]>
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))) { |
d39b10eb CB |
203 | struct lxc_list *cur; |
204 | bool matched = false; | |
205 | ||
80090207 CLG |
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 | ||
d39b10eb | 225 | /* Keep state clients that wait on reboots. */ |
3ee9e4fb CB |
226 | if (conf) { |
227 | lxc_list_for_each(cur, &conf->state_clients) { | |
228 | struct lxc_state_client *client = cur->elem; | |
d39b10eb | 229 | |
3ee9e4fb CB |
230 | if (client->clientfd != fd) |
231 | continue; | |
d39b10eb | 232 | |
3ee9e4fb CB |
233 | matched = true; |
234 | break; | |
235 | } | |
d39b10eb CB |
236 | } |
237 | ||
238 | if (matched) | |
239 | continue; | |
240 | ||
858377e4 SH |
241 | if (current_config && fd == current_config->logfd) |
242 | continue; | |
243 | ||
80090207 CLG |
244 | if (match_fd(fd)) |
245 | continue; | |
80090207 | 246 | |
d2cf4c37 | 247 | if (closeall) { |
b119f362 SH |
248 | close(fd); |
249 | closedir(dir); | |
28d9e29e | 250 | INFO("Closed inherited fd %d", fd); |
b119f362 SH |
251 | goto restart; |
252 | } | |
28d9e29e | 253 | WARN("Inherited fd %d", fd); |
80090207 CLG |
254 | } |
255 | ||
408da065 | 256 | /* Only enable syslog at this point to avoid the above logging function |
64c57ea1 BD |
257 | * to open a new fd and make the check_inherited function enter an |
258 | * infinite loop. | |
259 | */ | |
260 | lxc_log_enable_syslog(); | |
261 | ||
92c7f629 GK |
262 | closedir(dir); /* cannot fail */ |
263 | return 0; | |
80090207 CLG |
264 | } |
265 | ||
83ee7875 | 266 | static int setup_signal_fd(sigset_t *oldmask) |
b0a33c1e | 267 | { |
268 | sigset_t mask; | |
269 | int fd; | |
270 | ||
408da065 | 271 | /* Block everything except serious error signals. */ |
f3304a29 FW |
272 | if (sigfillset(&mask) || |
273 | sigdelset(&mask, SIGILL) || | |
274 | sigdelset(&mask, SIGSEGV) || | |
275 | sigdelset(&mask, SIGBUS) || | |
b5159817 | 276 | sigdelset(&mask, SIGWINCH) || |
f3304a29 | 277 | sigprocmask(SIG_BLOCK, &mask, oldmask)) { |
408da065 | 278 | SYSERROR("Failed to set signal mask."); |
b0a33c1e | 279 | return -1; |
280 | } | |
281 | ||
282 | fd = signalfd(-1, &mask, 0); | |
283 | if (fd < 0) { | |
408da065 | 284 | SYSERROR("Failed to create signal file descriptor."); |
b0a33c1e | 285 | return -1; |
286 | } | |
287 | ||
288 | if (fcntl(fd, F_SETFD, FD_CLOEXEC)) { | |
408da065 | 289 | SYSERROR("Failed to set FD_CLOEXEC on the signal file descriptor: %d.", fd); |
b0a33c1e | 290 | close(fd); |
291 | return -1; | |
292 | } | |
293 | ||
408da065 | 294 | DEBUG("Set SIGCHLD handler with file descriptor: %d.", fd); |
1ac470c0 | 295 | |
b0a33c1e | 296 | return fd; |
297 | } | |
298 | ||
84c92abd | 299 | static int signal_handler(int fd, uint32_t events, void *data, |
f3787121 | 300 | struct lxc_epoll_descr *descr) |
b0a33c1e | 301 | { |
15cd25fd | 302 | struct signalfd_siginfo siginfo; |
80507ee8 | 303 | siginfo_t info; |
15cd25fd | 304 | int ret; |
82d89dce | 305 | pid_t *pid = data; |
80507ee8 | 306 | bool init_died = false; |
15cd25fd DL |
307 | |
308 | ret = read(fd, &siginfo, sizeof(siginfo)); | |
309 | if (ret < 0) { | |
408da065 | 310 | ERROR("Failed to read signal info from signal file descriptor: %d.", fd); |
15cd25fd DL |
311 | return -1; |
312 | } | |
313 | ||
314 | if (ret != sizeof(siginfo)) { | |
408da065 | 315 | ERROR("Unexpected size for siginfo struct."); |
15cd25fd DL |
316 | return -1; |
317 | } | |
318 | ||
408da065 | 319 | /* Check whether init is running. */ |
80507ee8 SH |
320 | info.si_pid = 0; |
321 | ret = waitid(P_PID, *pid, &info, WEXITED | WNOWAIT | WNOHANG); | |
408da065 | 322 | if (ret == 0 && info.si_pid == *pid) |
80507ee8 | 323 | init_died = true; |
80507ee8 | 324 | |
f3304a29 FW |
325 | if (siginfo.ssi_signo != SIGCHLD) { |
326 | kill(*pid, siginfo.ssi_signo); | |
408da065 | 327 | INFO("Forwarded signal %d to pid %d.", siginfo.ssi_signo, *pid); |
80507ee8 | 328 | return init_died ? 1 : 0; |
f3304a29 FW |
329 | } |
330 | ||
408da065 CB |
331 | if (siginfo.ssi_code == CLD_STOPPED) { |
332 | INFO("Container init process was stopped."); | |
333 | return init_died ? 1 : 0; | |
334 | } else if (siginfo.ssi_code == CLD_CONTINUED) { | |
335 | INFO("Container init process was continued."); | |
80507ee8 | 336 | return init_died ? 1 : 0; |
15cd25fd | 337 | } |
1ac470c0 | 338 | |
408da065 CB |
339 | /* More robustness, protect ourself from a SIGCHLD sent |
340 | * by a process different from the container init. | |
82d89dce DL |
341 | */ |
342 | if (siginfo.ssi_pid != *pid) { | |
af4c0f05 | 343 | NOTICE("Received SIGCHLD from pid %d instead of container init %d.", siginfo.ssi_pid, *pid); |
80507ee8 | 344 | return init_died ? 1 : 0; |
82d89dce DL |
345 | } |
346 | ||
408da065 | 347 | DEBUG("Container init process %d exited.", *pid); |
b0a33c1e | 348 | return 1; |
349 | } | |
350 | ||
5e5576a4 CB |
351 | static int lxc_serve_state_clients(const char *name, |
352 | struct lxc_handler *handler, | |
353 | lxc_state_t state) | |
66aeffc7 | 354 | { |
dbc9832d CB |
355 | ssize_t ret; |
356 | struct lxc_list *cur, *next; | |
d39b10eb | 357 | struct lxc_state_client *client; |
dbc9832d CB |
358 | struct lxc_msg msg = {.type = lxc_msg_state, .value = state}; |
359 | ||
bc631984 | 360 | process_lock(); |
66aeffc7 | 361 | handler->state = state; |
9dfa4041 | 362 | TRACE("Set container state to %s", lxc_state2str(state)); |
dbc9832d | 363 | |
d39b10eb | 364 | if (lxc_list_empty(&handler->conf->state_clients)) { |
9dfa4041 | 365 | TRACE("No state clients registered"); |
dbc9832d | 366 | process_unlock(); |
65ee1673 | 367 | lxc_monitor_send_state(name, state, handler->lxcpath); |
dbc9832d CB |
368 | return 0; |
369 | } | |
370 | ||
371 | strncpy(msg.name, name, sizeof(msg.name)); | |
372 | msg.name[sizeof(msg.name) - 1] = 0; | |
373 | ||
d39b10eb | 374 | lxc_list_for_each_safe(cur, &handler->conf->state_clients, next) { |
dbc9832d CB |
375 | client = cur->elem; |
376 | ||
d39b10eb | 377 | if (client->states[state] == 0) { |
9dfa4041 | 378 | TRACE("State %s not registered for state client %d", |
dbc9832d CB |
379 | lxc_state2str(state), client->clientfd); |
380 | continue; | |
381 | } | |
382 | ||
9dfa4041 | 383 | TRACE("Sending state %s to state client %d", |
dbc9832d CB |
384 | lxc_state2str(state), client->clientfd); |
385 | ||
386 | again: | |
387 | ret = send(client->clientfd, &msg, sizeof(msg), 0); | |
ee8377bd CB |
388 | if (ret <= 0) { |
389 | if (errno == EINTR) { | |
390 | TRACE("Caught EINTR; retrying"); | |
dbc9832d | 391 | goto again; |
ee8377bd | 392 | } |
dbc9832d | 393 | |
9dfa4041 CB |
394 | ERROR("%s - Failed to send message to client", |
395 | strerror(errno)); | |
dbc9832d CB |
396 | } |
397 | ||
398 | /* kick client from list */ | |
399 | close(client->clientfd); | |
400 | lxc_list_del(cur); | |
401 | free(cur->elem); | |
402 | free(cur); | |
403 | } | |
404 | process_unlock(); | |
405 | ||
5e5576a4 CB |
406 | return 0; |
407 | } | |
408 | ||
409 | static int lxc_serve_state_socket_pair(const char *name, | |
410 | struct lxc_handler *handler, | |
411 | lxc_state_t state) | |
412 | { | |
413 | ssize_t ret; | |
414 | ||
415 | if (!handler->backgrounded || | |
416 | handler->state_socket_pair[1] < 0 || | |
417 | state == STARTING) | |
418 | return 0; | |
419 | ||
420 | /* Close read end of the socket pair. */ | |
421 | close(handler->state_socket_pair[0]); | |
422 | handler->state_socket_pair[0] = -1; | |
423 | ||
ee8377bd | 424 | again: |
5e5576a4 CB |
425 | ret = lxc_abstract_unix_send_credential(handler->state_socket_pair[1], |
426 | &(int){state}, sizeof(int)); | |
ee8377bd CB |
427 | if (ret != sizeof(int)) { |
428 | if (errno == EINTR) | |
429 | goto again; | |
5e5576a4 CB |
430 | SYSERROR("Failed to send state to %d", |
431 | handler->state_socket_pair[1]); | |
ee8377bd CB |
432 | return -1; |
433 | } | |
5e5576a4 CB |
434 | |
435 | TRACE("Sent container state \"%s\" to %d", lxc_state2str(state), | |
436 | handler->state_socket_pair[1]); | |
437 | ||
438 | /* Close write end of the socket pair. */ | |
439 | close(handler->state_socket_pair[1]); | |
440 | handler->state_socket_pair[1] = -1; | |
441 | ||
442 | return 0; | |
443 | } | |
444 | ||
445 | int lxc_set_state(const char *name, struct lxc_handler *handler, | |
446 | lxc_state_t state) | |
447 | { | |
448 | int ret; | |
449 | ||
450 | ret = lxc_serve_state_socket_pair(name, handler, state); | |
451 | if (ret < 0) { | |
452 | ERROR("Failed to synchronize via anonymous pair of unix sockets"); | |
453 | return -1; | |
454 | } | |
455 | ||
456 | ret = lxc_serve_state_clients(name, handler, state); | |
457 | if (ret < 0) | |
458 | return -1; | |
459 | ||
dbc9832d CB |
460 | /* This function will try to connect to the legacy lxc-monitord state |
461 | * server and only exists for backwards compatibility. | |
462 | */ | |
9123e471 | 463 | lxc_monitor_send_state(name, state, handler->lxcpath); |
dbc9832d | 464 | |
66aeffc7 DL |
465 | return 0; |
466 | } | |
467 | ||
735f2c6e | 468 | int lxc_poll(const char *name, struct lxc_handler *handler) |
b0a33c1e | 469 | { |
ca5f7926 DL |
470 | int sigfd = handler->sigfd; |
471 | int pid = handler->pid; | |
b0a33c1e | 472 | struct lxc_epoll_descr descr; |
473 | ||
a9e61274 | 474 | if (lxc_mainloop_open(&descr)) { |
408da065 | 475 | ERROR("Failed to create LXC mainloop."); |
50c8bf05 | 476 | goto out_sigfd; |
b0a33c1e | 477 | } |
478 | ||
83ee7875 | 479 | if (lxc_mainloop_add_handler(&descr, sigfd, signal_handler, &pid)) { |
408da065 | 480 | ERROR("Failed to add signal handler with file descriptor %d to LXC mainloop.", sigfd); |
b0a33c1e | 481 | goto out_mainloop_open; |
482 | } | |
483 | ||
da41561c | 484 | if (lxc_console_mainloop_add(&descr, handler->conf)) { |
408da065 | 485 | ERROR("Failed to add console handler to LXC mainloop."); |
63376d7d DL |
486 | goto out_mainloop_open; |
487 | } | |
488 | ||
ef6e34ee | 489 | if (lxc_cmd_mainloop_add(name, &descr, handler)) { |
408da065 | 490 | ERROR("Failed to add command handler to LXC mainloop."); |
96fa1ff0 | 491 | goto out_mainloop_open; |
563f2f2c DL |
492 | } |
493 | ||
3f903c04 | 494 | TRACE("lxc mainloop is ready"); |
b0a33c1e | 495 | |
e51d4895 | 496 | return lxc_mainloop(&descr, -1); |
b0a33c1e | 497 | |
498 | out_mainloop_open: | |
499 | lxc_mainloop_close(&descr); | |
408da065 | 500 | |
b0a33c1e | 501 | out_sigfd: |
502 | close(sigfd); | |
408da065 | 503 | |
c3e13372 | 504 | return -1; |
b0a33c1e | 505 | } |
506 | ||
41784e4e CB |
507 | void lxc_zero_handler(struct lxc_handler *handler) |
508 | { | |
509 | int i; | |
510 | ||
511 | memset(handler, 0, sizeof(struct lxc_handler)); | |
512 | ||
513 | handler->clone_flags = -1; | |
514 | ||
515 | handler->pinfd = -1; | |
516 | ||
517 | handler->sigfd = -1; | |
518 | ||
519 | for (i = 0; i < LXC_NS_MAX; i++) | |
520 | handler->nsfd[i] = -1; | |
521 | ||
522 | handler->data_sock[0] = -1; | |
523 | handler->data_sock[1] = -1; | |
524 | ||
525 | handler->state_socket_pair[0] = -1; | |
526 | handler->state_socket_pair[1] = -1; | |
527 | ||
528 | handler->sync_sock[0] = -1; | |
529 | handler->sync_sock[1] = -1; | |
530 | } | |
531 | ||
f2e07cb6 CB |
532 | void lxc_free_handler(struct lxc_handler *handler) |
533 | { | |
534 | if (handler->conf && handler->conf->maincmd_fd) | |
535 | close(handler->conf->maincmd_fd); | |
536 | ||
5e5576a4 CB |
537 | if (handler->state_socket_pair[0] >= 0) |
538 | close(handler->state_socket_pair[0]); | |
539 | ||
540 | if (handler->state_socket_pair[1] >= 0) | |
541 | close(handler->state_socket_pair[1]); | |
542 | ||
f2e07cb6 CB |
543 | handler->conf = NULL; |
544 | free(handler); | |
545 | } | |
546 | ||
aa460476 | 547 | struct lxc_handler *lxc_init_handler(const char *name, struct lxc_conf *conf, |
5e5576a4 | 548 | const char *lxcpath, bool daemonize) |
59eb99ba | 549 | { |
5e5576a4 | 550 | int i, ret; |
3a0f472d DL |
551 | struct lxc_handler *handler; |
552 | ||
553 | handler = malloc(sizeof(*handler)); | |
aa460476 CB |
554 | if (!handler) { |
555 | ERROR("failed to allocate memory"); | |
3a0f472d | 556 | return NULL; |
aa460476 | 557 | } |
59eb99ba DL |
558 | |
559 | memset(handler, 0, sizeof(*handler)); | |
560 | ||
790255cf CB |
561 | /* Note that am_unpriv() checks the effective uid. We probably don't |
562 | * care if we are real root only if we are running as root so this | |
563 | * should be fine. | |
564 | */ | |
d0fbc7ba | 565 | handler->am_root = !am_unpriv(); |
c6012571 | 566 | handler->data_sock[0] = handler->data_sock[1] = -1; |
fae349da | 567 | handler->conf = conf; |
9123e471 | 568 | handler->lxcpath = lxcpath; |
5c068da9 | 569 | handler->pinfd = -1; |
5e5576a4 | 570 | handler->state_socket_pair[0] = handler->state_socket_pair[1] = -1; |
d39b10eb CB |
571 | if (handler->conf->reboot == 0) |
572 | lxc_list_init(&handler->conf->state_clients); | |
fae349da | 573 | |
b6b2b194 WB |
574 | for (i = 0; i < LXC_NS_MAX; i++) |
575 | handler->nsfd[i] = -1; | |
576 | ||
f0ecc19d | 577 | handler->name = name; |
3bdf52d7 | 578 | |
5e5576a4 CB |
579 | if (daemonize && !handler->conf->reboot) { |
580 | /* Create socketpair() to synchronize on daemonized startup. | |
581 | * When the container reboots we don't need to synchronize again | |
582 | * currently so don't open another socketpair(). | |
583 | */ | |
584 | ret = socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0, | |
585 | handler->state_socket_pair); | |
586 | if (ret < 0) { | |
587 | ERROR("Failed to create anonymous pair of unix sockets"); | |
588 | goto on_error; | |
589 | } | |
590 | TRACE("Created anonymous pair {%d,%d} of unix sockets", | |
591 | handler->state_socket_pair[0], | |
592 | handler->state_socket_pair[1]); | |
593 | } | |
594 | ||
d39b10eb CB |
595 | if (handler->conf->reboot == 0) { |
596 | handler->conf->maincmd_fd = lxc_cmd_init(name, lxcpath, "command"); | |
597 | if (handler->conf->maincmd_fd < 0) { | |
598 | ERROR("Failed to set up command socket"); | |
599 | goto on_error; | |
600 | } | |
aa460476 | 601 | } |
9dfa4041 | 602 | TRACE("Unix domain socket %d for command server is ready", |
aa460476 CB |
603 | handler->conf->maincmd_fd); |
604 | ||
605 | return handler; | |
606 | ||
f2e07cb6 CB |
607 | on_error: |
608 | lxc_free_handler(handler); | |
aa460476 CB |
609 | |
610 | return NULL; | |
611 | } | |
612 | ||
613 | int lxc_init(const char *name, struct lxc_handler *handler) | |
614 | { | |
a2c09be0 | 615 | int ret; |
4a03ded4 | 616 | const char *loglevel; |
aa460476 CB |
617 | struct lxc_conf *conf = handler->conf; |
618 | ||
619 | lsm_init(); | |
620 | TRACE("initialized LSM"); | |
d2e30e99 | 621 | |
8f2c3a70 | 622 | if (lxc_read_seccomp_config(conf) != 0) { |
408da065 | 623 | ERROR("Failed loading seccomp policy."); |
d2e30e99 | 624 | goto out_close_maincmd_fd; |
8f2c3a70 | 625 | } |
dbc9832d | 626 | TRACE("read seccomp policy"); |
8f2c3a70 | 627 | |
408da065 | 628 | /* Begin by setting the state to STARTING. */ |
25c2aca5 | 629 | if (lxc_set_state(name, handler, STARTING)) { |
408da065 | 630 | ERROR("Failed to set state for container \"%s\" to \"%s\".", name, lxc_state2str(STARTING)); |
051151de | 631 | goto out_close_maincmd_fd; |
0ad19a3f | 632 | } |
dbc9832d | 633 | TRACE("set container state to \"STARTING\""); |
0ad19a3f | 634 | |
408da065 CB |
635 | /* Start of environment variable setup for hooks. */ |
636 | if (name && setenv("LXC_NAME", name, 1)) | |
637 | SYSERROR("Failed to set environment variable: LXC_NAME=%s.", name); | |
638 | ||
639 | if (conf->rcfile && setenv("LXC_CONFIG_FILE", conf->rcfile, 1)) | |
640 | SYSERROR("Failed to set environment variable: LXC_CONFIG_FILE=%s.", conf->rcfile); | |
641 | ||
642 | if (conf->rootfs.mount && setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1)) | |
643 | SYSERROR("Failed to set environment variable: LXC_ROOTFS_MOUNT=%s.", conf->rootfs.mount); | |
644 | ||
645 | if (conf->rootfs.path && setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1)) | |
646 | SYSERROR("Failed to set environment variable: LXC_ROOTFS_PATH=%s.", conf->rootfs.path); | |
647 | ||
648 | if (conf->console.path && setenv("LXC_CONSOLE", conf->console.path, 1)) | |
649 | SYSERROR("Failed to set environment variable: LXC_CONSOLE=%s.", conf->console.path); | |
650 | ||
651 | if (conf->console.log_path && setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1)) | |
652 | SYSERROR("Failed to set environment variable: LXC_CONSOLE_LOGPATH=%s.", conf->console.log_path); | |
653 | ||
654 | if (setenv("LXC_CGNS_AWARE", "1", 1)) | |
655 | SYSERROR("Failed to set environment variable LXC_CGNS_AWARE=1."); | |
b8f88d9b | 656 | |
4a03ded4 CB |
657 | loglevel = lxc_log_priority_to_string(lxc_log_get_level()); |
658 | if (setenv("LXC_LOG_LEVEL", loglevel, 1)) | |
659 | SYSERROR("Failed to set environment variable LXC_LOG_LEVEL=%s", loglevel); | |
a2c09be0 CB |
660 | |
661 | if (conf->hooks_version == 0) | |
662 | ret = setenv("LXC_HOOK_VERSION", "0", 1); | |
663 | else | |
664 | ret = setenv("LXC_HOOK_VERSION", "1", 1); | |
665 | if (ret < 0) | |
666 | SYSERROR("Failed to set environment variable LXC_HOOK_VERSION=%u", conf->hooks_version); | |
408da065 | 667 | /* End of environment variable setup for hooks. */ |
f7bee6c6 | 668 | |
dbc9832d CB |
669 | TRACE("set environment variables"); |
670 | ||
14a7b0f9 | 671 | if (run_lxc_hooks(name, "pre-start", conf, NULL)) { |
408da065 | 672 | ERROR("Failed to run lxc.hook.pre-start for container \"%s\".", name); |
773fb9ca SH |
673 | goto out_aborting; |
674 | } | |
dbc9832d | 675 | TRACE("ran pre-start hooks"); |
26ddeedd | 676 | |
408da065 CB |
677 | /* The signal fd has to be created before forking otherwise if the child |
678 | * process exits before we setup the signal fd, the event will be lost | |
679 | * and the command will be stuck. | |
680 | */ | |
83ee7875 | 681 | handler->sigfd = setup_signal_fd(&handler->oldmask); |
59eb99ba | 682 | if (handler->sigfd < 0) { |
408da065 | 683 | ERROR("Failed to setup SIGCHLD fd handler."); |
b5159817 DE |
684 | goto out_delete_tty; |
685 | } | |
dbc9832d | 686 | TRACE("set up signal fd"); |
b5159817 | 687 | |
408da065 | 688 | /* Do this after setting up signals since it might unblock SIGWINCH. */ |
b5159817 | 689 | if (lxc_console_create(conf)) { |
408da065 | 690 | ERROR("Failed to create console for container \"%s\".", name); |
b5159817 | 691 | goto out_restore_sigmask; |
b0a33c1e | 692 | } |
dbc9832d | 693 | TRACE("created console"); |
b0a33c1e | 694 | |
54117de5 | 695 | if (lxc_ttys_shift_ids(conf) < 0) { |
408da065 | 696 | ERROR("Failed to shift tty into container."); |
c4d10a05 SH |
697 | goto out_restore_sigmask; |
698 | } | |
dbc9832d | 699 | TRACE("shifted tty ids"); |
c4d10a05 | 700 | |
dbc9832d | 701 | INFO("container \"%s\" is initialized", name); |
aa460476 | 702 | return 0; |
59eb99ba | 703 | |
b5159817 DE |
704 | out_restore_sigmask: |
705 | sigprocmask(SIG_SETMASK, &handler->oldmask, NULL); | |
59eb99ba | 706 | out_delete_tty: |
fae349da | 707 | lxc_delete_tty(&conf->tty_info); |
59eb99ba | 708 | out_aborting: |
25c2aca5 | 709 | lxc_set_state(name, handler, ABORTING); |
d2e30e99 DE |
710 | out_close_maincmd_fd: |
711 | close(conf->maincmd_fd); | |
712 | conf->maincmd_fd = -1; | |
aa460476 | 713 | return -1; |
59eb99ba DL |
714 | } |
715 | ||
3b72c4a0 | 716 | void lxc_fini(const char *name, struct lxc_handler *handler) |
59eb99ba | 717 | { |
b3286b62 | 718 | int i, rc; |
18b3b9c1 | 719 | pid_t self; |
dbc9832d | 720 | struct lxc_list *cur, *next; |
dbc9832d | 721 | char *namespaces[LXC_NS_MAX + 1]; |
b3286b62 | 722 | size_t namespace_count = 0; |
b6b2b194 | 723 | |
408da065 CB |
724 | /* The STOPPING state is there for future cleanup code which can take |
725 | * awhile. | |
59eb99ba | 726 | */ |
25c2aca5 | 727 | lxc_set_state(name, handler, STOPPING); |
b6b2b194 | 728 | |
0059379f | 729 | self = lxc_raw_getpid(); |
b3286b62 | 730 | for (i = 0; i < LXC_NS_MAX; i++) { |
18b3b9c1 CB |
731 | if (handler->nsfd[i] < 0) |
732 | continue; | |
733 | ||
734 | if (handler->conf->hooks_version == 0) | |
735 | rc = asprintf(&namespaces[namespace_count], | |
736 | "%s:/proc/%d/fd/%d", ns_info[i].proc_name, | |
737 | self, handler->nsfd[i]); | |
738 | else | |
739 | rc = asprintf(&namespaces[namespace_count], | |
740 | "/proc/%d/fd/%d", self, handler->nsfd[i]); | |
741 | if (rc == -1) { | |
742 | SYSERROR("Failed to allocate memory."); | |
743 | break; | |
b3286b62 | 744 | } |
18b3b9c1 CB |
745 | |
746 | if (handler->conf->hooks_version == 0) { | |
747 | namespace_count++; | |
748 | continue; | |
749 | } | |
750 | ||
751 | rc = setenv(ns_info[i].env_name, namespaces[namespace_count], 1); | |
752 | if (rc < 0) | |
753 | SYSERROR("Failed to set environment variable %s=%s", | |
754 | ns_info[i].env_name, namespaces[namespace_count]); | |
755 | else | |
756 | TRACE("Set environment variable %s=%s", | |
757 | ns_info[i].env_name, namespaces[namespace_count]); | |
758 | ||
759 | namespace_count++; | |
b3286b62 WB |
760 | } |
761 | namespaces[namespace_count] = NULL; | |
c154af98 | 762 | |
408da065 CB |
763 | if (handler->conf->reboot && setenv("LXC_TARGET", "reboot", 1)) |
764 | SYSERROR("Failed to set environment variable: LXC_TARGET=reboot."); | |
765 | ||
766 | if (!handler->conf->reboot && setenv("LXC_TARGET", "stop", 1)) | |
767 | SYSERROR("Failed to set environment variable: LXC_TARGET=stop."); | |
c154af98 | 768 | |
18b3b9c1 | 769 | if (handler->conf->hooks_version == 0) |
14a7b0f9 | 770 | rc = run_lxc_hooks(name, "stop", handler->conf, namespaces); |
18b3b9c1 | 771 | else |
14a7b0f9 | 772 | rc = run_lxc_hooks(name, "stop", handler->conf, NULL); |
c154af98 | 773 | |
b3286b62 WB |
774 | while (namespace_count--) |
775 | free(namespaces[namespace_count]); | |
28d9e29e | 776 | |
b6b2b194 | 777 | for (i = 0; i < LXC_NS_MAX; i++) { |
28d9e29e CB |
778 | if (handler->nsfd[i] < 0) |
779 | continue; | |
738d0deb | 780 | |
28d9e29e CB |
781 | close(handler->nsfd[i]); |
782 | handler->nsfd[i] = -1; | |
738d0deb CB |
783 | } |
784 | ||
4288b79f CB |
785 | cgroup_destroy(handler); |
786 | ||
f893d898 CB |
787 | /* This function will try to connect to the legacy lxc-monitord state |
788 | * server and only exists for backwards compatibility. | |
789 | */ | |
790 | lxc_monitor_send_state(name, STOPPED, handler->lxcpath); | |
59eb99ba | 791 | |
d39b10eb | 792 | if (handler->conf->reboot == 0) { |
f893d898 CB |
793 | /* For all new state clients simply close the command socket. |
794 | * This will inform all state clients that the container is | |
795 | * STOPPED and also prevents a race between a open()/close() on | |
796 | * the command socket causing a new process to get ECONNREFUSED | |
797 | * because we haven't yet closed the command socket. | |
798 | */ | |
d39b10eb CB |
799 | close(handler->conf->maincmd_fd); |
800 | handler->conf->maincmd_fd = -1; | |
f893d898 CB |
801 | } else { |
802 | lxc_set_state(name, handler, STOPPED); | |
d39b10eb | 803 | } |
4288b79f | 804 | |
14a7b0f9 | 805 | if (run_lxc_hooks(name, "post-stop", handler->conf, NULL)) { |
408da065 | 806 | ERROR("Failed to run lxc.hook.post-stop for container \"%s\".", name); |
f3787121 CB |
807 | if (handler->conf->reboot) { |
808 | WARN("Container will be stopped instead of rebooted."); | |
809 | handler->conf->reboot = 0; | |
408da065 CB |
810 | if (setenv("LXC_TARGET", "stop", 1)) |
811 | WARN("Failed to set environment variable: LXC_TARGET=stop."); | |
f3787121 CB |
812 | } |
813 | } | |
26ddeedd | 814 | |
408da065 | 815 | /* Reset mask set by setup_signal_fd. */ |
8f64a3f6 | 816 | if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) |
408da065 | 817 | WARN("Failed to restore signal mask."); |
8f64a3f6 | 818 | |
b5159817 | 819 | lxc_console_delete(&handler->conf->console); |
b2431939 | 820 | lxc_delete_tty(&handler->conf->tty_info); |
dbc9832d | 821 | |
dbc9832d CB |
822 | /* The command socket is now closed, no more state clients can register |
823 | * themselves from now on. So free the list of state clients. | |
824 | */ | |
d39b10eb CB |
825 | lxc_list_for_each_safe(cur, &handler->conf->state_clients, next) { |
826 | struct lxc_state_client *client = cur->elem; | |
827 | ||
828 | /* Keep state clients that want to be notified about reboots. */ | |
829 | if ((handler->conf->reboot > 0) && (client->states[RUNNING] == 2)) | |
830 | continue; | |
831 | ||
dbc9832d CB |
832 | /* close state client socket */ |
833 | close(client->clientfd); | |
834 | lxc_list_del(cur); | |
835 | free(cur->elem); | |
836 | free(cur); | |
837 | } | |
838 | ||
c6012571 CB |
839 | if (handler->data_sock[0] != -1) { |
840 | close(handler->data_sock[0]); | |
841 | close(handler->data_sock[1]); | |
e8bd4e43 | 842 | } |
2c5f2ede CB |
843 | |
844 | if (handler->conf->ephemeral == 1 && handler->conf->reboot != 1) | |
28272964 | 845 | lxc_destroy_container_on_signal(handler, name); |
2c5f2ede | 846 | |
b2431939 | 847 | free(handler); |
59eb99ba DL |
848 | } |
849 | ||
735f2c6e | 850 | void lxc_abort(const char *name, struct lxc_handler *handler) |
59eb99ba | 851 | { |
73e608b2 SH |
852 | int ret, status; |
853 | ||
25c2aca5 | 854 | lxc_set_state(name, handler, ABORTING); |
7d9fb3e9 DL |
855 | if (handler->pid > 0) |
856 | kill(handler->pid, SIGKILL); | |
408da065 CB |
857 | while ((ret = waitpid(-1, &status, 0)) > 0) { |
858 | ; | |
859 | } | |
59eb99ba DL |
860 | } |
861 | ||
ffe1e01a | 862 | static int do_start(void *data) |
50e98013 | 863 | { |
5af9369b | 864 | int fd, ret; |
7c661726 | 865 | struct lxc_list *iterator; |
7a55c157 | 866 | char path[PATH_MAX]; |
fa3a5b22 | 867 | struct lxc_handler *handler = data; |
928b1f04 YT |
868 | bool have_cap_setgid; |
869 | uid_t new_uid; | |
870 | gid_t new_gid; | |
5af9369b | 871 | int devnull_fd = -1; |
50e98013 DL |
872 | |
873 | if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) { | |
408da065 | 874 | SYSERROR("Failed to set signal mask."); |
9d7f9e52 | 875 | return -1; |
50e98013 DL |
876 | } |
877 | ||
408da065 CB |
878 | /* This prctl must be before the synchro, so if the parent dies before |
879 | * we set the parent death signal, we will detect its death with the | |
880 | * synchro right after, otherwise we have a window where the parent can | |
881 | * exit before we set the pdeath signal leading to a unsupervized | |
882 | * container. | |
743ecd2e DL |
883 | */ |
884 | if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) { | |
408da065 | 885 | SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL."); |
743ecd2e DL |
886 | return -1; |
887 | } | |
888 | ||
3c22086f | 889 | lxc_sync_fini_parent(handler); |
50e98013 | 890 | |
408da065 | 891 | /* Don't leak the pinfd to the container. */ |
d39b10eb | 892 | if (handler->pinfd >= 0) |
0d03360a | 893 | close(handler->pinfd); |
2b0e17e4 | 894 | |
5b1e83cb SH |
895 | if (lxc_sync_wait_parent(handler, LXC_SYNC_STARTUP)) |
896 | return -1; | |
897 | ||
408da065 CB |
898 | /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See |
899 | * https://github.com/lxc/lxd/issues/1978. | |
900 | */ | |
5b1e83cb | 901 | if ((handler->clone_flags & (CLONE_NEWNET | CLONE_NEWUSER)) == |
408da065 | 902 | (CLONE_NEWNET | CLONE_NEWUSER)) { |
5b1e83cb SH |
903 | ret = unshare(CLONE_NEWNET); |
904 | if (ret < 0) { | |
408da065 | 905 | SYSERROR("Failed to unshare CLONE_NEWNET."); |
5b1e83cb SH |
906 | goto out_warn_father; |
907 | } | |
408da065 | 908 | INFO("Unshared CLONE_NEWNET."); |
5b1e83cb SH |
909 | } |
910 | ||
408da065 CB |
911 | /* Tell the parent task it can begin to configure the container and wait |
912 | * for it to finish. | |
3c22086f CLG |
913 | */ |
914 | if (lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE)) | |
9d7f9e52 | 915 | return -1; |
50e98013 | 916 | |
7ab1ba02 CB |
917 | if (lxc_network_recv_veth_names_from_parent(handler) < 0) { |
918 | ERROR("Failed to receive veth names from parent"); | |
658979c5 | 919 | goto out_warn_father; |
7ab1ba02 | 920 | } |
658979c5 | 921 | |
408da065 | 922 | /* If we are in a new user namespace, become root there to have |
d08f8d2f | 923 | * privilege over our namespace. |
f6d3e3e4 SH |
924 | */ |
925 | if (!lxc_list_empty(&handler->conf->id_map)) { | |
d08f8d2f | 926 | if (lxc_switch_uid_gid(0, 0) < 0) |
f6d3e3e4 | 927 | goto out_warn_father; |
d08f8d2f CB |
928 | |
929 | /* Drop groups only after we switched to a valid gid in the new | |
930 | * user namespace. | |
931 | */ | |
932 | if (lxc_setgroups(0, NULL) < 0) | |
c476bdce | 933 | goto out_warn_father; |
f6d3e3e4 SH |
934 | } |
935 | ||
99b71824 | 936 | if (access(handler->lxcpath, X_OK)) { |
c8154066 SH |
937 | print_top_failing_dir(handler->lxcpath); |
938 | goto out_warn_father; | |
939 | } | |
940 | ||
a91cde21 | 941 | ret = snprintf(path, sizeof(path), "%s/dev/null", handler->conf->rootfs.mount); |
408da065 | 942 | if (ret < 0 || ret >= sizeof(path)) |
7a55c157 | 943 | goto out_warn_father; |
7a55c157 TA |
944 | |
945 | /* In order to checkpoint restore, we need to have everything in the | |
946 | * same mount namespace. However, some containers may not have a | |
947 | * reasonable /dev (in particular, they may not have /dev/null), so we | |
948 | * can't set init's std fds to /dev/null by opening it from inside the | |
949 | * container. | |
950 | * | |
951 | * If that's the case, fall back to using the host's /dev/null. This | |
952 | * means that migration won't work, but at least we won't spew output | |
953 | * where it isn't wanted. | |
954 | */ | |
955 | if (handler->backgrounded && !handler->conf->autodev && access(path, F_OK) < 0) { | |
c44de748 AM |
956 | devnull_fd = open_devnull(); |
957 | ||
958 | if (devnull_fd < 0) | |
959 | goto out_warn_father; | |
408da065 CB |
960 | WARN("Using /dev/null from the host for container init's " |
961 | "standard file descriptors. Migration will not work."); | |
c44de748 AM |
962 | } |
963 | ||
408da065 | 964 | /* Ask father to setup cgroups and wait for him to finish. */ |
544a48a0 | 965 | if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP)) |
c44de748 | 966 | goto out_error; |
544a48a0 | 967 | |
deefdf8a CB |
968 | /* Unshare cgroup namespace after we have setup our cgroups. If we do it |
969 | * earlier we end up with a wrong view of /proc/self/cgroup. For | |
970 | * example, assume we unshare(CLONE_NEWCGROUP) first, and then create | |
971 | * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then | |
972 | * /proc/self/cgroup would show us: | |
973 | * | |
974 | * 8:cpuset:/lxc/c | |
975 | * | |
976 | * whereas it should actually show | |
977 | * | |
978 | * 8:cpuset:/ | |
979 | */ | |
5af9369b | 980 | if (handler->clone_flags & CLONE_NEWCGROUP) { |
deefdf8a CB |
981 | if (unshare(CLONE_NEWCGROUP) < 0) { |
982 | INFO("Failed to unshare CLONE_NEWCGROUP."); | |
983 | goto out_warn_father; | |
984 | } | |
985 | INFO("Unshared CLONE_NEWCGROUP."); | |
986 | } | |
987 | ||
149857af CB |
988 | /* Add the requested environment variables to the current environment to |
989 | * allow them to be used by the various hooks, such as the start hook | |
990 | * above. | |
98ff08ca | 991 | */ |
98ff08ca CB |
992 | lxc_list_for_each(iterator, &handler->conf->environment) { |
993 | if (putenv((char *)iterator->elem)) { | |
994 | SYSERROR("Failed to set environment variable: %s.", (char *)iterator->elem); | |
995 | goto out_warn_father; | |
996 | } | |
997 | } | |
998 | ||
f4152036 | 999 | /* Setup the container, ip, names, utsname, ... */ |
3b988b33 | 1000 | ret = lxc_setup(handler); |
7729f8e5 CB |
1001 | close(handler->data_sock[1]); |
1002 | if (ret < 0) { | |
f4152036 | 1003 | ERROR("Failed to setup container \"%s\".", handler->name); |
5af9369b | 1004 | close(handler->data_sock[0]); |
f4152036 CB |
1005 | goto out_warn_father; |
1006 | } | |
1007 | ||
5af9369b | 1008 | if (handler->clone_flags & CLONE_NEWCGROUP) { |
0059379f | 1009 | fd = lxc_preserve_ns(lxc_raw_getpid(), "cgroup"); |
5af9369b CB |
1010 | if (fd < 0) { |
1011 | ERROR("%s - Failed to preserve cgroup namespace", strerror(errno)); | |
1012 | close(handler->data_sock[0]); | |
1013 | goto out_warn_father; | |
1014 | } | |
1015 | ||
1016 | ret = lxc_abstract_unix_send_fds(handler->data_sock[0], &fd, 1, NULL, 0); | |
1017 | close(fd); | |
1018 | if (ret < 0) { | |
1019 | ERROR("%s - Failed to preserve cgroup namespace", strerror(errno)); | |
1020 | close(handler->data_sock[0]); | |
1021 | goto out_warn_father; | |
1022 | } | |
1023 | } | |
1024 | close(handler->data_sock[0]); | |
1025 | ||
408da065 | 1026 | /* Set the label to change to when we exec(2) the container's init. */ |
7aff4f43 | 1027 | if (lsm_process_label_set(NULL, handler->conf, 1, 1) < 0) |
e075f5d9 | 1028 | goto out_warn_father; |
5112cd70 | 1029 | |
029cdff5 | 1030 | /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it |
408da065 CB |
1031 | * before we aren't allowed anymore. |
1032 | */ | |
029cdff5 CB |
1033 | if (handler->conf->no_new_privs) { |
1034 | if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) { | |
1035 | SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges."); | |
1036 | goto out_warn_father; | |
1037 | } | |
1038 | DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges."); | |
1039 | } | |
1040 | ||
0d9acb99 DE |
1041 | /* Some init's such as busybox will set sane tty settings on stdin, |
1042 | * stdout, stderr which it thinks is the console. We already set them | |
1043 | * the way we wanted on the real terminal, and we want init to do its | |
408da065 CB |
1044 | * setup on its console ie. the pty allocated in lxc_console_create() so |
1045 | * make sure that that pty is stdin,stdout,stderr. | |
0d9acb99 | 1046 | */ |
5d113f65 CB |
1047 | if (handler->conf->console.slave >= 0) { |
1048 | if (handler->backgrounded || handler->conf->is_execute == 0) | |
1049 | ret = set_stdfds(handler->conf->console.slave); | |
1050 | else | |
1051 | ret = lxc_console_set_stdfds(handler->conf->console.slave); | |
1052 | if (ret < 0) { | |
c5b93afb LF |
1053 | ERROR("Failed to redirect std{in,out,err} to pty file " |
1054 | "descriptor %d", | |
1055 | handler->conf->console.slave); | |
1056 | goto out_warn_father; | |
1057 | } | |
5d113f65 | 1058 | } |
0d9acb99 | 1059 | |
408da065 | 1060 | /* If we mounted a temporary proc, then unmount it now. */ |
5112cd70 | 1061 | tmp_proc_unmount(handler->conf); |
e075f5d9 | 1062 | |
8f2c3a70 SH |
1063 | if (lxc_seccomp_load(handler->conf) != 0) |
1064 | goto out_warn_father; | |
1065 | ||
14a7b0f9 | 1066 | if (run_lxc_hooks(handler->name, "start", handler->conf, NULL)) { |
408da065 | 1067 | ERROR("Failed to run lxc.hook.start for container \"%s\".", handler->name); |
773fb9ca SH |
1068 | goto out_warn_father; |
1069 | } | |
fc25b815 | 1070 | |
773fb9ca | 1071 | close(handler->sigfd); |
26ddeedd | 1072 | |
7a55c157 TA |
1073 | if (devnull_fd < 0) { |
1074 | devnull_fd = open_devnull(); | |
1075 | ||
1076 | if (devnull_fd < 0) | |
1077 | goto out_warn_father; | |
1078 | } | |
1079 | ||
c5b93afb LF |
1080 | if (handler->conf->console.slave < 0 && handler->backgrounded) |
1081 | if (set_stdfds(devnull_fd) < 0) { | |
1082 | ERROR("Failed to redirect std{in,out,err} to " | |
1083 | "\"/dev/null\""); | |
1084 | goto out_warn_father; | |
1085 | } | |
507cee36 | 1086 | |
c44de748 AM |
1087 | if (devnull_fd >= 0) { |
1088 | close(devnull_fd); | |
1089 | devnull_fd = -1; | |
1090 | } | |
1091 | ||
8c9a7665 TA |
1092 | setsid(); |
1093 | ||
3c491553 L |
1094 | if (handler->conf->init_cwd && chdir(handler->conf->init_cwd)) { |
1095 | SYSERROR("Could not change directory to \"%s\"", handler->conf->init_cwd); | |
1096 | goto out_warn_father; | |
1097 | } | |
1098 | ||
f4152036 CB |
1099 | if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP_LIMITS)) |
1100 | goto out_warn_father; | |
1101 | ||
149857af CB |
1102 | /* Reset the environment variables the user requested in a clear |
1103 | * environment. | |
1104 | */ | |
1105 | if (clearenv()) { | |
1106 | SYSERROR("Failed to clear environment."); | |
1107 | /* Don't error out though. */ | |
1108 | } | |
1109 | ||
1110 | lxc_list_for_each(iterator, &handler->conf->environment) { | |
1111 | if (putenv((char *)iterator->elem)) { | |
1112 | SYSERROR("Failed to set environment variable: %s.", (char *)iterator->elem); | |
1113 | goto out_warn_father; | |
1114 | } | |
1115 | } | |
1116 | ||
98ff08ca CB |
1117 | if (putenv("container=lxc")) { |
1118 | SYSERROR("Failed to set environment variable: container=lxc."); | |
1119 | goto out_warn_father; | |
1120 | } | |
1121 | ||
1122 | if (handler->conf->pty_names) { | |
1123 | if (putenv(handler->conf->pty_names)) { | |
1124 | SYSERROR("Failed to set environment variable for container ptys."); | |
1125 | goto out_warn_father; | |
1126 | } | |
1127 | } | |
1128 | ||
76bdf299 CB |
1129 | /* The container has been setup. We can now switch to an unprivileged |
1130 | * uid/gid. | |
1131 | */ | |
928b1f04 YT |
1132 | new_uid = handler->conf->init_uid; |
1133 | new_gid = handler->conf->init_gid; | |
76bdf299 | 1134 | |
928b1f04 YT |
1135 | /* If we are in a new user namespace we already dropped all |
1136 | * groups when we switched to root in the new user namespace | |
1137 | * further above. Only drop groups if we can, so ensure that we | |
1138 | * have necessary privilege. | |
1139 | */ | |
1140 | #if HAVE_LIBCAP | |
1141 | have_cap_setgid = lxc_proc_cap_is_set(CAP_SETGID, CAP_EFFECTIVE); | |
1142 | #else | |
1143 | have_cap_setgid = false; | |
1144 | #endif | |
1145 | if (lxc_list_empty(&handler->conf->id_map) && have_cap_setgid) { | |
1146 | if (lxc_setgroups(0, NULL) < 0) | |
76bdf299 CB |
1147 | goto out_warn_father; |
1148 | } | |
1149 | ||
928b1f04 YT |
1150 | if (lxc_switch_uid_gid(new_uid, new_gid) < 0) |
1151 | goto out_warn_father; | |
1152 | ||
408da065 CB |
1153 | /* After this call, we are in error because this ops should not return |
1154 | * as it execs. | |
1155 | */ | |
c4ea60df | 1156 | handler->ops->start(handler, handler->data); |
50e98013 DL |
1157 | |
1158 | out_warn_father: | |
408da065 CB |
1159 | /* We want the parent to know something went wrong, so we return a |
1160 | * special error code. | |
1161 | */ | |
d1ccb562 | 1162 | lxc_sync_wake_parent(handler, LXC_SYNC_ERROR); |
c44de748 AM |
1163 | |
1164 | out_error: | |
1165 | if (devnull_fd >= 0) | |
1166 | close(devnull_fd); | |
1167 | ||
50e98013 DL |
1168 | return -1; |
1169 | } | |
1170 | ||
ae467c54 | 1171 | static int lxc_recv_ttys_from_child(struct lxc_handler *handler) |
e8bd4e43 | 1172 | { |
ae467c54 | 1173 | int i; |
ae467c54 CB |
1174 | struct lxc_pty_info *pty_info; |
1175 | int ret = -1; | |
c6012571 | 1176 | int sock = handler->data_sock[1]; |
e8bd4e43 | 1177 | struct lxc_conf *conf = handler->conf; |
e8bd4e43 SH |
1178 | struct lxc_tty_info *tty_info = &conf->tty_info; |
1179 | ||
1180 | if (!conf->tty) | |
1181 | return 0; | |
1182 | ||
408da065 CB |
1183 | tty_info->pty_info = malloc(sizeof(*tty_info->pty_info) * conf->tty); |
1184 | if (!tty_info->pty_info) | |
e8bd4e43 | 1185 | return -1; |
e8bd4e43 | 1186 | |
672c1e58 CB |
1187 | for (i = 0; i < conf->tty; i++) { |
1188 | int ttyfds[2]; | |
1189 | ||
1190 | ret = lxc_abstract_unix_recv_fds(sock, ttyfds, 2, NULL, 0); | |
1191 | if (ret < 0) | |
1192 | break; | |
ae467c54 | 1193 | |
672c1e58 | 1194 | pty_info = &tty_info->pty_info[i]; |
e8bd4e43 | 1195 | pty_info->busy = 0; |
672c1e58 CB |
1196 | pty_info->master = ttyfds[0]; |
1197 | pty_info->slave = ttyfds[1]; | |
1198 | TRACE("Received pty with master fd %d and slave fd %d from " | |
ae467c54 | 1199 | "parent", pty_info->master, pty_info->slave); |
e8bd4e43 | 1200 | } |
ae467c54 | 1201 | if (ret < 0) |
672c1e58 | 1202 | ERROR("Failed to receive %d ttys from child: %s", conf->tty, |
ae467c54 CB |
1203 | strerror(errno)); |
1204 | else | |
672c1e58 CB |
1205 | TRACE("Received %d ttys from child", conf->tty); |
1206 | ||
1207 | tty_info->nbtty = conf->tty; | |
ae467c54 CB |
1208 | |
1209 | return ret; | |
e8bd4e43 SH |
1210 | } |
1211 | ||
5af9369b | 1212 | int resolve_clone_flags(struct lxc_handler *handler) |
f813849c | 1213 | { |
03df7ab5 | 1214 | handler->clone_flags = CLONE_NEWNS; |
f813849c | 1215 | |
fa3a5b22 CB |
1216 | if (!handler->conf->inherit_ns[LXC_NS_USER]) { |
1217 | if (!lxc_list_empty(&handler->conf->id_map)) | |
1218 | handler->clone_flags |= CLONE_NEWUSER; | |
1219 | } else { | |
1220 | INFO("Inheriting user namespace"); | |
1221 | } | |
f813849c | 1222 | |
28d9e29e | 1223 | if (!handler->conf->inherit_ns[LXC_NS_NET]) { |
408da065 | 1224 | if (!lxc_requests_empty_network(handler)) |
f813849c TA |
1225 | handler->clone_flags |= CLONE_NEWNET; |
1226 | } else { | |
28d9e29e | 1227 | INFO("Inheriting net namespace"); |
f813849c TA |
1228 | } |
1229 | ||
28d9e29e | 1230 | if (!handler->conf->inherit_ns[LXC_NS_IPC]) |
f813849c | 1231 | handler->clone_flags |= CLONE_NEWIPC; |
408da065 | 1232 | else |
28d9e29e | 1233 | INFO("Inheriting ipc namespace"); |
f813849c | 1234 | |
28d9e29e | 1235 | if (!handler->conf->inherit_ns[LXC_NS_UTS]) |
f813849c | 1236 | handler->clone_flags |= CLONE_NEWUTS; |
408da065 | 1237 | else |
28d9e29e | 1238 | INFO("Inheriting uts namespace"); |
03df7ab5 | 1239 | |
28d9e29e | 1240 | if (!handler->conf->inherit_ns[LXC_NS_PID]) |
03df7ab5 L |
1241 | handler->clone_flags |= CLONE_NEWPID; |
1242 | else | |
28d9e29e | 1243 | INFO("Inheriting pid namespace"); |
5af9369b CB |
1244 | |
1245 | if (cgns_supported()) { | |
1246 | if (!handler->conf->inherit_ns[LXC_NS_CGROUP]) | |
1247 | handler->clone_flags |= CLONE_NEWCGROUP; | |
1248 | else | |
1249 | INFO("Inheriting cgroup namespace"); | |
1250 | } else if (handler->conf->inherit_ns[LXC_NS_CGROUP]) { | |
1251 | return -EINVAL; | |
1252 | } | |
1253 | ||
1254 | return 0; | |
f813849c TA |
1255 | } |
1256 | ||
8deca6c9 CB |
1257 | /* Note that this function is used with clone(CLONE_VM). Some glibc versions |
1258 | * used to reset the pid/tid to -1 when CLONE_VM was used without CLONE_THREAD. | |
1259 | * But since the memory between parent and child is shared on CLONE_VM this | |
1260 | * would invalidate the getpid() cache that glibc used to maintain and so | |
1261 | * getpid() in the child would return the parent's pid. This is all fixed in | |
1262 | * newer glibc versions where the getpid() cache is removed and the pid/tid is | |
1263 | * not reset anymore. | |
1264 | * However, if for whatever reason you - dear commiter - somehow need to get the | |
1265 | * pid of the dummy intermediate process for do_share_ns() you need to call | |
0059379f CB |
1266 | * lxc_raw_getpid(). The next lxc_raw_clone() call does not employ CLONE_VM and |
1267 | * will be fine. | |
8deca6c9 CB |
1268 | */ |
1269 | static inline int do_share_ns(void *arg) | |
fa3a5b22 | 1270 | { |
8deca6c9 | 1271 | int i, flags, ret; |
fa3a5b22 CB |
1272 | struct lxc_handler *handler = arg; |
1273 | ||
8deca6c9 CB |
1274 | for (i = 0; i < LXC_NS_MAX; i++) { |
1275 | if (handler->nsfd[i] < 0) | |
1276 | continue; | |
fa3a5b22 | 1277 | |
8deca6c9 | 1278 | ret = setns(handler->nsfd[i], 0); |
fa3a5b22 | 1279 | if (ret < 0) |
8deca6c9 | 1280 | return -1; |
fa3a5b22 | 1281 | |
8deca6c9 | 1282 | DEBUG("Inherited %s namespace", ns_info[i].proc_name); |
fa3a5b22 CB |
1283 | } |
1284 | ||
8deca6c9 CB |
1285 | flags = handler->on_clone_flags; |
1286 | flags |= CLONE_PARENT; | |
ac2ba696 | 1287 | handler->pid = lxc_raw_clone_cb(do_start, handler, flags); |
8deca6c9 | 1288 | if (handler->pid < 0) |
fa3a5b22 CB |
1289 | return -1; |
1290 | ||
8deca6c9 | 1291 | return 0; |
fa3a5b22 CB |
1292 | } |
1293 | ||
480588e6 CB |
1294 | /* lxc_spawn() performs crucial setup tasks and clone()s the new process which |
1295 | * exec()s the requested container binary. | |
1296 | * Note that lxc_spawn() runs in the parent namespaces. Any operations performed | |
1297 | * right here should be double checked if they'd pose a security risk. (For | |
1298 | * example, any {u}mount() operations performed here will be reflected on the | |
1299 | * host!) | |
1300 | */ | |
74a3920a | 1301 | static int lxc_spawn(struct lxc_handler *handler) |
59eb99ba | 1302 | { |
8deca6c9 | 1303 | int i, ret; |
08dd2805 | 1304 | char pidstr[20]; |
57927bf2 CB |
1305 | bool wants_to_map_ids; |
1306 | struct lxc_list *id_map; | |
6e5fc7a5 | 1307 | const char *name = handler->name; |
28d9e29e | 1308 | const char *lxcpath = handler->lxcpath; |
8deca6c9 | 1309 | bool cgroups_connected = false, share_ns = false; |
28d9e29e | 1310 | struct lxc_conf *conf = handler->conf; |
9f30a190 | 1311 | |
28d9e29e | 1312 | id_map = &conf->id_map; |
57927bf2 | 1313 | wants_to_map_ids = !lxc_list_empty(id_map); |
2f2623ec | 1314 | |
28d9e29e CB |
1315 | for (i = 0; i < LXC_NS_MAX; i++) { |
1316 | if (!conf->inherit_ns[i]) | |
1317 | continue; | |
1318 | ||
28d9e29e CB |
1319 | handler->nsfd[i] = lxc_inherit_namespace(conf->inherit_ns[i], lxcpath, ns_info[i].proc_name); |
1320 | if (handler->nsfd[i] < 0) | |
1321 | return -1; | |
fa3a5b22 | 1322 | |
8deca6c9 | 1323 | share_ns = true; |
28d9e29e | 1324 | } |
50e98013 | 1325 | |
3c22086f | 1326 | if (lxc_sync_init(handler)) |
9d7f9e52 | 1327 | return -1; |
0ad19a3f | 1328 | |
b9f522c5 | 1329 | ret = socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0, |
790255cf | 1330 | handler->data_sock); |
c6012571 | 1331 | if (ret < 0) { |
e8bd4e43 SH |
1332 | lxc_sync_fini(handler); |
1333 | return -1; | |
1334 | } | |
1335 | ||
5af9369b CB |
1336 | ret = resolve_clone_flags(handler); |
1337 | if (ret < 0) { | |
1338 | lxc_sync_fini(handler); | |
1339 | return -1; | |
1340 | } | |
9f30a190 | 1341 | |
f813849c | 1342 | if (handler->clone_flags & CLONE_NEWNET) { |
28d9e29e | 1343 | if (!lxc_list_empty(&conf->network)) { |
26b797f3 | 1344 | |
9f30a190 MM |
1345 | /* Find gateway addresses from the link device, which is |
1346 | * no longer accessible inside the container. Do this | |
1347 | * before creating network interfaces, since goto | |
408da065 CB |
1348 | * out_delete_net does not work before lxc_clone. |
1349 | */ | |
9f30a190 | 1350 | if (lxc_find_gateway_addresses(handler)) { |
408da065 | 1351 | ERROR("Failed to find gateway addresses."); |
9f30a190 MM |
1352 | lxc_sync_fini(handler); |
1353 | return -1; | |
1354 | } | |
1355 | ||
408da065 CB |
1356 | /* That should be done before the clone because we will |
1357 | * fill the netdev index and use them in the child. | |
9f30a190 | 1358 | */ |
811ef482 | 1359 | if (lxc_create_network_priv(handler)) { |
408da065 | 1360 | ERROR("Failed to create the network."); |
9f30a190 MM |
1361 | lxc_sync_fini(handler); |
1362 | return -1; | |
1363 | } | |
19a26f82 | 1364 | } |
82d5ae15 DL |
1365 | } |
1366 | ||
d4ef7c50 | 1367 | if (!cgroup_init(handler)) { |
408da065 | 1368 | ERROR("Failed initializing cgroup support."); |
33ad9f1a CS |
1369 | goto out_delete_net; |
1370 | } | |
1371 | ||
7e4dfe0b SH |
1372 | cgroups_connected = true; |
1373 | ||
d4ef7c50 | 1374 | if (!cgroup_create(handler)) { |
408da065 | 1375 | ERROR("Failed creating cgroups."); |
47d8fb3b CS |
1376 | goto out_delete_net; |
1377 | } | |
1378 | ||
408da065 CB |
1379 | /* If the rootfs is not a blockdev, prevent the container from marking |
1380 | * it readonly. | |
1381 | * If the container is unprivileged then skip rootfs pinning. | |
0c547523 | 1382 | */ |
0ee35059 | 1383 | if (!wants_to_map_ids) { |
28d9e29e | 1384 | handler->pinfd = pin_rootfs(conf->rootfs.path); |
5e32a990 | 1385 | if (handler->pinfd == -1) |
408da065 | 1386 | INFO("Failed to pin the rootfs for container \"%s\".", handler->name); |
5e32a990 | 1387 | } |
0c547523 | 1388 | |
408da065 | 1389 | /* Create a process in a new set of namespaces. */ |
8deca6c9 | 1390 | handler->on_clone_flags = handler->clone_flags; |
408da065 CB |
1391 | if (handler->clone_flags & CLONE_NEWUSER) { |
1392 | /* If CLONE_NEWUSER and CLONE_NEWNET was requested, we need to | |
1393 | * clone a new user namespace first and only later unshare our | |
1394 | * network namespace to ensure that network devices ownership is | |
1395 | * set up correctly. | |
1396 | */ | |
8deca6c9 | 1397 | handler->on_clone_flags &= ~CLONE_NEWNET; |
408da065 | 1398 | } |
5af9369b CB |
1399 | /* The cgroup namespace gets unshare()ed not clone()ed. */ |
1400 | handler->on_clone_flags &= ~CLONE_NEWCGROUP; | |
732375f5 | 1401 | |
4e232466 CB |
1402 | if (share_ns) { |
1403 | pid_t attacher_pid; | |
1404 | ||
1405 | attacher_pid = lxc_clone(do_share_ns, handler, CLONE_VFORK | CLONE_VM | CLONE_FILES); | |
1406 | if (attacher_pid < 0) { | |
1407 | SYSERROR(LXC_CLONE_ERROR); | |
1408 | goto out_delete_net; | |
1409 | } | |
1410 | ||
1411 | ret = wait_for_pid(attacher_pid); | |
1412 | if (ret < 0) { | |
1413 | SYSERROR("Intermediate process failed"); | |
1414 | goto out_delete_net; | |
1415 | } | |
1416 | } else { | |
ac2ba696 | 1417 | handler->pid = lxc_raw_clone_cb(do_start, handler, handler->on_clone_flags); |
4e232466 CB |
1418 | } |
1419 | if (handler->pid < 0) { | |
1420 | SYSERROR(LXC_CLONE_ERROR); | |
7fef7a06 | 1421 | goto out_delete_net; |
0ad19a3f | 1422 | } |
4e232466 | 1423 | |
fa3a5b22 | 1424 | TRACE("Cloned child process %d", handler->pid); |
1bd8d726 | 1425 | |
9662e444 | 1426 | for (i = 0; i < LXC_NS_MAX; i++) |
8deca6c9 | 1427 | if (handler->on_clone_flags & ns_info[i].clone_flag) |
28d9e29e | 1428 | INFO("Cloned %s", ns_info[i].flag_name); |
0ad19a3f | 1429 | |
fa3a5b22 | 1430 | if (!preserve_ns(handler->nsfd, handler->clone_flags & ~CLONE_NEWNET, handler->pid)) { |
28d9e29e CB |
1431 | ERROR("Failed to preserve cloned namespaces for lxc.hook.stop"); |
1432 | goto out_delete_net; | |
1433 | } | |
b6b2b194 | 1434 | |
3c22086f CLG |
1435 | lxc_sync_fini_child(handler); |
1436 | ||
408da065 CB |
1437 | /* Map the container uids. The container became an invalid userid the |
1438 | * moment it was cloned with CLONE_NEWUSER. This call doesn't change | |
1439 | * anything immediately, but allows the container to setuid(0) (0 being | |
1440 | * mapped to something else on the host.) later to become a valid uid | |
1441 | * again. | |
1442 | */ | |
fa3a5b22 CB |
1443 | if (wants_to_map_ids) { |
1444 | if (!handler->conf->inherit_ns[LXC_NS_USER]) { | |
1445 | ret = lxc_map_ids(id_map, handler->pid); | |
1446 | if (ret < 0) { | |
1447 | ERROR("Failed to set up id mapping."); | |
1448 | goto out_delete_net; | |
1449 | } | |
1450 | } | |
5b1e83cb SH |
1451 | } |
1452 | ||
9e2bdae4 | 1453 | if (lxc_sync_wake_child(handler, LXC_SYNC_STARTUP)) |
5b1e83cb | 1454 | goto out_delete_net; |
5b1e83cb | 1455 | |
9e2bdae4 | 1456 | if (lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE)) |
5b1e83cb | 1457 | goto out_delete_net; |
0ad19a3f | 1458 | |
d4ef7c50 | 1459 | if (!cgroup_create_legacy(handler)) { |
408da065 | 1460 | ERROR("Failed to setup legacy cgroups for container \"%s\".", name); |
ae5c8b8e | 1461 | goto out_delete_net; |
33ad9f1a | 1462 | } |
9daf6f5d | 1463 | if (!cgroup_setup_limits(handler, false)) { |
408da065 | 1464 | ERROR("Failed to setup cgroup limits for container \"%s\".", name); |
6031a6e5 DE |
1465 | goto out_delete_net; |
1466 | } | |
1467 | ||
d4ef7c50 | 1468 | if (!cgroup_enter(handler)) |
7fef7a06 | 1469 | goto out_delete_net; |
218d4250 | 1470 | |
0996e18a SH |
1471 | if (!cgroup_chown(handler)) |
1472 | goto out_delete_net; | |
1473 | ||
fa3a5b22 CB |
1474 | /* Now we're ready to preserve the network namespace */ |
1475 | ret = lxc_preserve_ns(handler->pid, "net"); | |
1476 | if (ret < 0) { | |
1477 | ERROR("%s - Failed to preserve net namespace", strerror(errno)); | |
1478 | goto out_delete_net; | |
1479 | } | |
1480 | handler->nsfd[LXC_NS_NET] = ret; | |
1481 | DEBUG("Preserved net namespace via fd %d", ret); | |
1482 | ||
408da065 | 1483 | /* Create the network configuration. */ |
d5088cf2 | 1484 | if (handler->clone_flags & CLONE_NEWNET) { |
74c6e2b0 CB |
1485 | if (lxc_network_move_created_netdev_priv(handler->lxcpath, |
1486 | handler->name, | |
28d9e29e | 1487 | &conf->network, |
74c6e2b0 | 1488 | handler->pid)) { |
408da065 | 1489 | ERROR("Failed to create the configured network."); |
7fef7a06 | 1490 | goto out_delete_net; |
82d5ae15 | 1491 | } |
74c6e2b0 CB |
1492 | |
1493 | if (lxc_create_network_unpriv(handler->lxcpath, handler->name, | |
28d9e29e | 1494 | &conf->network, |
74c6e2b0 CB |
1495 | handler->pid)) { |
1496 | ERROR("Failed to create the configured network."); | |
1497 | goto out_delete_net; | |
1498 | } | |
0ad19a3f | 1499 | } |
1500 | ||
7ab1ba02 CB |
1501 | if (lxc_network_send_veth_names_to_child(handler) < 0) { |
1502 | ERROR("Failed to send veth names to child"); | |
1503 | goto out_delete_net; | |
658979c5 SH |
1504 | } |
1505 | ||
61d7a733 YT |
1506 | if (!lxc_list_empty(&conf->procs)) { |
1507 | ret = setup_proc_filesystem(&conf->procs, handler->pid); | |
1508 | if (ret < 0) | |
1509 | goto out_delete_net; | |
1510 | } | |
1511 | ||
408da065 CB |
1512 | /* Tell the child to continue its initialization. We'll get |
1513 | * LXC_SYNC_CGROUP when it is ready for us to setup cgroups. | |
3c22086f CLG |
1514 | */ |
1515 | if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE)) | |
544a48a0 SH |
1516 | goto out_delete_net; |
1517 | ||
28d9e29e | 1518 | if (!lxc_list_empty(&conf->limits) && setup_resource_limits(&conf->limits, handler->pid)) { |
c6d09e15 | 1519 | ERROR("failed to setup resource limits for '%s'", name); |
84ff3af7 | 1520 | goto out_delete_net; |
c6d09e15 WB |
1521 | } |
1522 | ||
f4152036 CB |
1523 | if (lxc_sync_barrier_child(handler, LXC_SYNC_CGROUP_UNSHARE)) |
1524 | goto out_delete_net; | |
1525 | ||
9daf6f5d | 1526 | if (!cgroup_setup_limits(handler, true)) { |
408da065 | 1527 | ERROR("Failed to setup the devices cgroup for container \"%s\".", name); |
b98f7d6e | 1528 | goto out_delete_net; |
544a48a0 | 1529 | } |
f4152036 | 1530 | TRACE("Set up cgroup device limits"); |
544a48a0 | 1531 | |
73d28d42 | 1532 | cgroup_disconnect(); |
7e4dfe0b | 1533 | cgroups_connected = false; |
73d28d42 | 1534 | |
08dd2805 SH |
1535 | snprintf(pidstr, 20, "%d", handler->pid); |
1536 | if (setenv("LXC_PID", pidstr, 1)) | |
1537 | SYSERROR("Failed to set environment variable: LXC_PID=%s.", pidstr); | |
1538 | ||
1539 | /* Run any host-side start hooks */ | |
14a7b0f9 | 1540 | if (run_lxc_hooks(name, "start-host", conf, NULL)) { |
08dd2805 SH |
1541 | ERROR("Failed to run lxc.hook.start-host for container \"%s\".", name); |
1542 | return -1; | |
1543 | } | |
1544 | ||
408da065 CB |
1545 | /* Tell the child to complete its initialization and wait for it to exec |
1546 | * or return an error. (The child will never return | |
08dd2805 | 1547 | * LXC_SYNC_READY_START+1. It will either close the sync pipe, causing |
408da065 CB |
1548 | * lxc_sync_barrier_child to return success, or return a different |
1549 | * value, causing us to error out). | |
544a48a0 | 1550 | */ |
08dd2805 | 1551 | if (lxc_sync_barrier_child(handler, LXC_SYNC_READY_START)) |
3c22086f | 1552 | return -1; |
0ad19a3f | 1553 | |
790255cf CB |
1554 | if (lxc_network_recv_name_and_ifindex_from_child(handler) < 0) { |
1555 | ERROR("Failed to receive names and ifindices for network " | |
1556 | "devices from child"); | |
1557 | goto out_delete_net; | |
1558 | } | |
1559 | ||
1560 | /* Now all networks are created, network devices are moved into place, | |
1561 | * and the correct names and ifindeces in the respective namespaces have | |
1562 | * been recorded. The corresponding structs have now all been filled. So | |
1563 | * log them for debugging purposes. | |
1564 | */ | |
28d9e29e | 1565 | lxc_log_configured_netdevs(conf); |
790255cf | 1566 | |
f4152036 CB |
1567 | /* Read tty fds allocated by child. */ |
1568 | if (lxc_recv_ttys_from_child(handler) < 0) { | |
1569 | ERROR("Failed to receive tty info from child process."); | |
1570 | goto out_delete_net; | |
1571 | } | |
1572 | ||
5af9369b CB |
1573 | if (handler->clone_flags & CLONE_NEWCGROUP) { |
1574 | ret = lxc_abstract_unix_recv_fds(handler->data_sock[1], | |
1575 | &handler->nsfd[LXC_NS_CGROUP], | |
1576 | 1, NULL, 0); | |
1577 | if (ret < 0) { | |
1578 | ERROR("%s - Failed to preserve cgroup namespace", strerror(errno)); | |
1579 | goto out_delete_net; | |
1580 | } | |
1581 | DEBUG("Preserved cgroup namespace via fd %d", handler->nsfd[LXC_NS_CGROUP]); | |
1582 | } | |
1583 | ||
23c53af9 | 1584 | if (handler->ops->post_start(handler, handler->data)) |
e6126dbe MN |
1585 | goto out_abort; |
1586 | ||
25c2aca5 | 1587 | if (lxc_set_state(name, handler, RUNNING)) { |
408da065 CB |
1588 | ERROR("Failed to set state for container \"%s\" to \"%s\".", name, |
1589 | lxc_state2str(RUNNING)); | |
59eb99ba | 1590 | goto out_abort; |
3f21c114 | 1591 | } |
22ebac19 | 1592 | |
3c22086f | 1593 | lxc_sync_fini(handler); |
0c547523 | 1594 | |
e6126dbe | 1595 | return 0; |
1ac470c0 | 1596 | |
7fef7a06 | 1597 | out_delete_net: |
7e4dfe0b SH |
1598 | if (cgroups_connected) |
1599 | cgroup_disconnect(); | |
1bd8d726 | 1600 | |
bb84beda CB |
1601 | if (handler->clone_flags & CLONE_NEWNET) |
1602 | lxc_delete_network(handler); | |
1bd8d726 | 1603 | |
59eb99ba DL |
1604 | out_abort: |
1605 | lxc_abort(name, handler); | |
3c22086f | 1606 | lxc_sync_fini(handler); |
5c068da9 SH |
1607 | if (handler->pinfd >= 0) { |
1608 | close(handler->pinfd); | |
1609 | handler->pinfd = -1; | |
1610 | } | |
1611 | ||
b79fcd86 | 1612 | return -1; |
59eb99ba | 1613 | } |
0ad19a3f | 1614 | |
aa460476 | 1615 | int __lxc_start(const char *name, struct lxc_handler *handler, |
507cee36 TA |
1616 | struct lxc_operations* ops, void *data, const char *lxcpath, |
1617 | bool backgrounded) | |
59eb99ba | 1618 | { |
59eb99ba | 1619 | int status; |
aa460476 | 1620 | int err = -1; |
aa460476 | 1621 | struct lxc_conf *conf = handler->conf; |
80090207 | 1622 | |
aa460476 | 1623 | if (lxc_init(name, handler) < 0) { |
408da065 | 1624 | ERROR("Failed to initialize container \"%s\".", name); |
66aeffc7 | 1625 | return -1; |
0ad19a3f | 1626 | } |
ee70bf78 CLG |
1627 | handler->ops = ops; |
1628 | handler->data = data; | |
507cee36 | 1629 | handler->backgrounded = backgrounded; |
e6126dbe | 1630 | |
76a26f55 | 1631 | if (!attach_block_device(handler->conf)) { |
408da065 | 1632 | ERROR("Failed to attach block device."); |
76a26f55 SH |
1633 | goto out_fini_nonet; |
1634 | } | |
1635 | ||
35120d9c | 1636 | if (geteuid() == 0 && !lxc_list_empty(&conf->id_map)) { |
408da065 | 1637 | /* If the backing store is a device, mount it here and now. */ |
35120d9c SH |
1638 | if (rootfs_is_blockdev(conf)) { |
1639 | if (unshare(CLONE_NEWNS) < 0) { | |
408da065 | 1640 | ERROR("Failed to unshare CLONE_NEWNS."); |
35120d9c SH |
1641 | goto out_fini_nonet; |
1642 | } | |
408da065 CB |
1643 | INFO("Unshared CLONE_NEWNS."); |
1644 | ||
6a0c909a | 1645 | remount_all_slave(); |
35120d9c | 1646 | if (do_rootfs_setup(conf, name, lxcpath) < 0) { |
408da065 | 1647 | ERROR("Error setting up rootfs mount as root before spawn."); |
35120d9c SH |
1648 | goto out_fini_nonet; |
1649 | } | |
408da065 | 1650 | INFO("Set up container rootfs as host root."); |
35120d9c SH |
1651 | } |
1652 | } | |
1653 | ||
23c53af9 | 1654 | err = lxc_spawn(handler); |
59eb99ba | 1655 | if (err) { |
408da065 | 1656 | ERROR("Failed to spawn container \"%s\".", name); |
76a26f55 | 1657 | goto out_detach_blockdev; |
0ad19a3f | 1658 | } |
a9e1109e CB |
1659 | /* close parent side of data socket */ |
1660 | close(handler->data_sock[0]); | |
1661 | handler->data_sock[0] = -1; | |
1662 | close(handler->data_sock[1]); | |
1663 | handler->data_sock[1] = -1; | |
0ad19a3f | 1664 | |
8bee8851 WB |
1665 | handler->conf->reboot = 0; |
1666 | ||
3a0f472d | 1667 | err = lxc_poll(name, handler); |
e043236e | 1668 | if (err) { |
408da065 | 1669 | ERROR("LXC mainloop exited with error: %d.", err); |
59eb99ba DL |
1670 | goto out_abort; |
1671 | } | |
0ad19a3f | 1672 | |
3a0f472d | 1673 | while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR) |
1bc5cc8c | 1674 | continue; |
e043236e | 1675 | |
408da065 CB |
1676 | /* If the child process exited but was not signaled, it didn't call |
1677 | * reboot. This should mean it was an lxc-execute which simply exited. | |
1678 | * In any case, treat it as a 'halt'. | |
8b004f07 | 1679 | */ |
d028235d | 1680 | if (WIFSIGNALED(status)) { |
8b004f07 SH |
1681 | switch(WTERMSIG(status)) { |
1682 | case SIGINT: /* halt */ | |
408da065 | 1683 | DEBUG("Container \"%s\" is halting.", name); |
8b004f07 SH |
1684 | break; |
1685 | case SIGHUP: /* reboot */ | |
408da065 | 1686 | DEBUG("Container \"%s\" is rebooting.", name); |
8b004f07 SH |
1687 | handler->conf->reboot = 1; |
1688 | break; | |
c2b9bd9e | 1689 | case SIGSYS: /* seccomp */ |
408da065 | 1690 | DEBUG("Container \"%s\" violated its seccomp policy.", name); |
c2b9bd9e | 1691 | break; |
8b004f07 | 1692 | default: |
408da065 | 1693 | DEBUG("Unknown exit status for container \"%s\" init %d.", name, WTERMSIG(status)); |
8b004f07 SH |
1694 | break; |
1695 | } | |
d028235d | 1696 | } |
828695d9 | 1697 | |
b809f232 CB |
1698 | err = lxc_restore_phys_nics_to_netns(handler); |
1699 | if (err < 0) | |
1700 | ERROR("Failed to move physical network devices back to parent " | |
1701 | "network namespace"); | |
ce5782df | 1702 | |
5c068da9 SH |
1703 | if (handler->pinfd >= 0) { |
1704 | close(handler->pinfd); | |
1705 | handler->pinfd = -1; | |
1706 | } | |
1707 | ||
1787abca | 1708 | lxc_monitor_send_exit_code(name, status, handler->lxcpath); |
3a0f472d | 1709 | err = lxc_error_set_and_log(handler->pid, status); |
1bd8d726 | 1710 | |
9d7f9e52 | 1711 | out_fini: |
bb84beda | 1712 | lxc_delete_network(handler); |
74a2b586 | 1713 | |
76a26f55 SH |
1714 | out_detach_blockdev: |
1715 | detach_block_device(handler->conf); | |
1716 | ||
74a2b586 | 1717 | out_fini_nonet: |
3a0f472d | 1718 | lxc_fini(name, handler); |
0ad19a3f | 1719 | return err; |
1720 | ||
59eb99ba | 1721 | out_abort: |
3a0f472d | 1722 | lxc_abort(name, handler); |
9d7f9e52 | 1723 | goto out_fini; |
0ad19a3f | 1724 | } |
ee70bf78 CLG |
1725 | |
1726 | struct start_args { | |
1727 | char *const *argv; | |
1728 | }; | |
1729 | ||
1730 | static int start(struct lxc_handler *handler, void* data) | |
1731 | { | |
1732 | struct start_args *arg = data; | |
1733 | ||
408da065 | 1734 | NOTICE("Exec'ing \"%s\".", arg->argv[0]); |
ee70bf78 CLG |
1735 | |
1736 | execvp(arg->argv[0], arg->argv); | |
408da065 | 1737 | SYSERROR("Failed to exec \"%s\".", arg->argv[0]); |
ee70bf78 CLG |
1738 | return 0; |
1739 | } | |
1740 | ||
1741 | static int post_start(struct lxc_handler *handler, void* data) | |
1742 | { | |
1743 | struct start_args *arg = data; | |
1744 | ||
408da065 | 1745 | NOTICE("Started \"%s\" with pid \"%d\".", arg->argv[0], handler->pid); |
ee70bf78 CLG |
1746 | return 0; |
1747 | } | |
1748 | ||
1749 | static struct lxc_operations start_ops = { | |
1750 | .start = start, | |
1751 | .post_start = post_start | |
1752 | }; | |
1753 | ||
aa460476 | 1754 | int lxc_start(const char *name, char *const argv[], struct lxc_handler *handler, |
507cee36 | 1755 | const char *lxcpath, bool backgrounded) |
ee70bf78 CLG |
1756 | { |
1757 | struct start_args start_arg = { | |
1758 | .argv = argv, | |
1759 | }; | |
1760 | ||
aa460476 | 1761 | return __lxc_start(name, handler, &start_ops, &start_arg, lxcpath, backgrounded); |
ee70bf78 | 1762 | } |
28272964 CB |
1763 | |
1764 | static void lxc_destroy_container_on_signal(struct lxc_handler *handler, | |
1765 | const char *name) | |
1766 | { | |
1767 | char destroy[MAXPATHLEN]; | |
1768 | bool bret = true; | |
1769 | int ret = 0; | |
f01f7975 | 1770 | struct lxc_container *c; |
df31363a | 1771 | if (handler->conf->rootfs.path && handler->conf->rootfs.mount) { |
790255cf | 1772 | bret = do_destroy_container(handler); |
28272964 | 1773 | if (!bret) { |
408da065 | 1774 | ERROR("Error destroying rootfs for container \"%s\".", name); |
28272964 CB |
1775 | return; |
1776 | } | |
1777 | } | |
408da065 | 1778 | INFO("Destroyed rootfs for container \"%s\".", name); |
28272964 CB |
1779 | |
1780 | ret = snprintf(destroy, MAXPATHLEN, "%s/%s", handler->lxcpath, name); | |
1781 | if (ret < 0 || ret >= MAXPATHLEN) { | |
408da065 | 1782 | ERROR("Error destroying directory for container \"%s\".", name); |
28272964 CB |
1783 | return; |
1784 | } | |
1785 | ||
f01f7975 CB |
1786 | c = lxc_container_new(name, handler->lxcpath); |
1787 | if (c) { | |
1788 | if (container_disk_lock(c)) { | |
408da065 | 1789 | INFO("Could not update lxc_snapshots file."); |
f01f7975 CB |
1790 | lxc_container_put(c); |
1791 | } else { | |
1792 | mod_all_rdeps(c, false); | |
1793 | container_disk_unlock(c); | |
1794 | lxc_container_put(c); | |
1795 | } | |
1796 | } | |
1797 | ||
d0fbc7ba | 1798 | if (!handler->am_root) |
94b0a3ac CB |
1799 | ret = userns_exec_full(handler->conf, lxc_rmdir_onedev_wrapper, |
1800 | destroy, "lxc_rmdir_onedev_wrapper"); | |
28272964 CB |
1801 | else |
1802 | ret = lxc_rmdir_onedev(destroy, NULL); | |
1803 | ||
1804 | if (ret < 0) { | |
408da065 | 1805 | ERROR("Error destroying directory for container \"%s\".", name); |
28272964 CB |
1806 | return; |
1807 | } | |
408da065 | 1808 | INFO("Destroyed directory for container \"%s\".", name); |
28272964 CB |
1809 | } |
1810 | ||
1811 | static int lxc_rmdir_onedev_wrapper(void *data) | |
1812 | { | |
1813 | char *arg = (char *) data; | |
1814 | return lxc_rmdir_onedev(arg, NULL); | |
1815 | } | |
1816 | ||
790255cf | 1817 | static bool do_destroy_container(struct lxc_handler *handler) { |
94b0a3ac CB |
1818 | int ret; |
1819 | ||
d0fbc7ba | 1820 | if (!handler->am_root) { |
94b0a3ac CB |
1821 | ret = userns_exec_full(handler->conf, storage_destroy_wrapper, |
1822 | handler->conf, "storage_destroy_wrapper"); | |
1823 | if (ret < 0) | |
d028235d | 1824 | return false; |
10bc1861 | 1825 | |
d028235d SG |
1826 | return true; |
1827 | } | |
10bc1861 | 1828 | |
790255cf | 1829 | return storage_destroy(handler->conf); |
28272964 | 1830 | } |