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>
30 #include "cgroups/cgroup.h"
31 #include "cgroups/cgroup_utils.h"
33 #include "commands_utils.h"
37 #include "confile_utils.h"
39 #include "file_utils.h"
43 #include "lxccontainer.h"
45 #include "lxcseccomp.h"
48 #include "memory_utils.h"
50 #include "namespace.h"
52 #include "process_utils.h"
54 #include "storage/storage.h"
55 #include "storage/storage_utils.h"
57 #include "syscall_wrappers.h"
62 #include <sys/capability.h>
69 lxc_log_define(start
, lxc
);
71 extern void mod_all_rdeps(struct lxc_container
*c
, bool inc
);
72 static bool do_destroy_container(struct lxc_handler
*handler
);
73 static int lxc_rmdir_onedev_wrapper(void *data
);
74 static void lxc_destroy_container_on_signal(struct lxc_handler
*handler
,
77 static void print_top_failing_dir(const char *path
)
79 __do_free
char *copy
= NULL
;
83 copy
= must_copy_string(path
);
85 e
= copy
+ strlen(path
);
88 while (p
< e
&& *p
== '/')
91 while (p
< e
&& *p
!= '/')
97 ret
= access(copy
, X_OK
);
99 SYSERROR("Could not access %s. Please grant it x access, or add an ACL for the container " "root", copy
);
106 static void lxc_put_nsfds(struct lxc_handler
*handler
)
108 for (int i
= 0; i
< LXC_NS_MAX
; i
++) {
109 if (handler
->nsfd
[i
] < 0)
112 close_prot_errno_disarm(handler
->nsfd
[i
]);
116 static int lxc_try_preserve_namespace(struct lxc_handler
*handler
,
117 lxc_namespace_t idx
, const char *ns
)
119 __do_close
int fd
= -EBADF
;
122 fd
= lxc_preserve_ns(handler
->pid
, ns
);
126 ret
= strnprintf(handler
->nsfd_paths
[idx
],
127 sizeof(handler
->nsfd_paths
[idx
]), "%s:/proc/%d/fd/%d",
128 ns_info
[idx
].proc_name
, handler
->monitor_pid
, fd
);
130 return ret_errno(EIO
);
133 * In case LXC is configured for exposing information to hooks as
134 * argv-style arguments prepare an argv array we can use.
136 handler
->hook_argv
[handler
->hook_argc
] = handler
->nsfd_paths
[idx
];
137 handler
->hook_argc
++;
139 DEBUG("Preserved %s namespace via fd %d and stashed path as %s",
140 ns_info
[idx
].proc_name
, fd
, handler
->nsfd_paths
[idx
]);
142 handler
->nsfd
[idx
] = move_fd(fd
);
146 /* lxc_try_preserve_namespaces: open /proc/@pid/ns/@ns for each namespace
147 * specified in ns_clone_flags.
148 * Return true on success, false on failure.
150 static bool lxc_try_preserve_namespaces(struct lxc_handler
*handler
,
153 for (lxc_namespace_t ns_idx
= 0; ns_idx
< LXC_NS_MAX
; ns_idx
++)
154 handler
->nsfd
[ns_idx
] = -EBADF
;
156 for (lxc_namespace_t ns_idx
= 0; ns_idx
< LXC_NS_MAX
; ns_idx
++) {
158 const char *ns
= ns_info
[ns_idx
].proc_name
;
160 if ((ns_clone_flags
& ns_info
[ns_idx
].clone_flag
) == 0)
163 ret
= lxc_try_preserve_namespace(handler
, ns_idx
,
164 ns_info
[ns_idx
].proc_name
);
166 if (ret
== -ENOENT
) {
167 SYSERROR("Kernel does not support preserving %s namespaces", ns
);
172 * Handle kernels that do not support interacting with
173 * namespaces through procfs.
175 lxc_put_nsfds(handler
);
176 return log_error_errno(false, errno
, "Failed to preserve %s namespace", ns
);
183 static inline bool match_stdfds(int fd
)
185 return (fd
== STDIN_FILENO
|| fd
== STDOUT_FILENO
|| fd
== STDERR_FILENO
);
189 static bool match_dlog_fds(struct dirent
*direntp
)
191 char path
[PATH_MAX
] = {0};
192 char link
[PATH_MAX
] = {0};
196 ret
= strnprintf(path
, sizeof(path
), "/proc/self/fd/%s", direntp
->d_name
);
198 return log_error(false, "Failed to create file descriptor name");
200 linklen
= readlink(path
, link
, PATH_MAX
);
202 return log_error(false, "Failed to read link path - \"%s\"", path
);
203 else if (linklen
>= PATH_MAX
)
204 return log_error(false, "The name of link path is too long - \"%s\"", path
);
206 if (strequal(link
, "/dev/log_main") ||
207 strequal(link
, "/dev/log_system") ||
208 strequal(link
, "/dev/log_radio"))
215 /* Parses the LISTEN_FDS environment variable value.
216 * The returned value is the highest fd number up to which the
217 * file descriptors must be passed to the container process.
219 * For example, if LISTEN_FDS=2 then 4 is returned and file descriptors 3 and 4
220 * MUST be passed to the container process (in addition to the standard streams)
221 * to support [socket activation][systemd-listen-fds].
223 static unsigned int get_listen_fds_max(void)
226 unsigned int num_fds
;
229 val
= getenv("LISTEN_FDS");
233 ret
= lxc_safe_uint(val
, &num_fds
);
235 return syserror_ret(0, "Failed to parse \"LISTEN_FDS=%s\" environment variable", val
);
237 return log_trace(num_fds
, "Parsed \"LISTEN_FDS=%s\" environment variable", val
);
240 int lxc_check_inherited(struct lxc_conf
*conf
, bool closeall
,
241 int *fds_to_ignore
, size_t len_fds
)
246 struct dirent
*direntp
;
247 unsigned int listen_fds_max
;
248 struct lxc_state_client
*client
, *nclient
;
250 if (conf
&& conf
->close_all_fds
)
253 listen_fds_max
= get_listen_fds_max();
256 * Disable syslog at this point to avoid the above logging
257 * function to open a new fd and make the check_inherited function
258 * enter an infinite loop.
260 lxc_log_syslog_disable();
263 dir
= opendir("/proc/self/fd");
265 return log_warn(-1, "Failed to open directory");
269 while ((direntp
= readdir(dir
))) {
271 bool matched
= false;
273 if (strequal(direntp
->d_name
, "."))
276 if (strequal(direntp
->d_name
, ".."))
279 ret
= lxc_safe_int(direntp
->d_name
, &fd
);
281 INFO("Could not parse file descriptor for \"%s\"", direntp
->d_name
);
285 for (i
= 0; i
< len_fds
; i
++)
286 if (fds_to_ignore
[i
] == fd
)
289 if (fd
== fddir
|| fd
== lxc_log_fd
||
290 (i
< len_fds
&& fd
== fds_to_ignore
[i
]))
293 /* Keep state clients that wait on reboots. */
295 list_for_each_entry_safe(client
, nclient
, &conf
->state_clients
, head
) {
296 if (client
->clientfd
!= fd
)
307 if (current_config
&& fd
== current_config
->logfd
)
310 if (match_stdfds(fd
))
314 if (match_dlog_fds(direntp
))
319 if (fd
<= listen_fds_max
) {
320 INFO("Inheriting fd %d (using the LISTEN_FDS environment variable)", fd
);
326 SYSINFO("Closed inherited fd %d", fd
);
328 INFO("Closed inherited fd %d", fd
);
332 WARN("Inherited fd %d", fd
);
337 * Only enable syslog at this point to avoid the above logging
338 * function to open a new fd and make the check_inherited function
339 * enter an infinite loop.
341 lxc_log_syslog_enable();
346 static int setup_signal_fd(sigset_t
*oldmask
)
350 const int signals
[] = {SIGBUS
, SIGILL
, SIGSEGV
, SIGWINCH
};
352 /* Block everything except serious error signals. */
353 ret
= sigfillset(&mask
);
357 for (int sig
= 0; sig
< (sizeof(signals
) / sizeof(signals
[0])); sig
++) {
358 ret
= sigdelset(&mask
, signals
[sig
]);
363 ret
= pthread_sigmask(SIG_BLOCK
, &mask
, oldmask
);
365 return log_error_errno(-EBADF
, errno
,
366 "Failed to set signal mask");
368 ret
= signalfd(-1, &mask
, SFD_CLOEXEC
);
370 return log_error_errno(-EBADF
,
371 errno
, "Failed to create signal file descriptor");
373 TRACE("Created signal file descriptor %d", ret
);
378 static int signal_handler(int fd
, uint32_t events
, void *data
,
379 struct lxc_async_descr
*descr
)
383 struct signalfd_siginfo siginfo
;
384 struct lxc_handler
*hdlr
= data
;
386 ret
= lxc_read_nointr(fd
, &siginfo
, sizeof(siginfo
));
388 return log_error(LXC_MAINLOOP_ERROR
, "Failed to read signal info from signal file descriptor %d", fd
);
390 if (ret
!= sizeof(siginfo
))
391 return log_error(LXC_MAINLOOP_ERROR
, "Unexpected size for struct signalfd_siginfo");
393 /* Check whether init is running. */
395 ret
= waitid(P_PID
, hdlr
->pid
, &info
, WEXITED
| WNOWAIT
| WNOHANG
);
396 if (ret
== 0 && info
.si_pid
== hdlr
->pid
)
397 hdlr
->init_died
= true;
399 TRACE("Received signal ssi_signo(%d) for ssi_pid(%d), si_signo(%d), si_pid(%d)",
400 siginfo
.ssi_signo
, siginfo
.ssi_pid
, info
.si_signo
, info
.si_pid
);
402 /* Try to figure out a reasonable exit status to report. */
403 if (hdlr
->init_died
) {
404 switch (info
.si_code
) {
406 hdlr
->exit_status
= info
.si_status
<< 8;
411 hdlr
->exit_status
= info
.si_status
<< 8 | 0x7f;
414 /* Huh? The waitid() told us it's dead *and* continued? */
415 WARN("Init %d dead and continued?", hdlr
->pid
);
416 hdlr
->exit_status
= 1;
419 ERROR("Unknown si_code: %d", info
.si_code
);
420 hdlr
->exit_status
= 1;
424 if (siginfo
.ssi_signo
== SIGHUP
) {
425 if (hdlr
->pidfd
>= 0)
426 lxc_raw_pidfd_send_signal(hdlr
->pidfd
, SIGTERM
, NULL
, 0);
428 kill(hdlr
->pid
, SIGTERM
);
429 INFO("Killing %d since terminal hung up", hdlr
->pid
);
430 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
431 : LXC_MAINLOOP_CONTINUE
;
434 if (siginfo
.ssi_signo
!= SIGCHLD
) {
435 if (hdlr
->pidfd
>= 0)
436 lxc_raw_pidfd_send_signal(hdlr
->pidfd
,
437 siginfo
.ssi_signo
, NULL
, 0);
439 kill(hdlr
->pid
, siginfo
.ssi_signo
);
440 INFO("Forwarded signal %d to pid %d", siginfo
.ssi_signo
, hdlr
->pid
);
441 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
442 : LXC_MAINLOOP_CONTINUE
;
445 /* More robustness, protect ourself from a SIGCHLD sent
446 * by a process different from the container init.
448 if (siginfo
.ssi_pid
!= hdlr
->pid
) {
449 NOTICE("Received %d from pid %d instead of container init %d",
450 siginfo
.ssi_signo
, siginfo
.ssi_pid
, hdlr
->pid
);
451 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
452 : LXC_MAINLOOP_CONTINUE
;
455 if (siginfo
.ssi_code
== CLD_STOPPED
) {
456 INFO("Container init process was stopped");
457 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
458 : LXC_MAINLOOP_CONTINUE
;
461 if (siginfo
.ssi_code
== CLD_CONTINUED
) {
462 INFO("Container init process was continued");
463 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
464 : LXC_MAINLOOP_CONTINUE
;
467 return log_debug(LXC_MAINLOOP_CLOSE
, "Container init process %d exited", hdlr
->pid
);
470 int lxc_serve_state_clients(const char *name
, struct lxc_handler
*handler
,
473 struct lxc_msg msg
= {
474 .type
= lxc_msg_state
,
479 struct lxc_state_client
*client
, *nclient
;
482 handler
->state
= RUNNING
;
484 handler
->state
= state
;
486 TRACE("Set container state to %s", lxc_state2str(state
));
488 if (list_empty(&handler
->conf
->state_clients
))
489 return log_trace(0, "No state clients registered");
491 retlen
= strlcpy(msg
.name
, name
, sizeof(msg
.name
));
492 if (retlen
>= sizeof(msg
.name
))
495 list_for_each_entry_safe(client
, nclient
, &handler
->conf
->state_clients
, head
) {
496 if (client
->states
[state
] == 0) {
497 TRACE("State %s not registered for state client %d",
498 lxc_state2str(state
), client
->clientfd
);
502 TRACE("Sending state %s to state client %d",
503 lxc_state2str(state
), client
->clientfd
);
505 ret
= lxc_send_nointr(client
->clientfd
, &msg
, sizeof(msg
), MSG_NOSIGNAL
);
507 SYSERROR("Failed to send message to client");
509 /* kick client from list */
510 list_del(&client
->head
);
511 close(client
->clientfd
);
518 static int lxc_serve_state_socket_pair(const char *name
,
519 struct lxc_handler
*handler
,
524 if (!handler
->daemonize
||
525 handler
->state_socket_pair
[1] < 0 ||
529 /* Close read end of the socket pair. */
530 close_prot_errno_disarm(handler
->state_socket_pair
[0]);
533 ret
= lxc_abstract_unix_send_credential(handler
->state_socket_pair
[1],
534 &(int){state
}, sizeof(int));
536 SYSERROR("Failed to send state to %d", handler
->state_socket_pair
[1]);
544 if (ret
!= sizeof(int))
545 return log_error(-1, "Message too long : %d", handler
->state_socket_pair
[1]);
547 TRACE("Sent container state \"%s\" to %d", lxc_state2str(state
),
548 handler
->state_socket_pair
[1]);
550 /* Close write end of the socket pair. */
551 close_prot_errno_disarm(handler
->state_socket_pair
[1]);
556 int lxc_set_state(const char *name
, struct lxc_handler
*handler
,
561 ret
= lxc_serve_state_socket_pair(name
, handler
, state
);
563 return log_error(-1, "Failed to synchronize via anonymous pair of unix sockets");
565 ret
= lxc_serve_state_clients(name
, handler
, state
);
569 /* This function will try to connect to the legacy lxc-monitord state
570 * server and only exists for backwards compatibility.
572 lxc_monitor_send_state(name
, state
, handler
->lxcpath
);
577 int lxc_poll(const char *name
, struct lxc_handler
*handler
)
580 struct lxc_terminal
*console
= &handler
->conf
->console
;
581 struct lxc_async_descr descr
, descr_console
;
583 if (!wants_console(console
))
586 ret
= lxc_mainloop_open(&descr
);
588 ERROR("Failed to create mainloop");
593 ret
= lxc_mainloop_open(&descr_console
);
595 ERROR("Failed to create console mainloop");
600 ret
= lxc_mainloop_add_handler(&descr
, handler
->sigfd
,
602 default_cleanup_handler
,
603 handler
, "signal_handler");
605 ERROR("Failed to add signal handler for %d to mainloop", handler
->sigfd
);
606 goto out_mainloop_console
;
609 ret
= lxc_seccomp_setup_proxy(&handler
->conf
->seccomp
, &descr
, handler
);
611 ERROR("Failed to setup seccomp proxy");
612 goto out_mainloop_console
;
616 ret
= lxc_terminal_mainloop_add(&descr
, console
);
618 ERROR("Failed to add console handlers to mainloop");
619 goto out_mainloop_console
;
623 ret
= lxc_cmd_mainloop_add(name
, &descr
, handler
);
625 ERROR("Failed to add command handler to mainloop");
626 goto out_mainloop_console
;
629 TRACE("Mainloop is ready");
631 ret
= lxc_mainloop(&descr
, -1);
632 close_prot_errno_disarm(descr
.epfd
);
633 if (ret
< 0 || !handler
->init_died
)
634 goto out_mainloop_console
;
637 ret
= lxc_terminal_mainloop_add(&descr_console
, console
);
639 ret
= lxc_mainloop(&descr_console
, 0);
642 out_mainloop_console
:
644 lxc_mainloop_close(&descr_console
);
645 TRACE("Closed console mainloop");
649 lxc_mainloop_close(&descr
);
650 TRACE("Closed mainloop");
653 TRACE("Closed signal file descriptor %d", handler
->sigfd
);
654 close_prot_errno_disarm(handler
->sigfd
);
659 void lxc_put_handler(struct lxc_handler
*handler
)
661 close_prot_errno_disarm(handler
->pidfd
);
662 close_prot_errno_disarm(handler
->sigfd
);
663 lxc_put_nsfds(handler
);
664 if (handler
->conf
&& handler
->conf
->reboot
== REBOOT_NONE
)
665 close_prot_errno_disarm(handler
->conf
->maincmd_fd
);
666 close_prot_errno_disarm(handler
->monitor_status_fd
);
667 close_prot_errno_disarm(handler
->state_socket_pair
[0]);
668 close_prot_errno_disarm(handler
->state_socket_pair
[1]);
669 cgroup_exit(handler
->cgroup_ops
);
670 if (handler
->conf
&& handler
->conf
->reboot
== REBOOT_NONE
)
671 free_disarm(handler
);
673 handler
->conf
= NULL
;
676 struct lxc_handler
*lxc_init_handler(struct lxc_handler
*old
,
677 const char *name
, struct lxc_conf
*conf
,
678 const char *lxcpath
, bool daemonize
)
682 struct lxc_handler
*handler
;
685 handler
= zalloc(sizeof(*handler
));
691 /* Note that am_guest_unpriv() checks the effective uid. We
692 * probably don't care if we are real root only if we are running
693 * as root so this should be fine.
695 handler
->am_root
= !am_guest_unpriv();
696 handler
->conf
= conf
;
697 handler
->lxcpath
= lxcpath
;
698 handler
->init_died
= false;
699 handler
->data_sock
[0] = -EBADF
;
700 handler
->data_sock
[1] = -EBADF
;
701 handler
->monitor_status_fd
= -EBADF
;
702 handler
->pidfd
= -EBADF
;
703 handler
->sigfd
= -EBADF
;
704 handler
->state_socket_pair
[0] = -EBADF
;
705 handler
->state_socket_pair
[1] = -EBADF
;
706 if (handler
->conf
->reboot
== REBOOT_NONE
)
707 INIT_LIST_HEAD(&handler
->conf
->state_clients
);
709 for (lxc_namespace_t idx
= 0; idx
< LXC_NS_MAX
; idx
++) {
710 handler
->nsfd
[idx
] = -EBADF
;
712 if (handler
->conf
->reboot
== REBOOT_NONE
)
715 handler
->nsfd_paths
[idx
][0] = '\0';
716 handler
->hook_argv
[idx
] = NULL
;
718 if (handler
->hook_argc
!= 0)
719 handler
->hook_argc
= 0;
722 handler
->name
= name
;
724 handler
->transient_pid
= lxc_raw_getpid();
726 handler
->transient_pid
= -1;
728 if (daemonize
&& handler
->conf
->reboot
== REBOOT_NONE
) {
729 /* Create socketpair() to synchronize on daemonized startup.
730 * When the container reboots we don't need to synchronize
731 * again currently so don't open another socketpair().
733 ret
= socketpair(AF_UNIX
, SOCK_STREAM
| SOCK_CLOEXEC
, 0,
734 handler
->state_socket_pair
);
736 ERROR("Failed to create anonymous pair of unix sockets");
740 TRACE("Created anonymous pair {%d,%d} of unix sockets",
741 handler
->state_socket_pair
[0],
742 handler
->state_socket_pair
[1]);
743 handler
->keep_fds
[nr_keep_fds
++] = handler
->state_socket_pair
[0];
744 handler
->keep_fds
[nr_keep_fds
++] = handler
->state_socket_pair
[1];
747 if (handler
->conf
->reboot
== REBOOT_NONE
) {
748 handler
->conf
->maincmd_fd
= lxc_server_init(name
, lxcpath
, "command");
749 if (handler
->conf
->maincmd_fd
< 0) {
750 ERROR("Failed to set up command socket");
753 handler
->keep_fds
[nr_keep_fds
++] = handler
->conf
->maincmd_fd
;
756 TRACE("Unix domain socket %d for command server is ready",
757 handler
->conf
->maincmd_fd
);
762 lxc_put_handler(handler
);
767 int lxc_init(const char *name
, struct lxc_handler
*handler
)
769 __do_close
int status_fd
= -EBADF
;
771 const char *loglevel
;
772 struct lxc_conf
*conf
= handler
->conf
;
774 handler
->monitor_pid
= lxc_raw_getpid();
775 status_fd
= open("/proc/self/status", O_RDONLY
| O_CLOEXEC
);
777 return log_error_errno(-1, errno
, "Failed to open monitor status fd");
779 handler
->lsm_ops
= lsm_init_static();
780 TRACE("Initialized LSM");
782 /* Begin by setting the state to STARTING. */
783 ret
= lxc_set_state(name
, handler
, STARTING
);
785 return log_error(-1, "Failed to set state to \"%s\"", lxc_state2str(STARTING
));
786 TRACE("Set container state to \"STARTING\"");
788 /* Start of environment variable setup for hooks. */
789 ret
= setenv("LXC_NAME", name
, 1);
791 SYSERROR("Failed to set environment variable: LXC_NAME=%s", name
);
794 ret
= setenv("LXC_CONFIG_FILE", conf
->rcfile
, 1);
796 SYSERROR("Failed to set environment variable: LXC_CONFIG_FILE=%s", conf
->rcfile
);
799 if (conf
->rootfs
.mount
) {
800 ret
= setenv("LXC_ROOTFS_MOUNT", conf
->rootfs
.mount
, 1);
802 SYSERROR("Failed to set environment variable: LXC_ROOTFS_MOUNT=%s", conf
->rootfs
.mount
);
805 if (conf
->rootfs
.path
) {
806 ret
= setenv("LXC_ROOTFS_PATH", conf
->rootfs
.path
, 1);
808 SYSERROR("Failed to set environment variable: LXC_ROOTFS_PATH=%s", conf
->rootfs
.path
);
811 if (conf
->console
.path
) {
812 ret
= setenv("LXC_CONSOLE", conf
->console
.path
, 1);
814 SYSERROR("Failed to set environment variable: LXC_CONSOLE=%s", conf
->console
.path
);
817 if (conf
->console
.log_path
) {
818 ret
= setenv("LXC_CONSOLE_LOGPATH", conf
->console
.log_path
, 1);
820 SYSERROR("Failed to set environment variable: LXC_CONSOLE_LOGPATH=%s", conf
->console
.log_path
);
823 if (cgns_supported()) {
824 ret
= setenv("LXC_CGNS_AWARE", "1", 1);
826 SYSERROR("Failed to set environment variable LXC_CGNS_AWARE=1");
829 loglevel
= lxc_log_priority_to_string(lxc_log_get_level());
830 ret
= setenv("LXC_LOG_LEVEL", loglevel
, 1);
832 SYSERROR("Set environment variable LXC_LOG_LEVEL=%s", loglevel
);
834 if (conf
->hooks_version
== 0)
835 ret
= setenv("LXC_HOOK_VERSION", "0", 1);
837 ret
= setenv("LXC_HOOK_VERSION", "1", 1);
839 SYSERROR("Failed to set environment variable LXC_HOOK_VERSION=%u", conf
->hooks_version
);
840 /* End of environment variable setup for hooks. */
842 TRACE("Set environment variables");
844 ret
= run_lxc_hooks(name
, "pre-start", conf
, NULL
);
846 return log_error(-1, "Failed to run lxc.hook.pre-start for container \"%s\"", name
);
847 TRACE("Ran pre-start hooks");
849 ret
= lxc_terminal_parent(conf
);
851 return log_error(-1, "Failed to allocate terminal");
853 /* The signal fd has to be created before forking otherwise if the child
854 * process exits before we setup the signal fd, the event will be lost
855 * and the command will be stuck.
857 handler
->sigfd
= setup_signal_fd(&handler
->oldmask
);
858 if (handler
->sigfd
< 0)
859 return log_error(-1, "Failed to setup SIGCHLD fd handler.");
860 TRACE("Set up signal fd");
862 handler
->cgroup_ops
= cgroup_init(handler
->conf
);
863 if (!handler
->cgroup_ops
) {
864 ERROR("Failed to initialize cgroup driver");
865 goto out_restore_sigmask
;
867 TRACE("Initialized cgroup driver");
869 ret
= lxc_read_seccomp_config(conf
);
871 ERROR("Failed to read seccomp policy");
872 goto out_restore_sigmask
;
874 TRACE("Read seccomp policy");
876 ret
= handler
->lsm_ops
->prepare(handler
->lsm_ops
, conf
, handler
->lxcpath
);
878 ERROR("Failed to initialize LSM");
879 goto out_restore_sigmask
;
881 TRACE("Initialized LSM");
883 INFO("Container \"%s\" is initialized", name
);
884 handler
->monitor_status_fd
= move_fd(status_fd
);
888 (void)pthread_sigmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
893 void lxc_expose_namespace_environment(const struct lxc_handler
*handler
)
895 for (lxc_namespace_t i
= 0; i
< LXC_NS_MAX
; i
++) {
899 if (handler
->nsfd
[i
] < 0)
902 fd_path
= handler
->nsfd_paths
[i
] + strcspn(handler
->nsfd_paths
[i
], "/");
903 ret
= setenv(ns_info
[i
].env_name
, fd_path
, 1);
905 SYSERROR("Failed to set environment variable %s=%s",
906 ns_info
[i
].env_name
, fd_path
);
908 TRACE("Set environment variable %s=%s",
909 ns_info
[i
].env_name
, fd_path
);
913 void lxc_end(struct lxc_handler
*handler
)
916 const char *name
= handler
->name
;
917 struct cgroup_ops
*cgroup_ops
= handler
->cgroup_ops
;
918 struct lxc_state_client
*client
, *nclient
;
920 /* The STOPPING state is there for future cleanup code which can take
923 lxc_set_state(name
, handler
, STOPPING
);
925 /* Passing information to hooks via environment variables. */
926 if (handler
->conf
->hooks_version
> 0)
927 lxc_expose_namespace_environment(handler
);
929 if (handler
->conf
->reboot
> REBOOT_NONE
) {
930 ret
= setenv("LXC_TARGET", "reboot", 1);
932 SYSERROR("Failed to set environment variable: LXC_TARGET=reboot");
935 if (handler
->conf
->reboot
== REBOOT_NONE
) {
936 ret
= setenv("LXC_TARGET", "stop", 1);
938 SYSERROR("Failed to set environment variable: LXC_TARGET=stop");
941 if (handler
->conf
->hooks_version
== 0)
942 ret
= run_lxc_hooks(name
, "stop", handler
->conf
, handler
->hook_argv
);
944 ret
= run_lxc_hooks(name
, "stop", handler
->conf
, NULL
);
946 ERROR("Failed to run \"lxc.hook.stop\" hook");
948 handler
->lsm_ops
->cleanup(handler
->lsm_ops
, handler
->conf
, handler
->lxcpath
);
951 cgroup_ops
->payload_destroy(cgroup_ops
, handler
);
952 cgroup_ops
->monitor_destroy(cgroup_ops
, handler
);
955 put_lxc_rootfs(&handler
->conf
->rootfs
, true);
957 if (handler
->conf
->reboot
== REBOOT_NONE
) {
958 /* For all new state clients simply close the command socket.
959 * This will inform all state clients that the container is
960 * STOPPED and also prevents a race between a open()/close() on
961 * the command socket causing a new process to get ECONNREFUSED
962 * because we haven't yet closed the command socket.
964 close_prot_errno_disarm(handler
->conf
->maincmd_fd
);
965 TRACE("Closed command socket");
967 /* This function will try to connect to the legacy lxc-monitord
968 * state server and only exists for backwards compatibility.
970 lxc_monitor_send_state(name
, STOPPED
, handler
->lxcpath
);
972 /* The command socket is closed so no one can acces the command
973 * socket anymore so there's no need to lock it.
975 handler
->state
= STOPPED
;
976 TRACE("Set container state to \"STOPPED\"");
978 lxc_set_state(name
, handler
, STOPPED
);
979 TRACE("Set container state to \"STOPPED\"");
982 /* Avoid lingering namespace references. */
983 lxc_put_nsfds(handler
);
985 ret
= run_lxc_hooks(name
, "post-stop", handler
->conf
, NULL
);
987 ERROR("Failed to run lxc.hook.post-stop for container \"%s\"", name
);
988 if (handler
->conf
->reboot
> REBOOT_NONE
) {
989 WARN("Container will be stopped instead of rebooted");
990 handler
->conf
->reboot
= REBOOT_NONE
;
992 ret
= setenv("LXC_TARGET", "stop", 1);
994 WARN("Failed to set environment variable: LXC_TARGET=stop");
998 /* Reset mask set by setup_signal_fd. */
999 ret
= pthread_sigmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
1001 SYSWARN("Failed to restore signal mask");
1003 lxc_terminal_delete(&handler
->conf
->console
);
1004 lxc_delete_tty(&handler
->conf
->ttys
);
1005 close_prot_errno_disarm(handler
->conf
->devpts_fd
);
1007 /* The command socket is now closed, no more state clients can register
1008 * themselves from now on. So free the list of state clients.
1010 list_for_each_entry_safe(client
, nclient
, &handler
->conf
->state_clients
, head
) {
1011 /* Keep state clients that want to be notified about reboots. */
1012 if ((handler
->conf
->reboot
> REBOOT_NONE
) &&
1013 (client
->states
[RUNNING
] == 2))
1016 /* close state client socket */
1017 list_del(&client
->head
);
1018 close(client
->clientfd
);
1022 if (handler
->conf
->ephemeral
== 1 && handler
->conf
->reboot
!= REBOOT_REQ
)
1023 lxc_destroy_container_on_signal(handler
, name
);
1025 lxc_put_handler(handler
);
1028 void lxc_abort(struct lxc_handler
*handler
)
1033 lxc_set_state(handler
->name
, handler
, ABORTING
);
1035 if (handler
->pidfd
>= 0) {
1036 ret
= lxc_raw_pidfd_send_signal(handler
->pidfd
, SIGKILL
, NULL
, 0);
1038 SYSWARN("Failed to send SIGKILL via pidfd %d for process %d",
1039 handler
->pidfd
, handler
->pid
);
1042 if ((!ret
|| errno
!= ESRCH
) && handler
->pid
> 0)
1043 if (kill(handler
->pid
, SIGKILL
))
1044 SYSWARN("Failed to send SIGKILL to %d", handler
->pid
);
1047 ret
= waitpid(-1, &status
, 0);
1051 static int do_start(void *data
)
1053 struct lxc_handler
*handler
= data
;
1054 __lxc_unused __do_close
int data_sock0
= handler
->data_sock
[0],
1055 data_sock1
= handler
->data_sock
[1];
1056 __do_close
int devnull_fd
= -EBADF
, status_fd
= -EBADF
;
1063 lxc_sync_fini_parent(handler
);
1065 if (lxc_abstract_unix_recv_one_fd(data_sock1
, &status_fd
, NULL
, 0) < 0) {
1066 ERROR("Failed to receive status file descriptor from parent process");
1067 goto out_warn_father
;
1070 /* This prctl must be before the synchro, so if the parent dies before
1071 * we set the parent death signal, we will detect its death with the
1072 * synchro right after, otherwise we have a window where the parent can
1073 * exit before we set the pdeath signal leading to a unsupervized
1076 ret
= lxc_set_death_signal(SIGKILL
, handler
->monitor_pid
, status_fd
);
1078 SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL");
1079 goto out_warn_father
;
1082 ret
= lxc_ambient_caps_up();
1084 ERROR("Failed to raise ambient capabilities");
1085 goto out_warn_father
;
1088 ret
= pthread_sigmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
1090 SYSERROR("Failed to set signal mask");
1091 goto out_warn_father
;
1094 if (!lxc_sync_wait_parent(handler
, START_SYNC_STARTUP
))
1095 goto out_warn_father
;
1097 /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See
1098 * https://github.com/lxc/lxd/issues/1978.
1100 if (handler
->ns_unshare_flags
& CLONE_NEWNET
) {
1101 ret
= unshare(CLONE_NEWNET
);
1103 SYSERROR("Failed to unshare CLONE_NEWNET");
1104 goto out_warn_father
;
1106 INFO("Unshared CLONE_NEWNET");
1109 /* If we are in a new user namespace, become root there to have
1110 * privilege over our namespace.
1112 if (!list_empty(&handler
->conf
->id_map
)) {
1113 if (!handler
->conf
->root_nsuid_map
)
1114 nsuid
= handler
->conf
->init_uid
;
1116 if (!handler
->conf
->root_nsgid_map
)
1117 nsgid
= handler
->conf
->init_gid
;
1119 /* Drop groups only after we switched to a valid gid in the new
1122 if (!lxc_drop_groups() &&
1123 (handler
->am_root
|| errno
!= EPERM
))
1124 goto out_warn_father
;
1126 if (!lxc_switch_uid_gid(nsuid
, nsgid
))
1127 goto out_warn_father
;
1129 ret
= prctl(PR_SET_DUMPABLE
, prctl_arg(1), prctl_arg(0),
1130 prctl_arg(0), prctl_arg(0));
1132 goto out_warn_father
;
1134 /* set{g,u}id() clears deathsignal */
1135 ret
= lxc_set_death_signal(SIGKILL
, handler
->monitor_pid
, status_fd
);
1137 SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL");
1138 goto out_warn_father
;
1142 ret
= access(handler
->lxcpath
, X_OK
);
1144 print_top_failing_dir(handler
->lxcpath
);
1145 goto out_warn_father
;
1148 /* In order to checkpoint restore, we need to have everything in the
1149 * same mount namespace. However, some containers may not have a
1150 * reasonable /dev (in particular, they may not have /dev/null), so we
1151 * can't set init's std fds to /dev/null by opening it from inside the
1154 * If that's the case, fall back to using the host's /dev/null. This
1155 * means that migration won't work, but at least we won't spew output
1156 * where it isn't wanted.
1158 if (handler
->daemonize
&& !handler
->conf
->autodev
) {
1159 char path
[PATH_MAX
];
1161 ret
= strnprintf(path
, sizeof(path
), "%s/dev/null",
1162 handler
->conf
->rootfs
.mount
);
1164 goto out_warn_father
;
1166 ret
= access(path
, F_OK
);
1168 devnull_fd
= open_devnull();
1171 goto out_warn_father
;
1172 WARN("Using /dev/null from the host for container init's standard file descriptors. Migration will not work");
1177 * Tell the parent task it can begin to configure the container and wait
1180 if (!lxc_sync_wake_parent(handler
, START_SYNC_CONFIGURE
))
1183 /* Unshare cgroup namespace after we have setup our cgroups. If we do it
1184 * earlier we end up with a wrong view of /proc/self/cgroup. For
1185 * example, assume we unshare(CLONE_NEWCGROUP) first, and then create
1186 * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then
1187 * /proc/self/cgroup would show us:
1191 * whereas it should actually show
1195 if (handler
->ns_unshare_flags
& CLONE_NEWCGROUP
) {
1196 ret
= unshare(CLONE_NEWCGROUP
);
1198 if (errno
!= EINVAL
) {
1199 SYSERROR("Failed to unshare CLONE_NEWCGROUP");
1200 goto out_warn_father
;
1203 handler
->ns_clone_flags
&= ~CLONE_NEWCGROUP
;
1204 SYSINFO("Kernel does not support CLONE_NEWCGROUP");
1206 INFO("Unshared CLONE_NEWCGROUP");
1210 if (handler
->ns_unshare_flags
& CLONE_NEWTIME
) {
1211 ret
= unshare(CLONE_NEWTIME
);
1213 if (errno
!= EINVAL
) {
1214 SYSERROR("Failed to unshare CLONE_NEWTIME");
1215 goto out_warn_father
;
1218 handler
->ns_clone_flags
&= ~CLONE_NEWTIME
;
1219 SYSINFO("Kernel does not support CLONE_NEWTIME");
1221 __do_close
int timens_fd
= -EBADF
;
1223 INFO("Unshared CLONE_NEWTIME");
1225 if (handler
->conf
->timens
.s_boot
)
1226 ret
= timens_offset_write(CLOCK_BOOTTIME
, handler
->conf
->timens
.s_boot
, 0);
1227 else if (handler
->conf
->timens
.ns_boot
)
1228 ret
= timens_offset_write(CLOCK_BOOTTIME
, 0, handler
->conf
->timens
.ns_boot
);
1230 SYSERROR("Failed to write CLONE_BOOTTIME offset");
1231 goto out_warn_father
;
1233 TRACE("Wrote CLOCK_BOOTTIME offset");
1235 if (handler
->conf
->timens
.s_monotonic
)
1236 ret
= timens_offset_write(CLOCK_MONOTONIC
, handler
->conf
->timens
.s_monotonic
, 0);
1237 else if (handler
->conf
->timens
.ns_monotonic
)
1238 ret
= timens_offset_write(CLOCK_MONOTONIC
, 0, handler
->conf
->timens
.ns_monotonic
);
1240 SYSERROR("Failed to write CLONE_MONOTONIC offset");
1241 goto out_warn_father
;
1243 TRACE("Wrote CLOCK_MONOTONIC offset");
1245 timens_fd
= open("/proc/self/ns/time_for_children", O_RDONLY
| O_CLOEXEC
);
1246 if (timens_fd
< 0) {
1247 SYSERROR("Failed to open \"/proc/self/ns/time_for_children\"");
1248 goto out_warn_father
;
1251 ret
= setns(timens_fd
, CLONE_NEWTIME
);
1253 SYSERROR("Failed to setns(%d(\"/proc/self/ns/time_for_children\"))", timens_fd
);
1254 goto out_warn_father
;
1260 * Add the requested environment variables to the current environment
1261 * to allow them to be used by the various hooks, such as the start
1264 ret
= lxc_set_environment(handler
->conf
);
1266 goto out_warn_father
;
1268 if (!lxc_sync_wait_parent(handler
, START_SYNC_POST_CONFIGURE
))
1269 goto out_warn_father
;
1271 /* Setup the container, ip, names, utsname, ... */
1272 ret
= lxc_setup(handler
);
1274 ERROR("Failed to setup container \"%s\"", handler
->name
);
1275 goto out_warn_father
;
1278 /* Set the label to change to when we exec(2) the container's init. */
1279 ret
= handler
->lsm_ops
->process_label_set(handler
->lsm_ops
, NULL
, handler
->conf
, true);
1281 goto out_warn_father
;
1283 /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it
1284 * before we aren't allowed anymore.
1286 if (handler
->conf
->no_new_privs
) {
1287 ret
= prctl(PR_SET_NO_NEW_PRIVS
, prctl_arg(1), prctl_arg(0),
1288 prctl_arg(0), prctl_arg(0));
1290 SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges");
1291 goto out_warn_father
;
1293 DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges");
1296 /* If we mounted a temporary proc, then unmount it now. */
1297 tmp_proc_unmount(handler
->conf
);
1299 ret
= lxc_seccomp_load(handler
->conf
);
1301 goto out_warn_father
;
1303 ret
= run_lxc_hooks(handler
->name
, "start", handler
->conf
, NULL
);
1305 ERROR("Failed to run lxc.hook.start for container \"%s\"",
1307 goto out_warn_father
;
1310 close_prot_errno_disarm(handler
->sigfd
);
1312 if (handler
->conf
->console
.pty
< 0 && handler
->daemonize
) {
1313 if (devnull_fd
< 0) {
1314 devnull_fd
= open_devnull();
1316 goto out_warn_father
;
1319 ret
= set_stdfds(devnull_fd
);
1321 ERROR("Failed to redirect std{in,out,err} to \"/dev/null\"");
1322 goto out_warn_father
;
1326 close_prot_errno_disarm(devnull_fd
);
1330 if (handler
->conf
->init_cwd
) {
1331 ret
= chdir(handler
->conf
->init_cwd
);
1333 SYSERROR("Could not change directory to \"%s\"",
1334 handler
->conf
->init_cwd
);
1335 goto out_warn_father
;
1339 if (!lxc_sync_barrier_parent(handler
, START_SYNC_CGROUP_LIMITS
))
1340 goto out_warn_father
;
1342 ret
= lxc_sync_fds_child(handler
);
1344 SYSERROR("Failed to sync file descriptors with parent");
1345 goto out_warn_father
;
1348 if (!lxc_sync_wait_parent(handler
, START_SYNC_READY_START
))
1349 goto out_warn_father
;
1351 /* Reset the environment variables the user requested in a clear
1355 /* Don't error out though. */
1357 SYSERROR("Failed to clear environment.");
1359 ret
= lxc_set_environment(handler
->conf
);
1361 goto out_warn_father
;
1363 ret
= putenv("container=lxc");
1365 SYSERROR("Failed to set environment variable: container=lxc");
1366 goto out_warn_father
;
1369 if (handler
->conf
->ttys
.tty_names
) {
1370 ret
= putenv(handler
->conf
->ttys
.tty_names
);
1372 SYSERROR("Failed to set environment variable for container ptys");
1373 goto out_warn_father
;
1377 /* The container has been setup. We can now switch to an unprivileged
1380 new_uid
= handler
->conf
->init_uid
;
1381 new_gid
= handler
->conf
->init_gid
;
1383 /* Avoid unnecessary syscalls. */
1384 if (new_uid
== nsuid
)
1385 new_uid
= LXC_INVALID_UID
;
1387 if (new_gid
== nsgid
)
1388 new_gid
= LXC_INVALID_GID
;
1390 /* Make sure that the processes STDIO is correctly owned by the user that we are switching to */
1391 ret
= fix_stdio_permissions(new_uid
);
1393 WARN("Failed to ajust stdio permissions");
1395 /* If we are in a new user namespace we already dropped all groups when
1396 * we switched to root in the new user namespace further above. Only
1397 * drop groups if we can, so ensure that we have necessary privilege.
1399 if (list_empty(&handler
->conf
->id_map
)) {
1401 if (lxc_proc_cap_is_set(CAP_SETGID
, CAP_EFFECTIVE
))
1404 if (handler
->conf
->init_groups
.size
> 0) {
1405 if (!lxc_setgroups(handler
->conf
->init_groups
.list
,
1406 handler
->conf
->init_groups
.size
))
1407 goto out_warn_father
;
1409 if (!lxc_drop_groups())
1410 goto out_warn_father
;
1415 if (!lxc_switch_uid_gid(new_uid
, new_gid
))
1416 goto out_warn_father
;
1418 ret
= lxc_ambient_caps_down();
1420 ERROR("Failed to clear ambient capabilities");
1421 goto out_warn_father
;
1424 if (handler
->conf
->monitor_signal_pdeath
!= SIGKILL
) {
1425 ret
= lxc_set_death_signal(handler
->conf
->monitor_signal_pdeath
,
1426 handler
->monitor_pid
, status_fd
);
1428 SYSERROR("Failed to set PR_SET_PDEATHSIG to %d",
1429 handler
->conf
->monitor_signal_pdeath
);
1430 goto out_warn_father
;
1435 * After this call, we are in error because this ops should not return
1438 handler
->ops
->start(handler
, handler
->data
);
1442 * We want the parent to know something went wrong, so we return a
1443 * special error code.
1445 lxc_sync_wake_parent(handler
, SYNC_ERROR
);
1451 int resolve_clone_flags(struct lxc_handler
*handler
)
1454 struct lxc_conf
*conf
= handler
->conf
;
1455 bool wants_timens
= conf
->timens
.s_boot
|| conf
->timens
.ns_boot
||
1456 conf
->timens
.s_monotonic
|| conf
->timens
.ns_monotonic
;
1458 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1459 if (conf
->ns_keep
) {
1460 if (!(conf
->ns_keep
& ns_info
[i
].clone_flag
))
1461 handler
->ns_clone_flags
|= ns_info
[i
].clone_flag
;
1462 } else if (conf
->ns_clone
) {
1463 if ((conf
->ns_clone
& ns_info
[i
].clone_flag
))
1464 handler
->ns_clone_flags
|= ns_info
[i
].clone_flag
;
1466 if (i
== LXC_NS_USER
&& list_empty(&handler
->conf
->id_map
))
1469 if (i
== LXC_NS_NET
&& lxc_requests_empty_network(handler
))
1472 if (i
== LXC_NS_CGROUP
&& !cgns_supported())
1475 if (i
== LXC_NS_TIME
&& !wants_timens
)
1478 handler
->ns_clone_flags
|= ns_info
[i
].clone_flag
;
1481 if (!conf
->ns_share
[i
])
1484 handler
->ns_clone_flags
&= ~ns_info
[i
].clone_flag
;
1485 TRACE("Sharing %s namespace", ns_info
[i
].proc_name
);
1488 if (wants_timens
&& (conf
->ns_keep
& ns_info
[LXC_NS_TIME
].clone_flag
))
1489 return log_trace_errno(-1, EINVAL
, "Requested to keep time namespace while also specifying offsets");
1491 /* Deal with namespaces that are unshared. */
1492 if (handler
->ns_clone_flags
& CLONE_NEWTIME
)
1493 handler
->ns_unshare_flags
|= CLONE_NEWTIME
;
1495 if (!pure_unified_layout(handler
->cgroup_ops
) && handler
->ns_clone_flags
& CLONE_NEWCGROUP
)
1496 handler
->ns_unshare_flags
|= CLONE_NEWCGROUP
;
1498 if ((handler
->ns_clone_flags
& (CLONE_NEWNET
| CLONE_NEWUSER
)) ==
1499 (CLONE_NEWNET
| CLONE_NEWUSER
))
1500 handler
->ns_unshare_flags
|= CLONE_NEWNET
;
1502 /* Deal with namespaces that are spawned. */
1503 handler
->ns_on_clone_flags
= handler
->ns_clone_flags
& ~handler
->ns_unshare_flags
;
1505 handler
->clone_flags
= handler
->ns_on_clone_flags
| CLONE_PIDFD
;
1510 /* Note that this function is used with clone(CLONE_VM). Some glibc versions
1511 * used to reset the pid/tid to -1 when CLONE_VM was used without CLONE_THREAD.
1512 * But since the memory between parent and child is shared on CLONE_VM this
1513 * would invalidate the getpid() cache that glibc used to maintain and so
1514 * getpid() in the child would return the parent's pid. This is all fixed in
1515 * newer glibc versions where the getpid() cache is removed and the pid/tid is
1516 * not reset anymore.
1517 * However, if for whatever reason you - dear committer - somehow need to get the
1518 * pid of the placeholder intermediate process for do_share_ns() you need to
1519 * call lxc_raw_getpid(). The next lxc_raw_clone() call does not employ
1520 * CLONE_VM and will be fine.
1522 static inline int do_share_ns(void *arg
)
1525 struct lxc_handler
*handler
= arg
;
1527 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1528 if (handler
->nsfd
[i
] < 0)
1531 ret
= setns(handler
->nsfd
[i
], 0);
1534 * Note that joining a user and/or mount namespace
1535 * requires the process is not multithreaded otherwise
1536 * setns() will fail here.
1538 SYSERROR("Failed to inherit %s namespace",
1539 ns_info
[i
].proc_name
);
1543 DEBUG("Inherited %s namespace", ns_info
[i
].proc_name
);
1546 flags
= handler
->ns_on_clone_flags
;
1547 flags
|= CLONE_PARENT
;
1548 handler
->pid
= lxc_raw_clone_cb(do_start
, handler
, CLONE_PIDFD
| flags
,
1550 if (handler
->pid
< 0)
1556 /* lxc_spawn() performs crucial setup tasks and clone()s the new process which
1557 * exec()s the requested container binary.
1558 * Note that lxc_spawn() runs in the parent namespaces. Any operations performed
1559 * right here should be double checked if they'd pose a security risk. (For
1560 * example, any {u}mount() operations performed here will be reflected on the
1563 static int lxc_spawn(struct lxc_handler
*handler
)
1565 __do_close
int data_sock0
= -EBADF
, data_sock1
= -EBADF
;
1568 bool wants_to_map_ids
;
1569 struct list_head
*id_map
;
1570 const char *name
= handler
->name
;
1571 const char *lxcpath
= handler
->lxcpath
;
1572 bool share_ns
= false;
1573 struct lxc_conf
*conf
= handler
->conf
;
1574 struct cgroup_ops
*cgroup_ops
= handler
->cgroup_ops
;
1576 id_map
= &conf
->id_map
;
1577 wants_to_map_ids
= !list_empty(id_map
);
1579 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1580 if (!conf
->ns_share
[i
])
1583 handler
->nsfd
[i
] = lxc_inherit_namespace(conf
->ns_share
[i
], lxcpath
, ns_info
[i
].proc_name
);
1584 if (handler
->nsfd
[i
] < 0)
1590 if (!lxc_sync_init(handler
))
1593 ret
= socketpair(AF_UNIX
, SOCK_STREAM
| SOCK_CLOEXEC
, 0,
1594 handler
->data_sock
);
1597 data_sock0
= handler
->data_sock
[0];
1598 data_sock1
= handler
->data_sock
[1];
1600 ret
= resolve_clone_flags(handler
);
1604 if (handler
->ns_clone_flags
& CLONE_NEWNET
) {
1605 ret
= lxc_find_gateway_addresses(handler
);
1607 ERROR("Failed to find gateway addresses");
1612 if (!cgroup_ops
->payload_create(cgroup_ops
, handler
)) {
1613 ERROR("Failed creating cgroups");
1614 goto out_delete_net
;
1617 /* Create a process in a new set of namespaces. */
1621 attacher_pid
= lxc_clone(do_share_ns
, handler
,
1622 CLONE_VFORK
| CLONE_VM
| CLONE_FILES
, NULL
);
1623 if (attacher_pid
< 0) {
1624 SYSERROR(LXC_CLONE_ERROR
);
1625 goto out_delete_net
;
1628 ret
= wait_for_pid(attacher_pid
);
1630 SYSERROR("Intermediate process failed");
1631 goto out_delete_net
;
1634 if (handler
->pid
< 0) {
1635 SYSERROR(LXC_CLONE_ERROR
);
1636 goto out_delete_net
;
1639 int cgroup_fd
= -EBADF
;
1641 struct lxc_clone_args clone_args
= {
1642 .flags
= handler
->clone_flags
,
1643 .pidfd
= ptr_to_u64(&handler
->pidfd
),
1644 .exit_signal
= SIGCHLD
,
1647 if (handler
->ns_clone_flags
& CLONE_NEWCGROUP
) {
1648 cgroup_fd
= cgroup_unified_fd(cgroup_ops
);
1649 if (cgroup_fd
>= 0) {
1650 handler
->clone_flags
|= CLONE_INTO_CGROUP
;
1651 clone_args
.flags
|= CLONE_INTO_CGROUP
;
1652 clone_args
.cgroup
= cgroup_fd
;
1656 /* Try to spawn directly into target cgroup. */
1657 handler
->pid
= lxc_clone3(&clone_args
, CLONE_ARGS_SIZE_VER2
);
1658 if (handler
->pid
< 0) {
1659 SYSTRACE("Failed to spawn container directly into target cgroup");
1661 /* Kernel might simply be too old for CLONE_INTO_CGROUP. */
1662 handler
->clone_flags
&= ~(CLONE_INTO_CGROUP
| CLONE_NEWCGROUP
);
1663 handler
->ns_on_clone_flags
&= ~CLONE_NEWCGROUP
;
1664 handler
->ns_unshare_flags
|= CLONE_NEWCGROUP
;
1666 clone_args
.flags
= handler
->clone_flags
;
1668 handler
->pid
= lxc_clone3(&clone_args
, CLONE_ARGS_SIZE_VER0
);
1669 } else if (cgroup_fd
>= 0) {
1670 TRACE("Spawned container directly into target cgroup via cgroup2 fd %d", cgroup_fd
);
1673 /* Kernel might be too old for clone3(). */
1674 if (handler
->pid
< 0) {
1675 SYSTRACE("Failed to spawn container via clone3()");
1678 * In contrast to all other architectures arm64 verifies that
1679 * the argument we use to retrieve the pidfd with is
1680 * initialized to 0. But we need to be able to initialize it to
1681 * a negative value such as our customary -EBADF so we can
1682 * detect whether this kernel supports pidfds. If the syscall
1683 * returns and the pidfd variable is set to something >= 0 then
1684 * we know this is a kernel supporting pidfds. But if we can't
1685 * set it to -EBADF then this won't work since 0 is a valid
1686 * file descriptor too. And since legacy clone silently ignores
1687 * unknown flags we are left without any way to detect support
1688 * for pidfds. So let's special-case arm64 to not fail starting
1691 #if defined(__aarch64__)
1692 handler
->pid
= lxc_raw_legacy_clone(handler
->clone_flags
& ~CLONE_PIDFD
, NULL
);
1694 handler
->pid
= lxc_raw_legacy_clone(handler
->clone_flags
, &handler
->pidfd
);
1698 if (handler
->pid
< 0) {
1699 SYSERROR(LXC_CLONE_ERROR
);
1700 goto out_delete_net
;
1703 if (handler
->pid
== 0) {
1704 (void)do_start(handler
);
1705 _exit(EXIT_FAILURE
);
1708 if (handler
->pidfd
< 0)
1709 handler
->clone_flags
&= ~CLONE_PIDFD
;
1710 TRACE("Cloned child process %d", handler
->pid
);
1712 /* Verify that we can actually make use of pidfds. */
1713 if (!lxc_can_use_pidfd(handler
->pidfd
))
1714 close_prot_errno_disarm(handler
->pidfd
);
1716 ret
= strnprintf(pidstr
, 20, "%d", handler
->pid
);
1718 goto out_delete_net
;
1720 ret
= setenv("LXC_PID", pidstr
, 1);
1722 SYSERROR("Failed to set environment variable: LXC_PID=%s", pidstr
);
1724 for (i
= 0; i
< LXC_NS_MAX
; i
++)
1725 if (handler
->ns_on_clone_flags
& ns_info
[i
].clone_flag
)
1726 INFO("Cloned %s", ns_info
[i
].flag_name
);
1728 if (!lxc_try_preserve_namespaces(handler
, handler
->ns_on_clone_flags
)) {
1729 ERROR("Failed to preserve cloned namespaces for lxc.hook.stop");
1730 goto out_delete_net
;
1733 lxc_sync_fini_child(handler
);
1735 if (lxc_abstract_unix_send_fds(handler
->data_sock
[0], &handler
->monitor_status_fd
, 1, NULL
, 0) < 0) {
1736 ERROR("Failed to send status file descriptor to child process");
1737 goto out_delete_net
;
1739 close_prot_errno_disarm(handler
->monitor_status_fd
);
1741 /* Map the container uids. The container became an invalid userid the
1742 * moment it was cloned with CLONE_NEWUSER. This call doesn't change
1743 * anything immediately, but allows the container to setuid(0) (0 being
1744 * mapped to something else on the host.) later to become a valid uid
1747 if (wants_to_map_ids
) {
1748 if (!handler
->conf
->ns_share
[LXC_NS_USER
] &&
1749 (handler
->conf
->ns_keep
& CLONE_NEWUSER
) == 0) {
1750 ret
= lxc_map_ids(id_map
, handler
->pid
);
1752 ERROR("Failed to set up id mapping.");
1753 goto out_delete_net
;
1758 if (!cgroup_ops
->setup_limits_legacy(cgroup_ops
, handler
->conf
, false)) {
1759 ERROR("Failed to setup cgroup limits for container \"%s\"", name
);
1760 goto out_delete_net
;
1763 if (!cgroup_ops
->payload_delegate_controllers(cgroup_ops
)) {
1764 ERROR("Failed to delegate controllers to payload cgroup");
1765 goto out_delete_net
;
1768 if (!cgroup_ops
->payload_enter(cgroup_ops
, handler
)) {
1769 ERROR("Failed to enter cgroups");
1770 goto out_delete_net
;
1773 if (!cgroup_ops
->setup_limits(cgroup_ops
, handler
)) {
1774 ERROR("Failed to setup cgroup limits for container \"%s\"", name
);
1775 goto out_delete_net
;
1778 if (!cgroup_ops
->chown(cgroup_ops
, handler
->conf
))
1779 goto out_delete_net
;
1781 if (!lxc_sync_barrier_child(handler
, START_SYNC_STARTUP
))
1782 goto out_delete_net
;
1784 /* If not done yet, we're now ready to preserve the network namespace */
1785 if (handler
->nsfd
[LXC_NS_NET
] < 0) {
1786 ret
= lxc_try_preserve_namespace(handler
, LXC_NS_NET
, "net");
1788 if (ret
!= -ENOENT
) {
1789 SYSERROR("Failed to preserve net namespace");
1790 goto out_delete_net
;
1794 ret
= lxc_netns_set_nsid(handler
->nsfd
[LXC_NS_NET
]);
1796 SYSWARN("Failed to allocate new network namespace id");
1798 TRACE("Allocated new network namespace id");
1800 /* Create the network configuration. */
1801 if (handler
->ns_clone_flags
& CLONE_NEWNET
) {
1802 ret
= lxc_create_network(handler
);
1804 ERROR("Failed to create the network");
1805 goto out_delete_net
;
1809 ret
= setup_proc_filesystem(conf
, handler
->pid
);
1811 ERROR("Failed to setup procfs limits");
1812 goto out_delete_net
;
1815 ret
= setup_resource_limits(conf
, handler
->pid
);
1817 ERROR("Failed to setup resource limits");
1818 goto out_delete_net
;
1821 /* Tell the child to continue its initialization. */
1822 if (!lxc_sync_wake_child(handler
, START_SYNC_POST_CONFIGURE
))
1823 goto out_delete_net
;
1825 ret
= lxc_rootfs_prepare_parent(handler
);
1827 ERROR("Failed to prepare rootfs");
1828 goto out_delete_net
;
1831 if (handler
->ns_clone_flags
& CLONE_NEWNET
) {
1832 ret
= lxc_network_send_to_child(handler
);
1834 SYSERROR("Failed to send veth names to child");
1835 goto out_delete_net
;
1839 if (!lxc_sync_wait_child(handler
, START_SYNC_IDMAPPED_MOUNTS
))
1840 goto out_delete_net
;
1842 ret
= lxc_idmapped_mounts_parent(handler
);
1844 ERROR("Failed to setup mount entries");
1845 goto out_delete_net
;
1848 if (!lxc_sync_wait_child(handler
, START_SYNC_CGROUP_LIMITS
))
1849 goto out_delete_net
;
1852 * With isolation the limiting devices cgroup was already setup, so
1853 * only setup devices here if we have no namespace directory.
1855 if (!handler
->conf
->cgroup_meta
.namespace_dir
&&
1856 !cgroup_ops
->setup_limits_legacy(cgroup_ops
, handler
->conf
, true)) {
1857 ERROR("Failed to setup legacy device cgroup controller limits");
1858 goto out_delete_net
;
1860 TRACE("Set up legacy device cgroup controller limits");
1862 if (!cgroup_ops
->devices_activate(cgroup_ops
, handler
)) {
1863 ERROR("Failed to setup cgroup2 device controller limits");
1864 goto out_delete_net
;
1866 TRACE("Set up cgroup2 device controller limits");
1868 cgroup_ops
->finalize(cgroup_ops
);
1869 TRACE("Finished setting up cgroups");
1871 /* Run any host-side start hooks */
1872 ret
= run_lxc_hooks(name
, "start-host", conf
, NULL
);
1874 ERROR("Failed to run lxc.hook.start-host");
1875 goto out_delete_net
;
1878 if (!lxc_sync_wake_child(handler
, START_SYNC_FDS
))
1879 goto out_delete_net
;
1881 if (handler
->ns_unshare_flags
& CLONE_NEWCGROUP
) {
1882 /* Now we're ready to preserve the cgroup namespace */
1883 ret
= lxc_try_preserve_namespace(handler
, LXC_NS_CGROUP
, "cgroup");
1885 if (ret
!= -ENOENT
) {
1886 SYSERROR("Failed to preserve cgroup namespace");
1887 goto out_delete_net
;
1892 if (handler
->ns_unshare_flags
& CLONE_NEWTIME
) {
1893 /* Now we're ready to preserve the time namespace */
1894 ret
= lxc_try_preserve_namespace(handler
, LXC_NS_TIME
, "time");
1896 if (ret
!= -ENOENT
) {
1897 SYSERROR("Failed to preserve time namespace");
1898 goto out_delete_net
;
1903 ret
= lxc_sync_fds_parent(handler
);
1905 SYSERROR("Failed to sync file descriptors with child");
1906 goto out_delete_net
;
1909 ret
= lxc_terminal_setup(conf
);
1911 SYSERROR("Failed to create console");
1912 goto out_delete_net
;
1916 * Tell the child to complete its initialization and wait for it to
1917 * exec or return an error. (The child will never return
1918 * START_SYNC_READY_START+1. It will either close the sync pipe,
1919 * causing lxc_sync_barrier_child to return success, or return a
1920 * different value, causing us to error out).
1922 if (!lxc_sync_barrier_child(handler
, START_SYNC_READY_START
))
1923 goto out_delete_net
;
1925 /* Now all networks are created, network devices are moved into place,
1926 * and the correct names and ifindices in the respective namespaces have
1927 * been recorded. The corresponding structs have now all been filled. So
1928 * log them for debugging purposes.
1930 lxc_log_configured_netdevs(conf
);
1932 ret
= handler
->ops
->post_start(handler
, handler
->data
);
1936 ret
= lxc_set_state(name
, handler
, RUNNING
);
1938 ERROR("Failed to set state to \"%s\"", lxc_state2str(RUNNING
));
1942 lxc_sync_fini(handler
);
1947 if (handler
->ns_clone_flags
& CLONE_NEWNET
)
1948 lxc_delete_network(handler
);
1954 lxc_sync_fini(handler
);
1959 int __lxc_start(struct lxc_handler
*handler
, struct lxc_operations
*ops
,
1960 void *data
, const char *lxcpath
, bool daemonize
, int *error_num
)
1963 const char *name
= handler
->name
;
1964 struct lxc_conf
*conf
= handler
->conf
;
1965 struct cgroup_ops
*cgroup_ops
;
1967 ret
= lxc_init(name
, handler
);
1969 ERROR("Failed to initialize container \"%s\"", name
);
1973 handler
->data
= data
;
1974 handler
->daemonize
= daemonize
;
1975 cgroup_ops
= handler
->cgroup_ops
;
1977 if (!attach_block_device(handler
->conf
)) {
1978 ERROR("Failed to attach block device");
1983 if (!cgroup_ops
->monitor_create(cgroup_ops
, handler
)) {
1984 ERROR("Failed to create monitor cgroup");
1989 if (!cgroup_ops
->monitor_delegate_controllers(cgroup_ops
)) {
1990 ERROR("Failed to delegate controllers to monitor cgroup");
1995 if (!cgroup_ops
->monitor_enter(cgroup_ops
, handler
)) {
1996 ERROR("Failed to enter monitor cgroup");
2001 /* If the rootfs is not a blockdev, prevent the container from marking
2003 * If the container is unprivileged then skip rootfs pinning.
2005 ret
= lxc_rootfs_init(conf
, !list_empty(&conf
->id_map
));
2007 ERROR("Failed to handle rootfs pinning for container \"%s\"", handler
->name
);
2012 if (geteuid() == 0 && !list_empty(&conf
->id_map
)) {
2014 * Most filesystems can't be mounted inside a userns so handle them here.
2016 if (rootfs_is_blockdev(conf
)) {
2017 ret
= unshare(CLONE_NEWNS
);
2019 ERROR("Failed to unshare CLONE_NEWNS");
2022 INFO("Unshared CLONE_NEWNS");
2024 ret
= lxc_setup_rootfs_prepare_root(conf
, name
, lxcpath
);
2026 ERROR("Error setting up rootfs mount as root before spawn");
2029 INFO("Set up container rootfs as host root");
2033 ret
= lxc_spawn(handler
);
2035 ERROR("Failed to spawn container \"%s\"", name
);
2036 goto out_detach_blockdev
;
2039 handler
->conf
->reboot
= REBOOT_NONE
;
2041 ret
= lxc_poll(name
, handler
);
2043 ERROR("LXC mainloop exited with error: %d", ret
);
2044 goto out_delete_network
;
2047 if (!handler
->init_died
&& handler
->pid
> 0) {
2048 ERROR("Child process is not killed");
2050 goto out_delete_network
;
2053 status
= lxc_wait_for_pid_status(handler
->pid
);
2055 SYSERROR("Failed to retrieve status for %d", handler
->pid
);
2057 /* If the child process exited but was not signaled, it didn't call
2058 * reboot. This should mean it was an lxc-execute which simply exited.
2059 * In any case, treat it as a 'halt'.
2061 if (WIFSIGNALED(status
)) {
2062 switch(WTERMSIG(status
)) {
2063 case SIGINT
: /* halt */
2064 DEBUG("Container \"%s\" is halting", name
);
2066 case SIGHUP
: /* reboot */
2067 DEBUG("Container \"%s\" is rebooting", name
);
2068 handler
->conf
->reboot
= REBOOT_REQ
;
2070 case SIGSYS
: /* seccomp */
2071 DEBUG("Container \"%s\" violated its seccomp policy", name
);
2074 DEBUG("Unknown exit status for container \"%s\" init %d", name
, WTERMSIG(status
));
2079 ret
= lxc_restore_phys_nics_to_netns(handler
);
2081 ERROR("Failed to move physical network devices back to parent network namespace");
2083 lxc_monitor_send_exit_code(name
, status
, handler
->lxcpath
);
2084 lxc_error_set_and_log(handler
->pid
, status
);
2086 *error_num
= handler
->exit_status
;
2088 lxc_delete_network(handler
);
2089 detach_block_device(handler
->conf
);
2098 out_detach_blockdev
:
2100 detach_block_device(handler
->conf
);
2106 lxc_restore_phys_nics_to_netns(handler
);
2107 lxc_delete_network(handler
);
2108 detach_block_device(handler
->conf
);
2117 static int start(struct lxc_handler
*handler
, void* data
)
2119 struct start_args
*arg
= data
;
2121 NOTICE("Exec'ing \"%s\"", arg
->argv
[0]);
2123 execvp(arg
->argv
[0], arg
->argv
);
2124 SYSERROR("Failed to exec \"%s\"", arg
->argv
[0]);
2128 static int post_start(struct lxc_handler
*handler
, void* data
)
2130 struct start_args
*arg
= data
;
2132 NOTICE("Started \"%s\" with pid \"%d\"", arg
->argv
[0], handler
->pid
);
2136 static struct lxc_operations start_ops
= {
2138 .post_start
= post_start
2141 int lxc_start(char *const argv
[], struct lxc_handler
*handler
,
2142 const char *lxcpath
, bool daemonize
, int *error_num
)
2144 struct start_args start_arg
= {
2148 TRACE("Doing lxc_start");
2149 return __lxc_start(handler
, &start_ops
, &start_arg
, lxcpath
, daemonize
, error_num
);
2152 static void lxc_destroy_container_on_signal(struct lxc_handler
*handler
,
2155 char destroy
[PATH_MAX
];
2156 struct lxc_container
*c
;
2160 if (handler
->conf
->rootfs
.path
&& handler
->conf
->rootfs
.mount
) {
2161 bret
= do_destroy_container(handler
);
2163 ERROR("Error destroying rootfs for container \"%s\"", name
);
2167 INFO("Destroyed rootfs for container \"%s\"", name
);
2169 ret
= strnprintf(destroy
, sizeof(destroy
), "%s/%s", handler
->lxcpath
, name
);
2171 ERROR("Error destroying directory for container \"%s\"", name
);
2175 c
= lxc_container_new(name
, handler
->lxcpath
);
2177 if (container_disk_lock(c
)) {
2178 INFO("Could not update lxc_snapshots file");
2179 lxc_container_put(c
);
2181 mod_all_rdeps(c
, false);
2182 container_disk_unlock(c
);
2183 lxc_container_put(c
);
2187 if (!handler
->am_root
)
2188 ret
= userns_exec_full(handler
->conf
, lxc_rmdir_onedev_wrapper
,
2189 destroy
, "lxc_rmdir_onedev_wrapper");
2191 ret
= lxc_rmdir_onedev(destroy
, NULL
);
2194 ERROR("Error destroying directory for container \"%s\"", name
);
2197 INFO("Destroyed directory for container \"%s\"", name
);
2200 static int lxc_rmdir_onedev_wrapper(void *data
)
2202 char *arg
= (char *) data
;
2203 return lxc_rmdir_onedev(arg
, NULL
);
2206 static bool do_destroy_container(struct lxc_handler
*handler
)
2210 if (!handler
->am_root
) {
2211 ret
= userns_exec_full(handler
->conf
, storage_destroy_wrapper
,
2212 handler
->conf
, "storage_destroy_wrapper");
2219 return storage_destroy(handler
->conf
);