]>
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; | |
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; | |
162 | } | |
163 | ||
62d05d9b | 164 | return true; |
9f30a190 MM |
165 | |
166 | error: | |
4d8ac866 CB |
167 | if (errno == ENOENT) |
168 | SYSERROR("Kernel does not support attaching to %s namespaces.", ns_info[i].proc_name); | |
169 | else | |
170 | SYSERROR("Failed to open file descriptor for %s namespace: %s.", ns_info[i].proc_name, strerror(errno)); | |
9f30a190 | 171 | close_ns(ns_fd); |
62d05d9b | 172 | return false; |
9f30a190 MM |
173 | } |
174 | ||
175 | static int attach_ns(const int ns_fd[LXC_NS_MAX]) { | |
176 | int i; | |
177 | ||
178 | for (i = 0; i < LXC_NS_MAX; i++) { | |
179 | if (ns_fd[i] < 0) | |
180 | continue; | |
181 | ||
182 | if (setns(ns_fd[i], 0) != 0) | |
183 | goto error; | |
184 | } | |
185 | return 0; | |
186 | ||
187 | error: | |
408da065 | 188 | SYSERROR("Failed to attach %s namespace.", ns_info[i].proc_name); |
9f30a190 MM |
189 | return -1; |
190 | } | |
191 | ||
80090207 CLG |
192 | static int match_fd(int fd) |
193 | { | |
194 | return (fd == 0 || fd == 1 || fd == 2); | |
195 | } | |
196 | ||
47a46cf1 CB |
197 | int lxc_check_inherited(struct lxc_conf *conf, bool closeall, |
198 | int *fds_to_ignore, size_t len_fds) | |
80090207 | 199 | { |
74f96976 | 200 | struct dirent *direntp; |
80090207 | 201 | int fd, fddir; |
47a46cf1 | 202 | size_t i; |
80090207 | 203 | DIR *dir; |
80090207 | 204 | |
d2cf4c37 SH |
205 | if (conf && conf->close_all_fds) |
206 | closeall = true; | |
207 | ||
b119f362 | 208 | restart: |
80090207 CLG |
209 | dir = opendir("/proc/self/fd"); |
210 | if (!dir) { | |
13277ec4 | 211 | WARN("Failed to open directory: %s.", strerror(errno)); |
80090207 CLG |
212 | return -1; |
213 | } | |
214 | ||
215 | fddir = dirfd(dir); | |
216 | ||
74f96976 | 217 | while ((direntp = readdir(dir))) { |
80090207 CLG |
218 | if (!direntp) |
219 | break; | |
220 | ||
221 | if (!strcmp(direntp->d_name, ".")) | |
222 | continue; | |
223 | ||
224 | if (!strcmp(direntp->d_name, "..")) | |
225 | continue; | |
226 | ||
d4cff0d2 CB |
227 | if (lxc_safe_int(direntp->d_name, &fd) < 0) { |
228 | INFO("Could not parse file descriptor for: %s", direntp->d_name); | |
229 | continue; | |
230 | } | |
80090207 | 231 | |
47a46cf1 CB |
232 | for (i = 0; i < len_fds; i++) |
233 | if (fds_to_ignore[i] == fd) | |
234 | break; | |
235 | ||
236 | if (fd == fddir || fd == lxc_log_fd || | |
237 | (i < len_fds && fd == fds_to_ignore[i])) | |
80090207 CLG |
238 | continue; |
239 | ||
858377e4 SH |
240 | if (current_config && fd == current_config->logfd) |
241 | continue; | |
242 | ||
80090207 CLG |
243 | if (match_fd(fd)) |
244 | continue; | |
80090207 | 245 | |
d2cf4c37 | 246 | if (closeall) { |
b119f362 SH |
247 | close(fd); |
248 | closedir(dir); | |
28d9e29e | 249 | INFO("Closed inherited fd %d", fd); |
b119f362 SH |
250 | goto restart; |
251 | } | |
28d9e29e | 252 | WARN("Inherited fd %d", fd); |
80090207 CLG |
253 | } |
254 | ||
408da065 | 255 | /* Only enable syslog at this point to avoid the above logging function |
64c57ea1 BD |
256 | * to open a new fd and make the check_inherited function enter an |
257 | * infinite loop. | |
258 | */ | |
259 | lxc_log_enable_syslog(); | |
260 | ||
92c7f629 GK |
261 | closedir(dir); /* cannot fail */ |
262 | return 0; | |
80090207 CLG |
263 | } |
264 | ||
83ee7875 | 265 | static int setup_signal_fd(sigset_t *oldmask) |
b0a33c1e | 266 | { |
267 | sigset_t mask; | |
268 | int fd; | |
269 | ||
408da065 | 270 | /* Block everything except serious error signals. */ |
f3304a29 FW |
271 | if (sigfillset(&mask) || |
272 | sigdelset(&mask, SIGILL) || | |
273 | sigdelset(&mask, SIGSEGV) || | |
274 | sigdelset(&mask, SIGBUS) || | |
b5159817 | 275 | sigdelset(&mask, SIGWINCH) || |
f3304a29 | 276 | sigprocmask(SIG_BLOCK, &mask, oldmask)) { |
408da065 | 277 | SYSERROR("Failed to set signal mask."); |
b0a33c1e | 278 | return -1; |
279 | } | |
280 | ||
281 | fd = signalfd(-1, &mask, 0); | |
282 | if (fd < 0) { | |
408da065 | 283 | SYSERROR("Failed to create signal file descriptor."); |
b0a33c1e | 284 | return -1; |
285 | } | |
286 | ||
287 | if (fcntl(fd, F_SETFD, FD_CLOEXEC)) { | |
408da065 | 288 | SYSERROR("Failed to set FD_CLOEXEC on the signal file descriptor: %d.", fd); |
b0a33c1e | 289 | close(fd); |
290 | return -1; | |
291 | } | |
292 | ||
408da065 | 293 | DEBUG("Set SIGCHLD handler with file descriptor: %d.", fd); |
1ac470c0 | 294 | |
b0a33c1e | 295 | return fd; |
296 | } | |
297 | ||
84c92abd | 298 | static int signal_handler(int fd, uint32_t events, void *data, |
f3787121 | 299 | struct lxc_epoll_descr *descr) |
b0a33c1e | 300 | { |
15cd25fd | 301 | struct signalfd_siginfo siginfo; |
80507ee8 | 302 | siginfo_t info; |
15cd25fd | 303 | int ret; |
82d89dce | 304 | pid_t *pid = data; |
80507ee8 | 305 | bool init_died = false; |
15cd25fd DL |
306 | |
307 | ret = read(fd, &siginfo, sizeof(siginfo)); | |
308 | if (ret < 0) { | |
408da065 | 309 | ERROR("Failed to read signal info from signal file descriptor: %d.", fd); |
15cd25fd DL |
310 | return -1; |
311 | } | |
312 | ||
313 | if (ret != sizeof(siginfo)) { | |
408da065 | 314 | ERROR("Unexpected size for siginfo struct."); |
15cd25fd DL |
315 | return -1; |
316 | } | |
317 | ||
408da065 | 318 | /* Check whether init is running. */ |
80507ee8 SH |
319 | info.si_pid = 0; |
320 | ret = waitid(P_PID, *pid, &info, WEXITED | WNOWAIT | WNOHANG); | |
408da065 | 321 | if (ret == 0 && info.si_pid == *pid) |
80507ee8 | 322 | init_died = true; |
80507ee8 | 323 | |
f3304a29 FW |
324 | if (siginfo.ssi_signo != SIGCHLD) { |
325 | kill(*pid, siginfo.ssi_signo); | |
408da065 | 326 | INFO("Forwarded signal %d to pid %d.", siginfo.ssi_signo, *pid); |
80507ee8 | 327 | return init_died ? 1 : 0; |
f3304a29 FW |
328 | } |
329 | ||
408da065 CB |
330 | if (siginfo.ssi_code == CLD_STOPPED) { |
331 | INFO("Container init process was stopped."); | |
332 | return init_died ? 1 : 0; | |
333 | } else if (siginfo.ssi_code == CLD_CONTINUED) { | |
334 | INFO("Container init process was continued."); | |
80507ee8 | 335 | return init_died ? 1 : 0; |
15cd25fd | 336 | } |
1ac470c0 | 337 | |
408da065 CB |
338 | /* More robustness, protect ourself from a SIGCHLD sent |
339 | * by a process different from the container init. | |
82d89dce DL |
340 | */ |
341 | if (siginfo.ssi_pid != *pid) { | |
af4c0f05 | 342 | NOTICE("Received SIGCHLD from pid %d instead of container init %d.", siginfo.ssi_pid, *pid); |
80507ee8 | 343 | return init_died ? 1 : 0; |
82d89dce DL |
344 | } |
345 | ||
408da065 | 346 | DEBUG("Container init process %d exited.", *pid); |
b0a33c1e | 347 | return 1; |
348 | } | |
349 | ||
5e5576a4 CB |
350 | static int lxc_serve_state_clients(const char *name, |
351 | struct lxc_handler *handler, | |
352 | lxc_state_t state) | |
66aeffc7 | 353 | { |
dbc9832d CB |
354 | ssize_t ret; |
355 | struct lxc_list *cur, *next; | |
356 | struct state_client *client; | |
357 | struct lxc_msg msg = {.type = lxc_msg_state, .value = state}; | |
358 | ||
359 | process_lock(); | |
5e5576a4 | 360 | |
dbc9832d | 361 | /* Only set state under process lock held so that we don't cause |
54446942 | 362 | * lxc_cmd_add_state_client() to miss a state. |
dbc9832d | 363 | */ |
66aeffc7 | 364 | handler->state = state; |
dbc9832d CB |
365 | TRACE("set container state to %s", lxc_state2str(state)); |
366 | ||
367 | if (lxc_list_empty(&handler->state_clients)) { | |
368 | TRACE("no state clients registered"); | |
369 | process_unlock(); | |
65ee1673 | 370 | lxc_monitor_send_state(name, state, handler->lxcpath); |
dbc9832d CB |
371 | return 0; |
372 | } | |
373 | ||
374 | strncpy(msg.name, name, sizeof(msg.name)); | |
375 | msg.name[sizeof(msg.name) - 1] = 0; | |
376 | ||
377 | lxc_list_for_each_safe(cur, &handler->state_clients, next) { | |
378 | client = cur->elem; | |
379 | ||
380 | if (!client->states[state]) { | |
381 | TRACE("state %s not registered for state client %d", | |
382 | lxc_state2str(state), client->clientfd); | |
383 | continue; | |
384 | } | |
385 | ||
386 | TRACE("sending state %s to state client %d", | |
387 | lxc_state2str(state), client->clientfd); | |
388 | ||
389 | again: | |
390 | ret = send(client->clientfd, &msg, sizeof(msg), 0); | |
ee8377bd CB |
391 | if (ret <= 0) { |
392 | if (errno == EINTR) { | |
393 | TRACE("Caught EINTR; retrying"); | |
dbc9832d | 394 | goto again; |
ee8377bd | 395 | } |
dbc9832d CB |
396 | |
397 | ERROR("failed to send message to client"); | |
398 | } | |
399 | ||
400 | /* kick client from list */ | |
401 | close(client->clientfd); | |
402 | lxc_list_del(cur); | |
403 | free(cur->elem); | |
404 | free(cur); | |
405 | } | |
406 | process_unlock(); | |
407 | ||
5e5576a4 CB |
408 | return 0; |
409 | } | |
410 | ||
411 | static int lxc_serve_state_socket_pair(const char *name, | |
412 | struct lxc_handler *handler, | |
413 | lxc_state_t state) | |
414 | { | |
415 | ssize_t ret; | |
416 | ||
417 | if (!handler->backgrounded || | |
418 | handler->state_socket_pair[1] < 0 || | |
419 | state == STARTING) | |
420 | return 0; | |
421 | ||
422 | /* Close read end of the socket pair. */ | |
423 | close(handler->state_socket_pair[0]); | |
424 | handler->state_socket_pair[0] = -1; | |
425 | ||
ee8377bd | 426 | again: |
5e5576a4 CB |
427 | ret = lxc_abstract_unix_send_credential(handler->state_socket_pair[1], |
428 | &(int){state}, sizeof(int)); | |
ee8377bd CB |
429 | if (ret != sizeof(int)) { |
430 | if (errno == EINTR) | |
431 | goto again; | |
5e5576a4 CB |
432 | SYSERROR("Failed to send state to %d", |
433 | handler->state_socket_pair[1]); | |
ee8377bd CB |
434 | return -1; |
435 | } | |
5e5576a4 CB |
436 | |
437 | TRACE("Sent container state \"%s\" to %d", lxc_state2str(state), | |
438 | handler->state_socket_pair[1]); | |
439 | ||
440 | /* Close write end of the socket pair. */ | |
441 | close(handler->state_socket_pair[1]); | |
442 | handler->state_socket_pair[1] = -1; | |
443 | ||
444 | return 0; | |
445 | } | |
446 | ||
447 | int lxc_set_state(const char *name, struct lxc_handler *handler, | |
448 | lxc_state_t state) | |
449 | { | |
450 | int ret; | |
451 | ||
452 | ret = lxc_serve_state_socket_pair(name, handler, state); | |
453 | if (ret < 0) { | |
454 | ERROR("Failed to synchronize via anonymous pair of unix sockets"); | |
455 | return -1; | |
456 | } | |
457 | ||
458 | ret = lxc_serve_state_clients(name, handler, state); | |
459 | if (ret < 0) | |
460 | return -1; | |
461 | ||
dbc9832d CB |
462 | /* This function will try to connect to the legacy lxc-monitord state |
463 | * server and only exists for backwards compatibility. | |
464 | */ | |
9123e471 | 465 | lxc_monitor_send_state(name, state, handler->lxcpath); |
dbc9832d | 466 | |
66aeffc7 DL |
467 | return 0; |
468 | } | |
469 | ||
735f2c6e | 470 | int lxc_poll(const char *name, struct lxc_handler *handler) |
b0a33c1e | 471 | { |
ca5f7926 DL |
472 | int sigfd = handler->sigfd; |
473 | int pid = handler->pid; | |
b0a33c1e | 474 | struct lxc_epoll_descr descr; |
475 | ||
a9e61274 | 476 | if (lxc_mainloop_open(&descr)) { |
408da065 | 477 | ERROR("Failed to create LXC mainloop."); |
50c8bf05 | 478 | goto out_sigfd; |
b0a33c1e | 479 | } |
480 | ||
83ee7875 | 481 | if (lxc_mainloop_add_handler(&descr, sigfd, signal_handler, &pid)) { |
408da065 | 482 | ERROR("Failed to add signal handler with file descriptor %d to LXC mainloop.", sigfd); |
b0a33c1e | 483 | goto out_mainloop_open; |
484 | } | |
485 | ||
da41561c | 486 | if (lxc_console_mainloop_add(&descr, handler->conf)) { |
408da065 | 487 | ERROR("Failed to add console handler to LXC mainloop."); |
63376d7d DL |
488 | goto out_mainloop_open; |
489 | } | |
490 | ||
ef6e34ee | 491 | if (lxc_cmd_mainloop_add(name, &descr, handler)) { |
408da065 | 492 | ERROR("Failed to add command handler to LXC mainloop."); |
96fa1ff0 | 493 | goto out_mainloop_open; |
563f2f2c DL |
494 | } |
495 | ||
3f903c04 | 496 | TRACE("lxc mainloop is ready"); |
b0a33c1e | 497 | |
e51d4895 | 498 | return lxc_mainloop(&descr, -1); |
b0a33c1e | 499 | |
500 | out_mainloop_open: | |
501 | lxc_mainloop_close(&descr); | |
408da065 | 502 | |
b0a33c1e | 503 | out_sigfd: |
504 | close(sigfd); | |
408da065 | 505 | |
c3e13372 | 506 | return -1; |
b0a33c1e | 507 | } |
508 | ||
f2e07cb6 CB |
509 | void lxc_free_handler(struct lxc_handler *handler) |
510 | { | |
511 | if (handler->conf && handler->conf->maincmd_fd) | |
512 | close(handler->conf->maincmd_fd); | |
513 | ||
5e5576a4 CB |
514 | if (handler->state_socket_pair[0] >= 0) |
515 | close(handler->state_socket_pair[0]); | |
516 | ||
517 | if (handler->state_socket_pair[1] >= 0) | |
518 | close(handler->state_socket_pair[1]); | |
519 | ||
f2e07cb6 CB |
520 | handler->conf = NULL; |
521 | free(handler); | |
522 | } | |
523 | ||
aa460476 | 524 | struct lxc_handler *lxc_init_handler(const char *name, struct lxc_conf *conf, |
5e5576a4 | 525 | const char *lxcpath, bool daemonize) |
59eb99ba | 526 | { |
5e5576a4 | 527 | int i, ret; |
3a0f472d DL |
528 | struct lxc_handler *handler; |
529 | ||
530 | handler = malloc(sizeof(*handler)); | |
aa460476 CB |
531 | if (!handler) { |
532 | ERROR("failed to allocate memory"); | |
3a0f472d | 533 | return NULL; |
aa460476 | 534 | } |
59eb99ba DL |
535 | |
536 | memset(handler, 0, sizeof(*handler)); | |
537 | ||
790255cf CB |
538 | /* Note that am_unpriv() checks the effective uid. We probably don't |
539 | * care if we are real root only if we are running as root so this | |
540 | * should be fine. | |
541 | */ | |
d0fbc7ba | 542 | handler->am_root = !am_unpriv(); |
c6012571 | 543 | handler->data_sock[0] = handler->data_sock[1] = -1; |
fae349da | 544 | handler->conf = conf; |
9123e471 | 545 | handler->lxcpath = lxcpath; |
5c068da9 | 546 | handler->pinfd = -1; |
5e5576a4 | 547 | handler->state_socket_pair[0] = handler->state_socket_pair[1] = -1; |
dbc9832d | 548 | lxc_list_init(&handler->state_clients); |
fae349da | 549 | |
b6b2b194 WB |
550 | for (i = 0; i < LXC_NS_MAX; i++) |
551 | handler->nsfd[i] = -1; | |
552 | ||
f0ecc19d | 553 | handler->name = name; |
3bdf52d7 | 554 | |
5e5576a4 CB |
555 | if (daemonize && !handler->conf->reboot) { |
556 | /* Create socketpair() to synchronize on daemonized startup. | |
557 | * When the container reboots we don't need to synchronize again | |
558 | * currently so don't open another socketpair(). | |
559 | */ | |
560 | ret = socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0, | |
561 | handler->state_socket_pair); | |
562 | if (ret < 0) { | |
563 | ERROR("Failed to create anonymous pair of unix sockets"); | |
564 | goto on_error; | |
565 | } | |
566 | TRACE("Created anonymous pair {%d,%d} of unix sockets", | |
567 | handler->state_socket_pair[0], | |
568 | handler->state_socket_pair[1]); | |
569 | } | |
570 | ||
aa460476 CB |
571 | if (lxc_cmd_init(name, handler, lxcpath)) { |
572 | ERROR("failed to set up command socket"); | |
f2e07cb6 | 573 | goto on_error; |
aa460476 CB |
574 | } |
575 | ||
576 | TRACE("unix domain socket %d for command server is ready", | |
577 | handler->conf->maincmd_fd); | |
578 | ||
579 | return handler; | |
580 | ||
f2e07cb6 CB |
581 | on_error: |
582 | lxc_free_handler(handler); | |
aa460476 CB |
583 | |
584 | return NULL; | |
585 | } | |
586 | ||
587 | int lxc_init(const char *name, struct lxc_handler *handler) | |
588 | { | |
589 | struct lxc_conf *conf = handler->conf; | |
590 | ||
591 | lsm_init(); | |
592 | TRACE("initialized LSM"); | |
d2e30e99 | 593 | |
8f2c3a70 | 594 | if (lxc_read_seccomp_config(conf) != 0) { |
408da065 | 595 | ERROR("Failed loading seccomp policy."); |
d2e30e99 | 596 | goto out_close_maincmd_fd; |
8f2c3a70 | 597 | } |
dbc9832d | 598 | TRACE("read seccomp policy"); |
8f2c3a70 | 599 | |
408da065 | 600 | /* Begin by setting the state to STARTING. */ |
25c2aca5 | 601 | if (lxc_set_state(name, handler, STARTING)) { |
408da065 | 602 | ERROR("Failed to set state for container \"%s\" to \"%s\".", name, lxc_state2str(STARTING)); |
051151de | 603 | goto out_close_maincmd_fd; |
0ad19a3f | 604 | } |
dbc9832d | 605 | TRACE("set container state to \"STARTING\""); |
0ad19a3f | 606 | |
408da065 CB |
607 | /* Start of environment variable setup for hooks. */ |
608 | if (name && setenv("LXC_NAME", name, 1)) | |
609 | SYSERROR("Failed to set environment variable: LXC_NAME=%s.", name); | |
610 | ||
611 | if (conf->rcfile && setenv("LXC_CONFIG_FILE", conf->rcfile, 1)) | |
612 | SYSERROR("Failed to set environment variable: LXC_CONFIG_FILE=%s.", conf->rcfile); | |
613 | ||
614 | if (conf->rootfs.mount && setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1)) | |
615 | SYSERROR("Failed to set environment variable: LXC_ROOTFS_MOUNT=%s.", conf->rootfs.mount); | |
616 | ||
617 | if (conf->rootfs.path && setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1)) | |
618 | SYSERROR("Failed to set environment variable: LXC_ROOTFS_PATH=%s.", conf->rootfs.path); | |
619 | ||
620 | if (conf->console.path && setenv("LXC_CONSOLE", conf->console.path, 1)) | |
621 | SYSERROR("Failed to set environment variable: LXC_CONSOLE=%s.", conf->console.path); | |
622 | ||
623 | if (conf->console.log_path && setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1)) | |
624 | SYSERROR("Failed to set environment variable: LXC_CONSOLE_LOGPATH=%s.", conf->console.log_path); | |
625 | ||
626 | if (setenv("LXC_CGNS_AWARE", "1", 1)) | |
627 | SYSERROR("Failed to set environment variable LXC_CGNS_AWARE=1."); | |
b8f88d9b CB |
628 | |
629 | if (setenv("LXC_LOG_LEVEL", lxc_log_priority_to_string(handler->conf->loglevel), 1)) | |
630 | SYSERROR("Failed to set environment variable LXC_CGNS_AWARE=1."); | |
408da065 | 631 | /* End of environment variable setup for hooks. */ |
f7bee6c6 | 632 | |
dbc9832d CB |
633 | TRACE("set environment variables"); |
634 | ||
283678ed | 635 | if (run_lxc_hooks(name, "pre-start", conf, handler->lxcpath, NULL)) { |
408da065 | 636 | ERROR("Failed to run lxc.hook.pre-start for container \"%s\".", name); |
773fb9ca SH |
637 | goto out_aborting; |
638 | } | |
dbc9832d | 639 | TRACE("ran pre-start hooks"); |
26ddeedd | 640 | |
408da065 CB |
641 | /* The signal fd has to be created before forking otherwise if the child |
642 | * process exits before we setup the signal fd, the event will be lost | |
643 | * and the command will be stuck. | |
644 | */ | |
83ee7875 | 645 | handler->sigfd = setup_signal_fd(&handler->oldmask); |
59eb99ba | 646 | if (handler->sigfd < 0) { |
408da065 | 647 | ERROR("Failed to setup SIGCHLD fd handler."); |
b5159817 DE |
648 | goto out_delete_tty; |
649 | } | |
dbc9832d | 650 | TRACE("set up signal fd"); |
b5159817 | 651 | |
408da065 | 652 | /* Do this after setting up signals since it might unblock SIGWINCH. */ |
b5159817 | 653 | if (lxc_console_create(conf)) { |
408da065 | 654 | ERROR("Failed to create console for container \"%s\".", name); |
b5159817 | 655 | goto out_restore_sigmask; |
b0a33c1e | 656 | } |
dbc9832d | 657 | TRACE("created console"); |
b0a33c1e | 658 | |
54117de5 | 659 | if (lxc_ttys_shift_ids(conf) < 0) { |
408da065 | 660 | ERROR("Failed to shift tty into container."); |
c4d10a05 SH |
661 | goto out_restore_sigmask; |
662 | } | |
dbc9832d | 663 | TRACE("shifted tty ids"); |
c4d10a05 | 664 | |
dbc9832d | 665 | INFO("container \"%s\" is initialized", name); |
aa460476 | 666 | return 0; |
59eb99ba | 667 | |
b5159817 DE |
668 | out_restore_sigmask: |
669 | sigprocmask(SIG_SETMASK, &handler->oldmask, NULL); | |
59eb99ba | 670 | out_delete_tty: |
fae349da | 671 | lxc_delete_tty(&conf->tty_info); |
59eb99ba | 672 | out_aborting: |
25c2aca5 | 673 | lxc_set_state(name, handler, ABORTING); |
d2e30e99 DE |
674 | out_close_maincmd_fd: |
675 | close(conf->maincmd_fd); | |
676 | conf->maincmd_fd = -1; | |
aa460476 | 677 | return -1; |
59eb99ba DL |
678 | } |
679 | ||
3b72c4a0 | 680 | void lxc_fini(const char *name, struct lxc_handler *handler) |
59eb99ba | 681 | { |
b3286b62 | 682 | int i, rc; |
dbc9832d | 683 | struct lxc_list *cur, *next; |
b3286b62 | 684 | pid_t self = getpid(); |
dbc9832d | 685 | char *namespaces[LXC_NS_MAX + 1]; |
b3286b62 | 686 | size_t namespace_count = 0; |
b6b2b194 | 687 | |
408da065 CB |
688 | /* The STOPPING state is there for future cleanup code which can take |
689 | * awhile. | |
59eb99ba | 690 | */ |
25c2aca5 | 691 | lxc_set_state(name, handler, STOPPING); |
b6b2b194 | 692 | |
b3286b62 WB |
693 | for (i = 0; i < LXC_NS_MAX; i++) { |
694 | if (handler->nsfd[i] != -1) { | |
695 | rc = asprintf(&namespaces[namespace_count], "%s:/proc/%d/fd/%d", | |
696 | ns_info[i].proc_name, self, handler->nsfd[i]); | |
697 | if (rc == -1) { | |
408da065 | 698 | SYSERROR("Failed to allocate memory."); |
b3286b62 WB |
699 | break; |
700 | } | |
701 | ++namespace_count; | |
702 | } | |
703 | } | |
704 | namespaces[namespace_count] = NULL; | |
c154af98 | 705 | |
408da065 CB |
706 | if (handler->conf->reboot && setenv("LXC_TARGET", "reboot", 1)) |
707 | SYSERROR("Failed to set environment variable: LXC_TARGET=reboot."); | |
708 | ||
709 | if (!handler->conf->reboot && setenv("LXC_TARGET", "stop", 1)) | |
710 | SYSERROR("Failed to set environment variable: LXC_TARGET=stop."); | |
c154af98 | 711 | |
b3286b62 | 712 | if (run_lxc_hooks(name, "stop", handler->conf, handler->lxcpath, namespaces)) |
408da065 | 713 | ERROR("Failed to run lxc.hook.stop for container \"%s\".", name); |
c154af98 | 714 | |
b3286b62 WB |
715 | while (namespace_count--) |
716 | free(namespaces[namespace_count]); | |
28d9e29e | 717 | |
b6b2b194 | 718 | for (i = 0; i < LXC_NS_MAX; i++) { |
28d9e29e CB |
719 | if (handler->nsfd[i] < 0) |
720 | continue; | |
738d0deb | 721 | |
28d9e29e CB |
722 | close(handler->nsfd[i]); |
723 | handler->nsfd[i] = -1; | |
738d0deb CB |
724 | } |
725 | ||
4288b79f CB |
726 | cgroup_destroy(handler); |
727 | ||
25c2aca5 | 728 | lxc_set_state(name, handler, STOPPED); |
59eb99ba | 729 | |
4288b79f CB |
730 | /* close command socket */ |
731 | close(handler->conf->maincmd_fd); | |
732 | handler->conf->maincmd_fd = -1; | |
733 | ||
f3787121 | 734 | if (run_lxc_hooks(name, "post-stop", handler->conf, handler->lxcpath, NULL)) { |
408da065 | 735 | ERROR("Failed to run lxc.hook.post-stop for container \"%s\".", name); |
f3787121 CB |
736 | if (handler->conf->reboot) { |
737 | WARN("Container will be stopped instead of rebooted."); | |
738 | handler->conf->reboot = 0; | |
408da065 CB |
739 | if (setenv("LXC_TARGET", "stop", 1)) |
740 | WARN("Failed to set environment variable: LXC_TARGET=stop."); | |
f3787121 CB |
741 | } |
742 | } | |
26ddeedd | 743 | |
408da065 | 744 | /* Reset mask set by setup_signal_fd. */ |
8f64a3f6 | 745 | if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) |
408da065 | 746 | WARN("Failed to restore signal mask."); |
8f64a3f6 | 747 | |
b5159817 | 748 | lxc_console_delete(&handler->conf->console); |
b2431939 | 749 | lxc_delete_tty(&handler->conf->tty_info); |
dbc9832d | 750 | |
dbc9832d CB |
751 | /* The command socket is now closed, no more state clients can register |
752 | * themselves from now on. So free the list of state clients. | |
753 | */ | |
754 | lxc_list_for_each_safe(cur, &handler->state_clients, next) { | |
755 | struct state_client *client = cur->elem; | |
756 | /* close state client socket */ | |
757 | close(client->clientfd); | |
758 | lxc_list_del(cur); | |
759 | free(cur->elem); | |
760 | free(cur); | |
761 | } | |
762 | ||
c6012571 CB |
763 | if (handler->data_sock[0] != -1) { |
764 | close(handler->data_sock[0]); | |
765 | close(handler->data_sock[1]); | |
e8bd4e43 | 766 | } |
2c5f2ede CB |
767 | |
768 | if (handler->conf->ephemeral == 1 && handler->conf->reboot != 1) | |
28272964 | 769 | lxc_destroy_container_on_signal(handler, name); |
2c5f2ede | 770 | |
b2431939 | 771 | free(handler); |
59eb99ba DL |
772 | } |
773 | ||
735f2c6e | 774 | void lxc_abort(const char *name, struct lxc_handler *handler) |
59eb99ba | 775 | { |
73e608b2 SH |
776 | int ret, status; |
777 | ||
25c2aca5 | 778 | lxc_set_state(name, handler, ABORTING); |
7d9fb3e9 DL |
779 | if (handler->pid > 0) |
780 | kill(handler->pid, SIGKILL); | |
408da065 CB |
781 | while ((ret = waitpid(-1, &status, 0)) > 0) { |
782 | ; | |
783 | } | |
59eb99ba DL |
784 | } |
785 | ||
ffe1e01a | 786 | static int do_start(void *data) |
50e98013 | 787 | { |
7c661726 | 788 | struct lxc_list *iterator; |
23c53af9 | 789 | struct lxc_handler *handler = data; |
7a55c157 TA |
790 | int devnull_fd = -1, ret; |
791 | char path[PATH_MAX]; | |
50e98013 DL |
792 | |
793 | if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) { | |
408da065 | 794 | SYSERROR("Failed to set signal mask."); |
9d7f9e52 | 795 | return -1; |
50e98013 DL |
796 | } |
797 | ||
408da065 CB |
798 | /* This prctl must be before the synchro, so if the parent dies before |
799 | * we set the parent death signal, we will detect its death with the | |
800 | * synchro right after, otherwise we have a window where the parent can | |
801 | * exit before we set the pdeath signal leading to a unsupervized | |
802 | * container. | |
743ecd2e DL |
803 | */ |
804 | if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) { | |
408da065 | 805 | SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL."); |
743ecd2e DL |
806 | return -1; |
807 | } | |
808 | ||
3c22086f | 809 | lxc_sync_fini_parent(handler); |
50e98013 | 810 | |
408da065 | 811 | /* Don't leak the pinfd to the container. */ |
025ed0f3 | 812 | if (handler->pinfd >= 0) { |
0d03360a | 813 | close(handler->pinfd); |
025ed0f3 | 814 | } |
2b0e17e4 | 815 | |
5b1e83cb SH |
816 | if (lxc_sync_wait_parent(handler, LXC_SYNC_STARTUP)) |
817 | return -1; | |
818 | ||
408da065 CB |
819 | /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See |
820 | * https://github.com/lxc/lxd/issues/1978. | |
821 | */ | |
5b1e83cb | 822 | if ((handler->clone_flags & (CLONE_NEWNET | CLONE_NEWUSER)) == |
408da065 | 823 | (CLONE_NEWNET | CLONE_NEWUSER)) { |
5b1e83cb SH |
824 | ret = unshare(CLONE_NEWNET); |
825 | if (ret < 0) { | |
408da065 | 826 | SYSERROR("Failed to unshare CLONE_NEWNET."); |
5b1e83cb SH |
827 | goto out_warn_father; |
828 | } | |
408da065 | 829 | INFO("Unshared CLONE_NEWNET."); |
5b1e83cb SH |
830 | } |
831 | ||
408da065 CB |
832 | /* Tell the parent task it can begin to configure the container and wait |
833 | * for it to finish. | |
3c22086f CLG |
834 | */ |
835 | if (lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE)) | |
9d7f9e52 | 836 | return -1; |
50e98013 | 837 | |
7ab1ba02 CB |
838 | if (lxc_network_recv_veth_names_from_parent(handler) < 0) { |
839 | ERROR("Failed to receive veth names from parent"); | |
658979c5 | 840 | goto out_warn_father; |
7ab1ba02 | 841 | } |
658979c5 | 842 | |
408da065 | 843 | /* If we are in a new user namespace, become root there to have |
d08f8d2f | 844 | * privilege over our namespace. |
f6d3e3e4 SH |
845 | */ |
846 | if (!lxc_list_empty(&handler->conf->id_map)) { | |
d08f8d2f | 847 | if (lxc_switch_uid_gid(0, 0) < 0) |
f6d3e3e4 | 848 | goto out_warn_father; |
d08f8d2f CB |
849 | |
850 | /* Drop groups only after we switched to a valid gid in the new | |
851 | * user namespace. | |
852 | */ | |
853 | if (lxc_setgroups(0, NULL) < 0) | |
c476bdce | 854 | goto out_warn_father; |
f6d3e3e4 SH |
855 | } |
856 | ||
99b71824 | 857 | if (access(handler->lxcpath, X_OK)) { |
c8154066 SH |
858 | print_top_failing_dir(handler->lxcpath); |
859 | goto out_warn_father; | |
860 | } | |
861 | ||
a91cde21 | 862 | ret = snprintf(path, sizeof(path), "%s/dev/null", handler->conf->rootfs.mount); |
408da065 | 863 | if (ret < 0 || ret >= sizeof(path)) |
7a55c157 | 864 | goto out_warn_father; |
7a55c157 TA |
865 | |
866 | /* In order to checkpoint restore, we need to have everything in the | |
867 | * same mount namespace. However, some containers may not have a | |
868 | * reasonable /dev (in particular, they may not have /dev/null), so we | |
869 | * can't set init's std fds to /dev/null by opening it from inside the | |
870 | * container. | |
871 | * | |
872 | * If that's the case, fall back to using the host's /dev/null. This | |
873 | * means that migration won't work, but at least we won't spew output | |
874 | * where it isn't wanted. | |
875 | */ | |
876 | if (handler->backgrounded && !handler->conf->autodev && access(path, F_OK) < 0) { | |
c44de748 AM |
877 | devnull_fd = open_devnull(); |
878 | ||
879 | if (devnull_fd < 0) | |
880 | goto out_warn_father; | |
408da065 CB |
881 | WARN("Using /dev/null from the host for container init's " |
882 | "standard file descriptors. Migration will not work."); | |
c44de748 AM |
883 | } |
884 | ||
408da065 | 885 | /* Ask father to setup cgroups and wait for him to finish. */ |
544a48a0 | 886 | if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP)) |
c44de748 | 887 | goto out_error; |
544a48a0 | 888 | |
deefdf8a CB |
889 | /* Unshare cgroup namespace after we have setup our cgroups. If we do it |
890 | * earlier we end up with a wrong view of /proc/self/cgroup. For | |
891 | * example, assume we unshare(CLONE_NEWCGROUP) first, and then create | |
892 | * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then | |
893 | * /proc/self/cgroup would show us: | |
894 | * | |
895 | * 8:cpuset:/lxc/c | |
896 | * | |
897 | * whereas it should actually show | |
898 | * | |
899 | * 8:cpuset:/ | |
900 | */ | |
901 | if (cgns_supported()) { | |
902 | if (unshare(CLONE_NEWCGROUP) < 0) { | |
903 | INFO("Failed to unshare CLONE_NEWCGROUP."); | |
904 | goto out_warn_father; | |
905 | } | |
906 | INFO("Unshared CLONE_NEWCGROUP."); | |
907 | } | |
908 | ||
149857af CB |
909 | /* Add the requested environment variables to the current environment to |
910 | * allow them to be used by the various hooks, such as the start hook | |
911 | * above. | |
98ff08ca | 912 | */ |
98ff08ca CB |
913 | lxc_list_for_each(iterator, &handler->conf->environment) { |
914 | if (putenv((char *)iterator->elem)) { | |
915 | SYSERROR("Failed to set environment variable: %s.", (char *)iterator->elem); | |
916 | goto out_warn_father; | |
917 | } | |
918 | } | |
919 | ||
f4152036 | 920 | /* Setup the container, ip, names, utsname, ... */ |
7f135597 | 921 | ret = lxc_setup_child(handler); |
7729f8e5 CB |
922 | close(handler->data_sock[0]); |
923 | close(handler->data_sock[1]); | |
924 | if (ret < 0) { | |
f4152036 CB |
925 | ERROR("Failed to setup container \"%s\".", handler->name); |
926 | goto out_warn_father; | |
927 | } | |
928 | ||
408da065 | 929 | /* Set the label to change to when we exec(2) the container's init. */ |
7aff4f43 | 930 | if (lsm_process_label_set(NULL, handler->conf, 1, 1) < 0) |
e075f5d9 | 931 | goto out_warn_father; |
5112cd70 | 932 | |
029cdff5 | 933 | /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it |
408da065 CB |
934 | * before we aren't allowed anymore. |
935 | */ | |
029cdff5 CB |
936 | if (handler->conf->no_new_privs) { |
937 | if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) { | |
938 | SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges."); | |
939 | goto out_warn_father; | |
940 | } | |
941 | DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges."); | |
942 | } | |
943 | ||
0d9acb99 DE |
944 | /* Some init's such as busybox will set sane tty settings on stdin, |
945 | * stdout, stderr which it thinks is the console. We already set them | |
946 | * the way we wanted on the real terminal, and we want init to do its | |
408da065 CB |
947 | * setup on its console ie. the pty allocated in lxc_console_create() so |
948 | * make sure that that pty is stdin,stdout,stderr. | |
0d9acb99 | 949 | */ |
c5b93afb LF |
950 | if (handler->conf->console.slave >= 0) |
951 | if (set_stdfds(handler->conf->console.slave) < 0) { | |
952 | ERROR("Failed to redirect std{in,out,err} to pty file " | |
953 | "descriptor %d", | |
954 | handler->conf->console.slave); | |
955 | goto out_warn_father; | |
956 | } | |
0d9acb99 | 957 | |
408da065 | 958 | /* If we mounted a temporary proc, then unmount it now. */ |
5112cd70 | 959 | tmp_proc_unmount(handler->conf); |
e075f5d9 | 960 | |
8f2c3a70 SH |
961 | if (lxc_seccomp_load(handler->conf) != 0) |
962 | goto out_warn_father; | |
963 | ||
283678ed | 964 | if (run_lxc_hooks(handler->name, "start", handler->conf, handler->lxcpath, NULL)) { |
408da065 | 965 | ERROR("Failed to run lxc.hook.start for container \"%s\".", handler->name); |
773fb9ca SH |
966 | goto out_warn_father; |
967 | } | |
fc25b815 | 968 | |
773fb9ca | 969 | close(handler->sigfd); |
26ddeedd | 970 | |
7a55c157 TA |
971 | if (devnull_fd < 0) { |
972 | devnull_fd = open_devnull(); | |
973 | ||
974 | if (devnull_fd < 0) | |
975 | goto out_warn_father; | |
976 | } | |
977 | ||
c5b93afb LF |
978 | if (handler->conf->console.slave < 0 && handler->backgrounded) |
979 | if (set_stdfds(devnull_fd) < 0) { | |
980 | ERROR("Failed to redirect std{in,out,err} to " | |
981 | "\"/dev/null\""); | |
982 | goto out_warn_father; | |
983 | } | |
507cee36 | 984 | |
c44de748 AM |
985 | if (devnull_fd >= 0) { |
986 | close(devnull_fd); | |
987 | devnull_fd = -1; | |
988 | } | |
989 | ||
8c9a7665 TA |
990 | setsid(); |
991 | ||
f4152036 CB |
992 | if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP_LIMITS)) |
993 | goto out_warn_father; | |
994 | ||
149857af CB |
995 | /* Reset the environment variables the user requested in a clear |
996 | * environment. | |
997 | */ | |
998 | if (clearenv()) { | |
999 | SYSERROR("Failed to clear environment."); | |
1000 | /* Don't error out though. */ | |
1001 | } | |
1002 | ||
1003 | lxc_list_for_each(iterator, &handler->conf->environment) { | |
1004 | if (putenv((char *)iterator->elem)) { | |
1005 | SYSERROR("Failed to set environment variable: %s.", (char *)iterator->elem); | |
1006 | goto out_warn_father; | |
1007 | } | |
1008 | } | |
1009 | ||
98ff08ca CB |
1010 | if (putenv("container=lxc")) { |
1011 | SYSERROR("Failed to set environment variable: container=lxc."); | |
1012 | goto out_warn_father; | |
1013 | } | |
1014 | ||
1015 | if (handler->conf->pty_names) { | |
1016 | if (putenv(handler->conf->pty_names)) { | |
1017 | SYSERROR("Failed to set environment variable for container ptys."); | |
1018 | goto out_warn_father; | |
1019 | } | |
1020 | } | |
1021 | ||
76bdf299 CB |
1022 | /* The container has been setup. We can now switch to an unprivileged |
1023 | * uid/gid. | |
1024 | */ | |
1025 | if (handler->conf->is_execute) { | |
1026 | bool have_cap_setgid; | |
1027 | uid_t new_uid = handler->conf->init_uid; | |
1028 | gid_t new_gid = handler->conf->init_gid; | |
1029 | ||
1030 | /* If we are in a new user namespace we already dropped all | |
1031 | * groups when we switched to root in the new user namespace | |
1032 | * further above. Only drop groups if we can, so ensure that we | |
1033 | * have necessary privilege. | |
1034 | */ | |
1035 | #if HAVE_LIBCAP | |
1036 | have_cap_setgid = lxc_proc_cap_is_set(CAP_SETGID, CAP_EFFECTIVE); | |
1037 | #else | |
1038 | have_cap_setgid = false; | |
1039 | #endif | |
1040 | if (lxc_list_empty(&handler->conf->id_map) && have_cap_setgid) { | |
1041 | if (lxc_setgroups(0, NULL) < 0) | |
1042 | goto out_warn_father; | |
1043 | } | |
1044 | ||
1045 | if (lxc_switch_uid_gid(new_uid, new_gid) < 0) | |
1046 | goto out_warn_father; | |
1047 | } | |
1048 | ||
408da065 CB |
1049 | /* After this call, we are in error because this ops should not return |
1050 | * as it execs. | |
1051 | */ | |
c4ea60df | 1052 | handler->ops->start(handler, handler->data); |
50e98013 DL |
1053 | |
1054 | out_warn_father: | |
408da065 CB |
1055 | /* We want the parent to know something went wrong, so we return a |
1056 | * special error code. | |
1057 | */ | |
d1ccb562 | 1058 | lxc_sync_wake_parent(handler, LXC_SYNC_ERROR); |
c44de748 AM |
1059 | |
1060 | out_error: | |
1061 | if (devnull_fd >= 0) | |
1062 | close(devnull_fd); | |
1063 | ||
50e98013 DL |
1064 | return -1; |
1065 | } | |
1066 | ||
ae467c54 | 1067 | static int lxc_recv_ttys_from_child(struct lxc_handler *handler) |
e8bd4e43 | 1068 | { |
ae467c54 | 1069 | int i; |
ae467c54 CB |
1070 | struct lxc_pty_info *pty_info; |
1071 | int ret = -1; | |
c6012571 | 1072 | int sock = handler->data_sock[1]; |
e8bd4e43 | 1073 | struct lxc_conf *conf = handler->conf; |
e8bd4e43 SH |
1074 | struct lxc_tty_info *tty_info = &conf->tty_info; |
1075 | ||
1076 | if (!conf->tty) | |
1077 | return 0; | |
1078 | ||
408da065 CB |
1079 | tty_info->pty_info = malloc(sizeof(*tty_info->pty_info) * conf->tty); |
1080 | if (!tty_info->pty_info) | |
e8bd4e43 | 1081 | return -1; |
e8bd4e43 | 1082 | |
672c1e58 CB |
1083 | for (i = 0; i < conf->tty; i++) { |
1084 | int ttyfds[2]; | |
1085 | ||
1086 | ret = lxc_abstract_unix_recv_fds(sock, ttyfds, 2, NULL, 0); | |
1087 | if (ret < 0) | |
1088 | break; | |
ae467c54 | 1089 | |
672c1e58 | 1090 | pty_info = &tty_info->pty_info[i]; |
e8bd4e43 | 1091 | pty_info->busy = 0; |
672c1e58 CB |
1092 | pty_info->master = ttyfds[0]; |
1093 | pty_info->slave = ttyfds[1]; | |
1094 | TRACE("Received pty with master fd %d and slave fd %d from " | |
ae467c54 | 1095 | "parent", pty_info->master, pty_info->slave); |
e8bd4e43 | 1096 | } |
ae467c54 | 1097 | if (ret < 0) |
672c1e58 | 1098 | ERROR("Failed to receive %d ttys from child: %s", conf->tty, |
ae467c54 CB |
1099 | strerror(errno)); |
1100 | else | |
672c1e58 CB |
1101 | TRACE("Received %d ttys from child", conf->tty); |
1102 | ||
1103 | tty_info->nbtty = conf->tty; | |
ae467c54 CB |
1104 | |
1105 | return ret; | |
e8bd4e43 SH |
1106 | } |
1107 | ||
f813849c TA |
1108 | void resolve_clone_flags(struct lxc_handler *handler) |
1109 | { | |
03df7ab5 | 1110 | handler->clone_flags = CLONE_NEWNS; |
f813849c | 1111 | |
408da065 | 1112 | if (!lxc_list_empty(&handler->conf->id_map)) |
f813849c | 1113 | handler->clone_flags |= CLONE_NEWUSER; |
f813849c | 1114 | |
28d9e29e | 1115 | if (!handler->conf->inherit_ns[LXC_NS_NET]) { |
408da065 | 1116 | if (!lxc_requests_empty_network(handler)) |
f813849c TA |
1117 | handler->clone_flags |= CLONE_NEWNET; |
1118 | } else { | |
28d9e29e | 1119 | INFO("Inheriting net namespace"); |
f813849c TA |
1120 | } |
1121 | ||
28d9e29e | 1122 | if (!handler->conf->inherit_ns[LXC_NS_IPC]) |
f813849c | 1123 | handler->clone_flags |= CLONE_NEWIPC; |
408da065 | 1124 | else |
28d9e29e | 1125 | INFO("Inheriting ipc namespace"); |
f813849c | 1126 | |
28d9e29e | 1127 | if (!handler->conf->inherit_ns[LXC_NS_UTS]) |
f813849c | 1128 | handler->clone_flags |= CLONE_NEWUTS; |
408da065 | 1129 | else |
28d9e29e | 1130 | INFO("Inheriting uts namespace"); |
03df7ab5 | 1131 | |
28d9e29e | 1132 | if (!handler->conf->inherit_ns[LXC_NS_PID]) |
03df7ab5 L |
1133 | handler->clone_flags |= CLONE_NEWPID; |
1134 | else | |
28d9e29e | 1135 | INFO("Inheriting pid namespace"); |
f813849c TA |
1136 | } |
1137 | ||
480588e6 CB |
1138 | /* lxc_spawn() performs crucial setup tasks and clone()s the new process which |
1139 | * exec()s the requested container binary. | |
1140 | * Note that lxc_spawn() runs in the parent namespaces. Any operations performed | |
1141 | * right here should be double checked if they'd pose a security risk. (For | |
1142 | * example, any {u}mount() operations performed here will be reflected on the | |
1143 | * host!) | |
1144 | */ | |
74a3920a | 1145 | static int lxc_spawn(struct lxc_handler *handler) |
59eb99ba | 1146 | { |
7ab1ba02 | 1147 | int i, flags, ret; |
08dd2805 | 1148 | char pidstr[20]; |
57927bf2 | 1149 | bool wants_to_map_ids; |
7ab1ba02 | 1150 | int saved_ns_fd[LXC_NS_MAX]; |
57927bf2 | 1151 | struct lxc_list *id_map; |
9e2bdae4 | 1152 | int preserve_mask = 0; |
6e5fc7a5 | 1153 | const char *name = handler->name; |
28d9e29e | 1154 | const char *lxcpath = handler->lxcpath; |
d0b915aa | 1155 | bool cgroups_connected = false; |
28d9e29e | 1156 | struct lxc_conf *conf = handler->conf; |
9f30a190 | 1157 | |
28d9e29e | 1158 | id_map = &conf->id_map; |
57927bf2 | 1159 | wants_to_map_ids = !lxc_list_empty(id_map); |
6e5fc7a5 | 1160 | memset(saved_ns_fd, -1, sizeof(int) * LXC_NS_MAX); |
2f2623ec | 1161 | |
28d9e29e CB |
1162 | for (i = 0; i < LXC_NS_MAX; i++) { |
1163 | if (!conf->inherit_ns[i]) | |
1164 | continue; | |
1165 | ||
1166 | preserve_mask |= ns_info[i].clone_flag; | |
1167 | ||
1168 | handler->nsfd[i] = lxc_inherit_namespace(conf->inherit_ns[i], lxcpath, ns_info[i].proc_name); | |
1169 | if (handler->nsfd[i] < 0) | |
1170 | return -1; | |
1171 | } | |
50e98013 | 1172 | |
3c22086f | 1173 | if (lxc_sync_init(handler)) |
9d7f9e52 | 1174 | return -1; |
0ad19a3f | 1175 | |
b9f522c5 | 1176 | ret = socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0, |
790255cf | 1177 | handler->data_sock); |
c6012571 | 1178 | if (ret < 0) { |
e8bd4e43 SH |
1179 | lxc_sync_fini(handler); |
1180 | return -1; | |
1181 | } | |
1182 | ||
f813849c | 1183 | resolve_clone_flags(handler); |
9f30a190 | 1184 | |
f813849c | 1185 | if (handler->clone_flags & CLONE_NEWNET) { |
28d9e29e | 1186 | if (!lxc_list_empty(&conf->network)) { |
26b797f3 | 1187 | |
9f30a190 MM |
1188 | /* Find gateway addresses from the link device, which is |
1189 | * no longer accessible inside the container. Do this | |
1190 | * before creating network interfaces, since goto | |
408da065 CB |
1191 | * out_delete_net does not work before lxc_clone. |
1192 | */ | |
9f30a190 | 1193 | if (lxc_find_gateway_addresses(handler)) { |
408da065 | 1194 | ERROR("Failed to find gateway addresses."); |
9f30a190 MM |
1195 | lxc_sync_fini(handler); |
1196 | return -1; | |
1197 | } | |
1198 | ||
408da065 CB |
1199 | /* That should be done before the clone because we will |
1200 | * fill the netdev index and use them in the child. | |
9f30a190 | 1201 | */ |
811ef482 | 1202 | if (lxc_create_network_priv(handler)) { |
408da065 | 1203 | ERROR("Failed to create the network."); |
9f30a190 MM |
1204 | lxc_sync_fini(handler); |
1205 | return -1; | |
1206 | } | |
19a26f82 | 1207 | } |
82d5ae15 DL |
1208 | } |
1209 | ||
d4ef7c50 | 1210 | if (!cgroup_init(handler)) { |
408da065 | 1211 | ERROR("Failed initializing cgroup support."); |
33ad9f1a CS |
1212 | goto out_delete_net; |
1213 | } | |
1214 | ||
7e4dfe0b SH |
1215 | cgroups_connected = true; |
1216 | ||
d4ef7c50 | 1217 | if (!cgroup_create(handler)) { |
408da065 | 1218 | ERROR("Failed creating cgroups."); |
47d8fb3b CS |
1219 | goto out_delete_net; |
1220 | } | |
1221 | ||
408da065 CB |
1222 | /* If the rootfs is not a blockdev, prevent the container from marking |
1223 | * it readonly. | |
1224 | * If the container is unprivileged then skip rootfs pinning. | |
0c547523 | 1225 | */ |
0ee35059 | 1226 | if (!wants_to_map_ids) { |
28d9e29e | 1227 | handler->pinfd = pin_rootfs(conf->rootfs.path); |
5e32a990 | 1228 | if (handler->pinfd == -1) |
408da065 | 1229 | INFO("Failed to pin the rootfs for container \"%s\".", handler->name); |
5e32a990 | 1230 | } |
0c547523 | 1231 | |
4d8ac866 | 1232 | if (!preserve_ns(saved_ns_fd, preserve_mask, getpid())) |
cd43d2d1 | 1233 | goto out_delete_net; |
28d9e29e CB |
1234 | else |
1235 | TRACE("Preserved inherited namespaces"); | |
4d8ac866 | 1236 | |
28d9e29e CB |
1237 | for (i = 0; i < LXC_NS_MAX; i++) { |
1238 | if (handler->nsfd[i] < 0) | |
1239 | continue; | |
1240 | ||
1241 | ret = setns(handler->nsfd[i], 0); | |
1242 | if (ret < 0) | |
1243 | goto out_delete_net; | |
1244 | } | |
9f30a190 | 1245 | |
408da065 | 1246 | /* Create a process in a new set of namespaces. */ |
5b1e83cb | 1247 | flags = handler->clone_flags; |
408da065 CB |
1248 | if (handler->clone_flags & CLONE_NEWUSER) { |
1249 | /* If CLONE_NEWUSER and CLONE_NEWNET was requested, we need to | |
1250 | * clone a new user namespace first and only later unshare our | |
1251 | * network namespace to ensure that network devices ownership is | |
1252 | * set up correctly. | |
1253 | */ | |
5b1e83cb | 1254 | flags &= ~CLONE_NEWNET; |
408da065 | 1255 | } |
732375f5 CB |
1256 | |
1257 | ret = lxc_setup_parent(handler); | |
1258 | if (ret < 0) | |
1259 | goto out_delete_net; | |
1260 | ||
9fac8fbb | 1261 | handler->pid = lxc_clone(do_start, handler, flags); |
59eb99ba | 1262 | if (handler->pid < 0) { |
408da065 | 1263 | SYSERROR("Failed to clone a new set of namespaces."); |
7fef7a06 | 1264 | goto out_delete_net; |
0ad19a3f | 1265 | } |
1bd8d726 | 1266 | |
9662e444 CB |
1267 | for (i = 0; i < LXC_NS_MAX; i++) |
1268 | if (flags & ns_info[i].clone_flag) | |
28d9e29e | 1269 | INFO("Cloned %s", ns_info[i].flag_name); |
0ad19a3f | 1270 | |
28d9e29e CB |
1271 | if (!preserve_ns(handler->nsfd, handler->clone_flags, handler->pid)) { |
1272 | ERROR("Failed to preserve cloned namespaces for lxc.hook.stop"); | |
1273 | goto out_delete_net; | |
1274 | } | |
b6b2b194 | 1275 | |
28d9e29e CB |
1276 | if (attach_ns(saved_ns_fd)) { |
1277 | ERROR("Failed to restore saved namespaces"); | |
1278 | goto out_delete_net; | |
1279 | } | |
9f30a190 | 1280 | |
3c22086f CLG |
1281 | lxc_sync_fini_child(handler); |
1282 | ||
408da065 CB |
1283 | /* Map the container uids. The container became an invalid userid the |
1284 | * moment it was cloned with CLONE_NEWUSER. This call doesn't change | |
1285 | * anything immediately, but allows the container to setuid(0) (0 being | |
1286 | * mapped to something else on the host.) later to become a valid uid | |
1287 | * again. | |
1288 | */ | |
57927bf2 | 1289 | if (wants_to_map_ids && lxc_map_ids(id_map, handler->pid)) { |
408da065 | 1290 | ERROR("Failed to set up id mapping."); |
5b1e83cb SH |
1291 | goto out_delete_net; |
1292 | } | |
1293 | ||
9e2bdae4 | 1294 | if (lxc_sync_wake_child(handler, LXC_SYNC_STARTUP)) |
5b1e83cb | 1295 | goto out_delete_net; |
5b1e83cb | 1296 | |
9e2bdae4 | 1297 | if (lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE)) |
5b1e83cb | 1298 | goto out_delete_net; |
0ad19a3f | 1299 | |
d4ef7c50 | 1300 | if (!cgroup_create_legacy(handler)) { |
408da065 | 1301 | ERROR("Failed to setup legacy cgroups for container \"%s\".", name); |
ae5c8b8e | 1302 | goto out_delete_net; |
33ad9f1a | 1303 | } |
9daf6f5d | 1304 | if (!cgroup_setup_limits(handler, false)) { |
408da065 | 1305 | ERROR("Failed to setup cgroup limits for container \"%s\".", name); |
6031a6e5 DE |
1306 | goto out_delete_net; |
1307 | } | |
1308 | ||
d4ef7c50 | 1309 | if (!cgroup_enter(handler)) |
7fef7a06 | 1310 | goto out_delete_net; |
218d4250 | 1311 | |
0996e18a SH |
1312 | if (!cgroup_chown(handler)) |
1313 | goto out_delete_net; | |
1314 | ||
408da065 | 1315 | /* Create the network configuration. */ |
d5088cf2 | 1316 | if (handler->clone_flags & CLONE_NEWNET) { |
74c6e2b0 CB |
1317 | if (lxc_network_move_created_netdev_priv(handler->lxcpath, |
1318 | handler->name, | |
28d9e29e | 1319 | &conf->network, |
74c6e2b0 | 1320 | handler->pid)) { |
408da065 | 1321 | ERROR("Failed to create the configured network."); |
7fef7a06 | 1322 | goto out_delete_net; |
82d5ae15 | 1323 | } |
74c6e2b0 CB |
1324 | |
1325 | if (lxc_create_network_unpriv(handler->lxcpath, handler->name, | |
28d9e29e | 1326 | &conf->network, |
74c6e2b0 CB |
1327 | handler->pid)) { |
1328 | ERROR("Failed to create the configured network."); | |
1329 | goto out_delete_net; | |
1330 | } | |
0ad19a3f | 1331 | } |
1332 | ||
7ab1ba02 CB |
1333 | if (lxc_network_send_veth_names_to_child(handler) < 0) { |
1334 | ERROR("Failed to send veth names to child"); | |
1335 | goto out_delete_net; | |
658979c5 SH |
1336 | } |
1337 | ||
408da065 CB |
1338 | /* Tell the child to continue its initialization. We'll get |
1339 | * LXC_SYNC_CGROUP when it is ready for us to setup cgroups. | |
3c22086f CLG |
1340 | */ |
1341 | if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE)) | |
544a48a0 SH |
1342 | goto out_delete_net; |
1343 | ||
28d9e29e | 1344 | if (!lxc_list_empty(&conf->limits) && setup_resource_limits(&conf->limits, handler->pid)) { |
c6d09e15 | 1345 | ERROR("failed to setup resource limits for '%s'", name); |
84ff3af7 | 1346 | goto out_delete_net; |
c6d09e15 WB |
1347 | } |
1348 | ||
f4152036 CB |
1349 | if (lxc_sync_barrier_child(handler, LXC_SYNC_CGROUP_UNSHARE)) |
1350 | goto out_delete_net; | |
1351 | ||
9daf6f5d | 1352 | if (!cgroup_setup_limits(handler, true)) { |
408da065 | 1353 | ERROR("Failed to setup the devices cgroup for container \"%s\".", name); |
b98f7d6e | 1354 | goto out_delete_net; |
544a48a0 | 1355 | } |
f4152036 | 1356 | TRACE("Set up cgroup device limits"); |
544a48a0 | 1357 | |
73d28d42 | 1358 | cgroup_disconnect(); |
7e4dfe0b | 1359 | cgroups_connected = false; |
73d28d42 | 1360 | |
08dd2805 SH |
1361 | snprintf(pidstr, 20, "%d", handler->pid); |
1362 | if (setenv("LXC_PID", pidstr, 1)) | |
1363 | SYSERROR("Failed to set environment variable: LXC_PID=%s.", pidstr); | |
1364 | ||
1365 | /* Run any host-side start hooks */ | |
28d9e29e | 1366 | if (run_lxc_hooks(name, "start-host", conf, handler->lxcpath, NULL)) { |
08dd2805 SH |
1367 | ERROR("Failed to run lxc.hook.start-host for container \"%s\".", name); |
1368 | return -1; | |
1369 | } | |
1370 | ||
408da065 CB |
1371 | /* Tell the child to complete its initialization and wait for it to exec |
1372 | * or return an error. (The child will never return | |
08dd2805 | 1373 | * LXC_SYNC_READY_START+1. It will either close the sync pipe, causing |
408da065 CB |
1374 | * lxc_sync_barrier_child to return success, or return a different |
1375 | * value, causing us to error out). | |
544a48a0 | 1376 | */ |
08dd2805 | 1377 | if (lxc_sync_barrier_child(handler, LXC_SYNC_READY_START)) |
3c22086f | 1378 | return -1; |
0ad19a3f | 1379 | |
790255cf CB |
1380 | if (lxc_network_recv_name_and_ifindex_from_child(handler) < 0) { |
1381 | ERROR("Failed to receive names and ifindices for network " | |
1382 | "devices from child"); | |
1383 | goto out_delete_net; | |
1384 | } | |
1385 | ||
1386 | /* Now all networks are created, network devices are moved into place, | |
1387 | * and the correct names and ifindeces in the respective namespaces have | |
1388 | * been recorded. The corresponding structs have now all been filled. So | |
1389 | * log them for debugging purposes. | |
1390 | */ | |
28d9e29e | 1391 | lxc_log_configured_netdevs(conf); |
790255cf | 1392 | |
f4152036 CB |
1393 | /* Read tty fds allocated by child. */ |
1394 | if (lxc_recv_ttys_from_child(handler) < 0) { | |
1395 | ERROR("Failed to receive tty info from child process."); | |
1396 | goto out_delete_net; | |
1397 | } | |
1398 | ||
23c53af9 | 1399 | if (handler->ops->post_start(handler, handler->data)) |
e6126dbe MN |
1400 | goto out_abort; |
1401 | ||
25c2aca5 | 1402 | if (lxc_set_state(name, handler, RUNNING)) { |
408da065 CB |
1403 | ERROR("Failed to set state for container \"%s\" to \"%s\".", name, |
1404 | lxc_state2str(RUNNING)); | |
59eb99ba | 1405 | goto out_abort; |
3f21c114 | 1406 | } |
22ebac19 | 1407 | |
3c22086f | 1408 | lxc_sync_fini(handler); |
0c547523 | 1409 | |
6e5fc7a5 CB |
1410 | for (i = 0; i < LXC_NS_MAX; i++) |
1411 | if (saved_ns_fd[i] != -1) | |
1412 | close(saved_ns_fd[i]); | |
1413 | ||
e6126dbe | 1414 | return 0; |
1ac470c0 | 1415 | |
7fef7a06 | 1416 | out_delete_net: |
6e5fc7a5 CB |
1417 | for (i = 0; i < LXC_NS_MAX; i++) |
1418 | if (saved_ns_fd[i] != -1) | |
1419 | close(saved_ns_fd[i]); | |
1420 | ||
7e4dfe0b SH |
1421 | if (cgroups_connected) |
1422 | cgroup_disconnect(); | |
1bd8d726 | 1423 | |
bb84beda CB |
1424 | if (handler->clone_flags & CLONE_NEWNET) |
1425 | lxc_delete_network(handler); | |
1bd8d726 | 1426 | |
59eb99ba DL |
1427 | out_abort: |
1428 | lxc_abort(name, handler); | |
3c22086f | 1429 | lxc_sync_fini(handler); |
5c068da9 SH |
1430 | if (handler->pinfd >= 0) { |
1431 | close(handler->pinfd); | |
1432 | handler->pinfd = -1; | |
1433 | } | |
1434 | ||
b79fcd86 | 1435 | return -1; |
59eb99ba | 1436 | } |
0ad19a3f | 1437 | |
aa460476 | 1438 | int __lxc_start(const char *name, struct lxc_handler *handler, |
507cee36 TA |
1439 | struct lxc_operations* ops, void *data, const char *lxcpath, |
1440 | bool backgrounded) | |
59eb99ba | 1441 | { |
59eb99ba | 1442 | int status; |
aa460476 | 1443 | int err = -1; |
aa460476 | 1444 | struct lxc_conf *conf = handler->conf; |
80090207 | 1445 | |
aa460476 | 1446 | if (lxc_init(name, handler) < 0) { |
408da065 | 1447 | ERROR("Failed to initialize container \"%s\".", name); |
66aeffc7 | 1448 | return -1; |
0ad19a3f | 1449 | } |
ee70bf78 CLG |
1450 | handler->ops = ops; |
1451 | handler->data = data; | |
507cee36 | 1452 | handler->backgrounded = backgrounded; |
e6126dbe | 1453 | |
76a26f55 | 1454 | if (!attach_block_device(handler->conf)) { |
408da065 | 1455 | ERROR("Failed to attach block device."); |
76a26f55 SH |
1456 | goto out_fini_nonet; |
1457 | } | |
1458 | ||
35120d9c | 1459 | if (geteuid() == 0 && !lxc_list_empty(&conf->id_map)) { |
408da065 | 1460 | /* If the backing store is a device, mount it here and now. */ |
35120d9c SH |
1461 | if (rootfs_is_blockdev(conf)) { |
1462 | if (unshare(CLONE_NEWNS) < 0) { | |
408da065 | 1463 | ERROR("Failed to unshare CLONE_NEWNS."); |
35120d9c SH |
1464 | goto out_fini_nonet; |
1465 | } | |
408da065 CB |
1466 | INFO("Unshared CLONE_NEWNS."); |
1467 | ||
6a0c909a | 1468 | remount_all_slave(); |
35120d9c | 1469 | if (do_rootfs_setup(conf, name, lxcpath) < 0) { |
408da065 | 1470 | ERROR("Error setting up rootfs mount as root before spawn."); |
35120d9c SH |
1471 | goto out_fini_nonet; |
1472 | } | |
408da065 | 1473 | INFO("Set up container rootfs as host root."); |
35120d9c SH |
1474 | } |
1475 | } | |
1476 | ||
23c53af9 | 1477 | err = lxc_spawn(handler); |
59eb99ba | 1478 | if (err) { |
408da065 | 1479 | ERROR("Failed to spawn container \"%s\".", name); |
76a26f55 | 1480 | goto out_detach_blockdev; |
0ad19a3f | 1481 | } |
1482 | ||
8bee8851 WB |
1483 | handler->conf->reboot = 0; |
1484 | ||
3a0f472d | 1485 | err = lxc_poll(name, handler); |
e043236e | 1486 | if (err) { |
408da065 | 1487 | ERROR("LXC mainloop exited with error: %d.", err); |
59eb99ba DL |
1488 | goto out_abort; |
1489 | } | |
0ad19a3f | 1490 | |
3a0f472d | 1491 | while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR) |
1bc5cc8c | 1492 | continue; |
e043236e | 1493 | |
408da065 CB |
1494 | /* If the child process exited but was not signaled, it didn't call |
1495 | * reboot. This should mean it was an lxc-execute which simply exited. | |
1496 | * In any case, treat it as a 'halt'. | |
8b004f07 | 1497 | */ |
d028235d | 1498 | if (WIFSIGNALED(status)) { |
8b004f07 SH |
1499 | switch(WTERMSIG(status)) { |
1500 | case SIGINT: /* halt */ | |
408da065 | 1501 | DEBUG("Container \"%s\" is halting.", name); |
8b004f07 SH |
1502 | break; |
1503 | case SIGHUP: /* reboot */ | |
408da065 | 1504 | DEBUG("Container \"%s\" is rebooting.", name); |
8b004f07 SH |
1505 | handler->conf->reboot = 1; |
1506 | break; | |
c2b9bd9e | 1507 | case SIGSYS: /* seccomp */ |
408da065 | 1508 | DEBUG("Container \"%s\" violated its seccomp policy.", name); |
c2b9bd9e | 1509 | break; |
8b004f07 | 1510 | default: |
408da065 | 1511 | DEBUG("Unknown exit status for container \"%s\" init %d.", name, WTERMSIG(status)); |
8b004f07 SH |
1512 | break; |
1513 | } | |
d028235d | 1514 | } |
828695d9 | 1515 | |
b809f232 CB |
1516 | err = lxc_restore_phys_nics_to_netns(handler); |
1517 | if (err < 0) | |
1518 | ERROR("Failed to move physical network devices back to parent " | |
1519 | "network namespace"); | |
ce5782df | 1520 | |
5c068da9 SH |
1521 | if (handler->pinfd >= 0) { |
1522 | close(handler->pinfd); | |
1523 | handler->pinfd = -1; | |
1524 | } | |
1525 | ||
1787abca | 1526 | lxc_monitor_send_exit_code(name, status, handler->lxcpath); |
3a0f472d | 1527 | err = lxc_error_set_and_log(handler->pid, status); |
1bd8d726 | 1528 | |
9d7f9e52 | 1529 | out_fini: |
bb84beda | 1530 | lxc_delete_network(handler); |
74a2b586 | 1531 | |
76a26f55 SH |
1532 | out_detach_blockdev: |
1533 | detach_block_device(handler->conf); | |
1534 | ||
74a2b586 | 1535 | out_fini_nonet: |
3a0f472d | 1536 | lxc_fini(name, handler); |
0ad19a3f | 1537 | return err; |
1538 | ||
59eb99ba | 1539 | out_abort: |
3a0f472d | 1540 | lxc_abort(name, handler); |
9d7f9e52 | 1541 | goto out_fini; |
0ad19a3f | 1542 | } |
ee70bf78 CLG |
1543 | |
1544 | struct start_args { | |
1545 | char *const *argv; | |
1546 | }; | |
1547 | ||
1548 | static int start(struct lxc_handler *handler, void* data) | |
1549 | { | |
1550 | struct start_args *arg = data; | |
1551 | ||
408da065 | 1552 | NOTICE("Exec'ing \"%s\".", arg->argv[0]); |
ee70bf78 CLG |
1553 | |
1554 | execvp(arg->argv[0], arg->argv); | |
408da065 | 1555 | SYSERROR("Failed to exec \"%s\".", arg->argv[0]); |
ee70bf78 CLG |
1556 | return 0; |
1557 | } | |
1558 | ||
1559 | static int post_start(struct lxc_handler *handler, void* data) | |
1560 | { | |
1561 | struct start_args *arg = data; | |
1562 | ||
408da065 | 1563 | NOTICE("Started \"%s\" with pid \"%d\".", arg->argv[0], handler->pid); |
ee70bf78 CLG |
1564 | return 0; |
1565 | } | |
1566 | ||
1567 | static struct lxc_operations start_ops = { | |
1568 | .start = start, | |
1569 | .post_start = post_start | |
1570 | }; | |
1571 | ||
aa460476 | 1572 | int lxc_start(const char *name, char *const argv[], struct lxc_handler *handler, |
507cee36 | 1573 | const char *lxcpath, bool backgrounded) |
ee70bf78 CLG |
1574 | { |
1575 | struct start_args start_arg = { | |
1576 | .argv = argv, | |
1577 | }; | |
1578 | ||
aa460476 | 1579 | return __lxc_start(name, handler, &start_ops, &start_arg, lxcpath, backgrounded); |
ee70bf78 | 1580 | } |
28272964 CB |
1581 | |
1582 | static void lxc_destroy_container_on_signal(struct lxc_handler *handler, | |
1583 | const char *name) | |
1584 | { | |
1585 | char destroy[MAXPATHLEN]; | |
1586 | bool bret = true; | |
1587 | int ret = 0; | |
f01f7975 | 1588 | struct lxc_container *c; |
df31363a | 1589 | if (handler->conf->rootfs.path && handler->conf->rootfs.mount) { |
790255cf | 1590 | bret = do_destroy_container(handler); |
28272964 | 1591 | if (!bret) { |
408da065 | 1592 | ERROR("Error destroying rootfs for container \"%s\".", name); |
28272964 CB |
1593 | return; |
1594 | } | |
1595 | } | |
408da065 | 1596 | INFO("Destroyed rootfs for container \"%s\".", name); |
28272964 CB |
1597 | |
1598 | ret = snprintf(destroy, MAXPATHLEN, "%s/%s", handler->lxcpath, name); | |
1599 | if (ret < 0 || ret >= MAXPATHLEN) { | |
408da065 | 1600 | ERROR("Error destroying directory for container \"%s\".", name); |
28272964 CB |
1601 | return; |
1602 | } | |
1603 | ||
f01f7975 CB |
1604 | c = lxc_container_new(name, handler->lxcpath); |
1605 | if (c) { | |
1606 | if (container_disk_lock(c)) { | |
408da065 | 1607 | INFO("Could not update lxc_snapshots file."); |
f01f7975 CB |
1608 | lxc_container_put(c); |
1609 | } else { | |
1610 | mod_all_rdeps(c, false); | |
1611 | container_disk_unlock(c); | |
1612 | lxc_container_put(c); | |
1613 | } | |
1614 | } | |
1615 | ||
d0fbc7ba | 1616 | if (!handler->am_root) |
94b0a3ac CB |
1617 | ret = userns_exec_full(handler->conf, lxc_rmdir_onedev_wrapper, |
1618 | destroy, "lxc_rmdir_onedev_wrapper"); | |
28272964 CB |
1619 | else |
1620 | ret = lxc_rmdir_onedev(destroy, NULL); | |
1621 | ||
1622 | if (ret < 0) { | |
408da065 | 1623 | ERROR("Error destroying directory for container \"%s\".", name); |
28272964 CB |
1624 | return; |
1625 | } | |
408da065 | 1626 | INFO("Destroyed directory for container \"%s\".", name); |
28272964 CB |
1627 | } |
1628 | ||
1629 | static int lxc_rmdir_onedev_wrapper(void *data) | |
1630 | { | |
1631 | char *arg = (char *) data; | |
1632 | return lxc_rmdir_onedev(arg, NULL); | |
1633 | } | |
1634 | ||
790255cf | 1635 | static bool do_destroy_container(struct lxc_handler *handler) { |
94b0a3ac CB |
1636 | int ret; |
1637 | ||
d0fbc7ba | 1638 | if (!handler->am_root) { |
94b0a3ac CB |
1639 | ret = userns_exec_full(handler->conf, storage_destroy_wrapper, |
1640 | handler->conf, "storage_destroy_wrapper"); | |
1641 | if (ret < 0) | |
d028235d | 1642 | return false; |
10bc1861 | 1643 | |
d028235d SG |
1644 | return true; |
1645 | } | |
10bc1861 | 1646 | |
790255cf | 1647 | return storage_destroy(handler->conf); |
28272964 | 1648 | } |