]>
Commit | Line | Data |
---|---|---|
0ad19a3f | 1 | /* |
2 | * lxc: linux Container library | |
3 | * | |
4 | * (C) Copyright IBM Corp. 2007, 2008 | |
5 | * | |
6 | * Authors: | |
9afe19d6 | 7 | * Daniel Lezcano <daniel.lezcano at free.fr> |
0ad19a3f | 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 | |
250b1eec | 21 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
0ad19a3f | 22 | */ |
23 | ||
f3787121 | 24 | #define _GNU_SOURCE |
f549edcc GK |
25 | #include "config.h" |
26 | ||
f3787121 | 27 | #include <alloca.h> |
0ad19a3f | 28 | #include <dirent.h> |
29 | #include <errno.h> | |
b0a33c1e | 30 | #include <fcntl.h> |
c476bdce | 31 | #include <grp.h> |
37515ebd | 32 | #include <poll.h> |
f3787121 CB |
33 | #include <signal.h> |
34 | #include <stdio.h> | |
35 | #include <stdlib.h> | |
36 | #include <string.h> | |
37 | #include <unistd.h> | |
0ad19a3f | 38 | #include <sys/file.h> |
f4d507d5 | 39 | #include <sys/mount.h> |
f3787121 | 40 | #include <sys/param.h> |
0ad19a3f | 41 | #include <sys/prctl.h> |
f3787121 CB |
42 | #include <sys/socket.h> |
43 | #include <sys/stat.h> | |
44 | #include <sys/syscall.h> | |
ddceb1f9 | 45 | #include <sys/types.h> |
b0a33c1e | 46 | #include <sys/un.h> |
f3787121 | 47 | #include <sys/wait.h> |
ff218c25 | 48 | |
e37dda71 | 49 | #if HAVE_LIBCAP |
495d2046 SG |
50 | #include <sys/capability.h> |
51 | #endif | |
52 | ||
955e2a02 | 53 | #ifndef HAVE_DECL_PR_CAPBSET_DROP |
656994bb MH |
54 | #define PR_CAPBSET_DROP 24 |
55 | #endif | |
56 | ||
955e2a02 CB |
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 | ||
f3787121 CB |
65 | #include "af_unix.h" |
66 | #include "caps.h" | |
563f2f2c | 67 | #include "cgroup.h" |
f3787121 | 68 | #include "commands.h" |
bbf5cf35 | 69 | #include "commands_utils.h" |
f3787121 CB |
70 | #include "conf.h" |
71 | #include "console.h" | |
e2bcd7db | 72 | #include "error.h" |
f3787121 | 73 | #include "log.h" |
f2d5a09d | 74 | #include "lxccontainer.h" |
f3787121 CB |
75 | #include "lxclock.h" |
76 | #include "lxcseccomp.h" | |
f3787121 | 77 | #include "mainloop.h" |
63376d7d | 78 | #include "monitor.h" |
f549edcc | 79 | #include "namespace.h" |
f3787121 | 80 | #include "start.h" |
28d832c4 | 81 | #include "storage.h" |
f2d5a09d | 82 | #include "storage_utils.h" |
f3787121 CB |
83 | #include "sync.h" |
84 | #include "utils.h" | |
fe4de9a6 | 85 | #include "lsm/lsm.h" |
36eb9bde CLG |
86 | |
87 | lxc_log_define(lxc_start, lxc); | |
88 | ||
f01f7975 | 89 | extern void mod_all_rdeps(struct lxc_container *c, bool inc); |
28272964 CB |
90 | static bool do_destroy_container(struct lxc_conf *conf); |
91 | static int lxc_rmdir_onedev_wrapper(void *data); | |
92 | static void lxc_destroy_container_on_signal(struct lxc_handler *handler, | |
93 | const char *name); | |
94 | ||
c8154066 SH |
95 | static void print_top_failing_dir(const char *path) |
96 | { | |
97 | size_t len = strlen(path); | |
f3787121 | 98 | char *copy = alloca(len + 1), *p, *e, saved; |
c8154066 SH |
99 | strcpy(copy, path); |
100 | ||
101 | p = copy; | |
102 | e = copy + len; | |
103 | while (p < e) { | |
f3787121 CB |
104 | while (p < e && *p == '/') |
105 | p++; | |
106 | while (p < e && *p != '/') | |
107 | p++; | |
c8154066 SH |
108 | saved = *p; |
109 | *p = '\0'; | |
110 | if (access(copy, X_OK)) { | |
408da065 CB |
111 | SYSERROR("Could not access %s. Please grant it x " |
112 | "access, or add an ACL for the container " | |
113 | "root.", copy); | |
c8154066 SH |
114 | return; |
115 | } | |
116 | *p = saved; | |
117 | } | |
118 | } | |
119 | ||
408da065 CB |
120 | static void close_ns(int ns_fd[LXC_NS_MAX]) |
121 | { | |
9f30a190 MM |
122 | int i; |
123 | ||
9f30a190 MM |
124 | for (i = 0; i < LXC_NS_MAX; i++) { |
125 | if (ns_fd[i] > -1) { | |
126 | close(ns_fd[i]); | |
127 | ns_fd[i] = -1; | |
128 | } | |
129 | } | |
9f30a190 MM |
130 | } |
131 | ||
4d8ac866 | 132 | /* preserve_ns: open /proc/@pid/ns/@ns for each namespace specified |
62d05d9b | 133 | * in clone_flags. |
4d8ac866 | 134 | * Return true on success, false on failure. |
62d05d9b | 135 | */ |
4d8ac866 | 136 | static bool preserve_ns(int ns_fd[LXC_NS_MAX], int clone_flags, pid_t pid) |
f3787121 | 137 | { |
62d05d9b | 138 | int i, ret; |
9f30a190 | 139 | |
9f30a190 MM |
140 | for (i = 0; i < LXC_NS_MAX; i++) |
141 | ns_fd[i] = -1; | |
142 | ||
4d8ac866 CB |
143 | ret = lxc_preserve_ns(pid, ""); |
144 | if (ret < 0) { | |
145 | SYSERROR("Kernel does not support attaching to namespaces."); | |
62d05d9b | 146 | return false; |
4d8ac866 CB |
147 | } else { |
148 | close(ret); | |
cd43d2d1 SH |
149 | } |
150 | ||
9f30a190 MM |
151 | for (i = 0; i < LXC_NS_MAX; i++) { |
152 | if ((clone_flags & ns_info[i].clone_flag) == 0) | |
153 | continue; | |
4d8ac866 | 154 | ns_fd[i] = lxc_preserve_ns(pid, ns_info[i].proc_name); |
9f30a190 MM |
155 | if (ns_fd[i] < 0) |
156 | goto error; | |
157 | } | |
158 | ||
62d05d9b | 159 | return true; |
9f30a190 MM |
160 | |
161 | error: | |
4d8ac866 CB |
162 | if (errno == ENOENT) |
163 | SYSERROR("Kernel does not support attaching to %s namespaces.", ns_info[i].proc_name); | |
164 | else | |
165 | SYSERROR("Failed to open file descriptor for %s namespace: %s.", ns_info[i].proc_name, strerror(errno)); | |
9f30a190 | 166 | close_ns(ns_fd); |
62d05d9b | 167 | return false; |
9f30a190 MM |
168 | } |
169 | ||
170 | static int attach_ns(const int ns_fd[LXC_NS_MAX]) { | |
171 | int i; | |
172 | ||
173 | for (i = 0; i < LXC_NS_MAX; i++) { | |
174 | if (ns_fd[i] < 0) | |
175 | continue; | |
176 | ||
177 | if (setns(ns_fd[i], 0) != 0) | |
178 | goto error; | |
179 | } | |
180 | return 0; | |
181 | ||
182 | error: | |
408da065 | 183 | SYSERROR("Failed to attach %s namespace.", ns_info[i].proc_name); |
9f30a190 MM |
184 | return -1; |
185 | } | |
186 | ||
80090207 CLG |
187 | static int match_fd(int fd) |
188 | { | |
189 | return (fd == 0 || fd == 1 || fd == 2); | |
190 | } | |
191 | ||
47a46cf1 CB |
192 | int lxc_check_inherited(struct lxc_conf *conf, bool closeall, |
193 | int *fds_to_ignore, size_t len_fds) | |
80090207 | 194 | { |
74f96976 | 195 | struct dirent *direntp; |
80090207 | 196 | int fd, fddir; |
47a46cf1 | 197 | size_t i; |
80090207 | 198 | DIR *dir; |
80090207 | 199 | |
d2cf4c37 SH |
200 | if (conf && conf->close_all_fds) |
201 | closeall = true; | |
202 | ||
b119f362 | 203 | restart: |
80090207 CLG |
204 | dir = opendir("/proc/self/fd"); |
205 | if (!dir) { | |
13277ec4 | 206 | WARN("Failed to open directory: %s.", strerror(errno)); |
80090207 CLG |
207 | return -1; |
208 | } | |
209 | ||
210 | fddir = dirfd(dir); | |
211 | ||
74f96976 | 212 | while ((direntp = readdir(dir))) { |
80090207 CLG |
213 | if (!direntp) |
214 | break; | |
215 | ||
216 | if (!strcmp(direntp->d_name, ".")) | |
217 | continue; | |
218 | ||
219 | if (!strcmp(direntp->d_name, "..")) | |
220 | continue; | |
221 | ||
d4cff0d2 CB |
222 | if (lxc_safe_int(direntp->d_name, &fd) < 0) { |
223 | INFO("Could not parse file descriptor for: %s", direntp->d_name); | |
224 | continue; | |
225 | } | |
80090207 | 226 | |
47a46cf1 CB |
227 | for (i = 0; i < len_fds; i++) |
228 | if (fds_to_ignore[i] == fd) | |
229 | break; | |
230 | ||
231 | if (fd == fddir || fd == lxc_log_fd || | |
232 | (i < len_fds && fd == fds_to_ignore[i])) | |
80090207 CLG |
233 | continue; |
234 | ||
858377e4 SH |
235 | if (current_config && fd == current_config->logfd) |
236 | continue; | |
237 | ||
80090207 CLG |
238 | if (match_fd(fd)) |
239 | continue; | |
80090207 | 240 | |
d2cf4c37 | 241 | if (closeall) { |
b119f362 SH |
242 | close(fd); |
243 | closedir(dir); | |
408da065 | 244 | INFO("Closed inherited fd: %d.", fd); |
b119f362 SH |
245 | goto restart; |
246 | } | |
408da065 | 247 | WARN("Inherited fd: %d.", fd); |
80090207 CLG |
248 | } |
249 | ||
408da065 | 250 | /* Only enable syslog at this point to avoid the above logging function |
64c57ea1 BD |
251 | * to open a new fd and make the check_inherited function enter an |
252 | * infinite loop. | |
253 | */ | |
254 | lxc_log_enable_syslog(); | |
255 | ||
92c7f629 GK |
256 | closedir(dir); /* cannot fail */ |
257 | return 0; | |
80090207 CLG |
258 | } |
259 | ||
83ee7875 | 260 | static int setup_signal_fd(sigset_t *oldmask) |
b0a33c1e | 261 | { |
262 | sigset_t mask; | |
263 | int fd; | |
264 | ||
408da065 | 265 | /* Block everything except serious error signals. */ |
f3304a29 FW |
266 | if (sigfillset(&mask) || |
267 | sigdelset(&mask, SIGILL) || | |
268 | sigdelset(&mask, SIGSEGV) || | |
269 | sigdelset(&mask, SIGBUS) || | |
b5159817 | 270 | sigdelset(&mask, SIGWINCH) || |
f3304a29 | 271 | sigprocmask(SIG_BLOCK, &mask, oldmask)) { |
408da065 | 272 | SYSERROR("Failed to set signal mask."); |
b0a33c1e | 273 | return -1; |
274 | } | |
275 | ||
276 | fd = signalfd(-1, &mask, 0); | |
277 | if (fd < 0) { | |
408da065 | 278 | SYSERROR("Failed to create signal file descriptor."); |
b0a33c1e | 279 | return -1; |
280 | } | |
281 | ||
282 | if (fcntl(fd, F_SETFD, FD_CLOEXEC)) { | |
408da065 | 283 | SYSERROR("Failed to set FD_CLOEXEC on the signal file descriptor: %d.", fd); |
b0a33c1e | 284 | close(fd); |
285 | return -1; | |
286 | } | |
287 | ||
408da065 | 288 | DEBUG("Set SIGCHLD handler with file descriptor: %d.", fd); |
1ac470c0 | 289 | |
b0a33c1e | 290 | return fd; |
291 | } | |
292 | ||
84c92abd | 293 | static int signal_handler(int fd, uint32_t events, void *data, |
f3787121 | 294 | struct lxc_epoll_descr *descr) |
b0a33c1e | 295 | { |
15cd25fd | 296 | struct signalfd_siginfo siginfo; |
80507ee8 | 297 | siginfo_t info; |
15cd25fd | 298 | int ret; |
82d89dce | 299 | pid_t *pid = data; |
80507ee8 | 300 | bool init_died = false; |
15cd25fd DL |
301 | |
302 | ret = read(fd, &siginfo, sizeof(siginfo)); | |
303 | if (ret < 0) { | |
408da065 | 304 | ERROR("Failed to read signal info from signal file descriptor: %d.", fd); |
15cd25fd DL |
305 | return -1; |
306 | } | |
307 | ||
308 | if (ret != sizeof(siginfo)) { | |
408da065 | 309 | ERROR("Unexpected size for siginfo struct."); |
15cd25fd DL |
310 | return -1; |
311 | } | |
312 | ||
408da065 | 313 | /* Check whether init is running. */ |
80507ee8 SH |
314 | info.si_pid = 0; |
315 | ret = waitid(P_PID, *pid, &info, WEXITED | WNOWAIT | WNOHANG); | |
408da065 | 316 | if (ret == 0 && info.si_pid == *pid) |
80507ee8 | 317 | init_died = true; |
80507ee8 | 318 | |
f3304a29 FW |
319 | if (siginfo.ssi_signo != SIGCHLD) { |
320 | kill(*pid, siginfo.ssi_signo); | |
408da065 | 321 | INFO("Forwarded signal %d to pid %d.", siginfo.ssi_signo, *pid); |
80507ee8 | 322 | return init_died ? 1 : 0; |
f3304a29 FW |
323 | } |
324 | ||
408da065 CB |
325 | if (siginfo.ssi_code == CLD_STOPPED) { |
326 | INFO("Container init process was stopped."); | |
327 | return init_died ? 1 : 0; | |
328 | } else if (siginfo.ssi_code == CLD_CONTINUED) { | |
329 | INFO("Container init process was continued."); | |
80507ee8 | 330 | return init_died ? 1 : 0; |
15cd25fd | 331 | } |
1ac470c0 | 332 | |
408da065 CB |
333 | /* More robustness, protect ourself from a SIGCHLD sent |
334 | * by a process different from the container init. | |
82d89dce DL |
335 | */ |
336 | if (siginfo.ssi_pid != *pid) { | |
af4c0f05 | 337 | NOTICE("Received SIGCHLD from pid %d instead of container init %d.", siginfo.ssi_pid, *pid); |
80507ee8 | 338 | return init_died ? 1 : 0; |
82d89dce DL |
339 | } |
340 | ||
408da065 | 341 | DEBUG("Container init process %d exited.", *pid); |
b0a33c1e | 342 | return 1; |
343 | } | |
344 | ||
5e5576a4 CB |
345 | static int lxc_serve_state_clients(const char *name, |
346 | struct lxc_handler *handler, | |
347 | lxc_state_t state) | |
66aeffc7 | 348 | { |
dbc9832d CB |
349 | ssize_t ret; |
350 | struct lxc_list *cur, *next; | |
351 | struct state_client *client; | |
352 | struct lxc_msg msg = {.type = lxc_msg_state, .value = state}; | |
353 | ||
354 | process_lock(); | |
5e5576a4 | 355 | |
dbc9832d | 356 | /* Only set state under process lock held so that we don't cause |
54446942 | 357 | * lxc_cmd_add_state_client() to miss a state. |
dbc9832d | 358 | */ |
66aeffc7 | 359 | handler->state = state; |
dbc9832d CB |
360 | TRACE("set container state to %s", lxc_state2str(state)); |
361 | ||
362 | if (lxc_list_empty(&handler->state_clients)) { | |
363 | TRACE("no state clients registered"); | |
364 | process_unlock(); | |
65ee1673 | 365 | lxc_monitor_send_state(name, state, handler->lxcpath); |
dbc9832d CB |
366 | return 0; |
367 | } | |
368 | ||
369 | strncpy(msg.name, name, sizeof(msg.name)); | |
370 | msg.name[sizeof(msg.name) - 1] = 0; | |
371 | ||
372 | lxc_list_for_each_safe(cur, &handler->state_clients, next) { | |
373 | client = cur->elem; | |
374 | ||
375 | if (!client->states[state]) { | |
376 | TRACE("state %s not registered for state client %d", | |
377 | lxc_state2str(state), client->clientfd); | |
378 | continue; | |
379 | } | |
380 | ||
381 | TRACE("sending state %s to state client %d", | |
382 | lxc_state2str(state), client->clientfd); | |
383 | ||
384 | again: | |
385 | ret = send(client->clientfd, &msg, sizeof(msg), 0); | |
ee8377bd CB |
386 | if (ret <= 0) { |
387 | if (errno == EINTR) { | |
388 | TRACE("Caught EINTR; retrying"); | |
dbc9832d | 389 | goto again; |
ee8377bd | 390 | } |
dbc9832d CB |
391 | |
392 | ERROR("failed to send message to client"); | |
393 | } | |
394 | ||
395 | /* kick client from list */ | |
396 | close(client->clientfd); | |
397 | lxc_list_del(cur); | |
398 | free(cur->elem); | |
399 | free(cur); | |
400 | } | |
401 | process_unlock(); | |
402 | ||
5e5576a4 CB |
403 | return 0; |
404 | } | |
405 | ||
406 | static int lxc_serve_state_socket_pair(const char *name, | |
407 | struct lxc_handler *handler, | |
408 | lxc_state_t state) | |
409 | { | |
410 | ssize_t ret; | |
411 | ||
412 | if (!handler->backgrounded || | |
413 | handler->state_socket_pair[1] < 0 || | |
414 | state == STARTING) | |
415 | return 0; | |
416 | ||
417 | /* Close read end of the socket pair. */ | |
418 | close(handler->state_socket_pair[0]); | |
419 | handler->state_socket_pair[0] = -1; | |
420 | ||
ee8377bd | 421 | again: |
5e5576a4 CB |
422 | ret = lxc_abstract_unix_send_credential(handler->state_socket_pair[1], |
423 | &(int){state}, sizeof(int)); | |
ee8377bd CB |
424 | if (ret != sizeof(int)) { |
425 | if (errno == EINTR) | |
426 | goto again; | |
5e5576a4 CB |
427 | SYSERROR("Failed to send state to %d", |
428 | handler->state_socket_pair[1]); | |
ee8377bd CB |
429 | return -1; |
430 | } | |
5e5576a4 CB |
431 | |
432 | TRACE("Sent container state \"%s\" to %d", lxc_state2str(state), | |
433 | handler->state_socket_pair[1]); | |
434 | ||
435 | /* Close write end of the socket pair. */ | |
436 | close(handler->state_socket_pair[1]); | |
437 | handler->state_socket_pair[1] = -1; | |
438 | ||
439 | return 0; | |
440 | } | |
441 | ||
442 | int lxc_set_state(const char *name, struct lxc_handler *handler, | |
443 | lxc_state_t state) | |
444 | { | |
445 | int ret; | |
446 | ||
447 | ret = lxc_serve_state_socket_pair(name, handler, state); | |
448 | if (ret < 0) { | |
449 | ERROR("Failed to synchronize via anonymous pair of unix sockets"); | |
450 | return -1; | |
451 | } | |
452 | ||
453 | ret = lxc_serve_state_clients(name, handler, state); | |
454 | if (ret < 0) | |
455 | return -1; | |
456 | ||
dbc9832d CB |
457 | /* This function will try to connect to the legacy lxc-monitord state |
458 | * server and only exists for backwards compatibility. | |
459 | */ | |
9123e471 | 460 | lxc_monitor_send_state(name, state, handler->lxcpath); |
dbc9832d | 461 | |
66aeffc7 DL |
462 | return 0; |
463 | } | |
464 | ||
735f2c6e | 465 | int lxc_poll(const char *name, struct lxc_handler *handler) |
b0a33c1e | 466 | { |
ca5f7926 DL |
467 | int sigfd = handler->sigfd; |
468 | int pid = handler->pid; | |
b0a33c1e | 469 | struct lxc_epoll_descr descr; |
470 | ||
a9e61274 | 471 | if (lxc_mainloop_open(&descr)) { |
408da065 | 472 | ERROR("Failed to create LXC mainloop."); |
50c8bf05 | 473 | goto out_sigfd; |
b0a33c1e | 474 | } |
475 | ||
83ee7875 | 476 | if (lxc_mainloop_add_handler(&descr, sigfd, signal_handler, &pid)) { |
408da065 | 477 | ERROR("Failed to add signal handler with file descriptor %d to LXC mainloop.", sigfd); |
b0a33c1e | 478 | goto out_mainloop_open; |
479 | } | |
480 | ||
da41561c | 481 | if (lxc_console_mainloop_add(&descr, handler->conf)) { |
408da065 | 482 | ERROR("Failed to add console handler to LXC mainloop."); |
63376d7d DL |
483 | goto out_mainloop_open; |
484 | } | |
485 | ||
ef6e34ee | 486 | if (lxc_cmd_mainloop_add(name, &descr, handler)) { |
408da065 | 487 | ERROR("Failed to add command handler to LXC mainloop."); |
96fa1ff0 | 488 | goto out_mainloop_open; |
563f2f2c DL |
489 | } |
490 | ||
3f903c04 | 491 | TRACE("lxc mainloop is ready"); |
b0a33c1e | 492 | |
e51d4895 | 493 | return lxc_mainloop(&descr, -1); |
b0a33c1e | 494 | |
495 | out_mainloop_open: | |
496 | lxc_mainloop_close(&descr); | |
408da065 | 497 | |
b0a33c1e | 498 | out_sigfd: |
499 | close(sigfd); | |
408da065 | 500 | |
c3e13372 | 501 | return -1; |
b0a33c1e | 502 | } |
503 | ||
f2e07cb6 CB |
504 | void lxc_free_handler(struct lxc_handler *handler) |
505 | { | |
506 | if (handler->conf && handler->conf->maincmd_fd) | |
507 | close(handler->conf->maincmd_fd); | |
508 | ||
5e5576a4 CB |
509 | if (handler->state_socket_pair[0] >= 0) |
510 | close(handler->state_socket_pair[0]); | |
511 | ||
512 | if (handler->state_socket_pair[1] >= 0) | |
513 | close(handler->state_socket_pair[1]); | |
514 | ||
f2e07cb6 CB |
515 | if (handler->name) |
516 | free(handler->name); | |
517 | ||
518 | handler->conf = NULL; | |
519 | free(handler); | |
520 | } | |
521 | ||
aa460476 | 522 | struct lxc_handler *lxc_init_handler(const char *name, struct lxc_conf *conf, |
5e5576a4 | 523 | const char *lxcpath, bool daemonize) |
59eb99ba | 524 | { |
5e5576a4 | 525 | int i, ret; |
3a0f472d DL |
526 | struct lxc_handler *handler; |
527 | ||
528 | handler = malloc(sizeof(*handler)); | |
aa460476 CB |
529 | if (!handler) { |
530 | ERROR("failed to allocate memory"); | |
3a0f472d | 531 | return NULL; |
aa460476 | 532 | } |
59eb99ba DL |
533 | |
534 | memset(handler, 0, sizeof(*handler)); | |
535 | ||
e8bd4e43 | 536 | handler->ttysock[0] = handler->ttysock[1] = -1; |
fae349da | 537 | handler->conf = conf; |
9123e471 | 538 | handler->lxcpath = lxcpath; |
5c068da9 | 539 | handler->pinfd = -1; |
5e5576a4 | 540 | handler->state_socket_pair[0] = handler->state_socket_pair[1] = -1; |
dbc9832d | 541 | lxc_list_init(&handler->state_clients); |
fae349da | 542 | |
b6b2b194 WB |
543 | for (i = 0; i < LXC_NS_MAX; i++) |
544 | handler->nsfd[i] = -1; | |
545 | ||
3bdf52d7 DL |
546 | handler->name = strdup(name); |
547 | if (!handler->name) { | |
aa460476 | 548 | ERROR("failed to allocate memory"); |
f2e07cb6 | 549 | goto on_error; |
3bdf52d7 DL |
550 | } |
551 | ||
5e5576a4 CB |
552 | if (daemonize && !handler->conf->reboot) { |
553 | /* Create socketpair() to synchronize on daemonized startup. | |
554 | * When the container reboots we don't need to synchronize again | |
555 | * currently so don't open another socketpair(). | |
556 | */ | |
557 | ret = socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0, | |
558 | handler->state_socket_pair); | |
559 | if (ret < 0) { | |
560 | ERROR("Failed to create anonymous pair of unix sockets"); | |
561 | goto on_error; | |
562 | } | |
563 | TRACE("Created anonymous pair {%d,%d} of unix sockets", | |
564 | handler->state_socket_pair[0], | |
565 | handler->state_socket_pair[1]); | |
566 | } | |
567 | ||
aa460476 CB |
568 | if (lxc_cmd_init(name, handler, lxcpath)) { |
569 | ERROR("failed to set up command socket"); | |
f2e07cb6 | 570 | goto on_error; |
aa460476 CB |
571 | } |
572 | ||
573 | TRACE("unix domain socket %d for command server is ready", | |
574 | handler->conf->maincmd_fd); | |
575 | ||
576 | return handler; | |
577 | ||
f2e07cb6 CB |
578 | on_error: |
579 | lxc_free_handler(handler); | |
aa460476 CB |
580 | |
581 | return NULL; | |
582 | } | |
583 | ||
584 | int lxc_init(const char *name, struct lxc_handler *handler) | |
585 | { | |
586 | struct lxc_conf *conf = handler->conf; | |
587 | ||
588 | lsm_init(); | |
589 | TRACE("initialized LSM"); | |
d2e30e99 | 590 | |
8f2c3a70 | 591 | if (lxc_read_seccomp_config(conf) != 0) { |
408da065 | 592 | ERROR("Failed loading seccomp policy."); |
d2e30e99 | 593 | goto out_close_maincmd_fd; |
8f2c3a70 | 594 | } |
dbc9832d | 595 | TRACE("read seccomp policy"); |
8f2c3a70 | 596 | |
408da065 | 597 | /* Begin by setting the state to STARTING. */ |
25c2aca5 | 598 | if (lxc_set_state(name, handler, STARTING)) { |
408da065 | 599 | ERROR("Failed to set state for container \"%s\" to \"%s\".", name, lxc_state2str(STARTING)); |
051151de | 600 | goto out_close_maincmd_fd; |
0ad19a3f | 601 | } |
dbc9832d | 602 | TRACE("set container state to \"STARTING\""); |
0ad19a3f | 603 | |
408da065 CB |
604 | /* Start of environment variable setup for hooks. */ |
605 | if (name && setenv("LXC_NAME", name, 1)) | |
606 | SYSERROR("Failed to set environment variable: LXC_NAME=%s.", name); | |
607 | ||
608 | if (conf->rcfile && setenv("LXC_CONFIG_FILE", conf->rcfile, 1)) | |
609 | SYSERROR("Failed to set environment variable: LXC_CONFIG_FILE=%s.", conf->rcfile); | |
610 | ||
611 | if (conf->rootfs.mount && setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1)) | |
612 | SYSERROR("Failed to set environment variable: LXC_ROOTFS_MOUNT=%s.", conf->rootfs.mount); | |
613 | ||
614 | if (conf->rootfs.path && setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1)) | |
615 | SYSERROR("Failed to set environment variable: LXC_ROOTFS_PATH=%s.", conf->rootfs.path); | |
616 | ||
617 | if (conf->console.path && setenv("LXC_CONSOLE", conf->console.path, 1)) | |
618 | SYSERROR("Failed to set environment variable: LXC_CONSOLE=%s.", conf->console.path); | |
619 | ||
620 | if (conf->console.log_path && setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1)) | |
621 | SYSERROR("Failed to set environment variable: LXC_CONSOLE_LOGPATH=%s.", conf->console.log_path); | |
622 | ||
623 | if (setenv("LXC_CGNS_AWARE", "1", 1)) | |
624 | SYSERROR("Failed to set environment variable LXC_CGNS_AWARE=1."); | |
625 | /* End of environment variable setup for hooks. */ | |
f7bee6c6 | 626 | |
dbc9832d CB |
627 | TRACE("set environment variables"); |
628 | ||
283678ed | 629 | if (run_lxc_hooks(name, "pre-start", conf, handler->lxcpath, NULL)) { |
408da065 | 630 | ERROR("Failed to run lxc.hook.pre-start for container \"%s\".", name); |
773fb9ca SH |
631 | goto out_aborting; |
632 | } | |
dbc9832d | 633 | TRACE("ran pre-start hooks"); |
26ddeedd | 634 | |
408da065 CB |
635 | /* The signal fd has to be created before forking otherwise if the child |
636 | * process exits before we setup the signal fd, the event will be lost | |
637 | * and the command will be stuck. | |
638 | */ | |
83ee7875 | 639 | handler->sigfd = setup_signal_fd(&handler->oldmask); |
59eb99ba | 640 | if (handler->sigfd < 0) { |
408da065 | 641 | ERROR("Failed to setup SIGCHLD fd handler."); |
b5159817 DE |
642 | goto out_delete_tty; |
643 | } | |
dbc9832d | 644 | TRACE("set up signal fd"); |
b5159817 | 645 | |
408da065 | 646 | /* Do this after setting up signals since it might unblock SIGWINCH. */ |
b5159817 | 647 | if (lxc_console_create(conf)) { |
408da065 | 648 | ERROR("Failed to create console for container \"%s\".", name); |
b5159817 | 649 | goto out_restore_sigmask; |
b0a33c1e | 650 | } |
dbc9832d | 651 | TRACE("created console"); |
b0a33c1e | 652 | |
54117de5 | 653 | if (lxc_ttys_shift_ids(conf) < 0) { |
408da065 | 654 | ERROR("Failed to shift tty into container."); |
c4d10a05 SH |
655 | goto out_restore_sigmask; |
656 | } | |
dbc9832d | 657 | TRACE("shifted tty ids"); |
c4d10a05 | 658 | |
dbc9832d | 659 | INFO("container \"%s\" is initialized", name); |
aa460476 | 660 | return 0; |
59eb99ba | 661 | |
b5159817 DE |
662 | out_restore_sigmask: |
663 | sigprocmask(SIG_SETMASK, &handler->oldmask, NULL); | |
59eb99ba | 664 | out_delete_tty: |
fae349da | 665 | lxc_delete_tty(&conf->tty_info); |
59eb99ba | 666 | out_aborting: |
25c2aca5 | 667 | lxc_set_state(name, handler, ABORTING); |
d2e30e99 DE |
668 | out_close_maincmd_fd: |
669 | close(conf->maincmd_fd); | |
670 | conf->maincmd_fd = -1; | |
aa460476 | 671 | return -1; |
59eb99ba DL |
672 | } |
673 | ||
3b72c4a0 | 674 | void lxc_fini(const char *name, struct lxc_handler *handler) |
59eb99ba | 675 | { |
b3286b62 | 676 | int i, rc; |
dbc9832d | 677 | struct lxc_list *cur, *next; |
b3286b62 | 678 | pid_t self = getpid(); |
dbc9832d | 679 | char *namespaces[LXC_NS_MAX + 1]; |
b3286b62 | 680 | size_t namespace_count = 0; |
b6b2b194 | 681 | |
408da065 CB |
682 | /* The STOPPING state is there for future cleanup code which can take |
683 | * awhile. | |
59eb99ba | 684 | */ |
25c2aca5 | 685 | lxc_set_state(name, handler, STOPPING); |
b6b2b194 | 686 | |
b3286b62 WB |
687 | for (i = 0; i < LXC_NS_MAX; i++) { |
688 | if (handler->nsfd[i] != -1) { | |
689 | rc = asprintf(&namespaces[namespace_count], "%s:/proc/%d/fd/%d", | |
690 | ns_info[i].proc_name, self, handler->nsfd[i]); | |
691 | if (rc == -1) { | |
408da065 | 692 | SYSERROR("Failed to allocate memory."); |
b3286b62 WB |
693 | break; |
694 | } | |
695 | ++namespace_count; | |
696 | } | |
697 | } | |
698 | namespaces[namespace_count] = NULL; | |
c154af98 | 699 | |
408da065 CB |
700 | if (handler->conf->reboot && setenv("LXC_TARGET", "reboot", 1)) |
701 | SYSERROR("Failed to set environment variable: LXC_TARGET=reboot."); | |
702 | ||
703 | if (!handler->conf->reboot && setenv("LXC_TARGET", "stop", 1)) | |
704 | SYSERROR("Failed to set environment variable: LXC_TARGET=stop."); | |
c154af98 | 705 | |
b3286b62 | 706 | if (run_lxc_hooks(name, "stop", handler->conf, handler->lxcpath, namespaces)) |
408da065 | 707 | ERROR("Failed to run lxc.hook.stop for container \"%s\".", name); |
c154af98 | 708 | |
b3286b62 WB |
709 | while (namespace_count--) |
710 | free(namespaces[namespace_count]); | |
b6b2b194 WB |
711 | for (i = 0; i < LXC_NS_MAX; i++) { |
712 | if (handler->nsfd[i] != -1) { | |
713 | close(handler->nsfd[i]); | |
714 | handler->nsfd[i] = -1; | |
715 | } | |
716 | } | |
738d0deb CB |
717 | |
718 | if (handler->netnsfd >= 0) { | |
719 | close(handler->netnsfd); | |
720 | handler->netnsfd = -1; | |
721 | } | |
722 | ||
4288b79f CB |
723 | cgroup_destroy(handler); |
724 | ||
25c2aca5 | 725 | lxc_set_state(name, handler, STOPPED); |
59eb99ba | 726 | |
4288b79f CB |
727 | /* close command socket */ |
728 | close(handler->conf->maincmd_fd); | |
729 | handler->conf->maincmd_fd = -1; | |
730 | ||
f3787121 | 731 | if (run_lxc_hooks(name, "post-stop", handler->conf, handler->lxcpath, NULL)) { |
408da065 | 732 | ERROR("Failed to run lxc.hook.post-stop for container \"%s\".", name); |
f3787121 CB |
733 | if (handler->conf->reboot) { |
734 | WARN("Container will be stopped instead of rebooted."); | |
735 | handler->conf->reboot = 0; | |
408da065 CB |
736 | if (setenv("LXC_TARGET", "stop", 1)) |
737 | WARN("Failed to set environment variable: LXC_TARGET=stop."); | |
f3787121 CB |
738 | } |
739 | } | |
26ddeedd | 740 | |
408da065 | 741 | /* Reset mask set by setup_signal_fd. */ |
8f64a3f6 | 742 | if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) |
408da065 | 743 | WARN("Failed to restore signal mask."); |
8f64a3f6 | 744 | |
b5159817 | 745 | lxc_console_delete(&handler->conf->console); |
b2431939 | 746 | lxc_delete_tty(&handler->conf->tty_info); |
dbc9832d | 747 | |
dbc9832d CB |
748 | /* The command socket is now closed, no more state clients can register |
749 | * themselves from now on. So free the list of state clients. | |
750 | */ | |
751 | lxc_list_for_each_safe(cur, &handler->state_clients, next) { | |
752 | struct state_client *client = cur->elem; | |
753 | /* close state client socket */ | |
754 | close(client->clientfd); | |
755 | lxc_list_del(cur); | |
756 | free(cur->elem); | |
757 | free(cur); | |
758 | } | |
759 | ||
e8bd4e43 SH |
760 | if (handler->ttysock[0] != -1) { |
761 | close(handler->ttysock[0]); | |
762 | close(handler->ttysock[1]); | |
763 | } | |
2c5f2ede CB |
764 | |
765 | if (handler->conf->ephemeral == 1 && handler->conf->reboot != 1) | |
28272964 | 766 | lxc_destroy_container_on_signal(handler, name); |
2c5f2ede | 767 | |
4288b79f | 768 | free(handler->name); |
b2431939 | 769 | free(handler); |
59eb99ba DL |
770 | } |
771 | ||
735f2c6e | 772 | void lxc_abort(const char *name, struct lxc_handler *handler) |
59eb99ba | 773 | { |
73e608b2 SH |
774 | int ret, status; |
775 | ||
25c2aca5 | 776 | lxc_set_state(name, handler, ABORTING); |
7d9fb3e9 DL |
777 | if (handler->pid > 0) |
778 | kill(handler->pid, SIGKILL); | |
408da065 CB |
779 | while ((ret = waitpid(-1, &status, 0)) > 0) { |
780 | ; | |
781 | } | |
59eb99ba DL |
782 | } |
783 | ||
408da065 CB |
784 | /* netpipe is used in the unprivileged case to transfer the ifindexes from |
785 | * parent to child | |
658979c5 SH |
786 | */ |
787 | static int netpipe = -1; | |
788 | ||
789 | static inline int count_veths(struct lxc_list *network) | |
790 | { | |
791 | struct lxc_list *iterator; | |
792 | struct lxc_netdev *netdev; | |
793 | int count = 0; | |
794 | ||
795 | lxc_list_for_each(iterator, network) { | |
796 | netdev = iterator->elem; | |
797 | if (netdev->type != LXC_NET_VETH) | |
798 | continue; | |
799 | count++; | |
800 | } | |
801 | return count; | |
802 | } | |
803 | ||
804 | static int read_unpriv_netifindex(struct lxc_list *network) | |
805 | { | |
806 | struct lxc_list *iterator; | |
807 | struct lxc_netdev *netdev; | |
808 | ||
809 | if (netpipe == -1) | |
810 | return 0; | |
811 | lxc_list_for_each(iterator, network) { | |
812 | netdev = iterator->elem; | |
813 | if (netdev->type != LXC_NET_VETH) | |
814 | continue; | |
815 | if (!(netdev->name = malloc(IFNAMSIZ))) { | |
408da065 | 816 | ERROR("Out of memory."); |
658979c5 SH |
817 | close(netpipe); |
818 | return -1; | |
819 | } | |
820 | if (read(netpipe, netdev->name, IFNAMSIZ) != IFNAMSIZ) { | |
821 | close(netpipe); | |
822 | return -1; | |
823 | } | |
824 | } | |
825 | close(netpipe); | |
826 | return 0; | |
827 | } | |
828 | ||
ffe1e01a | 829 | static int do_start(void *data) |
50e98013 | 830 | { |
7c661726 | 831 | struct lxc_list *iterator; |
23c53af9 | 832 | struct lxc_handler *handler = data; |
7a55c157 TA |
833 | int devnull_fd = -1, ret; |
834 | char path[PATH_MAX]; | |
50e98013 DL |
835 | |
836 | if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) { | |
408da065 | 837 | SYSERROR("Failed to set signal mask."); |
9d7f9e52 | 838 | return -1; |
50e98013 DL |
839 | } |
840 | ||
408da065 CB |
841 | /* This prctl must be before the synchro, so if the parent dies before |
842 | * we set the parent death signal, we will detect its death with the | |
843 | * synchro right after, otherwise we have a window where the parent can | |
844 | * exit before we set the pdeath signal leading to a unsupervized | |
845 | * container. | |
743ecd2e DL |
846 | */ |
847 | if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) { | |
408da065 | 848 | SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL."); |
743ecd2e DL |
849 | return -1; |
850 | } | |
851 | ||
3c22086f | 852 | lxc_sync_fini_parent(handler); |
50e98013 | 853 | |
408da065 | 854 | /* Don't leak the pinfd to the container. */ |
025ed0f3 | 855 | if (handler->pinfd >= 0) { |
0d03360a | 856 | close(handler->pinfd); |
025ed0f3 | 857 | } |
2b0e17e4 | 858 | |
5b1e83cb SH |
859 | if (lxc_sync_wait_parent(handler, LXC_SYNC_STARTUP)) |
860 | return -1; | |
861 | ||
408da065 CB |
862 | /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See |
863 | * https://github.com/lxc/lxd/issues/1978. | |
864 | */ | |
5b1e83cb | 865 | if ((handler->clone_flags & (CLONE_NEWNET | CLONE_NEWUSER)) == |
408da065 | 866 | (CLONE_NEWNET | CLONE_NEWUSER)) { |
5b1e83cb SH |
867 | ret = unshare(CLONE_NEWNET); |
868 | if (ret < 0) { | |
408da065 | 869 | SYSERROR("Failed to unshare CLONE_NEWNET."); |
5b1e83cb SH |
870 | goto out_warn_father; |
871 | } | |
408da065 | 872 | INFO("Unshared CLONE_NEWNET."); |
5b1e83cb SH |
873 | } |
874 | ||
408da065 CB |
875 | /* Tell the parent task it can begin to configure the container and wait |
876 | * for it to finish. | |
3c22086f CLG |
877 | */ |
878 | if (lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE)) | |
9d7f9e52 | 879 | return -1; |
50e98013 | 880 | |
658979c5 SH |
881 | if (read_unpriv_netifindex(&handler->conf->network) < 0) |
882 | goto out_warn_father; | |
883 | ||
408da065 | 884 | /* If we are in a new user namespace, become root there to have |
d08f8d2f | 885 | * privilege over our namespace. |
f6d3e3e4 SH |
886 | */ |
887 | if (!lxc_list_empty(&handler->conf->id_map)) { | |
d08f8d2f | 888 | if (lxc_switch_uid_gid(0, 0) < 0) |
f6d3e3e4 | 889 | goto out_warn_father; |
d08f8d2f CB |
890 | |
891 | /* Drop groups only after we switched to a valid gid in the new | |
892 | * user namespace. | |
893 | */ | |
894 | if (lxc_setgroups(0, NULL) < 0) | |
c476bdce | 895 | goto out_warn_father; |
f6d3e3e4 SH |
896 | } |
897 | ||
99b71824 | 898 | if (access(handler->lxcpath, X_OK)) { |
c8154066 SH |
899 | print_top_failing_dir(handler->lxcpath); |
900 | goto out_warn_father; | |
901 | } | |
902 | ||
a91cde21 | 903 | ret = snprintf(path, sizeof(path), "%s/dev/null", handler->conf->rootfs.mount); |
408da065 | 904 | if (ret < 0 || ret >= sizeof(path)) |
7a55c157 | 905 | goto out_warn_father; |
7a55c157 TA |
906 | |
907 | /* In order to checkpoint restore, we need to have everything in the | |
908 | * same mount namespace. However, some containers may not have a | |
909 | * reasonable /dev (in particular, they may not have /dev/null), so we | |
910 | * can't set init's std fds to /dev/null by opening it from inside the | |
911 | * container. | |
912 | * | |
913 | * If that's the case, fall back to using the host's /dev/null. This | |
914 | * means that migration won't work, but at least we won't spew output | |
915 | * where it isn't wanted. | |
916 | */ | |
917 | if (handler->backgrounded && !handler->conf->autodev && access(path, F_OK) < 0) { | |
c44de748 AM |
918 | devnull_fd = open_devnull(); |
919 | ||
920 | if (devnull_fd < 0) | |
921 | goto out_warn_father; | |
408da065 CB |
922 | WARN("Using /dev/null from the host for container init's " |
923 | "standard file descriptors. Migration will not work."); | |
c44de748 AM |
924 | } |
925 | ||
408da065 | 926 | /* Ask father to setup cgroups and wait for him to finish. */ |
544a48a0 | 927 | if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP)) |
c44de748 | 928 | goto out_error; |
544a48a0 | 929 | |
deefdf8a CB |
930 | /* Unshare cgroup namespace after we have setup our cgroups. If we do it |
931 | * earlier we end up with a wrong view of /proc/self/cgroup. For | |
932 | * example, assume we unshare(CLONE_NEWCGROUP) first, and then create | |
933 | * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then | |
934 | * /proc/self/cgroup would show us: | |
935 | * | |
936 | * 8:cpuset:/lxc/c | |
937 | * | |
938 | * whereas it should actually show | |
939 | * | |
940 | * 8:cpuset:/ | |
941 | */ | |
942 | if (cgns_supported()) { | |
943 | if (unshare(CLONE_NEWCGROUP) < 0) { | |
944 | INFO("Failed to unshare CLONE_NEWCGROUP."); | |
945 | goto out_warn_father; | |
946 | } | |
947 | INFO("Unshared CLONE_NEWCGROUP."); | |
948 | } | |
949 | ||
f4152036 CB |
950 | /* Setup the container, ip, names, utsname, ... */ |
951 | if (lxc_setup(handler)) { | |
952 | ERROR("Failed to setup container \"%s\".", handler->name); | |
953 | goto out_warn_father; | |
954 | } | |
955 | ||
408da065 | 956 | /* Set the label to change to when we exec(2) the container's init. */ |
7aff4f43 | 957 | if (lsm_process_label_set(NULL, handler->conf, 1, 1) < 0) |
e075f5d9 | 958 | goto out_warn_father; |
5112cd70 | 959 | |
029cdff5 | 960 | /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it |
408da065 CB |
961 | * before we aren't allowed anymore. |
962 | */ | |
029cdff5 CB |
963 | if (handler->conf->no_new_privs) { |
964 | if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) { | |
965 | SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges."); | |
966 | goto out_warn_father; | |
967 | } | |
968 | DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges."); | |
969 | } | |
970 | ||
0d9acb99 DE |
971 | /* Some init's such as busybox will set sane tty settings on stdin, |
972 | * stdout, stderr which it thinks is the console. We already set them | |
973 | * the way we wanted on the real terminal, and we want init to do its | |
408da065 CB |
974 | * setup on its console ie. the pty allocated in lxc_console_create() so |
975 | * make sure that that pty is stdin,stdout,stderr. | |
0d9acb99 | 976 | */ |
c5b93afb LF |
977 | if (handler->conf->console.slave >= 0) |
978 | if (set_stdfds(handler->conf->console.slave) < 0) { | |
979 | ERROR("Failed to redirect std{in,out,err} to pty file " | |
980 | "descriptor %d", | |
981 | handler->conf->console.slave); | |
982 | goto out_warn_father; | |
983 | } | |
0d9acb99 | 984 | |
408da065 | 985 | /* If we mounted a temporary proc, then unmount it now. */ |
5112cd70 | 986 | tmp_proc_unmount(handler->conf); |
e075f5d9 | 987 | |
8f2c3a70 SH |
988 | if (lxc_seccomp_load(handler->conf) != 0) |
989 | goto out_warn_father; | |
990 | ||
283678ed | 991 | if (run_lxc_hooks(handler->name, "start", handler->conf, handler->lxcpath, NULL)) { |
408da065 | 992 | ERROR("Failed to run lxc.hook.start for container \"%s\".", handler->name); |
773fb9ca SH |
993 | goto out_warn_father; |
994 | } | |
fc25b815 | 995 | |
d08f8d2f CB |
996 | /* The container has been setup. We can now switch to an unprivileged |
997 | * uid/gid. | |
998 | */ | |
999 | if (handler->conf->is_execute) { | |
87bf0db0 CB |
1000 | bool have_cap_setgid; |
1001 | uid_t new_uid = handler->conf->init_uid; | |
1002 | gid_t new_gid = handler->conf->init_gid; | |
d08f8d2f CB |
1003 | |
1004 | /* If we are in a new user namespace we already dropped all | |
1005 | * groups when we switched to root in the new user namespace | |
1006 | * further above. Only drop groups if we can, so ensure that we | |
1007 | * have necessary privilege. | |
1008 | */ | |
e37dda71 | 1009 | #if HAVE_LIBCAP |
207c4c71 | 1010 | have_cap_setgid = lxc_proc_cap_is_set(CAP_SETGID, CAP_EFFECTIVE); |
df11e022 BN |
1011 | #else |
1012 | have_cap_setgid = false; | |
1013 | #endif | |
87bf0db0 | 1014 | if (lxc_list_empty(&handler->conf->id_map) && have_cap_setgid) { |
d08f8d2f CB |
1015 | if (lxc_setgroups(0, NULL) < 0) |
1016 | goto out_warn_father; | |
1017 | } | |
1018 | ||
1019 | if (lxc_switch_uid_gid(new_uid, new_gid) < 0) | |
1020 | goto out_warn_father; | |
1021 | } | |
1022 | ||
408da065 CB |
1023 | /* The clearenv() and putenv() calls have been moved here to allow us to |
1024 | * use environment variables passed to the various hooks, such as the | |
1025 | * start hook above. Not all of the variables like CONFIG_PATH or ROOTFS | |
1026 | * are valid in this context but others are. | |
1027 | */ | |
f7bee6c6 | 1028 | if (clearenv()) { |
408da065 CB |
1029 | SYSERROR("Failed to clear environment."); |
1030 | /* Don't error out though. */ | |
f7bee6c6 MW |
1031 | } |
1032 | ||
7c661726 MP |
1033 | lxc_list_for_each(iterator, &handler->conf->environment) { |
1034 | if (putenv((char *)iterator->elem)) { | |
408da065 | 1035 | SYSERROR("Failed to set environment variable: %s.", (char *)iterator->elem); |
7c661726 MP |
1036 | goto out_warn_father; |
1037 | } | |
1038 | } | |
1039 | ||
f7bee6c6 | 1040 | if (putenv("container=lxc")) { |
408da065 | 1041 | SYSERROR("Failed to set environment variable: container=lxc."); |
c4ea60df | 1042 | goto out_warn_father; |
f7bee6c6 MW |
1043 | } |
1044 | ||
393903d1 SH |
1045 | if (handler->conf->pty_names) { |
1046 | if (putenv(handler->conf->pty_names)) { | |
408da065 | 1047 | SYSERROR("Failed to set environment variable for container ptys."); |
393903d1 SH |
1048 | goto out_warn_father; |
1049 | } | |
1050 | } | |
1051 | ||
773fb9ca | 1052 | close(handler->sigfd); |
26ddeedd | 1053 | |
7a55c157 TA |
1054 | if (devnull_fd < 0) { |
1055 | devnull_fd = open_devnull(); | |
1056 | ||
1057 | if (devnull_fd < 0) | |
1058 | goto out_warn_father; | |
1059 | } | |
1060 | ||
c5b93afb LF |
1061 | if (handler->conf->console.slave < 0 && handler->backgrounded) |
1062 | if (set_stdfds(devnull_fd) < 0) { | |
1063 | ERROR("Failed to redirect std{in,out,err} to " | |
1064 | "\"/dev/null\""); | |
1065 | goto out_warn_father; | |
1066 | } | |
507cee36 | 1067 | |
c44de748 AM |
1068 | if (devnull_fd >= 0) { |
1069 | close(devnull_fd); | |
1070 | devnull_fd = -1; | |
1071 | } | |
1072 | ||
8c9a7665 TA |
1073 | setsid(); |
1074 | ||
f4152036 CB |
1075 | if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP_LIMITS)) |
1076 | goto out_warn_father; | |
1077 | ||
408da065 CB |
1078 | /* After this call, we are in error because this ops should not return |
1079 | * as it execs. | |
1080 | */ | |
c4ea60df | 1081 | handler->ops->start(handler, handler->data); |
50e98013 DL |
1082 | |
1083 | out_warn_father: | |
408da065 CB |
1084 | /* We want the parent to know something went wrong, so we return a |
1085 | * special error code. | |
1086 | */ | |
d1ccb562 | 1087 | lxc_sync_wake_parent(handler, LXC_SYNC_ERROR); |
c44de748 AM |
1088 | |
1089 | out_error: | |
1090 | if (devnull_fd >= 0) | |
1091 | close(devnull_fd); | |
1092 | ||
50e98013 DL |
1093 | return -1; |
1094 | } | |
1095 | ||
74a3920a | 1096 | static int save_phys_nics(struct lxc_conf *conf) |
7b35f3d6 SH |
1097 | { |
1098 | struct lxc_list *iterator; | |
40f2f8a2 | 1099 | int am_root = (getuid() == 0); |
7b35f3d6 | 1100 | |
40f2f8a2 LQ |
1101 | if (!am_root) |
1102 | return 0; | |
2c5f2ede | 1103 | |
7b35f3d6 SH |
1104 | lxc_list_for_each(iterator, &conf->network) { |
1105 | struct lxc_netdev *netdev = iterator->elem; | |
1106 | ||
1107 | if (netdev->type != LXC_NET_PHYS) | |
1108 | continue; | |
1109 | conf->saved_nics = realloc(conf->saved_nics, | |
1110 | (conf->num_savednics+1)*sizeof(struct saved_nic)); | |
408da065 | 1111 | if (!conf->saved_nics) |
7b35f3d6 | 1112 | return -1; |
7b35f3d6 SH |
1113 | conf->saved_nics[conf->num_savednics].ifindex = netdev->ifindex; |
1114 | conf->saved_nics[conf->num_savednics].orig_name = strdup(netdev->link); | |
408da065 | 1115 | if (!conf->saved_nics[conf->num_savednics].orig_name) |
7b35f3d6 | 1116 | return -1; |
408da065 | 1117 | INFO("Stored saved_nic #%d idx %d name %s.", conf->num_savednics, |
7b35f3d6 SH |
1118 | conf->saved_nics[conf->num_savednics].ifindex, |
1119 | conf->saved_nics[conf->num_savednics].orig_name); | |
1120 | conf->num_savednics++; | |
1121 | } | |
1122 | ||
1123 | return 0; | |
1124 | } | |
1125 | ||
ae467c54 | 1126 | static int lxc_recv_ttys_from_child(struct lxc_handler *handler) |
e8bd4e43 | 1127 | { |
ae467c54 CB |
1128 | int i; |
1129 | int *ttyfds; | |
1130 | struct lxc_pty_info *pty_info; | |
1131 | int ret = -1; | |
f07fa8df | 1132 | int sock = handler->ttysock[1]; |
e8bd4e43 | 1133 | struct lxc_conf *conf = handler->conf; |
e8bd4e43 | 1134 | struct lxc_tty_info *tty_info = &conf->tty_info; |
ae467c54 | 1135 | size_t num_ttyfds = (2 * conf->tty); |
e8bd4e43 SH |
1136 | |
1137 | if (!conf->tty) | |
1138 | return 0; | |
1139 | ||
408da065 CB |
1140 | tty_info->pty_info = malloc(sizeof(*tty_info->pty_info) * conf->tty); |
1141 | if (!tty_info->pty_info) | |
e8bd4e43 | 1142 | return -1; |
e8bd4e43 | 1143 | |
ae467c54 CB |
1144 | ttyfds = malloc(num_ttyfds * sizeof(int)); |
1145 | if (!ttyfds) | |
1146 | return -1; | |
1147 | ||
1148 | ret = lxc_abstract_unix_recv_fds(sock, ttyfds, num_ttyfds, NULL, 0); | |
1149 | for (i = 0; (ret >= 0 && *ttyfds != -1) && (i < num_ttyfds); i++) { | |
1150 | pty_info = &tty_info->pty_info[i / 2]; | |
e8bd4e43 | 1151 | pty_info->busy = 0; |
ae467c54 CB |
1152 | pty_info->slave = ttyfds[i++]; |
1153 | pty_info->master = ttyfds[i]; | |
1154 | TRACE("received pty with master fd %d and slave fd %d from " | |
1155 | "parent", pty_info->master, pty_info->slave); | |
e8bd4e43 | 1156 | } |
ae467c54 | 1157 | |
e8bd4e43 SH |
1158 | tty_info->nbtty = conf->tty; |
1159 | ||
ae467c54 CB |
1160 | free(ttyfds); |
1161 | ||
1162 | if (ret < 0) | |
1163 | ERROR("failed to receive %d ttys from child: %s", conf->tty, | |
1164 | strerror(errno)); | |
1165 | else | |
1166 | TRACE("received %d ttys from child", conf->tty); | |
1167 | ||
1168 | return ret; | |
e8bd4e43 SH |
1169 | } |
1170 | ||
f813849c TA |
1171 | void resolve_clone_flags(struct lxc_handler *handler) |
1172 | { | |
1173 | handler->clone_flags = CLONE_NEWPID | CLONE_NEWNS; | |
1174 | ||
408da065 | 1175 | if (!lxc_list_empty(&handler->conf->id_map)) |
f813849c | 1176 | handler->clone_flags |= CLONE_NEWUSER; |
f813849c TA |
1177 | |
1178 | if (handler->conf->inherit_ns_fd[LXC_NS_NET] == -1) { | |
408da065 | 1179 | if (!lxc_requests_empty_network(handler)) |
f813849c TA |
1180 | handler->clone_flags |= CLONE_NEWNET; |
1181 | } else { | |
198cbbaa | 1182 | INFO("Inheriting a NET namespace."); |
f813849c TA |
1183 | } |
1184 | ||
408da065 | 1185 | if (handler->conf->inherit_ns_fd[LXC_NS_IPC] == -1) |
f813849c | 1186 | handler->clone_flags |= CLONE_NEWIPC; |
408da065 | 1187 | else |
198cbbaa | 1188 | INFO("Inheriting an IPC namespace."); |
f813849c | 1189 | |
408da065 | 1190 | if (handler->conf->inherit_ns_fd[LXC_NS_UTS] == -1) |
f813849c | 1191 | handler->clone_flags |= CLONE_NEWUTS; |
408da065 | 1192 | else |
198cbbaa | 1193 | INFO("Inheriting a UTS namespace."); |
f813849c TA |
1194 | } |
1195 | ||
480588e6 CB |
1196 | /* lxc_spawn() performs crucial setup tasks and clone()s the new process which |
1197 | * exec()s the requested container binary. | |
1198 | * Note that lxc_spawn() runs in the parent namespaces. Any operations performed | |
1199 | * right here should be double checked if they'd pose a security risk. (For | |
1200 | * example, any {u}mount() operations performed here will be reflected on the | |
1201 | * host!) | |
1202 | */ | |
74a3920a | 1203 | static int lxc_spawn(struct lxc_handler *handler) |
59eb99ba | 1204 | { |
b98f7d6e | 1205 | int failed_before_rename = 0; |
ffe1e01a | 1206 | const char *name = handler->name; |
7e4dfe0b | 1207 | bool cgroups_connected = false; |
9f30a190 | 1208 | int saved_ns_fd[LXC_NS_MAX]; |
5b1e83cb | 1209 | int preserve_mask = 0, i, flags; |
658979c5 | 1210 | int netpipepair[2], nveths; |
57927bf2 CB |
1211 | bool wants_to_map_ids; |
1212 | struct lxc_list *id_map; | |
9f30a190 | 1213 | |
2f2623ec | 1214 | netpipe = -1; |
57927bf2 CB |
1215 | id_map = &handler->conf->id_map; |
1216 | wants_to_map_ids = !lxc_list_empty(id_map); | |
2f2623ec | 1217 | |
9f30a190 | 1218 | for (i = 0; i < LXC_NS_MAX; i++) |
6c544cb3 | 1219 | if (handler->conf->inherit_ns_fd[i] != -1) |
9f30a190 | 1220 | preserve_mask |= ns_info[i].clone_flag; |
50e98013 | 1221 | |
3c22086f | 1222 | if (lxc_sync_init(handler)) |
9d7f9e52 | 1223 | return -1; |
0ad19a3f | 1224 | |
e8bd4e43 SH |
1225 | if (socketpair(AF_UNIX, SOCK_DGRAM, 0, handler->ttysock) < 0) { |
1226 | lxc_sync_fini(handler); | |
1227 | return -1; | |
1228 | } | |
1229 | ||
f813849c | 1230 | resolve_clone_flags(handler); |
9f30a190 | 1231 | |
f813849c | 1232 | if (handler->clone_flags & CLONE_NEWNET) { |
26b797f3 SH |
1233 | if (!lxc_list_empty(&handler->conf->network)) { |
1234 | ||
9f30a190 MM |
1235 | /* Find gateway addresses from the link device, which is |
1236 | * no longer accessible inside the container. Do this | |
1237 | * before creating network interfaces, since goto | |
408da065 CB |
1238 | * out_delete_net does not work before lxc_clone. |
1239 | */ | |
9f30a190 | 1240 | if (lxc_find_gateway_addresses(handler)) { |
408da065 | 1241 | ERROR("Failed to find gateway addresses."); |
9f30a190 MM |
1242 | lxc_sync_fini(handler); |
1243 | return -1; | |
1244 | } | |
1245 | ||
408da065 CB |
1246 | /* That should be done before the clone because we will |
1247 | * fill the netdev index and use them in the child. | |
9f30a190 | 1248 | */ |
e337179a | 1249 | if (lxc_setup_networks_in_parent_namespaces(handler)) { |
408da065 | 1250 | ERROR("Failed to create the network."); |
9f30a190 MM |
1251 | lxc_sync_fini(handler); |
1252 | return -1; | |
1253 | } | |
19a26f82 MK |
1254 | } |
1255 | ||
9f30a190 | 1256 | if (save_phys_nics(handler->conf)) { |
408da065 | 1257 | ERROR("Failed to save physical nic info."); |
9f30a190 | 1258 | goto out_abort; |
82d5ae15 DL |
1259 | } |
1260 | } | |
1261 | ||
d4ef7c50 | 1262 | if (!cgroup_init(handler)) { |
408da065 | 1263 | ERROR("Failed initializing cgroup support."); |
33ad9f1a CS |
1264 | goto out_delete_net; |
1265 | } | |
1266 | ||
7e4dfe0b SH |
1267 | cgroups_connected = true; |
1268 | ||
d4ef7c50 | 1269 | if (!cgroup_create(handler)) { |
408da065 | 1270 | ERROR("Failed creating cgroups."); |
47d8fb3b CS |
1271 | goto out_delete_net; |
1272 | } | |
1273 | ||
408da065 CB |
1274 | /* If the rootfs is not a blockdev, prevent the container from marking |
1275 | * it readonly. | |
1276 | * If the container is unprivileged then skip rootfs pinning. | |
0c547523 | 1277 | */ |
0ee35059 | 1278 | if (!wants_to_map_ids) { |
5e32a990 ÇO |
1279 | handler->pinfd = pin_rootfs(handler->conf->rootfs.path); |
1280 | if (handler->pinfd == -1) | |
408da065 | 1281 | INFO("Failed to pin the rootfs for container \"%s\".", handler->name); |
5e32a990 | 1282 | } |
0c547523 | 1283 | |
4d8ac866 | 1284 | if (!preserve_ns(saved_ns_fd, preserve_mask, getpid())) |
cd43d2d1 | 1285 | goto out_delete_net; |
4d8ac866 | 1286 | |
cd43d2d1 SH |
1287 | if (attach_ns(handler->conf->inherit_ns_fd) < 0) |
1288 | goto out_delete_net; | |
9f30a190 | 1289 | |
658979c5 SH |
1290 | if (am_unpriv() && (nveths = count_veths(&handler->conf->network))) { |
1291 | if (pipe(netpipepair) < 0) { | |
408da065 | 1292 | SYSERROR("Failed to create pipe."); |
658979c5 SH |
1293 | goto out_delete_net; |
1294 | } | |
408da065 | 1295 | /* Store netpipe in the global var for do_start's use. */ |
658979c5 SH |
1296 | netpipe = netpipepair[0]; |
1297 | } | |
1298 | ||
408da065 | 1299 | /* Create a process in a new set of namespaces. */ |
5b1e83cb | 1300 | flags = handler->clone_flags; |
408da065 CB |
1301 | if (handler->clone_flags & CLONE_NEWUSER) { |
1302 | /* If CLONE_NEWUSER and CLONE_NEWNET was requested, we need to | |
1303 | * clone a new user namespace first and only later unshare our | |
1304 | * network namespace to ensure that network devices ownership is | |
1305 | * set up correctly. | |
1306 | */ | |
5b1e83cb | 1307 | flags &= ~CLONE_NEWNET; |
408da065 | 1308 | } |
9fac8fbb | 1309 | handler->pid = lxc_clone(do_start, handler, flags); |
59eb99ba | 1310 | if (handler->pid < 0) { |
408da065 | 1311 | SYSERROR("Failed to clone a new set of namespaces."); |
7fef7a06 | 1312 | goto out_delete_net; |
0ad19a3f | 1313 | } |
9662e444 CB |
1314 | for (i = 0; i < LXC_NS_MAX; i++) |
1315 | if (flags & ns_info[i].clone_flag) | |
1316 | INFO("Cloned %s.", ns_info[i].flag_name); | |
0ad19a3f | 1317 | |
4d8ac866 CB |
1318 | if (!preserve_ns(handler->nsfd, handler->clone_flags | preserve_mask, handler->pid)) |
1319 | INFO("Failed to preserve namespace for lxc.hook.stop."); | |
b6b2b194 | 1320 | |
cd43d2d1 | 1321 | if (attach_ns(saved_ns_fd)) |
408da065 | 1322 | WARN("Failed to restore saved namespaces."); |
9f30a190 | 1323 | |
3c22086f CLG |
1324 | lxc_sync_fini_child(handler); |
1325 | ||
408da065 CB |
1326 | /* Map the container uids. The container became an invalid userid the |
1327 | * moment it was cloned with CLONE_NEWUSER. This call doesn't change | |
1328 | * anything immediately, but allows the container to setuid(0) (0 being | |
1329 | * mapped to something else on the host.) later to become a valid uid | |
1330 | * again. | |
1331 | */ | |
57927bf2 | 1332 | if (wants_to_map_ids && lxc_map_ids(id_map, handler->pid)) { |
408da065 | 1333 | ERROR("Failed to set up id mapping."); |
5b1e83cb SH |
1334 | goto out_delete_net; |
1335 | } | |
1336 | ||
1337 | if (lxc_sync_wake_child(handler, LXC_SYNC_STARTUP)) { | |
99a6af52 | 1338 | failed_before_rename = 1; |
5b1e83cb SH |
1339 | goto out_delete_net; |
1340 | } | |
1341 | ||
1342 | if (lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE)) { | |
1343 | failed_before_rename = 1; | |
1344 | goto out_delete_net; | |
1345 | } | |
0ad19a3f | 1346 | |
d4ef7c50 | 1347 | if (!cgroup_create_legacy(handler)) { |
408da065 | 1348 | ERROR("Failed to setup legacy cgroups for container \"%s\".", name); |
ae5c8b8e | 1349 | goto out_delete_net; |
33ad9f1a | 1350 | } |
9daf6f5d | 1351 | if (!cgroup_setup_limits(handler, false)) { |
408da065 | 1352 | ERROR("Failed to setup cgroup limits for container \"%s\".", name); |
6031a6e5 DE |
1353 | goto out_delete_net; |
1354 | } | |
1355 | ||
d4ef7c50 | 1356 | if (!cgroup_enter(handler)) |
7fef7a06 | 1357 | goto out_delete_net; |
218d4250 | 1358 | |
0996e18a SH |
1359 | if (!cgroup_chown(handler)) |
1360 | goto out_delete_net; | |
1361 | ||
99a6af52 MN |
1362 | if (failed_before_rename) |
1363 | goto out_delete_net; | |
1364 | ||
408da065 | 1365 | /* Create the network configuration. */ |
d5088cf2 | 1366 | if (handler->clone_flags & CLONE_NEWNET) { |
c43cbc04 | 1367 | if (lxc_assign_network(handler->lxcpath, handler->name, |
408da065 CB |
1368 | &handler->conf->network, handler->pid)) { |
1369 | ERROR("Failed to create the configured network."); | |
7fef7a06 | 1370 | goto out_delete_net; |
82d5ae15 | 1371 | } |
0ad19a3f | 1372 | } |
1373 | ||
658979c5 SH |
1374 | if (netpipe != -1) { |
1375 | struct lxc_list *iterator; | |
1376 | struct lxc_netdev *netdev; | |
1377 | ||
1378 | close(netpipe); | |
1379 | lxc_list_for_each(iterator, &handler->conf->network) { | |
1380 | netdev = iterator->elem; | |
1381 | if (netdev->type != LXC_NET_VETH) | |
1382 | continue; | |
1383 | if (write(netpipepair[1], netdev->name, IFNAMSIZ) != IFNAMSIZ) { | |
408da065 | 1384 | ERROR("Error writing veth name to container."); |
658979c5 SH |
1385 | goto out_delete_net; |
1386 | } | |
1387 | } | |
1388 | close(netpipepair[1]); | |
1389 | } | |
1390 | ||
408da065 CB |
1391 | /* Tell the child to continue its initialization. We'll get |
1392 | * LXC_SYNC_CGROUP when it is ready for us to setup cgroups. | |
3c22086f CLG |
1393 | */ |
1394 | if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE)) | |
544a48a0 SH |
1395 | goto out_delete_net; |
1396 | ||
c6d09e15 WB |
1397 | if (!lxc_list_empty(&handler->conf->limits) && setup_resource_limits(&handler->conf->limits, handler->pid)) { |
1398 | ERROR("failed to setup resource limits for '%s'", name); | |
84ff3af7 | 1399 | goto out_delete_net; |
c6d09e15 WB |
1400 | } |
1401 | ||
f4152036 CB |
1402 | if (lxc_sync_barrier_child(handler, LXC_SYNC_CGROUP_UNSHARE)) |
1403 | goto out_delete_net; | |
1404 | ||
9daf6f5d | 1405 | if (!cgroup_setup_limits(handler, true)) { |
408da065 | 1406 | ERROR("Failed to setup the devices cgroup for container \"%s\".", name); |
b98f7d6e | 1407 | goto out_delete_net; |
544a48a0 | 1408 | } |
f4152036 | 1409 | TRACE("Set up cgroup device limits"); |
544a48a0 | 1410 | |
73d28d42 | 1411 | cgroup_disconnect(); |
7e4dfe0b | 1412 | cgroups_connected = false; |
73d28d42 | 1413 | |
408da065 CB |
1414 | /* Tell the child to complete its initialization and wait for it to exec |
1415 | * or return an error. (The child will never return | |
1416 | * LXC_SYNC_POST_CGROUP+1. It will either close the sync pipe, causing | |
1417 | * lxc_sync_barrier_child to return success, or return a different | |
1418 | * value, causing us to error out). | |
544a48a0 SH |
1419 | */ |
1420 | if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CGROUP)) | |
3c22086f | 1421 | return -1; |
0ad19a3f | 1422 | |
f4152036 CB |
1423 | /* Read tty fds allocated by child. */ |
1424 | if (lxc_recv_ttys_from_child(handler) < 0) { | |
1425 | ERROR("Failed to receive tty info from child process."); | |
1426 | goto out_delete_net; | |
1427 | } | |
1428 | ||
23c53af9 | 1429 | if (handler->ops->post_start(handler, handler->data)) |
e6126dbe MN |
1430 | goto out_abort; |
1431 | ||
25c2aca5 | 1432 | if (lxc_set_state(name, handler, RUNNING)) { |
408da065 CB |
1433 | ERROR("Failed to set state for container \"%s\" to \"%s\".", name, |
1434 | lxc_state2str(RUNNING)); | |
59eb99ba | 1435 | goto out_abort; |
3f21c114 | 1436 | } |
22ebac19 | 1437 | |
3c22086f | 1438 | lxc_sync_fini(handler); |
738d0deb | 1439 | handler->netnsfd = lxc_preserve_ns(handler->pid, "net"); |
0c547523 | 1440 | |
e6126dbe | 1441 | return 0; |
1ac470c0 | 1442 | |
7fef7a06 | 1443 | out_delete_net: |
7e4dfe0b SH |
1444 | if (cgroups_connected) |
1445 | cgroup_disconnect(); | |
d5088cf2 | 1446 | if (handler->clone_flags & CLONE_NEWNET) |
74a2b586 | 1447 | lxc_delete_network(handler); |
59eb99ba DL |
1448 | out_abort: |
1449 | lxc_abort(name, handler); | |
3c22086f | 1450 | lxc_sync_fini(handler); |
5c068da9 SH |
1451 | if (handler->pinfd >= 0) { |
1452 | close(handler->pinfd); | |
1453 | handler->pinfd = -1; | |
1454 | } | |
1455 | ||
b79fcd86 | 1456 | return -1; |
59eb99ba | 1457 | } |
0ad19a3f | 1458 | |
aa460476 | 1459 | int __lxc_start(const char *name, struct lxc_handler *handler, |
507cee36 TA |
1460 | struct lxc_operations* ops, void *data, const char *lxcpath, |
1461 | bool backgrounded) | |
59eb99ba | 1462 | { |
59eb99ba | 1463 | int status; |
aa460476 | 1464 | int err = -1; |
358daf49 | 1465 | bool removed_all_netdevs = true; |
aa460476 | 1466 | struct lxc_conf *conf = handler->conf; |
80090207 | 1467 | |
aa460476 | 1468 | if (lxc_init(name, handler) < 0) { |
408da065 | 1469 | ERROR("Failed to initialize container \"%s\".", name); |
66aeffc7 | 1470 | return -1; |
0ad19a3f | 1471 | } |
ee70bf78 CLG |
1472 | handler->ops = ops; |
1473 | handler->data = data; | |
507cee36 | 1474 | handler->backgrounded = backgrounded; |
738d0deb | 1475 | handler->netnsfd = -1; |
e6126dbe | 1476 | |
76a26f55 | 1477 | if (!attach_block_device(handler->conf)) { |
408da065 | 1478 | ERROR("Failed to attach block device."); |
76a26f55 SH |
1479 | goto out_fini_nonet; |
1480 | } | |
1481 | ||
35120d9c | 1482 | if (geteuid() == 0 && !lxc_list_empty(&conf->id_map)) { |
408da065 | 1483 | /* If the backing store is a device, mount it here and now. */ |
35120d9c SH |
1484 | if (rootfs_is_blockdev(conf)) { |
1485 | if (unshare(CLONE_NEWNS) < 0) { | |
408da065 | 1486 | ERROR("Failed to unshare CLONE_NEWNS."); |
35120d9c SH |
1487 | goto out_fini_nonet; |
1488 | } | |
408da065 CB |
1489 | INFO("Unshared CLONE_NEWNS."); |
1490 | ||
6a0c909a | 1491 | remount_all_slave(); |
35120d9c | 1492 | if (do_rootfs_setup(conf, name, lxcpath) < 0) { |
408da065 | 1493 | ERROR("Error setting up rootfs mount as root before spawn."); |
35120d9c SH |
1494 | goto out_fini_nonet; |
1495 | } | |
408da065 | 1496 | INFO("Set up container rootfs as host root."); |
35120d9c SH |
1497 | } |
1498 | } | |
1499 | ||
23c53af9 | 1500 | err = lxc_spawn(handler); |
59eb99ba | 1501 | if (err) { |
408da065 | 1502 | ERROR("Failed to spawn container \"%s\".", name); |
76a26f55 | 1503 | goto out_detach_blockdev; |
0ad19a3f | 1504 | } |
1505 | ||
8bee8851 WB |
1506 | handler->conf->reboot = 0; |
1507 | ||
3a0f472d | 1508 | err = lxc_poll(name, handler); |
e043236e | 1509 | if (err) { |
408da065 | 1510 | ERROR("LXC mainloop exited with error: %d.", err); |
738d0deb CB |
1511 | if (handler->netnsfd >= 0) { |
1512 | close(handler->netnsfd); | |
1513 | handler->netnsfd = -1; | |
1514 | } | |
59eb99ba DL |
1515 | goto out_abort; |
1516 | } | |
0ad19a3f | 1517 | |
3a0f472d | 1518 | while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR) |
1bc5cc8c | 1519 | continue; |
e043236e | 1520 | |
408da065 CB |
1521 | /* If the child process exited but was not signaled, it didn't call |
1522 | * reboot. This should mean it was an lxc-execute which simply exited. | |
1523 | * In any case, treat it as a 'halt'. | |
8b004f07 | 1524 | */ |
d028235d | 1525 | if (WIFSIGNALED(status)) { |
8b004f07 SH |
1526 | switch(WTERMSIG(status)) { |
1527 | case SIGINT: /* halt */ | |
408da065 | 1528 | DEBUG("Container \"%s\" is halting.", name); |
8b004f07 SH |
1529 | break; |
1530 | case SIGHUP: /* reboot */ | |
408da065 | 1531 | DEBUG("Container \"%s\" is rebooting.", name); |
8b004f07 SH |
1532 | handler->conf->reboot = 1; |
1533 | break; | |
c2b9bd9e | 1534 | case SIGSYS: /* seccomp */ |
408da065 | 1535 | DEBUG("Container \"%s\" violated its seccomp policy.", name); |
c2b9bd9e | 1536 | break; |
8b004f07 | 1537 | default: |
408da065 | 1538 | DEBUG("Unknown exit status for container \"%s\" init %d.", name, WTERMSIG(status)); |
8b004f07 SH |
1539 | break; |
1540 | } | |
d028235d | 1541 | } |
828695d9 | 1542 | |
ce5782df | 1543 | DEBUG("Pushing physical nics back to host namespace"); |
738d0deb | 1544 | lxc_restore_phys_nics_to_netns(handler->netnsfd, handler->conf); |
ce5782df | 1545 | |
408da065 | 1546 | DEBUG("Tearing down virtual network devices used by container \"%s\".", name); |
358daf49 | 1547 | removed_all_netdevs = lxc_delete_network(handler); |
ce5782df | 1548 | |
5c068da9 SH |
1549 | if (handler->pinfd >= 0) { |
1550 | close(handler->pinfd); | |
1551 | handler->pinfd = -1; | |
1552 | } | |
1553 | ||
1787abca | 1554 | lxc_monitor_send_exit_code(name, status, handler->lxcpath); |
3a0f472d | 1555 | err = lxc_error_set_and_log(handler->pid, status); |
9d7f9e52 | 1556 | out_fini: |
358daf49 | 1557 | if (!removed_all_netdevs) { |
408da065 | 1558 | DEBUG("Failed tearing down network devices used by container. Trying again!"); |
358daf49 CB |
1559 | removed_all_netdevs = lxc_delete_network(handler); |
1560 | if (!removed_all_netdevs) | |
1561 | DEBUG("Failed tearing down network devices used by container. Not trying again!"); | |
1562 | } | |
74a2b586 | 1563 | |
76a26f55 SH |
1564 | out_detach_blockdev: |
1565 | detach_block_device(handler->conf); | |
1566 | ||
74a2b586 | 1567 | out_fini_nonet: |
3a0f472d | 1568 | lxc_fini(name, handler); |
0ad19a3f | 1569 | return err; |
1570 | ||
59eb99ba | 1571 | out_abort: |
3a0f472d | 1572 | lxc_abort(name, handler); |
9d7f9e52 | 1573 | goto out_fini; |
0ad19a3f | 1574 | } |
ee70bf78 CLG |
1575 | |
1576 | struct start_args { | |
1577 | char *const *argv; | |
1578 | }; | |
1579 | ||
1580 | static int start(struct lxc_handler *handler, void* data) | |
1581 | { | |
1582 | struct start_args *arg = data; | |
1583 | ||
408da065 | 1584 | NOTICE("Exec'ing \"%s\".", arg->argv[0]); |
ee70bf78 CLG |
1585 | |
1586 | execvp(arg->argv[0], arg->argv); | |
408da065 | 1587 | SYSERROR("Failed to exec \"%s\".", arg->argv[0]); |
ee70bf78 CLG |
1588 | return 0; |
1589 | } | |
1590 | ||
1591 | static int post_start(struct lxc_handler *handler, void* data) | |
1592 | { | |
1593 | struct start_args *arg = data; | |
1594 | ||
408da065 | 1595 | NOTICE("Started \"%s\" with pid \"%d\".", arg->argv[0], handler->pid); |
ee70bf78 CLG |
1596 | return 0; |
1597 | } | |
1598 | ||
1599 | static struct lxc_operations start_ops = { | |
1600 | .start = start, | |
1601 | .post_start = post_start | |
1602 | }; | |
1603 | ||
aa460476 | 1604 | int lxc_start(const char *name, char *const argv[], struct lxc_handler *handler, |
507cee36 | 1605 | const char *lxcpath, bool backgrounded) |
ee70bf78 CLG |
1606 | { |
1607 | struct start_args start_arg = { | |
1608 | .argv = argv, | |
1609 | }; | |
1610 | ||
aa460476 | 1611 | return __lxc_start(name, handler, &start_ops, &start_arg, lxcpath, backgrounded); |
ee70bf78 | 1612 | } |
28272964 CB |
1613 | |
1614 | static void lxc_destroy_container_on_signal(struct lxc_handler *handler, | |
1615 | const char *name) | |
1616 | { | |
1617 | char destroy[MAXPATHLEN]; | |
1618 | bool bret = true; | |
1619 | int ret = 0; | |
f01f7975 | 1620 | struct lxc_container *c; |
df31363a | 1621 | if (handler->conf->rootfs.path && handler->conf->rootfs.mount) { |
28272964 CB |
1622 | bret = do_destroy_container(handler->conf); |
1623 | if (!bret) { | |
408da065 | 1624 | ERROR("Error destroying rootfs for container \"%s\".", name); |
28272964 CB |
1625 | return; |
1626 | } | |
1627 | } | |
408da065 | 1628 | INFO("Destroyed rootfs for container \"%s\".", name); |
28272964 CB |
1629 | |
1630 | ret = snprintf(destroy, MAXPATHLEN, "%s/%s", handler->lxcpath, name); | |
1631 | if (ret < 0 || ret >= MAXPATHLEN) { | |
408da065 | 1632 | ERROR("Error destroying directory for container \"%s\".", name); |
28272964 CB |
1633 | return; |
1634 | } | |
1635 | ||
f01f7975 CB |
1636 | c = lxc_container_new(name, handler->lxcpath); |
1637 | if (c) { | |
1638 | if (container_disk_lock(c)) { | |
408da065 | 1639 | INFO("Could not update lxc_snapshots file."); |
f01f7975 CB |
1640 | lxc_container_put(c); |
1641 | } else { | |
1642 | mod_all_rdeps(c, false); | |
1643 | container_disk_unlock(c); | |
1644 | lxc_container_put(c); | |
1645 | } | |
1646 | } | |
1647 | ||
28272964 | 1648 | if (am_unpriv()) |
c9b7c33e CB |
1649 | ret = userns_exec_1(handler->conf, lxc_rmdir_onedev_wrapper, |
1650 | destroy, "lxc_rmdir_onedev_wrapper"); | |
28272964 CB |
1651 | else |
1652 | ret = lxc_rmdir_onedev(destroy, NULL); | |
1653 | ||
1654 | if (ret < 0) { | |
408da065 | 1655 | ERROR("Error destroying directory for container \"%s\".", name); |
28272964 CB |
1656 | return; |
1657 | } | |
408da065 | 1658 | INFO("Destroyed directory for container \"%s\".", name); |
28272964 CB |
1659 | } |
1660 | ||
1661 | static int lxc_rmdir_onedev_wrapper(void *data) | |
1662 | { | |
1663 | char *arg = (char *) data; | |
1664 | return lxc_rmdir_onedev(arg, NULL); | |
1665 | } | |
1666 | ||
1667 | static bool do_destroy_container(struct lxc_conf *conf) { | |
d028235d | 1668 | if (am_unpriv()) { |
10bc1861 CB |
1669 | if (userns_exec_1(conf, storage_destroy_wrapper, conf, |
1670 | "storage_destroy_wrapper") < 0) | |
d028235d | 1671 | return false; |
10bc1861 | 1672 | |
d028235d SG |
1673 | return true; |
1674 | } | |
10bc1861 CB |
1675 | |
1676 | return storage_destroy(conf); | |
28272964 | 1677 | } |