag
))
1465 handler
->ns_clone_flags
|= ns_info
[i
].clone_flag
;
1466 } else if (conf
->ns_clone
) {
1467 if ((conf
->ns_clone
& ns_info
[i
].clone_flag
))
1468 handler
->ns_clone_flags
|= ns_info
[i
].clone_flag
;
1470 if (i
== LXC_NS_USER
&& lxc_list_empty(&handler
->conf
->id_map
))
1473 if (i
== LXC_NS_NET
&& lxc_requests_empty_network(handler
))
1476 if (i
== LXC_NS_CGROUP
&& !cgns_supported())
1479 handler
->ns_clone_flags
|= ns_info
[i
].clone_flag
;
1482 if (!conf
->ns_share
[i
])
1485 handler
->ns_clone_flags
&= ~ns_info
[i
].clone_flag
;
1486 TRACE("Sharing %s namespace", ns_info
[i
].proc_name
);
1492 /* Note that this function is used with clone(CLONE_VM). Some glibc versions
1493 * used to reset the pid/tid to -1 when CLONE_VM was used without CLONE_THREAD.
1494 * But since the memory between parent and child is shared on CLONE_VM this
1495 * would invalidate the getpid() cache that glibc used to maintain and so
1496 * getpid() in the child would return the parent's pid. This is all fixed in
1497 * newer glibc versions where the getpid() cache is removed and the pid/tid is
1498 * not reset anymore.
1499 * However, if for whatever reason you - dear committer - somehow need to get the
1500 * pid of the dummy intermediate process for do_share_ns() you need to call
1501 * lxc_raw_getpid(). The next lxc_raw_clone() call does not employ CLONE_VM and
1504 static inline int do_share_ns(void *arg
)
1507 struct lxc_handler
*handler
= arg
;
1509 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1510 if (handler
->nsfd
[i
] < 0)
1513 ret
= setns(handler
->nsfd
[i
], 0);
1516 * Note that joining a user and/or mount namespace
1517 * requires the process is not multithreaded otherwise
1518 * setns() will fail here.
1520 SYSERROR("Failed to inherit %s namespace",
1521 ns_info
[i
].proc_name
);
1525 DEBUG("Inherited %s namespace", ns_info
[i
].proc_name
);
1528 flags
= handler
->ns_on_clone_flags
;
1529 flags
|= CLONE_PARENT
;
1530 handler
->pid
= lxc_raw_clone_cb(do_start
, handler
, CLONE_PIDFD
| flags
,
1532 if (handler
->pid
< 0)
1538 /* lxc_spawn() performs crucial setup tasks and clone()s the new process which
1539 * exec()s the requested container binary.
1540 * Note that lxc_spawn() runs in the parent namespaces. Any operations performed
1541 * right here should be double checked if they'd pose a security risk. (For
1542 * example, any {u}mount() operations performed here will be reflected on the
1545 static int lxc_spawn(struct lxc_handler
*handler
)
1547 __do_close
int data_sock0
= -EBADF
, data_sock1
= -EBADF
;
1550 bool wants_to_map_ids
;
1551 struct lxc_list
*id_map
;
1552 const char *name
= handler
->name
;
1553 const char *lxcpath
= handler
->lxcpath
;
1554 bool share_ns
= false;
1555 struct lxc_conf
*conf
= handler
->conf
;
1556 struct cgroup_ops
*cgroup_ops
= handler
->cgroup_ops
;
1558 id_map
= &conf
->id_map
;
1559 wants_to_map_ids
= !lxc_list_empty(id_map
);
1561 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1562 if (!conf
->ns_share
[i
])
1565 handler
->nsfd
[i
] = lxc_inherit_namespace(conf
->ns_share
[i
], lxcpath
, ns_info
[i
].proc_name
);
1566 if (handler
->nsfd
[i
] < 0)
1572 ret
= lxc_sync_init(handler
);
1576 ret
= socketpair(AF_UNIX
, SOCK_STREAM
| SOCK_CLOEXEC
, 0,
1577 handler
->data_sock
);
1580 data_sock0
= handler
->data_sock
[0];
1581 data_sock1
= handler
->data_sock
[1];
1583 ret
= resolve_clone_flags(handler
);
1587 if (handler
->ns_clone_flags
& CLONE_NEWNET
) {
1588 ret
= lxc_find_gateway_addresses(handler
);
1590 ERROR("Failed to find gateway addresses");
1595 if (!cgroup_ops
->payload_create(cgroup_ops
, handler
)) {
1596 ERROR("Failed creating cgroups");
1597 goto out_delete_net
;
1600 /* If the rootfs is not a blockdev, prevent the container from marking
1602 * If the container is unprivileged then skip rootfs pinning.
1604 if (!wants_to_map_ids
) {
1605 handler
->pinfd
= pin_rootfs(conf
->rootfs
.path
);
1606 if (handler
->pinfd
== -EBADF
)
1607 INFO("Failed to pin the rootfs for container \"%s\"", handler
->name
);
1610 /* Create a process in a new set of namespaces. */
1611 handler
->ns_on_clone_flags
= handler
->ns_clone_flags
;
1612 if (handler
->ns_clone_flags
& CLONE_NEWUSER
) {
1613 /* If CLONE_NEWUSER and CLONE_NEWNET was requested, we need to
1614 * clone a new user namespace first and only later unshare our
1615 * network namespace to ensure that network devices ownership is
1618 handler
->ns_on_clone_flags
&= ~CLONE_NEWNET
;
1620 /* The cgroup namespace gets unshare()ed not clone()ed. */
1621 handler
->ns_on_clone_flags
&= ~CLONE_NEWCGROUP
;
1626 attacher_pid
= lxc_clone(do_share_ns
, handler
,
1627 CLONE_VFORK
| CLONE_VM
| CLONE_FILES
, NULL
);
1628 if (attacher_pid
< 0) {
1629 SYSERROR(LXC_CLONE_ERROR
);
1630 goto out_delete_net
;
1633 ret
= wait_for_pid(attacher_pid
);
1635 SYSERROR("Intermediate process failed");
1636 goto out_delete_net
;
1639 handler
->pid
= lxc_raw_clone_cb(do_start
, handler
,
1640 CLONE_PIDFD
| handler
->ns_on_clone_flags
,
1643 if (handler
->pid
< 0) {
1644 SYSERROR(LXC_CLONE_ERROR
);
1645 goto out_delete_net
;
1647 TRACE("Cloned child process %d", handler
->pid
);
1649 /* Verify that we can actually make use of pidfds. */
1650 if (!lxc_can_use_pidfd(handler
->pidfd
))
1651 close_prot_errno_disarm(handler
->pidfd
);
1653 ret
= snprintf(pidstr
, 20, "%d", handler
->pid
);
1654 if (ret
< 0 || ret
>= 20)
1655 goto out_delete_net
;
1657 ret
= setenv("LXC_PID", pidstr
, 1);
1659 SYSERROR("Failed to set environment variable: LXC_PID=%s", pidstr
);
1661 for (i
= 0; i
< LXC_NS_MAX
; i
++)
1662 if (handler
->ns_on_clone_flags
& ns_info
[i
].clone_flag
)
1663 INFO("Cloned %s", ns_info
[i
].flag_name
);
1665 if (!lxc_try_preserve_namespaces(handler
, handler
->ns_on_clone_flags
, handler
->pid
)) {
1666 ERROR("Failed to preserve cloned namespaces for lxc.hook.stop");
1667 goto out_delete_net
;
1670 lxc_sync_fini_child(handler
);
1672 if (lxc_abstract_unix_send_fds(handler
->data_sock
[0], &handler
->monitor_status_fd
, 1, NULL
, 0) < 0) {
1673 ERROR("Failed to send status file descriptor to child process");
1674 goto out_delete_net
;
1676 close_prot_errno_disarm(handler
->monitor_status_fd
);
1678 /* Map the container uids. The container became an invalid userid the
1679 * moment it was cloned with CLONE_NEWUSER. This call doesn't change
1680 * anything immediately, but allows the container to setuid(0) (0 being
1681 * mapped to something else on the host.) later to become a valid uid
1684 if (wants_to_map_ids
) {
1685 if (!handler
->conf
->ns_share
[LXC_NS_USER
] &&
1686 (handler
->conf
->ns_keep
& CLONE_NEWUSER
) == 0) {
1687 ret
= lxc_map_ids(id_map
, handler
->pid
);
1689 ERROR("Failed to set up id mapping.");
1690 goto out_delete_net
;
1695 ret
= lxc_sync_wake_child(handler
, LXC_SYNC_STARTUP
);
1697 goto out_delete_net
;
1699 ret
= lxc_sync_wait_child(handler
, LXC_SYNC_CONFIGURE
);
1701 goto out_delete_net
;
1703 if (!cgroup_ops
->setup_limits_legacy(cgroup_ops
, handler
->conf
, false)) {
1704 ERROR("Failed to setup cgroup limits for container \"%s\"", name
);
1705 goto out_delete_net
;
1708 if (!cgroup_ops
->payload_enter(cgroup_ops
, handler
)) {
1709 ERROR("Failed to enter cgroups");
1710 goto out_delete_net
;
1713 if (!cgroup_ops
->payload_delegate_controllers(cgroup_ops
)) {
1714 ERROR("Failed to delegate controllers to payload cgroup");
1715 goto out_delete_net
;
1718 if (!cgroup_ops
->setup_limits(cgroup_ops
, handler
)) {
1719 ERROR("Failed to setup cgroup limits for container \"%s\"", name
);
1720 goto out_delete_net
;
1723 if (!cgroup_ops
->chown(cgroup_ops
, handler
->conf
))
1724 goto out_delete_net
;
1726 /* If not done yet, we're now ready to preserve the network namespace */
1727 if (handler
->nsfd
[LXC_NS_NET
] < 0) {
1728 ret
= lxc_try_preserve_ns(handler
->pid
, "net");
1730 if (ret
!= -EOPNOTSUPP
) {
1731 SYSERROR("Failed to preserve net namespace");
1732 goto out_delete_net
;
1735 handler
->nsfd
[LXC_NS_NET
] = ret
;
1736 DEBUG("Preserved net namespace via fd %d", ret
);
1739 ret
= lxc_netns_set_nsid(handler
->nsfd
[LXC_NS_NET
]);
1741 SYSWARN("Failed to allocate new network namespace id");
1743 TRACE("Allocated new network namespace id");
1745 /* Create the network configuration. */
1746 if (handler
->ns_clone_flags
& CLONE_NEWNET
) {
1747 ret
= lxc_create_network(handler
);
1749 ERROR("Failed to create the network");
1750 goto out_delete_net
;
1753 ret
= lxc_network_send_to_child(handler
);
1755 ERROR("Failed to send veth names to child");
1756 goto out_delete_net
;
1760 if (!lxc_list_empty(&conf
->procs
)) {
1761 ret
= setup_proc_filesystem(&conf
->procs
, handler
->pid
);
1763 goto out_delete_net
;
1766 /* Tell the child to continue its initialization. We'll get
1767 * LXC_SYNC_CGROUP when it is ready for us to setup cgroups.
1769 ret
= lxc_sync_barrier_child(handler
, LXC_SYNC_POST_CONFIGURE
);
1771 goto out_delete_net
;
1773 if (!lxc_list_empty(&conf
->limits
)) {
1774 ret
= setup_resource_limits(&conf
->limits
, handler
->pid
);
1776 ERROR("Failed to setup resource limits");
1777 goto out_delete_net
;
1781 ret
= lxc_sync_barrier_child(handler
, LXC_SYNC_CGROUP_UNSHARE
);
1783 goto out_delete_net
;
1785 if (!cgroup_ops
->setup_limits_legacy(cgroup_ops
, handler
->conf
, true)) {
1786 ERROR("Failed to setup legacy device cgroup controller limits");
1787 goto out_delete_net
;
1789 TRACE("Set up legacy device cgroup controller limits");
1791 if (!cgroup_ops
->devices_activate(cgroup_ops
, handler
)) {
1792 ERROR("Failed to setup cgroup2 device controller limits");
1793 goto out_delete_net
;
1795 TRACE("Set up cgroup2 device controller limits");
1797 if (handler
->ns_clone_flags
& CLONE_NEWCGROUP
) {
1798 /* Now we're ready to preserve the cgroup namespace */
1799 ret
= lxc_try_preserve_ns(handler
->pid
, "cgroup");
1801 if (ret
!= -EOPNOTSUPP
) {
1802 SYSERROR("Failed to preserve cgroup namespace");
1803 goto out_delete_net
;
1806 handler
->nsfd
[LXC_NS_CGROUP
] = ret
;
1807 DEBUG("Preserved cgroup namespace via fd %d", ret
);
1811 cgroup_ops
->payload_finalize(cgroup_ops
);
1812 TRACE("Finished setting up cgroups");
1814 /* Run any host-side start hooks */
1815 ret
= run_lxc_hooks(name
, "start-host", conf
, NULL
);
1817 ERROR("Failed to run lxc.hook.start-host");
1818 goto out_delete_net
;
1821 /* Tell the child to complete its initialization and wait for it to exec
1822 * or return an error. (The child will never return
1823 * LXC_SYNC_READY_START+1. It will either close the sync pipe, causing
1824 * lxc_sync_barrier_child to return success, or return a different
1825 * value, causing us to error out).
1827 ret
= lxc_sync_barrier_child(handler
, LXC_SYNC_READY_START
);
1829 goto out_delete_net
;
1831 if (handler
->ns_clone_flags
& CLONE_NEWNET
) {
1832 ret
= lxc_network_recv_name_and_ifindex_from_child(handler
);
1834 ERROR("Failed to receive names and ifindices for network devices from child");
1835 goto out_delete_net
;
1839 /* Now all networks are created, network devices are moved into place,
1840 * and the correct names and ifindices in the respective namespaces have
1841 * been recorded. The corresponding structs have now all been filled. So
1842 * log them for debugging purposes.
1844 lxc_log_configured_netdevs(conf
);
1846 /* Read tty fds allocated by child. */
1847 ret
= lxc_recv_ttys_from_child(handler
);
1849 ERROR("Failed to receive tty info from child process");
1850 goto out_delete_net
;
1853 ret
= lxc_seccomp_recv_notifier_fd(&handler
->conf
->seccomp
, data_sock1
);
1855 SYSERROR("Failed to receive seccomp notify fd from child");
1856 goto out_delete_net
;
1859 ret
= handler
->ops
->post_start(handler
, handler
->data
);
1863 ret
= lxc_set_state(name
, handler
, RUNNING
);
1865 ERROR("Failed to set state to \"%s\"", lxc_state2str(RUNNING
));
1869 lxc_sync_fini(handler
);
1874 if (handler
->ns_clone_flags
& CLONE_NEWNET
)
1875 lxc_delete_network(handler
);
1881 lxc_sync_fini(handler
);
1882 close_prot_errno_disarm(handler
->pinfd
);
1887 int __lxc_start(struct lxc_handler
*handler
, struct lxc_operations
*ops
,
1888 void *data
, const char *lxcpath
, bool daemonize
, int *error_num
)
1891 const char *name
= handler
->name
;
1892 struct lxc_conf
*conf
= handler
->conf
;
1893 struct cgroup_ops
*cgroup_ops
;
1895 ret
= lxc_init(name
, handler
);
1897 ERROR("Failed to initialize container \"%s\"", name
);
1901 handler
->data
= data
;
1902 handler
->daemonize
= daemonize
;
1903 cgroup_ops
= handler
->cgroup_ops
;
1905 if (!attach_block_device(handler
->conf
)) {
1906 ERROR("Failed to attach block device");
1911 if (!cgroup_ops
->monitor_create(cgroup_ops
, handler
)) {
1912 ERROR("Failed to create monitor cgroup");
1917 if (!cgroup_ops
->monitor_enter(cgroup_ops
, handler
)) {
1918 ERROR("Failed to enter monitor cgroup");
1923 if (!cgroup_ops
->monitor_delegate_controllers(cgroup_ops
)) {
1924 ERROR("Failed to delegate controllers to monitor cgroup");
1929 if (geteuid() == 0 && !lxc_list_empty(&conf
->id_map
)) {
1930 /* If the backing store is a device, mount it here and now. */
1931 if (rootfs_is_blockdev(conf
)) {
1932 ret
= unshare(CLONE_NEWNS
);
1934 ERROR("Failed to unshare CLONE_NEWNS");
1937 INFO("Unshared CLONE_NEWNS");
1939 remount_all_slave();
1940 ret
= lxc_setup_rootfs_prepare_root(conf
, name
, lxcpath
);
1942 ERROR("Error setting up rootfs mount as root before spawn");
1945 INFO("Set up container rootfs as host root");
1949 ret
= lxc_spawn(handler
);
1951 ERROR("Failed to spawn container \"%s\"", name
);
1952 goto out_detach_blockdev
;
1955 handler
->conf
->reboot
= REBOOT_NONE
;
1957 ret
= lxc_poll(name
, handler
);
1959 ERROR("LXC mainloop exited with error: %d", ret
);
1960 goto out_delete_network
;
1963 if (!handler
->init_died
&& handler
->pid
> 0) {
1964 ERROR("Child process is not killed");
1966 goto out_delete_network
;
1969 status
= lxc_wait_for_pid_status(handler
->pid
);
1971 SYSERROR("Failed to retrieve status for %d", handler
->pid
);
1973 /* If the child process exited but was not signaled, it didn't call
1974 * reboot. This should mean it was an lxc-execute which simply exited.
1975 * In any case, treat it as a 'halt'.
1977 if (WIFSIGNALED(status
)) {
1978 switch(WTERMSIG(status
)) {
1979 case SIGINT
: /* halt */
1980 DEBUG("Container \"%s\" is halting", name
);
1982 case SIGHUP
: /* reboot */
1983 DEBUG("Container \"%s\" is rebooting", name
);
1984 handler
->conf
->reboot
= REBOOT_REQ
;
1986 case SIGSYS
: /* seccomp */
1987 DEBUG("Container \"%s\" violated its seccomp policy", name
);
1990 DEBUG("Unknown exit status for container \"%s\" init %d", name
, WTERMSIG(status
));
1995 ret
= lxc_restore_phys_nics_to_netns(handler
);
1997 ERROR("Failed to move physical network devices back to parent network namespace");
1999 close_prot_errno_disarm(handler
->pinfd
);
2001 lxc_monitor_send_exit_code(name
, status
, handler
->lxcpath
);
2002 lxc_error_set_and_log(handler
->pid
, status
);
2004 *error_num
= handler
->exit_status
;
2006 /* These are not the droids you are looking for. */
2008 lxc_delete_network(handler
);
2011 detach_block_device(handler
->conf
);
2018 /* These are the droids you are looking for. */
2021 goto __private_goto3
;
2023 out_detach_blockdev
:
2025 goto __private_goto2
;
2029 goto __private_goto1
;
2036 static int start(struct lxc_handler
*handler
, void* data
)
2038 struct start_args
*arg
= data
;
2040 NOTICE("Exec'ing \"%s\"", arg
->argv
[0]);
2042 execvp(arg
->argv
[0], arg
->argv
);
2043 SYSERROR("Failed to exec \"%s\"", arg
->argv
[0]);
2047 static int post_start(struct lxc_handler
*handler
, void* data
)
2049 struct start_args
*arg
= data
;
2051 NOTICE("Started \"%s\" with pid \"%d\"", arg
->argv
[0], handler
->pid
);
2055 static struct lxc_operations start_ops
= {
2057 .post_start
= post_start
2060 int lxc_start(char *const argv
[], struct lxc_handler
*handler
,
2061 const char *lxcpath
, bool daemonize
, int *error_num
)
2063 struct start_args start_arg
= {
2067 TRACE("Doing lxc_start");
2068 return __lxc_start(handler
, &start_ops
, &start_arg
, lxcpath
, daemonize
, error_num
);
2071 static void lxc_destroy_container_on_signal(struct lxc_handler
*handler
,
2074 char destroy
[PATH_MAX
];
2075 struct lxc_container
*c
;
2079 if (handler
->conf
->rootfs
.path
&& handler
->conf
->rootfs
.mount
) {
2080 bret
= do_destroy_container(handler
);
2082 ERROR("Error destroying rootfs for container \"%s\"", name
);
2086 INFO("Destroyed rootfs for container \"%s\"", name
);
2088 ret
= snprintf(destroy
, PATH_MAX
, "%s/%s", handler
->lxcpath
, name
);
2089 if (ret
< 0 || ret
>= PATH_MAX
) {
2090 ERROR("Error destroying directory for container \"%s\"", name
);
2094 c
= lxc_container_new(name
, handler
->lxcpath
);
2096 if (container_disk_lock(c
)) {
2097 INFO("Could not update lxc_snapshots file");
2098 lxc_container_put(c
);
2100 mod_all_rdeps(c
, false);
2101 container_disk_unlock(c
);
2102 lxc_container_put(c
);
2106 if (!handler
->am_root
)
2107 ret
= userns_exec_full(handler
->conf
, lxc_rmdir_onedev_wrapper
,
2108 destroy
, "lxc_rmdir_onedev_wrapper");
2110 ret
= lxc_rmdir_onedev(destroy
, NULL
);
2113 ERROR("Error destroying directory for container \"%s\"", name
);
2116 INFO("Destroyed directory for container \"%s\"", name
);
2119 static int lxc_rmdir_onedev_wrapper(void *data
)
2121 char *arg
= (char *) data
;
2122 return lxc_rmdir_onedev(arg
, NULL
);
2125 static bool do_destroy_container(struct lxc_handler
*handler
)
2129 if (!handler
->am_root
) {
2130 ret
= userns_exec_full(handler
->conf
, storage_destroy_wrapper
,
2131 handler
->conf
, "storage_destroy_wrapper");
2138 return storage_destroy(handler
->conf
);