]>
Commit | Line | Data |
---|---|---|
1 | /* SPDX-License-Identifier: LGPL-2.1+ */ | |
2 | ||
3 | #ifndef _GNU_SOURCE | |
4 | #define _GNU_SOURCE 1 | |
5 | #endif | |
6 | #include <dirent.h> | |
7 | #include <errno.h> | |
8 | #include <fcntl.h> | |
9 | #include <grp.h> | |
10 | #include <poll.h> | |
11 | #include <pthread.h> | |
12 | #include <signal.h> | |
13 | #include <stdio.h> | |
14 | #include <stdlib.h> | |
15 | #include <string.h> | |
16 | #include <sys/file.h> | |
17 | #include <sys/mount.h> | |
18 | #include <sys/param.h> | |
19 | #include <sys/prctl.h> | |
20 | #include <sys/socket.h> | |
21 | #include <sys/stat.h> | |
22 | #include <sys/syscall.h> | |
23 | #include <sys/types.h> | |
24 | #include <sys/un.h> | |
25 | #include <sys/wait.h> | |
26 | #include <unistd.h> | |
27 | ||
28 | #include "af_unix.h" | |
29 | #include "caps.h" | |
30 | #include "cgroup.h" | |
31 | #include "commands.h" | |
32 | #include "commands_utils.h" | |
33 | #include "conf.h" | |
34 | #include "config.h" | |
35 | #include "confile_utils.h" | |
36 | #include "error.h" | |
37 | #include "file_utils.h" | |
38 | #include "list.h" | |
39 | #include "log.h" | |
40 | #include "lsm/lsm.h" | |
41 | #include "lxccontainer.h" | |
42 | #include "lxclock.h" | |
43 | #include "lxcseccomp.h" | |
44 | #include "macro.h" | |
45 | #include "mainloop.h" | |
46 | #include "memory_utils.h" | |
47 | #include "monitor.h" | |
48 | #include "namespace.h" | |
49 | #include "network.h" | |
50 | #include "raw_syscalls.h" | |
51 | #include "start.h" | |
52 | #include "storage/storage.h" | |
53 | #include "storage/storage_utils.h" | |
54 | #include "sync.h" | |
55 | #include "syscall_wrappers.h" | |
56 | #include "terminal.h" | |
57 | #include "utils.h" | |
58 | ||
59 | #if HAVE_LIBCAP | |
60 | #include <sys/capability.h> | |
61 | #endif | |
62 | ||
63 | #ifndef HAVE_STRLCPY | |
64 | #include "include/strlcpy.h" | |
65 | #endif | |
66 | ||
67 | lxc_log_define(start, lxc); | |
68 | ||
69 | extern void mod_all_rdeps(struct lxc_container *c, bool inc); | |
70 | static bool do_destroy_container(struct lxc_handler *handler); | |
71 | static int lxc_rmdir_onedev_wrapper(void *data); | |
72 | static void lxc_destroy_container_on_signal(struct lxc_handler *handler, | |
73 | const char *name); | |
74 | ||
75 | static void print_top_failing_dir(const char *path) | |
76 | { | |
77 | __do_free char *copy = NULL; | |
78 | int ret; | |
79 | char *e, *p, saved; | |
80 | ||
81 | copy = must_copy_string(path); | |
82 | p = copy; | |
83 | e = copy + strlen(path); | |
84 | ||
85 | while (p < e) { | |
86 | while (p < e && *p == '/') | |
87 | p++; | |
88 | ||
89 | while (p < e && *p != '/') | |
90 | p++; | |
91 | ||
92 | saved = *p; | |
93 | *p = '\0'; | |
94 | ||
95 | ret = access(copy, X_OK); | |
96 | if (ret != 0) { | |
97 | SYSERROR("Could not access %s. Please grant it x access, or add an ACL for the container " "root", copy); | |
98 | return; | |
99 | } | |
100 | *p = saved; | |
101 | } | |
102 | } | |
103 | ||
104 | static void lxc_put_nsfds(struct lxc_handler *handler) | |
105 | { | |
106 | for (int i = 0; i < LXC_NS_MAX; i++) { | |
107 | if (handler->nsfd[i] < 0) | |
108 | continue; | |
109 | ||
110 | close_prot_errno_disarm(handler->nsfd[i]); | |
111 | } | |
112 | } | |
113 | ||
114 | static int lxc_try_preserve_ns(const int pid, const char *ns) | |
115 | { | |
116 | int fd; | |
117 | ||
118 | fd = lxc_preserve_ns(pid, ns); | |
119 | if (fd < 0) { | |
120 | if (errno != ENOENT) | |
121 | return log_error_errno(-EINVAL, | |
122 | errno, "Failed to preserve %s namespace", | |
123 | ns); | |
124 | ||
125 | return log_warn_errno(-EOPNOTSUPP, | |
126 | errno, "Kernel does not support preserving %s namespaces", | |
127 | ns); | |
128 | } | |
129 | ||
130 | return fd; | |
131 | } | |
132 | ||
133 | /* lxc_try_preserve_namespaces: open /proc/@pid/ns/@ns for each namespace | |
134 | * specified in ns_clone_flags. | |
135 | * Return true on success, false on failure. | |
136 | */ | |
137 | static bool lxc_try_preserve_namespaces(struct lxc_handler *handler, | |
138 | int ns_clone_flags, pid_t pid) | |
139 | { | |
140 | int i; | |
141 | ||
142 | for (i = 0; i < LXC_NS_MAX; i++) | |
143 | handler->nsfd[i] = -EBADF; | |
144 | ||
145 | for (i = 0; i < LXC_NS_MAX; i++) { | |
146 | int fd; | |
147 | ||
148 | if ((ns_clone_flags & ns_info[i].clone_flag) == 0) | |
149 | continue; | |
150 | ||
151 | fd = lxc_try_preserve_ns(pid, ns_info[i].proc_name); | |
152 | if (fd < 0) { | |
153 | /* Do not fail to start container on kernels that do | |
154 | * not support interacting with namespaces through | |
155 | * /proc. | |
156 | */ | |
157 | if (fd == -EOPNOTSUPP) | |
158 | continue; | |
159 | ||
160 | lxc_put_nsfds(handler); | |
161 | return false; | |
162 | } | |
163 | ||
164 | handler->nsfd[i] = fd; | |
165 | DEBUG("Preserved %s namespace via fd %d", ns_info[i].proc_name, | |
166 | handler->nsfd[i]); | |
167 | } | |
168 | ||
169 | return true; | |
170 | } | |
171 | ||
172 | static inline bool match_stdfds(int fd) | |
173 | { | |
174 | return (fd == STDIN_FILENO || fd == STDOUT_FILENO || fd == STDERR_FILENO); | |
175 | } | |
176 | ||
177 | #ifdef HAVE_DLOG | |
178 | static bool match_dlog_fds(struct dirent *direntp) | |
179 | { | |
180 | char path[PATH_MAX] = {0}; | |
181 | char link[PATH_MAX] = {0}; | |
182 | ssize_t linklen; | |
183 | int ret; | |
184 | ||
185 | ret = snprintf(path, PATH_MAX, "/proc/self/fd/%s", direntp->d_name); | |
186 | if (ret < 0 || ret >= PATH_MAX) | |
187 | return log_error(false, "Failed to create file descriptor name"); | |
188 | ||
189 | linklen = readlink(path, link, PATH_MAX); | |
190 | if (linklen < 0) | |
191 | return log_error(false, "Failed to read link path - \"%s\"", path); | |
192 | else if (linklen >= PATH_MAX) | |
193 | return log_error(false, "The name of link path is too long - \"%s\"", path); | |
194 | ||
195 | if (strcmp(link, "/dev/log_main") == 0 || | |
196 | strcmp(link, "/dev/log_system") == 0 || | |
197 | strcmp(link, "/dev/log_radio") == 0) | |
198 | return true; | |
199 | ||
200 | return false; | |
201 | } | |
202 | #endif | |
203 | ||
204 | int lxc_check_inherited(struct lxc_conf *conf, bool closeall, | |
205 | int *fds_to_ignore, size_t len_fds) | |
206 | { | |
207 | int fd, fddir; | |
208 | size_t i; | |
209 | DIR *dir; | |
210 | struct dirent *direntp; | |
211 | ||
212 | if (conf && conf->close_all_fds) | |
213 | closeall = true; | |
214 | ||
215 | restart: | |
216 | dir = opendir("/proc/self/fd"); | |
217 | if (!dir) | |
218 | return log_warn(-1, "Failed to open directory"); | |
219 | ||
220 | fddir = dirfd(dir); | |
221 | ||
222 | while ((direntp = readdir(dir))) { | |
223 | int ret; | |
224 | struct lxc_list *cur; | |
225 | bool matched = false; | |
226 | ||
227 | if (strcmp(direntp->d_name, ".") == 0) | |
228 | continue; | |
229 | ||
230 | if (strcmp(direntp->d_name, "..") == 0) | |
231 | continue; | |
232 | ||
233 | ret = lxc_safe_int(direntp->d_name, &fd); | |
234 | if (ret < 0) { | |
235 | INFO("Could not parse file descriptor for \"%s\"", direntp->d_name); | |
236 | continue; | |
237 | } | |
238 | ||
239 | for (i = 0; i < len_fds; i++) | |
240 | if (fds_to_ignore[i] == fd) | |
241 | break; | |
242 | ||
243 | if (fd == fddir || fd == lxc_log_fd || | |
244 | (i < len_fds && fd == fds_to_ignore[i])) | |
245 | continue; | |
246 | ||
247 | /* Keep state clients that wait on reboots. */ | |
248 | if (conf) { | |
249 | lxc_list_for_each(cur, &conf->state_clients) { | |
250 | struct lxc_state_client *client = cur->elem; | |
251 | ||
252 | if (client->clientfd != fd) | |
253 | continue; | |
254 | ||
255 | matched = true; | |
256 | break; | |
257 | } | |
258 | } | |
259 | ||
260 | if (matched) | |
261 | continue; | |
262 | ||
263 | if (current_config && fd == current_config->logfd) | |
264 | continue; | |
265 | ||
266 | if (match_stdfds(fd)) | |
267 | continue; | |
268 | ||
269 | #ifdef HAVE_DLOG | |
270 | if (match_dlog_fds(direntp)) | |
271 | continue; | |
272 | ||
273 | #endif | |
274 | if (closeall) { | |
275 | close(fd); | |
276 | closedir(dir); | |
277 | INFO("Closed inherited fd %d", fd); | |
278 | goto restart; | |
279 | } | |
280 | WARN("Inherited fd %d", fd); | |
281 | } | |
282 | ||
283 | /* Only enable syslog at this point to avoid the above logging function | |
284 | * to open a new fd and make the check_inherited function enter an | |
285 | * infinite loop. | |
286 | */ | |
287 | lxc_log_enable_syslog(); | |
288 | ||
289 | closedir(dir); /* cannot fail */ | |
290 | return 0; | |
291 | } | |
292 | ||
293 | static int setup_signal_fd(sigset_t *oldmask) | |
294 | { | |
295 | int ret; | |
296 | sigset_t mask; | |
297 | const int signals[] = {SIGBUS, SIGILL, SIGSEGV, SIGWINCH}; | |
298 | ||
299 | /* Block everything except serious error signals. */ | |
300 | ret = sigfillset(&mask); | |
301 | if (ret < 0) | |
302 | return -EBADF; | |
303 | ||
304 | for (int sig = 0; sig < (sizeof(signals) / sizeof(signals[0])); sig++) { | |
305 | ret = sigdelset(&mask, signals[sig]); | |
306 | if (ret < 0) | |
307 | return -EBADF; | |
308 | } | |
309 | ||
310 | ret = pthread_sigmask(SIG_BLOCK, &mask, oldmask); | |
311 | if (ret < 0) | |
312 | return log_error_errno(-EBADF, errno, | |
313 | "Failed to set signal mask"); | |
314 | ||
315 | ret = signalfd(-1, &mask, SFD_CLOEXEC); | |
316 | if (ret < 0) | |
317 | return log_error_errno(-EBADF, | |
318 | errno, "Failed to create signal file descriptor"); | |
319 | ||
320 | TRACE("Created signal file descriptor %d", ret); | |
321 | ||
322 | return ret; | |
323 | } | |
324 | ||
325 | static int signal_handler(int fd, uint32_t events, void *data, | |
326 | struct lxc_epoll_descr *descr) | |
327 | { | |
328 | int ret; | |
329 | siginfo_t info; | |
330 | struct signalfd_siginfo siginfo; | |
331 | struct lxc_handler *hdlr = data; | |
332 | ||
333 | ret = lxc_read_nointr(fd, &siginfo, sizeof(siginfo)); | |
334 | if (ret < 0) | |
335 | return log_error(LXC_MAINLOOP_ERROR, "Failed to read signal info from signal file descriptor %d", fd); | |
336 | ||
337 | if (ret != sizeof(siginfo)) | |
338 | return log_error(-EINVAL, "Unexpected size for struct signalfd_siginfo"); | |
339 | ||
340 | /* Check whether init is running. */ | |
341 | info.si_pid = 0; | |
342 | ret = waitid(P_PID, hdlr->pid, &info, WEXITED | WNOWAIT | WNOHANG); | |
343 | if (ret == 0 && info.si_pid == hdlr->pid) | |
344 | hdlr->init_died = true; | |
345 | ||
346 | /* Try to figure out a reasonable exit status to report. */ | |
347 | if (hdlr->init_died) { | |
348 | switch (info.si_code) { | |
349 | case CLD_EXITED: | |
350 | hdlr->exit_status = info.si_status << 8; | |
351 | break; | |
352 | case CLD_KILLED: | |
353 | case CLD_DUMPED: | |
354 | case CLD_STOPPED: | |
355 | hdlr->exit_status = info.si_status << 8 | 0x7f; | |
356 | break; | |
357 | case CLD_CONTINUED: | |
358 | /* Huh? The waitid() told us it's dead *and* continued? */ | |
359 | WARN("Init %d dead and continued?", hdlr->pid); | |
360 | hdlr->exit_status = 1; | |
361 | break; | |
362 | default: | |
363 | ERROR("Unknown si_code: %d", info.si_code); | |
364 | hdlr->exit_status = 1; | |
365 | } | |
366 | } | |
367 | ||
368 | if (siginfo.ssi_signo == SIGHUP) { | |
369 | if (hdlr->pidfd >= 0) | |
370 | lxc_raw_pidfd_send_signal(hdlr->pidfd, SIGTERM, NULL, 0); | |
371 | else | |
372 | kill(hdlr->pid, SIGTERM); | |
373 | INFO("Killing %d since terminal hung up", hdlr->pid); | |
374 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE | |
375 | : LXC_MAINLOOP_CONTINUE; | |
376 | } | |
377 | ||
378 | if (siginfo.ssi_signo != SIGCHLD) { | |
379 | if (hdlr->pidfd >= 0) | |
380 | lxc_raw_pidfd_send_signal(hdlr->pidfd, | |
381 | siginfo.ssi_signo, NULL, 0); | |
382 | else | |
383 | kill(hdlr->pid, siginfo.ssi_signo); | |
384 | INFO("Forwarded signal %d to pid %d", siginfo.ssi_signo, hdlr->pid); | |
385 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE | |
386 | : LXC_MAINLOOP_CONTINUE; | |
387 | } | |
388 | ||
389 | /* More robustness, protect ourself from a SIGCHLD sent | |
390 | * by a process different from the container init. | |
391 | */ | |
392 | if (siginfo.ssi_pid != hdlr->pid) { | |
393 | NOTICE("Received %d from pid %d instead of container init %d", | |
394 | siginfo.ssi_signo, siginfo.ssi_pid, hdlr->pid); | |
395 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE | |
396 | : LXC_MAINLOOP_CONTINUE; | |
397 | } | |
398 | ||
399 | if (siginfo.ssi_code == CLD_STOPPED) { | |
400 | INFO("Container init process was stopped"); | |
401 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE | |
402 | : LXC_MAINLOOP_CONTINUE; | |
403 | } | |
404 | ||
405 | if (siginfo.ssi_code == CLD_CONTINUED) { | |
406 | INFO("Container init process was continued"); | |
407 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE | |
408 | : LXC_MAINLOOP_CONTINUE; | |
409 | } | |
410 | ||
411 | return log_debug(LXC_MAINLOOP_CLOSE, "Container init process %d exited", hdlr->pid); | |
412 | } | |
413 | ||
414 | int lxc_serve_state_clients(const char *name, struct lxc_handler *handler, | |
415 | lxc_state_t state) | |
416 | { | |
417 | size_t retlen; | |
418 | ssize_t ret; | |
419 | struct lxc_list *cur, *next; | |
420 | struct lxc_msg msg = {.type = lxc_msg_state, .value = state}; | |
421 | ||
422 | if (state == THAWED) | |
423 | handler->state = RUNNING; | |
424 | else | |
425 | handler->state = state; | |
426 | ||
427 | TRACE("Set container state to %s", lxc_state2str(state)); | |
428 | ||
429 | if (lxc_list_empty(&handler->conf->state_clients)) | |
430 | return log_trace(0, "No state clients registered"); | |
431 | ||
432 | retlen = strlcpy(msg.name, name, sizeof(msg.name)); | |
433 | if (retlen >= sizeof(msg.name)) | |
434 | return -E2BIG; | |
435 | ||
436 | lxc_list_for_each_safe(cur, &handler->conf->state_clients, next) { | |
437 | struct lxc_state_client *client = cur->elem; | |
438 | ||
439 | if (client->states[state] == 0) { | |
440 | TRACE("State %s not registered for state client %d", | |
441 | lxc_state2str(state), client->clientfd); | |
442 | continue; | |
443 | } | |
444 | ||
445 | TRACE("Sending state %s to state client %d", | |
446 | lxc_state2str(state), client->clientfd); | |
447 | ||
448 | ret = lxc_send_nointr(client->clientfd, &msg, sizeof(msg), MSG_NOSIGNAL); | |
449 | if (ret <= 0) | |
450 | SYSERROR("Failed to send message to client"); | |
451 | ||
452 | /* kick client from list */ | |
453 | lxc_list_del(cur); | |
454 | close(client->clientfd); | |
455 | free(cur->elem); | |
456 | free(cur); | |
457 | } | |
458 | ||
459 | return 0; | |
460 | } | |
461 | ||
462 | static int lxc_serve_state_socket_pair(const char *name, | |
463 | struct lxc_handler *handler, | |
464 | lxc_state_t state) | |
465 | { | |
466 | ssize_t ret; | |
467 | ||
468 | if (!handler->daemonize || | |
469 | handler->state_socket_pair[1] < 0 || | |
470 | state == STARTING) | |
471 | return 0; | |
472 | ||
473 | /* Close read end of the socket pair. */ | |
474 | close_prot_errno_disarm(handler->state_socket_pair[0]); | |
475 | ||
476 | again: | |
477 | ret = lxc_abstract_unix_send_credential(handler->state_socket_pair[1], | |
478 | &(int){state}, sizeof(int)); | |
479 | if (ret < 0) { | |
480 | SYSERROR("Failed to send state to %d", handler->state_socket_pair[1]); | |
481 | ||
482 | if (errno == EINTR) | |
483 | goto again; | |
484 | ||
485 | return -1; | |
486 | } | |
487 | ||
488 | if (ret != sizeof(int)) | |
489 | return log_error(-1, "Message too long : %d", handler->state_socket_pair[1]); | |
490 | ||
491 | TRACE("Sent container state \"%s\" to %d", lxc_state2str(state), | |
492 | handler->state_socket_pair[1]); | |
493 | ||
494 | /* Close write end of the socket pair. */ | |
495 | close_prot_errno_disarm(handler->state_socket_pair[1]); | |
496 | ||
497 | return 0; | |
498 | } | |
499 | ||
500 | int lxc_set_state(const char *name, struct lxc_handler *handler, | |
501 | lxc_state_t state) | |
502 | { | |
503 | int ret; | |
504 | ||
505 | ret = lxc_serve_state_socket_pair(name, handler, state); | |
506 | if (ret < 0) | |
507 | return log_error(-1, "Failed to synchronize via anonymous pair of unix sockets"); | |
508 | ||
509 | ret = lxc_serve_state_clients(name, handler, state); | |
510 | if (ret < 0) | |
511 | return -1; | |
512 | ||
513 | /* This function will try to connect to the legacy lxc-monitord state | |
514 | * server and only exists for backwards compatibility. | |
515 | */ | |
516 | lxc_monitor_send_state(name, state, handler->lxcpath); | |
517 | ||
518 | return 0; | |
519 | } | |
520 | ||
521 | int lxc_poll(const char *name, struct lxc_handler *handler) | |
522 | { | |
523 | int ret; | |
524 | bool has_console = true; | |
525 | struct lxc_epoll_descr descr, descr_console; | |
526 | ||
527 | if (handler->conf->console.path && | |
528 | strcmp(handler->conf->console.path, "none") == 0) | |
529 | has_console = false; | |
530 | ||
531 | ret = lxc_mainloop_open(&descr); | |
532 | if (ret < 0) { | |
533 | ERROR("Failed to create mainloop"); | |
534 | goto out_sigfd; | |
535 | } | |
536 | ||
537 | if (has_console) { | |
538 | ret = lxc_mainloop_open(&descr_console); | |
539 | if (ret < 0) { | |
540 | ERROR("Failed to create console mainloop"); | |
541 | goto out_mainloop; | |
542 | } | |
543 | } | |
544 | ||
545 | ret = lxc_mainloop_add_handler(&descr, handler->sigfd, signal_handler, handler); | |
546 | if (ret < 0) { | |
547 | ERROR("Failed to add signal handler for %d to mainloop", handler->sigfd); | |
548 | goto out_mainloop_console; | |
549 | } | |
550 | ||
551 | ret = lxc_seccomp_setup_proxy(&handler->conf->seccomp, &descr, handler); | |
552 | if (ret < 0) { | |
553 | ERROR("Failed to setup seccomp proxy"); | |
554 | goto out_mainloop_console; | |
555 | } | |
556 | ||
557 | if (has_console) { | |
558 | struct lxc_terminal *console = &handler->conf->console; | |
559 | ||
560 | ret = lxc_terminal_mainloop_add(&descr, console); | |
561 | if (ret < 0) { | |
562 | ERROR("Failed to add console handlers to mainloop"); | |
563 | goto out_mainloop_console; | |
564 | } | |
565 | ||
566 | ret = lxc_terminal_mainloop_add(&descr_console, console); | |
567 | if (ret < 0) { | |
568 | ERROR("Failed to add console handlers to console mainloop"); | |
569 | goto out_mainloop_console; | |
570 | } | |
571 | ||
572 | handler->conf->console.descr = &descr; | |
573 | } | |
574 | ||
575 | ret = lxc_cmd_mainloop_add(name, &descr, handler); | |
576 | if (ret < 0) { | |
577 | ERROR("Failed to add command handler to mainloop"); | |
578 | goto out_mainloop_console; | |
579 | } | |
580 | ||
581 | TRACE("Mainloop is ready"); | |
582 | ||
583 | ret = lxc_mainloop(&descr, -1); | |
584 | close_prot_errno_disarm(descr.epfd); | |
585 | if (ret < 0 || !handler->init_died) | |
586 | goto out_mainloop_console; | |
587 | ||
588 | if (has_console) | |
589 | ret = lxc_mainloop(&descr_console, 0); | |
590 | ||
591 | out_mainloop_console: | |
592 | if (has_console) { | |
593 | lxc_mainloop_close(&descr_console); | |
594 | TRACE("Closed console mainloop"); | |
595 | } | |
596 | ||
597 | out_mainloop: | |
598 | lxc_mainloop_close(&descr); | |
599 | TRACE("Closed mainloop"); | |
600 | ||
601 | out_sigfd: | |
602 | TRACE("Closed signal file descriptor %d", handler->sigfd); | |
603 | close_prot_errno_disarm(handler->sigfd); | |
604 | ||
605 | return ret; | |
606 | } | |
607 | ||
608 | void lxc_zero_handler(struct lxc_handler *handler) | |
609 | { | |
610 | memset(handler, 0, sizeof(struct lxc_handler)); | |
611 | ||
612 | handler->state = STOPPED; | |
613 | ||
614 | handler->pinfd = -EBADF; | |
615 | ||
616 | handler->pidfd = -EBADF; | |
617 | ||
618 | handler->sigfd = -EBADF; | |
619 | ||
620 | for (int i = 0; i < LXC_NS_MAX; i++) | |
621 | handler->nsfd[i] = -EBADF; | |
622 | ||
623 | handler->data_sock[0] = -EBADF; | |
624 | handler->data_sock[1] = -EBADF; | |
625 | ||
626 | handler->state_socket_pair[0] = -EBADF; | |
627 | handler->state_socket_pair[1] = -EBADF; | |
628 | ||
629 | handler->sync_sock[0] = -EBADF; | |
630 | handler->sync_sock[1] = -EBADF; | |
631 | } | |
632 | ||
633 | void lxc_free_handler(struct lxc_handler *handler) | |
634 | { | |
635 | close_prot_errno_disarm(handler->pinfd); | |
636 | close_prot_errno_disarm(handler->pidfd); | |
637 | close_prot_errno_disarm(handler->sigfd); | |
638 | lxc_put_nsfds(handler); | |
639 | if (handler->conf && handler->conf->reboot == REBOOT_NONE) | |
640 | close_prot_errno_disarm(handler->conf->maincmd_fd); | |
641 | close_prot_errno_disarm(handler->monitor_status_fd); | |
642 | close_prot_errno_disarm(handler->state_socket_pair[0]); | |
643 | close_prot_errno_disarm(handler->state_socket_pair[1]); | |
644 | cgroup_exit(handler->cgroup_ops); | |
645 | handler->conf = NULL; | |
646 | free_disarm(handler); | |
647 | } | |
648 | ||
649 | struct lxc_handler *lxc_init_handler(const char *name, struct lxc_conf *conf, | |
650 | const char *lxcpath, bool daemonize) | |
651 | { | |
652 | int ret; | |
653 | struct lxc_handler *handler; | |
654 | ||
655 | handler = malloc(sizeof(*handler)); | |
656 | if (!handler) | |
657 | return NULL; | |
658 | ||
659 | memset(handler, 0, sizeof(*handler)); | |
660 | ||
661 | /* Note that am_guest_unpriv() checks the effective uid. We | |
662 | * probably don't care if we are real root only if we are running | |
663 | * as root so this should be fine. | |
664 | */ | |
665 | handler->am_root = !am_guest_unpriv(); | |
666 | handler->conf = conf; | |
667 | handler->lxcpath = lxcpath; | |
668 | handler->init_died = false; | |
669 | handler->data_sock[0] = -EBADF; | |
670 | handler->data_sock[1] = -EBADF; | |
671 | handler->monitor_status_fd = -EBADF; | |
672 | handler->pinfd = -EBADF; | |
673 | handler->pidfd = -EBADF; | |
674 | handler->sigfd = -EBADF; | |
675 | handler->state_socket_pair[0] = -EBADF; | |
676 | handler->state_socket_pair[1] = -EBADF; | |
677 | if (handler->conf->reboot == REBOOT_NONE) | |
678 | lxc_list_init(&handler->conf->state_clients); | |
679 | ||
680 | for (int i = 0; i < LXC_NS_MAX; i++) | |
681 | handler->nsfd[i] = -EBADF; | |
682 | ||
683 | handler->name = name; | |
684 | if (daemonize) | |
685 | handler->transient_pid = lxc_raw_getpid(); | |
686 | else | |
687 | handler->transient_pid = -1; | |
688 | ||
689 | if (daemonize && handler->conf->reboot == REBOOT_NONE) { | |
690 | /* Create socketpair() to synchronize on daemonized startup. | |
691 | * When the container reboots we don't need to synchronize | |
692 | * again currently so don't open another socketpair(). | |
693 | */ | |
694 | ret = socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0, | |
695 | handler->state_socket_pair); | |
696 | if (ret < 0) { | |
697 | ERROR("Failed to create anonymous pair of unix sockets"); | |
698 | goto on_error; | |
699 | } | |
700 | ||
701 | TRACE("Created anonymous pair {%d,%d} of unix sockets", | |
702 | handler->state_socket_pair[0], | |
703 | handler->state_socket_pair[1]); | |
704 | } | |
705 | ||
706 | if (handler->conf->reboot == REBOOT_NONE) { | |
707 | handler->conf->maincmd_fd = lxc_cmd_init(name, lxcpath, "command"); | |
708 | if (handler->conf->maincmd_fd < 0) { | |
709 | ERROR("Failed to set up command socket"); | |
710 | goto on_error; | |
711 | } | |
712 | } | |
713 | ||
714 | TRACE("Unix domain socket %d for command server is ready", | |
715 | handler->conf->maincmd_fd); | |
716 | ||
717 | return handler; | |
718 | ||
719 | on_error: | |
720 | lxc_free_handler(handler); | |
721 | ||
722 | return NULL; | |
723 | } | |
724 | ||
725 | int lxc_init(const char *name, struct lxc_handler *handler) | |
726 | { | |
727 | __do_close int status_fd = -EBADF; | |
728 | int ret; | |
729 | const char *loglevel; | |
730 | struct lxc_conf *conf = handler->conf; | |
731 | ||
732 | handler->monitor_pid = lxc_raw_getpid(); | |
733 | status_fd = open("/proc/self/status", O_RDONLY | O_CLOEXEC); | |
734 | if (status_fd < 0) | |
735 | return log_error_errno(-1, errno, "Failed to open monitor status fd"); | |
736 | ||
737 | lsm_init(); | |
738 | TRACE("Initialized LSM"); | |
739 | ||
740 | ret = lxc_read_seccomp_config(conf); | |
741 | if (ret < 0) | |
742 | return log_error(-1, "Failed loading seccomp policy"); | |
743 | TRACE("Read seccomp policy"); | |
744 | ||
745 | /* Begin by setting the state to STARTING. */ | |
746 | ret = lxc_set_state(name, handler, STARTING); | |
747 | if (ret < 0) | |
748 | return log_error(-1, "Failed to set state to \"%s\"", lxc_state2str(STARTING)); | |
749 | TRACE("Set container state to \"STARTING\""); | |
750 | ||
751 | /* Start of environment variable setup for hooks. */ | |
752 | ret = setenv("LXC_NAME", name, 1); | |
753 | if (ret < 0) | |
754 | SYSERROR("Failed to set environment variable: LXC_NAME=%s", name); | |
755 | ||
756 | if (conf->rcfile) { | |
757 | ret = setenv("LXC_CONFIG_FILE", conf->rcfile, 1); | |
758 | if (ret < 0) | |
759 | SYSERROR("Failed to set environment variable: LXC_CONFIG_FILE=%s", conf->rcfile); | |
760 | } | |
761 | ||
762 | if (conf->rootfs.mount) { | |
763 | ret = setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1); | |
764 | if (ret < 0) | |
765 | SYSERROR("Failed to set environment variable: LXC_ROOTFS_MOUNT=%s", conf->rootfs.mount); | |
766 | } | |
767 | ||
768 | if (conf->rootfs.path) { | |
769 | ret = setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1); | |
770 | if (ret < 0) | |
771 | SYSERROR("Failed to set environment variable: LXC_ROOTFS_PATH=%s", conf->rootfs.path); | |
772 | } | |
773 | ||
774 | if (conf->console.path) { | |
775 | ret = setenv("LXC_CONSOLE", conf->console.path, 1); | |
776 | if (ret < 0) | |
777 | SYSERROR("Failed to set environment variable: LXC_CONSOLE=%s", conf->console.path); | |
778 | } | |
779 | ||
780 | if (conf->console.log_path) { | |
781 | ret = setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1); | |
782 | if (ret < 0) | |
783 | SYSERROR("Failed to set environment variable: LXC_CONSOLE_LOGPATH=%s", conf->console.log_path); | |
784 | } | |
785 | ||
786 | if (cgns_supported()) { | |
787 | ret = setenv("LXC_CGNS_AWARE", "1", 1); | |
788 | if (ret < 0) | |
789 | SYSERROR("Failed to set environment variable LXC_CGNS_AWARE=1"); | |
790 | } | |
791 | ||
792 | loglevel = lxc_log_priority_to_string(lxc_log_get_level()); | |
793 | ret = setenv("LXC_LOG_LEVEL", loglevel, 1); | |
794 | if (ret < 0) | |
795 | SYSERROR("Set environment variable LXC_LOG_LEVEL=%s", loglevel); | |
796 | ||
797 | if (conf->hooks_version == 0) | |
798 | ret = setenv("LXC_HOOK_VERSION", "0", 1); | |
799 | else | |
800 | ret = setenv("LXC_HOOK_VERSION", "1", 1); | |
801 | if (ret < 0) | |
802 | SYSERROR("Failed to set environment variable LXC_HOOK_VERSION=%u", conf->hooks_version); | |
803 | /* End of environment variable setup for hooks. */ | |
804 | ||
805 | TRACE("Set environment variables"); | |
806 | ||
807 | ret = run_lxc_hooks(name, "pre-start", conf, NULL); | |
808 | if (ret < 0) | |
809 | return log_error(-1, "Failed to run lxc.hook.pre-start for container \"%s\"", name); | |
810 | TRACE("Ran pre-start hooks"); | |
811 | ||
812 | /* The signal fd has to be created before forking otherwise if the child | |
813 | * process exits before we setup the signal fd, the event will be lost | |
814 | * and the command will be stuck. | |
815 | */ | |
816 | handler->sigfd = setup_signal_fd(&handler->oldmask); | |
817 | if (handler->sigfd < 0) | |
818 | return log_error(-1, "Failed to setup SIGCHLD fd handler."); | |
819 | TRACE("Set up signal fd"); | |
820 | ||
821 | /* Do this after setting up signals since it might unblock SIGWINCH. */ | |
822 | ret = lxc_terminal_setup(conf); | |
823 | if (ret < 0) { | |
824 | ERROR("Failed to create console"); | |
825 | goto out_restore_sigmask; | |
826 | } | |
827 | TRACE("Created console"); | |
828 | ||
829 | ret = lxc_terminal_map_ids(conf, &conf->console); | |
830 | if (ret < 0) { | |
831 | ERROR("Failed to chown console"); | |
832 | goto out_delete_terminal; | |
833 | } | |
834 | TRACE("Chowned console"); | |
835 | ||
836 | handler->cgroup_ops = cgroup_init(handler->conf); | |
837 | if (!handler->cgroup_ops) { | |
838 | ERROR("Failed to initialize cgroup driver"); | |
839 | goto out_delete_terminal; | |
840 | } | |
841 | TRACE("Initialized cgroup driver"); | |
842 | ||
843 | ret = lsm_process_prepare(conf, handler->lxcpath); | |
844 | if (ret < 0) { | |
845 | ERROR("Failed to initialize LSM"); | |
846 | goto out_delete_terminal; | |
847 | } | |
848 | TRACE("Initialized LSM"); | |
849 | ||
850 | INFO("Container \"%s\" is initialized", name); | |
851 | handler->monitor_status_fd = move_fd(status_fd); | |
852 | return 0; | |
853 | ||
854 | out_delete_terminal: | |
855 | lxc_terminal_delete(&handler->conf->console); | |
856 | ||
857 | out_restore_sigmask: | |
858 | (void)pthread_sigmask(SIG_SETMASK, &handler->oldmask, NULL); | |
859 | ||
860 | return -1; | |
861 | } | |
862 | ||
863 | void lxc_fini(struct lxc_handler *handler) | |
864 | { | |
865 | int ret; | |
866 | pid_t self; | |
867 | struct lxc_list *cur, *next; | |
868 | char *namespaces[LXC_NS_MAX + 1]; | |
869 | size_t namespace_count = 0; | |
870 | const char *name = handler->name; | |
871 | struct cgroup_ops *cgroup_ops = handler->cgroup_ops; | |
872 | ||
873 | /* The STOPPING state is there for future cleanup code which can take | |
874 | * awhile. | |
875 | */ | |
876 | lxc_set_state(name, handler, STOPPING); | |
877 | ||
878 | self = lxc_raw_getpid(); | |
879 | for (int i = 0; i < LXC_NS_MAX; i++) { | |
880 | if (handler->nsfd[i] < 0) | |
881 | continue; | |
882 | ||
883 | if (handler->conf->hooks_version == 0) | |
884 | ret = asprintf(&namespaces[namespace_count], | |
885 | "%s:/proc/%d/fd/%d", ns_info[i].proc_name, | |
886 | self, handler->nsfd[i]); | |
887 | else | |
888 | ret = asprintf(&namespaces[namespace_count], | |
889 | "/proc/%d/fd/%d", self, handler->nsfd[i]); | |
890 | if (ret < 0) { | |
891 | SYSERROR("Failed to allocate memory"); | |
892 | break; | |
893 | } | |
894 | ||
895 | if (handler->conf->hooks_version == 0) { | |
896 | namespace_count++; | |
897 | continue; | |
898 | } | |
899 | ||
900 | ret = setenv(ns_info[i].env_name, namespaces[namespace_count], 1); | |
901 | if (ret < 0) | |
902 | SYSERROR("Failed to set environment variable %s=%s", | |
903 | ns_info[i].env_name, namespaces[namespace_count]); | |
904 | else | |
905 | TRACE("Set environment variable %s=%s", | |
906 | ns_info[i].env_name, namespaces[namespace_count]); | |
907 | ||
908 | namespace_count++; | |
909 | } | |
910 | namespaces[namespace_count] = NULL; | |
911 | ||
912 | if (handler->conf->reboot > REBOOT_NONE) { | |
913 | ret = setenv("LXC_TARGET", "reboot", 1); | |
914 | if (ret < 0) | |
915 | SYSERROR("Failed to set environment variable: LXC_TARGET=reboot"); | |
916 | } | |
917 | ||
918 | if (handler->conf->reboot == REBOOT_NONE) { | |
919 | ret = setenv("LXC_TARGET", "stop", 1); | |
920 | if (ret < 0) | |
921 | SYSERROR("Failed to set environment variable: LXC_TARGET=stop"); | |
922 | } | |
923 | ||
924 | if (handler->conf->hooks_version == 0) | |
925 | ret = run_lxc_hooks(name, "stop", handler->conf, namespaces); | |
926 | else | |
927 | ret = run_lxc_hooks(name, "stop", handler->conf, NULL); | |
928 | if (ret < 0) | |
929 | ERROR("Failed to run \"lxc.hook.stop\" hook"); | |
930 | ||
931 | while (namespace_count--) | |
932 | free(namespaces[namespace_count]); | |
933 | ||
934 | lsm_process_cleanup(handler->conf, handler->lxcpath); | |
935 | ||
936 | if (cgroup_ops) { | |
937 | cgroup_ops->payload_destroy(cgroup_ops, handler); | |
938 | cgroup_ops->monitor_destroy(cgroup_ops, handler); | |
939 | } | |
940 | ||
941 | if (handler->conf->reboot == REBOOT_NONE) { | |
942 | /* For all new state clients simply close the command socket. | |
943 | * This will inform all state clients that the container is | |
944 | * STOPPED and also prevents a race between a open()/close() on | |
945 | * the command socket causing a new process to get ECONNREFUSED | |
946 | * because we haven't yet closed the command socket. | |
947 | */ | |
948 | close_prot_errno_disarm(handler->conf->maincmd_fd); | |
949 | TRACE("Closed command socket"); | |
950 | ||
951 | /* This function will try to connect to the legacy lxc-monitord | |
952 | * state server and only exists for backwards compatibility. | |
953 | */ | |
954 | lxc_monitor_send_state(name, STOPPED, handler->lxcpath); | |
955 | ||
956 | /* The command socket is closed so no one can acces the command | |
957 | * socket anymore so there's no need to lock it. | |
958 | */ | |
959 | handler->state = STOPPED; | |
960 | TRACE("Set container state to \"STOPPED\""); | |
961 | } else { | |
962 | lxc_set_state(name, handler, STOPPED); | |
963 | TRACE("Set container state to \"STOPPED\""); | |
964 | } | |
965 | ||
966 | /* Avoid lingering namespace references. */ | |
967 | lxc_put_nsfds(handler); | |
968 | ||
969 | ret = run_lxc_hooks(name, "post-stop", handler->conf, NULL); | |
970 | if (ret < 0) { | |
971 | ERROR("Failed to run lxc.hook.post-stop for container \"%s\"", name); | |
972 | if (handler->conf->reboot > REBOOT_NONE) { | |
973 | WARN("Container will be stopped instead of rebooted"); | |
974 | handler->conf->reboot = REBOOT_NONE; | |
975 | ||
976 | ret = setenv("LXC_TARGET", "stop", 1); | |
977 | if (ret < 0) | |
978 | WARN("Failed to set environment variable: LXC_TARGET=stop"); | |
979 | } | |
980 | } | |
981 | ||
982 | /* Reset mask set by setup_signal_fd. */ | |
983 | ret = pthread_sigmask(SIG_SETMASK, &handler->oldmask, NULL); | |
984 | if (ret < 0) | |
985 | SYSWARN("Failed to restore signal mask"); | |
986 | ||
987 | lxc_terminal_delete(&handler->conf->console); | |
988 | lxc_delete_tty(&handler->conf->ttys); | |
989 | ||
990 | /* The command socket is now closed, no more state clients can register | |
991 | * themselves from now on. So free the list of state clients. | |
992 | */ | |
993 | lxc_list_for_each_safe(cur, &handler->conf->state_clients, next) { | |
994 | struct lxc_state_client *client = cur->elem; | |
995 | ||
996 | /* Keep state clients that want to be notified about reboots. */ | |
997 | if ((handler->conf->reboot > REBOOT_NONE) && | |
998 | (client->states[RUNNING] == 2)) | |
999 | continue; | |
1000 | ||
1001 | /* close state client socket */ | |
1002 | lxc_list_del(cur); | |
1003 | close(client->clientfd); | |
1004 | free(cur->elem); | |
1005 | free(cur); | |
1006 | } | |
1007 | ||
1008 | if (handler->conf->ephemeral == 1 && handler->conf->reboot != REBOOT_REQ) | |
1009 | lxc_destroy_container_on_signal(handler, name); | |
1010 | ||
1011 | lxc_free_handler(handler); | |
1012 | } | |
1013 | ||
1014 | void lxc_abort(struct lxc_handler *handler) | |
1015 | { | |
1016 | int ret = 0; | |
1017 | int status; | |
1018 | ||
1019 | lxc_set_state(handler->name, handler, ABORTING); | |
1020 | ||
1021 | if (handler->pidfd >= 0) { | |
1022 | ret = lxc_raw_pidfd_send_signal(handler->pidfd, SIGKILL, NULL, 0); | |
1023 | if (ret) | |
1024 | SYSWARN("Failed to send SIGKILL via pidfd %d for process %d", | |
1025 | handler->pidfd, handler->pid); | |
1026 | } | |
1027 | ||
1028 | if ((!ret || errno != ESRCH) && handler->pid > 0) | |
1029 | if (kill(handler->pid, SIGKILL)) | |
1030 | SYSWARN("Failed to send SIGKILL to %d", handler->pid); | |
1031 | ||
1032 | do { | |
1033 | ret = waitpid(-1, &status, 0); | |
1034 | } while (ret > 0); | |
1035 | } | |
1036 | ||
1037 | static int do_start(void *data) | |
1038 | { | |
1039 | struct lxc_handler *handler = data; | |
1040 | __lxc_unused __do_close int data_sock0 = handler->data_sock[0], | |
1041 | data_sock1 = handler->data_sock[1]; | |
1042 | __do_close int status_fd = -EBADF; | |
1043 | int ret; | |
1044 | uid_t new_uid; | |
1045 | gid_t new_gid; | |
1046 | struct lxc_list *iterator; | |
1047 | uid_t nsuid = 0; | |
1048 | gid_t nsgid = 0; | |
1049 | int devnull_fd = -1; | |
1050 | ||
1051 | lxc_sync_fini_parent(handler); | |
1052 | ||
1053 | if (lxc_abstract_unix_recv_fds(data_sock1, &status_fd, 1, NULL, 0) < 0) { | |
1054 | ERROR("Failed to receive status file descriptor to child process"); | |
1055 | goto out_warn_father; | |
1056 | } | |
1057 | ||
1058 | /* This prctl must be before the synchro, so if the parent dies before | |
1059 | * we set the parent death signal, we will detect its death with the | |
1060 | * synchro right after, otherwise we have a window where the parent can | |
1061 | * exit before we set the pdeath signal leading to a unsupervized | |
1062 | * container. | |
1063 | */ | |
1064 | ret = lxc_set_death_signal(SIGKILL, handler->monitor_pid, status_fd); | |
1065 | if (ret < 0) { | |
1066 | SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL"); | |
1067 | goto out_warn_father; | |
1068 | } | |
1069 | ||
1070 | ret = lxc_ambient_caps_up(); | |
1071 | if (ret < 0) { | |
1072 | ERROR("Failed to raise ambient capabilities"); | |
1073 | goto out_warn_father; | |
1074 | } | |
1075 | ||
1076 | ret = pthread_sigmask(SIG_SETMASK, &handler->oldmask, NULL); | |
1077 | if (ret < 0) { | |
1078 | SYSERROR("Failed to set signal mask"); | |
1079 | goto out_warn_father; | |
1080 | } | |
1081 | ||
1082 | /* Don't leak the pinfd to the container. */ | |
1083 | close_prot_errno_disarm(handler->pinfd); | |
1084 | ||
1085 | ret = lxc_sync_wait_parent(handler, LXC_SYNC_STARTUP); | |
1086 | if (ret < 0) | |
1087 | goto out_warn_father; | |
1088 | ||
1089 | /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See | |
1090 | * https://github.com/lxc/lxd/issues/1978. | |
1091 | */ | |
1092 | if ((handler->ns_clone_flags & (CLONE_NEWNET | CLONE_NEWUSER)) == | |
1093 | (CLONE_NEWNET | CLONE_NEWUSER)) { | |
1094 | ret = unshare(CLONE_NEWNET); | |
1095 | if (ret < 0) { | |
1096 | SYSERROR("Failed to unshare CLONE_NEWNET"); | |
1097 | goto out_warn_father; | |
1098 | } | |
1099 | INFO("Unshared CLONE_NEWNET"); | |
1100 | } | |
1101 | ||
1102 | /* Tell the parent task it can begin to configure the container and wait | |
1103 | * for it to finish. | |
1104 | */ | |
1105 | ret = lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE); | |
1106 | if (ret < 0) | |
1107 | goto out_error; | |
1108 | ||
1109 | if (handler->ns_clone_flags & CLONE_NEWNET) { | |
1110 | ret = lxc_network_recv_from_parent(handler); | |
1111 | if (ret < 0) { | |
1112 | ERROR("Failed to receive veth names from parent"); | |
1113 | goto out_warn_father; | |
1114 | } | |
1115 | } | |
1116 | ||
1117 | /* If we are in a new user namespace, become root there to have | |
1118 | * privilege over our namespace. | |
1119 | */ | |
1120 | if (!lxc_list_empty(&handler->conf->id_map)) { | |
1121 | if (!handler->conf->root_nsuid_map) | |
1122 | nsuid = handler->conf->init_uid; | |
1123 | ||
1124 | if (!handler->conf->root_nsgid_map) | |
1125 | nsgid = handler->conf->init_gid; | |
1126 | ||
1127 | /* Drop groups only after we switched to a valid gid in the new | |
1128 | * user namespace. | |
1129 | */ | |
1130 | if (!lxc_setgroups(0, NULL) && | |
1131 | (handler->am_root || errno != EPERM)) | |
1132 | goto out_warn_father; | |
1133 | ||
1134 | if (!lxc_switch_uid_gid(nsuid, nsgid)) | |
1135 | goto out_warn_father; | |
1136 | ||
1137 | ret = prctl(PR_SET_DUMPABLE, prctl_arg(1), prctl_arg(0), | |
1138 | prctl_arg(0), prctl_arg(0)); | |
1139 | if (ret < 0) | |
1140 | goto out_warn_father; | |
1141 | ||
1142 | /* set{g,u}id() clears deathsignal */ | |
1143 | ret = lxc_set_death_signal(SIGKILL, handler->monitor_pid, status_fd); | |
1144 | if (ret < 0) { | |
1145 | SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL"); | |
1146 | goto out_warn_father; | |
1147 | } | |
1148 | } | |
1149 | ||
1150 | ret = access(handler->lxcpath, X_OK); | |
1151 | if (ret != 0) { | |
1152 | print_top_failing_dir(handler->lxcpath); | |
1153 | goto out_warn_father; | |
1154 | } | |
1155 | ||
1156 | /* In order to checkpoint restore, we need to have everything in the | |
1157 | * same mount namespace. However, some containers may not have a | |
1158 | * reasonable /dev (in particular, they may not have /dev/null), so we | |
1159 | * can't set init's std fds to /dev/null by opening it from inside the | |
1160 | * container. | |
1161 | * | |
1162 | * If that's the case, fall back to using the host's /dev/null. This | |
1163 | * means that migration won't work, but at least we won't spew output | |
1164 | * where it isn't wanted. | |
1165 | */ | |
1166 | if (handler->daemonize && !handler->conf->autodev) { | |
1167 | char path[PATH_MAX]; | |
1168 | ||
1169 | ret = snprintf(path, sizeof(path), "%s/dev/null", | |
1170 | handler->conf->rootfs.mount); | |
1171 | if (ret < 0 || ret >= sizeof(path)) | |
1172 | goto out_warn_father; | |
1173 | ||
1174 | ret = access(path, F_OK); | |
1175 | if (ret != 0) { | |
1176 | devnull_fd = open_devnull(); | |
1177 | ||
1178 | if (devnull_fd < 0) | |
1179 | goto out_warn_father; | |
1180 | WARN("Using /dev/null from the host for container init's standard file descriptors. Migration will not work"); | |
1181 | } | |
1182 | } | |
1183 | ||
1184 | /* Ask father to setup cgroups and wait for him to finish. */ | |
1185 | ret = lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP); | |
1186 | if (ret < 0) | |
1187 | goto out_error; | |
1188 | ||
1189 | /* Unshare cgroup namespace after we have setup our cgroups. If we do it | |
1190 | * earlier we end up with a wrong view of /proc/self/cgroup. For | |
1191 | * example, assume we unshare(CLONE_NEWCGROUP) first, and then create | |
1192 | * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then | |
1193 | * /proc/self/cgroup would show us: | |
1194 | * | |
1195 | * 8:cpuset:/lxc/c | |
1196 | * | |
1197 | * whereas it should actually show | |
1198 | * | |
1199 | * 8:cpuset:/ | |
1200 | */ | |
1201 | if (handler->ns_clone_flags & CLONE_NEWCGROUP) { | |
1202 | ret = unshare(CLONE_NEWCGROUP); | |
1203 | if (ret < 0) { | |
1204 | if (errno != EINVAL) { | |
1205 | SYSERROR("Failed to unshare CLONE_NEWCGROUP"); | |
1206 | goto out_warn_father; | |
1207 | } | |
1208 | ||
1209 | handler->ns_clone_flags &= ~CLONE_NEWCGROUP; | |
1210 | SYSINFO("Kernel does not support CLONE_NEWCGROUP"); | |
1211 | } else { | |
1212 | INFO("Unshared CLONE_NEWCGROUP"); | |
1213 | } | |
1214 | } | |
1215 | ||
1216 | /* Add the requested environment variables to the current environment to | |
1217 | * allow them to be used by the various hooks, such as the start hook | |
1218 | * below. | |
1219 | */ | |
1220 | lxc_list_for_each(iterator, &handler->conf->environment) { | |
1221 | ret = putenv((char *)iterator->elem); | |
1222 | if (ret < 0) { | |
1223 | SYSERROR("Failed to set environment variable: %s", | |
1224 | (char *)iterator->elem); | |
1225 | goto out_warn_father; | |
1226 | } | |
1227 | } | |
1228 | ||
1229 | /* Setup the container, ip, names, utsname, ... */ | |
1230 | ret = lxc_setup(handler); | |
1231 | if (ret < 0) { | |
1232 | ERROR("Failed to setup container \"%s\"", handler->name); | |
1233 | goto out_warn_father; | |
1234 | } | |
1235 | ||
1236 | /* Set the label to change to when we exec(2) the container's init. */ | |
1237 | ret = lsm_process_label_set(NULL, handler->conf, true); | |
1238 | if (ret < 0) | |
1239 | goto out_warn_father; | |
1240 | ||
1241 | /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it | |
1242 | * before we aren't allowed anymore. | |
1243 | */ | |
1244 | if (handler->conf->no_new_privs) { | |
1245 | ret = prctl(PR_SET_NO_NEW_PRIVS, prctl_arg(1), prctl_arg(0), | |
1246 | prctl_arg(0), prctl_arg(0)); | |
1247 | if (ret < 0) { | |
1248 | SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges"); | |
1249 | goto out_warn_father; | |
1250 | } | |
1251 | DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges"); | |
1252 | } | |
1253 | ||
1254 | /* Some init's such as busybox will set sane tty settings on stdin, | |
1255 | * stdout, stderr which it thinks is the console. We already set them | |
1256 | * the way we wanted on the real terminal, and we want init to do its | |
1257 | * setup on its console ie. the pty allocated in lxc_terminal_setup() so | |
1258 | * make sure that that pty is stdin,stdout,stderr. | |
1259 | */ | |
1260 | if (handler->conf->console.slave >= 0) { | |
1261 | if (handler->daemonize || !handler->conf->is_execute) | |
1262 | ret = set_stdfds(handler->conf->console.slave); | |
1263 | else | |
1264 | ret = lxc_terminal_set_stdfds(handler->conf->console.slave); | |
1265 | if (ret < 0) { | |
1266 | ERROR("Failed to redirect std{in,out,err} to pty file descriptor %d", | |
1267 | handler->conf->console.slave); | |
1268 | goto out_warn_father; | |
1269 | } | |
1270 | } | |
1271 | ||
1272 | /* If we mounted a temporary proc, then unmount it now. */ | |
1273 | tmp_proc_unmount(handler->conf); | |
1274 | ||
1275 | ret = lxc_seccomp_load(handler->conf); | |
1276 | if (ret < 0) | |
1277 | goto out_warn_father; | |
1278 | ||
1279 | ret = lxc_seccomp_send_notifier_fd(&handler->conf->seccomp, data_sock0); | |
1280 | if (ret < 0) { | |
1281 | SYSERROR("Failed to send seccomp notify fd to parent"); | |
1282 | goto out_warn_father; | |
1283 | } | |
1284 | ||
1285 | ret = run_lxc_hooks(handler->name, "start", handler->conf, NULL); | |
1286 | if (ret < 0) { | |
1287 | ERROR("Failed to run lxc.hook.start for container \"%s\"", | |
1288 | handler->name); | |
1289 | goto out_warn_father; | |
1290 | } | |
1291 | ||
1292 | close_prot_errno_disarm(handler->sigfd); | |
1293 | ||
1294 | if (handler->conf->console.slave < 0 && handler->daemonize) { | |
1295 | if (devnull_fd < 0) { | |
1296 | devnull_fd = open_devnull(); | |
1297 | if (devnull_fd < 0) | |
1298 | goto out_warn_father; | |
1299 | } | |
1300 | ||
1301 | ret = set_stdfds(devnull_fd); | |
1302 | if (ret < 0) { | |
1303 | ERROR("Failed to redirect std{in,out,err} to \"/dev/null\""); | |
1304 | goto out_warn_father; | |
1305 | } | |
1306 | } | |
1307 | ||
1308 | close_prot_errno_disarm(devnull_fd); | |
1309 | ||
1310 | setsid(); | |
1311 | ||
1312 | if (handler->conf->init_cwd) { | |
1313 | ret = chdir(handler->conf->init_cwd); | |
1314 | if (ret < 0) { | |
1315 | SYSERROR("Could not change directory to \"%s\"", | |
1316 | handler->conf->init_cwd); | |
1317 | goto out_warn_father; | |
1318 | } | |
1319 | } | |
1320 | ||
1321 | ret = lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP_LIMITS); | |
1322 | if (ret < 0) | |
1323 | goto out_warn_father; | |
1324 | ||
1325 | /* Reset the environment variables the user requested in a clear | |
1326 | * environment. | |
1327 | */ | |
1328 | ret = clearenv(); | |
1329 | /* Don't error out though. */ | |
1330 | if (ret < 0) | |
1331 | SYSERROR("Failed to clear environment."); | |
1332 | ||
1333 | lxc_list_for_each(iterator, &handler->conf->environment) { | |
1334 | ret = putenv((char *)iterator->elem); | |
1335 | if (ret < 0) { | |
1336 | SYSERROR("Failed to set environment variable: %s", | |
1337 | (char *)iterator->elem); | |
1338 | goto out_warn_father; | |
1339 | } | |
1340 | } | |
1341 | ||
1342 | ret = putenv("container=lxc"); | |
1343 | if (ret < 0) { | |
1344 | SYSERROR("Failed to set environment variable: container=lxc"); | |
1345 | goto out_warn_father; | |
1346 | } | |
1347 | ||
1348 | if (handler->conf->ttys.tty_names) { | |
1349 | ret = putenv(handler->conf->ttys.tty_names); | |
1350 | if (ret < 0) { | |
1351 | SYSERROR("Failed to set environment variable for container ptys"); | |
1352 | goto out_warn_father; | |
1353 | } | |
1354 | } | |
1355 | ||
1356 | /* The container has been setup. We can now switch to an unprivileged | |
1357 | * uid/gid. | |
1358 | */ | |
1359 | new_uid = handler->conf->init_uid; | |
1360 | new_gid = handler->conf->init_gid; | |
1361 | ||
1362 | /* Avoid unnecessary syscalls. */ | |
1363 | if (new_uid == nsuid) | |
1364 | new_uid = LXC_INVALID_UID; | |
1365 | ||
1366 | if (new_gid == nsgid) | |
1367 | new_gid = LXC_INVALID_GID; | |
1368 | ||
1369 | /* If we are in a new user namespace we already dropped all groups when | |
1370 | * we switched to root in the new user namespace further above. Only | |
1371 | * drop groups if we can, so ensure that we have necessary privilege. | |
1372 | */ | |
1373 | if (lxc_list_empty(&handler->conf->id_map)) | |
1374 | #if HAVE_LIBCAP | |
1375 | if (lxc_proc_cap_is_set(CAP_SETGID, CAP_EFFECTIVE)) | |
1376 | #endif | |
1377 | if (!lxc_setgroups(0, NULL)) | |
1378 | goto out_warn_father; | |
1379 | ||
1380 | if (!lxc_switch_uid_gid(new_uid, new_gid)) | |
1381 | goto out_warn_father; | |
1382 | ||
1383 | ret = lxc_ambient_caps_down(); | |
1384 | if (ret < 0) { | |
1385 | ERROR("Failed to clear ambient capabilities"); | |
1386 | goto out_warn_father; | |
1387 | } | |
1388 | ||
1389 | if (handler->conf->monitor_signal_pdeath != SIGKILL) { | |
1390 | ret = lxc_set_death_signal(handler->conf->monitor_signal_pdeath, | |
1391 | handler->monitor_pid, status_fd); | |
1392 | if (ret < 0) { | |
1393 | SYSERROR("Failed to set PR_SET_PDEATHSIG to %d", | |
1394 | handler->conf->monitor_signal_pdeath); | |
1395 | goto out_warn_father; | |
1396 | } | |
1397 | } | |
1398 | ||
1399 | /* After this call, we are in error because this ops should not return | |
1400 | * as it execs. | |
1401 | */ | |
1402 | handler->ops->start(handler, handler->data); | |
1403 | ||
1404 | out_warn_father: | |
1405 | /* We want the parent to know something went wrong, so we return a | |
1406 | * special error code. | |
1407 | */ | |
1408 | lxc_sync_wake_parent(handler, LXC_SYNC_ERROR); | |
1409 | ||
1410 | out_error: | |
1411 | close_prot_errno_disarm(devnull_fd); | |
1412 | ||
1413 | return -1; | |
1414 | } | |
1415 | ||
1416 | static int lxc_recv_ttys_from_child(struct lxc_handler *handler) | |
1417 | { | |
1418 | int i; | |
1419 | struct lxc_terminal_info *tty; | |
1420 | int ret = -1; | |
1421 | int sock = handler->data_sock[1]; | |
1422 | struct lxc_conf *conf = handler->conf; | |
1423 | struct lxc_tty_info *ttys = &conf->ttys; | |
1424 | ||
1425 | if (!conf->ttys.max) | |
1426 | return 0; | |
1427 | ||
1428 | ttys->tty = malloc(sizeof(*ttys->tty) * ttys->max); | |
1429 | if (!ttys->tty) | |
1430 | return -1; | |
1431 | ||
1432 | for (i = 0; i < conf->ttys.max; i++) { | |
1433 | int ttyfds[2]; | |
1434 | ||
1435 | ret = lxc_abstract_unix_recv_fds(sock, ttyfds, 2, NULL, 0); | |
1436 | if (ret < 0) | |
1437 | break; | |
1438 | ||
1439 | tty = &ttys->tty[i]; | |
1440 | tty->busy = -1; | |
1441 | tty->master = ttyfds[0]; | |
1442 | tty->slave = ttyfds[1]; | |
1443 | TRACE("Received pty with master fd %d and slave fd %d from child", tty->master, tty->slave); | |
1444 | } | |
1445 | ||
1446 | if (ret < 0) | |
1447 | SYSERROR("Failed to receive %zu ttys from child", ttys->max); | |
1448 | else | |
1449 | TRACE("Received %zu ttys from child", ttys->max); | |
1450 | ||
1451 | return ret; | |
1452 | } | |
1453 | ||
1454 | int resolve_clone_flags(struct lxc_handler *handler) | |
1455 | { | |
1456 | int i; | |
1457 | struct lxc_conf *conf = handler->conf; | |
1458 | ||
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; | |
1466 | } else { | |
1467 | if (i == LXC_NS_USER && lxc_list_empty(&handler->conf->id_map)) | |
1468 | continue; | |
1469 | ||
1470 | if (i == LXC_NS_NET && lxc_requests_empty_network(handler)) | |
1471 | continue; | |
1472 | ||
1473 | if (i == LXC_NS_CGROUP && !cgns_supported()) | |
1474 | continue; | |
1475 | ||
1476 | handler->ns_clone_flags |= ns_info[i].clone_flag; | |
1477 | } | |
1478 | ||
1479 | if (!conf->ns_share[i]) | |
1480 | continue; | |
1481 | ||
1482 | handler->ns_clone_flags &= ~ns_info[i].clone_flag; | |
1483 | TRACE("Sharing %s namespace", ns_info[i].proc_name); | |
1484 | } | |
1485 | ||
1486 | return 0; | |
1487 | } | |
1488 | ||
1489 | /* Note that this function is used with clone(CLONE_VM). Some glibc versions | |
1490 | * used to reset the pid/tid to -1 when CLONE_VM was used without CLONE_THREAD. | |
1491 | * But since the memory between parent and child is shared on CLONE_VM this | |
1492 | * would invalidate the getpid() cache that glibc used to maintain and so | |
1493 | * getpid() in the child would return the parent's pid. This is all fixed in | |
1494 | * newer glibc versions where the getpid() cache is removed and the pid/tid is | |
1495 | * not reset anymore. | |
1496 | * However, if for whatever reason you - dear committer - somehow need to get the | |
1497 | * pid of the dummy intermediate process for do_share_ns() you need to call | |
1498 | * lxc_raw_getpid(). The next lxc_raw_clone() call does not employ CLONE_VM and | |
1499 | * will be fine. | |
1500 | */ | |
1501 | static inline int do_share_ns(void *arg) | |
1502 | { | |
1503 | int i, flags, ret; | |
1504 | struct lxc_handler *handler = arg; | |
1505 | ||
1506 | for (i = 0; i < LXC_NS_MAX; i++) { | |
1507 | if (handler->nsfd[i] < 0) | |
1508 | continue; | |
1509 | ||
1510 | ret = setns(handler->nsfd[i], 0); | |
1511 | if (ret < 0) { | |
1512 | /* | |
1513 | * Note that joining a user and/or mount namespace | |
1514 | * requires the process is not multithreaded otherwise | |
1515 | * setns() will fail here. | |
1516 | */ | |
1517 | SYSERROR("Failed to inherit %s namespace", | |
1518 | ns_info[i].proc_name); | |
1519 | return -1; | |
1520 | } | |
1521 | ||
1522 | DEBUG("Inherited %s namespace", ns_info[i].proc_name); | |
1523 | } | |
1524 | ||
1525 | flags = handler->ns_on_clone_flags; | |
1526 | flags |= CLONE_PARENT; | |
1527 | handler->pid = lxc_raw_clone_cb(do_start, handler, CLONE_PIDFD | flags, | |
1528 | &handler->pidfd); | |
1529 | if (handler->pid < 0) | |
1530 | return -1; | |
1531 | ||
1532 | return 0; | |
1533 | } | |
1534 | ||
1535 | /* lxc_spawn() performs crucial setup tasks and clone()s the new process which | |
1536 | * exec()s the requested container binary. | |
1537 | * Note that lxc_spawn() runs in the parent namespaces. Any operations performed | |
1538 | * right here should be double checked if they'd pose a security risk. (For | |
1539 | * example, any {u}mount() operations performed here will be reflected on the | |
1540 | * host!) | |
1541 | */ | |
1542 | static int lxc_spawn(struct lxc_handler *handler) | |
1543 | { | |
1544 | __do_close int data_sock0 = -EBADF, data_sock1 = -EBADF; | |
1545 | int i, ret; | |
1546 | char pidstr[20]; | |
1547 | bool wants_to_map_ids; | |
1548 | struct lxc_list *id_map; | |
1549 | const char *name = handler->name; | |
1550 | const char *lxcpath = handler->lxcpath; | |
1551 | bool share_ns = false; | |
1552 | struct lxc_conf *conf = handler->conf; | |
1553 | struct cgroup_ops *cgroup_ops = handler->cgroup_ops; | |
1554 | ||
1555 | id_map = &conf->id_map; | |
1556 | wants_to_map_ids = !lxc_list_empty(id_map); | |
1557 | ||
1558 | for (i = 0; i < LXC_NS_MAX; i++) { | |
1559 | if (!conf->ns_share[i]) | |
1560 | continue; | |
1561 | ||
1562 | handler->nsfd[i] = lxc_inherit_namespace(conf->ns_share[i], lxcpath, ns_info[i].proc_name); | |
1563 | if (handler->nsfd[i] < 0) | |
1564 | return -1; | |
1565 | ||
1566 | share_ns = true; | |
1567 | } | |
1568 | ||
1569 | ret = lxc_sync_init(handler); | |
1570 | if (ret < 0) | |
1571 | return -1; | |
1572 | ||
1573 | ret = socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0, | |
1574 | handler->data_sock); | |
1575 | if (ret < 0) | |
1576 | goto out_sync_fini; | |
1577 | data_sock0 = handler->data_sock[0]; | |
1578 | data_sock1 = handler->data_sock[1]; | |
1579 | ||
1580 | ret = resolve_clone_flags(handler); | |
1581 | if (ret < 0) | |
1582 | goto out_sync_fini; | |
1583 | ||
1584 | if (handler->ns_clone_flags & CLONE_NEWNET) { | |
1585 | ret = lxc_find_gateway_addresses(handler); | |
1586 | if (ret) { | |
1587 | ERROR("Failed to find gateway addresses"); | |
1588 | goto out_sync_fini; | |
1589 | } | |
1590 | } | |
1591 | ||
1592 | if (!cgroup_ops->payload_create(cgroup_ops, handler)) { | |
1593 | ERROR("Failed creating cgroups"); | |
1594 | goto out_delete_net; | |
1595 | } | |
1596 | ||
1597 | /* If the rootfs is not a blockdev, prevent the container from marking | |
1598 | * it readonly. | |
1599 | * If the container is unprivileged then skip rootfs pinning. | |
1600 | */ | |
1601 | if (!wants_to_map_ids) { | |
1602 | handler->pinfd = pin_rootfs(conf->rootfs.path); | |
1603 | if (handler->pinfd == -EBADF) | |
1604 | INFO("Failed to pin the rootfs for container \"%s\"", handler->name); | |
1605 | } | |
1606 | ||
1607 | /* Create a process in a new set of namespaces. */ | |
1608 | handler->ns_on_clone_flags = handler->ns_clone_flags; | |
1609 | if (handler->ns_clone_flags & CLONE_NEWUSER) { | |
1610 | /* If CLONE_NEWUSER and CLONE_NEWNET was requested, we need to | |
1611 | * clone a new user namespace first and only later unshare our | |
1612 | * network namespace to ensure that network devices ownership is | |
1613 | * set up correctly. | |
1614 | */ | |
1615 | handler->ns_on_clone_flags &= ~CLONE_NEWNET; | |
1616 | } | |
1617 | /* The cgroup namespace gets unshare()ed not clone()ed. */ | |
1618 | handler->ns_on_clone_flags &= ~CLONE_NEWCGROUP; | |
1619 | ||
1620 | if (share_ns) { | |
1621 | pid_t attacher_pid; | |
1622 | ||
1623 | attacher_pid = lxc_clone(do_share_ns, handler, | |
1624 | CLONE_VFORK | CLONE_VM | CLONE_FILES, NULL); | |
1625 | if (attacher_pid < 0) { | |
1626 | SYSERROR(LXC_CLONE_ERROR); | |
1627 | goto out_delete_net; | |
1628 | } | |
1629 | ||
1630 | ret = wait_for_pid(attacher_pid); | |
1631 | if (ret < 0) { | |
1632 | SYSERROR("Intermediate process failed"); | |
1633 | goto out_delete_net; | |
1634 | } | |
1635 | } else { | |
1636 | handler->pid = lxc_raw_clone_cb(do_start, handler, | |
1637 | CLONE_PIDFD | handler->ns_on_clone_flags, | |
1638 | &handler->pidfd); | |
1639 | } | |
1640 | if (handler->pid < 0) { | |
1641 | SYSERROR(LXC_CLONE_ERROR); | |
1642 | goto out_delete_net; | |
1643 | } | |
1644 | TRACE("Cloned child process %d", handler->pid); | |
1645 | ||
1646 | /* Verify that we can actually make use of pidfds. */ | |
1647 | if (!lxc_can_use_pidfd(handler->pidfd)) | |
1648 | close_prot_errno_disarm(handler->pidfd); | |
1649 | ||
1650 | ret = snprintf(pidstr, 20, "%d", handler->pid); | |
1651 | if (ret < 0 || ret >= 20) | |
1652 | goto out_delete_net; | |
1653 | ||
1654 | ret = setenv("LXC_PID", pidstr, 1); | |
1655 | if (ret < 0) | |
1656 | SYSERROR("Failed to set environment variable: LXC_PID=%s", pidstr); | |
1657 | ||
1658 | for (i = 0; i < LXC_NS_MAX; i++) | |
1659 | if (handler->ns_on_clone_flags & ns_info[i].clone_flag) | |
1660 | INFO("Cloned %s", ns_info[i].flag_name); | |
1661 | ||
1662 | if (!lxc_try_preserve_namespaces(handler, handler->ns_on_clone_flags, handler->pid)) { | |
1663 | ERROR("Failed to preserve cloned namespaces for lxc.hook.stop"); | |
1664 | goto out_delete_net; | |
1665 | } | |
1666 | ||
1667 | lxc_sync_fini_child(handler); | |
1668 | ||
1669 | if (lxc_abstract_unix_send_fds(handler->data_sock[0], &handler->monitor_status_fd, 1, NULL, 0) < 0) { | |
1670 | ERROR("Failed to send status file descriptor to child process"); | |
1671 | goto out_delete_net; | |
1672 | } | |
1673 | close_prot_errno_disarm(handler->monitor_status_fd); | |
1674 | ||
1675 | /* Map the container uids. The container became an invalid userid the | |
1676 | * moment it was cloned with CLONE_NEWUSER. This call doesn't change | |
1677 | * anything immediately, but allows the container to setuid(0) (0 being | |
1678 | * mapped to something else on the host.) later to become a valid uid | |
1679 | * again. | |
1680 | */ | |
1681 | if (wants_to_map_ids) { | |
1682 | if (!handler->conf->ns_share[LXC_NS_USER] && | |
1683 | (handler->conf->ns_keep & CLONE_NEWUSER) == 0) { | |
1684 | ret = lxc_map_ids(id_map, handler->pid); | |
1685 | if (ret < 0) { | |
1686 | ERROR("Failed to set up id mapping."); | |
1687 | goto out_delete_net; | |
1688 | } | |
1689 | } | |
1690 | } | |
1691 | ||
1692 | ret = lxc_sync_wake_child(handler, LXC_SYNC_STARTUP); | |
1693 | if (ret < 0) | |
1694 | goto out_delete_net; | |
1695 | ||
1696 | ret = lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE); | |
1697 | if (ret < 0) | |
1698 | goto out_delete_net; | |
1699 | ||
1700 | if (!cgroup_ops->setup_limits_legacy(cgroup_ops, handler->conf, false)) { | |
1701 | ERROR("Failed to setup cgroup limits for container \"%s\"", name); | |
1702 | goto out_delete_net; | |
1703 | } | |
1704 | ||
1705 | if (!cgroup_ops->payload_enter(cgroup_ops, handler)) { | |
1706 | goto out_delete_net; | |
1707 | } | |
1708 | ||
1709 | if (!cgroup_ops->payload_delegate_controllers(cgroup_ops)) { | |
1710 | ERROR("Failed to delegate controllers to payload cgroup"); | |
1711 | goto out_delete_net; | |
1712 | } | |
1713 | ||
1714 | if (!cgroup_ops->setup_limits(cgroup_ops, handler)) { | |
1715 | ERROR("Failed to setup cgroup limits for container \"%s\"", name); | |
1716 | goto out_delete_net; | |
1717 | } | |
1718 | ||
1719 | if (!cgroup_ops->chown(cgroup_ops, handler->conf)) | |
1720 | goto out_delete_net; | |
1721 | ||
1722 | /* If not done yet, we're now ready to preserve the network namespace */ | |
1723 | if (handler->nsfd[LXC_NS_NET] < 0) { | |
1724 | ret = lxc_try_preserve_ns(handler->pid, "net"); | |
1725 | if (ret < 0) { | |
1726 | if (ret != -EOPNOTSUPP) { | |
1727 | SYSERROR("Failed to preserve net namespace"); | |
1728 | goto out_delete_net; | |
1729 | } | |
1730 | } else { | |
1731 | handler->nsfd[LXC_NS_NET] = ret; | |
1732 | DEBUG("Preserved net namespace via fd %d", ret); | |
1733 | } | |
1734 | } | |
1735 | ret = lxc_netns_set_nsid(handler->nsfd[LXC_NS_NET]); | |
1736 | if (ret < 0) | |
1737 | SYSWARN("Failed to allocate new network namespace id"); | |
1738 | else | |
1739 | TRACE("Allocated new network namespace id"); | |
1740 | ||
1741 | /* Create the network configuration. */ | |
1742 | if (handler->ns_clone_flags & CLONE_NEWNET) { | |
1743 | ret = lxc_create_network(handler); | |
1744 | if (ret < 0) { | |
1745 | ERROR("Failed to create the network"); | |
1746 | goto out_delete_net; | |
1747 | } | |
1748 | ||
1749 | ret = lxc_network_send_to_child(handler); | |
1750 | if (ret < 0) { | |
1751 | ERROR("Failed to send veth names to child"); | |
1752 | goto out_delete_net; | |
1753 | } | |
1754 | } | |
1755 | ||
1756 | if (!lxc_list_empty(&conf->procs)) { | |
1757 | ret = setup_proc_filesystem(&conf->procs, handler->pid); | |
1758 | if (ret < 0) | |
1759 | goto out_delete_net; | |
1760 | } | |
1761 | ||
1762 | /* Tell the child to continue its initialization. We'll get | |
1763 | * LXC_SYNC_CGROUP when it is ready for us to setup cgroups. | |
1764 | */ | |
1765 | ret = lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE); | |
1766 | if (ret < 0) | |
1767 | goto out_delete_net; | |
1768 | ||
1769 | if (!lxc_list_empty(&conf->limits)) { | |
1770 | ret = setup_resource_limits(&conf->limits, handler->pid); | |
1771 | if (ret < 0) { | |
1772 | ERROR("Failed to setup resource limits"); | |
1773 | goto out_delete_net; | |
1774 | } | |
1775 | } | |
1776 | ||
1777 | ret = lxc_sync_barrier_child(handler, LXC_SYNC_CGROUP_UNSHARE); | |
1778 | if (ret < 0) | |
1779 | goto out_delete_net; | |
1780 | ||
1781 | if (!cgroup_ops->setup_limits_legacy(cgroup_ops, handler->conf, true)) { | |
1782 | ERROR("Failed to setup legacy device cgroup controller limits"); | |
1783 | goto out_delete_net; | |
1784 | } | |
1785 | TRACE("Set up legacy device cgroup controller limits"); | |
1786 | ||
1787 | if (!cgroup_ops->devices_activate(cgroup_ops, handler)) { | |
1788 | ERROR("Failed to setup cgroup2 device controller limits"); | |
1789 | goto out_delete_net; | |
1790 | } | |
1791 | TRACE("Set up cgroup2 device controller limits"); | |
1792 | ||
1793 | if (handler->ns_clone_flags & CLONE_NEWCGROUP) { | |
1794 | /* Now we're ready to preserve the cgroup namespace */ | |
1795 | ret = lxc_try_preserve_ns(handler->pid, "cgroup"); | |
1796 | if (ret < 0) { | |
1797 | if (ret != -EOPNOTSUPP) { | |
1798 | SYSERROR("Failed to preserve cgroup namespace"); | |
1799 | goto out_delete_net; | |
1800 | } | |
1801 | } else { | |
1802 | handler->nsfd[LXC_NS_CGROUP] = ret; | |
1803 | DEBUG("Preserved cgroup namespace via fd %d", ret); | |
1804 | } | |
1805 | } | |
1806 | ||
1807 | cgroup_ops->payload_finalize(cgroup_ops); | |
1808 | TRACE("Finished setting up cgroups"); | |
1809 | ||
1810 | /* Run any host-side start hooks */ | |
1811 | ret = run_lxc_hooks(name, "start-host", conf, NULL); | |
1812 | if (ret < 0) { | |
1813 | ERROR("Failed to run lxc.hook.start-host"); | |
1814 | goto out_delete_net; | |
1815 | } | |
1816 | ||
1817 | /* Tell the child to complete its initialization and wait for it to exec | |
1818 | * or return an error. (The child will never return | |
1819 | * LXC_SYNC_READY_START+1. It will either close the sync pipe, causing | |
1820 | * lxc_sync_barrier_child to return success, or return a different | |
1821 | * value, causing us to error out). | |
1822 | */ | |
1823 | ret = lxc_sync_barrier_child(handler, LXC_SYNC_READY_START); | |
1824 | if (ret < 0) | |
1825 | goto out_delete_net; | |
1826 | ||
1827 | if (handler->ns_clone_flags & CLONE_NEWNET) { | |
1828 | ret = lxc_network_recv_name_and_ifindex_from_child(handler); | |
1829 | if (ret < 0) { | |
1830 | ERROR("Failed to receive names and ifindices for network devices from child"); | |
1831 | goto out_delete_net; | |
1832 | } | |
1833 | } | |
1834 | ||
1835 | /* Now all networks are created, network devices are moved into place, | |
1836 | * and the correct names and ifindices in the respective namespaces have | |
1837 | * been recorded. The corresponding structs have now all been filled. So | |
1838 | * log them for debugging purposes. | |
1839 | */ | |
1840 | lxc_log_configured_netdevs(conf); | |
1841 | ||
1842 | /* Read tty fds allocated by child. */ | |
1843 | ret = lxc_recv_ttys_from_child(handler); | |
1844 | if (ret < 0) { | |
1845 | ERROR("Failed to receive tty info from child process"); | |
1846 | goto out_delete_net; | |
1847 | } | |
1848 | ||
1849 | ret = lxc_seccomp_recv_notifier_fd(&handler->conf->seccomp, data_sock1); | |
1850 | if (ret < 0) { | |
1851 | SYSERROR("Failed to receive seccomp notify fd from child"); | |
1852 | goto out_delete_net; | |
1853 | } | |
1854 | ||
1855 | ret = handler->ops->post_start(handler, handler->data); | |
1856 | if (ret < 0) | |
1857 | goto out_abort; | |
1858 | ||
1859 | ret = lxc_set_state(name, handler, RUNNING); | |
1860 | if (ret < 0) { | |
1861 | ERROR("Failed to set state to \"%s\"", lxc_state2str(RUNNING)); | |
1862 | goto out_abort; | |
1863 | } | |
1864 | ||
1865 | lxc_sync_fini(handler); | |
1866 | ||
1867 | return 0; | |
1868 | ||
1869 | out_delete_net: | |
1870 | if (handler->ns_clone_flags & CLONE_NEWNET) | |
1871 | lxc_delete_network(handler); | |
1872 | ||
1873 | out_abort: | |
1874 | lxc_abort(handler); | |
1875 | ||
1876 | out_sync_fini: | |
1877 | lxc_sync_fini(handler); | |
1878 | close_prot_errno_disarm(handler->pinfd); | |
1879 | ||
1880 | return -1; | |
1881 | } | |
1882 | ||
1883 | int __lxc_start(struct lxc_handler *handler, struct lxc_operations *ops, | |
1884 | void *data, const char *lxcpath, bool daemonize, int *error_num) | |
1885 | { | |
1886 | int ret, status; | |
1887 | const char *name = handler->name; | |
1888 | struct lxc_conf *conf = handler->conf; | |
1889 | struct cgroup_ops *cgroup_ops; | |
1890 | ||
1891 | ret = lxc_init(name, handler); | |
1892 | if (ret < 0) { | |
1893 | ERROR("Failed to initialize container \"%s\"", name); | |
1894 | goto out_abort; | |
1895 | } | |
1896 | handler->ops = ops; | |
1897 | handler->data = data; | |
1898 | handler->daemonize = daemonize; | |
1899 | cgroup_ops = handler->cgroup_ops; | |
1900 | ||
1901 | if (!attach_block_device(handler->conf)) { | |
1902 | ERROR("Failed to attach block device"); | |
1903 | ret = -1; | |
1904 | goto out_abort; | |
1905 | } | |
1906 | ||
1907 | if (!cgroup_ops->monitor_create(cgroup_ops, handler)) { | |
1908 | ERROR("Failed to create monitor cgroup"); | |
1909 | ret = -1; | |
1910 | goto out_abort; | |
1911 | } | |
1912 | ||
1913 | if (!cgroup_ops->monitor_enter(cgroup_ops, handler)) { | |
1914 | ERROR("Failed to enter monitor cgroup"); | |
1915 | ret = -1; | |
1916 | goto out_abort; | |
1917 | } | |
1918 | ||
1919 | if (!cgroup_ops->monitor_delegate_controllers(cgroup_ops)) { | |
1920 | ERROR("Failed to delegate controllers to monitor cgroup"); | |
1921 | ret = -1; | |
1922 | goto out_abort; | |
1923 | } | |
1924 | ||
1925 | if (geteuid() == 0 && !lxc_list_empty(&conf->id_map)) { | |
1926 | /* If the backing store is a device, mount it here and now. */ | |
1927 | if (rootfs_is_blockdev(conf)) { | |
1928 | ret = unshare(CLONE_NEWNS); | |
1929 | if (ret < 0) { | |
1930 | ERROR("Failed to unshare CLONE_NEWNS"); | |
1931 | goto out_abort; | |
1932 | } | |
1933 | INFO("Unshared CLONE_NEWNS"); | |
1934 | ||
1935 | remount_all_slave(); | |
1936 | ret = lxc_setup_rootfs_prepare_root(conf, name, lxcpath); | |
1937 | if (ret < 0) { | |
1938 | ERROR("Error setting up rootfs mount as root before spawn"); | |
1939 | goto out_abort; | |
1940 | } | |
1941 | INFO("Set up container rootfs as host root"); | |
1942 | } | |
1943 | } | |
1944 | ||
1945 | ret = lxc_spawn(handler); | |
1946 | if (ret < 0) { | |
1947 | ERROR("Failed to spawn container \"%s\"", name); | |
1948 | goto out_detach_blockdev; | |
1949 | } | |
1950 | ||
1951 | handler->conf->reboot = REBOOT_NONE; | |
1952 | ||
1953 | ret = lxc_poll(name, handler); | |
1954 | if (ret) { | |
1955 | ERROR("LXC mainloop exited with error: %d", ret); | |
1956 | goto out_delete_network; | |
1957 | } | |
1958 | ||
1959 | if (!handler->init_died && handler->pid > 0) { | |
1960 | ERROR("Child process is not killed"); | |
1961 | ret = -1; | |
1962 | goto out_delete_network; | |
1963 | } | |
1964 | ||
1965 | status = lxc_wait_for_pid_status(handler->pid); | |
1966 | if (status < 0) | |
1967 | SYSERROR("Failed to retrieve status for %d", handler->pid); | |
1968 | ||
1969 | /* If the child process exited but was not signaled, it didn't call | |
1970 | * reboot. This should mean it was an lxc-execute which simply exited. | |
1971 | * In any case, treat it as a 'halt'. | |
1972 | */ | |
1973 | if (WIFSIGNALED(status)) { | |
1974 | switch(WTERMSIG(status)) { | |
1975 | case SIGINT: /* halt */ | |
1976 | DEBUG("Container \"%s\" is halting", name); | |
1977 | break; | |
1978 | case SIGHUP: /* reboot */ | |
1979 | DEBUG("Container \"%s\" is rebooting", name); | |
1980 | handler->conf->reboot = REBOOT_REQ; | |
1981 | break; | |
1982 | case SIGSYS: /* seccomp */ | |
1983 | DEBUG("Container \"%s\" violated its seccomp policy", name); | |
1984 | break; | |
1985 | default: | |
1986 | DEBUG("Unknown exit status for container \"%s\" init %d", name, WTERMSIG(status)); | |
1987 | break; | |
1988 | } | |
1989 | } | |
1990 | ||
1991 | ret = lxc_restore_phys_nics_to_netns(handler); | |
1992 | if (ret < 0) | |
1993 | ERROR("Failed to move physical network devices back to parent network namespace"); | |
1994 | ||
1995 | close_prot_errno_disarm(handler->pinfd); | |
1996 | ||
1997 | lxc_monitor_send_exit_code(name, status, handler->lxcpath); | |
1998 | lxc_error_set_and_log(handler->pid, status); | |
1999 | if (error_num) | |
2000 | *error_num = handler->exit_status; | |
2001 | ||
2002 | /* These are not the droids you are looking for. */ | |
2003 | __private_goto1: | |
2004 | lxc_delete_network(handler); | |
2005 | ||
2006 | __private_goto2: | |
2007 | detach_block_device(handler->conf); | |
2008 | ||
2009 | __private_goto3: | |
2010 | lxc_fini(handler); | |
2011 | ||
2012 | return ret; | |
2013 | ||
2014 | /* These are the droids you are looking for. */ | |
2015 | out_abort: | |
2016 | lxc_abort(handler); | |
2017 | goto __private_goto3; | |
2018 | ||
2019 | out_detach_blockdev: | |
2020 | lxc_abort(handler); | |
2021 | goto __private_goto2; | |
2022 | ||
2023 | out_delete_network: | |
2024 | lxc_abort(handler); | |
2025 | goto __private_goto1; | |
2026 | } | |
2027 | ||
2028 | struct start_args { | |
2029 | char *const *argv; | |
2030 | }; | |
2031 | ||
2032 | static int start(struct lxc_handler *handler, void* data) | |
2033 | { | |
2034 | struct start_args *arg = data; | |
2035 | ||
2036 | NOTICE("Exec'ing \"%s\"", arg->argv[0]); | |
2037 | ||
2038 | execvp(arg->argv[0], arg->argv); | |
2039 | SYSERROR("Failed to exec \"%s\"", arg->argv[0]); | |
2040 | return 0; | |
2041 | } | |
2042 | ||
2043 | static int post_start(struct lxc_handler *handler, void* data) | |
2044 | { | |
2045 | struct start_args *arg = data; | |
2046 | ||
2047 | NOTICE("Started \"%s\" with pid \"%d\"", arg->argv[0], handler->pid); | |
2048 | return 0; | |
2049 | } | |
2050 | ||
2051 | static struct lxc_operations start_ops = { | |
2052 | .start = start, | |
2053 | .post_start = post_start | |
2054 | }; | |
2055 | ||
2056 | int lxc_start(char *const argv[], struct lxc_handler *handler, | |
2057 | const char *lxcpath, bool daemonize, int *error_num) | |
2058 | { | |
2059 | struct start_args start_arg = { | |
2060 | .argv = argv, | |
2061 | }; | |
2062 | ||
2063 | TRACE("Doing lxc_start"); | |
2064 | return __lxc_start(handler, &start_ops, &start_arg, lxcpath, daemonize, error_num); | |
2065 | } | |
2066 | ||
2067 | static void lxc_destroy_container_on_signal(struct lxc_handler *handler, | |
2068 | const char *name) | |
2069 | { | |
2070 | char destroy[PATH_MAX]; | |
2071 | struct lxc_container *c; | |
2072 | int ret = 0; | |
2073 | bool bret = true; | |
2074 | ||
2075 | if (handler->conf->rootfs.path && handler->conf->rootfs.mount) { | |
2076 | bret = do_destroy_container(handler); | |
2077 | if (!bret) { | |
2078 | ERROR("Error destroying rootfs for container \"%s\"", name); | |
2079 | return; | |
2080 | } | |
2081 | } | |
2082 | INFO("Destroyed rootfs for container \"%s\"", name); | |
2083 | ||
2084 | ret = snprintf(destroy, PATH_MAX, "%s/%s", handler->lxcpath, name); | |
2085 | if (ret < 0 || ret >= PATH_MAX) { | |
2086 | ERROR("Error destroying directory for container \"%s\"", name); | |
2087 | return; | |
2088 | } | |
2089 | ||
2090 | c = lxc_container_new(name, handler->lxcpath); | |
2091 | if (c) { | |
2092 | if (container_disk_lock(c)) { | |
2093 | INFO("Could not update lxc_snapshots file"); | |
2094 | lxc_container_put(c); | |
2095 | } else { | |
2096 | mod_all_rdeps(c, false); | |
2097 | container_disk_unlock(c); | |
2098 | lxc_container_put(c); | |
2099 | } | |
2100 | } | |
2101 | ||
2102 | if (!handler->am_root) | |
2103 | ret = userns_exec_full(handler->conf, lxc_rmdir_onedev_wrapper, | |
2104 | destroy, "lxc_rmdir_onedev_wrapper"); | |
2105 | else | |
2106 | ret = lxc_rmdir_onedev(destroy, NULL); | |
2107 | ||
2108 | if (ret < 0) { | |
2109 | ERROR("Error destroying directory for container \"%s\"", name); | |
2110 | return; | |
2111 | } | |
2112 | INFO("Destroyed directory for container \"%s\"", name); | |
2113 | } | |
2114 | ||
2115 | static int lxc_rmdir_onedev_wrapper(void *data) | |
2116 | { | |
2117 | char *arg = (char *) data; | |
2118 | return lxc_rmdir_onedev(arg, NULL); | |
2119 | } | |
2120 | ||
2121 | static bool do_destroy_container(struct lxc_handler *handler) | |
2122 | { | |
2123 | int ret; | |
2124 | ||
2125 | if (!handler->am_root) { | |
2126 | ret = userns_exec_full(handler->conf, storage_destroy_wrapper, | |
2127 | handler->conf, "storage_destroy_wrapper"); | |
2128 | if (ret < 0) | |
2129 | return false; | |
2130 | ||
2131 | return true; | |
2132 | } | |
2133 | ||
2134 | return storage_destroy(handler->conf); | |
2135 | } |