1 /* SPDX-License-Identifier: LGPL-2.1+ */
16 #include <sys/mount.h>
17 #include <sys/param.h>
18 #include <sys/prctl.h>
19 #include <sys/socket.h>
21 #include <sys/syscall.h>
22 #include <sys/types.h>
30 #include "attach_options.h"
32 #include "cgroups/cgroup.h"
33 #include "cgroups/cgroup_utils.h"
35 #include "commands_utils.h"
38 #include "confile_utils.h"
40 #include "file_utils.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
++) {
155 const char *ns
= ns_info
[ns_idx
].proc_name
;
157 if ((ns_clone_flags
& ns_info
[ns_idx
].clone_flag
) == 0)
160 ret
= lxc_try_preserve_namespace(handler
, ns_idx
,
161 ns_info
[ns_idx
].proc_name
);
163 if (ret
== -ENOENT
) {
164 SYSERROR("Kernel does not support preserving %s namespaces", ns
);
169 * Handle kernels that do not support interacting with
170 * namespaces through procfs.
172 lxc_put_nsfds(handler
);
173 return log_error_errno(false, errno
, "Failed to preserve %s namespace", ns
);
180 static inline bool match_stdfds(int fd
)
182 return (fd
== STDIN_FILENO
|| fd
== STDOUT_FILENO
|| fd
== STDERR_FILENO
);
186 static bool match_dlog_fds(struct dirent
*direntp
)
188 char path
[PATH_MAX
] = {0};
189 char link
[PATH_MAX
] = {0};
193 ret
= strnprintf(path
, sizeof(path
), "/proc/self/fd/%s", direntp
->d_name
);
195 return log_error(false, "Failed to create file descriptor name");
197 linklen
= readlink(path
, link
, PATH_MAX
);
199 return log_error(false, "Failed to read link path - \"%s\"", path
);
200 else if (linklen
>= PATH_MAX
)
201 return log_error(false, "The name of link path is too long - \"%s\"", path
);
203 if (strequal(link
, "/dev/log_main") ||
204 strequal(link
, "/dev/log_system") ||
205 strequal(link
, "/dev/log_radio"))
212 /* Parses the LISTEN_FDS environment variable value.
213 * The returned value is the highest fd number up to which the
214 * file descriptors must be passed to the container process.
216 * For example, if LISTEN_FDS=2 then 4 is returned and file descriptors 3 and 4
217 * MUST be passed to the container process (in addition to the standard streams)
218 * to support [socket activation][systemd-listen-fds].
220 static unsigned int get_listen_fds_max(void)
223 unsigned int num_fds
;
226 val
= getenv("LISTEN_FDS");
230 ret
= lxc_safe_uint(val
, &num_fds
);
232 return syserror_ret(0, "Failed to parse \"LISTEN_FDS=%s\" environment variable", val
);
234 return log_trace(num_fds
, "Parsed \"LISTEN_FDS=%s\" environment variable", val
);
237 int lxc_check_inherited(struct lxc_conf
*conf
, bool closeall
,
238 int *fds_to_ignore
, size_t len_fds
)
243 struct dirent
*direntp
;
244 unsigned int listen_fds_max
;
245 struct lxc_state_client
*client
, *nclient
;
247 if (conf
&& conf
->close_all_fds
)
250 listen_fds_max
= get_listen_fds_max();
253 * Disable syslog at this point to avoid the above logging
254 * function to open a new fd and make the check_inherited function
255 * enter an infinite loop.
257 lxc_log_syslog_disable();
260 dir
= opendir("/proc/self/fd");
262 return log_warn(-1, "Failed to open directory");
266 while ((direntp
= readdir(dir
))) {
268 bool matched
= false;
270 if (strequal(direntp
->d_name
, "."))
273 if (strequal(direntp
->d_name
, ".."))
276 ret
= lxc_safe_int(direntp
->d_name
, &fd
);
278 INFO("Could not parse file descriptor for \"%s\"", direntp
->d_name
);
282 for (i
= 0; i
< len_fds
; i
++)
283 if (fds_to_ignore
[i
] == fd
)
286 if (fd
== fddir
|| fd
== lxc_log_fd
||
287 (i
< len_fds
&& fd
== fds_to_ignore
[i
]))
290 /* Keep state clients that wait on reboots. */
292 list_for_each_entry_safe(client
, nclient
, &conf
->state_clients
, head
) {
293 if (client
->clientfd
!= fd
)
304 if (current_config
&& fd
== current_config
->logfd
)
307 if (match_stdfds(fd
))
311 if (match_dlog_fds(direntp
))
316 if ((size_t)fd
<= listen_fds_max
) {
317 INFO("Inheriting fd %d (using the LISTEN_FDS environment variable)", fd
);
323 SYSINFO("Closed inherited fd %d", fd
);
325 INFO("Closed inherited fd %d", fd
);
329 WARN("Inherited fd %d", fd
);
334 * Only enable syslog at this point to avoid the above logging
335 * function to open a new fd and make the check_inherited function
336 * enter an infinite loop.
338 lxc_log_syslog_enable();
343 static int setup_signal_fd(sigset_t
*oldmask
)
347 const int signals
[] = {SIGBUS
, SIGILL
, SIGSEGV
, SIGWINCH
};
349 /* Block everything except serious error signals. */
350 ret
= sigfillset(&mask
);
354 for (size_t sig
= 0; sig
< (sizeof(signals
) / sizeof(signals
[0])); sig
++) {
355 ret
= sigdelset(&mask
, signals
[sig
]);
360 ret
= pthread_sigmask(SIG_BLOCK
, &mask
, oldmask
);
362 return log_error_errno(-EBADF
, errno
,
363 "Failed to set signal mask");
365 ret
= signalfd(-1, &mask
, SFD_CLOEXEC
);
367 return log_error_errno(-EBADF
,
368 errno
, "Failed to create signal file descriptor");
370 TRACE("Created signal file descriptor %d", ret
);
375 static int signal_handler(int fd
, uint32_t events
, void *data
,
376 struct lxc_async_descr
*descr
)
380 struct signalfd_siginfo siginfo
;
381 struct lxc_handler
*hdlr
= data
;
383 ret
= lxc_read_nointr(fd
, &siginfo
, sizeof(siginfo
));
385 return log_error(LXC_MAINLOOP_ERROR
, "Failed to read signal info from signal file descriptor %d", fd
);
387 if (ret
!= sizeof(siginfo
))
388 return log_error(LXC_MAINLOOP_ERROR
, "Unexpected size for struct signalfd_siginfo");
390 /* Check whether init is running. */
392 ret
= waitid(P_PID
, hdlr
->pid
, &info
, WEXITED
| WNOWAIT
| WNOHANG
);
393 if (ret
== 0 && info
.si_pid
== hdlr
->pid
)
394 hdlr
->init_died
= true;
396 TRACE("Received signal ssi_signo(%d) for ssi_pid(%d), si_signo(%d), si_pid(%d)",
397 siginfo
.ssi_signo
, siginfo
.ssi_pid
, info
.si_signo
, info
.si_pid
);
399 /* Try to figure out a reasonable exit status to report. */
400 if (hdlr
->init_died
) {
401 switch (info
.si_code
) {
403 hdlr
->exit_status
= info
.si_status
<< 8;
408 hdlr
->exit_status
= info
.si_status
<< 8 | 0x7f;
411 /* Huh? The waitid() told us it's dead *and* continued? */
412 WARN("Init %d dead and continued?", hdlr
->pid
);
413 hdlr
->exit_status
= 1;
416 ERROR("Unknown si_code: %d", info
.si_code
);
417 hdlr
->exit_status
= 1;
421 if (siginfo
.ssi_signo
== SIGHUP
) {
422 if (hdlr
->pidfd
>= 0)
423 lxc_raw_pidfd_send_signal(hdlr
->pidfd
, SIGTERM
, NULL
, 0);
425 kill(hdlr
->pid
, SIGTERM
);
426 INFO("Killing %d since terminal hung up", hdlr
->pid
);
427 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
428 : LXC_MAINLOOP_CONTINUE
;
431 if (siginfo
.ssi_signo
!= SIGCHLD
) {
432 if (hdlr
->pidfd
>= 0)
433 lxc_raw_pidfd_send_signal(hdlr
->pidfd
,
434 siginfo
.ssi_signo
, NULL
, 0);
436 kill(hdlr
->pid
, siginfo
.ssi_signo
);
437 INFO("Forwarded signal %d to pid %d", siginfo
.ssi_signo
, hdlr
->pid
);
438 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
439 : LXC_MAINLOOP_CONTINUE
;
442 /* More robustness, protect ourself from a SIGCHLD sent
443 * by a process different from the container init.
445 if ((__u64
)siginfo
.ssi_pid
!= (__u64
)hdlr
->pid
) {
446 NOTICE("Received %d from pid %d instead of container init %d",
447 siginfo
.ssi_signo
, siginfo
.ssi_pid
, hdlr
->pid
);
448 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
449 : LXC_MAINLOOP_CONTINUE
;
452 if (siginfo
.ssi_code
== CLD_STOPPED
) {
453 INFO("Container init process was stopped");
454 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
455 : LXC_MAINLOOP_CONTINUE
;
458 if (siginfo
.ssi_code
== CLD_CONTINUED
) {
459 INFO("Container init process was continued");
460 return hdlr
->init_died
? LXC_MAINLOOP_CLOSE
461 : LXC_MAINLOOP_CONTINUE
;
464 return log_debug(LXC_MAINLOOP_CLOSE
, "Container init process %d exited", hdlr
->pid
);
467 int lxc_serve_state_clients(const char *name
, struct lxc_handler
*handler
,
470 struct lxc_msg msg
= {
471 .type
= lxc_msg_state
,
476 struct lxc_state_client
*client
, *nclient
;
479 handler
->state
= RUNNING
;
481 handler
->state
= state
;
483 TRACE("Set container state to %s", lxc_state2str(state
));
485 if (list_empty(&handler
->conf
->state_clients
))
486 return log_trace(0, "No state clients registered");
488 retlen
= strlcpy(msg
.name
, name
, sizeof(msg
.name
));
489 if (retlen
>= sizeof(msg
.name
))
492 list_for_each_entry_safe(client
, nclient
, &handler
->conf
->state_clients
, head
) {
493 if (client
->states
[state
] == 0) {
494 TRACE("State %s not registered for state client %d",
495 lxc_state2str(state
), client
->clientfd
);
499 TRACE("Sending state %s to state client %d",
500 lxc_state2str(state
), client
->clientfd
);
502 ret
= lxc_send_nointr(client
->clientfd
, &msg
, sizeof(msg
), MSG_NOSIGNAL
);
504 SYSERROR("Failed to send message to client");
506 /* kick client from list */
507 list_del(&client
->head
);
508 close(client
->clientfd
);
515 static int lxc_serve_state_socket_pair(const char *name
,
516 struct lxc_handler
*handler
,
521 if (!handler
->daemonize
||
522 handler
->state_socket_pair
[1] < 0 ||
526 /* Close read end of the socket pair. */
527 close_prot_errno_disarm(handler
->state_socket_pair
[0]);
530 ret
= lxc_abstract_unix_send_credential(handler
->state_socket_pair
[1],
531 &(int){state
}, sizeof(int));
533 SYSERROR("Failed to send state to %d", handler
->state_socket_pair
[1]);
541 if (ret
!= sizeof(int))
542 return log_error(-1, "Message too long : %d", handler
->state_socket_pair
[1]);
544 TRACE("Sent container state \"%s\" to %d", lxc_state2str(state
),
545 handler
->state_socket_pair
[1]);
547 /* Close write end of the socket pair. */
548 close_prot_errno_disarm(handler
->state_socket_pair
[1]);
553 int lxc_set_state(const char *name
, struct lxc_handler
*handler
,
558 ret
= lxc_serve_state_socket_pair(name
, handler
, state
);
560 return log_error(-1, "Failed to synchronize via anonymous pair of unix sockets");
562 ret
= lxc_serve_state_clients(name
, handler
, state
);
566 /* This function will try to connect to the legacy lxc-monitord state
567 * server and only exists for backwards compatibility.
569 lxc_monitor_send_state(name
, state
, handler
->lxcpath
);
574 int lxc_poll(const char *name
, struct lxc_handler
*handler
)
577 struct lxc_terminal
*console
= &handler
->conf
->console
;
578 struct lxc_async_descr descr
, descr_console
;
580 if (!wants_console(console
))
583 ret
= lxc_mainloop_open(&descr
);
585 ERROR("Failed to create mainloop");
590 ret
= lxc_mainloop_open(&descr_console
);
592 ERROR("Failed to create console mainloop");
597 ret
= lxc_mainloop_add_handler(&descr
, handler
->sigfd
,
599 default_cleanup_handler
,
600 handler
, "signal_handler");
602 ERROR("Failed to add signal handler for %d to mainloop", handler
->sigfd
);
603 goto out_mainloop_console
;
606 ret
= lxc_seccomp_setup_proxy(&handler
->conf
->seccomp
, &descr
, handler
);
608 ERROR("Failed to setup seccomp proxy");
609 goto out_mainloop_console
;
613 ret
= lxc_terminal_mainloop_add(&descr
, console
);
615 ERROR("Failed to add console handlers to mainloop");
616 goto out_mainloop_console
;
620 ret
= lxc_cmd_mainloop_add(name
, &descr
, handler
);
622 ERROR("Failed to add command handler to mainloop");
623 goto out_mainloop_console
;
626 TRACE("Mainloop is ready");
628 ret
= lxc_mainloop(&descr
, -1);
629 if (descr
.type
== LXC_MAINLOOP_EPOLL
)
630 close_prot_errno_disarm(descr
.epfd
);
631 if (ret
< 0 || !handler
->init_died
)
632 goto out_mainloop_console
;
635 ret
= lxc_terminal_mainloop_add(&descr_console
, console
);
637 ret
= lxc_mainloop(&descr_console
, 0);
640 out_mainloop_console
:
642 lxc_mainloop_close(&descr_console
);
643 TRACE("Closed console mainloop");
647 lxc_mainloop_close(&descr
);
648 TRACE("Closed mainloop");
651 TRACE("Closed signal file descriptor %d", handler
->sigfd
);
652 close_prot_errno_disarm(handler
->sigfd
);
657 void lxc_put_handler(struct lxc_handler
*handler
)
659 close_prot_errno_disarm(handler
->pidfd
);
660 close_prot_errno_disarm(handler
->sigfd
);
661 lxc_put_nsfds(handler
);
662 if (handler
->conf
&& handler
->conf
->reboot
== REBOOT_NONE
)
663 close_prot_errno_disarm(handler
->conf
->maincmd_fd
);
664 close_prot_errno_disarm(handler
->monitor_status_fd
);
665 close_prot_errno_disarm(handler
->state_socket_pair
[0]);
666 close_prot_errno_disarm(handler
->state_socket_pair
[1]);
667 cgroup_exit(handler
->cgroup_ops
);
668 if (handler
->conf
&& handler
->conf
->reboot
== REBOOT_NONE
)
669 free_disarm(handler
);
671 handler
->conf
= NULL
;
674 struct lxc_handler
*lxc_init_handler(struct lxc_handler
*old
,
675 const char *name
, struct lxc_conf
*conf
,
676 const char *lxcpath
, bool daemonize
)
680 struct lxc_handler
*handler
;
683 handler
= zalloc(sizeof(*handler
));
689 /* Note that am_guest_unpriv() checks the effective uid. We
690 * probably don't care if we are real root only if we are running
691 * as root so this should be fine.
693 handler
->am_root
= !am_guest_unpriv();
694 handler
->conf
= conf
;
695 handler
->lxcpath
= lxcpath
;
696 handler
->init_died
= false;
697 handler
->data_sock
[0] = -EBADF
;
698 handler
->data_sock
[1] = -EBADF
;
699 handler
->monitor_status_fd
= -EBADF
;
700 handler
->pidfd
= -EBADF
;
701 handler
->sigfd
= -EBADF
;
702 handler
->state_socket_pair
[0] = -EBADF
;
703 handler
->state_socket_pair
[1] = -EBADF
;
704 if (handler
->conf
->reboot
== REBOOT_NONE
)
705 INIT_LIST_HEAD(&handler
->conf
->state_clients
);
707 for (lxc_namespace_t idx
= 0; idx
< LXC_NS_MAX
; idx
++) {
708 handler
->nsfd
[idx
] = -EBADF
;
710 if (handler
->conf
->reboot
== REBOOT_NONE
)
713 handler
->nsfd_paths
[idx
][0] = '\0';
714 handler
->hook_argv
[idx
] = NULL
;
716 if (handler
->hook_argc
!= 0)
717 handler
->hook_argc
= 0;
720 handler
->name
= name
;
722 handler
->transient_pid
= lxc_raw_getpid();
724 handler
->transient_pid
= -1;
726 if (daemonize
&& handler
->conf
->reboot
== REBOOT_NONE
) {
727 /* Create socketpair() to synchronize on daemonized startup.
728 * When the container reboots we don't need to synchronize
729 * again currently so don't open another socketpair().
731 ret
= socketpair(AF_UNIX
, SOCK_STREAM
| SOCK_CLOEXEC
, 0,
732 handler
->state_socket_pair
);
734 ERROR("Failed to create anonymous pair of unix sockets");
738 TRACE("Created anonymous pair {%d,%d} of unix sockets",
739 handler
->state_socket_pair
[0],
740 handler
->state_socket_pair
[1]);
741 handler
->keep_fds
[nr_keep_fds
++] = handler
->state_socket_pair
[0];
742 handler
->keep_fds
[nr_keep_fds
++] = handler
->state_socket_pair
[1];
745 if (handler
->conf
->reboot
== REBOOT_NONE
) {
746 handler
->conf
->maincmd_fd
= lxc_server_init(name
, lxcpath
, "command");
747 if (handler
->conf
->maincmd_fd
< 0) {
748 ERROR("Failed to set up command socket");
751 handler
->keep_fds
[nr_keep_fds
++] = handler
->conf
->maincmd_fd
;
754 TRACE("Unix domain socket %d for command server is ready",
755 handler
->conf
->maincmd_fd
);
760 lxc_put_handler(handler
);
765 int lxc_init(const char *name
, struct lxc_handler
*handler
)
767 __do_close
int status_fd
= -EBADF
;
769 const char *loglevel
;
770 struct lxc_conf
*conf
= handler
->conf
;
772 handler
->monitor_pid
= lxc_raw_getpid();
773 status_fd
= open("/proc/self/status", O_RDONLY
| O_CLOEXEC
);
775 return log_error_errno(-1, errno
, "Failed to open monitor status fd");
777 handler
->lsm_ops
= lsm_init_static();
778 TRACE("Initialized LSM");
780 /* Begin by setting the state to STARTING. */
781 ret
= lxc_set_state(name
, handler
, STARTING
);
783 return log_error(-1, "Failed to set state to \"%s\"", lxc_state2str(STARTING
));
784 TRACE("Set container state to \"STARTING\"");
786 /* Start of environment variable setup for hooks. */
787 ret
= setenv("LXC_NAME", name
, 1);
789 SYSERROR("Failed to set environment variable: LXC_NAME=%s", name
);
792 ret
= setenv("LXC_CONFIG_FILE", conf
->rcfile
, 1);
794 SYSERROR("Failed to set environment variable: LXC_CONFIG_FILE=%s", conf
->rcfile
);
797 if (conf
->rootfs
.mount
) {
798 ret
= setenv("LXC_ROOTFS_MOUNT", conf
->rootfs
.mount
, 1);
800 SYSERROR("Failed to set environment variable: LXC_ROOTFS_MOUNT=%s", conf
->rootfs
.mount
);
803 if (conf
->rootfs
.path
) {
804 ret
= setenv("LXC_ROOTFS_PATH", conf
->rootfs
.path
, 1);
806 SYSERROR("Failed to set environment variable: LXC_ROOTFS_PATH=%s", conf
->rootfs
.path
);
809 if (conf
->console
.path
) {
810 ret
= setenv("LXC_CONSOLE", conf
->console
.path
, 1);
812 SYSERROR("Failed to set environment variable: LXC_CONSOLE=%s", conf
->console
.path
);
815 if (conf
->console
.log_path
) {
816 ret
= setenv("LXC_CONSOLE_LOGPATH", conf
->console
.log_path
, 1);
818 SYSERROR("Failed to set environment variable: LXC_CONSOLE_LOGPATH=%s", conf
->console
.log_path
);
821 if (cgns_supported()) {
822 ret
= setenv("LXC_CGNS_AWARE", "1", 1);
824 SYSERROR("Failed to set environment variable LXC_CGNS_AWARE=1");
827 loglevel
= lxc_log_priority_to_string(lxc_log_get_level());
828 ret
= setenv("LXC_LOG_LEVEL", loglevel
, 1);
830 SYSERROR("Set environment variable LXC_LOG_LEVEL=%s", loglevel
);
832 if (conf
->hooks_version
== 0)
833 ret
= setenv("LXC_HOOK_VERSION", "0", 1);
835 ret
= setenv("LXC_HOOK_VERSION", "1", 1);
837 SYSERROR("Failed to set environment variable LXC_HOOK_VERSION=%u", conf
->hooks_version
);
838 /* End of environment variable setup for hooks. */
840 TRACE("Set environment variables");
842 ret
= run_lxc_hooks(name
, "pre-start", conf
, NULL
);
844 return log_error(-1, "Failed to run lxc.hook.pre-start for container \"%s\"", name
);
845 TRACE("Ran pre-start hooks");
847 ret
= lxc_terminal_parent(conf
);
849 return log_error(-1, "Failed to allocate terminal");
851 /* The signal fd has to be created before forking otherwise if the child
852 * process exits before we setup the signal fd, the event will be lost
853 * and the command will be stuck.
855 handler
->sigfd
= setup_signal_fd(&handler
->oldmask
);
856 if (handler
->sigfd
< 0)
857 return log_error(-1, "Failed to setup SIGCHLD fd handler.");
858 TRACE("Set up signal fd");
860 handler
->cgroup_ops
= cgroup_init(handler
->conf
);
861 if (!handler
->cgroup_ops
) {
862 ERROR("Failed to initialize cgroup driver");
863 goto out_restore_sigmask
;
865 TRACE("Initialized cgroup driver");
867 ret
= lxc_read_seccomp_config(conf
);
869 ERROR("Failed to read seccomp policy");
870 goto out_restore_sigmask
;
872 TRACE("Read seccomp policy");
874 ret
= handler
->lsm_ops
->prepare(handler
->lsm_ops
, conf
, handler
->lxcpath
);
876 ERROR("Failed to initialize LSM");
877 goto out_restore_sigmask
;
879 TRACE("Initialized LSM");
881 INFO("Container \"%s\" is initialized", name
);
882 handler
->monitor_status_fd
= move_fd(status_fd
);
886 (void)pthread_sigmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
891 void lxc_expose_namespace_environment(const struct lxc_handler
*handler
)
893 for (lxc_namespace_t i
= 0; i
< LXC_NS_MAX
; i
++) {
897 if (handler
->nsfd
[i
] < 0)
900 fd_path
= handler
->nsfd_paths
[i
] + strcspn(handler
->nsfd_paths
[i
], "/");
901 ret
= setenv(ns_info
[i
].env_name
, fd_path
, 1);
903 SYSERROR("Failed to set environment variable %s=%s",
904 ns_info
[i
].env_name
, fd_path
);
906 TRACE("Set environment variable %s=%s",
907 ns_info
[i
].env_name
, fd_path
);
911 void lxc_end(struct lxc_handler
*handler
)
914 const char *name
= handler
->name
;
915 struct cgroup_ops
*cgroup_ops
= handler
->cgroup_ops
;
916 struct lxc_state_client
*client
, *nclient
;
918 /* The STOPPING state is there for future cleanup code which can take
921 lxc_set_state(name
, handler
, STOPPING
);
923 /* Passing information to hooks via environment variables. */
924 if (handler
->conf
->hooks_version
> 0)
925 lxc_expose_namespace_environment(handler
);
927 if (handler
->conf
->reboot
> REBOOT_NONE
) {
928 ret
= setenv("LXC_TARGET", "reboot", 1);
930 SYSERROR("Failed to set environment variable: LXC_TARGET=reboot");
933 if (handler
->conf
->reboot
== REBOOT_NONE
) {
934 ret
= setenv("LXC_TARGET", "stop", 1);
936 SYSERROR("Failed to set environment variable: LXC_TARGET=stop");
939 if (handler
->conf
->hooks_version
== 0)
940 ret
= run_lxc_hooks(name
, "stop", handler
->conf
, handler
->hook_argv
);
942 ret
= run_lxc_hooks(name
, "stop", handler
->conf
, NULL
);
944 ERROR("Failed to run \"lxc.hook.stop\" hook");
946 handler
->lsm_ops
->cleanup(handler
->lsm_ops
, handler
->conf
, handler
->lxcpath
);
949 cgroup_ops
->payload_destroy(cgroup_ops
, handler
);
950 cgroup_ops
->monitor_destroy(cgroup_ops
, handler
);
953 put_lxc_rootfs(&handler
->conf
->rootfs
, true);
955 if (handler
->conf
->reboot
== REBOOT_NONE
) {
956 /* For all new state clients simply close the command socket.
957 * This will inform all state clients that the container is
958 * STOPPED and also prevents a race between a open()/close() on
959 * the command socket causing a new process to get ECONNREFUSED
960 * because we haven't yet closed the command socket.
962 close_prot_errno_disarm(handler
->conf
->maincmd_fd
);
963 TRACE("Closed command socket");
965 /* This function will try to connect to the legacy lxc-monitord
966 * state server and only exists for backwards compatibility.
968 lxc_monitor_send_state(name
, STOPPED
, handler
->lxcpath
);
970 /* The command socket is closed so no one can acces the command
971 * socket anymore so there's no need to lock it.
973 handler
->state
= STOPPED
;
974 TRACE("Set container state to \"STOPPED\"");
976 lxc_set_state(name
, handler
, STOPPED
);
977 TRACE("Set container state to \"STOPPED\"");
980 /* Avoid lingering namespace references. */
981 lxc_put_nsfds(handler
);
983 ret
= run_lxc_hooks(name
, "post-stop", handler
->conf
, NULL
);
985 ERROR("Failed to run lxc.hook.post-stop for container \"%s\"", name
);
986 if (handler
->conf
->reboot
> REBOOT_NONE
) {
987 WARN("Container will be stopped instead of rebooted");
988 handler
->conf
->reboot
= REBOOT_NONE
;
990 ret
= setenv("LXC_TARGET", "stop", 1);
992 WARN("Failed to set environment variable: LXC_TARGET=stop");
996 /* Reset mask set by setup_signal_fd. */
997 ret
= pthread_sigmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
999 SYSWARN("Failed to restore signal mask");
1001 lxc_terminal_delete(&handler
->conf
->console
);
1002 lxc_delete_tty(&handler
->conf
->ttys
);
1003 close_prot_errno_disarm(handler
->conf
->devpts_fd
);
1005 /* The command socket is now closed, no more state clients can register
1006 * themselves from now on. So free the list of state clients.
1008 list_for_each_entry_safe(client
, nclient
, &handler
->conf
->state_clients
, head
) {
1009 /* Keep state clients that want to be notified about reboots. */
1010 if ((handler
->conf
->reboot
> REBOOT_NONE
) &&
1011 (client
->states
[RUNNING
] == 2))
1014 /* close state client socket */
1015 list_del(&client
->head
);
1016 close(client
->clientfd
);
1020 if (handler
->conf
->ephemeral
== 1 && handler
->conf
->reboot
!= REBOOT_REQ
)
1021 lxc_destroy_container_on_signal(handler
, name
);
1023 lxc_put_handler(handler
);
1026 void lxc_abort(struct lxc_handler
*handler
)
1031 lxc_set_state(handler
->name
, handler
, ABORTING
);
1033 if (handler
->pidfd
>= 0) {
1034 ret
= lxc_raw_pidfd_send_signal(handler
->pidfd
, SIGKILL
, NULL
, 0);
1036 SYSWARN("Failed to send SIGKILL via pidfd %d for process %d",
1037 handler
->pidfd
, handler
->pid
);
1040 if ((!ret
|| errno
!= ESRCH
) && handler
->pid
> 0)
1041 if (kill(handler
->pid
, SIGKILL
))
1042 SYSWARN("Failed to send SIGKILL to %d", handler
->pid
);
1045 ret
= waitpid(-1, &status
, 0);
1049 static int do_start(void *data
)
1051 struct lxc_handler
*handler
= data
;
1052 __lxc_unused __do_close
int data_sock0
= handler
->data_sock
[0],
1053 data_sock1
= handler
->data_sock
[1];
1054 __do_close
int devnull_fd
= -EBADF
, status_fd
= -EBADF
;
1061 lxc_sync_fini_parent(handler
);
1063 if (lxc_abstract_unix_recv_one_fd(data_sock1
, &status_fd
, NULL
, 0) < 0) {
1064 ERROR("Failed to receive status file descriptor from parent process");
1065 goto out_warn_father
;
1068 /* This prctl must be before the synchro, so if the parent dies before
1069 * we set the parent death signal, we will detect its death with the
1070 * synchro right after, otherwise we have a window where the parent can
1071 * exit before we set the pdeath signal leading to a unsupervized
1074 ret
= lxc_set_death_signal(SIGKILL
, handler
->monitor_pid
, status_fd
);
1076 SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL");
1077 goto out_warn_father
;
1080 ret
= lxc_ambient_caps_up();
1082 ERROR("Failed to raise ambient capabilities");
1083 goto out_warn_father
;
1086 ret
= pthread_sigmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
1088 SYSERROR("Failed to set signal mask");
1089 goto out_warn_father
;
1092 if (!lxc_sync_wait_parent(handler
, START_SYNC_STARTUP
))
1093 goto out_warn_father
;
1095 /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See
1096 * https://github.com/lxc/lxd/issues/1978.
1098 if (handler
->ns_unshare_flags
& CLONE_NEWNET
) {
1099 ret
= unshare(CLONE_NEWNET
);
1101 SYSERROR("Failed to unshare CLONE_NEWNET");
1102 goto out_warn_father
;
1104 INFO("Unshared CLONE_NEWNET");
1107 /* If we are in a new user namespace, become root there to have
1108 * privilege over our namespace.
1110 if (!list_empty(&handler
->conf
->id_map
)) {
1111 if (!handler
->conf
->root_nsuid_map
)
1112 nsuid
= handler
->conf
->init_uid
;
1114 if (!handler
->conf
->root_nsgid_map
)
1115 nsgid
= handler
->conf
->init_gid
;
1117 /* Drop groups only after we switched to a valid gid in the new
1120 if (!lxc_drop_groups() &&
1121 (handler
->am_root
|| errno
!= EPERM
))
1122 goto out_warn_father
;
1124 if (!lxc_switch_uid_gid(nsuid
, nsgid
))
1125 goto out_warn_father
;
1127 ret
= prctl(PR_SET_DUMPABLE
, prctl_arg(1), prctl_arg(0),
1128 prctl_arg(0), prctl_arg(0));
1130 goto out_warn_father
;
1132 /* set{g,u}id() clears deathsignal */
1133 ret
= lxc_set_death_signal(SIGKILL
, handler
->monitor_pid
, status_fd
);
1135 SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL");
1136 goto out_warn_father
;
1140 ret
= access(handler
->lxcpath
, X_OK
);
1142 print_top_failing_dir(handler
->lxcpath
);
1143 goto out_warn_father
;
1146 /* In order to checkpoint restore, we need to have everything in the
1147 * same mount namespace. However, some containers may not have a
1148 * reasonable /dev (in particular, they may not have /dev/null), so we
1149 * can't set init's std fds to /dev/null by opening it from inside the
1152 * If that's the case, fall back to using the host's /dev/null. This
1153 * means that migration won't work, but at least we won't spew output
1154 * where it isn't wanted.
1156 if (handler
->daemonize
&& !handler
->conf
->autodev
) {
1157 char path
[PATH_MAX
];
1159 ret
= strnprintf(path
, sizeof(path
), "%s/dev/null",
1160 handler
->conf
->rootfs
.mount
);
1162 goto out_warn_father
;
1164 ret
= access(path
, F_OK
);
1166 devnull_fd
= open_devnull();
1169 goto out_warn_father
;
1170 WARN("Using /dev/null from the host for container init's standard file descriptors. Migration will not work");
1175 * Tell the parent task it can begin to configure the container and wait
1178 if (!lxc_sync_wake_parent(handler
, START_SYNC_CONFIGURE
))
1181 /* Unshare cgroup namespace after we have setup our cgroups. If we do it
1182 * earlier we end up with a wrong view of /proc/self/cgroup. For
1183 * example, assume we unshare(CLONE_NEWCGROUP) first, and then create
1184 * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then
1185 * /proc/self/cgroup would show us:
1189 * whereas it should actually show
1193 if (handler
->ns_unshare_flags
& CLONE_NEWCGROUP
) {
1194 ret
= unshare(CLONE_NEWCGROUP
);
1196 if (errno
!= EINVAL
) {
1197 SYSERROR("Failed to unshare CLONE_NEWCGROUP");
1198 goto out_warn_father
;
1201 handler
->ns_clone_flags
&= ~CLONE_NEWCGROUP
;
1202 SYSINFO("Kernel does not support CLONE_NEWCGROUP");
1204 INFO("Unshared CLONE_NEWCGROUP");
1208 if (handler
->ns_unshare_flags
& CLONE_NEWTIME
) {
1209 ret
= unshare(CLONE_NEWTIME
);
1211 if (errno
!= EINVAL
) {
1212 SYSERROR("Failed to unshare CLONE_NEWTIME");
1213 goto out_warn_father
;
1216 handler
->ns_clone_flags
&= ~CLONE_NEWTIME
;
1217 SYSINFO("Kernel does not support CLONE_NEWTIME");
1219 __do_close
int timens_fd
= -EBADF
;
1221 INFO("Unshared CLONE_NEWTIME");
1223 if (handler
->conf
->timens
.s_boot
)
1224 ret
= timens_offset_write(CLOCK_BOOTTIME
, handler
->conf
->timens
.s_boot
, 0);
1225 else if (handler
->conf
->timens
.ns_boot
)
1226 ret
= timens_offset_write(CLOCK_BOOTTIME
, 0, handler
->conf
->timens
.ns_boot
);
1228 SYSERROR("Failed to write CLONE_BOOTTIME offset");
1229 goto out_warn_father
;
1231 TRACE("Wrote CLOCK_BOOTTIME offset");
1233 if (handler
->conf
->timens
.s_monotonic
)
1234 ret
= timens_offset_write(CLOCK_MONOTONIC
, handler
->conf
->timens
.s_monotonic
, 0);
1235 else if (handler
->conf
->timens
.ns_monotonic
)
1236 ret
= timens_offset_write(CLOCK_MONOTONIC
, 0, handler
->conf
->timens
.ns_monotonic
);
1238 SYSERROR("Failed to write CLONE_MONOTONIC offset");
1239 goto out_warn_father
;
1241 TRACE("Wrote CLOCK_MONOTONIC offset");
1243 timens_fd
= open("/proc/self/ns/time_for_children", O_RDONLY
| O_CLOEXEC
);
1244 if (timens_fd
< 0) {
1245 SYSERROR("Failed to open \"/proc/self/ns/time_for_children\"");
1246 goto out_warn_father
;
1249 ret
= setns(timens_fd
, CLONE_NEWTIME
);
1251 SYSERROR("Failed to setns(%d(\"/proc/self/ns/time_for_children\"))", timens_fd
);
1252 goto out_warn_father
;
1258 * Add the requested environment variables to the current environment
1259 * to allow them to be used by the various hooks, such as the start
1262 ret
= lxc_set_environment(handler
->conf
);
1264 goto out_warn_father
;
1266 if (!lxc_sync_wait_parent(handler
, START_SYNC_POST_CONFIGURE
))
1267 goto out_warn_father
;
1269 /* Setup the container, ip, names, utsname, ... */
1270 ret
= lxc_setup(handler
);
1272 ERROR("Failed to setup container \"%s\"", handler
->name
);
1273 goto out_warn_father
;
1276 /* Set the label to change to when we exec(2) the container's init. */
1277 ret
= handler
->lsm_ops
->process_label_set(handler
->lsm_ops
, NULL
, handler
->conf
, true);
1279 goto out_warn_father
;
1281 /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it
1282 * before we aren't allowed anymore.
1284 if (handler
->conf
->no_new_privs
) {
1285 ret
= prctl(PR_SET_NO_NEW_PRIVS
, prctl_arg(1), prctl_arg(0),
1286 prctl_arg(0), prctl_arg(0));
1288 SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges");
1289 goto out_warn_father
;
1291 DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges");
1294 /* If we mounted a temporary proc, then unmount it now. */
1295 tmp_proc_unmount(handler
->conf
);
1297 ret
= lxc_seccomp_load(handler
->conf
);
1299 goto out_warn_father
;
1301 ret
= run_lxc_hooks(handler
->name
, "start", handler
->conf
, NULL
);
1303 ERROR("Failed to run lxc.hook.start for container \"%s\"",
1305 goto out_warn_father
;
1308 close_prot_errno_disarm(handler
->sigfd
);
1310 if (handler
->conf
->console
.pty
< 0 && handler
->daemonize
) {
1311 if (devnull_fd
< 0) {
1312 devnull_fd
= open_devnull();
1314 goto out_warn_father
;
1317 ret
= set_stdfds(devnull_fd
);
1319 ERROR("Failed to redirect std{in,out,err} to \"/dev/null\"");
1320 goto out_warn_father
;
1324 close_prot_errno_disarm(devnull_fd
);
1328 if (handler
->conf
->init_cwd
) {
1329 ret
= chdir(handler
->conf
->init_cwd
);
1331 SYSERROR("Could not change directory to \"%s\"",
1332 handler
->conf
->init_cwd
);
1333 goto out_warn_father
;
1337 if (!lxc_sync_barrier_parent(handler
, START_SYNC_CGROUP_LIMITS
))
1338 goto out_warn_father
;
1340 ret
= lxc_sync_fds_child(handler
);
1342 SYSERROR("Failed to sync file descriptors with parent");
1343 goto out_warn_father
;
1346 if (!lxc_sync_wait_parent(handler
, START_SYNC_READY_START
))
1347 goto out_warn_father
;
1349 /* Reset the environment variables the user requested in a clear
1353 /* Don't error out though. */
1355 SYSERROR("Failed to clear environment.");
1357 ret
= lxc_set_environment(handler
->conf
);
1359 goto out_warn_father
;
1361 ret
= putenv("container=lxc");
1363 SYSERROR("Failed to set environment variable: container=lxc");
1364 goto out_warn_father
;
1367 if (handler
->conf
->ttys
.tty_names
) {
1368 ret
= putenv(handler
->conf
->ttys
.tty_names
);
1370 SYSERROR("Failed to set environment variable for container ptys");
1371 goto out_warn_father
;
1375 /* The container has been setup. We can now switch to an unprivileged
1378 new_uid
= handler
->conf
->init_uid
;
1379 new_gid
= handler
->conf
->init_gid
;
1381 /* Avoid unnecessary syscalls. */
1382 if (new_uid
== nsuid
)
1383 new_uid
= LXC_INVALID_UID
;
1385 if (new_gid
== nsgid
)
1386 new_gid
= LXC_INVALID_GID
;
1388 /* Make sure that the processes STDIO is correctly owned by the user that we are switching to */
1389 ret
= fix_stdio_permissions(new_uid
);
1391 WARN("Failed to ajust stdio permissions");
1393 /* If we are in a new user namespace we already dropped all groups when
1394 * we switched to root in the new user namespace further above. Only
1395 * drop groups if we can, so ensure that we have necessary privilege.
1397 if (list_empty(&handler
->conf
->id_map
)) {
1399 if (lxc_proc_cap_is_set(CAP_SETGID
, CAP_EFFECTIVE
))
1402 if (handler
->conf
->init_groups
.size
> 0) {
1403 if (!lxc_setgroups(handler
->conf
->init_groups
.list
,
1404 handler
->conf
->init_groups
.size
))
1405 goto out_warn_father
;
1407 if (!lxc_drop_groups())
1408 goto out_warn_father
;
1413 if (!lxc_switch_uid_gid(new_uid
, new_gid
))
1414 goto out_warn_father
;
1416 ret
= lxc_ambient_caps_down();
1418 ERROR("Failed to clear ambient capabilities");
1419 goto out_warn_father
;
1422 if (handler
->conf
->monitor_signal_pdeath
!= SIGKILL
) {
1423 ret
= lxc_set_death_signal(handler
->conf
->monitor_signal_pdeath
,
1424 handler
->monitor_pid
, status_fd
);
1426 SYSERROR("Failed to set PR_SET_PDEATHSIG to %d",
1427 handler
->conf
->monitor_signal_pdeath
);
1428 goto out_warn_father
;
1433 * After this call, we are in error because this ops should not return
1436 handler
->ops
->start(handler
, handler
->data
);
1440 * We want the parent to know something went wrong, so we return a
1441 * special error code.
1443 lxc_sync_wake_parent(handler
, SYNC_ERROR
);
1449 int resolve_clone_flags(struct lxc_handler
*handler
)
1452 struct lxc_conf
*conf
= handler
->conf
;
1453 bool wants_timens
= conf
->timens
.s_boot
|| conf
->timens
.ns_boot
||
1454 conf
->timens
.s_monotonic
|| conf
->timens
.ns_monotonic
;
1456 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1457 if (conf
->ns_keep
) {
1458 if (!(conf
->ns_keep
& ns_info
[i
].clone_flag
))
1459 handler
->ns_clone_flags
|= ns_info
[i
].clone_flag
;
1460 } else if (conf
->ns_clone
) {
1461 if ((conf
->ns_clone
& ns_info
[i
].clone_flag
))
1462 handler
->ns_clone_flags
|= ns_info
[i
].clone_flag
;
1464 if (i
== LXC_NS_USER
&& list_empty(&handler
->conf
->id_map
))
1467 if (i
== LXC_NS_NET
&& lxc_requests_empty_network(handler
))
1470 if (i
== LXC_NS_CGROUP
&& !cgns_supported())
1473 if (i
== LXC_NS_TIME
&& !wants_timens
)
1476 handler
->ns_clone_flags
|= ns_info
[i
].clone_flag
;
1479 if (!conf
->ns_share
[i
])
1482 handler
->ns_clone_flags
&= ~ns_info
[i
].clone_flag
;
1483 TRACE("Sharing %s namespace", ns_info
[i
].proc_name
);
1486 if (wants_timens
&& (conf
->ns_keep
& ns_info
[LXC_NS_TIME
].clone_flag
))
1487 return log_trace_errno(-1, EINVAL
, "Requested to keep time namespace while also specifying offsets");
1489 /* Deal with namespaces that are unshared. */
1490 if (handler
->ns_clone_flags
& CLONE_NEWTIME
)
1491 handler
->ns_unshare_flags
|= CLONE_NEWTIME
;
1493 if (!pure_unified_layout(handler
->cgroup_ops
) && handler
->ns_clone_flags
& CLONE_NEWCGROUP
)
1494 handler
->ns_unshare_flags
|= CLONE_NEWCGROUP
;
1496 if ((handler
->ns_clone_flags
& (CLONE_NEWNET
| CLONE_NEWUSER
)) ==
1497 (CLONE_NEWNET
| CLONE_NEWUSER
))
1498 handler
->ns_unshare_flags
|= CLONE_NEWNET
;
1500 /* Deal with namespaces that are spawned. */
1501 handler
->ns_on_clone_flags
= handler
->ns_clone_flags
& ~handler
->ns_unshare_flags
;
1503 handler
->clone_flags
= handler
->ns_on_clone_flags
| CLONE_PIDFD
;
1508 /* Note that this function is used with clone(CLONE_VM). Some glibc versions
1509 * used to reset the pid/tid to -1 when CLONE_VM was used without CLONE_THREAD.
1510 * But since the memory between parent and child is shared on CLONE_VM this
1511 * would invalidate the getpid() cache that glibc used to maintain and so
1512 * getpid() in the child would return the parent's pid. This is all fixed in
1513 * newer glibc versions where the getpid() cache is removed and the pid/tid is
1514 * not reset anymore.
1515 * However, if for whatever reason you - dear committer - somehow need to get the
1516 * pid of the placeholder intermediate process for do_share_ns() you need to
1517 * call lxc_raw_getpid(). The next lxc_raw_clone() call does not employ
1518 * CLONE_VM and will be fine.
1520 static inline int do_share_ns(void *arg
)
1523 struct lxc_handler
*handler
= arg
;
1525 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1526 if (handler
->nsfd
[i
] < 0)
1529 ret
= setns(handler
->nsfd
[i
], 0);
1532 * Note that joining a user and/or mount namespace
1533 * requires the process is not multithreaded otherwise
1534 * setns() will fail here.
1536 SYSERROR("Failed to inherit %s namespace",
1537 ns_info
[i
].proc_name
);
1541 DEBUG("Inherited %s namespace", ns_info
[i
].proc_name
);
1544 flags
= handler
->ns_on_clone_flags
;
1545 flags
|= CLONE_PARENT
;
1546 handler
->pid
= lxc_raw_clone_cb(do_start
, handler
, CLONE_PIDFD
| flags
,
1548 if (handler
->pid
< 0)
1554 static int core_scheduling(struct lxc_handler
*handler
)
1556 struct lxc_conf
*conf
= handler
->conf
;
1559 if (!conf
->sched_core
)
1560 return log_trace(0, "No new core scheduling domain requested");
1562 if (!(handler
->ns_clone_flags
& CLONE_NEWPID
))
1563 return syserror_set(-EINVAL
, "Core scheduling currently requires a separate pid namespace");
1565 ret
= core_scheduling_cookie_create_threadgroup(handler
->pid
);
1567 if (ret
== -ENODEV
) {
1568 INFO("The kernel doesn't support or doesn't use simultaneous multithreading (SMT)");
1569 conf
->sched_core
= false;
1573 return syserror("The kernel does not support core scheduling");
1575 return syserror("Failed to create new core scheduling domain");
1578 ret
= core_scheduling_cookie_get(handler
->pid
, &conf
->sched_core_cookie
);
1579 if (ret
|| !core_scheduling_cookie_valid(conf
->sched_core_cookie
))
1580 return syserror("Failed to retrieve core scheduling domain cookie");
1582 TRACE("Created new core scheduling domain with cookie %llu",
1583 (llu
)conf
->sched_core_cookie
);
1588 static bool inherits_namespaces(const struct lxc_handler
*handler
)
1590 struct lxc_conf
*conf
= handler
->conf
;
1592 for (lxc_namespace_t i
= 0; i
< LXC_NS_MAX
; i
++) {
1593 if (conf
->ns_share
[i
])
1600 static inline void resolve_cgroup_clone_flags(struct lxc_handler
*handler
)
1602 handler
->clone_flags
&= ~(CLONE_INTO_CGROUP
| CLONE_NEWCGROUP
);
1603 handler
->ns_on_clone_flags
&= ~(CLONE_INTO_CGROUP
| CLONE_NEWCGROUP
);
1604 handler
->ns_unshare_flags
|= CLONE_NEWCGROUP
;
1607 /* lxc_spawn() performs crucial setup tasks and clone()s the new process which
1608 * exec()s the requested container binary.
1609 * Note that lxc_spawn() runs in the parent namespaces. Any operations performed
1610 * right here should be double checked if they'd pose a security risk. (For
1611 * example, any {u}mount() operations performed here will be reflected on the
1614 static int lxc_spawn(struct lxc_handler
*handler
)
1616 __do_close
int data_sock0
= -EBADF
, data_sock1
= -EBADF
;
1619 bool wants_to_map_ids
;
1620 struct list_head
*id_map
;
1621 const char *name
= handler
->name
;
1622 struct lxc_conf
*conf
= handler
->conf
;
1623 struct cgroup_ops
*cgroup_ops
= handler
->cgroup_ops
;
1625 id_map
= &conf
->id_map
;
1626 wants_to_map_ids
= !list_empty(id_map
);
1628 if (!lxc_sync_init(handler
))
1631 ret
= socketpair(AF_UNIX
, SOCK_STREAM
| SOCK_CLOEXEC
, 0,
1632 handler
->data_sock
);
1635 data_sock0
= handler
->data_sock
[0];
1636 data_sock1
= handler
->data_sock
[1];
1638 if (handler
->ns_clone_flags
& CLONE_NEWNET
) {
1639 ret
= lxc_find_gateway_addresses(handler
);
1641 ERROR("Failed to find gateway addresses");
1646 if (!cgroup_ops
->payload_create(cgroup_ops
, handler
)) {
1647 ERROR("Failed creating cgroups");
1648 goto out_delete_net
;
1651 /* Create a process in a new set of namespaces. */
1652 if (inherits_namespaces(handler
)) {
1655 resolve_cgroup_clone_flags(handler
);
1656 attacher_pid
= lxc_clone(do_share_ns
, handler
,
1657 CLONE_VFORK
| CLONE_VM
| CLONE_FILES
, NULL
);
1658 if (attacher_pid
< 0) {
1659 SYSERROR(LXC_CLONE_ERROR
);
1660 goto out_delete_net
;
1663 ret
= wait_for_pid(attacher_pid
);
1665 SYSERROR("Intermediate process failed");
1666 goto out_delete_net
;
1669 if (handler
->pid
< 0) {
1670 SYSERROR(LXC_CLONE_ERROR
);
1671 goto out_delete_net
;
1674 int cgroup_fd
= -EBADF
;
1676 struct clone_args clone_args
= {
1677 .flags
= handler
->clone_flags
,
1678 .pidfd
= ptr_to_u64(&handler
->pidfd
),
1679 .exit_signal
= SIGCHLD
,
1682 if (handler
->ns_clone_flags
& CLONE_NEWCGROUP
) {
1683 cgroup_fd
= cgroup_unified_fd(cgroup_ops
);
1684 if (cgroup_fd
>= 0) {
1685 handler
->clone_flags
|= CLONE_INTO_CGROUP
;
1686 clone_args
.flags
|= CLONE_INTO_CGROUP
;
1687 clone_args
.cgroup
= cgroup_fd
;
1691 /* Try to spawn directly into target cgroup. */
1692 handler
->pid
= lxc_clone3(&clone_args
, CLONE_ARGS_SIZE_VER2
);
1693 if (handler
->pid
< 0) {
1694 SYSTRACE("Failed to spawn container directly into target cgroup");
1696 /* Kernel might simply be too old for CLONE_INTO_CGROUP. */
1697 resolve_cgroup_clone_flags(handler
);
1698 clone_args
.flags
= handler
->clone_flags
;
1700 handler
->pid
= lxc_clone3(&clone_args
, CLONE_ARGS_SIZE_VER0
);
1701 } else if (cgroup_fd
>= 0) {
1702 TRACE("Spawned container directly into target cgroup via cgroup2 fd %d", cgroup_fd
);
1705 /* Kernel might be too old for clone3(). */
1706 if (handler
->pid
< 0) {
1707 SYSTRACE("Failed to spawn container via clone3()");
1710 * In contrast to all other architectures arm64 verifies that
1711 * the argument we use to retrieve the pidfd with is
1712 * initialized to 0. But we need to be able to initialize it to
1713 * a negative value such as our customary -EBADF so we can
1714 * detect whether this kernel supports pidfds. If the syscall
1715 * returns and the pidfd variable is set to something >= 0 then
1716 * we know this is a kernel supporting pidfds. But if we can't
1717 * set it to -EBADF then this won't work since 0 is a valid
1718 * file descriptor too. And since legacy clone silently ignores
1719 * unknown flags we are left without any way to detect support
1720 * for pidfds. So let's special-case arm64 to not fail starting
1723 #if defined(__aarch64__)
1724 handler
->pid
= lxc_raw_legacy_clone(handler
->clone_flags
& ~CLONE_PIDFD
, NULL
);
1726 handler
->pid
= lxc_raw_legacy_clone(handler
->clone_flags
, &handler
->pidfd
);
1730 if (handler
->pid
< 0) {
1731 SYSERROR(LXC_CLONE_ERROR
);
1732 goto out_delete_net
;
1735 if (handler
->pid
== 0) {
1736 (void)do_start(handler
);
1737 _exit(EXIT_FAILURE
);
1740 if (handler
->pidfd
< 0)
1741 handler
->clone_flags
&= ~CLONE_PIDFD
;
1742 TRACE("Cloned child process %d", handler
->pid
);
1744 ret
= core_scheduling(handler
);
1746 goto out_delete_net
;
1748 /* Verify that we can actually make use of pidfds. */
1749 if (!lxc_can_use_pidfd(handler
->pidfd
))
1750 close_prot_errno_disarm(handler
->pidfd
);
1752 ret
= strnprintf(pidstr
, 20, "%d", handler
->pid
);
1754 goto out_delete_net
;
1756 ret
= setenv("LXC_PID", pidstr
, 1);
1758 SYSERROR("Failed to set environment variable: LXC_PID=%s", pidstr
);
1760 for (i
= 0; i
< LXC_NS_MAX
; i
++)
1761 if (handler
->ns_on_clone_flags
& ns_info
[i
].clone_flag
)
1762 INFO("Cloned %s", ns_info
[i
].flag_name
);
1764 if (!lxc_try_preserve_namespaces(handler
, handler
->ns_on_clone_flags
)) {
1765 ERROR("Failed to preserve cloned namespaces for lxc.hook.stop");
1766 goto out_delete_net
;
1769 lxc_sync_fini_child(handler
);
1771 if (lxc_abstract_unix_send_fds(handler
->data_sock
[0], &handler
->monitor_status_fd
, 1, NULL
, 0) < 0) {
1772 ERROR("Failed to send status file descriptor to child process");
1773 goto out_delete_net
;
1775 close_prot_errno_disarm(handler
->monitor_status_fd
);
1777 /* Map the container uids. The container became an invalid userid the
1778 * moment it was cloned with CLONE_NEWUSER. This call doesn't change
1779 * anything immediately, but allows the container to setuid(0) (0 being
1780 * mapped to something else on the host.) later to become a valid uid
1783 if (wants_to_map_ids
) {
1784 if (!handler
->conf
->ns_share
[LXC_NS_USER
] &&
1785 (handler
->conf
->ns_keep
& CLONE_NEWUSER
) == 0) {
1786 ret
= lxc_map_ids(id_map
, handler
->pid
);
1788 ERROR("Failed to set up id mapping.");
1789 goto out_delete_net
;
1794 if (!cgroup_ops
->setup_limits_legacy(cgroup_ops
, handler
->conf
, false)) {
1795 ERROR("Failed to setup cgroup limits for container \"%s\"", name
);
1796 goto out_delete_net
;
1799 if (!cgroup_ops
->payload_delegate_controllers(cgroup_ops
)) {
1800 ERROR("Failed to delegate controllers to payload cgroup");
1801 goto out_delete_net
;
1804 if (!cgroup_ops
->payload_enter(cgroup_ops
, handler
)) {
1805 ERROR("Failed to enter cgroups");
1806 goto out_delete_net
;
1809 if (!cgroup_ops
->setup_limits(cgroup_ops
, handler
)) {
1810 ERROR("Failed to setup cgroup limits for container \"%s\"", name
);
1811 goto out_delete_net
;
1814 if (!cgroup_ops
->chown(cgroup_ops
, handler
->conf
))
1815 goto out_delete_net
;
1817 if (!lxc_sync_barrier_child(handler
, START_SYNC_STARTUP
))
1818 goto out_delete_net
;
1820 /* If not done yet, we're now ready to preserve the network namespace */
1821 if (handler
->nsfd
[LXC_NS_NET
] < 0) {
1822 ret
= lxc_try_preserve_namespace(handler
, LXC_NS_NET
, "net");
1824 if (ret
!= -ENOENT
) {
1825 SYSERROR("Failed to preserve net namespace");
1826 goto out_delete_net
;
1830 ret
= lxc_netns_set_nsid(handler
->nsfd
[LXC_NS_NET
]);
1832 SYSWARN("Failed to allocate new network namespace id");
1834 TRACE("Allocated new network namespace id");
1836 /* Create the network configuration. */
1837 if (handler
->ns_clone_flags
& CLONE_NEWNET
) {
1838 ret
= lxc_create_network(handler
);
1840 ERROR("Failed to create the network");
1841 goto out_delete_net
;
1845 ret
= setup_proc_filesystem(conf
, handler
->pid
);
1847 ERROR("Failed to setup procfs limits");
1848 goto out_delete_net
;
1851 ret
= setup_resource_limits(conf
, handler
->pid
);
1853 ERROR("Failed to setup resource limits");
1854 goto out_delete_net
;
1857 /* Tell the child to continue its initialization. */
1858 if (!lxc_sync_wake_child(handler
, START_SYNC_POST_CONFIGURE
))
1859 goto out_delete_net
;
1861 ret
= lxc_rootfs_prepare_parent(handler
);
1863 ERROR("Failed to prepare rootfs");
1864 goto out_delete_net
;
1867 if (handler
->ns_clone_flags
& CLONE_NEWNET
) {
1868 ret
= lxc_network_send_to_child(handler
);
1870 SYSERROR("Failed to send veth names to child");
1871 goto out_delete_net
;
1875 if (!lxc_sync_wait_child(handler
, START_SYNC_IDMAPPED_MOUNTS
))
1876 goto out_delete_net
;
1878 ret
= lxc_idmapped_mounts_parent(handler
);
1880 ERROR("Failed to setup mount entries");
1881 goto out_delete_net
;
1884 if (!lxc_sync_wait_child(handler
, START_SYNC_CGROUP_LIMITS
))
1885 goto out_delete_net
;
1888 * With isolation the limiting devices cgroup was already setup, so
1889 * only setup devices here if we have no namespace directory.
1891 if (!handler
->conf
->cgroup_meta
.namespace_dir
&&
1892 !cgroup_ops
->setup_limits_legacy(cgroup_ops
, handler
->conf
, true)) {
1893 ERROR("Failed to setup legacy device cgroup controller limits");
1894 goto out_delete_net
;
1896 TRACE("Set up legacy device cgroup controller limits");
1898 if (!cgroup_ops
->devices_activate(cgroup_ops
, handler
)) {
1899 ERROR("Failed to setup cgroup2 device controller limits");
1900 goto out_delete_net
;
1902 TRACE("Set up cgroup2 device controller limits");
1904 cgroup_ops
->finalize(cgroup_ops
);
1905 TRACE("Finished setting up cgroups");
1907 /* Run any host-side start hooks */
1908 ret
= run_lxc_hooks(name
, "start-host", conf
, NULL
);
1910 ERROR("Failed to run lxc.hook.start-host");
1911 goto out_delete_net
;
1914 if (!lxc_sync_wake_child(handler
, START_SYNC_FDS
))
1915 goto out_delete_net
;
1917 if (handler
->ns_unshare_flags
& CLONE_NEWCGROUP
) {
1918 /* Now we're ready to preserve the cgroup namespace */
1919 ret
= lxc_try_preserve_namespace(handler
, LXC_NS_CGROUP
, "cgroup");
1921 if (ret
!= -ENOENT
) {
1922 SYSERROR("Failed to preserve cgroup namespace");
1923 goto out_delete_net
;
1928 if (handler
->ns_unshare_flags
& CLONE_NEWTIME
) {
1929 /* Now we're ready to preserve the time namespace */
1930 ret
= lxc_try_preserve_namespace(handler
, LXC_NS_TIME
, "time");
1932 if (ret
!= -ENOENT
) {
1933 SYSERROR("Failed to preserve time namespace");
1934 goto out_delete_net
;
1939 ret
= lxc_sync_fds_parent(handler
);
1941 SYSERROR("Failed to sync file descriptors with child");
1942 goto out_delete_net
;
1945 ret
= lxc_terminal_setup(conf
);
1947 SYSERROR("Failed to create console");
1948 goto out_delete_net
;
1952 * Tell the child to complete its initialization and wait for it to
1953 * exec or return an error. (The child will never return
1954 * START_SYNC_READY_START+1. It will either close the sync pipe,
1955 * causing lxc_sync_barrier_child to return success, or return a
1956 * different value, causing us to error out).
1958 if (!lxc_sync_barrier_child(handler
, START_SYNC_READY_START
))
1959 goto out_delete_net
;
1961 /* Now all networks are created, network devices are moved into place,
1962 * and the correct names and ifindices in the respective namespaces have
1963 * been recorded. The corresponding structs have now all been filled. So
1964 * log them for debugging purposes.
1966 lxc_log_configured_netdevs(conf
);
1968 ret
= handler
->ops
->post_start(handler
, handler
->data
);
1972 ret
= lxc_set_state(name
, handler
, RUNNING
);
1974 ERROR("Failed to set state to \"%s\"", lxc_state2str(RUNNING
));
1978 lxc_sync_fini(handler
);
1983 if (handler
->ns_clone_flags
& CLONE_NEWNET
)
1984 lxc_delete_network(handler
);
1990 lxc_sync_fini(handler
);
1995 static int lxc_inherit_namespaces(struct lxc_handler
*handler
)
1997 const char *lxcpath
= handler
->lxcpath
;
1998 struct lxc_conf
*conf
= handler
->conf
;
2000 for (lxc_namespace_t i
= 0; i
< LXC_NS_MAX
; i
++) {
2001 if (!conf
->ns_share
[i
])
2004 handler
->nsfd
[i
] = lxc_inherit_namespace(conf
->ns_share
[i
],
2006 ns_info
[i
].proc_name
);
2007 if (handler
->nsfd
[i
] < 0)
2010 TRACE("Recording inherited %s namespace with fd %d",
2011 ns_info
[i
].proc_name
, handler
->nsfd
[i
]);
2017 int __lxc_start(struct lxc_handler
*handler
, struct lxc_operations
*ops
,
2018 void *data
, const char *lxcpath
, bool daemonize
, int *error_num
)
2021 const char *name
= handler
->name
;
2022 struct lxc_conf
*conf
= handler
->conf
;
2023 struct cgroup_ops
*cgroup_ops
;
2025 ret
= lxc_init(name
, handler
);
2027 ERROR("Failed to initialize container \"%s\"", name
);
2031 handler
->data
= data
;
2032 handler
->daemonize
= daemonize
;
2033 cgroup_ops
= handler
->cgroup_ops
;
2035 if (!attach_block_device(handler
->conf
)) {
2036 ERROR("Failed to attach block device");
2041 if (!cgroup_ops
->monitor_create(cgroup_ops
, handler
)) {
2042 ERROR("Failed to create monitor cgroup");
2047 if (!cgroup_ops
->monitor_delegate_controllers(cgroup_ops
)) {
2048 ERROR("Failed to delegate controllers to monitor cgroup");
2053 if (!cgroup_ops
->monitor_enter(cgroup_ops
, handler
)) {
2054 ERROR("Failed to enter monitor cgroup");
2059 ret
= resolve_clone_flags(handler
);
2061 ERROR("Failed to resolve clone flags");
2066 ret
= lxc_inherit_namespaces(handler
);
2068 SYSERROR("Failed to record inherited namespaces");
2073 /* If the rootfs is not a blockdev, prevent the container from marking
2075 * If the container is unprivileged then skip rootfs pinning.
2077 ret
= lxc_rootfs_init(conf
, !list_empty(&conf
->id_map
));
2079 ERROR("Failed to handle rootfs pinning for container \"%s\"", handler
->name
);
2084 if (geteuid() == 0 && !list_empty(&conf
->id_map
)) {
2086 * Most filesystems can't be mounted inside a userns so handle them here.
2088 if (rootfs_is_blockdev(conf
)) {
2089 ret
= unshare(CLONE_NEWNS
);
2091 ERROR("Failed to unshare CLONE_NEWNS");
2094 INFO("Unshared CLONE_NEWNS");
2096 ret
= lxc_setup_rootfs_prepare_root(conf
, name
, lxcpath
);
2098 ERROR("Error setting up rootfs mount as root before spawn");
2101 INFO("Set up container rootfs as host root");
2105 ret
= lxc_spawn(handler
);
2107 ERROR("Failed to spawn container \"%s\"", name
);
2108 goto out_detach_blockdev
;
2111 handler
->conf
->reboot
= REBOOT_NONE
;
2113 ret
= lxc_poll(name
, handler
);
2115 ERROR("LXC mainloop exited with error: %d", ret
);
2116 goto out_delete_network
;
2119 if (!handler
->init_died
&& handler
->pid
> 0) {
2120 ERROR("Child process is not killed");
2122 goto out_delete_network
;
2125 status
= lxc_wait_for_pid_status(handler
->pid
);
2127 SYSERROR("Failed to retrieve status for %d", handler
->pid
);
2129 /* If the child process exited but was not signaled, it didn't call
2130 * reboot. This should mean it was an lxc-execute which simply exited.
2131 * In any case, treat it as a 'halt'.
2133 if (WIFSIGNALED(status
)) {
2134 int signal_nr
= WTERMSIG(status
);
2136 case SIGINT
: /* halt */
2137 DEBUG("%s(%d) - Container \"%s\" is halting", signal_name(signal_nr
), signal_nr
, name
);
2139 case SIGHUP
: /* reboot */
2140 DEBUG("%s(%d) - Container \"%s\" is rebooting", signal_name(signal_nr
), signal_nr
, name
);
2141 handler
->conf
->reboot
= REBOOT_REQ
;
2143 case SIGSYS
: /* seccomp */
2144 DEBUG("%s(%d) - Container \"%s\" violated its seccomp policy", signal_name(signal_nr
), signal_nr
, name
);
2147 DEBUG("%s(%d) - Container \"%s\" init exited", signal_name(signal_nr
), signal_nr
, name
);
2152 ret
= lxc_restore_phys_nics_to_netns(handler
);
2154 ERROR("Failed to move physical network devices back to parent network namespace");
2156 lxc_monitor_send_exit_code(name
, status
, handler
->lxcpath
);
2157 lxc_error_set_and_log(handler
->pid
, status
);
2159 *error_num
= handler
->exit_status
;
2161 lxc_delete_network(handler
);
2162 detach_block_device(handler
->conf
);
2171 out_detach_blockdev
:
2173 detach_block_device(handler
->conf
);
2179 lxc_restore_phys_nics_to_netns(handler
);
2180 lxc_delete_network(handler
);
2181 detach_block_device(handler
->conf
);
2190 static int start(struct lxc_handler
*handler
, void* data
)
2192 struct start_args
*arg
= data
;
2194 NOTICE("Exec'ing \"%s\"", arg
->argv
[0]);
2196 execvp(arg
->argv
[0], arg
->argv
);
2197 SYSERROR("Failed to exec \"%s\"", arg
->argv
[0]);
2201 static int post_start(struct lxc_handler
*handler
, void* data
)
2203 struct start_args
*arg
= data
;
2205 NOTICE("Started \"%s\" with pid \"%d\"", arg
->argv
[0], handler
->pid
);
2209 static struct lxc_operations start_ops
= {
2211 .post_start
= post_start
2214 int lxc_start(char *const argv
[], struct lxc_handler
*handler
,
2215 const char *lxcpath
, bool daemonize
, int *error_num
)
2217 struct start_args start_arg
= {
2221 TRACE("Doing lxc_start");
2222 return __lxc_start(handler
, &start_ops
, &start_arg
, lxcpath
, daemonize
, error_num
);
2225 static void lxc_destroy_container_on_signal(struct lxc_handler
*handler
,
2228 char destroy
[PATH_MAX
];
2229 struct lxc_container
*c
;
2233 if (handler
->conf
->rootfs
.path
&& handler
->conf
->rootfs
.mount
) {
2234 bret
= do_destroy_container(handler
);
2236 ERROR("Error destroying rootfs for container \"%s\"", name
);
2240 INFO("Destroyed rootfs for container \"%s\"", name
);
2242 ret
= strnprintf(destroy
, sizeof(destroy
), "%s/%s", handler
->lxcpath
, name
);
2244 ERROR("Error destroying directory for container \"%s\"", name
);
2248 c
= lxc_container_new(name
, handler
->lxcpath
);
2250 if (container_disk_lock(c
)) {
2251 INFO("Could not update lxc_snapshots file");
2252 lxc_container_put(c
);
2254 mod_all_rdeps(c
, false);
2255 container_disk_unlock(c
);
2256 lxc_container_put(c
);
2260 if (!handler
->am_root
)
2261 ret
= userns_exec_full(handler
->conf
, lxc_rmdir_onedev_wrapper
,
2262 destroy
, "lxc_rmdir_onedev_wrapper");
2264 ret
= lxc_rmdir_onedev(destroy
, NULL
);
2267 ERROR("Error destroying directory for container \"%s\"", name
);
2270 INFO("Destroyed directory for container \"%s\"", name
);
2273 static int lxc_rmdir_onedev_wrapper(void *data
)
2275 char *arg
= (char *) data
;
2276 return lxc_rmdir_onedev(arg
, NULL
);
2279 static bool do_destroy_container(struct lxc_handler
*handler
)
2283 if (!handler
->am_root
) {
2284 ret
= userns_exec_full(handler
->conf
, storage_destroy_wrapper
,
2285 handler
->conf
, "storage_destroy_wrapper");
2292 return storage_destroy(handler
->conf
);