]>
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 | ||
995acccc | 55 | #if !HAVE_DECL_PR_CAPBSET_DROP |
656994bb MH |
56 | #define PR_CAPBSET_DROP 24 |
57 | #endif | |
58 | ||
995acccc | 59 | #if !HAVE_DECL_PR_SET_NO_NEW_PRIVS |
955e2a02 CB |
60 | #define PR_SET_NO_NEW_PRIVS 38 |
61 | #endif | |
62 | ||
995acccc | 63 | #if !HAVE_DECL_PR_GET_NO_NEW_PRIVS |
955e2a02 CB |
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; | |
28d9e29e | 158 | |
4d8ac866 | 159 | ns_fd[i] = lxc_preserve_ns(pid, ns_info[i].proc_name); |
9f30a190 MM |
160 | if (ns_fd[i] < 0) |
161 | goto error; | |
fa3a5b22 CB |
162 | |
163 | DEBUG("Preserved %s namespace via fd %d", ns_info[i].proc_name, ns_fd[i]); | |
9f30a190 MM |
164 | } |
165 | ||
62d05d9b | 166 | return true; |
9f30a190 MM |
167 | |
168 | error: | |
4d8ac866 CB |
169 | if (errno == ENOENT) |
170 | SYSERROR("Kernel does not support attaching to %s namespaces.", ns_info[i].proc_name); | |
171 | else | |
172 | SYSERROR("Failed to open file descriptor for %s namespace: %s.", ns_info[i].proc_name, strerror(errno)); | |
9f30a190 | 173 | close_ns(ns_fd); |
62d05d9b | 174 | return false; |
9f30a190 MM |
175 | } |
176 | ||
80090207 CLG |
177 | static int match_fd(int fd) |
178 | { | |
179 | return (fd == 0 || fd == 1 || fd == 2); | |
180 | } | |
181 | ||
47a46cf1 CB |
182 | int lxc_check_inherited(struct lxc_conf *conf, bool closeall, |
183 | int *fds_to_ignore, size_t len_fds) | |
80090207 | 184 | { |
74f96976 | 185 | struct dirent *direntp; |
80090207 | 186 | int fd, fddir; |
47a46cf1 | 187 | size_t i; |
80090207 | 188 | DIR *dir; |
80090207 | 189 | |
d2cf4c37 SH |
190 | if (conf && conf->close_all_fds) |
191 | closeall = true; | |
192 | ||
b119f362 | 193 | restart: |
80090207 CLG |
194 | dir = opendir("/proc/self/fd"); |
195 | if (!dir) { | |
13277ec4 | 196 | WARN("Failed to open directory: %s.", strerror(errno)); |
80090207 CLG |
197 | return -1; |
198 | } | |
199 | ||
200 | fddir = dirfd(dir); | |
201 | ||
74f96976 | 202 | while ((direntp = readdir(dir))) { |
80090207 CLG |
203 | if (!direntp) |
204 | break; | |
205 | ||
206 | if (!strcmp(direntp->d_name, ".")) | |
207 | continue; | |
208 | ||
209 | if (!strcmp(direntp->d_name, "..")) | |
210 | continue; | |
211 | ||
d4cff0d2 CB |
212 | if (lxc_safe_int(direntp->d_name, &fd) < 0) { |
213 | INFO("Could not parse file descriptor for: %s", direntp->d_name); | |
214 | continue; | |
215 | } | |
80090207 | 216 | |
47a46cf1 CB |
217 | for (i = 0; i < len_fds; i++) |
218 | if (fds_to_ignore[i] == fd) | |
219 | break; | |
220 | ||
221 | if (fd == fddir || fd == lxc_log_fd || | |
222 | (i < len_fds && fd == fds_to_ignore[i])) | |
80090207 CLG |
223 | continue; |
224 | ||
858377e4 SH |
225 | if (current_config && fd == current_config->logfd) |
226 | continue; | |
227 | ||
80090207 CLG |
228 | if (match_fd(fd)) |
229 | continue; | |
80090207 | 230 | |
d2cf4c37 | 231 | if (closeall) { |
b119f362 SH |
232 | close(fd); |
233 | closedir(dir); | |
28d9e29e | 234 | INFO("Closed inherited fd %d", fd); |
b119f362 SH |
235 | goto restart; |
236 | } | |
28d9e29e | 237 | WARN("Inherited fd %d", fd); |
80090207 CLG |
238 | } |
239 | ||
408da065 | 240 | /* Only enable syslog at this point to avoid the above logging function |
64c57ea1 BD |
241 | * to open a new fd and make the check_inherited function enter an |
242 | * infinite loop. | |
243 | */ | |
244 | lxc_log_enable_syslog(); | |
245 | ||
92c7f629 GK |
246 | closedir(dir); /* cannot fail */ |
247 | return 0; | |
80090207 CLG |
248 | } |
249 | ||
83ee7875 | 250 | static int setup_signal_fd(sigset_t *oldmask) |
b0a33c1e | 251 | { |
252 | sigset_t mask; | |
253 | int fd; | |
254 | ||
408da065 | 255 | /* Block everything except serious error signals. */ |
f3304a29 FW |
256 | if (sigfillset(&mask) || |
257 | sigdelset(&mask, SIGILL) || | |
258 | sigdelset(&mask, SIGSEGV) || | |
259 | sigdelset(&mask, SIGBUS) || | |
b5159817 | 260 | sigdelset(&mask, SIGWINCH) || |
f3304a29 | 261 | sigprocmask(SIG_BLOCK, &mask, oldmask)) { |
408da065 | 262 | SYSERROR("Failed to set signal mask."); |
b0a33c1e | 263 | return -1; |
264 | } | |
265 | ||
266 | fd = signalfd(-1, &mask, 0); | |
267 | if (fd < 0) { | |
408da065 | 268 | SYSERROR("Failed to create signal file descriptor."); |
b0a33c1e | 269 | return -1; |
270 | } | |
271 | ||
272 | if (fcntl(fd, F_SETFD, FD_CLOEXEC)) { | |
408da065 | 273 | SYSERROR("Failed to set FD_CLOEXEC on the signal file descriptor: %d.", fd); |
b0a33c1e | 274 | close(fd); |
275 | return -1; | |
276 | } | |
277 | ||
408da065 | 278 | DEBUG("Set SIGCHLD handler with file descriptor: %d.", fd); |
1ac470c0 | 279 | |
b0a33c1e | 280 | return fd; |
281 | } | |
282 | ||
84c92abd | 283 | static int signal_handler(int fd, uint32_t events, void *data, |
f3787121 | 284 | struct lxc_epoll_descr *descr) |
b0a33c1e | 285 | { |
15cd25fd | 286 | struct signalfd_siginfo siginfo; |
80507ee8 | 287 | siginfo_t info; |
15cd25fd | 288 | int ret; |
82d89dce | 289 | pid_t *pid = data; |
80507ee8 | 290 | bool init_died = false; |
15cd25fd DL |
291 | |
292 | ret = read(fd, &siginfo, sizeof(siginfo)); | |
293 | if (ret < 0) { | |
408da065 | 294 | ERROR("Failed to read signal info from signal file descriptor: %d.", fd); |
15cd25fd DL |
295 | return -1; |
296 | } | |
297 | ||
298 | if (ret != sizeof(siginfo)) { | |
408da065 | 299 | ERROR("Unexpected size for siginfo struct."); |
15cd25fd DL |
300 | return -1; |
301 | } | |
302 | ||
408da065 | 303 | /* Check whether init is running. */ |
80507ee8 SH |
304 | info.si_pid = 0; |
305 | ret = waitid(P_PID, *pid, &info, WEXITED | WNOWAIT | WNOHANG); | |
408da065 | 306 | if (ret == 0 && info.si_pid == *pid) |
80507ee8 | 307 | init_died = true; |
80507ee8 | 308 | |
f3304a29 FW |
309 | if (siginfo.ssi_signo != SIGCHLD) { |
310 | kill(*pid, siginfo.ssi_signo); | |
408da065 | 311 | INFO("Forwarded signal %d to pid %d.", siginfo.ssi_signo, *pid); |
80507ee8 | 312 | return init_died ? 1 : 0; |
f3304a29 FW |
313 | } |
314 | ||
408da065 CB |
315 | if (siginfo.ssi_code == CLD_STOPPED) { |
316 | INFO("Container init process was stopped."); | |
317 | return init_died ? 1 : 0; | |
318 | } else if (siginfo.ssi_code == CLD_CONTINUED) { | |
319 | INFO("Container init process was continued."); | |
80507ee8 | 320 | return init_died ? 1 : 0; |
15cd25fd | 321 | } |
1ac470c0 | 322 | |
408da065 CB |
323 | /* More robustness, protect ourself from a SIGCHLD sent |
324 | * by a process different from the container init. | |
82d89dce DL |
325 | */ |
326 | if (siginfo.ssi_pid != *pid) { | |
af4c0f05 | 327 | NOTICE("Received SIGCHLD from pid %d instead of container init %d.", siginfo.ssi_pid, *pid); |
80507ee8 | 328 | return init_died ? 1 : 0; |
82d89dce DL |
329 | } |
330 | ||
408da065 | 331 | DEBUG("Container init process %d exited.", *pid); |
b0a33c1e | 332 | return 1; |
333 | } | |
334 | ||
5e5576a4 CB |
335 | static int lxc_serve_state_clients(const char *name, |
336 | struct lxc_handler *handler, | |
337 | lxc_state_t state) | |
66aeffc7 | 338 | { |
dbc9832d CB |
339 | ssize_t ret; |
340 | struct lxc_list *cur, *next; | |
341 | struct state_client *client; | |
342 | struct lxc_msg msg = {.type = lxc_msg_state, .value = state}; | |
343 | ||
344 | process_lock(); | |
5e5576a4 | 345 | |
dbc9832d | 346 | /* Only set state under process lock held so that we don't cause |
54446942 | 347 | * lxc_cmd_add_state_client() to miss a state. |
dbc9832d | 348 | */ |
66aeffc7 | 349 | handler->state = state; |
9dfa4041 | 350 | TRACE("Set container state to %s", lxc_state2str(state)); |
dbc9832d CB |
351 | |
352 | if (lxc_list_empty(&handler->state_clients)) { | |
9dfa4041 | 353 | TRACE("No state clients registered"); |
dbc9832d | 354 | process_unlock(); |
65ee1673 | 355 | lxc_monitor_send_state(name, state, handler->lxcpath); |
dbc9832d CB |
356 | return 0; |
357 | } | |
358 | ||
359 | strncpy(msg.name, name, sizeof(msg.name)); | |
360 | msg.name[sizeof(msg.name) - 1] = 0; | |
361 | ||
362 | lxc_list_for_each_safe(cur, &handler->state_clients, next) { | |
363 | client = cur->elem; | |
364 | ||
365 | if (!client->states[state]) { | |
9dfa4041 | 366 | TRACE("State %s not registered for state client %d", |
dbc9832d CB |
367 | lxc_state2str(state), client->clientfd); |
368 | continue; | |
369 | } | |
370 | ||
9dfa4041 | 371 | TRACE("Sending state %s to state client %d", |
dbc9832d CB |
372 | lxc_state2str(state), client->clientfd); |
373 | ||
374 | again: | |
375 | ret = send(client->clientfd, &msg, sizeof(msg), 0); | |
ee8377bd CB |
376 | if (ret <= 0) { |
377 | if (errno == EINTR) { | |
378 | TRACE("Caught EINTR; retrying"); | |
dbc9832d | 379 | goto again; |
ee8377bd | 380 | } |
dbc9832d | 381 | |
9dfa4041 CB |
382 | ERROR("%s - Failed to send message to client", |
383 | strerror(errno)); | |
dbc9832d CB |
384 | } |
385 | ||
386 | /* kick client from list */ | |
387 | close(client->clientfd); | |
388 | lxc_list_del(cur); | |
389 | free(cur->elem); | |
390 | free(cur); | |
391 | } | |
392 | process_unlock(); | |
393 | ||
5e5576a4 CB |
394 | return 0; |
395 | } | |
396 | ||
397 | static int lxc_serve_state_socket_pair(const char *name, | |
398 | struct lxc_handler *handler, | |
399 | lxc_state_t state) | |
400 | { | |
401 | ssize_t ret; | |
402 | ||
403 | if (!handler->backgrounded || | |
404 | handler->state_socket_pair[1] < 0 || | |
405 | state == STARTING) | |
406 | return 0; | |
407 | ||
408 | /* Close read end of the socket pair. */ | |
409 | close(handler->state_socket_pair[0]); | |
410 | handler->state_socket_pair[0] = -1; | |
411 | ||
ee8377bd | 412 | again: |
5e5576a4 CB |
413 | ret = lxc_abstract_unix_send_credential(handler->state_socket_pair[1], |
414 | &(int){state}, sizeof(int)); | |
ee8377bd CB |
415 | if (ret != sizeof(int)) { |
416 | if (errno == EINTR) | |
417 | goto again; | |
5e5576a4 CB |
418 | SYSERROR("Failed to send state to %d", |
419 | handler->state_socket_pair[1]); | |
ee8377bd CB |
420 | return -1; |
421 | } | |
5e5576a4 CB |
422 | |
423 | TRACE("Sent container state \"%s\" to %d", lxc_state2str(state), | |
424 | handler->state_socket_pair[1]); | |
425 | ||
426 | /* Close write end of the socket pair. */ | |
427 | close(handler->state_socket_pair[1]); | |
428 | handler->state_socket_pair[1] = -1; | |
429 | ||
430 | return 0; | |
431 | } | |
432 | ||
433 | int lxc_set_state(const char *name, struct lxc_handler *handler, | |
434 | lxc_state_t state) | |
435 | { | |
436 | int ret; | |
437 | ||
438 | ret = lxc_serve_state_socket_pair(name, handler, state); | |
439 | if (ret < 0) { | |
440 | ERROR("Failed to synchronize via anonymous pair of unix sockets"); | |
441 | return -1; | |
442 | } | |
443 | ||
444 | ret = lxc_serve_state_clients(name, handler, state); | |
445 | if (ret < 0) | |
446 | return -1; | |
447 | ||
dbc9832d CB |
448 | /* This function will try to connect to the legacy lxc-monitord state |
449 | * server and only exists for backwards compatibility. | |
450 | */ | |
9123e471 | 451 | lxc_monitor_send_state(name, state, handler->lxcpath); |
dbc9832d | 452 | |
66aeffc7 DL |
453 | return 0; |
454 | } | |
455 | ||
735f2c6e | 456 | int lxc_poll(const char *name, struct lxc_handler *handler) |
b0a33c1e | 457 | { |
ca5f7926 DL |
458 | int sigfd = handler->sigfd; |
459 | int pid = handler->pid; | |
b0a33c1e | 460 | struct lxc_epoll_descr descr; |
461 | ||
a9e61274 | 462 | if (lxc_mainloop_open(&descr)) { |
408da065 | 463 | ERROR("Failed to create LXC mainloop."); |
50c8bf05 | 464 | goto out_sigfd; |
b0a33c1e | 465 | } |
466 | ||
83ee7875 | 467 | if (lxc_mainloop_add_handler(&descr, sigfd, signal_handler, &pid)) { |
408da065 | 468 | ERROR("Failed to add signal handler with file descriptor %d to LXC mainloop.", sigfd); |
b0a33c1e | 469 | goto out_mainloop_open; |
470 | } | |
471 | ||
da41561c | 472 | if (lxc_console_mainloop_add(&descr, handler->conf)) { |
408da065 | 473 | ERROR("Failed to add console handler to LXC mainloop."); |
63376d7d DL |
474 | goto out_mainloop_open; |
475 | } | |
476 | ||
ef6e34ee | 477 | if (lxc_cmd_mainloop_add(name, &descr, handler)) { |
408da065 | 478 | ERROR("Failed to add command handler to LXC mainloop."); |
96fa1ff0 | 479 | goto out_mainloop_open; |
563f2f2c DL |
480 | } |
481 | ||
3f903c04 | 482 | TRACE("lxc mainloop is ready"); |
b0a33c1e | 483 | |
e51d4895 | 484 | return lxc_mainloop(&descr, -1); |
b0a33c1e | 485 | |
486 | out_mainloop_open: | |
487 | lxc_mainloop_close(&descr); | |
408da065 | 488 | |
b0a33c1e | 489 | out_sigfd: |
490 | close(sigfd); | |
408da065 | 491 | |
c3e13372 | 492 | return -1; |
b0a33c1e | 493 | } |
494 | ||
f2e07cb6 CB |
495 | void lxc_free_handler(struct lxc_handler *handler) |
496 | { | |
497 | if (handler->conf && handler->conf->maincmd_fd) | |
498 | close(handler->conf->maincmd_fd); | |
499 | ||
5e5576a4 CB |
500 | if (handler->state_socket_pair[0] >= 0) |
501 | close(handler->state_socket_pair[0]); | |
502 | ||
503 | if (handler->state_socket_pair[1] >= 0) | |
504 | close(handler->state_socket_pair[1]); | |
505 | ||
f2e07cb6 CB |
506 | handler->conf = NULL; |
507 | free(handler); | |
508 | } | |
509 | ||
aa460476 | 510 | struct lxc_handler *lxc_init_handler(const char *name, struct lxc_conf *conf, |
5e5576a4 | 511 | const char *lxcpath, bool daemonize) |
59eb99ba | 512 | { |
5e5576a4 | 513 | int i, ret; |
3a0f472d DL |
514 | struct lxc_handler *handler; |
515 | ||
516 | handler = malloc(sizeof(*handler)); | |
aa460476 CB |
517 | if (!handler) { |
518 | ERROR("failed to allocate memory"); | |
3a0f472d | 519 | return NULL; |
aa460476 | 520 | } |
59eb99ba DL |
521 | |
522 | memset(handler, 0, sizeof(*handler)); | |
523 | ||
790255cf CB |
524 | /* Note that am_unpriv() checks the effective uid. We probably don't |
525 | * care if we are real root only if we are running as root so this | |
526 | * should be fine. | |
527 | */ | |
d0fbc7ba | 528 | handler->am_root = !am_unpriv(); |
c6012571 | 529 | handler->data_sock[0] = handler->data_sock[1] = -1; |
fae349da | 530 | handler->conf = conf; |
9123e471 | 531 | handler->lxcpath = lxcpath; |
5c068da9 | 532 | handler->pinfd = -1; |
5e5576a4 | 533 | handler->state_socket_pair[0] = handler->state_socket_pair[1] = -1; |
dbc9832d | 534 | lxc_list_init(&handler->state_clients); |
fae349da | 535 | |
b6b2b194 WB |
536 | for (i = 0; i < LXC_NS_MAX; i++) |
537 | handler->nsfd[i] = -1; | |
538 | ||
f0ecc19d | 539 | handler->name = name; |
3bdf52d7 | 540 | |
5e5576a4 CB |
541 | if (daemonize && !handler->conf->reboot) { |
542 | /* Create socketpair() to synchronize on daemonized startup. | |
543 | * When the container reboots we don't need to synchronize again | |
544 | * currently so don't open another socketpair(). | |
545 | */ | |
546 | ret = socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0, | |
547 | handler->state_socket_pair); | |
548 | if (ret < 0) { | |
549 | ERROR("Failed to create anonymous pair of unix sockets"); | |
550 | goto on_error; | |
551 | } | |
552 | TRACE("Created anonymous pair {%d,%d} of unix sockets", | |
553 | handler->state_socket_pair[0], | |
554 | handler->state_socket_pair[1]); | |
555 | } | |
556 | ||
9dfa4041 CB |
557 | handler->conf->maincmd_fd = lxc_cmd_init(name, lxcpath, "command"); |
558 | if (handler->conf->maincmd_fd < 0) { | |
559 | ERROR("Failed to set up command socket"); | |
f2e07cb6 | 560 | goto on_error; |
aa460476 | 561 | } |
9dfa4041 | 562 | TRACE("Unix domain socket %d for command server is ready", |
aa460476 CB |
563 | handler->conf->maincmd_fd); |
564 | ||
565 | return handler; | |
566 | ||
f2e07cb6 CB |
567 | on_error: |
568 | lxc_free_handler(handler); | |
aa460476 CB |
569 | |
570 | return NULL; | |
571 | } | |
572 | ||
573 | int lxc_init(const char *name, struct lxc_handler *handler) | |
574 | { | |
575 | struct lxc_conf *conf = handler->conf; | |
576 | ||
577 | lsm_init(); | |
578 | TRACE("initialized LSM"); | |
d2e30e99 | 579 | |
8f2c3a70 | 580 | if (lxc_read_seccomp_config(conf) != 0) { |
408da065 | 581 | ERROR("Failed loading seccomp policy."); |
d2e30e99 | 582 | goto out_close_maincmd_fd; |
8f2c3a70 | 583 | } |
dbc9832d | 584 | TRACE("read seccomp policy"); |
8f2c3a70 | 585 | |
408da065 | 586 | /* Begin by setting the state to STARTING. */ |
25c2aca5 | 587 | if (lxc_set_state(name, handler, STARTING)) { |
408da065 | 588 | ERROR("Failed to set state for container \"%s\" to \"%s\".", name, lxc_state2str(STARTING)); |
051151de | 589 | goto out_close_maincmd_fd; |
0ad19a3f | 590 | } |
dbc9832d | 591 | TRACE("set container state to \"STARTING\""); |
0ad19a3f | 592 | |
408da065 CB |
593 | /* Start of environment variable setup for hooks. */ |
594 | if (name && setenv("LXC_NAME", name, 1)) | |
595 | SYSERROR("Failed to set environment variable: LXC_NAME=%s.", name); | |
596 | ||
597 | if (conf->rcfile && setenv("LXC_CONFIG_FILE", conf->rcfile, 1)) | |
598 | SYSERROR("Failed to set environment variable: LXC_CONFIG_FILE=%s.", conf->rcfile); | |
599 | ||
600 | if (conf->rootfs.mount && setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1)) | |
601 | SYSERROR("Failed to set environment variable: LXC_ROOTFS_MOUNT=%s.", conf->rootfs.mount); | |
602 | ||
603 | if (conf->rootfs.path && setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1)) | |
604 | SYSERROR("Failed to set environment variable: LXC_ROOTFS_PATH=%s.", conf->rootfs.path); | |
605 | ||
606 | if (conf->console.path && setenv("LXC_CONSOLE", conf->console.path, 1)) | |
607 | SYSERROR("Failed to set environment variable: LXC_CONSOLE=%s.", conf->console.path); | |
608 | ||
609 | if (conf->console.log_path && setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1)) | |
610 | SYSERROR("Failed to set environment variable: LXC_CONSOLE_LOGPATH=%s.", conf->console.log_path); | |
611 | ||
612 | if (setenv("LXC_CGNS_AWARE", "1", 1)) | |
613 | SYSERROR("Failed to set environment variable LXC_CGNS_AWARE=1."); | |
b8f88d9b CB |
614 | |
615 | if (setenv("LXC_LOG_LEVEL", lxc_log_priority_to_string(handler->conf->loglevel), 1)) | |
616 | SYSERROR("Failed to set environment variable LXC_CGNS_AWARE=1."); | |
408da065 | 617 | /* End of environment variable setup for hooks. */ |
f7bee6c6 | 618 | |
dbc9832d CB |
619 | TRACE("set environment variables"); |
620 | ||
283678ed | 621 | if (run_lxc_hooks(name, "pre-start", conf, handler->lxcpath, NULL)) { |
408da065 | 622 | ERROR("Failed to run lxc.hook.pre-start for container \"%s\".", name); |
773fb9ca SH |
623 | goto out_aborting; |
624 | } | |
dbc9832d | 625 | TRACE("ran pre-start hooks"); |
26ddeedd | 626 | |
408da065 CB |
627 | /* The signal fd has to be created before forking otherwise if the child |
628 | * process exits before we setup the signal fd, the event will be lost | |
629 | * and the command will be stuck. | |
630 | */ | |
83ee7875 | 631 | handler->sigfd = setup_signal_fd(&handler->oldmask); |
59eb99ba | 632 | if (handler->sigfd < 0) { |
408da065 | 633 | ERROR("Failed to setup SIGCHLD fd handler."); |
b5159817 DE |
634 | goto out_delete_tty; |
635 | } | |
dbc9832d | 636 | TRACE("set up signal fd"); |
b5159817 | 637 | |
408da065 | 638 | /* Do this after setting up signals since it might unblock SIGWINCH. */ |
b5159817 | 639 | if (lxc_console_create(conf)) { |
408da065 | 640 | ERROR("Failed to create console for container \"%s\".", name); |
b5159817 | 641 | goto out_restore_sigmask; |
b0a33c1e | 642 | } |
dbc9832d | 643 | TRACE("created console"); |
b0a33c1e | 644 | |
54117de5 | 645 | if (lxc_ttys_shift_ids(conf) < 0) { |
408da065 | 646 | ERROR("Failed to shift tty into container."); |
c4d10a05 SH |
647 | goto out_restore_sigmask; |
648 | } | |
dbc9832d | 649 | TRACE("shifted tty ids"); |
c4d10a05 | 650 | |
dbc9832d | 651 | INFO("container \"%s\" is initialized", name); |
aa460476 | 652 | return 0; |
59eb99ba | 653 | |
b5159817 DE |
654 | out_restore_sigmask: |
655 | sigprocmask(SIG_SETMASK, &handler->oldmask, NULL); | |
59eb99ba | 656 | out_delete_tty: |
fae349da | 657 | lxc_delete_tty(&conf->tty_info); |
59eb99ba | 658 | out_aborting: |
25c2aca5 | 659 | lxc_set_state(name, handler, ABORTING); |
d2e30e99 DE |
660 | out_close_maincmd_fd: |
661 | close(conf->maincmd_fd); | |
662 | conf->maincmd_fd = -1; | |
aa460476 | 663 | return -1; |
59eb99ba DL |
664 | } |
665 | ||
3b72c4a0 | 666 | void lxc_fini(const char *name, struct lxc_handler *handler) |
59eb99ba | 667 | { |
b3286b62 | 668 | int i, rc; |
dbc9832d | 669 | struct lxc_list *cur, *next; |
b3286b62 | 670 | pid_t self = getpid(); |
dbc9832d | 671 | char *namespaces[LXC_NS_MAX + 1]; |
b3286b62 | 672 | size_t namespace_count = 0; |
b6b2b194 | 673 | |
408da065 CB |
674 | /* The STOPPING state is there for future cleanup code which can take |
675 | * awhile. | |
59eb99ba | 676 | */ |
25c2aca5 | 677 | lxc_set_state(name, handler, STOPPING); |
b6b2b194 | 678 | |
b3286b62 WB |
679 | for (i = 0; i < LXC_NS_MAX; i++) { |
680 | if (handler->nsfd[i] != -1) { | |
681 | rc = asprintf(&namespaces[namespace_count], "%s:/proc/%d/fd/%d", | |
682 | ns_info[i].proc_name, self, handler->nsfd[i]); | |
683 | if (rc == -1) { | |
408da065 | 684 | SYSERROR("Failed to allocate memory."); |
b3286b62 WB |
685 | break; |
686 | } | |
687 | ++namespace_count; | |
688 | } | |
689 | } | |
690 | namespaces[namespace_count] = NULL; | |
c154af98 | 691 | |
408da065 CB |
692 | if (handler->conf->reboot && setenv("LXC_TARGET", "reboot", 1)) |
693 | SYSERROR("Failed to set environment variable: LXC_TARGET=reboot."); | |
694 | ||
695 | if (!handler->conf->reboot && setenv("LXC_TARGET", "stop", 1)) | |
696 | SYSERROR("Failed to set environment variable: LXC_TARGET=stop."); | |
c154af98 | 697 | |
b3286b62 | 698 | if (run_lxc_hooks(name, "stop", handler->conf, handler->lxcpath, namespaces)) |
408da065 | 699 | ERROR("Failed to run lxc.hook.stop for container \"%s\".", name); |
c154af98 | 700 | |
b3286b62 WB |
701 | while (namespace_count--) |
702 | free(namespaces[namespace_count]); | |
28d9e29e | 703 | |
b6b2b194 | 704 | for (i = 0; i < LXC_NS_MAX; i++) { |
28d9e29e CB |
705 | if (handler->nsfd[i] < 0) |
706 | continue; | |
738d0deb | 707 | |
28d9e29e CB |
708 | close(handler->nsfd[i]); |
709 | handler->nsfd[i] = -1; | |
738d0deb CB |
710 | } |
711 | ||
4288b79f CB |
712 | cgroup_destroy(handler); |
713 | ||
25c2aca5 | 714 | lxc_set_state(name, handler, STOPPED); |
59eb99ba | 715 | |
4288b79f CB |
716 | /* close command socket */ |
717 | close(handler->conf->maincmd_fd); | |
718 | handler->conf->maincmd_fd = -1; | |
719 | ||
f3787121 | 720 | if (run_lxc_hooks(name, "post-stop", handler->conf, handler->lxcpath, NULL)) { |
408da065 | 721 | ERROR("Failed to run lxc.hook.post-stop for container \"%s\".", name); |
f3787121 CB |
722 | if (handler->conf->reboot) { |
723 | WARN("Container will be stopped instead of rebooted."); | |
724 | handler->conf->reboot = 0; | |
408da065 CB |
725 | if (setenv("LXC_TARGET", "stop", 1)) |
726 | WARN("Failed to set environment variable: LXC_TARGET=stop."); | |
f3787121 CB |
727 | } |
728 | } | |
26ddeedd | 729 | |
408da065 | 730 | /* Reset mask set by setup_signal_fd. */ |
8f64a3f6 | 731 | if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) |
408da065 | 732 | WARN("Failed to restore signal mask."); |
8f64a3f6 | 733 | |
b5159817 | 734 | lxc_console_delete(&handler->conf->console); |
b2431939 | 735 | lxc_delete_tty(&handler->conf->tty_info); |
dbc9832d | 736 | |
dbc9832d CB |
737 | /* The command socket is now closed, no more state clients can register |
738 | * themselves from now on. So free the list of state clients. | |
739 | */ | |
740 | lxc_list_for_each_safe(cur, &handler->state_clients, next) { | |
741 | struct state_client *client = cur->elem; | |
742 | /* close state client socket */ | |
743 | close(client->clientfd); | |
744 | lxc_list_del(cur); | |
745 | free(cur->elem); | |
746 | free(cur); | |
747 | } | |
748 | ||
c6012571 CB |
749 | if (handler->data_sock[0] != -1) { |
750 | close(handler->data_sock[0]); | |
751 | close(handler->data_sock[1]); | |
e8bd4e43 | 752 | } |
2c5f2ede CB |
753 | |
754 | if (handler->conf->ephemeral == 1 && handler->conf->reboot != 1) | |
28272964 | 755 | lxc_destroy_container_on_signal(handler, name); |
2c5f2ede | 756 | |
b2431939 | 757 | free(handler); |
59eb99ba DL |
758 | } |
759 | ||
735f2c6e | 760 | void lxc_abort(const char *name, struct lxc_handler *handler) |
59eb99ba | 761 | { |
73e608b2 SH |
762 | int ret, status; |
763 | ||
25c2aca5 | 764 | lxc_set_state(name, handler, ABORTING); |
7d9fb3e9 DL |
765 | if (handler->pid > 0) |
766 | kill(handler->pid, SIGKILL); | |
408da065 CB |
767 | while ((ret = waitpid(-1, &status, 0)) > 0) { |
768 | ; | |
769 | } | |
59eb99ba DL |
770 | } |
771 | ||
ffe1e01a | 772 | static int do_start(void *data) |
50e98013 | 773 | { |
fa3a5b22 | 774 | int ret; |
7c661726 | 775 | struct lxc_list *iterator; |
7a55c157 | 776 | char path[PATH_MAX]; |
fa3a5b22 CB |
777 | int devnull_fd = -1; |
778 | struct lxc_handler *handler = data; | |
928b1f04 YT |
779 | bool have_cap_setgid; |
780 | uid_t new_uid; | |
781 | gid_t new_gid; | |
50e98013 DL |
782 | |
783 | if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) { | |
408da065 | 784 | SYSERROR("Failed to set signal mask."); |
9d7f9e52 | 785 | return -1; |
50e98013 DL |
786 | } |
787 | ||
408da065 CB |
788 | /* This prctl must be before the synchro, so if the parent dies before |
789 | * we set the parent death signal, we will detect its death with the | |
790 | * synchro right after, otherwise we have a window where the parent can | |
791 | * exit before we set the pdeath signal leading to a unsupervized | |
792 | * container. | |
743ecd2e DL |
793 | */ |
794 | if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) { | |
408da065 | 795 | SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL."); |
743ecd2e DL |
796 | return -1; |
797 | } | |
798 | ||
3c22086f | 799 | lxc_sync_fini_parent(handler); |
50e98013 | 800 | |
408da065 | 801 | /* Don't leak the pinfd to the container. */ |
025ed0f3 | 802 | if (handler->pinfd >= 0) { |
0d03360a | 803 | close(handler->pinfd); |
025ed0f3 | 804 | } |
2b0e17e4 | 805 | |
5b1e83cb SH |
806 | if (lxc_sync_wait_parent(handler, LXC_SYNC_STARTUP)) |
807 | return -1; | |
808 | ||
408da065 CB |
809 | /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See |
810 | * https://github.com/lxc/lxd/issues/1978. | |
811 | */ | |
5b1e83cb | 812 | if ((handler->clone_flags & (CLONE_NEWNET | CLONE_NEWUSER)) == |
408da065 | 813 | (CLONE_NEWNET | CLONE_NEWUSER)) { |
5b1e83cb SH |
814 | ret = unshare(CLONE_NEWNET); |
815 | if (ret < 0) { | |
408da065 | 816 | SYSERROR("Failed to unshare CLONE_NEWNET."); |
5b1e83cb SH |
817 | goto out_warn_father; |
818 | } | |
408da065 | 819 | INFO("Unshared CLONE_NEWNET."); |
5b1e83cb SH |
820 | } |
821 | ||
408da065 CB |
822 | /* Tell the parent task it can begin to configure the container and wait |
823 | * for it to finish. | |
3c22086f CLG |
824 | */ |
825 | if (lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE)) | |
9d7f9e52 | 826 | return -1; |
50e98013 | 827 | |
7ab1ba02 CB |
828 | if (lxc_network_recv_veth_names_from_parent(handler) < 0) { |
829 | ERROR("Failed to receive veth names from parent"); | |
658979c5 | 830 | goto out_warn_father; |
7ab1ba02 | 831 | } |
658979c5 | 832 | |
408da065 | 833 | /* If we are in a new user namespace, become root there to have |
d08f8d2f | 834 | * privilege over our namespace. |
f6d3e3e4 SH |
835 | */ |
836 | if (!lxc_list_empty(&handler->conf->id_map)) { | |
d08f8d2f | 837 | if (lxc_switch_uid_gid(0, 0) < 0) |
f6d3e3e4 | 838 | goto out_warn_father; |
d08f8d2f CB |
839 | |
840 | /* Drop groups only after we switched to a valid gid in the new | |
841 | * user namespace. | |
842 | */ | |
843 | if (lxc_setgroups(0, NULL) < 0) | |
c476bdce | 844 | goto out_warn_father; |
f6d3e3e4 SH |
845 | } |
846 | ||
99b71824 | 847 | if (access(handler->lxcpath, X_OK)) { |
c8154066 SH |
848 | print_top_failing_dir(handler->lxcpath); |
849 | goto out_warn_father; | |
850 | } | |
851 | ||
a91cde21 | 852 | ret = snprintf(path, sizeof(path), "%s/dev/null", handler->conf->rootfs.mount); |
408da065 | 853 | if (ret < 0 || ret >= sizeof(path)) |
7a55c157 | 854 | goto out_warn_father; |
7a55c157 TA |
855 | |
856 | /* In order to checkpoint restore, we need to have everything in the | |
857 | * same mount namespace. However, some containers may not have a | |
858 | * reasonable /dev (in particular, they may not have /dev/null), so we | |
859 | * can't set init's std fds to /dev/null by opening it from inside the | |
860 | * container. | |
861 | * | |
862 | * If that's the case, fall back to using the host's /dev/null. This | |
863 | * means that migration won't work, but at least we won't spew output | |
864 | * where it isn't wanted. | |
865 | */ | |
866 | if (handler->backgrounded && !handler->conf->autodev && access(path, F_OK) < 0) { | |
c44de748 AM |
867 | devnull_fd = open_devnull(); |
868 | ||
869 | if (devnull_fd < 0) | |
870 | goto out_warn_father; | |
408da065 CB |
871 | WARN("Using /dev/null from the host for container init's " |
872 | "standard file descriptors. Migration will not work."); | |
c44de748 AM |
873 | } |
874 | ||
408da065 | 875 | /* Ask father to setup cgroups and wait for him to finish. */ |
544a48a0 | 876 | if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP)) |
c44de748 | 877 | goto out_error; |
544a48a0 | 878 | |
deefdf8a CB |
879 | /* Unshare cgroup namespace after we have setup our cgroups. If we do it |
880 | * earlier we end up with a wrong view of /proc/self/cgroup. For | |
881 | * example, assume we unshare(CLONE_NEWCGROUP) first, and then create | |
882 | * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then | |
883 | * /proc/self/cgroup would show us: | |
884 | * | |
885 | * 8:cpuset:/lxc/c | |
886 | * | |
887 | * whereas it should actually show | |
888 | * | |
889 | * 8:cpuset:/ | |
890 | */ | |
891 | if (cgns_supported()) { | |
892 | if (unshare(CLONE_NEWCGROUP) < 0) { | |
893 | INFO("Failed to unshare CLONE_NEWCGROUP."); | |
894 | goto out_warn_father; | |
895 | } | |
896 | INFO("Unshared CLONE_NEWCGROUP."); | |
897 | } | |
898 | ||
149857af CB |
899 | /* Add the requested environment variables to the current environment to |
900 | * allow them to be used by the various hooks, such as the start hook | |
901 | * above. | |
98ff08ca | 902 | */ |
98ff08ca CB |
903 | lxc_list_for_each(iterator, &handler->conf->environment) { |
904 | if (putenv((char *)iterator->elem)) { | |
905 | SYSERROR("Failed to set environment variable: %s.", (char *)iterator->elem); | |
906 | goto out_warn_father; | |
907 | } | |
908 | } | |
909 | ||
f4152036 | 910 | /* Setup the container, ip, names, utsname, ... */ |
3b988b33 | 911 | ret = lxc_setup(handler); |
7729f8e5 CB |
912 | close(handler->data_sock[0]); |
913 | close(handler->data_sock[1]); | |
914 | if (ret < 0) { | |
f4152036 CB |
915 | ERROR("Failed to setup container \"%s\".", handler->name); |
916 | goto out_warn_father; | |
917 | } | |
918 | ||
408da065 | 919 | /* Set the label to change to when we exec(2) the container's init. */ |
7aff4f43 | 920 | if (lsm_process_label_set(NULL, handler->conf, 1, 1) < 0) |
e075f5d9 | 921 | goto out_warn_father; |
5112cd70 | 922 | |
029cdff5 | 923 | /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it |
408da065 CB |
924 | * before we aren't allowed anymore. |
925 | */ | |
029cdff5 CB |
926 | if (handler->conf->no_new_privs) { |
927 | if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) { | |
928 | SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges."); | |
929 | goto out_warn_father; | |
930 | } | |
931 | DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges."); | |
932 | } | |
933 | ||
0d9acb99 DE |
934 | /* Some init's such as busybox will set sane tty settings on stdin, |
935 | * stdout, stderr which it thinks is the console. We already set them | |
936 | * the way we wanted on the real terminal, and we want init to do its | |
408da065 CB |
937 | * setup on its console ie. the pty allocated in lxc_console_create() so |
938 | * make sure that that pty is stdin,stdout,stderr. | |
0d9acb99 | 939 | */ |
c5b93afb LF |
940 | if (handler->conf->console.slave >= 0) |
941 | if (set_stdfds(handler->conf->console.slave) < 0) { | |
942 | ERROR("Failed to redirect std{in,out,err} to pty file " | |
943 | "descriptor %d", | |
944 | handler->conf->console.slave); | |
945 | goto out_warn_father; | |
946 | } | |
0d9acb99 | 947 | |
408da065 | 948 | /* If we mounted a temporary proc, then unmount it now. */ |
5112cd70 | 949 | tmp_proc_unmount(handler->conf); |
e075f5d9 | 950 | |
8f2c3a70 SH |
951 | if (lxc_seccomp_load(handler->conf) != 0) |
952 | goto out_warn_father; | |
953 | ||
283678ed | 954 | if (run_lxc_hooks(handler->name, "start", handler->conf, handler->lxcpath, NULL)) { |
408da065 | 955 | ERROR("Failed to run lxc.hook.start for container \"%s\".", handler->name); |
773fb9ca SH |
956 | goto out_warn_father; |
957 | } | |
fc25b815 | 958 | |
773fb9ca | 959 | close(handler->sigfd); |
26ddeedd | 960 | |
7a55c157 TA |
961 | if (devnull_fd < 0) { |
962 | devnull_fd = open_devnull(); | |
963 | ||
964 | if (devnull_fd < 0) | |
965 | goto out_warn_father; | |
966 | } | |
967 | ||
c5b93afb LF |
968 | if (handler->conf->console.slave < 0 && handler->backgrounded) |
969 | if (set_stdfds(devnull_fd) < 0) { | |
970 | ERROR("Failed to redirect std{in,out,err} to " | |
971 | "\"/dev/null\""); | |
972 | goto out_warn_father; | |
973 | } | |
507cee36 | 974 | |
c44de748 AM |
975 | if (devnull_fd >= 0) { |
976 | close(devnull_fd); | |
977 | devnull_fd = -1; | |
978 | } | |
979 | ||
8c9a7665 TA |
980 | setsid(); |
981 | ||
f4152036 CB |
982 | if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP_LIMITS)) |
983 | goto out_warn_father; | |
984 | ||
149857af CB |
985 | /* Reset the environment variables the user requested in a clear |
986 | * environment. | |
987 | */ | |
988 | if (clearenv()) { | |
989 | SYSERROR("Failed to clear environment."); | |
990 | /* Don't error out though. */ | |
991 | } | |
992 | ||
993 | lxc_list_for_each(iterator, &handler->conf->environment) { | |
994 | if (putenv((char *)iterator->elem)) { | |
995 | SYSERROR("Failed to set environment variable: %s.", (char *)iterator->elem); | |
996 | goto out_warn_father; | |
997 | } | |
998 | } | |
999 | ||
98ff08ca CB |
1000 | if (putenv("container=lxc")) { |
1001 | SYSERROR("Failed to set environment variable: container=lxc."); | |
1002 | goto out_warn_father; | |
1003 | } | |
1004 | ||
1005 | if (handler->conf->pty_names) { | |
1006 | if (putenv(handler->conf->pty_names)) { | |
1007 | SYSERROR("Failed to set environment variable for container ptys."); | |
1008 | goto out_warn_father; | |
1009 | } | |
1010 | } | |
1011 | ||
76bdf299 CB |
1012 | /* The container has been setup. We can now switch to an unprivileged |
1013 | * uid/gid. | |
1014 | */ | |
928b1f04 YT |
1015 | new_uid = handler->conf->init_uid; |
1016 | new_gid = handler->conf->init_gid; | |
76bdf299 | 1017 | |
928b1f04 YT |
1018 | /* If we are in a new user namespace we already dropped all |
1019 | * groups when we switched to root in the new user namespace | |
1020 | * further above. Only drop groups if we can, so ensure that we | |
1021 | * have necessary privilege. | |
1022 | */ | |
1023 | #if HAVE_LIBCAP | |
1024 | have_cap_setgid = lxc_proc_cap_is_set(CAP_SETGID, CAP_EFFECTIVE); | |
1025 | #else | |
1026 | have_cap_setgid = false; | |
1027 | #endif | |
1028 | if (lxc_list_empty(&handler->conf->id_map) && have_cap_setgid) { | |
1029 | if (lxc_setgroups(0, NULL) < 0) | |
76bdf299 CB |
1030 | goto out_warn_father; |
1031 | } | |
1032 | ||
928b1f04 YT |
1033 | if (lxc_switch_uid_gid(new_uid, new_gid) < 0) |
1034 | goto out_warn_father; | |
1035 | ||
408da065 CB |
1036 | /* After this call, we are in error because this ops should not return |
1037 | * as it execs. | |
1038 | */ | |
c4ea60df | 1039 | handler->ops->start(handler, handler->data); |
50e98013 DL |
1040 | |
1041 | out_warn_father: | |
408da065 CB |
1042 | /* We want the parent to know something went wrong, so we return a |
1043 | * special error code. | |
1044 | */ | |
d1ccb562 | 1045 | lxc_sync_wake_parent(handler, LXC_SYNC_ERROR); |
c44de748 AM |
1046 | |
1047 | out_error: | |
1048 | if (devnull_fd >= 0) | |
1049 | close(devnull_fd); | |
1050 | ||
50e98013 DL |
1051 | return -1; |
1052 | } | |
1053 | ||
ae467c54 | 1054 | static int lxc_recv_ttys_from_child(struct lxc_handler *handler) |
e8bd4e43 | 1055 | { |
ae467c54 | 1056 | int i; |
ae467c54 CB |
1057 | struct lxc_pty_info *pty_info; |
1058 | int ret = -1; | |
c6012571 | 1059 | int sock = handler->data_sock[1]; |
e8bd4e43 | 1060 | struct lxc_conf *conf = handler->conf; |
e8bd4e43 SH |
1061 | struct lxc_tty_info *tty_info = &conf->tty_info; |
1062 | ||
1063 | if (!conf->tty) | |
1064 | return 0; | |
1065 | ||
408da065 CB |
1066 | tty_info->pty_info = malloc(sizeof(*tty_info->pty_info) * conf->tty); |
1067 | if (!tty_info->pty_info) | |
e8bd4e43 | 1068 | return -1; |
e8bd4e43 | 1069 | |
672c1e58 CB |
1070 | for (i = 0; i < conf->tty; i++) { |
1071 | int ttyfds[2]; | |
1072 | ||
1073 | ret = lxc_abstract_unix_recv_fds(sock, ttyfds, 2, NULL, 0); | |
1074 | if (ret < 0) | |
1075 | break; | |
ae467c54 | 1076 | |
672c1e58 | 1077 | pty_info = &tty_info->pty_info[i]; |
e8bd4e43 | 1078 | pty_info->busy = 0; |
672c1e58 CB |
1079 | pty_info->master = ttyfds[0]; |
1080 | pty_info->slave = ttyfds[1]; | |
1081 | TRACE("Received pty with master fd %d and slave fd %d from " | |
ae467c54 | 1082 | "parent", pty_info->master, pty_info->slave); |
e8bd4e43 | 1083 | } |
ae467c54 | 1084 | if (ret < 0) |
672c1e58 | 1085 | ERROR("Failed to receive %d ttys from child: %s", conf->tty, |
ae467c54 CB |
1086 | strerror(errno)); |
1087 | else | |
672c1e58 CB |
1088 | TRACE("Received %d ttys from child", conf->tty); |
1089 | ||
1090 | tty_info->nbtty = conf->tty; | |
ae467c54 CB |
1091 | |
1092 | return ret; | |
e8bd4e43 SH |
1093 | } |
1094 | ||
f813849c TA |
1095 | void resolve_clone_flags(struct lxc_handler *handler) |
1096 | { | |
03df7ab5 | 1097 | handler->clone_flags = CLONE_NEWNS; |
f813849c | 1098 | |
fa3a5b22 CB |
1099 | if (!handler->conf->inherit_ns[LXC_NS_USER]) { |
1100 | if (!lxc_list_empty(&handler->conf->id_map)) | |
1101 | handler->clone_flags |= CLONE_NEWUSER; | |
1102 | } else { | |
1103 | INFO("Inheriting user namespace"); | |
1104 | } | |
f813849c | 1105 | |
28d9e29e | 1106 | if (!handler->conf->inherit_ns[LXC_NS_NET]) { |
408da065 | 1107 | if (!lxc_requests_empty_network(handler)) |
f813849c TA |
1108 | handler->clone_flags |= CLONE_NEWNET; |
1109 | } else { | |
28d9e29e | 1110 | INFO("Inheriting net namespace"); |
f813849c TA |
1111 | } |
1112 | ||
28d9e29e | 1113 | if (!handler->conf->inherit_ns[LXC_NS_IPC]) |
f813849c | 1114 | handler->clone_flags |= CLONE_NEWIPC; |
408da065 | 1115 | else |
28d9e29e | 1116 | INFO("Inheriting ipc namespace"); |
f813849c | 1117 | |
28d9e29e | 1118 | if (!handler->conf->inherit_ns[LXC_NS_UTS]) |
f813849c | 1119 | handler->clone_flags |= CLONE_NEWUTS; |
408da065 | 1120 | else |
28d9e29e | 1121 | INFO("Inheriting uts namespace"); |
03df7ab5 | 1122 | |
28d9e29e | 1123 | if (!handler->conf->inherit_ns[LXC_NS_PID]) |
03df7ab5 L |
1124 | handler->clone_flags |= CLONE_NEWPID; |
1125 | else | |
28d9e29e | 1126 | INFO("Inheriting pid namespace"); |
f813849c TA |
1127 | } |
1128 | ||
fa3a5b22 CB |
1129 | static pid_t lxc_fork_attach_clone(int (*fn)(void *), void *arg, int flags) |
1130 | { | |
1131 | int i, r, ret; | |
1132 | pid_t pid, init_pid; | |
1133 | struct lxc_handler *handler = arg; | |
1134 | ||
1135 | pid = fork(); | |
1136 | if (pid < 0) | |
1137 | return -1; | |
1138 | ||
1139 | if (!pid) { | |
1140 | for (i = 0; i < LXC_NS_MAX; i++) { | |
1141 | if (handler->nsfd[i] < 0) | |
1142 | continue; | |
1143 | ||
1144 | ret = setns(handler->nsfd[i], 0); | |
1145 | if (ret < 0) | |
1146 | exit(EXIT_FAILURE); | |
1147 | ||
1148 | DEBUG("Inherited %s namespace", ns_info[i].proc_name); | |
1149 | } | |
1150 | ||
1151 | init_pid = lxc_clone(do_start, handler, flags); | |
1152 | ret = lxc_write_nointr(handler->data_sock[1], &init_pid, | |
1153 | sizeof(init_pid)); | |
1154 | if (ret < 0) | |
1155 | exit(EXIT_FAILURE); | |
1156 | ||
1157 | exit(EXIT_SUCCESS); | |
1158 | } | |
1159 | ||
1160 | r = lxc_read_nointr(handler->data_sock[0], &init_pid, sizeof(init_pid)); | |
1161 | ret = wait_for_pid(pid); | |
1162 | if (ret < 0 || r < 0) | |
1163 | return -1; | |
1164 | ||
1165 | return init_pid; | |
1166 | } | |
1167 | ||
480588e6 CB |
1168 | /* lxc_spawn() performs crucial setup tasks and clone()s the new process which |
1169 | * exec()s the requested container binary. | |
1170 | * Note that lxc_spawn() runs in the parent namespaces. Any operations performed | |
1171 | * right here should be double checked if they'd pose a security risk. (For | |
1172 | * example, any {u}mount() operations performed here will be reflected on the | |
1173 | * host!) | |
1174 | */ | |
74a3920a | 1175 | static int lxc_spawn(struct lxc_handler *handler) |
59eb99ba | 1176 | { |
7ab1ba02 | 1177 | int i, flags, ret; |
08dd2805 | 1178 | char pidstr[20]; |
57927bf2 CB |
1179 | bool wants_to_map_ids; |
1180 | struct lxc_list *id_map; | |
6e5fc7a5 | 1181 | const char *name = handler->name; |
28d9e29e | 1182 | const char *lxcpath = handler->lxcpath; |
fa3a5b22 | 1183 | bool cgroups_connected = false, fork_before_clone = false; |
28d9e29e | 1184 | struct lxc_conf *conf = handler->conf; |
9f30a190 | 1185 | |
28d9e29e | 1186 | id_map = &conf->id_map; |
57927bf2 | 1187 | wants_to_map_ids = !lxc_list_empty(id_map); |
2f2623ec | 1188 | |
28d9e29e CB |
1189 | for (i = 0; i < LXC_NS_MAX; i++) { |
1190 | if (!conf->inherit_ns[i]) | |
1191 | continue; | |
1192 | ||
28d9e29e CB |
1193 | handler->nsfd[i] = lxc_inherit_namespace(conf->inherit_ns[i], lxcpath, ns_info[i].proc_name); |
1194 | if (handler->nsfd[i] < 0) | |
1195 | return -1; | |
fa3a5b22 CB |
1196 | |
1197 | fork_before_clone = true; | |
28d9e29e | 1198 | } |
50e98013 | 1199 | |
3c22086f | 1200 | if (lxc_sync_init(handler)) |
9d7f9e52 | 1201 | return -1; |
0ad19a3f | 1202 | |
b9f522c5 | 1203 | ret = socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0, |
790255cf | 1204 | handler->data_sock); |
c6012571 | 1205 | if (ret < 0) { |
e8bd4e43 SH |
1206 | lxc_sync_fini(handler); |
1207 | return -1; | |
1208 | } | |
1209 | ||
f813849c | 1210 | resolve_clone_flags(handler); |
9f30a190 | 1211 | |
f813849c | 1212 | if (handler->clone_flags & CLONE_NEWNET) { |
28d9e29e | 1213 | if (!lxc_list_empty(&conf->network)) { |
26b797f3 | 1214 | |
9f30a190 MM |
1215 | /* Find gateway addresses from the link device, which is |
1216 | * no longer accessible inside the container. Do this | |
1217 | * before creating network interfaces, since goto | |
408da065 CB |
1218 | * out_delete_net does not work before lxc_clone. |
1219 | */ | |
9f30a190 | 1220 | if (lxc_find_gateway_addresses(handler)) { |
408da065 | 1221 | ERROR("Failed to find gateway addresses."); |
9f30a190 MM |
1222 | lxc_sync_fini(handler); |
1223 | return -1; | |
1224 | } | |
1225 | ||
408da065 CB |
1226 | /* That should be done before the clone because we will |
1227 | * fill the netdev index and use them in the child. | |
9f30a190 | 1228 | */ |
811ef482 | 1229 | if (lxc_create_network_priv(handler)) { |
408da065 | 1230 | ERROR("Failed to create the network."); |
9f30a190 MM |
1231 | lxc_sync_fini(handler); |
1232 | return -1; | |
1233 | } | |
19a26f82 | 1234 | } |
82d5ae15 DL |
1235 | } |
1236 | ||
d4ef7c50 | 1237 | if (!cgroup_init(handler)) { |
408da065 | 1238 | ERROR("Failed initializing cgroup support."); |
33ad9f1a CS |
1239 | goto out_delete_net; |
1240 | } | |
1241 | ||
7e4dfe0b SH |
1242 | cgroups_connected = true; |
1243 | ||
d4ef7c50 | 1244 | if (!cgroup_create(handler)) { |
408da065 | 1245 | ERROR("Failed creating cgroups."); |
47d8fb3b CS |
1246 | goto out_delete_net; |
1247 | } | |
1248 | ||
408da065 CB |
1249 | /* If the rootfs is not a blockdev, prevent the container from marking |
1250 | * it readonly. | |
1251 | * If the container is unprivileged then skip rootfs pinning. | |
0c547523 | 1252 | */ |
0ee35059 | 1253 | if (!wants_to_map_ids) { |
28d9e29e | 1254 | handler->pinfd = pin_rootfs(conf->rootfs.path); |
5e32a990 | 1255 | if (handler->pinfd == -1) |
408da065 | 1256 | INFO("Failed to pin the rootfs for container \"%s\".", handler->name); |
5e32a990 | 1257 | } |
0c547523 | 1258 | |
408da065 | 1259 | /* Create a process in a new set of namespaces. */ |
5b1e83cb | 1260 | flags = handler->clone_flags; |
408da065 CB |
1261 | if (handler->clone_flags & CLONE_NEWUSER) { |
1262 | /* If CLONE_NEWUSER and CLONE_NEWNET was requested, we need to | |
1263 | * clone a new user namespace first and only later unshare our | |
1264 | * network namespace to ensure that network devices ownership is | |
1265 | * set up correctly. | |
1266 | */ | |
5b1e83cb | 1267 | flags &= ~CLONE_NEWNET; |
408da065 | 1268 | } |
732375f5 | 1269 | |
fa3a5b22 CB |
1270 | if (fork_before_clone) |
1271 | handler->pid = lxc_fork_attach_clone(do_start, handler, flags | CLONE_PARENT); | |
1272 | else | |
1273 | handler->pid = lxc_clone(do_start, handler, flags); | |
59eb99ba | 1274 | if (handler->pid < 0) { |
408da065 | 1275 | SYSERROR("Failed to clone a new set of namespaces."); |
7fef7a06 | 1276 | goto out_delete_net; |
0ad19a3f | 1277 | } |
fa3a5b22 | 1278 | TRACE("Cloned child process %d", handler->pid); |
1bd8d726 | 1279 | |
9662e444 CB |
1280 | for (i = 0; i < LXC_NS_MAX; i++) |
1281 | if (flags & ns_info[i].clone_flag) | |
28d9e29e | 1282 | INFO("Cloned %s", ns_info[i].flag_name); |
0ad19a3f | 1283 | |
fa3a5b22 | 1284 | if (!preserve_ns(handler->nsfd, handler->clone_flags & ~CLONE_NEWNET, handler->pid)) { |
28d9e29e CB |
1285 | ERROR("Failed to preserve cloned namespaces for lxc.hook.stop"); |
1286 | goto out_delete_net; | |
1287 | } | |
b6b2b194 | 1288 | |
3c22086f CLG |
1289 | lxc_sync_fini_child(handler); |
1290 | ||
408da065 CB |
1291 | /* Map the container uids. The container became an invalid userid the |
1292 | * moment it was cloned with CLONE_NEWUSER. This call doesn't change | |
1293 | * anything immediately, but allows the container to setuid(0) (0 being | |
1294 | * mapped to something else on the host.) later to become a valid uid | |
1295 | * again. | |
1296 | */ | |
fa3a5b22 CB |
1297 | if (wants_to_map_ids) { |
1298 | if (!handler->conf->inherit_ns[LXC_NS_USER]) { | |
1299 | ret = lxc_map_ids(id_map, handler->pid); | |
1300 | if (ret < 0) { | |
1301 | ERROR("Failed to set up id mapping."); | |
1302 | goto out_delete_net; | |
1303 | } | |
1304 | } | |
5b1e83cb SH |
1305 | } |
1306 | ||
9e2bdae4 | 1307 | if (lxc_sync_wake_child(handler, LXC_SYNC_STARTUP)) |
5b1e83cb | 1308 | goto out_delete_net; |
5b1e83cb | 1309 | |
9e2bdae4 | 1310 | if (lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE)) |
5b1e83cb | 1311 | goto out_delete_net; |
0ad19a3f | 1312 | |
d4ef7c50 | 1313 | if (!cgroup_create_legacy(handler)) { |
408da065 | 1314 | ERROR("Failed to setup legacy cgroups for container \"%s\".", name); |
ae5c8b8e | 1315 | goto out_delete_net; |
33ad9f1a | 1316 | } |
9daf6f5d | 1317 | if (!cgroup_setup_limits(handler, false)) { |
408da065 | 1318 | ERROR("Failed to setup cgroup limits for container \"%s\".", name); |
6031a6e5 DE |
1319 | goto out_delete_net; |
1320 | } | |
1321 | ||
d4ef7c50 | 1322 | if (!cgroup_enter(handler)) |
7fef7a06 | 1323 | goto out_delete_net; |
218d4250 | 1324 | |
0996e18a SH |
1325 | if (!cgroup_chown(handler)) |
1326 | goto out_delete_net; | |
1327 | ||
fa3a5b22 CB |
1328 | /* Now we're ready to preserve the network namespace */ |
1329 | ret = lxc_preserve_ns(handler->pid, "net"); | |
1330 | if (ret < 0) { | |
1331 | ERROR("%s - Failed to preserve net namespace", strerror(errno)); | |
1332 | goto out_delete_net; | |
1333 | } | |
1334 | handler->nsfd[LXC_NS_NET] = ret; | |
1335 | DEBUG("Preserved net namespace via fd %d", ret); | |
1336 | ||
408da065 | 1337 | /* Create the network configuration. */ |
d5088cf2 | 1338 | if (handler->clone_flags & CLONE_NEWNET) { |
74c6e2b0 CB |
1339 | if (lxc_network_move_created_netdev_priv(handler->lxcpath, |
1340 | handler->name, | |
28d9e29e | 1341 | &conf->network, |
74c6e2b0 | 1342 | handler->pid)) { |
408da065 | 1343 | ERROR("Failed to create the configured network."); |
7fef7a06 | 1344 | goto out_delete_net; |
82d5ae15 | 1345 | } |
74c6e2b0 CB |
1346 | |
1347 | if (lxc_create_network_unpriv(handler->lxcpath, handler->name, | |
28d9e29e | 1348 | &conf->network, |
74c6e2b0 CB |
1349 | handler->pid)) { |
1350 | ERROR("Failed to create the configured network."); | |
1351 | goto out_delete_net; | |
1352 | } | |
0ad19a3f | 1353 | } |
1354 | ||
7ab1ba02 CB |
1355 | if (lxc_network_send_veth_names_to_child(handler) < 0) { |
1356 | ERROR("Failed to send veth names to child"); | |
1357 | goto out_delete_net; | |
658979c5 SH |
1358 | } |
1359 | ||
408da065 CB |
1360 | /* Tell the child to continue its initialization. We'll get |
1361 | * LXC_SYNC_CGROUP when it is ready for us to setup cgroups. | |
3c22086f CLG |
1362 | */ |
1363 | if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE)) | |
544a48a0 SH |
1364 | goto out_delete_net; |
1365 | ||
28d9e29e | 1366 | if (!lxc_list_empty(&conf->limits) && setup_resource_limits(&conf->limits, handler->pid)) { |
c6d09e15 | 1367 | ERROR("failed to setup resource limits for '%s'", name); |
84ff3af7 | 1368 | goto out_delete_net; |
c6d09e15 WB |
1369 | } |
1370 | ||
f4152036 CB |
1371 | if (lxc_sync_barrier_child(handler, LXC_SYNC_CGROUP_UNSHARE)) |
1372 | goto out_delete_net; | |
1373 | ||
9daf6f5d | 1374 | if (!cgroup_setup_limits(handler, true)) { |
408da065 | 1375 | ERROR("Failed to setup the devices cgroup for container \"%s\".", name); |
b98f7d6e | 1376 | goto out_delete_net; |
544a48a0 | 1377 | } |
f4152036 | 1378 | TRACE("Set up cgroup device limits"); |
544a48a0 | 1379 | |
73d28d42 | 1380 | cgroup_disconnect(); |
7e4dfe0b | 1381 | cgroups_connected = false; |
73d28d42 | 1382 | |
08dd2805 SH |
1383 | snprintf(pidstr, 20, "%d", handler->pid); |
1384 | if (setenv("LXC_PID", pidstr, 1)) | |
1385 | SYSERROR("Failed to set environment variable: LXC_PID=%s.", pidstr); | |
1386 | ||
1387 | /* Run any host-side start hooks */ | |
28d9e29e | 1388 | if (run_lxc_hooks(name, "start-host", conf, handler->lxcpath, NULL)) { |
08dd2805 SH |
1389 | ERROR("Failed to run lxc.hook.start-host for container \"%s\".", name); |
1390 | return -1; | |
1391 | } | |
1392 | ||
408da065 CB |
1393 | /* Tell the child to complete its initialization and wait for it to exec |
1394 | * or return an error. (The child will never return | |
08dd2805 | 1395 | * LXC_SYNC_READY_START+1. It will either close the sync pipe, causing |
408da065 CB |
1396 | * lxc_sync_barrier_child to return success, or return a different |
1397 | * value, causing us to error out). | |
544a48a0 | 1398 | */ |
08dd2805 | 1399 | if (lxc_sync_barrier_child(handler, LXC_SYNC_READY_START)) |
3c22086f | 1400 | return -1; |
0ad19a3f | 1401 | |
baed7c1b CB |
1402 | if (cgns_supported()) { |
1403 | ret = lxc_preserve_ns(handler->pid, "cgroup"); | |
1404 | if (ret < 0) { | |
1405 | ERROR("%s - Failed to preserve cgroup namespace", strerror(errno)); | |
1406 | goto out_delete_net; | |
1407 | } | |
1408 | handler->nsfd[LXC_NS_CGROUP] = ret; | |
1409 | DEBUG("Preserved cgroup namespace via fd %d", ret); | |
fa3a5b22 | 1410 | } |
fa3a5b22 | 1411 | |
790255cf CB |
1412 | if (lxc_network_recv_name_and_ifindex_from_child(handler) < 0) { |
1413 | ERROR("Failed to receive names and ifindices for network " | |
1414 | "devices from child"); | |
1415 | goto out_delete_net; | |
1416 | } | |
1417 | ||
1418 | /* Now all networks are created, network devices are moved into place, | |
1419 | * and the correct names and ifindeces in the respective namespaces have | |
1420 | * been recorded. The corresponding structs have now all been filled. So | |
1421 | * log them for debugging purposes. | |
1422 | */ | |
28d9e29e | 1423 | lxc_log_configured_netdevs(conf); |
790255cf | 1424 | |
f4152036 CB |
1425 | /* Read tty fds allocated by child. */ |
1426 | if (lxc_recv_ttys_from_child(handler) < 0) { | |
1427 | ERROR("Failed to receive tty info from child process."); | |
1428 | goto out_delete_net; | |
1429 | } | |
1430 | ||
23c53af9 | 1431 | if (handler->ops->post_start(handler, handler->data)) |
e6126dbe MN |
1432 | goto out_abort; |
1433 | ||
25c2aca5 | 1434 | if (lxc_set_state(name, handler, RUNNING)) { |
408da065 CB |
1435 | ERROR("Failed to set state for container \"%s\" to \"%s\".", name, |
1436 | lxc_state2str(RUNNING)); | |
59eb99ba | 1437 | goto out_abort; |
3f21c114 | 1438 | } |
22ebac19 | 1439 | |
3c22086f | 1440 | lxc_sync_fini(handler); |
0c547523 | 1441 | |
e6126dbe | 1442 | return 0; |
1ac470c0 | 1443 | |
7fef7a06 | 1444 | out_delete_net: |
7e4dfe0b SH |
1445 | if (cgroups_connected) |
1446 | cgroup_disconnect(); | |
1bd8d726 | 1447 | |
bb84beda CB |
1448 | if (handler->clone_flags & CLONE_NEWNET) |
1449 | lxc_delete_network(handler); | |
1bd8d726 | 1450 | |
59eb99ba DL |
1451 | out_abort: |
1452 | lxc_abort(name, handler); | |
3c22086f | 1453 | lxc_sync_fini(handler); |
5c068da9 SH |
1454 | if (handler->pinfd >= 0) { |
1455 | close(handler->pinfd); | |
1456 | handler->pinfd = -1; | |
1457 | } | |
1458 | ||
b79fcd86 | 1459 | return -1; |
59eb99ba | 1460 | } |
0ad19a3f | 1461 | |
aa460476 | 1462 | int __lxc_start(const char *name, struct lxc_handler *handler, |
507cee36 TA |
1463 | struct lxc_operations* ops, void *data, const char *lxcpath, |
1464 | bool backgrounded) | |
59eb99ba | 1465 | { |
59eb99ba | 1466 | int status; |
aa460476 | 1467 | int err = -1; |
aa460476 | 1468 | struct lxc_conf *conf = handler->conf; |
80090207 | 1469 | |
aa460476 | 1470 | if (lxc_init(name, handler) < 0) { |
408da065 | 1471 | ERROR("Failed to initialize container \"%s\".", name); |
66aeffc7 | 1472 | return -1; |
0ad19a3f | 1473 | } |
ee70bf78 CLG |
1474 | handler->ops = ops; |
1475 | handler->data = data; | |
507cee36 | 1476 | handler->backgrounded = backgrounded; |
e6126dbe | 1477 | |
76a26f55 | 1478 | if (!attach_block_device(handler->conf)) { |
408da065 | 1479 | ERROR("Failed to attach block device."); |
76a26f55 SH |
1480 | goto out_fini_nonet; |
1481 | } | |
1482 | ||
35120d9c | 1483 | if (geteuid() == 0 && !lxc_list_empty(&conf->id_map)) { |
408da065 | 1484 | /* If the backing store is a device, mount it here and now. */ |
35120d9c SH |
1485 | if (rootfs_is_blockdev(conf)) { |
1486 | if (unshare(CLONE_NEWNS) < 0) { | |
408da065 | 1487 | ERROR("Failed to unshare CLONE_NEWNS."); |
35120d9c SH |
1488 | goto out_fini_nonet; |
1489 | } | |
408da065 CB |
1490 | INFO("Unshared CLONE_NEWNS."); |
1491 | ||
6a0c909a | 1492 | remount_all_slave(); |
35120d9c | 1493 | if (do_rootfs_setup(conf, name, lxcpath) < 0) { |
408da065 | 1494 | ERROR("Error setting up rootfs mount as root before spawn."); |
35120d9c SH |
1495 | goto out_fini_nonet; |
1496 | } | |
408da065 | 1497 | INFO("Set up container rootfs as host root."); |
35120d9c SH |
1498 | } |
1499 | } | |
1500 | ||
23c53af9 | 1501 | err = lxc_spawn(handler); |
59eb99ba | 1502 | if (err) { |
408da065 | 1503 | ERROR("Failed to spawn container \"%s\".", name); |
76a26f55 | 1504 | goto out_detach_blockdev; |
0ad19a3f | 1505 | } |
a9e1109e CB |
1506 | /* close parent side of data socket */ |
1507 | close(handler->data_sock[0]); | |
1508 | handler->data_sock[0] = -1; | |
1509 | close(handler->data_sock[1]); | |
1510 | handler->data_sock[1] = -1; | |
0ad19a3f | 1511 | |
8bee8851 WB |
1512 | handler->conf->reboot = 0; |
1513 | ||
3a0f472d | 1514 | err = lxc_poll(name, handler); |
e043236e | 1515 | if (err) { |
408da065 | 1516 | ERROR("LXC mainloop exited with error: %d.", err); |
59eb99ba DL |
1517 | goto out_abort; |
1518 | } | |
0ad19a3f | 1519 | |
3a0f472d | 1520 | while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR) |
1bc5cc8c | 1521 | continue; |
e043236e | 1522 | |
408da065 CB |
1523 | /* If the child process exited but was not signaled, it didn't call |
1524 | * reboot. This should mean it was an lxc-execute which simply exited. | |
1525 | * In any case, treat it as a 'halt'. | |
8b004f07 | 1526 | */ |
d028235d | 1527 | if (WIFSIGNALED(status)) { |
8b004f07 SH |
1528 | switch(WTERMSIG(status)) { |
1529 | case SIGINT: /* halt */ | |
408da065 | 1530 | DEBUG("Container \"%s\" is halting.", name); |
8b004f07 SH |
1531 | break; |
1532 | case SIGHUP: /* reboot */ | |
408da065 | 1533 | DEBUG("Container \"%s\" is rebooting.", name); |
8b004f07 SH |
1534 | handler->conf->reboot = 1; |
1535 | break; | |
c2b9bd9e | 1536 | case SIGSYS: /* seccomp */ |
408da065 | 1537 | DEBUG("Container \"%s\" violated its seccomp policy.", name); |
c2b9bd9e | 1538 | break; |
8b004f07 | 1539 | default: |
408da065 | 1540 | DEBUG("Unknown exit status for container \"%s\" init %d.", name, WTERMSIG(status)); |
8b004f07 SH |
1541 | break; |
1542 | } | |
d028235d | 1543 | } |
828695d9 | 1544 | |
b809f232 CB |
1545 | err = lxc_restore_phys_nics_to_netns(handler); |
1546 | if (err < 0) | |
1547 | ERROR("Failed to move physical network devices back to parent " | |
1548 | "network namespace"); | |
ce5782df | 1549 | |
5c068da9 SH |
1550 | if (handler->pinfd >= 0) { |
1551 | close(handler->pinfd); | |
1552 | handler->pinfd = -1; | |
1553 | } | |
1554 | ||
1787abca | 1555 | lxc_monitor_send_exit_code(name, status, handler->lxcpath); |
3a0f472d | 1556 | err = lxc_error_set_and_log(handler->pid, status); |
1bd8d726 | 1557 | |
9d7f9e52 | 1558 | out_fini: |
bb84beda | 1559 | lxc_delete_network(handler); |
74a2b586 | 1560 | |
76a26f55 SH |
1561 | out_detach_blockdev: |
1562 | detach_block_device(handler->conf); | |
1563 | ||
74a2b586 | 1564 | out_fini_nonet: |
3a0f472d | 1565 | lxc_fini(name, handler); |
0ad19a3f | 1566 | return err; |
1567 | ||
59eb99ba | 1568 | out_abort: |
3a0f472d | 1569 | lxc_abort(name, handler); |
9d7f9e52 | 1570 | goto out_fini; |
0ad19a3f | 1571 | } |
ee70bf78 CLG |
1572 | |
1573 | struct start_args { | |
1574 | char *const *argv; | |
1575 | }; | |
1576 | ||
1577 | static int start(struct lxc_handler *handler, void* data) | |
1578 | { | |
1579 | struct start_args *arg = data; | |
1580 | ||
408da065 | 1581 | NOTICE("Exec'ing \"%s\".", arg->argv[0]); |
ee70bf78 CLG |
1582 | |
1583 | execvp(arg->argv[0], arg->argv); | |
408da065 | 1584 | SYSERROR("Failed to exec \"%s\".", arg->argv[0]); |
ee70bf78 CLG |
1585 | return 0; |
1586 | } | |
1587 | ||
1588 | static int post_start(struct lxc_handler *handler, void* data) | |
1589 | { | |
1590 | struct start_args *arg = data; | |
1591 | ||
408da065 | 1592 | NOTICE("Started \"%s\" with pid \"%d\".", arg->argv[0], handler->pid); |
ee70bf78 CLG |
1593 | return 0; |
1594 | } | |
1595 | ||
1596 | static struct lxc_operations start_ops = { | |
1597 | .start = start, | |
1598 | .post_start = post_start | |
1599 | }; | |
1600 | ||
aa460476 | 1601 | int lxc_start(const char *name, char *const argv[], struct lxc_handler *handler, |
507cee36 | 1602 | const char *lxcpath, bool backgrounded) |
ee70bf78 CLG |
1603 | { |
1604 | struct start_args start_arg = { | |
1605 | .argv = argv, | |
1606 | }; | |
1607 | ||
aa460476 | 1608 | return __lxc_start(name, handler, &start_ops, &start_arg, lxcpath, backgrounded); |
ee70bf78 | 1609 | } |
28272964 CB |
1610 | |
1611 | static void lxc_destroy_container_on_signal(struct lxc_handler *handler, | |
1612 | const char *name) | |
1613 | { | |
1614 | char destroy[MAXPATHLEN]; | |
1615 | bool bret = true; | |
1616 | int ret = 0; | |
f01f7975 | 1617 | struct lxc_container *c; |
df31363a | 1618 | if (handler->conf->rootfs.path && handler->conf->rootfs.mount) { |
790255cf | 1619 | bret = do_destroy_container(handler); |
28272964 | 1620 | if (!bret) { |
408da065 | 1621 | ERROR("Error destroying rootfs for container \"%s\".", name); |
28272964 CB |
1622 | return; |
1623 | } | |
1624 | } | |
408da065 | 1625 | INFO("Destroyed rootfs for container \"%s\".", name); |
28272964 CB |
1626 | |
1627 | ret = snprintf(destroy, MAXPATHLEN, "%s/%s", handler->lxcpath, name); | |
1628 | if (ret < 0 || ret >= MAXPATHLEN) { | |
408da065 | 1629 | ERROR("Error destroying directory for container \"%s\".", name); |
28272964 CB |
1630 | return; |
1631 | } | |
1632 | ||
f01f7975 CB |
1633 | c = lxc_container_new(name, handler->lxcpath); |
1634 | if (c) { | |
1635 | if (container_disk_lock(c)) { | |
408da065 | 1636 | INFO("Could not update lxc_snapshots file."); |
f01f7975 CB |
1637 | lxc_container_put(c); |
1638 | } else { | |
1639 | mod_all_rdeps(c, false); | |
1640 | container_disk_unlock(c); | |
1641 | lxc_container_put(c); | |
1642 | } | |
1643 | } | |
1644 | ||
d0fbc7ba | 1645 | if (!handler->am_root) |
94b0a3ac CB |
1646 | ret = userns_exec_full(handler->conf, lxc_rmdir_onedev_wrapper, |
1647 | destroy, "lxc_rmdir_onedev_wrapper"); | |
28272964 CB |
1648 | else |
1649 | ret = lxc_rmdir_onedev(destroy, NULL); | |
1650 | ||
1651 | if (ret < 0) { | |
408da065 | 1652 | ERROR("Error destroying directory for container \"%s\".", name); |
28272964 CB |
1653 | return; |
1654 | } | |
408da065 | 1655 | INFO("Destroyed directory for container \"%s\".", name); |
28272964 CB |
1656 | } |
1657 | ||
1658 | static int lxc_rmdir_onedev_wrapper(void *data) | |
1659 | { | |
1660 | char *arg = (char *) data; | |
1661 | return lxc_rmdir_onedev(arg, NULL); | |
1662 | } | |
1663 | ||
790255cf | 1664 | static bool do_destroy_container(struct lxc_handler *handler) { |
94b0a3ac CB |
1665 | int ret; |
1666 | ||
d0fbc7ba | 1667 | if (!handler->am_root) { |
94b0a3ac CB |
1668 | ret = userns_exec_full(handler->conf, storage_destroy_wrapper, |
1669 | handler->conf, "storage_destroy_wrapper"); | |
1670 | if (ret < 0) | |
d028235d | 1671 | return false; |
10bc1861 | 1672 | |
d028235d SG |
1673 | return true; |
1674 | } | |
10bc1861 | 1675 | |
790255cf | 1676 | return storage_destroy(handler->conf); |
28272964 | 1677 | } |