]>
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> |
35a02107 CB |
8 | * Serge Hallyn <serge@hallyn.com> |
9 | * Christian Brauner <christian.brauner@ubuntu.com> | |
0ad19a3f | 10 | * |
11 | * This library is free software; you can redistribute it and/or | |
12 | * modify it under the terms of the GNU Lesser General Public | |
13 | * License as published by the Free Software Foundation; either | |
14 | * version 2.1 of the License, or (at your option) any later version. | |
15 | * | |
16 | * This library is distributed in the hope that it will be useful, | |
17 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
18 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
19 | * Lesser General Public License for more details. | |
20 | * | |
21 | * You should have received a copy of the GNU Lesser General Public | |
22 | * License along with this library; if not, write to the Free Software | |
250b1eec | 23 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
0ad19a3f | 24 | */ |
25 | ||
f3787121 | 26 | #define _GNU_SOURCE |
f549edcc GK |
27 | #include "config.h" |
28 | ||
f3787121 | 29 | #include <alloca.h> |
0ad19a3f | 30 | #include <dirent.h> |
31 | #include <errno.h> | |
b0a33c1e | 32 | #include <fcntl.h> |
c476bdce | 33 | #include <grp.h> |
37515ebd | 34 | #include <poll.h> |
f3787121 CB |
35 | #include <signal.h> |
36 | #include <stdio.h> | |
37 | #include <stdlib.h> | |
38 | #include <string.h> | |
39 | #include <unistd.h> | |
0ad19a3f | 40 | #include <sys/file.h> |
f4d507d5 | 41 | #include <sys/mount.h> |
f3787121 | 42 | #include <sys/param.h> |
0ad19a3f | 43 | #include <sys/prctl.h> |
f3787121 CB |
44 | #include <sys/socket.h> |
45 | #include <sys/stat.h> | |
46 | #include <sys/syscall.h> | |
ddceb1f9 | 47 | #include <sys/types.h> |
b0a33c1e | 48 | #include <sys/un.h> |
f3787121 | 49 | #include <sys/wait.h> |
ff218c25 | 50 | |
e37dda71 | 51 | #if HAVE_LIBCAP |
495d2046 SG |
52 | #include <sys/capability.h> |
53 | #endif | |
54 | ||
955e2a02 | 55 | #ifndef HAVE_DECL_PR_CAPBSET_DROP |
656994bb MH |
56 | #define PR_CAPBSET_DROP 24 |
57 | #endif | |
58 | ||
955e2a02 CB |
59 | #ifndef HAVE_DECL_PR_SET_NO_NEW_PRIVS |
60 | #define PR_SET_NO_NEW_PRIVS 38 | |
61 | #endif | |
62 | ||
63 | #ifndef HAVE_DECL_PR_GET_NO_NEW_PRIVS | |
64 | #define PR_GET_NO_NEW_PRIVS 39 | |
65 | #endif | |
66 | ||
f3787121 CB |
67 | #include "af_unix.h" |
68 | #include "caps.h" | |
563f2f2c | 69 | #include "cgroup.h" |
f3787121 | 70 | #include "commands.h" |
bbf5cf35 | 71 | #include "commands_utils.h" |
f3787121 | 72 | #include "conf.h" |
74c6e2b0 | 73 | #include "confile_utils.h" |
f3787121 | 74 | #include "console.h" |
e2bcd7db | 75 | #include "error.h" |
f3787121 | 76 | #include "log.h" |
f2d5a09d | 77 | #include "lxccontainer.h" |
f3787121 CB |
78 | #include "lxclock.h" |
79 | #include "lxcseccomp.h" | |
f3787121 | 80 | #include "mainloop.h" |
63376d7d | 81 | #include "monitor.h" |
f549edcc | 82 | #include "namespace.h" |
811ef482 | 83 | #include "network.h" |
f3787121 | 84 | #include "start.h" |
28d832c4 | 85 | #include "storage.h" |
f2d5a09d | 86 | #include "storage_utils.h" |
f3787121 CB |
87 | #include "sync.h" |
88 | #include "utils.h" | |
fe4de9a6 | 89 | #include "lsm/lsm.h" |
36eb9bde CLG |
90 | |
91 | lxc_log_define(lxc_start, lxc); | |
92 | ||
f01f7975 | 93 | extern void mod_all_rdeps(struct lxc_container *c, bool inc); |
790255cf | 94 | static bool do_destroy_container(struct lxc_handler *handler); |
28272964 CB |
95 | static int lxc_rmdir_onedev_wrapper(void *data); |
96 | static void lxc_destroy_container_on_signal(struct lxc_handler *handler, | |
97 | const char *name); | |
98 | ||
c8154066 SH |
99 | static void print_top_failing_dir(const char *path) |
100 | { | |
101 | size_t len = strlen(path); | |
f3787121 | 102 | char *copy = alloca(len + 1), *p, *e, saved; |
c8154066 SH |
103 | strcpy(copy, path); |
104 | ||
105 | p = copy; | |
106 | e = copy + len; | |
107 | while (p < e) { | |
f3787121 CB |
108 | while (p < e && *p == '/') |
109 | p++; | |
110 | while (p < e && *p != '/') | |
111 | p++; | |
c8154066 SH |
112 | saved = *p; |
113 | *p = '\0'; | |
114 | if (access(copy, X_OK)) { | |
408da065 CB |
115 | SYSERROR("Could not access %s. Please grant it x " |
116 | "access, or add an ACL for the container " | |
117 | "root.", copy); | |
c8154066 SH |
118 | return; |
119 | } | |
120 | *p = saved; | |
121 | } | |
122 | } | |
123 | ||
408da065 CB |
124 | static void close_ns(int ns_fd[LXC_NS_MAX]) |
125 | { | |
9f30a190 MM |
126 | int i; |
127 | ||
9f30a190 MM |
128 | for (i = 0; i < LXC_NS_MAX; i++) { |
129 | if (ns_fd[i] > -1) { | |
130 | close(ns_fd[i]); | |
131 | ns_fd[i] = -1; | |
132 | } | |
133 | } | |
9f30a190 MM |
134 | } |
135 | ||
4d8ac866 | 136 | /* preserve_ns: open /proc/@pid/ns/@ns for each namespace specified |
62d05d9b | 137 | * in clone_flags. |
4d8ac866 | 138 | * Return true on success, false on failure. |
62d05d9b | 139 | */ |
4d8ac866 | 140 | static bool preserve_ns(int ns_fd[LXC_NS_MAX], int clone_flags, pid_t pid) |
f3787121 | 141 | { |
62d05d9b | 142 | int i, ret; |
9f30a190 | 143 | |
9f30a190 MM |
144 | for (i = 0; i < LXC_NS_MAX; i++) |
145 | ns_fd[i] = -1; | |
146 | ||
4d8ac866 CB |
147 | ret = lxc_preserve_ns(pid, ""); |
148 | if (ret < 0) { | |
149 | SYSERROR("Kernel does not support attaching to namespaces."); | |
62d05d9b | 150 | return false; |
4d8ac866 CB |
151 | } else { |
152 | close(ret); | |
cd43d2d1 SH |
153 | } |
154 | ||
9f30a190 MM |
155 | for (i = 0; i < LXC_NS_MAX; i++) { |
156 | if ((clone_flags & ns_info[i].clone_flag) == 0) | |
157 | continue; | |
4d8ac866 | 158 | ns_fd[i] = lxc_preserve_ns(pid, ns_info[i].proc_name); |
9f30a190 MM |
159 | if (ns_fd[i] < 0) |
160 | goto error; | |
161 | } | |
162 | ||
62d05d9b | 163 | return true; |
9f30a190 MM |
164 | |
165 | error: | |
4d8ac866 CB |
166 | if (errno == ENOENT) |
167 | SYSERROR("Kernel does not support attaching to %s namespaces.", ns_info[i].proc_name); | |
168 | else | |
169 | SYSERROR("Failed to open file descriptor for %s namespace: %s.", ns_info[i].proc_name, strerror(errno)); | |
9f30a190 | 170 | close_ns(ns_fd); |
62d05d9b | 171 | return false; |
9f30a190 MM |
172 | } |
173 | ||
174 | static int attach_ns(const int ns_fd[LXC_NS_MAX]) { | |
175 | int i; | |
176 | ||
177 | for (i = 0; i < LXC_NS_MAX; i++) { | |
178 | if (ns_fd[i] < 0) | |
179 | continue; | |
180 | ||
181 | if (setns(ns_fd[i], 0) != 0) | |
182 | goto error; | |
183 | } | |
184 | return 0; | |
185 | ||
186 | error: | |
408da065 | 187 | SYSERROR("Failed to attach %s namespace.", ns_info[i].proc_name); |
9f30a190 MM |
188 | return -1; |
189 | } | |
190 | ||
80090207 CLG |
191 | static int match_fd(int fd) |
192 | { | |
193 | return (fd == 0 || fd == 1 || fd == 2); | |
194 | } | |
195 | ||
47a46cf1 CB |
196 | int lxc_check_inherited(struct lxc_conf *conf, bool closeall, |
197 | int *fds_to_ignore, size_t len_fds) | |
80090207 | 198 | { |
74f96976 | 199 | struct dirent *direntp; |
80090207 | 200 | int fd, fddir; |
47a46cf1 | 201 | size_t i; |
80090207 | 202 | DIR *dir; |
80090207 | 203 | |
d2cf4c37 SH |
204 | if (conf && conf->close_all_fds) |
205 | closeall = true; | |
206 | ||
b119f362 | 207 | restart: |
80090207 CLG |
208 | dir = opendir("/proc/self/fd"); |
209 | if (!dir) { | |
13277ec4 | 210 | WARN("Failed to open directory: %s.", strerror(errno)); |
80090207 CLG |
211 | return -1; |
212 | } | |
213 | ||
214 | fddir = dirfd(dir); | |
215 | ||
74f96976 | 216 | while ((direntp = readdir(dir))) { |
80090207 CLG |
217 | if (!direntp) |
218 | break; | |
219 | ||
220 | if (!strcmp(direntp->d_name, ".")) | |
221 | continue; | |
222 | ||
223 | if (!strcmp(direntp->d_name, "..")) | |
224 | continue; | |
225 | ||
d4cff0d2 CB |
226 | if (lxc_safe_int(direntp->d_name, &fd) < 0) { |
227 | INFO("Could not parse file descriptor for: %s", direntp->d_name); | |
228 | continue; | |
229 | } | |
80090207 | 230 | |
47a46cf1 CB |
231 | for (i = 0; i < len_fds; i++) |
232 | if (fds_to_ignore[i] == fd) | |
233 | break; | |
234 | ||
235 | if (fd == fddir || fd == lxc_log_fd || | |
236 | (i < len_fds && fd == fds_to_ignore[i])) | |
80090207 CLG |
237 | continue; |
238 | ||
858377e4 SH |
239 | if (current_config && fd == current_config->logfd) |
240 | continue; | |
241 | ||
80090207 CLG |
242 | if (match_fd(fd)) |
243 | continue; | |
80090207 | 244 | |
d2cf4c37 | 245 | if (closeall) { |
b119f362 SH |
246 | close(fd); |
247 | closedir(dir); | |
408da065 | 248 | INFO("Closed inherited fd: %d.", fd); |
b119f362 SH |
249 | goto restart; |
250 | } | |
408da065 | 251 | WARN("Inherited fd: %d.", fd); |
80090207 CLG |
252 | } |
253 | ||
408da065 | 254 | /* Only enable syslog at this point to avoid the above logging function |
64c57ea1 BD |
255 | * to open a new fd and make the check_inherited function enter an |
256 | * infinite loop. | |
257 | */ | |
258 | lxc_log_enable_syslog(); | |
259 | ||
92c7f629 GK |
260 | closedir(dir); /* cannot fail */ |
261 | return 0; | |
80090207 CLG |
262 | } |
263 | ||
83ee7875 | 264 | static int setup_signal_fd(sigset_t *oldmask) |
b0a33c1e | 265 | { |
266 | sigset_t mask; | |
267 | int fd; | |
268 | ||
408da065 | 269 | /* Block everything except serious error signals. */ |
f3304a29 FW |
270 | if (sigfillset(&mask) || |
271 | sigdelset(&mask, SIGILL) || | |
272 | sigdelset(&mask, SIGSEGV) || | |
273 | sigdelset(&mask, SIGBUS) || | |
b5159817 | 274 | sigdelset(&mask, SIGWINCH) || |
f3304a29 | 275 | sigprocmask(SIG_BLOCK, &mask, oldmask)) { |
408da065 | 276 | SYSERROR("Failed to set signal mask."); |
b0a33c1e | 277 | return -1; |
278 | } | |
279 | ||
280 | fd = signalfd(-1, &mask, 0); | |
281 | if (fd < 0) { | |
408da065 | 282 | SYSERROR("Failed to create signal file descriptor."); |
b0a33c1e | 283 | return -1; |
284 | } | |
285 | ||
286 | if (fcntl(fd, F_SETFD, FD_CLOEXEC)) { | |
408da065 | 287 | SYSERROR("Failed to set FD_CLOEXEC on the signal file descriptor: %d.", fd); |
b0a33c1e | 288 | close(fd); |
289 | return -1; | |
290 | } | |
291 | ||
408da065 | 292 | DEBUG("Set SIGCHLD handler with file descriptor: %d.", fd); |
1ac470c0 | 293 | |
b0a33c1e | 294 | return fd; |
295 | } | |
296 | ||
84c92abd | 297 | static int signal_handler(int fd, uint32_t events, void *data, |
f3787121 | 298 | struct lxc_epoll_descr *descr) |
b0a33c1e | 299 | { |
15cd25fd | 300 | struct signalfd_siginfo siginfo; |
80507ee8 | 301 | siginfo_t info; |
15cd25fd | 302 | int ret; |
82d89dce | 303 | pid_t *pid = data; |
80507ee8 | 304 | bool init_died = false; |
15cd25fd DL |
305 | |
306 | ret = read(fd, &siginfo, sizeof(siginfo)); | |
307 | if (ret < 0) { | |
408da065 | 308 | ERROR("Failed to read signal info from signal file descriptor: %d.", fd); |
15cd25fd DL |
309 | return -1; |
310 | } | |
311 | ||
312 | if (ret != sizeof(siginfo)) { | |
408da065 | 313 | ERROR("Unexpected size for siginfo struct."); |
15cd25fd DL |
314 | return -1; |
315 | } | |
316 | ||
408da065 | 317 | /* Check whether init is running. */ |
80507ee8 SH |
318 | info.si_pid = 0; |
319 | ret = waitid(P_PID, *pid, &info, WEXITED | WNOWAIT | WNOHANG); | |
408da065 | 320 | if (ret == 0 && info.si_pid == *pid) |
80507ee8 | 321 | init_died = true; |
80507ee8 | 322 | |
f3304a29 FW |
323 | if (siginfo.ssi_signo != SIGCHLD) { |
324 | kill(*pid, siginfo.ssi_signo); | |
408da065 | 325 | INFO("Forwarded signal %d to pid %d.", siginfo.ssi_signo, *pid); |
80507ee8 | 326 | return init_died ? 1 : 0; |
f3304a29 FW |
327 | } |
328 | ||
408da065 CB |
329 | if (siginfo.ssi_code == CLD_STOPPED) { |
330 | INFO("Container init process was stopped."); | |
331 | return init_died ? 1 : 0; | |
332 | } else if (siginfo.ssi_code == CLD_CONTINUED) { | |
333 | INFO("Container init process was continued."); | |
80507ee8 | 334 | return init_died ? 1 : 0; |
15cd25fd | 335 | } |
1ac470c0 | 336 | |
408da065 CB |
337 | /* More robustness, protect ourself from a SIGCHLD sent |
338 | * by a process different from the container init. | |
82d89dce DL |
339 | */ |
340 | if (siginfo.ssi_pid != *pid) { | |
af4c0f05 | 341 | NOTICE("Received SIGCHLD from pid %d instead of container init %d.", siginfo.ssi_pid, *pid); |
80507ee8 | 342 | return init_died ? 1 : 0; |
82d89dce DL |
343 | } |
344 | ||
408da065 | 345 | DEBUG("Container init process %d exited.", *pid); |
b0a33c1e | 346 | return 1; |
347 | } | |
348 | ||
5e5576a4 CB |
349 | static int lxc_serve_state_clients(const char *name, |
350 | struct lxc_handler *handler, | |
351 | lxc_state_t state) | |
66aeffc7 | 352 | { |
dbc9832d CB |
353 | ssize_t ret; |
354 | struct lxc_list *cur, *next; | |
355 | struct state_client *client; | |
356 | struct lxc_msg msg = {.type = lxc_msg_state, .value = state}; | |
357 | ||
358 | process_lock(); | |
5e5576a4 | 359 | |
dbc9832d | 360 | /* Only set state under process lock held so that we don't cause |
54446942 | 361 | * lxc_cmd_add_state_client() to miss a state. |
dbc9832d | 362 | */ |
66aeffc7 | 363 | handler->state = state; |
dbc9832d CB |
364 | TRACE("set container state to %s", lxc_state2str(state)); |
365 | ||
366 | if (lxc_list_empty(&handler->state_clients)) { | |
367 | TRACE("no state clients registered"); | |
368 | process_unlock(); | |
65ee1673 | 369 | lxc_monitor_send_state(name, state, handler->lxcpath); |
dbc9832d CB |
370 | return 0; |
371 | } | |
372 | ||
373 | strncpy(msg.name, name, sizeof(msg.name)); | |
374 | msg.name[sizeof(msg.name) - 1] = 0; | |
375 | ||
376 | lxc_list_for_each_safe(cur, &handler->state_clients, next) { | |
377 | client = cur->elem; | |
378 | ||
379 | if (!client->states[state]) { | |
380 | TRACE("state %s not registered for state client %d", | |
381 | lxc_state2str(state), client->clientfd); | |
382 | continue; | |
383 | } | |
384 | ||
385 | TRACE("sending state %s to state client %d", | |
386 | lxc_state2str(state), client->clientfd); | |
387 | ||
388 | again: | |
389 | ret = send(client->clientfd, &msg, sizeof(msg), 0); | |
ee8377bd CB |
390 | if (ret <= 0) { |
391 | if (errno == EINTR) { | |
392 | TRACE("Caught EINTR; retrying"); | |
dbc9832d | 393 | goto again; |
ee8377bd | 394 | } |
dbc9832d CB |
395 | |
396 | ERROR("failed to send message to client"); | |
397 | } | |
398 | ||
399 | /* kick client from list */ | |
400 | close(client->clientfd); | |
401 | lxc_list_del(cur); | |
402 | free(cur->elem); | |
403 | free(cur); | |
404 | } | |
405 | process_unlock(); | |
406 | ||
5e5576a4 CB |
407 | return 0; |
408 | } | |
409 | ||
410 | static int lxc_serve_state_socket_pair(const char *name, | |
411 | struct lxc_handler *handler, | |
412 | lxc_state_t state) | |
413 | { | |
414 | ssize_t ret; | |
415 | ||
416 | if (!handler->backgrounded || | |
417 | handler->state_socket_pair[1] < 0 || | |
418 | state == STARTING) | |
419 | return 0; | |
420 | ||
421 | /* Close read end of the socket pair. */ | |
422 | close(handler->state_socket_pair[0]); | |
423 | handler->state_socket_pair[0] = -1; | |
424 | ||
ee8377bd | 425 | again: |
5e5576a4 CB |
426 | ret = lxc_abstract_unix_send_credential(handler->state_socket_pair[1], |
427 | &(int){state}, sizeof(int)); | |
ee8377bd CB |
428 | if (ret != sizeof(int)) { |
429 | if (errno == EINTR) | |
430 | goto again; | |
5e5576a4 CB |
431 | SYSERROR("Failed to send state to %d", |
432 | handler->state_socket_pair[1]); | |
ee8377bd CB |
433 | return -1; |
434 | } | |
5e5576a4 CB |
435 | |
436 | TRACE("Sent container state \"%s\" to %d", lxc_state2str(state), | |
437 | handler->state_socket_pair[1]); | |
438 | ||
439 | /* Close write end of the socket pair. */ | |
440 | close(handler->state_socket_pair[1]); | |
441 | handler->state_socket_pair[1] = -1; | |
442 | ||
443 | return 0; | |
444 | } | |
445 | ||
446 | int lxc_set_state(const char *name, struct lxc_handler *handler, | |
447 | lxc_state_t state) | |
448 | { | |
449 | int ret; | |
450 | ||
451 | ret = lxc_serve_state_socket_pair(name, handler, state); | |
452 | if (ret < 0) { | |
453 | ERROR("Failed to synchronize via anonymous pair of unix sockets"); | |
454 | return -1; | |
455 | } | |
456 | ||
457 | ret = lxc_serve_state_clients(name, handler, state); | |
458 | if (ret < 0) | |
459 | return -1; | |
460 | ||
dbc9832d CB |
461 | /* This function will try to connect to the legacy lxc-monitord state |
462 | * server and only exists for backwards compatibility. | |
463 | */ | |
9123e471 | 464 | lxc_monitor_send_state(name, state, handler->lxcpath); |
dbc9832d | 465 | |
66aeffc7 DL |
466 | return 0; |
467 | } | |
468 | ||
735f2c6e | 469 | int lxc_poll(const char *name, struct lxc_handler *handler) |
b0a33c1e | 470 | { |
ca5f7926 DL |
471 | int sigfd = handler->sigfd; |
472 | int pid = handler->pid; | |
b0a33c1e | 473 | struct lxc_epoll_descr descr; |
474 | ||
a9e61274 | 475 | if (lxc_mainloop_open(&descr)) { |
408da065 | 476 | ERROR("Failed to create LXC mainloop."); |
50c8bf05 | 477 | goto out_sigfd; |
b0a33c1e | 478 | } |
479 | ||
83ee7875 | 480 | if (lxc_mainloop_add_handler(&descr, sigfd, signal_handler, &pid)) { |
408da065 | 481 | ERROR("Failed to add signal handler with file descriptor %d to LXC mainloop.", sigfd); |
b0a33c1e | 482 | goto out_mainloop_open; |
483 | } | |
484 | ||
da41561c | 485 | if (lxc_console_mainloop_add(&descr, handler->conf)) { |
408da065 | 486 | ERROR("Failed to add console handler to LXC mainloop."); |
63376d7d DL |
487 | goto out_mainloop_open; |
488 | } | |
489 | ||
ef6e34ee | 490 | if (lxc_cmd_mainloop_add(name, &descr, handler)) { |
408da065 | 491 | ERROR("Failed to add command handler to LXC mainloop."); |
96fa1ff0 | 492 | goto out_mainloop_open; |
563f2f2c DL |
493 | } |
494 | ||
3f903c04 | 495 | TRACE("lxc mainloop is ready"); |
b0a33c1e | 496 | |
e51d4895 | 497 | return lxc_mainloop(&descr, -1); |
b0a33c1e | 498 | |
499 | out_mainloop_open: | |
500 | lxc_mainloop_close(&descr); | |
408da065 | 501 | |
b0a33c1e | 502 | out_sigfd: |
503 | close(sigfd); | |
408da065 | 504 | |
c3e13372 | 505 | return -1; |
b0a33c1e | 506 | } |
507 | ||
f2e07cb6 CB |
508 | void lxc_free_handler(struct lxc_handler *handler) |
509 | { | |
510 | if (handler->conf && handler->conf->maincmd_fd) | |
511 | close(handler->conf->maincmd_fd); | |
512 | ||
5e5576a4 CB |
513 | if (handler->state_socket_pair[0] >= 0) |
514 | close(handler->state_socket_pair[0]); | |
515 | ||
516 | if (handler->state_socket_pair[1] >= 0) | |
517 | close(handler->state_socket_pair[1]); | |
518 | ||
f2e07cb6 CB |
519 | if (handler->name) |
520 | free(handler->name); | |
521 | ||
522 | handler->conf = NULL; | |
523 | free(handler); | |
524 | } | |
525 | ||
aa460476 | 526 | struct lxc_handler *lxc_init_handler(const char *name, struct lxc_conf *conf, |
5e5576a4 | 527 | const char *lxcpath, bool daemonize) |
59eb99ba | 528 | { |
5e5576a4 | 529 | int i, ret; |
3a0f472d DL |
530 | struct lxc_handler *handler; |
531 | ||
532 | handler = malloc(sizeof(*handler)); | |
aa460476 CB |
533 | if (!handler) { |
534 | ERROR("failed to allocate memory"); | |
3a0f472d | 535 | return NULL; |
aa460476 | 536 | } |
59eb99ba DL |
537 | |
538 | memset(handler, 0, sizeof(*handler)); | |
539 | ||
790255cf CB |
540 | /* Note that am_unpriv() checks the effective uid. We probably don't |
541 | * care if we are real root only if we are running as root so this | |
542 | * should be fine. | |
543 | */ | |
d0fbc7ba | 544 | handler->am_root = !am_unpriv(); |
c6012571 | 545 | handler->data_sock[0] = handler->data_sock[1] = -1; |
fae349da | 546 | handler->conf = conf; |
9123e471 | 547 | handler->lxcpath = lxcpath; |
5c068da9 | 548 | handler->pinfd = -1; |
5e5576a4 | 549 | handler->state_socket_pair[0] = handler->state_socket_pair[1] = -1; |
dbc9832d | 550 | lxc_list_init(&handler->state_clients); |
fae349da | 551 | |
b6b2b194 WB |
552 | for (i = 0; i < LXC_NS_MAX; i++) |
553 | handler->nsfd[i] = -1; | |
554 | ||
3bdf52d7 DL |
555 | handler->name = strdup(name); |
556 | if (!handler->name) { | |
aa460476 | 557 | ERROR("failed to allocate memory"); |
f2e07cb6 | 558 | goto on_error; |
3bdf52d7 DL |
559 | } |
560 | ||
5e5576a4 CB |
561 | if (daemonize && !handler->conf->reboot) { |
562 | /* Create socketpair() to synchronize on daemonized startup. | |
563 | * When the container reboots we don't need to synchronize again | |
564 | * currently so don't open another socketpair(). | |
565 | */ | |
566 | ret = socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0, | |
567 | handler->state_socket_pair); | |
568 | if (ret < 0) { | |
569 | ERROR("Failed to create anonymous pair of unix sockets"); | |
570 | goto on_error; | |
571 | } | |
572 | TRACE("Created anonymous pair {%d,%d} of unix sockets", | |
573 | handler->state_socket_pair[0], | |
574 | handler->state_socket_pair[1]); | |
575 | } | |
576 | ||
aa460476 CB |
577 | if (lxc_cmd_init(name, handler, lxcpath)) { |
578 | ERROR("failed to set up command socket"); | |
f2e07cb6 | 579 | goto on_error; |
aa460476 CB |
580 | } |
581 | ||
582 | TRACE("unix domain socket %d for command server is ready", | |
583 | handler->conf->maincmd_fd); | |
584 | ||
585 | return handler; | |
586 | ||
f2e07cb6 CB |
587 | on_error: |
588 | lxc_free_handler(handler); | |
aa460476 CB |
589 | |
590 | return NULL; | |
591 | } | |
592 | ||
593 | int lxc_init(const char *name, struct lxc_handler *handler) | |
594 | { | |
595 | struct lxc_conf *conf = handler->conf; | |
596 | ||
597 | lsm_init(); | |
598 | TRACE("initialized LSM"); | |
d2e30e99 | 599 | |
8f2c3a70 | 600 | if (lxc_read_seccomp_config(conf) != 0) { |
408da065 | 601 | ERROR("Failed loading seccomp policy."); |
d2e30e99 | 602 | goto out_close_maincmd_fd; |
8f2c3a70 | 603 | } |
dbc9832d | 604 | TRACE("read seccomp policy"); |
8f2c3a70 | 605 | |
408da065 | 606 | /* Begin by setting the state to STARTING. */ |
25c2aca5 | 607 | if (lxc_set_state(name, handler, STARTING)) { |
408da065 | 608 | ERROR("Failed to set state for container \"%s\" to \"%s\".", name, lxc_state2str(STARTING)); |
051151de | 609 | goto out_close_maincmd_fd; |
0ad19a3f | 610 | } |
dbc9832d | 611 | TRACE("set container state to \"STARTING\""); |
0ad19a3f | 612 | |
408da065 CB |
613 | /* Start of environment variable setup for hooks. */ |
614 | if (name && setenv("LXC_NAME", name, 1)) | |
615 | SYSERROR("Failed to set environment variable: LXC_NAME=%s.", name); | |
616 | ||
617 | if (conf->rcfile && setenv("LXC_CONFIG_FILE", conf->rcfile, 1)) | |
618 | SYSERROR("Failed to set environment variable: LXC_CONFIG_FILE=%s.", conf->rcfile); | |
619 | ||
620 | if (conf->rootfs.mount && setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1)) | |
621 | SYSERROR("Failed to set environment variable: LXC_ROOTFS_MOUNT=%s.", conf->rootfs.mount); | |
622 | ||
623 | if (conf->rootfs.path && setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1)) | |
624 | SYSERROR("Failed to set environment variable: LXC_ROOTFS_PATH=%s.", conf->rootfs.path); | |
625 | ||
626 | if (conf->console.path && setenv("LXC_CONSOLE", conf->console.path, 1)) | |
627 | SYSERROR("Failed to set environment variable: LXC_CONSOLE=%s.", conf->console.path); | |
628 | ||
629 | if (conf->console.log_path && setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1)) | |
630 | SYSERROR("Failed to set environment variable: LXC_CONSOLE_LOGPATH=%s.", conf->console.log_path); | |
631 | ||
632 | if (setenv("LXC_CGNS_AWARE", "1", 1)) | |
633 | SYSERROR("Failed to set environment variable LXC_CGNS_AWARE=1."); | |
634 | /* End of environment variable setup for hooks. */ | |
f7bee6c6 | 635 | |
dbc9832d CB |
636 | TRACE("set environment variables"); |
637 | ||
283678ed | 638 | if (run_lxc_hooks(name, "pre-start", conf, handler->lxcpath, NULL)) { |
408da065 | 639 | ERROR("Failed to run lxc.hook.pre-start for container \"%s\".", name); |
773fb9ca SH |
640 | goto out_aborting; |
641 | } | |
dbc9832d | 642 | TRACE("ran pre-start hooks"); |
26ddeedd | 643 | |
408da065 CB |
644 | /* The signal fd has to be created before forking otherwise if the child |
645 | * process exits before we setup the signal fd, the event will be lost | |
646 | * and the command will be stuck. | |
647 | */ | |
83ee7875 | 648 | handler->sigfd = setup_signal_fd(&handler->oldmask); |
59eb99ba | 649 | if (handler->sigfd < 0) { |
408da065 | 650 | ERROR("Failed to setup SIGCHLD fd handler."); |
b5159817 DE |
651 | goto out_delete_tty; |
652 | } | |
dbc9832d | 653 | TRACE("set up signal fd"); |
b5159817 | 654 | |
408da065 | 655 | /* Do this after setting up signals since it might unblock SIGWINCH. */ |
b5159817 | 656 | if (lxc_console_create(conf)) { |
408da065 | 657 | ERROR("Failed to create console for container \"%s\".", name); |
b5159817 | 658 | goto out_restore_sigmask; |
b0a33c1e | 659 | } |
dbc9832d | 660 | TRACE("created console"); |
b0a33c1e | 661 | |
54117de5 | 662 | if (lxc_ttys_shift_ids(conf) < 0) { |
408da065 | 663 | ERROR("Failed to shift tty into container."); |
c4d10a05 SH |
664 | goto out_restore_sigmask; |
665 | } | |
dbc9832d | 666 | TRACE("shifted tty ids"); |
c4d10a05 | 667 | |
dbc9832d | 668 | INFO("container \"%s\" is initialized", name); |
aa460476 | 669 | return 0; |
59eb99ba | 670 | |
b5159817 DE |
671 | out_restore_sigmask: |
672 | sigprocmask(SIG_SETMASK, &handler->oldmask, NULL); | |
59eb99ba | 673 | out_delete_tty: |
fae349da | 674 | lxc_delete_tty(&conf->tty_info); |
59eb99ba | 675 | out_aborting: |
25c2aca5 | 676 | lxc_set_state(name, handler, ABORTING); |
d2e30e99 DE |
677 | out_close_maincmd_fd: |
678 | close(conf->maincmd_fd); | |
679 | conf->maincmd_fd = -1; | |
aa460476 | 680 | return -1; |
59eb99ba DL |
681 | } |
682 | ||
3b72c4a0 | 683 | void lxc_fini(const char *name, struct lxc_handler *handler) |
59eb99ba | 684 | { |
b3286b62 | 685 | int i, rc; |
dbc9832d | 686 | struct lxc_list *cur, *next; |
b3286b62 | 687 | pid_t self = getpid(); |
dbc9832d | 688 | char *namespaces[LXC_NS_MAX + 1]; |
b3286b62 | 689 | size_t namespace_count = 0; |
b6b2b194 | 690 | |
408da065 CB |
691 | /* The STOPPING state is there for future cleanup code which can take |
692 | * awhile. | |
59eb99ba | 693 | */ |
25c2aca5 | 694 | lxc_set_state(name, handler, STOPPING); |
b6b2b194 | 695 | |
b3286b62 WB |
696 | for (i = 0; i < LXC_NS_MAX; i++) { |
697 | if (handler->nsfd[i] != -1) { | |
698 | rc = asprintf(&namespaces[namespace_count], "%s:/proc/%d/fd/%d", | |
699 | ns_info[i].proc_name, self, handler->nsfd[i]); | |
700 | if (rc == -1) { | |
408da065 | 701 | SYSERROR("Failed to allocate memory."); |
b3286b62 WB |
702 | break; |
703 | } | |
704 | ++namespace_count; | |
705 | } | |
706 | } | |
707 | namespaces[namespace_count] = NULL; | |
c154af98 | 708 | |
408da065 CB |
709 | if (handler->conf->reboot && setenv("LXC_TARGET", "reboot", 1)) |
710 | SYSERROR("Failed to set environment variable: LXC_TARGET=reboot."); | |
711 | ||
712 | if (!handler->conf->reboot && setenv("LXC_TARGET", "stop", 1)) | |
713 | SYSERROR("Failed to set environment variable: LXC_TARGET=stop."); | |
c154af98 | 714 | |
b3286b62 | 715 | if (run_lxc_hooks(name, "stop", handler->conf, handler->lxcpath, namespaces)) |
408da065 | 716 | ERROR("Failed to run lxc.hook.stop for container \"%s\".", name); |
c154af98 | 717 | |
b3286b62 WB |
718 | while (namespace_count--) |
719 | free(namespaces[namespace_count]); | |
b6b2b194 WB |
720 | for (i = 0; i < LXC_NS_MAX; i++) { |
721 | if (handler->nsfd[i] != -1) { | |
722 | close(handler->nsfd[i]); | |
723 | handler->nsfd[i] = -1; | |
724 | } | |
725 | } | |
738d0deb CB |
726 | |
727 | if (handler->netnsfd >= 0) { | |
728 | close(handler->netnsfd); | |
729 | handler->netnsfd = -1; | |
730 | } | |
731 | ||
4288b79f CB |
732 | cgroup_destroy(handler); |
733 | ||
25c2aca5 | 734 | lxc_set_state(name, handler, STOPPED); |
59eb99ba | 735 | |
4288b79f CB |
736 | /* close command socket */ |
737 | close(handler->conf->maincmd_fd); | |
738 | handler->conf->maincmd_fd = -1; | |
739 | ||
f3787121 | 740 | if (run_lxc_hooks(name, "post-stop", handler->conf, handler->lxcpath, NULL)) { |
408da065 | 741 | ERROR("Failed to run lxc.hook.post-stop for container \"%s\".", name); |
f3787121 CB |
742 | if (handler->conf->reboot) { |
743 | WARN("Container will be stopped instead of rebooted."); | |
744 | handler->conf->reboot = 0; | |
408da065 CB |
745 | if (setenv("LXC_TARGET", "stop", 1)) |
746 | WARN("Failed to set environment variable: LXC_TARGET=stop."); | |
f3787121 CB |
747 | } |
748 | } | |
26ddeedd | 749 | |
408da065 | 750 | /* Reset mask set by setup_signal_fd. */ |
8f64a3f6 | 751 | if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) |
408da065 | 752 | WARN("Failed to restore signal mask."); |
8f64a3f6 | 753 | |
b5159817 | 754 | lxc_console_delete(&handler->conf->console); |
b2431939 | 755 | lxc_delete_tty(&handler->conf->tty_info); |
dbc9832d | 756 | |
dbc9832d CB |
757 | /* The command socket is now closed, no more state clients can register |
758 | * themselves from now on. So free the list of state clients. | |
759 | */ | |
760 | lxc_list_for_each_safe(cur, &handler->state_clients, next) { | |
761 | struct state_client *client = cur->elem; | |
762 | /* close state client socket */ | |
763 | close(client->clientfd); | |
764 | lxc_list_del(cur); | |
765 | free(cur->elem); | |
766 | free(cur); | |
767 | } | |
768 | ||
c6012571 CB |
769 | if (handler->data_sock[0] != -1) { |
770 | close(handler->data_sock[0]); | |
771 | close(handler->data_sock[1]); | |
e8bd4e43 | 772 | } |
2c5f2ede CB |
773 | |
774 | if (handler->conf->ephemeral == 1 && handler->conf->reboot != 1) | |
28272964 | 775 | lxc_destroy_container_on_signal(handler, name); |
2c5f2ede | 776 | |
4288b79f | 777 | free(handler->name); |
b2431939 | 778 | free(handler); |
59eb99ba DL |
779 | } |
780 | ||
735f2c6e | 781 | void lxc_abort(const char *name, struct lxc_handler *handler) |
59eb99ba | 782 | { |
73e608b2 SH |
783 | int ret, status; |
784 | ||
25c2aca5 | 785 | lxc_set_state(name, handler, ABORTING); |
7d9fb3e9 DL |
786 | if (handler->pid > 0) |
787 | kill(handler->pid, SIGKILL); | |
408da065 CB |
788 | while ((ret = waitpid(-1, &status, 0)) > 0) { |
789 | ; | |
790 | } | |
59eb99ba DL |
791 | } |
792 | ||
ffe1e01a | 793 | static int do_start(void *data) |
50e98013 | 794 | { |
7c661726 | 795 | struct lxc_list *iterator; |
23c53af9 | 796 | struct lxc_handler *handler = data; |
7a55c157 TA |
797 | int devnull_fd = -1, ret; |
798 | char path[PATH_MAX]; | |
50e98013 DL |
799 | |
800 | if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) { | |
408da065 | 801 | SYSERROR("Failed to set signal mask."); |
9d7f9e52 | 802 | return -1; |
50e98013 DL |
803 | } |
804 | ||
408da065 CB |
805 | /* This prctl must be before the synchro, so if the parent dies before |
806 | * we set the parent death signal, we will detect its death with the | |
807 | * synchro right after, otherwise we have a window where the parent can | |
808 | * exit before we set the pdeath signal leading to a unsupervized | |
809 | * container. | |
743ecd2e DL |
810 | */ |
811 | if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) { | |
408da065 | 812 | SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL."); |
743ecd2e DL |
813 | return -1; |
814 | } | |
815 | ||
3c22086f | 816 | lxc_sync_fini_parent(handler); |
50e98013 | 817 | |
408da065 | 818 | /* Don't leak the pinfd to the container. */ |
025ed0f3 | 819 | if (handler->pinfd >= 0) { |
0d03360a | 820 | close(handler->pinfd); |
025ed0f3 | 821 | } |
2b0e17e4 | 822 | |
5b1e83cb SH |
823 | if (lxc_sync_wait_parent(handler, LXC_SYNC_STARTUP)) |
824 | return -1; | |
825 | ||
408da065 CB |
826 | /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See |
827 | * https://github.com/lxc/lxd/issues/1978. | |
828 | */ | |
5b1e83cb | 829 | if ((handler->clone_flags & (CLONE_NEWNET | CLONE_NEWUSER)) == |
408da065 | 830 | (CLONE_NEWNET | CLONE_NEWUSER)) { |
5b1e83cb SH |
831 | ret = unshare(CLONE_NEWNET); |
832 | if (ret < 0) { | |
408da065 | 833 | SYSERROR("Failed to unshare CLONE_NEWNET."); |
5b1e83cb SH |
834 | goto out_warn_father; |
835 | } | |
408da065 | 836 | INFO("Unshared CLONE_NEWNET."); |
5b1e83cb SH |
837 | } |
838 | ||
408da065 CB |
839 | /* Tell the parent task it can begin to configure the container and wait |
840 | * for it to finish. | |
3c22086f CLG |
841 | */ |
842 | if (lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE)) | |
9d7f9e52 | 843 | return -1; |
50e98013 | 844 | |
7ab1ba02 CB |
845 | if (lxc_network_recv_veth_names_from_parent(handler) < 0) { |
846 | ERROR("Failed to receive veth names from parent"); | |
658979c5 | 847 | goto out_warn_father; |
7ab1ba02 | 848 | } |
658979c5 | 849 | |
408da065 | 850 | /* If we are in a new user namespace, become root there to have |
d08f8d2f | 851 | * privilege over our namespace. |
f6d3e3e4 SH |
852 | */ |
853 | if (!lxc_list_empty(&handler->conf->id_map)) { | |
d08f8d2f | 854 | if (lxc_switch_uid_gid(0, 0) < 0) |
f6d3e3e4 | 855 | goto out_warn_father; |
d08f8d2f CB |
856 | |
857 | /* Drop groups only after we switched to a valid gid in the new | |
858 | * user namespace. | |
859 | */ | |
860 | if (lxc_setgroups(0, NULL) < 0) | |
c476bdce | 861 | goto out_warn_father; |
f6d3e3e4 SH |
862 | } |
863 | ||
99b71824 | 864 | if (access(handler->lxcpath, X_OK)) { |
c8154066 SH |
865 | print_top_failing_dir(handler->lxcpath); |
866 | goto out_warn_father; | |
867 | } | |
868 | ||
a91cde21 | 869 | ret = snprintf(path, sizeof(path), "%s/dev/null", handler->conf->rootfs.mount); |
408da065 | 870 | if (ret < 0 || ret >= sizeof(path)) |
7a55c157 | 871 | goto out_warn_father; |
7a55c157 TA |
872 | |
873 | /* In order to checkpoint restore, we need to have everything in the | |
874 | * same mount namespace. However, some containers may not have a | |
875 | * reasonable /dev (in particular, they may not have /dev/null), so we | |
876 | * can't set init's std fds to /dev/null by opening it from inside the | |
877 | * container. | |
878 | * | |
879 | * If that's the case, fall back to using the host's /dev/null. This | |
880 | * means that migration won't work, but at least we won't spew output | |
881 | * where it isn't wanted. | |
882 | */ | |
883 | if (handler->backgrounded && !handler->conf->autodev && access(path, F_OK) < 0) { | |
c44de748 AM |
884 | devnull_fd = open_devnull(); |
885 | ||
886 | if (devnull_fd < 0) | |
887 | goto out_warn_father; | |
408da065 CB |
888 | WARN("Using /dev/null from the host for container init's " |
889 | "standard file descriptors. Migration will not work."); | |
c44de748 AM |
890 | } |
891 | ||
408da065 | 892 | /* Ask father to setup cgroups and wait for him to finish. */ |
544a48a0 | 893 | if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP)) |
c44de748 | 894 | goto out_error; |
544a48a0 | 895 | |
deefdf8a CB |
896 | /* Unshare cgroup namespace after we have setup our cgroups. If we do it |
897 | * earlier we end up with a wrong view of /proc/self/cgroup. For | |
898 | * example, assume we unshare(CLONE_NEWCGROUP) first, and then create | |
899 | * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then | |
900 | * /proc/self/cgroup would show us: | |
901 | * | |
902 | * 8:cpuset:/lxc/c | |
903 | * | |
904 | * whereas it should actually show | |
905 | * | |
906 | * 8:cpuset:/ | |
907 | */ | |
908 | if (cgns_supported()) { | |
909 | if (unshare(CLONE_NEWCGROUP) < 0) { | |
910 | INFO("Failed to unshare CLONE_NEWCGROUP."); | |
911 | goto out_warn_father; | |
912 | } | |
913 | INFO("Unshared CLONE_NEWCGROUP."); | |
914 | } | |
915 | ||
f4152036 | 916 | /* Setup the container, ip, names, utsname, ... */ |
7729f8e5 CB |
917 | ret = lxc_setup(handler); |
918 | close(handler->data_sock[0]); | |
919 | close(handler->data_sock[1]); | |
920 | if (ret < 0) { | |
f4152036 CB |
921 | ERROR("Failed to setup container \"%s\".", handler->name); |
922 | goto out_warn_father; | |
923 | } | |
924 | ||
408da065 | 925 | /* Set the label to change to when we exec(2) the container's init. */ |
7aff4f43 | 926 | if (lsm_process_label_set(NULL, handler->conf, 1, 1) < 0) |
e075f5d9 | 927 | goto out_warn_father; |
5112cd70 | 928 | |
029cdff5 | 929 | /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it |
408da065 CB |
930 | * before we aren't allowed anymore. |
931 | */ | |
029cdff5 CB |
932 | if (handler->conf->no_new_privs) { |
933 | if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) { | |
934 | SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges."); | |
935 | goto out_warn_father; | |
936 | } | |
937 | DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges."); | |
938 | } | |
939 | ||
0d9acb99 DE |
940 | /* Some init's such as busybox will set sane tty settings on stdin, |
941 | * stdout, stderr which it thinks is the console. We already set them | |
942 | * the way we wanted on the real terminal, and we want init to do its | |
408da065 CB |
943 | * setup on its console ie. the pty allocated in lxc_console_create() so |
944 | * make sure that that pty is stdin,stdout,stderr. | |
0d9acb99 | 945 | */ |
c5b93afb LF |
946 | if (handler->conf->console.slave >= 0) |
947 | if (set_stdfds(handler->conf->console.slave) < 0) { | |
948 | ERROR("Failed to redirect std{in,out,err} to pty file " | |
949 | "descriptor %d", | |
950 | handler->conf->console.slave); | |
951 | goto out_warn_father; | |
952 | } | |
0d9acb99 | 953 | |
408da065 | 954 | /* If we mounted a temporary proc, then unmount it now. */ |
5112cd70 | 955 | tmp_proc_unmount(handler->conf); |
e075f5d9 | 956 | |
8f2c3a70 SH |
957 | if (lxc_seccomp_load(handler->conf) != 0) |
958 | goto out_warn_father; | |
959 | ||
283678ed | 960 | if (run_lxc_hooks(handler->name, "start", handler->conf, handler->lxcpath, NULL)) { |
408da065 | 961 | ERROR("Failed to run lxc.hook.start for container \"%s\".", handler->name); |
773fb9ca SH |
962 | goto out_warn_father; |
963 | } | |
fc25b815 | 964 | |
d08f8d2f CB |
965 | /* The container has been setup. We can now switch to an unprivileged |
966 | * uid/gid. | |
967 | */ | |
968 | if (handler->conf->is_execute) { | |
87bf0db0 CB |
969 | bool have_cap_setgid; |
970 | uid_t new_uid = handler->conf->init_uid; | |
971 | gid_t new_gid = handler->conf->init_gid; | |
d08f8d2f CB |
972 | |
973 | /* If we are in a new user namespace we already dropped all | |
974 | * groups when we switched to root in the new user namespace | |
975 | * further above. Only drop groups if we can, so ensure that we | |
976 | * have necessary privilege. | |
977 | */ | |
e37dda71 | 978 | #if HAVE_LIBCAP |
207c4c71 | 979 | have_cap_setgid = lxc_proc_cap_is_set(CAP_SETGID, CAP_EFFECTIVE); |
df11e022 BN |
980 | #else |
981 | have_cap_setgid = false; | |
982 | #endif | |
87bf0db0 | 983 | if (lxc_list_empty(&handler->conf->id_map) && have_cap_setgid) { |
d08f8d2f CB |
984 | if (lxc_setgroups(0, NULL) < 0) |
985 | goto out_warn_father; | |
986 | } | |
987 | ||
988 | if (lxc_switch_uid_gid(new_uid, new_gid) < 0) | |
989 | goto out_warn_father; | |
990 | } | |
991 | ||
408da065 CB |
992 | /* The clearenv() and putenv() calls have been moved here to allow us to |
993 | * use environment variables passed to the various hooks, such as the | |
994 | * start hook above. Not all of the variables like CONFIG_PATH or ROOTFS | |
995 | * are valid in this context but others are. | |
996 | */ | |
f7bee6c6 | 997 | if (clearenv()) { |
408da065 CB |
998 | SYSERROR("Failed to clear environment."); |
999 | /* Don't error out though. */ | |
f7bee6c6 MW |
1000 | } |
1001 | ||
7c661726 MP |
1002 | lxc_list_for_each(iterator, &handler->conf->environment) { |
1003 | if (putenv((char *)iterator->elem)) { | |
408da065 | 1004 | SYSERROR("Failed to set environment variable: %s.", (char *)iterator->elem); |
7c661726 MP |
1005 | goto out_warn_father; |
1006 | } | |
1007 | } | |
1008 | ||
f7bee6c6 | 1009 | if (putenv("container=lxc")) { |
408da065 | 1010 | SYSERROR("Failed to set environment variable: container=lxc."); |
c4ea60df | 1011 | goto out_warn_father; |
f7bee6c6 MW |
1012 | } |
1013 | ||
393903d1 SH |
1014 | if (handler->conf->pty_names) { |
1015 | if (putenv(handler->conf->pty_names)) { | |
408da065 | 1016 | SYSERROR("Failed to set environment variable for container ptys."); |
393903d1 SH |
1017 | goto out_warn_father; |
1018 | } | |
1019 | } | |
1020 | ||
773fb9ca | 1021 | close(handler->sigfd); |
26ddeedd | 1022 | |
7a55c157 TA |
1023 | if (devnull_fd < 0) { |
1024 | devnull_fd = open_devnull(); | |
1025 | ||
1026 | if (devnull_fd < 0) | |
1027 | goto out_warn_father; | |
1028 | } | |
1029 | ||
c5b93afb LF |
1030 | if (handler->conf->console.slave < 0 && handler->backgrounded) |
1031 | if (set_stdfds(devnull_fd) < 0) { | |
1032 | ERROR("Failed to redirect std{in,out,err} to " | |
1033 | "\"/dev/null\""); | |
1034 | goto out_warn_father; | |
1035 | } | |
507cee36 | 1036 | |
c44de748 AM |
1037 | if (devnull_fd >= 0) { |
1038 | close(devnull_fd); | |
1039 | devnull_fd = -1; | |
1040 | } | |
1041 | ||
8c9a7665 TA |
1042 | setsid(); |
1043 | ||
f4152036 CB |
1044 | if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP_LIMITS)) |
1045 | goto out_warn_father; | |
1046 | ||
408da065 CB |
1047 | /* After this call, we are in error because this ops should not return |
1048 | * as it execs. | |
1049 | */ | |
c4ea60df | 1050 | handler->ops->start(handler, handler->data); |
50e98013 DL |
1051 | |
1052 | out_warn_father: | |
408da065 CB |
1053 | /* We want the parent to know something went wrong, so we return a |
1054 | * special error code. | |
1055 | */ | |
d1ccb562 | 1056 | lxc_sync_wake_parent(handler, LXC_SYNC_ERROR); |
c44de748 AM |
1057 | |
1058 | out_error: | |
1059 | if (devnull_fd >= 0) | |
1060 | close(devnull_fd); | |
1061 | ||
50e98013 DL |
1062 | return -1; |
1063 | } | |
1064 | ||
ae467c54 | 1065 | static int lxc_recv_ttys_from_child(struct lxc_handler *handler) |
e8bd4e43 | 1066 | { |
ae467c54 | 1067 | int i; |
ae467c54 CB |
1068 | struct lxc_pty_info *pty_info; |
1069 | int ret = -1; | |
c6012571 | 1070 | int sock = handler->data_sock[1]; |
e8bd4e43 | 1071 | struct lxc_conf *conf = handler->conf; |
e8bd4e43 SH |
1072 | struct lxc_tty_info *tty_info = &conf->tty_info; |
1073 | ||
1074 | if (!conf->tty) | |
1075 | return 0; | |
1076 | ||
408da065 CB |
1077 | tty_info->pty_info = malloc(sizeof(*tty_info->pty_info) * conf->tty); |
1078 | if (!tty_info->pty_info) | |
e8bd4e43 | 1079 | return -1; |
e8bd4e43 | 1080 | |
672c1e58 CB |
1081 | for (i = 0; i < conf->tty; i++) { |
1082 | int ttyfds[2]; | |
1083 | ||
1084 | ret = lxc_abstract_unix_recv_fds(sock, ttyfds, 2, NULL, 0); | |
1085 | if (ret < 0) | |
1086 | break; | |
ae467c54 | 1087 | |
672c1e58 | 1088 | pty_info = &tty_info->pty_info[i]; |
e8bd4e43 | 1089 | pty_info->busy = 0; |
672c1e58 CB |
1090 | pty_info->master = ttyfds[0]; |
1091 | pty_info->slave = ttyfds[1]; | |
1092 | TRACE("Received pty with master fd %d and slave fd %d from " | |
ae467c54 | 1093 | "parent", pty_info->master, pty_info->slave); |
e8bd4e43 | 1094 | } |
ae467c54 | 1095 | if (ret < 0) |
672c1e58 | 1096 | ERROR("Failed to receive %d ttys from child: %s", conf->tty, |
ae467c54 CB |
1097 | strerror(errno)); |
1098 | else | |
672c1e58 CB |
1099 | TRACE("Received %d ttys from child", conf->tty); |
1100 | ||
1101 | tty_info->nbtty = conf->tty; | |
ae467c54 CB |
1102 | |
1103 | return ret; | |
e8bd4e43 SH |
1104 | } |
1105 | ||
f813849c TA |
1106 | void resolve_clone_flags(struct lxc_handler *handler) |
1107 | { | |
1108 | handler->clone_flags = CLONE_NEWPID | CLONE_NEWNS; | |
1109 | ||
408da065 | 1110 | if (!lxc_list_empty(&handler->conf->id_map)) |
f813849c | 1111 | handler->clone_flags |= CLONE_NEWUSER; |
f813849c TA |
1112 | |
1113 | if (handler->conf->inherit_ns_fd[LXC_NS_NET] == -1) { | |
408da065 | 1114 | if (!lxc_requests_empty_network(handler)) |
f813849c TA |
1115 | handler->clone_flags |= CLONE_NEWNET; |
1116 | } else { | |
198cbbaa | 1117 | INFO("Inheriting a NET namespace."); |
f813849c TA |
1118 | } |
1119 | ||
408da065 | 1120 | if (handler->conf->inherit_ns_fd[LXC_NS_IPC] == -1) |
f813849c | 1121 | handler->clone_flags |= CLONE_NEWIPC; |
408da065 | 1122 | else |
198cbbaa | 1123 | INFO("Inheriting an IPC namespace."); |
f813849c | 1124 | |
408da065 | 1125 | if (handler->conf->inherit_ns_fd[LXC_NS_UTS] == -1) |
f813849c | 1126 | handler->clone_flags |= CLONE_NEWUTS; |
408da065 | 1127 | else |
198cbbaa | 1128 | INFO("Inheriting a UTS namespace."); |
f813849c TA |
1129 | } |
1130 | ||
480588e6 CB |
1131 | /* lxc_spawn() performs crucial setup tasks and clone()s the new process which |
1132 | * exec()s the requested container binary. | |
1133 | * Note that lxc_spawn() runs in the parent namespaces. Any operations performed | |
1134 | * right here should be double checked if they'd pose a security risk. (For | |
1135 | * example, any {u}mount() operations performed here will be reflected on the | |
1136 | * host!) | |
1137 | */ | |
74a3920a | 1138 | static int lxc_spawn(struct lxc_handler *handler) |
59eb99ba | 1139 | { |
7ab1ba02 | 1140 | int i, flags, ret; |
ffe1e01a | 1141 | const char *name = handler->name; |
57927bf2 | 1142 | bool wants_to_map_ids; |
7ab1ba02 | 1143 | int saved_ns_fd[LXC_NS_MAX]; |
57927bf2 | 1144 | struct lxc_list *id_map; |
d0b915aa CB |
1145 | int failed_before_rename = 0, preserve_mask = 0; |
1146 | bool cgroups_connected = false; | |
9f30a190 | 1147 | |
57927bf2 CB |
1148 | id_map = &handler->conf->id_map; |
1149 | wants_to_map_ids = !lxc_list_empty(id_map); | |
2f2623ec | 1150 | |
9f30a190 | 1151 | for (i = 0; i < LXC_NS_MAX; i++) |
6c544cb3 | 1152 | if (handler->conf->inherit_ns_fd[i] != -1) |
9f30a190 | 1153 | preserve_mask |= ns_info[i].clone_flag; |
50e98013 | 1154 | |
3c22086f | 1155 | if (lxc_sync_init(handler)) |
9d7f9e52 | 1156 | return -1; |
0ad19a3f | 1157 | |
b9f522c5 | 1158 | ret = socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0, |
790255cf | 1159 | handler->data_sock); |
c6012571 | 1160 | if (ret < 0) { |
e8bd4e43 SH |
1161 | lxc_sync_fini(handler); |
1162 | return -1; | |
1163 | } | |
1164 | ||
f813849c | 1165 | resolve_clone_flags(handler); |
9f30a190 | 1166 | |
f813849c | 1167 | if (handler->clone_flags & CLONE_NEWNET) { |
26b797f3 SH |
1168 | if (!lxc_list_empty(&handler->conf->network)) { |
1169 | ||
9f30a190 MM |
1170 | /* Find gateway addresses from the link device, which is |
1171 | * no longer accessible inside the container. Do this | |
1172 | * before creating network interfaces, since goto | |
408da065 CB |
1173 | * out_delete_net does not work before lxc_clone. |
1174 | */ | |
9f30a190 | 1175 | if (lxc_find_gateway_addresses(handler)) { |
408da065 | 1176 | ERROR("Failed to find gateway addresses."); |
9f30a190 MM |
1177 | lxc_sync_fini(handler); |
1178 | return -1; | |
1179 | } | |
1180 | ||
408da065 CB |
1181 | /* That should be done before the clone because we will |
1182 | * fill the netdev index and use them in the child. | |
9f30a190 | 1183 | */ |
811ef482 | 1184 | if (lxc_create_network_priv(handler)) { |
408da065 | 1185 | ERROR("Failed to create the network."); |
9f30a190 MM |
1186 | lxc_sync_fini(handler); |
1187 | return -1; | |
1188 | } | |
19a26f82 | 1189 | } |
82d5ae15 DL |
1190 | } |
1191 | ||
d4ef7c50 | 1192 | if (!cgroup_init(handler)) { |
408da065 | 1193 | ERROR("Failed initializing cgroup support."); |
33ad9f1a CS |
1194 | goto out_delete_net; |
1195 | } | |
1196 | ||
7e4dfe0b SH |
1197 | cgroups_connected = true; |
1198 | ||
d4ef7c50 | 1199 | if (!cgroup_create(handler)) { |
408da065 | 1200 | ERROR("Failed creating cgroups."); |
47d8fb3b CS |
1201 | goto out_delete_net; |
1202 | } | |
1203 | ||
408da065 CB |
1204 | /* If the rootfs is not a blockdev, prevent the container from marking |
1205 | * it readonly. | |
1206 | * If the container is unprivileged then skip rootfs pinning. | |
0c547523 | 1207 | */ |
0ee35059 | 1208 | if (!wants_to_map_ids) { |
5e32a990 ÇO |
1209 | handler->pinfd = pin_rootfs(handler->conf->rootfs.path); |
1210 | if (handler->pinfd == -1) | |
408da065 | 1211 | INFO("Failed to pin the rootfs for container \"%s\".", handler->name); |
5e32a990 | 1212 | } |
0c547523 | 1213 | |
4d8ac866 | 1214 | if (!preserve_ns(saved_ns_fd, preserve_mask, getpid())) |
cd43d2d1 | 1215 | goto out_delete_net; |
4d8ac866 | 1216 | |
cd43d2d1 SH |
1217 | if (attach_ns(handler->conf->inherit_ns_fd) < 0) |
1218 | goto out_delete_net; | |
9f30a190 | 1219 | |
408da065 | 1220 | /* Create a process in a new set of namespaces. */ |
5b1e83cb | 1221 | flags = handler->clone_flags; |
408da065 CB |
1222 | if (handler->clone_flags & CLONE_NEWUSER) { |
1223 | /* If CLONE_NEWUSER and CLONE_NEWNET was requested, we need to | |
1224 | * clone a new user namespace first and only later unshare our | |
1225 | * network namespace to ensure that network devices ownership is | |
1226 | * set up correctly. | |
1227 | */ | |
5b1e83cb | 1228 | flags &= ~CLONE_NEWNET; |
408da065 | 1229 | } |
9fac8fbb | 1230 | handler->pid = lxc_clone(do_start, handler, flags); |
59eb99ba | 1231 | if (handler->pid < 0) { |
408da065 | 1232 | SYSERROR("Failed to clone a new set of namespaces."); |
7fef7a06 | 1233 | goto out_delete_net; |
0ad19a3f | 1234 | } |
1bd8d726 | 1235 | |
9662e444 CB |
1236 | for (i = 0; i < LXC_NS_MAX; i++) |
1237 | if (flags & ns_info[i].clone_flag) | |
1238 | INFO("Cloned %s.", ns_info[i].flag_name); | |
0ad19a3f | 1239 | |
4d8ac866 CB |
1240 | if (!preserve_ns(handler->nsfd, handler->clone_flags | preserve_mask, handler->pid)) |
1241 | INFO("Failed to preserve namespace for lxc.hook.stop."); | |
b6b2b194 | 1242 | |
cd43d2d1 | 1243 | if (attach_ns(saved_ns_fd)) |
408da065 | 1244 | WARN("Failed to restore saved namespaces."); |
9f30a190 | 1245 | |
3c22086f CLG |
1246 | lxc_sync_fini_child(handler); |
1247 | ||
408da065 CB |
1248 | /* Map the container uids. The container became an invalid userid the |
1249 | * moment it was cloned with CLONE_NEWUSER. This call doesn't change | |
1250 | * anything immediately, but allows the container to setuid(0) (0 being | |
1251 | * mapped to something else on the host.) later to become a valid uid | |
1252 | * again. | |
1253 | */ | |
57927bf2 | 1254 | if (wants_to_map_ids && lxc_map_ids(id_map, handler->pid)) { |
408da065 | 1255 | ERROR("Failed to set up id mapping."); |
5b1e83cb SH |
1256 | goto out_delete_net; |
1257 | } | |
1258 | ||
1259 | if (lxc_sync_wake_child(handler, LXC_SYNC_STARTUP)) { | |
99a6af52 | 1260 | failed_before_rename = 1; |
5b1e83cb SH |
1261 | goto out_delete_net; |
1262 | } | |
1263 | ||
1264 | if (lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE)) { | |
1265 | failed_before_rename = 1; | |
1266 | goto out_delete_net; | |
1267 | } | |
0ad19a3f | 1268 | |
d4ef7c50 | 1269 | if (!cgroup_create_legacy(handler)) { |
408da065 | 1270 | ERROR("Failed to setup legacy cgroups for container \"%s\".", name); |
ae5c8b8e | 1271 | goto out_delete_net; |
33ad9f1a | 1272 | } |
9daf6f5d | 1273 | if (!cgroup_setup_limits(handler, false)) { |
408da065 | 1274 | ERROR("Failed to setup cgroup limits for container \"%s\".", name); |
6031a6e5 DE |
1275 | goto out_delete_net; |
1276 | } | |
1277 | ||
d4ef7c50 | 1278 | if (!cgroup_enter(handler)) |
7fef7a06 | 1279 | goto out_delete_net; |
218d4250 | 1280 | |
0996e18a SH |
1281 | if (!cgroup_chown(handler)) |
1282 | goto out_delete_net; | |
1283 | ||
99a6af52 MN |
1284 | if (failed_before_rename) |
1285 | goto out_delete_net; | |
1286 | ||
1bd8d726 CB |
1287 | handler->netnsfd = lxc_preserve_ns(handler->pid, "net"); |
1288 | if (handler->netnsfd < 0) { | |
1289 | ERROR("Failed to preserve network namespace"); | |
1290 | goto out_delete_net; | |
1291 | } | |
1292 | ||
408da065 | 1293 | /* Create the network configuration. */ |
d5088cf2 | 1294 | if (handler->clone_flags & CLONE_NEWNET) { |
74c6e2b0 CB |
1295 | if (lxc_network_move_created_netdev_priv(handler->lxcpath, |
1296 | handler->name, | |
1297 | &handler->conf->network, | |
1298 | handler->pid)) { | |
408da065 | 1299 | ERROR("Failed to create the configured network."); |
7fef7a06 | 1300 | goto out_delete_net; |
82d5ae15 | 1301 | } |
74c6e2b0 CB |
1302 | |
1303 | if (lxc_create_network_unpriv(handler->lxcpath, handler->name, | |
1304 | &handler->conf->network, | |
1305 | handler->pid)) { | |
1306 | ERROR("Failed to create the configured network."); | |
1307 | goto out_delete_net; | |
1308 | } | |
0ad19a3f | 1309 | } |
1310 | ||
7ab1ba02 CB |
1311 | if (lxc_network_send_veth_names_to_child(handler) < 0) { |
1312 | ERROR("Failed to send veth names to child"); | |
1313 | goto out_delete_net; | |
658979c5 SH |
1314 | } |
1315 | ||
408da065 CB |
1316 | /* Tell the child to continue its initialization. We'll get |
1317 | * LXC_SYNC_CGROUP when it is ready for us to setup cgroups. | |
3c22086f CLG |
1318 | */ |
1319 | if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE)) | |
544a48a0 SH |
1320 | goto out_delete_net; |
1321 | ||
c6d09e15 WB |
1322 | if (!lxc_list_empty(&handler->conf->limits) && setup_resource_limits(&handler->conf->limits, handler->pid)) { |
1323 | ERROR("failed to setup resource limits for '%s'", name); | |
84ff3af7 | 1324 | goto out_delete_net; |
c6d09e15 WB |
1325 | } |
1326 | ||
f4152036 CB |
1327 | if (lxc_sync_barrier_child(handler, LXC_SYNC_CGROUP_UNSHARE)) |
1328 | goto out_delete_net; | |
1329 | ||
9daf6f5d | 1330 | if (!cgroup_setup_limits(handler, true)) { |
408da065 | 1331 | ERROR("Failed to setup the devices cgroup for container \"%s\".", name); |
b98f7d6e | 1332 | goto out_delete_net; |
544a48a0 | 1333 | } |
f4152036 | 1334 | TRACE("Set up cgroup device limits"); |
544a48a0 | 1335 | |
73d28d42 | 1336 | cgroup_disconnect(); |
7e4dfe0b | 1337 | cgroups_connected = false; |
73d28d42 | 1338 | |
408da065 CB |
1339 | /* Tell the child to complete its initialization and wait for it to exec |
1340 | * or return an error. (The child will never return | |
1341 | * LXC_SYNC_POST_CGROUP+1. It will either close the sync pipe, causing | |
1342 | * lxc_sync_barrier_child to return success, or return a different | |
1343 | * value, causing us to error out). | |
544a48a0 SH |
1344 | */ |
1345 | if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CGROUP)) | |
3c22086f | 1346 | return -1; |
0ad19a3f | 1347 | |
790255cf CB |
1348 | if (lxc_network_recv_name_and_ifindex_from_child(handler) < 0) { |
1349 | ERROR("Failed to receive names and ifindices for network " | |
1350 | "devices from child"); | |
1351 | goto out_delete_net; | |
1352 | } | |
1353 | ||
1354 | /* Now all networks are created, network devices are moved into place, | |
1355 | * and the correct names and ifindeces in the respective namespaces have | |
1356 | * been recorded. The corresponding structs have now all been filled. So | |
1357 | * log them for debugging purposes. | |
1358 | */ | |
1359 | lxc_log_configured_netdevs(handler->conf); | |
1360 | ||
f4152036 CB |
1361 | /* Read tty fds allocated by child. */ |
1362 | if (lxc_recv_ttys_from_child(handler) < 0) { | |
1363 | ERROR("Failed to receive tty info from child process."); | |
1364 | goto out_delete_net; | |
1365 | } | |
1366 | ||
23c53af9 | 1367 | if (handler->ops->post_start(handler, handler->data)) |
e6126dbe MN |
1368 | goto out_abort; |
1369 | ||
25c2aca5 | 1370 | if (lxc_set_state(name, handler, RUNNING)) { |
408da065 CB |
1371 | ERROR("Failed to set state for container \"%s\" to \"%s\".", name, |
1372 | lxc_state2str(RUNNING)); | |
59eb99ba | 1373 | goto out_abort; |
3f21c114 | 1374 | } |
22ebac19 | 1375 | |
3c22086f | 1376 | lxc_sync_fini(handler); |
0c547523 | 1377 | |
e6126dbe | 1378 | return 0; |
1ac470c0 | 1379 | |
7fef7a06 | 1380 | out_delete_net: |
7e4dfe0b SH |
1381 | if (cgroups_connected) |
1382 | cgroup_disconnect(); | |
1bd8d726 CB |
1383 | |
1384 | if (handler->clone_flags & CLONE_NEWNET) { | |
1385 | DEBUG("Tearing down network devices"); | |
1386 | if (!lxc_delete_network_priv(handler)) | |
1387 | DEBUG("Failed tearing down network devices"); | |
1388 | ||
1389 | if (!lxc_delete_network_unpriv(handler)) | |
1390 | DEBUG("Failed tearing down network devices"); | |
1391 | } | |
1392 | ||
59eb99ba DL |
1393 | out_abort: |
1394 | lxc_abort(name, handler); | |
3c22086f | 1395 | lxc_sync_fini(handler); |
5c068da9 SH |
1396 | if (handler->pinfd >= 0) { |
1397 | close(handler->pinfd); | |
1398 | handler->pinfd = -1; | |
1399 | } | |
1400 | ||
1bd8d726 CB |
1401 | if (handler->netnsfd >= 0) { |
1402 | close(handler->netnsfd); | |
1403 | handler->netnsfd = -1; | |
1404 | } | |
1405 | ||
b79fcd86 | 1406 | return -1; |
59eb99ba | 1407 | } |
0ad19a3f | 1408 | |
aa460476 | 1409 | int __lxc_start(const char *name, struct lxc_handler *handler, |
507cee36 TA |
1410 | struct lxc_operations* ops, void *data, const char *lxcpath, |
1411 | bool backgrounded) | |
59eb99ba | 1412 | { |
59eb99ba | 1413 | int status; |
aa460476 | 1414 | int err = -1; |
aa460476 | 1415 | struct lxc_conf *conf = handler->conf; |
80090207 | 1416 | |
aa460476 | 1417 | if (lxc_init(name, handler) < 0) { |
408da065 | 1418 | ERROR("Failed to initialize container \"%s\".", name); |
66aeffc7 | 1419 | return -1; |
0ad19a3f | 1420 | } |
ee70bf78 CLG |
1421 | handler->ops = ops; |
1422 | handler->data = data; | |
507cee36 | 1423 | handler->backgrounded = backgrounded; |
738d0deb | 1424 | handler->netnsfd = -1; |
e6126dbe | 1425 | |
76a26f55 | 1426 | if (!attach_block_device(handler->conf)) { |
408da065 | 1427 | ERROR("Failed to attach block device."); |
76a26f55 SH |
1428 | goto out_fini_nonet; |
1429 | } | |
1430 | ||
35120d9c | 1431 | if (geteuid() == 0 && !lxc_list_empty(&conf->id_map)) { |
408da065 | 1432 | /* If the backing store is a device, mount it here and now. */ |
35120d9c SH |
1433 | if (rootfs_is_blockdev(conf)) { |
1434 | if (unshare(CLONE_NEWNS) < 0) { | |
408da065 | 1435 | ERROR("Failed to unshare CLONE_NEWNS."); |
35120d9c SH |
1436 | goto out_fini_nonet; |
1437 | } | |
408da065 CB |
1438 | INFO("Unshared CLONE_NEWNS."); |
1439 | ||
6a0c909a | 1440 | remount_all_slave(); |
35120d9c | 1441 | if (do_rootfs_setup(conf, name, lxcpath) < 0) { |
408da065 | 1442 | ERROR("Error setting up rootfs mount as root before spawn."); |
35120d9c SH |
1443 | goto out_fini_nonet; |
1444 | } | |
408da065 | 1445 | INFO("Set up container rootfs as host root."); |
35120d9c SH |
1446 | } |
1447 | } | |
1448 | ||
23c53af9 | 1449 | err = lxc_spawn(handler); |
59eb99ba | 1450 | if (err) { |
408da065 | 1451 | ERROR("Failed to spawn container \"%s\".", name); |
76a26f55 | 1452 | goto out_detach_blockdev; |
0ad19a3f | 1453 | } |
1454 | ||
8bee8851 WB |
1455 | handler->conf->reboot = 0; |
1456 | ||
3a0f472d | 1457 | err = lxc_poll(name, handler); |
e043236e | 1458 | if (err) { |
408da065 | 1459 | ERROR("LXC mainloop exited with error: %d.", err); |
59eb99ba DL |
1460 | goto out_abort; |
1461 | } | |
0ad19a3f | 1462 | |
3a0f472d | 1463 | while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR) |
1bc5cc8c | 1464 | continue; |
e043236e | 1465 | |
408da065 CB |
1466 | /* If the child process exited but was not signaled, it didn't call |
1467 | * reboot. This should mean it was an lxc-execute which simply exited. | |
1468 | * In any case, treat it as a 'halt'. | |
8b004f07 | 1469 | */ |
d028235d | 1470 | if (WIFSIGNALED(status)) { |
8b004f07 SH |
1471 | switch(WTERMSIG(status)) { |
1472 | case SIGINT: /* halt */ | |
408da065 | 1473 | DEBUG("Container \"%s\" is halting.", name); |
8b004f07 SH |
1474 | break; |
1475 | case SIGHUP: /* reboot */ | |
408da065 | 1476 | DEBUG("Container \"%s\" is rebooting.", name); |
8b004f07 SH |
1477 | handler->conf->reboot = 1; |
1478 | break; | |
c2b9bd9e | 1479 | case SIGSYS: /* seccomp */ |
408da065 | 1480 | DEBUG("Container \"%s\" violated its seccomp policy.", name); |
c2b9bd9e | 1481 | break; |
8b004f07 | 1482 | default: |
408da065 | 1483 | DEBUG("Unknown exit status for container \"%s\" init %d.", name, WTERMSIG(status)); |
8b004f07 SH |
1484 | break; |
1485 | } | |
d028235d | 1486 | } |
828695d9 | 1487 | |
b809f232 CB |
1488 | err = lxc_restore_phys_nics_to_netns(handler); |
1489 | if (err < 0) | |
1490 | ERROR("Failed to move physical network devices back to parent " | |
1491 | "network namespace"); | |
ce5782df | 1492 | |
5c068da9 SH |
1493 | if (handler->pinfd >= 0) { |
1494 | close(handler->pinfd); | |
1495 | handler->pinfd = -1; | |
1496 | } | |
1497 | ||
1787abca | 1498 | lxc_monitor_send_exit_code(name, status, handler->lxcpath); |
3a0f472d | 1499 | err = lxc_error_set_and_log(handler->pid, status); |
1bd8d726 | 1500 | |
9d7f9e52 | 1501 | out_fini: |
1bd8d726 CB |
1502 | DEBUG("Tearing down network devices"); |
1503 | if (!lxc_delete_network_priv(handler)) | |
1504 | DEBUG("Failed tearing down network devices"); | |
1505 | ||
1506 | if (!lxc_delete_network_unpriv(handler)) | |
1507 | DEBUG("Failed tearing down network devices"); | |
1508 | ||
1509 | if (handler->netnsfd >= 0) { | |
1510 | close(handler->netnsfd); | |
1511 | handler->netnsfd = -1; | |
358daf49 | 1512 | } |
74a2b586 | 1513 | |
76a26f55 SH |
1514 | out_detach_blockdev: |
1515 | detach_block_device(handler->conf); | |
1516 | ||
74a2b586 | 1517 | out_fini_nonet: |
3a0f472d | 1518 | lxc_fini(name, handler); |
0ad19a3f | 1519 | return err; |
1520 | ||
59eb99ba | 1521 | out_abort: |
3a0f472d | 1522 | lxc_abort(name, handler); |
9d7f9e52 | 1523 | goto out_fini; |
0ad19a3f | 1524 | } |
ee70bf78 CLG |
1525 | |
1526 | struct start_args { | |
1527 | char *const *argv; | |
1528 | }; | |
1529 | ||
1530 | static int start(struct lxc_handler *handler, void* data) | |
1531 | { | |
1532 | struct start_args *arg = data; | |
1533 | ||
408da065 | 1534 | NOTICE("Exec'ing \"%s\".", arg->argv[0]); |
ee70bf78 CLG |
1535 | |
1536 | execvp(arg->argv[0], arg->argv); | |
408da065 | 1537 | SYSERROR("Failed to exec \"%s\".", arg->argv[0]); |
ee70bf78 CLG |
1538 | return 0; |
1539 | } | |
1540 | ||
1541 | static int post_start(struct lxc_handler *handler, void* data) | |
1542 | { | |
1543 | struct start_args *arg = data; | |
1544 | ||
408da065 | 1545 | NOTICE("Started \"%s\" with pid \"%d\".", arg->argv[0], handler->pid); |
ee70bf78 CLG |
1546 | return 0; |
1547 | } | |
1548 | ||
1549 | static struct lxc_operations start_ops = { | |
1550 | .start = start, | |
1551 | .post_start = post_start | |
1552 | }; | |
1553 | ||
aa460476 | 1554 | int lxc_start(const char *name, char *const argv[], struct lxc_handler *handler, |
507cee36 | 1555 | const char *lxcpath, bool backgrounded) |
ee70bf78 CLG |
1556 | { |
1557 | struct start_args start_arg = { | |
1558 | .argv = argv, | |
1559 | }; | |
1560 | ||
aa460476 | 1561 | return __lxc_start(name, handler, &start_ops, &start_arg, lxcpath, backgrounded); |
ee70bf78 | 1562 | } |
28272964 CB |
1563 | |
1564 | static void lxc_destroy_container_on_signal(struct lxc_handler *handler, | |
1565 | const char *name) | |
1566 | { | |
1567 | char destroy[MAXPATHLEN]; | |
1568 | bool bret = true; | |
1569 | int ret = 0; | |
f01f7975 | 1570 | struct lxc_container *c; |
df31363a | 1571 | if (handler->conf->rootfs.path && handler->conf->rootfs.mount) { |
790255cf | 1572 | bret = do_destroy_container(handler); |
28272964 | 1573 | if (!bret) { |
408da065 | 1574 | ERROR("Error destroying rootfs for container \"%s\".", name); |
28272964 CB |
1575 | return; |
1576 | } | |
1577 | } | |
408da065 | 1578 | INFO("Destroyed rootfs for container \"%s\".", name); |
28272964 CB |
1579 | |
1580 | ret = snprintf(destroy, MAXPATHLEN, "%s/%s", handler->lxcpath, name); | |
1581 | if (ret < 0 || ret >= MAXPATHLEN) { | |
408da065 | 1582 | ERROR("Error destroying directory for container \"%s\".", name); |
28272964 CB |
1583 | return; |
1584 | } | |
1585 | ||
f01f7975 CB |
1586 | c = lxc_container_new(name, handler->lxcpath); |
1587 | if (c) { | |
1588 | if (container_disk_lock(c)) { | |
408da065 | 1589 | INFO("Could not update lxc_snapshots file."); |
f01f7975 CB |
1590 | lxc_container_put(c); |
1591 | } else { | |
1592 | mod_all_rdeps(c, false); | |
1593 | container_disk_unlock(c); | |
1594 | lxc_container_put(c); | |
1595 | } | |
1596 | } | |
1597 | ||
d0fbc7ba | 1598 | if (!handler->am_root) |
c9b7c33e CB |
1599 | ret = userns_exec_1(handler->conf, lxc_rmdir_onedev_wrapper, |
1600 | destroy, "lxc_rmdir_onedev_wrapper"); | |
28272964 CB |
1601 | else |
1602 | ret = lxc_rmdir_onedev(destroy, NULL); | |
1603 | ||
1604 | if (ret < 0) { | |
408da065 | 1605 | ERROR("Error destroying directory for container \"%s\".", name); |
28272964 CB |
1606 | return; |
1607 | } | |
408da065 | 1608 | INFO("Destroyed directory for container \"%s\".", name); |
28272964 CB |
1609 | } |
1610 | ||
1611 | static int lxc_rmdir_onedev_wrapper(void *data) | |
1612 | { | |
1613 | char *arg = (char *) data; | |
1614 | return lxc_rmdir_onedev(arg, NULL); | |
1615 | } | |
1616 | ||
790255cf | 1617 | static bool do_destroy_container(struct lxc_handler *handler) { |
d0fbc7ba | 1618 | if (!handler->am_root) { |
790255cf CB |
1619 | if (userns_exec_1(handler->conf, storage_destroy_wrapper, |
1620 | handler->conf, "storage_destroy_wrapper") < 0) | |
d028235d | 1621 | return false; |
10bc1861 | 1622 | |
d028235d SG |
1623 | return true; |
1624 | } | |
10bc1861 | 1625 | |
790255cf | 1626 | return storage_destroy(handler->conf); |
28272964 | 1627 | } |