]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * lxc: linux Container library | |
3 | * | |
4 | * (C) Copyright IBM Corp. 2007, 2008 | |
5 | * | |
6 | * Authors: | |
7 | * Daniel Lezcano <daniel.lezcano at free.fr> | |
8 | * | |
9 | * This library is free software; you can redistribute it and/or | |
10 | * modify it under the terms of the GNU Lesser General Public | |
11 | * License as published by the Free Software Foundation; either | |
12 | * version 2.1 of the License, or (at your option) any later version. | |
13 | * | |
14 | * This library is distributed in the hope that it will be useful, | |
15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
17 | * Lesser General Public License for more details. | |
18 | * | |
19 | * You should have received a copy of the GNU Lesser General Public | |
20 | * License along with this library; if not, write to the Free Software | |
21 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | |
22 | */ | |
23 | ||
24 | #define _GNU_SOURCE | |
25 | #include "config.h" | |
26 | ||
27 | #include <alloca.h> | |
28 | #include <dirent.h> | |
29 | #include <errno.h> | |
30 | #include <fcntl.h> | |
31 | #include <grp.h> | |
32 | #include <poll.h> | |
33 | #include <signal.h> | |
34 | #include <stdio.h> | |
35 | #include <stdlib.h> | |
36 | #include <string.h> | |
37 | #include <unistd.h> | |
38 | #include <sys/file.h> | |
39 | #include <sys/mount.h> | |
40 | #include <sys/param.h> | |
41 | #include <sys/prctl.h> | |
42 | #include <sys/socket.h> | |
43 | #include <sys/stat.h> | |
44 | #include <sys/syscall.h> | |
45 | #include <sys/types.h> | |
46 | #include <sys/un.h> | |
47 | #include <sys/wait.h> | |
48 | ||
49 | #if HAVE_LIBCAP | |
50 | #include <sys/capability.h> | |
51 | #endif | |
52 | ||
53 | #ifndef HAVE_DECL_PR_CAPBSET_DROP | |
54 | #define PR_CAPBSET_DROP 24 | |
55 | #endif | |
56 | ||
57 | #ifndef HAVE_DECL_PR_SET_NO_NEW_PRIVS | |
58 | #define PR_SET_NO_NEW_PRIVS 38 | |
59 | #endif | |
60 | ||
61 | #ifndef HAVE_DECL_PR_GET_NO_NEW_PRIVS | |
62 | #define PR_GET_NO_NEW_PRIVS 39 | |
63 | #endif | |
64 | ||
65 | #include "af_unix.h" | |
66 | #include "bdev.h" | |
67 | #include "caps.h" | |
68 | #include "cgroup.h" | |
69 | #include "commands.h" | |
70 | #include "commands_utils.h" | |
71 | #include "conf.h" | |
72 | #include "console.h" | |
73 | #include "error.h" | |
74 | #include "log.h" | |
75 | #include "lxccontainer.h" | |
76 | #include "lxclock.h" | |
77 | #include "lxcseccomp.h" | |
78 | #include "lxcutmp.h" | |
79 | #include "mainloop.h" | |
80 | #include "monitor.h" | |
81 | #include "namespace.h" | |
82 | #include "start.h" | |
83 | #include "storage_utils.h" | |
84 | #include "sync.h" | |
85 | #include "utils.h" | |
86 | #include "lsm/lsm.h" | |
87 | ||
88 | lxc_log_define(lxc_start, lxc); | |
89 | ||
90 | extern void mod_all_rdeps(struct lxc_container *c, bool inc); | |
91 | static bool do_destroy_container(struct lxc_conf *conf); | |
92 | static int lxc_rmdir_onedev_wrapper(void *data); | |
93 | static void lxc_destroy_container_on_signal(struct lxc_handler *handler, | |
94 | const char *name); | |
95 | ||
96 | static void print_top_failing_dir(const char *path) | |
97 | { | |
98 | size_t len = strlen(path); | |
99 | char *copy = alloca(len + 1), *p, *e, saved; | |
100 | strcpy(copy, path); | |
101 | ||
102 | p = copy; | |
103 | e = copy + len; | |
104 | while (p < e) { | |
105 | while (p < e && *p == '/') | |
106 | p++; | |
107 | while (p < e && *p != '/') | |
108 | p++; | |
109 | saved = *p; | |
110 | *p = '\0'; | |
111 | if (access(copy, X_OK)) { | |
112 | SYSERROR("Could not access %s. Please grant it x " | |
113 | "access, or add an ACL for the container " | |
114 | "root.", copy); | |
115 | return; | |
116 | } | |
117 | *p = saved; | |
118 | } | |
119 | } | |
120 | ||
121 | static void close_ns(int ns_fd[LXC_NS_MAX]) | |
122 | { | |
123 | int i; | |
124 | ||
125 | for (i = 0; i < LXC_NS_MAX; i++) { | |
126 | if (ns_fd[i] > -1) { | |
127 | close(ns_fd[i]); | |
128 | ns_fd[i] = -1; | |
129 | } | |
130 | } | |
131 | } | |
132 | ||
133 | /* preserve_ns: open /proc/@pid/ns/@ns for each namespace specified | |
134 | * in clone_flags. | |
135 | * Return true on success, false on failure. | |
136 | */ | |
137 | static bool preserve_ns(int ns_fd[LXC_NS_MAX], int clone_flags, pid_t pid) | |
138 | { | |
139 | int i, ret; | |
140 | ||
141 | for (i = 0; i < LXC_NS_MAX; i++) | |
142 | ns_fd[i] = -1; | |
143 | ||
144 | ret = lxc_preserve_ns(pid, ""); | |
145 | if (ret < 0) { | |
146 | SYSERROR("Kernel does not support attaching to namespaces."); | |
147 | return false; | |
148 | } else { | |
149 | close(ret); | |
150 | } | |
151 | ||
152 | for (i = 0; i < LXC_NS_MAX; i++) { | |
153 | if ((clone_flags & ns_info[i].clone_flag) == 0) | |
154 | continue; | |
155 | ns_fd[i] = lxc_preserve_ns(pid, ns_info[i].proc_name); | |
156 | if (ns_fd[i] < 0) | |
157 | goto error; | |
158 | } | |
159 | ||
160 | return true; | |
161 | ||
162 | error: | |
163 | if (errno == ENOENT) | |
164 | SYSERROR("Kernel does not support attaching to %s namespaces.", ns_info[i].proc_name); | |
165 | else | |
166 | SYSERROR("Failed to open file descriptor for %s namespace: %s.", ns_info[i].proc_name, strerror(errno)); | |
167 | close_ns(ns_fd); | |
168 | return false; | |
169 | } | |
170 | ||
171 | static int attach_ns(const int ns_fd[LXC_NS_MAX]) { | |
172 | int i; | |
173 | ||
174 | for (i = 0; i < LXC_NS_MAX; i++) { | |
175 | if (ns_fd[i] < 0) | |
176 | continue; | |
177 | ||
178 | if (setns(ns_fd[i], 0) != 0) | |
179 | goto error; | |
180 | } | |
181 | return 0; | |
182 | ||
183 | error: | |
184 | SYSERROR("Failed to attach %s namespace.", ns_info[i].proc_name); | |
185 | return -1; | |
186 | } | |
187 | ||
188 | static int match_fd(int fd) | |
189 | { | |
190 | return (fd == 0 || fd == 1 || fd == 2); | |
191 | } | |
192 | ||
193 | int lxc_check_inherited(struct lxc_conf *conf, bool closeall, | |
194 | int *fds_to_ignore, size_t len_fds) | |
195 | { | |
196 | struct dirent *direntp; | |
197 | int fd, fddir; | |
198 | size_t i; | |
199 | DIR *dir; | |
200 | ||
201 | if (conf && conf->close_all_fds) | |
202 | closeall = true; | |
203 | ||
204 | restart: | |
205 | dir = opendir("/proc/self/fd"); | |
206 | if (!dir) { | |
207 | WARN("Failed to open directory: %s.", strerror(errno)); | |
208 | return -1; | |
209 | } | |
210 | ||
211 | fddir = dirfd(dir); | |
212 | ||
213 | while ((direntp = readdir(dir))) { | |
214 | if (!direntp) | |
215 | break; | |
216 | ||
217 | if (!strcmp(direntp->d_name, ".")) | |
218 | continue; | |
219 | ||
220 | if (!strcmp(direntp->d_name, "..")) | |
221 | continue; | |
222 | ||
223 | if (lxc_safe_int(direntp->d_name, &fd) < 0) { | |
224 | INFO("Could not parse file descriptor for: %s", direntp->d_name); | |
225 | continue; | |
226 | } | |
227 | ||
228 | for (i = 0; i < len_fds; i++) | |
229 | if (fds_to_ignore[i] == fd) | |
230 | break; | |
231 | ||
232 | if (fd == fddir || fd == lxc_log_fd || | |
233 | (i < len_fds && fd == fds_to_ignore[i])) | |
234 | continue; | |
235 | ||
236 | if (current_config && fd == current_config->logfd) | |
237 | continue; | |
238 | ||
239 | if (match_fd(fd)) | |
240 | continue; | |
241 | ||
242 | if (closeall) { | |
243 | close(fd); | |
244 | closedir(dir); | |
245 | INFO("Closed inherited fd: %d.", fd); | |
246 | goto restart; | |
247 | } | |
248 | WARN("Inherited fd: %d.", fd); | |
249 | } | |
250 | ||
251 | /* Only enable syslog at this point to avoid the above logging function | |
252 | * to open a new fd and make the check_inherited function enter an | |
253 | * infinite loop. | |
254 | */ | |
255 | lxc_log_enable_syslog(); | |
256 | ||
257 | closedir(dir); /* cannot fail */ | |
258 | return 0; | |
259 | } | |
260 | ||
261 | static int setup_signal_fd(sigset_t *oldmask) | |
262 | { | |
263 | sigset_t mask; | |
264 | int fd; | |
265 | ||
266 | /* Block everything except serious error signals. */ | |
267 | if (sigfillset(&mask) || | |
268 | sigdelset(&mask, SIGILL) || | |
269 | sigdelset(&mask, SIGSEGV) || | |
270 | sigdelset(&mask, SIGBUS) || | |
271 | sigdelset(&mask, SIGWINCH) || | |
272 | sigprocmask(SIG_BLOCK, &mask, oldmask)) { | |
273 | SYSERROR("Failed to set signal mask."); | |
274 | return -1; | |
275 | } | |
276 | ||
277 | fd = signalfd(-1, &mask, 0); | |
278 | if (fd < 0) { | |
279 | SYSERROR("Failed to create signal file descriptor."); | |
280 | return -1; | |
281 | } | |
282 | ||
283 | if (fcntl(fd, F_SETFD, FD_CLOEXEC)) { | |
284 | SYSERROR("Failed to set FD_CLOEXEC on the signal file descriptor: %d.", fd); | |
285 | close(fd); | |
286 | return -1; | |
287 | } | |
288 | ||
289 | DEBUG("Set SIGCHLD handler with file descriptor: %d.", fd); | |
290 | ||
291 | return fd; | |
292 | } | |
293 | ||
294 | static int signal_handler(int fd, uint32_t events, void *data, | |
295 | struct lxc_epoll_descr *descr) | |
296 | { | |
297 | struct signalfd_siginfo siginfo; | |
298 | siginfo_t info; | |
299 | int ret; | |
300 | pid_t *pid = data; | |
301 | bool init_died = false; | |
302 | ||
303 | ret = read(fd, &siginfo, sizeof(siginfo)); | |
304 | if (ret < 0) { | |
305 | ERROR("Failed to read signal info from signal file descriptor: %d.", fd); | |
306 | return -1; | |
307 | } | |
308 | ||
309 | if (ret != sizeof(siginfo)) { | |
310 | ERROR("Unexpected size for siginfo struct."); | |
311 | return -1; | |
312 | } | |
313 | ||
314 | /* Check whether init is running. */ | |
315 | info.si_pid = 0; | |
316 | ret = waitid(P_PID, *pid, &info, WEXITED | WNOWAIT | WNOHANG); | |
317 | if (ret == 0 && info.si_pid == *pid) | |
318 | init_died = true; | |
319 | ||
320 | if (siginfo.ssi_signo != SIGCHLD) { | |
321 | kill(*pid, siginfo.ssi_signo); | |
322 | INFO("Forwarded signal %d to pid %d.", siginfo.ssi_signo, *pid); | |
323 | return init_died ? 1 : 0; | |
324 | } | |
325 | ||
326 | if (siginfo.ssi_code == CLD_STOPPED) { | |
327 | INFO("Container init process was stopped."); | |
328 | return init_died ? 1 : 0; | |
329 | } else if (siginfo.ssi_code == CLD_CONTINUED) { | |
330 | INFO("Container init process was continued."); | |
331 | return init_died ? 1 : 0; | |
332 | } | |
333 | ||
334 | /* More robustness, protect ourself from a SIGCHLD sent | |
335 | * by a process different from the container init. | |
336 | */ | |
337 | if (siginfo.ssi_pid != *pid) { | |
338 | NOTICE("Received SIGCHLD from pid %d instead of container init %d.", siginfo.ssi_pid, *pid); | |
339 | return init_died ? 1 : 0; | |
340 | } | |
341 | ||
342 | DEBUG("Container init process %d exited.", *pid); | |
343 | return 1; | |
344 | } | |
345 | ||
346 | static int lxc_serve_state_clients(const char *name, | |
347 | struct lxc_handler *handler, | |
348 | lxc_state_t state) | |
349 | { | |
350 | ssize_t ret; | |
351 | struct lxc_list *cur, *next; | |
352 | struct state_client *client; | |
353 | struct lxc_msg msg = {.type = lxc_msg_state, .value = state}; | |
354 | ||
355 | process_lock(); | |
356 | ||
357 | /* Only set state under process lock held so that we don't cause | |
358 | * lxc_cmd_add_state_client() to miss a state. | |
359 | */ | |
360 | handler->state = state; | |
361 | TRACE("set container state to %s", lxc_state2str(state)); | |
362 | ||
363 | if (lxc_list_empty(&handler->state_clients)) { | |
364 | TRACE("no state clients registered"); | |
365 | process_unlock(); | |
366 | lxc_monitor_send_state(name, state, handler->lxcpath); | |
367 | return 0; | |
368 | } | |
369 | ||
370 | strncpy(msg.name, name, sizeof(msg.name)); | |
371 | msg.name[sizeof(msg.name) - 1] = 0; | |
372 | ||
373 | lxc_list_for_each_safe(cur, &handler->state_clients, next) { | |
374 | client = cur->elem; | |
375 | ||
376 | if (!client->states[state]) { | |
377 | TRACE("state %s not registered for state client %d", | |
378 | lxc_state2str(state), client->clientfd); | |
379 | continue; | |
380 | } | |
381 | ||
382 | TRACE("sending state %s to state client %d", | |
383 | lxc_state2str(state), client->clientfd); | |
384 | ||
385 | again: | |
386 | ret = send(client->clientfd, &msg, sizeof(msg), 0); | |
387 | if (ret < 0) { | |
388 | if (errno == EINTR) | |
389 | goto again; | |
390 | ||
391 | ERROR("failed to send message to client"); | |
392 | } | |
393 | ||
394 | /* kick client from list */ | |
395 | close(client->clientfd); | |
396 | lxc_list_del(cur); | |
397 | free(cur->elem); | |
398 | free(cur); | |
399 | } | |
400 | process_unlock(); | |
401 | ||
402 | return 0; | |
403 | } | |
404 | ||
405 | static int lxc_serve_state_socket_pair(const char *name, | |
406 | struct lxc_handler *handler, | |
407 | lxc_state_t state) | |
408 | { | |
409 | ssize_t ret; | |
410 | ||
411 | if (!handler->backgrounded || | |
412 | handler->state_socket_pair[1] < 0 || | |
413 | state == STARTING) | |
414 | return 0; | |
415 | ||
416 | /* Close read end of the socket pair. */ | |
417 | close(handler->state_socket_pair[0]); | |
418 | handler->state_socket_pair[0] = -1; | |
419 | ||
420 | ret = lxc_abstract_unix_send_credential(handler->state_socket_pair[1], | |
421 | &(int){state}, sizeof(int)); | |
422 | if (ret != sizeof(int)) | |
423 | SYSERROR("Failed to send state to %d", | |
424 | handler->state_socket_pair[1]); | |
425 | ||
426 | TRACE("Sent container state \"%s\" to %d", lxc_state2str(state), | |
427 | handler->state_socket_pair[1]); | |
428 | ||
429 | /* Close write end of the socket pair. */ | |
430 | close(handler->state_socket_pair[1]); | |
431 | handler->state_socket_pair[1] = -1; | |
432 | ||
433 | return 0; | |
434 | } | |
435 | ||
436 | int lxc_set_state(const char *name, struct lxc_handler *handler, | |
437 | lxc_state_t state) | |
438 | { | |
439 | int ret; | |
440 | ||
441 | ret = lxc_serve_state_socket_pair(name, handler, state); | |
442 | if (ret < 0) { | |
443 | ERROR("Failed to synchronize via anonymous pair of unix sockets"); | |
444 | return -1; | |
445 | } | |
446 | ||
447 | ret = lxc_serve_state_clients(name, handler, state); | |
448 | if (ret < 0) | |
449 | return -1; | |
450 | ||
451 | /* This function will try to connect to the legacy lxc-monitord state | |
452 | * server and only exists for backwards compatibility. | |
453 | */ | |
454 | lxc_monitor_send_state(name, state, handler->lxcpath); | |
455 | ||
456 | return 0; | |
457 | } | |
458 | ||
459 | int lxc_poll(const char *name, struct lxc_handler *handler) | |
460 | { | |
461 | int sigfd = handler->sigfd; | |
462 | int pid = handler->pid; | |
463 | struct lxc_epoll_descr descr; | |
464 | ||
465 | if (lxc_mainloop_open(&descr)) { | |
466 | ERROR("Failed to create LXC mainloop."); | |
467 | goto out_sigfd; | |
468 | } | |
469 | ||
470 | if (lxc_mainloop_add_handler(&descr, sigfd, signal_handler, &pid)) { | |
471 | ERROR("Failed to add signal handler with file descriptor %d to LXC mainloop.", sigfd); | |
472 | goto out_mainloop_open; | |
473 | } | |
474 | ||
475 | if (lxc_console_mainloop_add(&descr, handler->conf)) { | |
476 | ERROR("Failed to add console handler to LXC mainloop."); | |
477 | goto out_mainloop_open; | |
478 | } | |
479 | ||
480 | if (lxc_cmd_mainloop_add(name, &descr, handler)) { | |
481 | ERROR("Failed to add command handler to LXC mainloop."); | |
482 | goto out_mainloop_open; | |
483 | } | |
484 | ||
485 | if (handler->conf->need_utmp_watch) { | |
486 | #if HAVE_LIBCAP | |
487 | if (lxc_utmp_mainloop_add(&descr, handler)) { | |
488 | ERROR("Failed to add utmp handler to LXC mainloop."); | |
489 | goto out_mainloop_open; | |
490 | } | |
491 | #else | |
492 | DEBUG("Not starting utmp handler as CAP_SYS_BOOT cannot be dropped without capabilities support."); | |
493 | #endif | |
494 | } | |
495 | TRACE("lxc mainloop is ready"); | |
496 | ||
497 | return lxc_mainloop(&descr, -1); | |
498 | ||
499 | out_mainloop_open: | |
500 | lxc_mainloop_close(&descr); | |
501 | ||
502 | out_sigfd: | |
503 | close(sigfd); | |
504 | ||
505 | return -1; | |
506 | } | |
507 | ||
508 | void lxc_free_handler(struct lxc_handler *handler) | |
509 | { | |
510 | if (handler->conf && handler->conf->maincmd_fd) | |
511 | close(handler->conf->maincmd_fd); | |
512 | ||
513 | if (handler->state_socket_pair[0] >= 0) | |
514 | close(handler->state_socket_pair[0]); | |
515 | ||
516 | if (handler->state_socket_pair[1] >= 0) | |
517 | close(handler->state_socket_pair[1]); | |
518 | ||
519 | if (handler->name) | |
520 | free(handler->name); | |
521 | ||
522 | handler->conf = NULL; | |
523 | free(handler); | |
524 | } | |
525 | ||
526 | struct lxc_handler *lxc_init_handler(const char *name, struct lxc_conf *conf, | |
527 | const char *lxcpath, bool daemonize) | |
528 | { | |
529 | int i, ret; | |
530 | struct lxc_handler *handler; | |
531 | ||
532 | handler = malloc(sizeof(*handler)); | |
533 | if (!handler) { | |
534 | ERROR("failed to allocate memory"); | |
535 | return NULL; | |
536 | } | |
537 | ||
538 | memset(handler, 0, sizeof(*handler)); | |
539 | ||
540 | handler->ttysock[0] = handler->ttysock[1] = -1; | |
541 | handler->conf = conf; | |
542 | handler->lxcpath = lxcpath; | |
543 | handler->pinfd = -1; | |
544 | handler->state_socket_pair[0] = handler->state_socket_pair[1] = -1; | |
545 | lxc_list_init(&handler->state_clients); | |
546 | ||
547 | for (i = 0; i < LXC_NS_MAX; i++) | |
548 | handler->nsfd[i] = -1; | |
549 | ||
550 | handler->name = strdup(name); | |
551 | if (!handler->name) { | |
552 | ERROR("failed to allocate memory"); | |
553 | goto on_error; | |
554 | } | |
555 | ||
556 | if (daemonize && !handler->conf->reboot) { | |
557 | /* Create socketpair() to synchronize on daemonized startup. | |
558 | * When the container reboots we don't need to synchronize again | |
559 | * currently so don't open another socketpair(). | |
560 | */ | |
561 | ret = socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0, | |
562 | handler->state_socket_pair); | |
563 | if (ret < 0) { | |
564 | ERROR("Failed to create anonymous pair of unix sockets"); | |
565 | goto on_error; | |
566 | } | |
567 | TRACE("Created anonymous pair {%d,%d} of unix sockets", | |
568 | handler->state_socket_pair[0], | |
569 | handler->state_socket_pair[1]); | |
570 | } | |
571 | ||
572 | if (lxc_cmd_init(name, handler, lxcpath)) { | |
573 | ERROR("failed to set up command socket"); | |
574 | goto on_error; | |
575 | } | |
576 | ||
577 | TRACE("unix domain socket %d for command server is ready", | |
578 | handler->conf->maincmd_fd); | |
579 | ||
580 | return handler; | |
581 | ||
582 | on_error: | |
583 | lxc_free_handler(handler); | |
584 | ||
585 | return NULL; | |
586 | } | |
587 | ||
588 | int lxc_init(const char *name, struct lxc_handler *handler) | |
589 | { | |
590 | struct lxc_conf *conf = handler->conf; | |
591 | ||
592 | lsm_init(); | |
593 | TRACE("initialized LSM"); | |
594 | ||
595 | if (lxc_read_seccomp_config(conf) != 0) { | |
596 | ERROR("Failed loading seccomp policy."); | |
597 | goto out_close_maincmd_fd; | |
598 | } | |
599 | TRACE("read seccomp policy"); | |
600 | ||
601 | /* Begin by setting the state to STARTING. */ | |
602 | if (lxc_set_state(name, handler, STARTING)) { | |
603 | ERROR("Failed to set state for container \"%s\" to \"%s\".", name, lxc_state2str(STARTING)); | |
604 | goto out_close_maincmd_fd; | |
605 | } | |
606 | TRACE("set container state to \"STARTING\""); | |
607 | ||
608 | /* Start of environment variable setup for hooks. */ | |
609 | if (name && setenv("LXC_NAME", name, 1)) | |
610 | SYSERROR("Failed to set environment variable: LXC_NAME=%s.", name); | |
611 | ||
612 | if (conf->rcfile && setenv("LXC_CONFIG_FILE", conf->rcfile, 1)) | |
613 | SYSERROR("Failed to set environment variable: LXC_CONFIG_FILE=%s.", conf->rcfile); | |
614 | ||
615 | if (conf->rootfs.mount && setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1)) | |
616 | SYSERROR("Failed to set environment variable: LXC_ROOTFS_MOUNT=%s.", conf->rootfs.mount); | |
617 | ||
618 | if (conf->rootfs.path && setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1)) | |
619 | SYSERROR("Failed to set environment variable: LXC_ROOTFS_PATH=%s.", conf->rootfs.path); | |
620 | ||
621 | if (conf->console.path && setenv("LXC_CONSOLE", conf->console.path, 1)) | |
622 | SYSERROR("Failed to set environment variable: LXC_CONSOLE=%s.", conf->console.path); | |
623 | ||
624 | if (conf->console.log_path && setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1)) | |
625 | SYSERROR("Failed to set environment variable: LXC_CONSOLE_LOGPATH=%s.", conf->console.log_path); | |
626 | ||
627 | if (setenv("LXC_CGNS_AWARE", "1", 1)) | |
628 | SYSERROR("Failed to set environment variable LXC_CGNS_AWARE=1."); | |
629 | /* End of environment variable setup for hooks. */ | |
630 | ||
631 | TRACE("set environment variables"); | |
632 | ||
633 | if (run_lxc_hooks(name, "pre-start", conf, handler->lxcpath, NULL)) { | |
634 | ERROR("Failed to run lxc.hook.pre-start for container \"%s\".", name); | |
635 | goto out_aborting; | |
636 | } | |
637 | TRACE("ran pre-start hooks"); | |
638 | ||
639 | /* The signal fd has to be created before forking otherwise if the child | |
640 | * process exits before we setup the signal fd, the event will be lost | |
641 | * and the command will be stuck. | |
642 | */ | |
643 | handler->sigfd = setup_signal_fd(&handler->oldmask); | |
644 | if (handler->sigfd < 0) { | |
645 | ERROR("Failed to setup SIGCHLD fd handler."); | |
646 | goto out_delete_tty; | |
647 | } | |
648 | TRACE("set up signal fd"); | |
649 | ||
650 | /* Do this after setting up signals since it might unblock SIGWINCH. */ | |
651 | if (lxc_console_create(conf)) { | |
652 | ERROR("Failed to create console for container \"%s\".", name); | |
653 | goto out_restore_sigmask; | |
654 | } | |
655 | TRACE("created console"); | |
656 | ||
657 | if (lxc_ttys_shift_ids(conf) < 0) { | |
658 | ERROR("Failed to shift tty into container."); | |
659 | goto out_restore_sigmask; | |
660 | } | |
661 | TRACE("shifted tty ids"); | |
662 | ||
663 | INFO("container \"%s\" is initialized", name); | |
664 | return 0; | |
665 | ||
666 | out_restore_sigmask: | |
667 | sigprocmask(SIG_SETMASK, &handler->oldmask, NULL); | |
668 | out_delete_tty: | |
669 | lxc_delete_tty(&conf->tty_info); | |
670 | out_aborting: | |
671 | lxc_set_state(name, handler, ABORTING); | |
672 | out_close_maincmd_fd: | |
673 | close(conf->maincmd_fd); | |
674 | conf->maincmd_fd = -1; | |
675 | return -1; | |
676 | } | |
677 | ||
678 | void lxc_fini(const char *name, struct lxc_handler *handler) | |
679 | { | |
680 | int i, rc; | |
681 | struct lxc_list *cur, *next; | |
682 | pid_t self = getpid(); | |
683 | char *namespaces[LXC_NS_MAX + 1]; | |
684 | size_t namespace_count = 0; | |
685 | ||
686 | /* The STOPPING state is there for future cleanup code which can take | |
687 | * awhile. | |
688 | */ | |
689 | lxc_set_state(name, handler, STOPPING); | |
690 | ||
691 | for (i = 0; i < LXC_NS_MAX; i++) { | |
692 | if (handler->nsfd[i] != -1) { | |
693 | rc = asprintf(&namespaces[namespace_count], "%s:/proc/%d/fd/%d", | |
694 | ns_info[i].proc_name, self, handler->nsfd[i]); | |
695 | if (rc == -1) { | |
696 | SYSERROR("Failed to allocate memory."); | |
697 | break; | |
698 | } | |
699 | ++namespace_count; | |
700 | } | |
701 | } | |
702 | namespaces[namespace_count] = NULL; | |
703 | ||
704 | if (handler->conf->reboot && setenv("LXC_TARGET", "reboot", 1)) | |
705 | SYSERROR("Failed to set environment variable: LXC_TARGET=reboot."); | |
706 | ||
707 | if (!handler->conf->reboot && setenv("LXC_TARGET", "stop", 1)) | |
708 | SYSERROR("Failed to set environment variable: LXC_TARGET=stop."); | |
709 | ||
710 | if (run_lxc_hooks(name, "stop", handler->conf, handler->lxcpath, namespaces)) | |
711 | ERROR("Failed to run lxc.hook.stop for container \"%s\".", name); | |
712 | ||
713 | while (namespace_count--) | |
714 | free(namespaces[namespace_count]); | |
715 | for (i = 0; i < LXC_NS_MAX; i++) { | |
716 | if (handler->nsfd[i] != -1) { | |
717 | close(handler->nsfd[i]); | |
718 | handler->nsfd[i] = -1; | |
719 | } | |
720 | } | |
721 | ||
722 | if (handler->netnsfd >= 0) { | |
723 | close(handler->netnsfd); | |
724 | handler->netnsfd = -1; | |
725 | } | |
726 | ||
727 | lxc_set_state(name, handler, STOPPED); | |
728 | ||
729 | if (run_lxc_hooks(name, "post-stop", handler->conf, handler->lxcpath, NULL)) { | |
730 | ERROR("Failed to run lxc.hook.post-stop for container \"%s\".", name); | |
731 | if (handler->conf->reboot) { | |
732 | WARN("Container will be stopped instead of rebooted."); | |
733 | handler->conf->reboot = 0; | |
734 | if (setenv("LXC_TARGET", "stop", 1)) | |
735 | WARN("Failed to set environment variable: LXC_TARGET=stop."); | |
736 | } | |
737 | } | |
738 | ||
739 | /* Reset mask set by setup_signal_fd. */ | |
740 | if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) | |
741 | WARN("Failed to restore signal mask."); | |
742 | ||
743 | lxc_console_delete(&handler->conf->console); | |
744 | lxc_delete_tty(&handler->conf->tty_info); | |
745 | ||
746 | /* close the command socket */ | |
747 | close(handler->conf->maincmd_fd); | |
748 | handler->conf->maincmd_fd = -1; | |
749 | ||
750 | /* The command socket is now closed, no more state clients can register | |
751 | * themselves from now on. So free the list of state clients. | |
752 | */ | |
753 | lxc_list_for_each_safe(cur, &handler->state_clients, next) { | |
754 | struct state_client *client = cur->elem; | |
755 | /* close state client socket */ | |
756 | close(client->clientfd); | |
757 | lxc_list_del(cur); | |
758 | free(cur->elem); | |
759 | free(cur); | |
760 | } | |
761 | ||
762 | free(handler->name); | |
763 | if (handler->ttysock[0] != -1) { | |
764 | close(handler->ttysock[0]); | |
765 | close(handler->ttysock[1]); | |
766 | } | |
767 | ||
768 | if (handler->conf->ephemeral == 1 && handler->conf->reboot != 1) | |
769 | lxc_destroy_container_on_signal(handler, name); | |
770 | ||
771 | cgroup_destroy(handler); | |
772 | free(handler); | |
773 | } | |
774 | ||
775 | void lxc_abort(const char *name, struct lxc_handler *handler) | |
776 | { | |
777 | int ret, status; | |
778 | ||
779 | lxc_set_state(name, handler, ABORTING); | |
780 | if (handler->pid > 0) | |
781 | kill(handler->pid, SIGKILL); | |
782 | while ((ret = waitpid(-1, &status, 0)) > 0) { | |
783 | ; | |
784 | } | |
785 | } | |
786 | ||
787 | #include <sys/reboot.h> | |
788 | #include <linux/reboot.h> | |
789 | ||
790 | /* reboot(LINUX_REBOOT_CMD_CAD_ON) will return -EINVAL in a child pid namespace | |
791 | * if container reboot support exists. Otherwise, it will either succeed or | |
792 | * return -EPERM. | |
793 | */ | |
794 | static int container_reboot_supported(void *arg) | |
795 | { | |
796 | int *cmd = arg; | |
797 | int ret; | |
798 | ||
799 | ret = reboot(*cmd); | |
800 | if (ret == -1 && errno == EINVAL) | |
801 | return 1; | |
802 | return 0; | |
803 | } | |
804 | ||
805 | static int must_drop_cap_sys_boot(struct lxc_conf *conf) | |
806 | { | |
807 | FILE *f; | |
808 | int ret, cmd, v, flags; | |
809 | long stack_size = 4096; | |
810 | void *stack = alloca(stack_size); | |
811 | int status; | |
812 | pid_t pid; | |
813 | ||
814 | f = fopen("/proc/sys/kernel/ctrl-alt-del", "r"); | |
815 | if (!f) { | |
816 | DEBUG("failed to open /proc/sys/kernel/ctrl-alt-del"); | |
817 | return 1; | |
818 | } | |
819 | ||
820 | ret = fscanf(f, "%d", &v); | |
821 | fclose(f); | |
822 | if (ret != 1) { | |
823 | DEBUG("Failed to read /proc/sys/kernel/ctrl-alt-del."); | |
824 | return 1; | |
825 | } | |
826 | cmd = v ? LINUX_REBOOT_CMD_CAD_ON : LINUX_REBOOT_CMD_CAD_OFF; | |
827 | ||
828 | flags = CLONE_NEWPID | SIGCHLD; | |
829 | if (!lxc_list_empty(&conf->id_map)) | |
830 | flags |= CLONE_NEWUSER; | |
831 | ||
832 | #ifdef __ia64__ | |
833 | pid = __clone2(container_reboot_supported, stack, stack_size, flags, &cmd); | |
834 | #else | |
835 | stack += stack_size; | |
836 | pid = clone(container_reboot_supported, stack, flags, &cmd); | |
837 | #endif | |
838 | if (pid < 0) { | |
839 | if (flags & CLONE_NEWUSER) | |
840 | ERROR("Failed to clone (%#x): %s (includes CLONE_NEWUSER).", flags, strerror(errno)); | |
841 | else | |
842 | ERROR("Failed to clone (%#x): %s.", flags, strerror(errno)); | |
843 | return -1; | |
844 | } | |
845 | if (wait(&status) < 0) { | |
846 | SYSERROR("Unexpected wait error: %s.", strerror(errno)); | |
847 | return -1; | |
848 | } | |
849 | ||
850 | if (WEXITSTATUS(status) != 1) | |
851 | return 1; | |
852 | ||
853 | return 0; | |
854 | } | |
855 | ||
856 | /* netpipe is used in the unprivileged case to transfer the ifindexes from | |
857 | * parent to child | |
858 | */ | |
859 | static int netpipe = -1; | |
860 | ||
861 | static inline int count_veths(struct lxc_list *network) | |
862 | { | |
863 | struct lxc_list *iterator; | |
864 | struct lxc_netdev *netdev; | |
865 | int count = 0; | |
866 | ||
867 | lxc_list_for_each(iterator, network) { | |
868 | netdev = iterator->elem; | |
869 | if (netdev->type != LXC_NET_VETH) | |
870 | continue; | |
871 | count++; | |
872 | } | |
873 | return count; | |
874 | } | |
875 | ||
876 | static int read_unpriv_netifindex(struct lxc_list *network) | |
877 | { | |
878 | struct lxc_list *iterator; | |
879 | struct lxc_netdev *netdev; | |
880 | ||
881 | if (netpipe == -1) | |
882 | return 0; | |
883 | lxc_list_for_each(iterator, network) { | |
884 | netdev = iterator->elem; | |
885 | if (netdev->type != LXC_NET_VETH) | |
886 | continue; | |
887 | if (!(netdev->name = malloc(IFNAMSIZ))) { | |
888 | ERROR("Out of memory."); | |
889 | close(netpipe); | |
890 | return -1; | |
891 | } | |
892 | if (read(netpipe, netdev->name, IFNAMSIZ) != IFNAMSIZ) { | |
893 | close(netpipe); | |
894 | return -1; | |
895 | } | |
896 | } | |
897 | close(netpipe); | |
898 | return 0; | |
899 | } | |
900 | ||
901 | static int do_start(void *data) | |
902 | { | |
903 | struct lxc_list *iterator; | |
904 | struct lxc_handler *handler = data; | |
905 | int devnull_fd = -1, ret; | |
906 | char path[PATH_MAX]; | |
907 | ||
908 | if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) { | |
909 | SYSERROR("Failed to set signal mask."); | |
910 | return -1; | |
911 | } | |
912 | ||
913 | /* This prctl must be before the synchro, so if the parent dies before | |
914 | * we set the parent death signal, we will detect its death with the | |
915 | * synchro right after, otherwise we have a window where the parent can | |
916 | * exit before we set the pdeath signal leading to a unsupervized | |
917 | * container. | |
918 | */ | |
919 | if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) { | |
920 | SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL."); | |
921 | return -1; | |
922 | } | |
923 | ||
924 | lxc_sync_fini_parent(handler); | |
925 | ||
926 | /* Don't leak the pinfd to the container. */ | |
927 | if (handler->pinfd >= 0) { | |
928 | close(handler->pinfd); | |
929 | } | |
930 | ||
931 | if (lxc_sync_wait_parent(handler, LXC_SYNC_STARTUP)) | |
932 | return -1; | |
933 | ||
934 | /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See | |
935 | * https://github.com/lxc/lxd/issues/1978. | |
936 | */ | |
937 | if ((handler->clone_flags & (CLONE_NEWNET | CLONE_NEWUSER)) == | |
938 | (CLONE_NEWNET | CLONE_NEWUSER)) { | |
939 | ret = unshare(CLONE_NEWNET); | |
940 | if (ret < 0) { | |
941 | SYSERROR("Failed to unshare CLONE_NEWNET."); | |
942 | goto out_warn_father; | |
943 | } | |
944 | INFO("Unshared CLONE_NEWNET."); | |
945 | } | |
946 | ||
947 | /* Tell the parent task it can begin to configure the container and wait | |
948 | * for it to finish. | |
949 | */ | |
950 | if (lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE)) | |
951 | return -1; | |
952 | ||
953 | if (read_unpriv_netifindex(&handler->conf->network) < 0) | |
954 | goto out_warn_father; | |
955 | ||
956 | /* If we are in a new user namespace, become root there to have | |
957 | * privilege over our namespace. | |
958 | */ | |
959 | if (!lxc_list_empty(&handler->conf->id_map)) { | |
960 | if (lxc_switch_uid_gid(0, 0) < 0) | |
961 | goto out_warn_father; | |
962 | ||
963 | /* Drop groups only after we switched to a valid gid in the new | |
964 | * user namespace. | |
965 | */ | |
966 | if (lxc_setgroups(0, NULL) < 0) | |
967 | goto out_warn_father; | |
968 | } | |
969 | ||
970 | if (access(handler->lxcpath, X_OK)) { | |
971 | print_top_failing_dir(handler->lxcpath); | |
972 | goto out_warn_father; | |
973 | } | |
974 | ||
975 | #if HAVE_LIBCAP | |
976 | if (handler->conf->need_utmp_watch) { | |
977 | if (prctl(PR_CAPBSET_DROP, CAP_SYS_BOOT, 0, 0, 0)) { | |
978 | SYSERROR("Failed to remove the CAP_SYS_BOOT capability."); | |
979 | goto out_warn_father; | |
980 | } | |
981 | DEBUG("Dropped the CAP_SYS_BOOT capability."); | |
982 | } | |
983 | #endif | |
984 | ||
985 | ret = snprintf(path, sizeof(path), "%s/dev/null", handler->conf->rootfs.mount); | |
986 | if (ret < 0 || ret >= sizeof(path)) | |
987 | goto out_warn_father; | |
988 | ||
989 | /* In order to checkpoint restore, we need to have everything in the | |
990 | * same mount namespace. However, some containers may not have a | |
991 | * reasonable /dev (in particular, they may not have /dev/null), so we | |
992 | * can't set init's std fds to /dev/null by opening it from inside the | |
993 | * container. | |
994 | * | |
995 | * If that's the case, fall back to using the host's /dev/null. This | |
996 | * means that migration won't work, but at least we won't spew output | |
997 | * where it isn't wanted. | |
998 | */ | |
999 | if (handler->backgrounded && !handler->conf->autodev && access(path, F_OK) < 0) { | |
1000 | devnull_fd = open_devnull(); | |
1001 | ||
1002 | if (devnull_fd < 0) | |
1003 | goto out_warn_father; | |
1004 | WARN("Using /dev/null from the host for container init's " | |
1005 | "standard file descriptors. Migration will not work."); | |
1006 | } | |
1007 | ||
1008 | /* Setup the container, ip, names, utsname, ... */ | |
1009 | if (lxc_setup(handler)) { | |
1010 | ERROR("Failed to setup container \"%s\".", handler->name); | |
1011 | goto out_warn_father; | |
1012 | } | |
1013 | ||
1014 | /* Ask father to setup cgroups and wait for him to finish. */ | |
1015 | if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP)) | |
1016 | goto out_error; | |
1017 | ||
1018 | /* Unshare cgroup namespace after we have setup our cgroups. If we do it | |
1019 | * earlier we end up with a wrong view of /proc/self/cgroup. For | |
1020 | * example, assume we unshare(CLONE_NEWCGROUP) first, and then create | |
1021 | * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then | |
1022 | * /proc/self/cgroup would show us: | |
1023 | * | |
1024 | * 8:cpuset:/lxc/c | |
1025 | * | |
1026 | * whereas it should actually show | |
1027 | * | |
1028 | * 8:cpuset:/ | |
1029 | */ | |
1030 | if (cgns_supported()) { | |
1031 | if (unshare(CLONE_NEWCGROUP) < 0) { | |
1032 | INFO("Failed to unshare CLONE_NEWCGROUP."); | |
1033 | goto out_warn_father; | |
1034 | } | |
1035 | INFO("Unshared CLONE_NEWCGROUP."); | |
1036 | } | |
1037 | ||
1038 | /* Set the label to change to when we exec(2) the container's init. */ | |
1039 | if (lsm_process_label_set(NULL, handler->conf, 1, 1) < 0) | |
1040 | goto out_warn_father; | |
1041 | ||
1042 | /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it | |
1043 | * before we aren't allowed anymore. | |
1044 | */ | |
1045 | if (handler->conf->no_new_privs) { | |
1046 | if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) { | |
1047 | SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges."); | |
1048 | goto out_warn_father; | |
1049 | } | |
1050 | DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges."); | |
1051 | } | |
1052 | ||
1053 | /* Some init's such as busybox will set sane tty settings on stdin, | |
1054 | * stdout, stderr which it thinks is the console. We already set them | |
1055 | * the way we wanted on the real terminal, and we want init to do its | |
1056 | * setup on its console ie. the pty allocated in lxc_console_create() so | |
1057 | * make sure that that pty is stdin,stdout,stderr. | |
1058 | */ | |
1059 | if (lxc_console_set_stdfds(handler->conf->console.slave) < 0) | |
1060 | goto out_warn_father; | |
1061 | ||
1062 | /* If we mounted a temporary proc, then unmount it now. */ | |
1063 | tmp_proc_unmount(handler->conf); | |
1064 | ||
1065 | if (lxc_seccomp_load(handler->conf) != 0) | |
1066 | goto out_warn_father; | |
1067 | ||
1068 | if (run_lxc_hooks(handler->name, "start", handler->conf, handler->lxcpath, NULL)) { | |
1069 | ERROR("Failed to run lxc.hook.start for container \"%s\".", handler->name); | |
1070 | goto out_warn_father; | |
1071 | } | |
1072 | ||
1073 | /* The container has been setup. We can now switch to an unprivileged | |
1074 | * uid/gid. | |
1075 | */ | |
1076 | if (handler->conf->is_execute) { | |
1077 | bool have_cap_setgid; | |
1078 | uid_t new_uid = handler->conf->init_uid; | |
1079 | gid_t new_gid = handler->conf->init_gid; | |
1080 | ||
1081 | /* If we are in a new user namespace we already dropped all | |
1082 | * groups when we switched to root in the new user namespace | |
1083 | * further above. Only drop groups if we can, so ensure that we | |
1084 | * have necessary privilege. | |
1085 | */ | |
1086 | #if HAVE_LIBCAP | |
1087 | have_cap_setgid = lxc_proc_cap_is_set(CAP_SETGID, CAP_EFFECTIVE); | |
1088 | #else | |
1089 | have_cap_setgid = false; | |
1090 | #endif | |
1091 | if (lxc_list_empty(&handler->conf->id_map) && have_cap_setgid) { | |
1092 | if (lxc_setgroups(0, NULL) < 0) | |
1093 | goto out_warn_father; | |
1094 | } | |
1095 | ||
1096 | if (lxc_switch_uid_gid(new_uid, new_gid) < 0) | |
1097 | goto out_warn_father; | |
1098 | } | |
1099 | ||
1100 | /* The clearenv() and putenv() calls have been moved here to allow us to | |
1101 | * use environment variables passed to the various hooks, such as the | |
1102 | * start hook above. Not all of the variables like CONFIG_PATH or ROOTFS | |
1103 | * are valid in this context but others are. | |
1104 | */ | |
1105 | if (clearenv()) { | |
1106 | SYSERROR("Failed to clear environment."); | |
1107 | /* Don't error out though. */ | |
1108 | } | |
1109 | ||
1110 | lxc_list_for_each(iterator, &handler->conf->environment) { | |
1111 | if (putenv((char *)iterator->elem)) { | |
1112 | SYSERROR("Failed to set environment variable: %s.", (char *)iterator->elem); | |
1113 | goto out_warn_father; | |
1114 | } | |
1115 | } | |
1116 | ||
1117 | if (putenv("container=lxc")) { | |
1118 | SYSERROR("Failed to set environment variable: container=lxc."); | |
1119 | goto out_warn_father; | |
1120 | } | |
1121 | ||
1122 | if (handler->conf->pty_names) { | |
1123 | if (putenv(handler->conf->pty_names)) { | |
1124 | SYSERROR("Failed to set environment variable for container ptys."); | |
1125 | goto out_warn_father; | |
1126 | } | |
1127 | } | |
1128 | ||
1129 | close(handler->sigfd); | |
1130 | ||
1131 | if (devnull_fd < 0) { | |
1132 | devnull_fd = open_devnull(); | |
1133 | ||
1134 | if (devnull_fd < 0) | |
1135 | goto out_warn_father; | |
1136 | } | |
1137 | ||
1138 | if (handler->backgrounded && set_stdfds(devnull_fd)) | |
1139 | goto out_warn_father; | |
1140 | ||
1141 | if (devnull_fd >= 0) { | |
1142 | close(devnull_fd); | |
1143 | devnull_fd = -1; | |
1144 | } | |
1145 | ||
1146 | setsid(); | |
1147 | ||
1148 | /* After this call, we are in error because this ops should not return | |
1149 | * as it execs. | |
1150 | */ | |
1151 | handler->ops->start(handler, handler->data); | |
1152 | ||
1153 | out_warn_father: | |
1154 | /* We want the parent to know something went wrong, so we return a | |
1155 | * special error code. | |
1156 | */ | |
1157 | lxc_sync_wake_parent(handler, LXC_SYNC_ERROR); | |
1158 | ||
1159 | out_error: | |
1160 | if (devnull_fd >= 0) | |
1161 | close(devnull_fd); | |
1162 | ||
1163 | return -1; | |
1164 | } | |
1165 | ||
1166 | static int save_phys_nics(struct lxc_conf *conf) | |
1167 | { | |
1168 | struct lxc_list *iterator; | |
1169 | int am_root = (getuid() == 0); | |
1170 | ||
1171 | if (!am_root) | |
1172 | return 0; | |
1173 | ||
1174 | lxc_list_for_each(iterator, &conf->network) { | |
1175 | struct lxc_netdev *netdev = iterator->elem; | |
1176 | ||
1177 | if (netdev->type != LXC_NET_PHYS) | |
1178 | continue; | |
1179 | conf->saved_nics = realloc(conf->saved_nics, | |
1180 | (conf->num_savednics+1)*sizeof(struct saved_nic)); | |
1181 | if (!conf->saved_nics) | |
1182 | return -1; | |
1183 | conf->saved_nics[conf->num_savednics].ifindex = netdev->ifindex; | |
1184 | conf->saved_nics[conf->num_savednics].orig_name = strdup(netdev->link); | |
1185 | if (!conf->saved_nics[conf->num_savednics].orig_name) | |
1186 | return -1; | |
1187 | INFO("Stored saved_nic #%d idx %d name %s.", conf->num_savednics, | |
1188 | conf->saved_nics[conf->num_savednics].ifindex, | |
1189 | conf->saved_nics[conf->num_savednics].orig_name); | |
1190 | conf->num_savednics++; | |
1191 | } | |
1192 | ||
1193 | return 0; | |
1194 | } | |
1195 | ||
1196 | static int lxc_recv_ttys_from_child(struct lxc_handler *handler) | |
1197 | { | |
1198 | int i; | |
1199 | int *ttyfds; | |
1200 | struct lxc_pty_info *pty_info; | |
1201 | int ret = -1; | |
1202 | int sock = handler->ttysock[1]; | |
1203 | struct lxc_conf *conf = handler->conf; | |
1204 | struct lxc_tty_info *tty_info = &conf->tty_info; | |
1205 | size_t num_ttyfds = (2 * conf->tty); | |
1206 | ||
1207 | if (!conf->tty) | |
1208 | return 0; | |
1209 | ||
1210 | tty_info->pty_info = malloc(sizeof(*tty_info->pty_info) * conf->tty); | |
1211 | if (!tty_info->pty_info) | |
1212 | return -1; | |
1213 | ||
1214 | ttyfds = malloc(num_ttyfds * sizeof(int)); | |
1215 | if (!ttyfds) | |
1216 | return -1; | |
1217 | ||
1218 | ret = lxc_abstract_unix_recv_fds(sock, ttyfds, num_ttyfds, NULL, 0); | |
1219 | for (i = 0; (ret >= 0 && *ttyfds != -1) && (i < num_ttyfds); i++) { | |
1220 | pty_info = &tty_info->pty_info[i / 2]; | |
1221 | pty_info->busy = 0; | |
1222 | pty_info->slave = ttyfds[i++]; | |
1223 | pty_info->master = ttyfds[i]; | |
1224 | TRACE("received pty with master fd %d and slave fd %d from " | |
1225 | "parent", pty_info->master, pty_info->slave); | |
1226 | } | |
1227 | ||
1228 | tty_info->nbtty = conf->tty; | |
1229 | ||
1230 | free(ttyfds); | |
1231 | ||
1232 | if (ret < 0) | |
1233 | ERROR("failed to receive %d ttys from child: %s", conf->tty, | |
1234 | strerror(errno)); | |
1235 | else | |
1236 | TRACE("received %d ttys from child", conf->tty); | |
1237 | ||
1238 | return ret; | |
1239 | } | |
1240 | ||
1241 | void resolve_clone_flags(struct lxc_handler *handler) | |
1242 | { | |
1243 | handler->clone_flags = CLONE_NEWPID | CLONE_NEWNS; | |
1244 | ||
1245 | if (!lxc_list_empty(&handler->conf->id_map)) | |
1246 | handler->clone_flags |= CLONE_NEWUSER; | |
1247 | ||
1248 | if (handler->conf->inherit_ns_fd[LXC_NS_NET] == -1) { | |
1249 | if (!lxc_requests_empty_network(handler)) | |
1250 | handler->clone_flags |= CLONE_NEWNET; | |
1251 | } else { | |
1252 | INFO("Inheriting a NET namespace."); | |
1253 | } | |
1254 | ||
1255 | if (handler->conf->inherit_ns_fd[LXC_NS_IPC] == -1) | |
1256 | handler->clone_flags |= CLONE_NEWIPC; | |
1257 | else | |
1258 | INFO("Inheriting an IPC namespace."); | |
1259 | ||
1260 | if (handler->conf->inherit_ns_fd[LXC_NS_UTS] == -1) | |
1261 | handler->clone_flags |= CLONE_NEWUTS; | |
1262 | else | |
1263 | INFO("Inheriting a UTS namespace."); | |
1264 | } | |
1265 | ||
1266 | /* lxc_spawn() performs crucial setup tasks and clone()s the new process which | |
1267 | * exec()s the requested container binary. | |
1268 | * Note that lxc_spawn() runs in the parent namespaces. Any operations performed | |
1269 | * right here should be double checked if they'd pose a security risk. (For | |
1270 | * example, any {u}mount() operations performed here will be reflected on the | |
1271 | * host!) | |
1272 | */ | |
1273 | static int lxc_spawn(struct lxc_handler *handler) | |
1274 | { | |
1275 | int failed_before_rename = 0; | |
1276 | const char *name = handler->name; | |
1277 | bool cgroups_connected = false; | |
1278 | int saved_ns_fd[LXC_NS_MAX]; | |
1279 | int preserve_mask = 0, i, flags; | |
1280 | int netpipepair[2], nveths; | |
1281 | bool wants_to_map_ids; | |
1282 | struct lxc_list *id_map; | |
1283 | ||
1284 | netpipe = -1; | |
1285 | id_map = &handler->conf->id_map; | |
1286 | wants_to_map_ids = !lxc_list_empty(id_map); | |
1287 | ||
1288 | for (i = 0; i < LXC_NS_MAX; i++) | |
1289 | if (handler->conf->inherit_ns_fd[i] != -1) | |
1290 | preserve_mask |= ns_info[i].clone_flag; | |
1291 | ||
1292 | if (lxc_sync_init(handler)) | |
1293 | return -1; | |
1294 | ||
1295 | if (socketpair(AF_UNIX, SOCK_DGRAM, 0, handler->ttysock) < 0) { | |
1296 | lxc_sync_fini(handler); | |
1297 | return -1; | |
1298 | } | |
1299 | ||
1300 | resolve_clone_flags(handler); | |
1301 | ||
1302 | if (handler->clone_flags & CLONE_NEWNET) { | |
1303 | if (!lxc_list_empty(&handler->conf->network)) { | |
1304 | ||
1305 | /* Find gateway addresses from the link device, which is | |
1306 | * no longer accessible inside the container. Do this | |
1307 | * before creating network interfaces, since goto | |
1308 | * out_delete_net does not work before lxc_clone. | |
1309 | */ | |
1310 | if (lxc_find_gateway_addresses(handler)) { | |
1311 | ERROR("Failed to find gateway addresses."); | |
1312 | lxc_sync_fini(handler); | |
1313 | return -1; | |
1314 | } | |
1315 | ||
1316 | /* That should be done before the clone because we will | |
1317 | * fill the netdev index and use them in the child. | |
1318 | */ | |
1319 | if (lxc_setup_networks_in_parent_namespaces(handler)) { | |
1320 | ERROR("Failed to create the network."); | |
1321 | lxc_sync_fini(handler); | |
1322 | return -1; | |
1323 | } | |
1324 | } | |
1325 | ||
1326 | if (save_phys_nics(handler->conf)) { | |
1327 | ERROR("Failed to save physical nic info."); | |
1328 | goto out_abort; | |
1329 | } | |
1330 | } | |
1331 | ||
1332 | if (!cgroup_init(handler)) { | |
1333 | ERROR("Failed initializing cgroup support."); | |
1334 | goto out_delete_net; | |
1335 | } | |
1336 | ||
1337 | cgroups_connected = true; | |
1338 | ||
1339 | if (!cgroup_create(handler)) { | |
1340 | ERROR("Failed creating cgroups."); | |
1341 | goto out_delete_net; | |
1342 | } | |
1343 | ||
1344 | /* If the rootfs is not a blockdev, prevent the container from marking | |
1345 | * it readonly. | |
1346 | * If the container is unprivileged then skip rootfs pinning. | |
1347 | */ | |
1348 | if (!wants_to_map_ids) { | |
1349 | handler->pinfd = pin_rootfs(handler->conf->rootfs.path); | |
1350 | if (handler->pinfd == -1) | |
1351 | INFO("Failed to pin the rootfs for container \"%s\".", handler->name); | |
1352 | } | |
1353 | ||
1354 | if (!preserve_ns(saved_ns_fd, preserve_mask, getpid())) | |
1355 | goto out_delete_net; | |
1356 | ||
1357 | if (attach_ns(handler->conf->inherit_ns_fd) < 0) | |
1358 | goto out_delete_net; | |
1359 | ||
1360 | if (am_unpriv() && (nveths = count_veths(&handler->conf->network))) { | |
1361 | if (pipe(netpipepair) < 0) { | |
1362 | SYSERROR("Failed to create pipe."); | |
1363 | goto out_delete_net; | |
1364 | } | |
1365 | /* Store netpipe in the global var for do_start's use. */ | |
1366 | netpipe = netpipepair[0]; | |
1367 | } | |
1368 | ||
1369 | /* Create a process in a new set of namespaces. */ | |
1370 | flags = handler->clone_flags; | |
1371 | if (handler->clone_flags & CLONE_NEWUSER) { | |
1372 | /* If CLONE_NEWUSER and CLONE_NEWNET was requested, we need to | |
1373 | * clone a new user namespace first and only later unshare our | |
1374 | * network namespace to ensure that network devices ownership is | |
1375 | * set up correctly. | |
1376 | */ | |
1377 | flags &= ~CLONE_NEWNET; | |
1378 | } | |
1379 | handler->pid = lxc_clone(do_start, handler, flags); | |
1380 | if (handler->pid < 0) { | |
1381 | SYSERROR("Failed to clone a new set of namespaces."); | |
1382 | goto out_delete_net; | |
1383 | } | |
1384 | for (i = 0; i < LXC_NS_MAX; i++) | |
1385 | if (flags & ns_info[i].clone_flag) | |
1386 | INFO("Cloned %s.", ns_info[i].flag_name); | |
1387 | ||
1388 | if (!preserve_ns(handler->nsfd, handler->clone_flags | preserve_mask, handler->pid)) | |
1389 | INFO("Failed to preserve namespace for lxc.hook.stop."); | |
1390 | ||
1391 | if (attach_ns(saved_ns_fd)) | |
1392 | WARN("Failed to restore saved namespaces."); | |
1393 | ||
1394 | lxc_sync_fini_child(handler); | |
1395 | ||
1396 | /* Map the container uids. The container became an invalid userid the | |
1397 | * moment it was cloned with CLONE_NEWUSER. This call doesn't change | |
1398 | * anything immediately, but allows the container to setuid(0) (0 being | |
1399 | * mapped to something else on the host.) later to become a valid uid | |
1400 | * again. | |
1401 | */ | |
1402 | if (wants_to_map_ids && lxc_map_ids(id_map, handler->pid)) { | |
1403 | ERROR("Failed to set up id mapping."); | |
1404 | goto out_delete_net; | |
1405 | } | |
1406 | ||
1407 | if (lxc_sync_wake_child(handler, LXC_SYNC_STARTUP)) { | |
1408 | failed_before_rename = 1; | |
1409 | goto out_delete_net; | |
1410 | } | |
1411 | ||
1412 | if (lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE)) { | |
1413 | failed_before_rename = 1; | |
1414 | goto out_delete_net; | |
1415 | } | |
1416 | ||
1417 | if (!cgroup_create_legacy(handler)) { | |
1418 | ERROR("Failed to setup legacy cgroups for container \"%s\".", name); | |
1419 | goto out_delete_net; | |
1420 | } | |
1421 | if (!cgroup_setup_limits(handler, false)) { | |
1422 | ERROR("Failed to setup cgroup limits for container \"%s\".", name); | |
1423 | goto out_delete_net; | |
1424 | } | |
1425 | ||
1426 | if (!cgroup_enter(handler)) | |
1427 | goto out_delete_net; | |
1428 | ||
1429 | if (!cgroup_chown(handler)) | |
1430 | goto out_delete_net; | |
1431 | ||
1432 | if (failed_before_rename) | |
1433 | goto out_delete_net; | |
1434 | ||
1435 | /* Create the network configuration. */ | |
1436 | if (handler->clone_flags & CLONE_NEWNET) { | |
1437 | if (lxc_assign_network(handler->lxcpath, handler->name, | |
1438 | &handler->conf->network, handler->pid)) { | |
1439 | ERROR("Failed to create the configured network."); | |
1440 | goto out_delete_net; | |
1441 | } | |
1442 | } | |
1443 | ||
1444 | if (netpipe != -1) { | |
1445 | struct lxc_list *iterator; | |
1446 | struct lxc_netdev *netdev; | |
1447 | ||
1448 | close(netpipe); | |
1449 | lxc_list_for_each(iterator, &handler->conf->network) { | |
1450 | netdev = iterator->elem; | |
1451 | if (netdev->type != LXC_NET_VETH) | |
1452 | continue; | |
1453 | if (write(netpipepair[1], netdev->name, IFNAMSIZ) != IFNAMSIZ) { | |
1454 | ERROR("Error writing veth name to container."); | |
1455 | goto out_delete_net; | |
1456 | } | |
1457 | } | |
1458 | close(netpipepair[1]); | |
1459 | } | |
1460 | ||
1461 | /* Tell the child to continue its initialization. We'll get | |
1462 | * LXC_SYNC_CGROUP when it is ready for us to setup cgroups. | |
1463 | */ | |
1464 | if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE)) | |
1465 | goto out_delete_net; | |
1466 | ||
1467 | if (!lxc_list_empty(&handler->conf->limits) && setup_resource_limits(&handler->conf->limits, handler->pid)) { | |
1468 | ERROR("failed to setup resource limits for '%s'", name); | |
1469 | goto out_delete_net; | |
1470 | } | |
1471 | ||
1472 | if (!cgroup_setup_limits(handler, true)) { | |
1473 | ERROR("Failed to setup the devices cgroup for container \"%s\".", name); | |
1474 | goto out_delete_net; | |
1475 | } | |
1476 | ||
1477 | cgroup_disconnect(); | |
1478 | cgroups_connected = false; | |
1479 | ||
1480 | /* Read tty fds allocated by child. */ | |
1481 | if (lxc_recv_ttys_from_child(handler) < 0) { | |
1482 | ERROR("Failed to receive tty info from child process."); | |
1483 | goto out_delete_net; | |
1484 | } | |
1485 | ||
1486 | /* Tell the child to complete its initialization and wait for it to exec | |
1487 | * or return an error. (The child will never return | |
1488 | * LXC_SYNC_POST_CGROUP+1. It will either close the sync pipe, causing | |
1489 | * lxc_sync_barrier_child to return success, or return a different | |
1490 | * value, causing us to error out). | |
1491 | */ | |
1492 | if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CGROUP)) | |
1493 | return -1; | |
1494 | ||
1495 | if (handler->ops->post_start(handler, handler->data)) | |
1496 | goto out_abort; | |
1497 | ||
1498 | if (lxc_set_state(name, handler, RUNNING)) { | |
1499 | ERROR("Failed to set state for container \"%s\" to \"%s\".", name, | |
1500 | lxc_state2str(RUNNING)); | |
1501 | goto out_abort; | |
1502 | } | |
1503 | ||
1504 | lxc_sync_fini(handler); | |
1505 | handler->netnsfd = lxc_preserve_ns(handler->pid, "net"); | |
1506 | ||
1507 | return 0; | |
1508 | ||
1509 | out_delete_net: | |
1510 | if (cgroups_connected) | |
1511 | cgroup_disconnect(); | |
1512 | if (handler->clone_flags & CLONE_NEWNET) | |
1513 | lxc_delete_network(handler); | |
1514 | out_abort: | |
1515 | lxc_abort(name, handler); | |
1516 | lxc_sync_fini(handler); | |
1517 | if (handler->pinfd >= 0) { | |
1518 | close(handler->pinfd); | |
1519 | handler->pinfd = -1; | |
1520 | } | |
1521 | ||
1522 | return -1; | |
1523 | } | |
1524 | ||
1525 | int __lxc_start(const char *name, struct lxc_handler *handler, | |
1526 | struct lxc_operations* ops, void *data, const char *lxcpath, | |
1527 | bool backgrounded) | |
1528 | { | |
1529 | int status; | |
1530 | int err = -1; | |
1531 | bool removed_all_netdevs = true; | |
1532 | struct lxc_conf *conf = handler->conf; | |
1533 | ||
1534 | if (lxc_init(name, handler) < 0) { | |
1535 | ERROR("Failed to initialize container \"%s\".", name); | |
1536 | return -1; | |
1537 | } | |
1538 | handler->ops = ops; | |
1539 | handler->data = data; | |
1540 | handler->backgrounded = backgrounded; | |
1541 | handler->netnsfd = -1; | |
1542 | ||
1543 | if (must_drop_cap_sys_boot(handler->conf)) { | |
1544 | #if HAVE_LIBCAP | |
1545 | DEBUG("Dropping CAP_SYS_BOOT capability."); | |
1546 | #else | |
1547 | DEBUG("Not dropping CAP_SYS_BOOT capability as capabilities aren't supported."); | |
1548 | #endif | |
1549 | } else { | |
1550 | DEBUG("Not dropping CAP_SYS_BOOT or watching utmp."); | |
1551 | handler->conf->need_utmp_watch = 0; | |
1552 | } | |
1553 | ||
1554 | if (!attach_block_device(handler->conf)) { | |
1555 | ERROR("Failed to attach block device."); | |
1556 | goto out_fini_nonet; | |
1557 | } | |
1558 | ||
1559 | if (geteuid() == 0 && !lxc_list_empty(&conf->id_map)) { | |
1560 | /* If the backing store is a device, mount it here and now. */ | |
1561 | if (rootfs_is_blockdev(conf)) { | |
1562 | if (unshare(CLONE_NEWNS) < 0) { | |
1563 | ERROR("Failed to unshare CLONE_NEWNS."); | |
1564 | goto out_fini_nonet; | |
1565 | } | |
1566 | INFO("Unshared CLONE_NEWNS."); | |
1567 | ||
1568 | remount_all_slave(); | |
1569 | if (do_rootfs_setup(conf, name, lxcpath) < 0) { | |
1570 | ERROR("Error setting up rootfs mount as root before spawn."); | |
1571 | goto out_fini_nonet; | |
1572 | } | |
1573 | INFO("Set up container rootfs as host root."); | |
1574 | } | |
1575 | } | |
1576 | ||
1577 | err = lxc_spawn(handler); | |
1578 | if (err) { | |
1579 | ERROR("Failed to spawn container \"%s\".", name); | |
1580 | goto out_detach_blockdev; | |
1581 | } | |
1582 | ||
1583 | handler->conf->reboot = 0; | |
1584 | ||
1585 | err = lxc_poll(name, handler); | |
1586 | if (err) { | |
1587 | ERROR("LXC mainloop exited with error: %d.", err); | |
1588 | if (handler->netnsfd >= 0) { | |
1589 | close(handler->netnsfd); | |
1590 | handler->netnsfd = -1; | |
1591 | } | |
1592 | goto out_abort; | |
1593 | } | |
1594 | ||
1595 | while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR) | |
1596 | continue; | |
1597 | ||
1598 | /* If the child process exited but was not signaled, it didn't call | |
1599 | * reboot. This should mean it was an lxc-execute which simply exited. | |
1600 | * In any case, treat it as a 'halt'. | |
1601 | */ | |
1602 | if (WIFSIGNALED(status)) { | |
1603 | switch(WTERMSIG(status)) { | |
1604 | case SIGINT: /* halt */ | |
1605 | DEBUG("Container \"%s\" is halting.", name); | |
1606 | break; | |
1607 | case SIGHUP: /* reboot */ | |
1608 | DEBUG("Container \"%s\" is rebooting.", name); | |
1609 | handler->conf->reboot = 1; | |
1610 | break; | |
1611 | case SIGSYS: /* seccomp */ | |
1612 | DEBUG("Container \"%s\" violated its seccomp policy.", name); | |
1613 | break; | |
1614 | default: | |
1615 | DEBUG("Unknown exit status for container \"%s\" init %d.", name, WTERMSIG(status)); | |
1616 | break; | |
1617 | } | |
1618 | } | |
1619 | ||
1620 | DEBUG("Pushing physical nics back to host namespace"); | |
1621 | lxc_restore_phys_nics_to_netns(handler->netnsfd, handler->conf); | |
1622 | ||
1623 | DEBUG("Tearing down virtual network devices used by container \"%s\".", name); | |
1624 | removed_all_netdevs = lxc_delete_network(handler); | |
1625 | ||
1626 | if (handler->pinfd >= 0) { | |
1627 | close(handler->pinfd); | |
1628 | handler->pinfd = -1; | |
1629 | } | |
1630 | ||
1631 | lxc_monitor_send_exit_code(name, status, handler->lxcpath); | |
1632 | err = lxc_error_set_and_log(handler->pid, status); | |
1633 | out_fini: | |
1634 | if (!removed_all_netdevs) { | |
1635 | DEBUG("Failed tearing down network devices used by container. Trying again!"); | |
1636 | removed_all_netdevs = lxc_delete_network(handler); | |
1637 | if (!removed_all_netdevs) | |
1638 | DEBUG("Failed tearing down network devices used by container. Not trying again!"); | |
1639 | } | |
1640 | ||
1641 | out_detach_blockdev: | |
1642 | detach_block_device(handler->conf); | |
1643 | ||
1644 | out_fini_nonet: | |
1645 | lxc_fini(name, handler); | |
1646 | return err; | |
1647 | ||
1648 | out_abort: | |
1649 | lxc_abort(name, handler); | |
1650 | goto out_fini; | |
1651 | } | |
1652 | ||
1653 | struct start_args { | |
1654 | char *const *argv; | |
1655 | }; | |
1656 | ||
1657 | static int start(struct lxc_handler *handler, void* data) | |
1658 | { | |
1659 | struct start_args *arg = data; | |
1660 | ||
1661 | NOTICE("Exec'ing \"%s\".", arg->argv[0]); | |
1662 | ||
1663 | execvp(arg->argv[0], arg->argv); | |
1664 | SYSERROR("Failed to exec \"%s\".", arg->argv[0]); | |
1665 | return 0; | |
1666 | } | |
1667 | ||
1668 | static int post_start(struct lxc_handler *handler, void* data) | |
1669 | { | |
1670 | struct start_args *arg = data; | |
1671 | ||
1672 | NOTICE("Started \"%s\" with pid \"%d\".", arg->argv[0], handler->pid); | |
1673 | return 0; | |
1674 | } | |
1675 | ||
1676 | static struct lxc_operations start_ops = { | |
1677 | .start = start, | |
1678 | .post_start = post_start | |
1679 | }; | |
1680 | ||
1681 | int lxc_start(const char *name, char *const argv[], struct lxc_handler *handler, | |
1682 | const char *lxcpath, bool backgrounded) | |
1683 | { | |
1684 | struct start_args start_arg = { | |
1685 | .argv = argv, | |
1686 | }; | |
1687 | ||
1688 | handler->conf->need_utmp_watch = 1; | |
1689 | return __lxc_start(name, handler, &start_ops, &start_arg, lxcpath, backgrounded); | |
1690 | } | |
1691 | ||
1692 | static void lxc_destroy_container_on_signal(struct lxc_handler *handler, | |
1693 | const char *name) | |
1694 | { | |
1695 | char destroy[MAXPATHLEN]; | |
1696 | bool bret = true; | |
1697 | int ret = 0; | |
1698 | struct lxc_container *c; | |
1699 | if (handler->conf->rootfs.path && handler->conf->rootfs.mount) { | |
1700 | bret = do_destroy_container(handler->conf); | |
1701 | if (!bret) { | |
1702 | ERROR("Error destroying rootfs for container \"%s\".", name); | |
1703 | return; | |
1704 | } | |
1705 | } | |
1706 | INFO("Destroyed rootfs for container \"%s\".", name); | |
1707 | ||
1708 | ret = snprintf(destroy, MAXPATHLEN, "%s/%s", handler->lxcpath, name); | |
1709 | if (ret < 0 || ret >= MAXPATHLEN) { | |
1710 | ERROR("Error destroying directory for container \"%s\".", name); | |
1711 | return; | |
1712 | } | |
1713 | ||
1714 | c = lxc_container_new(name, handler->lxcpath); | |
1715 | if (c) { | |
1716 | if (container_disk_lock(c)) { | |
1717 | INFO("Could not update lxc_snapshots file."); | |
1718 | lxc_container_put(c); | |
1719 | } else { | |
1720 | mod_all_rdeps(c, false); | |
1721 | container_disk_unlock(c); | |
1722 | lxc_container_put(c); | |
1723 | } | |
1724 | } | |
1725 | ||
1726 | if (am_unpriv()) | |
1727 | ret = userns_exec_1(handler->conf, lxc_rmdir_onedev_wrapper, | |
1728 | destroy, "lxc_rmdir_onedev_wrapper"); | |
1729 | else | |
1730 | ret = lxc_rmdir_onedev(destroy, NULL); | |
1731 | ||
1732 | if (ret < 0) { | |
1733 | ERROR("Error destroying directory for container \"%s\".", name); | |
1734 | return; | |
1735 | } | |
1736 | INFO("Destroyed directory for container \"%s\".", name); | |
1737 | } | |
1738 | ||
1739 | static int lxc_rmdir_onedev_wrapper(void *data) | |
1740 | { | |
1741 | char *arg = (char *) data; | |
1742 | return lxc_rmdir_onedev(arg, NULL); | |
1743 | } | |
1744 | ||
1745 | static bool do_destroy_container(struct lxc_conf *conf) { | |
1746 | if (am_unpriv()) { | |
1747 | if (userns_exec_1(conf, bdev_destroy_wrapper, conf, | |
1748 | "bdev_destroy_wrapper") < 0) | |
1749 | return false; | |
1750 | return true; | |
1751 | } | |
1752 | return bdev_destroy(conf); | |
1753 | } |