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
++)
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 ((size_t)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 (size_t 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 ((__u64
)siginfo
.ssi_pid
!= (__u64
)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 if (descr
.type
== LXC_MAINLOOP_EPOLL
)
633 close_prot_errno_disarm(descr
.epfd
);
634 if (ret
< 0 || !handler
->init_died
)
635 goto out_mainloop_console
;
638 ret
= lxc_terminal_mainloop_add(&descr_console
, console
);
640 ret
= lxc_mainloop(&descr_console
, 0);
643 out_mainloop_console
:
645 lxc_mainloop_close(&descr_console
);
646 TRACE("Closed console mainloop");
650 lxc_mainloop_close(&descr
);
651 TRACE("Closed mainloop");
654 TRACE("Closed signal file descriptor %d", handler
->sigfd
);
655 close_prot_errno_disarm(handler
->sigfd
);
660 void lxc_put_handler(struct lxc_handler
*handler
)
662 close_prot_errno_disarm(handler
->pidfd
);
663 close_prot_errno_disarm(handler
->sigfd
);
664 lxc_put_nsfds(handler
);
665 if (handler
->conf
&& handler
->conf
->reboot
== REBOOT_NONE
)
666 close_prot_errno_disarm(handler
->conf
->maincmd_fd
);
667 close_prot_errno_disarm(handler
->monitor_status_fd
);
668 close_prot_errno_disarm(handler
->state_socket_pair
[0]);
669 close_prot_errno_disarm(handler
->state_socket_pair
[1]);
670 cgroup_exit(handler
->cgroup_ops
);
671 if (handler
->conf
&& handler
->conf
->reboot
== REBOOT_NONE
)
672 free_disarm(handler
);
674 handler
->conf
= NULL
;
677 struct lxc_handler
*lxc_init_handler(struct lxc_handler
*old
,
678 const char *name
, struct lxc_conf
*conf
,
679 const char *lxcpath
, bool daemonize
)
683 struct lxc_handler
*handler
;
686 handler
= zalloc(sizeof(*handler
));
692 /* Note that am_guest_unpriv() checks the effective uid. We
693 * probably don't care if we are real root only if we are running
694 * as root so this should be fine.
696 handler
->am_root
= !am_guest_unpriv();
697 handler
->conf
= conf
;
698 handler
->lxcpath
= lxcpath
;
699 handler
->init_died
= false;
700 handler
->data_sock
[0] = -EBADF
;
701 handler
->data_sock
[1] = -EBADF
;
702 handler
->monitor_status_fd
= -EBADF
;
703 handler
->pidfd
= -EBADF
;
704 handler
->sigfd
= -EBADF
;
705 handler
->state_socket_pair
[0] = -EBADF
;
706 handler
->state_socket_pair
[1] = -EBADF
;
707 if (handler
->conf
->reboot
== REBOOT_NONE
)
708 INIT_LIST_HEAD(&handler
->conf
->state_clients
);
710 for (lxc_namespace_t idx
= 0; idx
< LXC_NS_MAX
; idx
++) {
711 handler
->nsfd
[idx
] = -EBADF
;
713 if (handler
->conf
->reboot
== REBOOT_NONE
)
716 handler
->nsfd_paths
[idx
][0] = '\0';
717 handler
->hook_argv
[idx
] = NULL
;
719 if (handler
->hook_argc
!= 0)
720 handler
->hook_argc
= 0;
723 handler
->name
= name
;
725 handler
->transient_pid
= lxc_raw_getpid();
727 handler
->transient_pid
= -1;
729 if (daemonize
&& handler
->conf
->reboot
== REBOOT_NONE
) {
730 /* Create socketpair() to synchronize on daemonized startup.
731 * When the container reboots we don't need to synchronize
732 * again currently so don't open another socketpair().
734 ret
= socketpair(AF_UNIX
, SOCK_STREAM
| SOCK_CLOEXEC
, 0,
735 handler
->state_socket_pair
);
737 ERROR("Failed to create anonymous pair of unix sockets");
741 TRACE("Created anonymous pair {%d,%d} of unix sockets",
742 handler
->state_socket_pair
[0],
743 handler
->state_socket_pair
[1]);
744 handler
->keep_fds
[nr_keep_fds
++] = handler
->state_socket_pair
[0];
745 handler
->keep_fds
[nr_keep_fds
++] = handler
->state_socket_pair
[1];
748 if (handler
->conf
->reboot
== REBOOT_NONE
) {
749 handler
->conf
->maincmd_fd
= lxc_server_init(name
, lxcpath
, "command");
750 if (handler
->conf
->maincmd_fd
< 0) {
751 ERROR("Failed to set up command socket");
754 handler
->keep_fds
[nr_keep_fds
++] = handler
->conf
->maincmd_fd
;
757 TRACE("Unix domain socket %d for command server is ready",
758 handler
->conf
->maincmd_fd
);
763 lxc_put_handler(handler
);
768 int lxc_init(const char *name
, struct lxc_handler
*handler
)
770 __do_close
int status_fd
= -EBADF
;
772 const char *loglevel
;
773 struct lxc_conf
*conf
= handler
->conf
;
775 handler
->monitor_pid
= lxc_raw_getpid();
776 status_fd
= open("/proc/self/status", O_RDONLY
| O_CLOEXEC
);
778 return log_error_errno(-1, errno
, "Failed to open monitor status fd");
780 handler
->lsm_ops
= lsm_init_static();
781 TRACE("Initialized LSM");
783 /* Begin by setting the state to STARTING. */
784 ret
= lxc_set_state(name
, handler
, STARTING
);
786 return log_error(-1, "Failed to set state to \"%s\"", lxc_state2str(STARTING
));
787 TRACE("Set container state to \"STARTING\"");
789 /* Start of environment variable setup for hooks. */
790 ret
= setenv("LXC_NAME", name
, 1);
792 SYSERROR("Failed to set environment variable: LXC_NAME=%s", name
);
795 ret
= setenv("LXC_CONFIG_FILE", conf
->rcfile
, 1);
797 SYSERROR("Failed to set environment variable: LXC_CONFIG_FILE=%s", conf
->rcfile
);
800 if (conf
->rootfs
.mount
) {
801 ret
= setenv("LXC_ROOTFS_MOUNT", conf
->rootfs
.mount
, 1);
803 SYSERROR("Failed to set environment variable: LXC_ROOTFS_MOUNT=%s", conf
->rootfs
.mount
);
806 if (conf
->rootfs
.path
) {
807 ret
= setenv("LXC_ROOTFS_PATH", conf
->rootfs
.path
, 1);
809 SYSERROR("Failed to set environment variable: LXC_ROOTFS_PATH=%s", conf
->rootfs
.path
);
812 if (conf
->console
.path
) {
813 ret
= setenv("LXC_CONSOLE", conf
->console
.path
, 1);
815 SYSERROR("Failed to set environment variable: LXC_CONSOLE=%s", conf
->console
.path
);
818 if (conf
->console
.log_path
) {
819 ret
= setenv("LXC_CONSOLE_LOGPATH", conf
->console
.log_path
, 1);
821 SYSERROR("Failed to set environment variable: LXC_CONSOLE_LOGPATH=%s", conf
->console
.log_path
);
824 if (cgns_supported()) {
825 ret
= setenv("LXC_CGNS_AWARE", "1", 1);
827 SYSERROR("Failed to set environment variable LXC_CGNS_AWARE=1");
830 loglevel
= lxc_log_priority_to_string(lxc_log_get_level());
831 ret
= setenv("LXC_LOG_LEVEL", loglevel
, 1);
833 SYSERROR("Set environment variable LXC_LOG_LEVEL=%s", loglevel
);
835 if (conf
->hooks_version
== 0)
836 ret
= setenv("LXC_HOOK_VERSION", "0", 1);
838 ret
= setenv("LXC_HOOK_VERSION", "1", 1);
840 SYSERROR("Failed to set environment variable LXC_HOOK_VERSION=%u", conf
->hooks_version
);
841 /* End of environment variable setup for hooks. */
843 TRACE("Set environment variables");
845 ret
= run_lxc_hooks(name
, "pre-start", conf
, NULL
);
847 return log_error(-1, "Failed to run lxc.hook.pre-start for container \"%s\"", name
);
848 TRACE("Ran pre-start hooks");
850 ret
= lxc_terminal_parent(conf
);
852 return log_error(-1, "Failed to allocate terminal");
854 /* The signal fd has to be created before forking otherwise if the child
855 * process exits before we setup the signal fd, the event will be lost
856 * and the command will be stuck.
858 handler
->sigfd
= setup_signal_fd(&handler
->oldmask
);
859 if (handler
->sigfd
< 0)
860 return log_error(-1, "Failed to setup SIGCHLD fd handler.");
861 TRACE("Set up signal fd");
863 handler
->cgroup_ops
= cgroup_init(handler
->conf
);
864 if (!handler
->cgroup_ops
) {
865 ERROR("Failed to initialize cgroup driver");
866 goto out_restore_sigmask
;
868 TRACE("Initialized cgroup driver");
870 ret
= lxc_read_seccomp_config(conf
);
872 ERROR("Failed to read seccomp policy");
873 goto out_restore_sigmask
;
875 TRACE("Read seccomp policy");
877 ret
= handler
->lsm_ops
->prepare(handler
->lsm_ops
, conf
, handler
->lxcpath
);
879 ERROR("Failed to initialize LSM");
880 goto out_restore_sigmask
;
882 TRACE("Initialized LSM");
884 INFO("Container \"%s\" is initialized", name
);
885 handler
->monitor_status_fd
= move_fd(status_fd
);
889 (void)pthread_sigmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
894 void lxc_expose_namespace_environment(const struct lxc_handler
*handler
)
896 for (lxc_namespace_t i
= 0; i
< LXC_NS_MAX
; i
++) {
900 if (handler
->nsfd
[i
] < 0)
903 fd_path
= handler
->nsfd_paths
[i
] + strcspn(handler
->nsfd_paths
[i
], "/");
904 ret
= setenv(ns_info
[i
].env_name
, fd_path
, 1);
906 SYSERROR("Failed to set environment variable %s=%s",
907 ns_info
[i
].env_name
, fd_path
);
909 TRACE("Set environment variable %s=%s",
910 ns_info
[i
].env_name
, fd_path
);
914 void lxc_end(struct lxc_handler
*handler
)
917 const char *name
= handler
->name
;
918 struct cgroup_ops
*cgroup_ops
= handler
->cgroup_ops
;
919 struct lxc_state_client
*client
, *nclient
;
921 /* The STOPPING state is there for future cleanup code which can take
924 lxc_set_state(name
, handler
, STOPPING
);
926 /* Passing information to hooks via environment variables. */
927 if (handler
->conf
->hooks_version
> 0)
928 lxc_expose_namespace_environment(handler
);
930 if (handler
->conf
->reboot
> REBOOT_NONE
) {
931 ret
= setenv("LXC_TARGET", "reboot", 1);
933 SYSERROR("Failed to set environment variable: LXC_TARGET=reboot");
936 if (handler
->conf
->reboot
== REBOOT_NONE
) {
937 ret
= setenv("LXC_TARGET", "stop", 1);
939 SYSERROR("Failed to set environment variable: LXC_TARGET=stop");
942 if (handler
->conf
->hooks_version
== 0)
943 ret
= run_lxc_hooks(name
, "stop", handler
->conf
, handler
->hook_argv
);
945 ret
= run_lxc_hooks(name
, "stop", handler
->conf
, NULL
);
947 ERROR("Failed to run \"lxc.hook.stop\" hook");
949 handler
->lsm_ops
->cleanup(handler
->lsm_ops
, handler
->conf
, handler
->lxcpath
);
952 cgroup_ops
->payload_destroy(cgroup_ops
, handler
);
953 cgroup_ops
->monitor_destroy(cgroup_ops
, handler
);
956 put_lxc_rootfs(&handler
->conf
->rootfs
, true);
958 if (handler
->conf
->reboot
== REBOOT_NONE
) {
959 /* For all new state clients simply close the command socket.
960 * This will inform all state clients that the container is
961 * STOPPED and also prevents a race between a open()/close() on
962 * the command socket causing a new process to get ECONNREFUSED
963 * because we haven't yet closed the command socket.
965 close_prot_errno_disarm(handler
->conf
->maincmd_fd
);
966 TRACE("Closed command socket");
968 /* This function will try to connect to the legacy lxc-monitord
969 * state server and only exists for backwards compatibility.
971 lxc_monitor_send_state(name
, STOPPED
, handler
->lxcpath
);
973 /* The command socket is closed so no one can acces the command
974 * socket anymore so there's no need to lock it.
976 handler
->state
= STOPPED
;
977 TRACE("Set container state to \"STOPPED\"");
979 lxc_set_state(name
, handler
, STOPPED
);
980 TRACE("Set container state to \"STOPPED\"");
983 /* Avoid lingering namespace references. */
984 lxc_put_nsfds(handler
);
986 ret
= run_lxc_hooks(name
, "post-stop", handler
->conf
, NULL
);
988 ERROR("Failed to run lxc.hook.post-stop for container \"%s\"", name
);
989 if (handler
->conf
->reboot
> REBOOT_NONE
) {
990 WARN("Container will be stopped instead of rebooted");
991 handler
->conf
->reboot
= REBOOT_NONE
;
993 ret
= setenv("LXC_TARGET", "stop", 1);
995 WARN("Failed to set environment variable: LXC_TARGET=stop");
999 /* Reset mask set by setup_signal_fd. */
1000 ret
= pthread_sigmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
1002 SYSWARN("Failed to restore signal mask");
1004 lxc_terminal_delete(&handler
->conf
->console
);
1005 lxc_delete_tty(&handler
->conf
->ttys
);
1006 close_prot_errno_disarm(handler
->conf
->devpts_fd
);
1008 /* The command socket is now closed, no more state clients can register
1009 * themselves from now on. So free the list of state clients.
1011 list_for_each_entry_safe(client
, nclient
, &handler
->conf
->state_clients
, head
) {
1012 /* Keep state clients that want to be notified about reboots. */
1013 if ((handler
->conf
->reboot
> REBOOT_NONE
) &&
1014 (client
->states
[RUNNING
] == 2))
1017 /* close state client socket */
1018 list_del(&client
->head
);
1019 close(client
->clientfd
);
1023 if (handler
->conf
->ephemeral
== 1 && handler
->conf
->reboot
!= REBOOT_REQ
)
1024 lxc_destroy_container_on_signal(handler
, name
);
1026 lxc_put_handler(handler
);
1029 void lxc_abort(struct lxc_handler
*handler
)
1034 lxc_set_state(handler
->name
, handler
, ABORTING
);
1036 if (handler
->pidfd
>= 0) {
1037 ret
= lxc_raw_pidfd_send_signal(handler
->pidfd
, SIGKILL
, NULL
, 0);
1039 SYSWARN("Failed to send SIGKILL via pidfd %d for process %d",
1040 handler
->pidfd
, handler
->pid
);
1043 if ((!ret
|| errno
!= ESRCH
) && handler
->pid
> 0)
1044 if (kill(handler
->pid
, SIGKILL
))
1045 SYSWARN("Failed to send SIGKILL to %d", handler
->pid
);
1048 ret
= waitpid(-1, &status
, 0);
1052 static int do_start(void *data
)
1054 struct lxc_handler
*handler
= data
;
1055 __lxc_unused __do_close
int data_sock0
= handler
->data_sock
[0],
1056 data_sock1
= handler
->data_sock
[1];
1057 __do_close
int devnull_fd
= -EBADF
, status_fd
= -EBADF
;
1064 lxc_sync_fini_parent(handler
);
1066 if (lxc_abstract_unix_recv_one_fd(data_sock1
, &status_fd
, NULL
, 0) < 0) {
1067 ERROR("Failed to receive status file descriptor from parent process");
1068 goto out_warn_father
;
1071 /* This prctl must be before the synchro, so if the parent dies before
1072 * we set the parent death signal, we will detect its death with the
1073 * synchro right after, otherwise we have a window where the parent can
1074 * exit before we set the pdeath signal leading to a unsupervized
1077 ret
= lxc_set_death_signal(SIGKILL
, handler
->monitor_pid
, status_fd
);
1079 SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL");
1080 goto out_warn_father
;
1083 ret
= lxc_ambient_caps_up();
1085 ERROR("Failed to raise ambient capabilities");
1086 goto out_warn_father
;
1089 ret
= pthread_sigmask(SIG_SETMASK
, &handler
->oldmask
, NULL
);
1091 SYSERROR("Failed to set signal mask");
1092 goto out_warn_father
;
1095 if (!lxc_sync_wait_parent(handler
, START_SYNC_STARTUP
))
1096 goto out_warn_father
;
1098 /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See
1099 * https://github.com/lxc/lxd/issues/1978.
1101 if (handler
->ns_unshare_flags
& CLONE_NEWNET
) {
1102 ret
= unshare(CLONE_NEWNET
);
1104 SYSERROR("Failed to unshare CLONE_NEWNET");
1105 goto out_warn_father
;
1107 INFO("Unshared CLONE_NEWNET");
1110 /* If we are in a new user namespace, become root there to have
1111 * privilege over our namespace.
1113 if (!list_empty(&handler
->conf
->id_map
)) {
1114 if (!handler
->conf
->root_nsuid_map
)
1115 nsuid
= handler
->conf
->init_uid
;
1117 if (!handler
->conf
->root_nsgid_map
)
1118 nsgid
= handler
->conf
->init_gid
;
1120 /* Drop groups only after we switched to a valid gid in the new
1123 if (!lxc_drop_groups() &&
1124 (handler
->am_root
|| errno
!= EPERM
))
1125 goto out_warn_father
;
1127 if (!lxc_switch_uid_gid(nsuid
, nsgid
))
1128 goto out_warn_father
;
1130 ret
= prctl(PR_SET_DUMPABLE
, prctl_arg(1), prctl_arg(0),
1131 prctl_arg(0), prctl_arg(0));
1133 goto out_warn_father
;
1135 /* set{g,u}id() clears deathsignal */
1136 ret
= lxc_set_death_signal(SIGKILL
, handler
->monitor_pid
, status_fd
);
1138 SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL");
1139 goto out_warn_father
;
1143 ret
= access(handler
->lxcpath
, X_OK
);
1145 print_top_failing_dir(handler
->lxcpath
);
1146 goto out_warn_father
;
1149 /* In order to checkpoint restore, we need to have everything in the
1150 * same mount namespace. However, some containers may not have a
1151 * reasonable /dev (in particular, they may not have /dev/null), so we
1152 * can't set init's std fds to /dev/null by opening it from inside the
1155 * If that's the case, fall back to using the host's /dev/null. This
1156 * means that migration won't work, but at least we won't spew output
1157 * where it isn't wanted.
1159 if (handler
->daemonize
&& !handler
->conf
->autodev
) {
1160 char path
[PATH_MAX
];
1162 ret
= strnprintf(path
, sizeof(path
), "%s/dev/null",
1163 handler
->conf
->rootfs
.mount
);
1165 goto out_warn_father
;
1167 ret
= access(path
, F_OK
);
1169 devnull_fd
= open_devnull();
1172 goto out_warn_father
;
1173 WARN("Using /dev/null from the host for container init's standard file descriptors. Migration will not work");
1178 * Tell the parent task it can begin to configure the container and wait
1181 if (!lxc_sync_wake_parent(handler
, START_SYNC_CONFIGURE
))
1184 /* Unshare cgroup namespace after we have setup our cgroups. If we do it
1185 * earlier we end up with a wrong view of /proc/self/cgroup. For
1186 * example, assume we unshare(CLONE_NEWCGROUP) first, and then create
1187 * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then
1188 * /proc/self/cgroup would show us:
1192 * whereas it should actually show
1196 if (handler
->ns_unshare_flags
& CLONE_NEWCGROUP
) {
1197 ret
= unshare(CLONE_NEWCGROUP
);
1199 if (errno
!= EINVAL
) {
1200 SYSERROR("Failed to unshare CLONE_NEWCGROUP");
1201 goto out_warn_father
;
1204 handler
->ns_clone_flags
&= ~CLONE_NEWCGROUP
;
1205 SYSINFO("Kernel does not support CLONE_NEWCGROUP");
1207 INFO("Unshared CLONE_NEWCGROUP");
1211 if (handler
->ns_unshare_flags
& CLONE_NEWTIME
) {
1212 ret
= unshare(CLONE_NEWTIME
);
1214 if (errno
!= EINVAL
) {
1215 SYSERROR("Failed to unshare CLONE_NEWTIME");
1216 goto out_warn_father
;
1219 handler
->ns_clone_flags
&= ~CLONE_NEWTIME
;
1220 SYSINFO("Kernel does not support CLONE_NEWTIME");
1222 __do_close
int timens_fd
= -EBADF
;
1224 INFO("Unshared CLONE_NEWTIME");
1226 if (handler
->conf
->timens
.s_boot
)
1227 ret
= timens_offset_write(CLOCK_BOOTTIME
, handler
->conf
->timens
.s_boot
, 0);
1228 else if (handler
->conf
->timens
.ns_boot
)
1229 ret
= timens_offset_write(CLOCK_BOOTTIME
, 0, handler
->conf
->timens
.ns_boot
);
1231 SYSERROR("Failed to write CLONE_BOOTTIME offset");
1232 goto out_warn_father
;
1234 TRACE("Wrote CLOCK_BOOTTIME offset");
1236 if (handler
->conf
->timens
.s_monotonic
)
1237 ret
= timens_offset_write(CLOCK_MONOTONIC
, handler
->conf
->timens
.s_monotonic
, 0);
1238 else if (handler
->conf
->timens
.ns_monotonic
)
1239 ret
= timens_offset_write(CLOCK_MONOTONIC
, 0, handler
->conf
->timens
.ns_monotonic
);
1241 SYSERROR("Failed to write CLONE_MONOTONIC offset");
1242 goto out_warn_father
;
1244 TRACE("Wrote CLOCK_MONOTONIC offset");
1246 timens_fd
= open("/proc/self/ns/time_for_children", O_RDONLY
| O_CLOEXEC
);
1247 if (timens_fd
< 0) {
1248 SYSERROR("Failed to open \"/proc/self/ns/time_for_children\"");
1249 goto out_warn_father
;
1252 ret
= setns(timens_fd
, CLONE_NEWTIME
);
1254 SYSERROR("Failed to setns(%d(\"/proc/self/ns/time_for_children\"))", timens_fd
);
1255 goto out_warn_father
;
1261 * Add the requested environment variables to the current environment
1262 * to allow them to be used by the various hooks, such as the start
1265 ret
= lxc_set_environment(handler
->conf
);
1267 goto out_warn_father
;
1269 if (!lxc_sync_wait_parent(handler
, START_SYNC_POST_CONFIGURE
))
1270 goto out_warn_father
;
1272 /* Setup the container, ip, names, utsname, ... */
1273 ret
= lxc_setup(handler
);
1275 ERROR("Failed to setup container \"%s\"", handler
->name
);
1276 goto out_warn_father
;
1279 /* Set the label to change to when we exec(2) the container's init. */
1280 ret
= handler
->lsm_ops
->process_label_set(handler
->lsm_ops
, NULL
, handler
->conf
, true);
1282 goto out_warn_father
;
1284 /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it
1285 * before we aren't allowed anymore.
1287 if (handler
->conf
->no_new_privs
) {
1288 ret
= prctl(PR_SET_NO_NEW_PRIVS
, prctl_arg(1), prctl_arg(0),
1289 prctl_arg(0), prctl_arg(0));
1291 SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges");
1292 goto out_warn_father
;
1294 DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges");
1297 /* If we mounted a temporary proc, then unmount it now. */
1298 tmp_proc_unmount(handler
->conf
);
1300 ret
= lxc_seccomp_load(handler
->conf
);
1302 goto out_warn_father
;
1304 ret
= run_lxc_hooks(handler
->name
, "start", handler
->conf
, NULL
);
1306 ERROR("Failed to run lxc.hook.start for container \"%s\"",
1308 goto out_warn_father
;
1311 close_prot_errno_disarm(handler
->sigfd
);
1313 if (handler
->conf
->console
.pty
< 0 && handler
->daemonize
) {
1314 if (devnull_fd
< 0) {
1315 devnull_fd
= open_devnull();
1317 goto out_warn_father
;
1320 ret
= set_stdfds(devnull_fd
);
1322 ERROR("Failed to redirect std{in,out,err} to \"/dev/null\"");
1323 goto out_warn_father
;
1327 close_prot_errno_disarm(devnull_fd
);
1331 if (handler
->conf
->init_cwd
) {
1332 ret
= chdir(handler
->conf
->init_cwd
);
1334 SYSERROR("Could not change directory to \"%s\"",
1335 handler
->conf
->init_cwd
);
1336 goto out_warn_father
;
1340 if (!lxc_sync_barrier_parent(handler
, START_SYNC_CGROUP_LIMITS
))
1341 goto out_warn_father
;
1343 ret
= lxc_sync_fds_child(handler
);
1345 SYSERROR("Failed to sync file descriptors with parent");
1346 goto out_warn_father
;
1349 if (!lxc_sync_wait_parent(handler
, START_SYNC_READY_START
))
1350 goto out_warn_father
;
1352 /* Reset the environment variables the user requested in a clear
1356 /* Don't error out though. */
1358 SYSERROR("Failed to clear environment.");
1360 ret
= lxc_set_environment(handler
->conf
);
1362 goto out_warn_father
;
1364 ret
= putenv("container=lxc");
1366 SYSERROR("Failed to set environment variable: container=lxc");
1367 goto out_warn_father
;
1370 if (handler
->conf
->ttys
.tty_names
) {
1371 ret
= putenv(handler
->conf
->ttys
.tty_names
);
1373 SYSERROR("Failed to set environment variable for container ptys");
1374 goto out_warn_father
;
1378 /* The container has been setup. We can now switch to an unprivileged
1381 new_uid
= handler
->conf
->init_uid
;
1382 new_gid
= handler
->conf
->init_gid
;
1384 /* Avoid unnecessary syscalls. */
1385 if (new_uid
== nsuid
)
1386 new_uid
= LXC_INVALID_UID
;
1388 if (new_gid
== nsgid
)
1389 new_gid
= LXC_INVALID_GID
;
1391 /* Make sure that the processes STDIO is correctly owned by the user that we are switching to */
1392 ret
= fix_stdio_permissions(new_uid
);
1394 WARN("Failed to ajust stdio permissions");
1396 /* If we are in a new user namespace we already dropped all groups when
1397 * we switched to root in the new user namespace further above. Only
1398 * drop groups if we can, so ensure that we have necessary privilege.
1400 if (list_empty(&handler
->conf
->id_map
)) {
1402 if (lxc_proc_cap_is_set(CAP_SETGID
, CAP_EFFECTIVE
))
1405 if (handler
->conf
->init_groups
.size
> 0) {
1406 if (!lxc_setgroups(handler
->conf
->init_groups
.list
,
1407 handler
->conf
->init_groups
.size
))
1408 goto out_warn_father
;
1410 if (!lxc_drop_groups())
1411 goto out_warn_father
;
1416 if (!lxc_switch_uid_gid(new_uid
, new_gid
))
1417 goto out_warn_father
;
1419 ret
= lxc_ambient_caps_down();
1421 ERROR("Failed to clear ambient capabilities");
1422 goto out_warn_father
;
1425 if (handler
->conf
->monitor_signal_pdeath
!= SIGKILL
) {
1426 ret
= lxc_set_death_signal(handler
->conf
->monitor_signal_pdeath
,
1427 handler
->monitor_pid
, status_fd
);
1429 SYSERROR("Failed to set PR_SET_PDEATHSIG to %d",
1430 handler
->conf
->monitor_signal_pdeath
);
1431 goto out_warn_father
;
1436 * After this call, we are in error because this ops should not return
1439 handler
->ops
->start(handler
, handler
->data
);
1443 * We want the parent to know something went wrong, so we return a
1444 * special error code.
1446 lxc_sync_wake_parent(handler
, SYNC_ERROR
);
1452 int resolve_clone_flags(struct lxc_handler
*handler
)
1455 struct lxc_conf
*conf
= handler
->conf
;
1456 bool wants_timens
= conf
->timens
.s_boot
|| conf
->timens
.ns_boot
||
1457 conf
->timens
.s_monotonic
|| conf
->timens
.ns_monotonic
;
1459 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1460 if (conf
->ns_keep
) {
1461 if (!(conf
->ns_keep
& ns_info
[i
].clone_flag
))
1462 handler
->ns_clone_flags
|= ns_info
[i
].clone_flag
;
1463 } else if (conf
->ns_clone
) {
1464 if ((conf
->ns_clone
& ns_info
[i
].clone_flag
))
1465 handler
->ns_clone_flags
|= ns_info
[i
].clone_flag
;
1467 if (i
== LXC_NS_USER
&& list_empty(&handler
->conf
->id_map
))
1470 if (i
== LXC_NS_NET
&& lxc_requests_empty_network(handler
))
1473 if (i
== LXC_NS_CGROUP
&& !cgns_supported())
1476 if (i
== LXC_NS_TIME
&& !wants_timens
)
1479 handler
->ns_clone_flags
|= ns_info
[i
].clone_flag
;
1482 if (!conf
->ns_share
[i
])
1485 handler
->ns_clone_flags
&= ~ns_info
[i
].clone_flag
;
1486 TRACE("Sharing %s namespace", ns_info
[i
].proc_name
);
1489 if (wants_timens
&& (conf
->ns_keep
& ns_info
[LXC_NS_TIME
].clone_flag
))
1490 return log_trace_errno(-1, EINVAL
, "Requested to keep time namespace while also specifying offsets");
1492 /* Deal with namespaces that are unshared. */
1493 if (handler
->ns_clone_flags
& CLONE_NEWTIME
)
1494 handler
->ns_unshare_flags
|= CLONE_NEWTIME
;
1496 if (!pure_unified_layout(handler
->cgroup_ops
) && handler
->ns_clone_flags
& CLONE_NEWCGROUP
)
1497 handler
->ns_unshare_flags
|= CLONE_NEWCGROUP
;
1499 if ((handler
->ns_clone_flags
& (CLONE_NEWNET
| CLONE_NEWUSER
)) ==
1500 (CLONE_NEWNET
| CLONE_NEWUSER
))
1501 handler
->ns_unshare_flags
|= CLONE_NEWNET
;
1503 /* Deal with namespaces that are spawned. */
1504 handler
->ns_on_clone_flags
= handler
->ns_clone_flags
& ~handler
->ns_unshare_flags
;
1506 handler
->clone_flags
= handler
->ns_on_clone_flags
| CLONE_PIDFD
;
1511 /* Note that this function is used with clone(CLONE_VM). Some glibc versions
1512 * used to reset the pid/tid to -1 when CLONE_VM was used without CLONE_THREAD.
1513 * But since the memory between parent and child is shared on CLONE_VM this
1514 * would invalidate the getpid() cache that glibc used to maintain and so
1515 * getpid() in the child would return the parent's pid. This is all fixed in
1516 * newer glibc versions where the getpid() cache is removed and the pid/tid is
1517 * not reset anymore.
1518 * However, if for whatever reason you - dear committer - somehow need to get the
1519 * pid of the placeholder intermediate process for do_share_ns() you need to
1520 * call lxc_raw_getpid(). The next lxc_raw_clone() call does not employ
1521 * CLONE_VM and will be fine.
1523 static inline int do_share_ns(void *arg
)
1526 struct lxc_handler
*handler
= arg
;
1528 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1529 if (handler
->nsfd
[i
] < 0)
1532 ret
= setns(handler
->nsfd
[i
], 0);
1535 * Note that joining a user and/or mount namespace
1536 * requires the process is not multithreaded otherwise
1537 * setns() will fail here.
1539 SYSERROR("Failed to inherit %s namespace",
1540 ns_info
[i
].proc_name
);
1544 DEBUG("Inherited %s namespace", ns_info
[i
].proc_name
);
1547 flags
= handler
->ns_on_clone_flags
;
1548 flags
|= CLONE_PARENT
;
1549 handler
->pid
= lxc_raw_clone_cb(do_start
, handler
, CLONE_PIDFD
| flags
,
1551 if (handler
->pid
< 0)
1557 static int core_scheduling(struct lxc_handler
*handler
)
1559 struct lxc_conf
*conf
= handler
->conf
;
1562 if (!conf
->sched_core
)
1563 return log_trace(0, "No new core scheduling domain requested");
1565 if (!(handler
->ns_clone_flags
& CLONE_NEWPID
))
1566 return syserror_set(-EINVAL
, "Core scheduling currently requires a separate pid namespace");
1568 ret
= core_scheduling_cookie_create_threadgroup(handler
->pid
);
1570 if (ret
== -ENODEV
) {
1571 INFO("The kernel doesn't support or doesn't use simultaneous multithreading (SMT)");
1572 conf
->sched_core
= false;
1576 return syserror("The kernel does not support core scheduling");
1578 return syserror("Failed to create new core scheduling domain");
1581 ret
= core_scheduling_cookie_get(handler
->pid
, &conf
->sched_core_cookie
);
1582 if (ret
|| !core_scheduling_cookie_valid(conf
->sched_core_cookie
))
1583 return syserror("Failed to retrieve core scheduling domain cookie");
1585 TRACE("Created new core scheduling domain with cookie %llu",
1586 (llu
)conf
->sched_core_cookie
);
1591 /* lxc_spawn() performs crucial setup tasks and clone()s the new process which
1592 * exec()s the requested container binary.
1593 * Note that lxc_spawn() runs in the parent namespaces. Any operations performed
1594 * right here should be double checked if they'd pose a security risk. (For
1595 * example, any {u}mount() operations performed here will be reflected on the
1598 static int lxc_spawn(struct lxc_handler
*handler
)
1600 __do_close
int data_sock0
= -EBADF
, data_sock1
= -EBADF
;
1603 bool wants_to_map_ids
;
1604 struct list_head
*id_map
;
1605 const char *name
= handler
->name
;
1606 const char *lxcpath
= handler
->lxcpath
;
1607 bool share_ns
= false;
1608 struct lxc_conf
*conf
= handler
->conf
;
1609 struct cgroup_ops
*cgroup_ops
= handler
->cgroup_ops
;
1611 id_map
= &conf
->id_map
;
1612 wants_to_map_ids
= !list_empty(id_map
);
1614 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1615 if (!conf
->ns_share
[i
])
1618 handler
->nsfd
[i
] = lxc_inherit_namespace(conf
->ns_share
[i
], lxcpath
, ns_info
[i
].proc_name
);
1619 if (handler
->nsfd
[i
] < 0)
1625 if (!lxc_sync_init(handler
))
1628 ret
= socketpair(AF_UNIX
, SOCK_STREAM
| SOCK_CLOEXEC
, 0,
1629 handler
->data_sock
);
1632 data_sock0
= handler
->data_sock
[0];
1633 data_sock1
= handler
->data_sock
[1];
1635 ret
= resolve_clone_flags(handler
);
1639 if (handler
->ns_clone_flags
& CLONE_NEWNET
) {
1640 ret
= lxc_find_gateway_addresses(handler
);
1642 ERROR("Failed to find gateway addresses");
1647 if (!cgroup_ops
->payload_create(cgroup_ops
, handler
)) {
1648 ERROR("Failed creating cgroups");
1649 goto out_delete_net
;
1652 /* Create a process in a new set of namespaces. */
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 lxc_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 handler
->clone_flags
&= ~(CLONE_INTO_CGROUP
| CLONE_NEWCGROUP
);
1698 handler
->ns_on_clone_flags
&= ~CLONE_NEWCGROUP
;
1699 handler
->ns_unshare_flags
|= CLONE_NEWCGROUP
;
1701 clone_args
.flags
= handler
->clone_flags
;
1703 handler
->pid
= lxc_clone3(&clone_args
, CLONE_ARGS_SIZE_VER0
);
1704 } else if (cgroup_fd
>= 0) {
1705 TRACE("Spawned container directly into target cgroup via cgroup2 fd %d", cgroup_fd
);
1708 /* Kernel might be too old for clone3(). */
1709 if (handler
->pid
< 0) {
1710 SYSTRACE("Failed to spawn container via clone3()");
1713 * In contrast to all other architectures arm64 verifies that
1714 * the argument we use to retrieve the pidfd with is
1715 * initialized to 0. But we need to be able to initialize it to
1716 * a negative value such as our customary -EBADF so we can
1717 * detect whether this kernel supports pidfds. If the syscall
1718 * returns and the pidfd variable is set to something >= 0 then
1719 * we know this is a kernel supporting pidfds. But if we can't
1720 * set it to -EBADF then this won't work since 0 is a valid
1721 * file descriptor too. And since legacy clone silently ignores
1722 * unknown flags we are left without any way to detect support
1723 * for pidfds. So let's special-case arm64 to not fail starting
1726 #if defined(__aarch64__)
1727 handler
->pid
= lxc_raw_legacy_clone(handler
->clone_flags
& ~CLONE_PIDFD
, NULL
);
1729 handler
->pid
= lxc_raw_legacy_clone(handler
->clone_flags
, &handler
->pidfd
);
1733 if (handler
->pid
< 0) {
1734 SYSERROR(LXC_CLONE_ERROR
);
1735 goto out_delete_net
;
1738 if (handler
->pid
== 0) {
1739 (void)do_start(handler
);
1740 _exit(EXIT_FAILURE
);
1743 if (handler
->pidfd
< 0)
1744 handler
->clone_flags
&= ~CLONE_PIDFD
;
1745 TRACE("Cloned child process %d", handler
->pid
);
1747 ret
= core_scheduling(handler
);
1749 goto out_delete_net
;
1751 /* Verify that we can actually make use of pidfds. */
1752 if (!lxc_can_use_pidfd(handler
->pidfd
))
1753 close_prot_errno_disarm(handler
->pidfd
);
1755 ret
= strnprintf(pidstr
, 20, "%d", handler
->pid
);
1757 goto out_delete_net
;
1759 ret
= setenv("LXC_PID", pidstr
, 1);
1761 SYSERROR("Failed to set environment variable: LXC_PID=%s", pidstr
);
1763 for (i
= 0; i
< LXC_NS_MAX
; i
++)
1764 if (handler
->ns_on_clone_flags
& ns_info
[i
].clone_flag
)
1765 INFO("Cloned %s", ns_info
[i
].flag_name
);
1767 if (!lxc_try_preserve_namespaces(handler
, handler
->ns_on_clone_flags
)) {
1768 ERROR("Failed to preserve cloned namespaces for lxc.hook.stop");
1769 goto out_delete_net
;
1772 lxc_sync_fini_child(handler
);
1774 if (lxc_abstract_unix_send_fds(handler
->data_sock
[0], &handler
->monitor_status_fd
, 1, NULL
, 0) < 0) {
1775 ERROR("Failed to send status file descriptor to child process");
1776 goto out_delete_net
;
1778 close_prot_errno_disarm(handler
->monitor_status_fd
);
1780 /* Map the container uids. The container became an invalid userid the
1781 * moment it was cloned with CLONE_NEWUSER. This call doesn't change
1782 * anything immediately, but allows the container to setuid(0) (0 being
1783 * mapped to something else on the host.) later to become a valid uid
1786 if (wants_to_map_ids
) {
1787 if (!handler
->conf
->ns_share
[LXC_NS_USER
] &&
1788 (handler
->conf
->ns_keep
& CLONE_NEWUSER
) == 0) {
1789 ret
= lxc_map_ids(id_map
, handler
->pid
);
1791 ERROR("Failed to set up id mapping.");
1792 goto out_delete_net
;
1797 if (!cgroup_ops
->setup_limits_legacy(cgroup_ops
, handler
->conf
, false)) {
1798 ERROR("Failed to setup cgroup limits for container \"%s\"", name
);
1799 goto out_delete_net
;
1802 if (!cgroup_ops
->payload_delegate_controllers(cgroup_ops
)) {
1803 ERROR("Failed to delegate controllers to payload cgroup");
1804 goto out_delete_net
;
1807 if (!cgroup_ops
->payload_enter(cgroup_ops
, handler
)) {
1808 ERROR("Failed to enter cgroups");
1809 goto out_delete_net
;
1812 if (!cgroup_ops
->setup_limits(cgroup_ops
, handler
)) {
1813 ERROR("Failed to setup cgroup limits for container \"%s\"", name
);
1814 goto out_delete_net
;
1817 if (!cgroup_ops
->chown(cgroup_ops
, handler
->conf
))
1818 goto out_delete_net
;
1820 if (!lxc_sync_barrier_child(handler
, START_SYNC_STARTUP
))
1821 goto out_delete_net
;
1823 /* If not done yet, we're now ready to preserve the network namespace */
1824 if (handler
->nsfd
[LXC_NS_NET
] < 0) {
1825 ret
= lxc_try_preserve_namespace(handler
, LXC_NS_NET
, "net");
1827 if (ret
!= -ENOENT
) {
1828 SYSERROR("Failed to preserve net namespace");
1829 goto out_delete_net
;
1833 ret
= lxc_netns_set_nsid(handler
->nsfd
[LXC_NS_NET
]);
1835 SYSWARN("Failed to allocate new network namespace id");
1837 TRACE("Allocated new network namespace id");
1839 /* Create the network configuration. */
1840 if (handler
->ns_clone_flags
& CLONE_NEWNET
) {
1841 ret
= lxc_create_network(handler
);
1843 ERROR("Failed to create the network");
1844 goto out_delete_net
;
1848 ret
= setup_proc_filesystem(conf
, handler
->pid
);
1850 ERROR("Failed to setup procfs limits");
1851 goto out_delete_net
;
1854 ret
= setup_resource_limits(conf
, handler
->pid
);
1856 ERROR("Failed to setup resource limits");
1857 goto out_delete_net
;
1860 /* Tell the child to continue its initialization. */
1861 if (!lxc_sync_wake_child(handler
, START_SYNC_POST_CONFIGURE
))
1862 goto out_delete_net
;
1864 ret
= lxc_rootfs_prepare_parent(handler
);
1866 ERROR("Failed to prepare rootfs");
1867 goto out_delete_net
;
1870 if (handler
->ns_clone_flags
& CLONE_NEWNET
) {
1871 ret
= lxc_network_send_to_child(handler
);
1873 SYSERROR("Failed to send veth names to child");
1874 goto out_delete_net
;
1878 if (!lxc_sync_wait_child(handler
, START_SYNC_IDMAPPED_MOUNTS
))
1879 goto out_delete_net
;
1881 ret
= lxc_idmapped_mounts_parent(handler
);
1883 ERROR("Failed to setup mount entries");
1884 goto out_delete_net
;
1887 if (!lxc_sync_wait_child(handler
, START_SYNC_CGROUP_LIMITS
))
1888 goto out_delete_net
;
1891 * With isolation the limiting devices cgroup was already setup, so
1892 * only setup devices here if we have no namespace directory.
1894 if (!handler
->conf
->cgroup_meta
.namespace_dir
&&
1895 !cgroup_ops
->setup_limits_legacy(cgroup_ops
, handler
->conf
, true)) {
1896 ERROR("Failed to setup legacy device cgroup controller limits");
1897 goto out_delete_net
;
1899 TRACE("Set up legacy device cgroup controller limits");
1901 if (!cgroup_ops
->devices_activate(cgroup_ops
, handler
)) {
1902 ERROR("Failed to setup cgroup2 device controller limits");
1903 goto out_delete_net
;
1905 TRACE("Set up cgroup2 device controller limits");
1907 cgroup_ops
->finalize(cgroup_ops
);
1908 TRACE("Finished setting up cgroups");
1910 /* Run any host-side start hooks */
1911 ret
= run_lxc_hooks(name
, "start-host", conf
, NULL
);
1913 ERROR("Failed to run lxc.hook.start-host");
1914 goto out_delete_net
;
1917 if (!lxc_sync_wake_child(handler
, START_SYNC_FDS
))
1918 goto out_delete_net
;
1920 if (handler
->ns_unshare_flags
& CLONE_NEWCGROUP
) {
1921 /* Now we're ready to preserve the cgroup namespace */
1922 ret
= lxc_try_preserve_namespace(handler
, LXC_NS_CGROUP
, "cgroup");
1924 if (ret
!= -ENOENT
) {
1925 SYSERROR("Failed to preserve cgroup namespace");
1926 goto out_delete_net
;
1931 if (handler
->ns_unshare_flags
& CLONE_NEWTIME
) {
1932 /* Now we're ready to preserve the time namespace */
1933 ret
= lxc_try_preserve_namespace(handler
, LXC_NS_TIME
, "time");
1935 if (ret
!= -ENOENT
) {
1936 SYSERROR("Failed to preserve time namespace");
1937 goto out_delete_net
;
1942 ret
= lxc_sync_fds_parent(handler
);
1944 SYSERROR("Failed to sync file descriptors with child");
1945 goto out_delete_net
;
1948 ret
= lxc_terminal_setup(conf
);
1950 SYSERROR("Failed to create console");
1951 goto out_delete_net
;
1955 * Tell the child to complete its initialization and wait for it to
1956 * exec or return an error. (The child will never return
1957 * START_SYNC_READY_START+1. It will either close the sync pipe,
1958 * causing lxc_sync_barrier_child to return success, or return a
1959 * different value, causing us to error out).
1961 if (!lxc_sync_barrier_child(handler
, START_SYNC_READY_START
))
1962 goto out_delete_net
;
1964 /* Now all networks are created, network devices are moved into place,
1965 * and the correct names and ifindices in the respective namespaces have
1966 * been recorded. The corresponding structs have now all been filled. So
1967 * log them for debugging purposes.
1969 lxc_log_configured_netdevs(conf
);
1971 ret
= handler
->ops
->post_start(handler
, handler
->data
);
1975 ret
= lxc_set_state(name
, handler
, RUNNING
);
1977 ERROR("Failed to set state to \"%s\"", lxc_state2str(RUNNING
));
1981 lxc_sync_fini(handler
);
1986 if (handler
->ns_clone_flags
& CLONE_NEWNET
)
1987 lxc_delete_network(handler
);
1993 lxc_sync_fini(handler
);
1998 int __lxc_start(struct lxc_handler
*handler
, struct lxc_operations
*ops
,
1999 void *data
, const char *lxcpath
, bool daemonize
, int *error_num
)
2002 const char *name
= handler
->name
;
2003 struct lxc_conf
*conf
= handler
->conf
;
2004 struct cgroup_ops
*cgroup_ops
;
2006 ret
= lxc_init(name
, handler
);
2008 ERROR("Failed to initialize container \"%s\"", name
);
2012 handler
->data
= data
;
2013 handler
->daemonize
= daemonize
;
2014 cgroup_ops
= handler
->cgroup_ops
;
2016 if (!attach_block_device(handler
->conf
)) {
2017 ERROR("Failed to attach block device");
2022 if (!cgroup_ops
->monitor_create(cgroup_ops
, handler
)) {
2023 ERROR("Failed to create monitor cgroup");
2028 if (!cgroup_ops
->monitor_delegate_controllers(cgroup_ops
)) {
2029 ERROR("Failed to delegate controllers to monitor cgroup");
2034 if (!cgroup_ops
->monitor_enter(cgroup_ops
, handler
)) {
2035 ERROR("Failed to enter monitor cgroup");
2040 /* If the rootfs is not a blockdev, prevent the container from marking
2042 * If the container is unprivileged then skip rootfs pinning.
2044 ret
= lxc_rootfs_init(conf
, !list_empty(&conf
->id_map
));
2046 ERROR("Failed to handle rootfs pinning for container \"%s\"", handler
->name
);
2051 if (geteuid() == 0 && !list_empty(&conf
->id_map
)) {
2053 * Most filesystems can't be mounted inside a userns so handle them here.
2055 if (rootfs_is_blockdev(conf
)) {
2056 ret
= unshare(CLONE_NEWNS
);
2058 ERROR("Failed to unshare CLONE_NEWNS");
2061 INFO("Unshared CLONE_NEWNS");
2063 ret
= lxc_setup_rootfs_prepare_root(conf
, name
, lxcpath
);
2065 ERROR("Error setting up rootfs mount as root before spawn");
2068 INFO("Set up container rootfs as host root");
2072 ret
= lxc_spawn(handler
);
2074 ERROR("Failed to spawn container \"%s\"", name
);
2075 goto out_detach_blockdev
;
2078 handler
->conf
->reboot
= REBOOT_NONE
;
2080 ret
= lxc_poll(name
, handler
);
2082 ERROR("LXC mainloop exited with error: %d", ret
);
2083 goto out_delete_network
;
2086 if (!handler
->init_died
&& handler
->pid
> 0) {
2087 ERROR("Child process is not killed");
2089 goto out_delete_network
;
2092 status
= lxc_wait_for_pid_status(handler
->pid
);
2094 SYSERROR("Failed to retrieve status for %d", handler
->pid
);
2096 /* If the child process exited but was not signaled, it didn't call
2097 * reboot. This should mean it was an lxc-execute which simply exited.
2098 * In any case, treat it as a 'halt'.
2100 if (WIFSIGNALED(status
)) {
2101 int signal_nr
= WTERMSIG(status
);
2103 case SIGINT
: /* halt */
2104 DEBUG("%s(%d) - Container \"%s\" is halting", signal_name(signal_nr
), signal_nr
, name
);
2106 case SIGHUP
: /* reboot */
2107 DEBUG("%s(%d) - Container \"%s\" is rebooting", signal_name(signal_nr
), signal_nr
, name
);
2108 handler
->conf
->reboot
= REBOOT_REQ
;
2110 case SIGSYS
: /* seccomp */
2111 DEBUG("%s(%d) - Container \"%s\" violated its seccomp policy", signal_name(signal_nr
), signal_nr
, name
);
2114 DEBUG("%s(%d) - Container \"%s\" init exited", signal_name(signal_nr
), signal_nr
, name
);
2119 ret
= lxc_restore_phys_nics_to_netns(handler
);
2121 ERROR("Failed to move physical network devices back to parent network namespace");
2123 lxc_monitor_send_exit_code(name
, status
, handler
->lxcpath
);
2124 lxc_error_set_and_log(handler
->pid
, status
);
2126 *error_num
= handler
->exit_status
;
2128 lxc_delete_network(handler
);
2129 detach_block_device(handler
->conf
);
2138 out_detach_blockdev
:
2140 detach_block_device(handler
->conf
);
2146 lxc_restore_phys_nics_to_netns(handler
);
2147 lxc_delete_network(handler
);
2148 detach_block_device(handler
->conf
);
2157 static int start(struct lxc_handler
*handler
, void* data
)
2159 struct start_args
*arg
= data
;
2161 NOTICE("Exec'ing \"%s\"", arg
->argv
[0]);
2163 execvp(arg
->argv
[0], arg
->argv
);
2164 SYSERROR("Failed to exec \"%s\"", arg
->argv
[0]);
2168 static int post_start(struct lxc_handler
*handler
, void* data
)
2170 struct start_args
*arg
= data
;
2172 NOTICE("Started \"%s\" with pid \"%d\"", arg
->argv
[0], handler
->pid
);
2176 static struct lxc_operations start_ops
= {
2178 .post_start
= post_start
2181 int lxc_start(char *const argv
[], struct lxc_handler
*handler
,
2182 const char *lxcpath
, bool daemonize
, int *error_num
)
2184 struct start_args start_arg
= {
2188 TRACE("Doing lxc_start");
2189 return __lxc_start(handler
, &start_ops
, &start_arg
, lxcpath
, daemonize
, error_num
);
2192 static void lxc_destroy_container_on_signal(struct lxc_handler
*handler
,
2195 char destroy
[PATH_MAX
];
2196 struct lxc_container
*c
;
2200 if (handler
->conf
->rootfs
.path
&& handler
->conf
->rootfs
.mount
) {
2201 bret
= do_destroy_container(handler
);
2203 ERROR("Error destroying rootfs for container \"%s\"", name
);
2207 INFO("Destroyed rootfs for container \"%s\"", name
);
2209 ret
= strnprintf(destroy
, sizeof(destroy
), "%s/%s", handler
->lxcpath
, name
);
2211 ERROR("Error destroying directory for container \"%s\"", name
);
2215 c
= lxc_container_new(name
, handler
->lxcpath
);
2217 if (container_disk_lock(c
)) {
2218 INFO("Could not update lxc_snapshots file");
2219 lxc_container_put(c
);
2221 mod_all_rdeps(c
, false);
2222 container_disk_unlock(c
);
2223 lxc_container_put(c
);
2227 if (!handler
->am_root
)
2228 ret
= userns_exec_full(handler
->conf
, lxc_rmdir_onedev_wrapper
,
2229 destroy
, "lxc_rmdir_onedev_wrapper");
2231 ret
= lxc_rmdir_onedev(destroy
, NULL
);
2234 ERROR("Error destroying directory for container \"%s\"", name
);
2237 INFO("Destroyed directory for container \"%s\"", name
);
2240 static int lxc_rmdir_onedev_wrapper(void *data
)
2242 char *arg
= (char *) data
;
2243 return lxc_rmdir_onedev(arg
, NULL
);
2246 static bool do_destroy_container(struct lxc_handler
*handler
)
2250 if (!handler
->am_root
) {
2251 ret
= userns_exec_full(handler
->conf
, storage_destroy_wrapper
,
2252 handler
->conf
, "storage_destroy_wrapper");
2259 return storage_destroy(handler
->conf
);