]>
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> |
b467714b | 35 | #include <pthread.h> |
f3787121 CB |
36 | #include <signal.h> |
37 | #include <stdio.h> | |
38 | #include <stdlib.h> | |
39 | #include <string.h> | |
40 | #include <unistd.h> | |
0ad19a3f | 41 | #include <sys/file.h> |
f4d507d5 | 42 | #include <sys/mount.h> |
f3787121 | 43 | #include <sys/param.h> |
0ad19a3f | 44 | #include <sys/prctl.h> |
f3787121 CB |
45 | #include <sys/socket.h> |
46 | #include <sys/stat.h> | |
47 | #include <sys/syscall.h> | |
ddceb1f9 | 48 | #include <sys/types.h> |
b0a33c1e | 49 | #include <sys/un.h> |
f3787121 | 50 | #include <sys/wait.h> |
ff218c25 | 51 | |
e37dda71 | 52 | #if HAVE_LIBCAP |
495d2046 SG |
53 | #include <sys/capability.h> |
54 | #endif | |
55 | ||
995acccc | 56 | #if !HAVE_DECL_PR_CAPBSET_DROP |
656994bb MH |
57 | #define PR_CAPBSET_DROP 24 |
58 | #endif | |
59 | ||
995acccc | 60 | #if !HAVE_DECL_PR_SET_NO_NEW_PRIVS |
955e2a02 CB |
61 | #define PR_SET_NO_NEW_PRIVS 38 |
62 | #endif | |
63 | ||
995acccc | 64 | #if !HAVE_DECL_PR_GET_NO_NEW_PRIVS |
955e2a02 CB |
65 | #define PR_GET_NO_NEW_PRIVS 39 |
66 | #endif | |
67 | ||
f3787121 CB |
68 | #include "af_unix.h" |
69 | #include "caps.h" | |
563f2f2c | 70 | #include "cgroup.h" |
f3787121 | 71 | #include "commands.h" |
bbf5cf35 | 72 | #include "commands_utils.h" |
f3787121 | 73 | #include "conf.h" |
74c6e2b0 | 74 | #include "confile_utils.h" |
e2bcd7db | 75 | #include "error.h" |
1cc8bd4b | 76 | #include "list.h" |
0ed9b1bc | 77 | #include "lsm/lsm.h" |
f3787121 | 78 | #include "log.h" |
f2d5a09d | 79 | #include "lxccontainer.h" |
f3787121 CB |
80 | #include "lxclock.h" |
81 | #include "lxcseccomp.h" | |
f3787121 | 82 | #include "mainloop.h" |
63376d7d | 83 | #include "monitor.h" |
f549edcc | 84 | #include "namespace.h" |
811ef482 | 85 | #include "network.h" |
f3787121 | 86 | #include "start.h" |
6be5397b CB |
87 | #include "storage/storage.h" |
88 | #include "storage/storage_utils.h" | |
0ed9b1bc CB |
89 | #include "sync.h" |
90 | #include "terminal.h" | |
91 | #include "utils.h" | |
36eb9bde | 92 | |
9de31d5a CB |
93 | #ifndef HAVE_STRLCPY |
94 | #include "include/strlcpy.h" | |
95 | #endif | |
96 | ||
ac2cecc4 | 97 | lxc_log_define(start, lxc); |
36eb9bde | 98 | |
f01f7975 | 99 | extern void mod_all_rdeps(struct lxc_container *c, bool inc); |
790255cf | 100 | static bool do_destroy_container(struct lxc_handler *handler); |
28272964 CB |
101 | static int lxc_rmdir_onedev_wrapper(void *data); |
102 | static void lxc_destroy_container_on_signal(struct lxc_handler *handler, | |
103 | const char *name); | |
104 | ||
c8154066 SH |
105 | static void print_top_failing_dir(const char *path) |
106 | { | |
9e5f5f2f CB |
107 | int ret; |
108 | size_t len; | |
109 | char *copy, *e, *p, saved; | |
c8154066 | 110 | |
9e5f5f2f CB |
111 | len = strlen(path); |
112 | copy = alloca(len + 1); | |
cbb9c7c7 DJ |
113 | (void)strlcpy(copy, path, len + 1); |
114 | ||
c8154066 SH |
115 | p = copy; |
116 | e = copy + len; | |
cbb9c7c7 | 117 | |
c8154066 | 118 | while (p < e) { |
f3787121 CB |
119 | while (p < e && *p == '/') |
120 | p++; | |
9e5f5f2f | 121 | |
f3787121 CB |
122 | while (p < e && *p != '/') |
123 | p++; | |
9e5f5f2f | 124 | |
c8154066 SH |
125 | saved = *p; |
126 | *p = '\0'; | |
9e5f5f2f CB |
127 | |
128 | ret = access(copy, X_OK); | |
129 | if (ret != 0) { | |
408da065 CB |
130 | SYSERROR("Could not access %s. Please grant it x " |
131 | "access, or add an ACL for the container " | |
9e5f5f2f | 132 | "root", copy); |
c8154066 SH |
133 | return; |
134 | } | |
135 | *p = saved; | |
136 | } | |
137 | } | |
138 | ||
5032bf39 | 139 | static void lxc_put_nsfds(struct lxc_handler *handler) |
408da065 | 140 | { |
9f30a190 MM |
141 | int i; |
142 | ||
9f30a190 | 143 | for (i = 0; i < LXC_NS_MAX; i++) { |
5032bf39 | 144 | if (handler->nsfd[i] < 0) |
39cd919c CB |
145 | continue; |
146 | ||
5032bf39 CB |
147 | close(handler->nsfd[i]); |
148 | handler->nsfd[i] = -EBADF; | |
9f30a190 | 149 | } |
9f30a190 MM |
150 | } |
151 | ||
4cb53844 CB |
152 | static int lxc_try_preserve_ns(const int pid, const char *ns) |
153 | { | |
154 | int fd; | |
155 | ||
156 | fd = lxc_preserve_ns(pid, ns); | |
157 | if (fd < 0) { | |
158 | if (errno != ENOENT) { | |
159 | SYSERROR("Failed to preserve %s namespace", ns); | |
160 | return -EINVAL; | |
161 | } | |
162 | ||
a24c5678 | 163 | SYSWARN("Kernel does not support preserving %s namespaces", ns); |
4cb53844 CB |
164 | return -EOPNOTSUPP; |
165 | } | |
166 | ||
167 | return fd; | |
168 | } | |
169 | ||
170 | /* lxc_try_preserve_namespaces: open /proc/@pid/ns/@ns for each namespace | |
171 | * specified in ns_clone_flags. | |
4d8ac866 | 172 | * Return true on success, false on failure. |
62d05d9b | 173 | */ |
4cb53844 CB |
174 | static bool lxc_try_preserve_namespaces(struct lxc_handler *handler, |
175 | int ns_clone_flags, pid_t pid) | |
f3787121 | 176 | { |
9fef3355 | 177 | int i; |
9f30a190 | 178 | |
9f30a190 | 179 | for (i = 0; i < LXC_NS_MAX; i++) |
5032bf39 | 180 | handler->nsfd[i] = -EBADF; |
cd43d2d1 | 181 | |
9f30a190 | 182 | for (i = 0; i < LXC_NS_MAX; i++) { |
4cb53844 CB |
183 | int fd; |
184 | ||
becad0ec | 185 | if ((ns_clone_flags & ns_info[i].clone_flag) == 0) |
9f30a190 | 186 | continue; |
28d9e29e | 187 | |
4cb53844 CB |
188 | fd = lxc_try_preserve_ns(pid, ns_info[i].proc_name); |
189 | if (fd < 0) { | |
190 | handler->nsfd[i] = -EBADF; | |
191 | ||
192 | /* Do not fail to start container on kernels that do | |
193 | * not support interacting with namespaces through | |
194 | * /proc. | |
195 | */ | |
196 | if (fd == -EOPNOTSUPP) | |
197 | continue; | |
fa3a5b22 | 198 | |
4cb53844 CB |
199 | lxc_put_nsfds(handler); |
200 | return false; | |
201 | } | |
202 | ||
203 | handler->nsfd[i] = fd; | |
204 | DEBUG("Preserved %s namespace via fd %d", ns_info[i].proc_name, | |
205 | handler->nsfd[i]); | |
9f30a190 MM |
206 | } |
207 | ||
62d05d9b | 208 | return true; |
9f30a190 MM |
209 | } |
210 | ||
ec1dc633 | 211 | static inline bool match_stdfds(int fd) |
80090207 | 212 | { |
ec1dc633 | 213 | return (fd == STDIN_FILENO || fd == STDOUT_FILENO || fd == STDERR_FILENO); |
80090207 CLG |
214 | } |
215 | ||
47a46cf1 CB |
216 | int lxc_check_inherited(struct lxc_conf *conf, bool closeall, |
217 | int *fds_to_ignore, size_t len_fds) | |
80090207 | 218 | { |
80090207 | 219 | int fd, fddir; |
47a46cf1 | 220 | size_t i; |
80090207 | 221 | DIR *dir; |
a5a70219 | 222 | struct dirent *direntp; |
80090207 | 223 | |
d2cf4c37 SH |
224 | if (conf && conf->close_all_fds) |
225 | closeall = true; | |
226 | ||
b119f362 | 227 | restart: |
80090207 CLG |
228 | dir = opendir("/proc/self/fd"); |
229 | if (!dir) { | |
a24c5678 | 230 | SYSWARN("Failed to open directory"); |
80090207 CLG |
231 | return -1; |
232 | } | |
233 | ||
234 | fddir = dirfd(dir); | |
235 | ||
74f96976 | 236 | while ((direntp = readdir(dir))) { |
a5a70219 | 237 | int ret; |
d39b10eb CB |
238 | struct lxc_list *cur; |
239 | bool matched = false; | |
240 | ||
a5a70219 | 241 | if (strcmp(direntp->d_name, ".") == 0) |
80090207 CLG |
242 | continue; |
243 | ||
a5a70219 | 244 | if (strcmp(direntp->d_name, "..") == 0) |
80090207 CLG |
245 | continue; |
246 | ||
a5a70219 CB |
247 | ret = lxc_safe_int(direntp->d_name, &fd); |
248 | if (ret < 0) { | |
249 | INFO("Could not parse file descriptor for \"%s\"", direntp->d_name); | |
d4cff0d2 CB |
250 | continue; |
251 | } | |
80090207 | 252 | |
47a46cf1 CB |
253 | for (i = 0; i < len_fds; i++) |
254 | if (fds_to_ignore[i] == fd) | |
255 | break; | |
256 | ||
257 | if (fd == fddir || fd == lxc_log_fd || | |
258 | (i < len_fds && fd == fds_to_ignore[i])) | |
80090207 CLG |
259 | continue; |
260 | ||
d39b10eb | 261 | /* Keep state clients that wait on reboots. */ |
3ee9e4fb CB |
262 | if (conf) { |
263 | lxc_list_for_each(cur, &conf->state_clients) { | |
264 | struct lxc_state_client *client = cur->elem; | |
d39b10eb | 265 | |
3ee9e4fb CB |
266 | if (client->clientfd != fd) |
267 | continue; | |
d39b10eb | 268 | |
3ee9e4fb CB |
269 | matched = true; |
270 | break; | |
271 | } | |
d39b10eb CB |
272 | } |
273 | ||
274 | if (matched) | |
275 | continue; | |
276 | ||
858377e4 SH |
277 | if (current_config && fd == current_config->logfd) |
278 | continue; | |
279 | ||
ec1dc633 | 280 | if (match_stdfds(fd)) |
80090207 | 281 | continue; |
80090207 | 282 | |
d2cf4c37 | 283 | if (closeall) { |
b119f362 SH |
284 | close(fd); |
285 | closedir(dir); | |
28d9e29e | 286 | INFO("Closed inherited fd %d", fd); |
b119f362 SH |
287 | goto restart; |
288 | } | |
28d9e29e | 289 | WARN("Inherited fd %d", fd); |
80090207 CLG |
290 | } |
291 | ||
408da065 | 292 | /* Only enable syslog at this point to avoid the above logging function |
64c57ea1 BD |
293 | * to open a new fd and make the check_inherited function enter an |
294 | * infinite loop. | |
295 | */ | |
296 | lxc_log_enable_syslog(); | |
297 | ||
92c7f629 GK |
298 | closedir(dir); /* cannot fail */ |
299 | return 0; | |
80090207 CLG |
300 | } |
301 | ||
83ee7875 | 302 | static int setup_signal_fd(sigset_t *oldmask) |
b0a33c1e | 303 | { |
7288dfb6 CB |
304 | int ret; |
305 | int sig; | |
b0a33c1e | 306 | sigset_t mask; |
7288dfb6 | 307 | const int signals[] = {SIGBUS, SIGILL, SIGSEGV, SIGWINCH}; |
b0a33c1e | 308 | |
408da065 | 309 | /* Block everything except serious error signals. */ |
df0795b1 CB |
310 | ret = sigfillset(&mask); |
311 | if (ret < 0) | |
312 | return -EBADF; | |
313 | ||
314 | for (sig = 0; sig < (sizeof(signals) / sizeof(signals[0])); sig++) { | |
315 | ret = sigdelset(&mask, signals[sig]); | |
316 | if (ret < 0) | |
317 | return -EBADF; | |
b0a33c1e | 318 | } |
319 | ||
b467714b | 320 | ret = pthread_sigmask(SIG_BLOCK, &mask, oldmask); |
df0795b1 CB |
321 | if (ret < 0) { |
322 | SYSERROR("Failed to set signal mask"); | |
323 | return -EBADF; | |
b0a33c1e | 324 | } |
325 | ||
df0795b1 CB |
326 | ret = signalfd(-1, &mask, SFD_CLOEXEC); |
327 | if (ret < 0) { | |
328 | SYSERROR("Failed to create signal file descriptor"); | |
329 | return -EBADF; | |
b0a33c1e | 330 | } |
331 | ||
df0795b1 | 332 | TRACE("Created signal file descriptor %d", ret); |
1ac470c0 | 333 | |
df0795b1 | 334 | return ret; |
b0a33c1e | 335 | } |
336 | ||
84c92abd | 337 | static int signal_handler(int fd, uint32_t events, void *data, |
f3787121 | 338 | struct lxc_epoll_descr *descr) |
b0a33c1e | 339 | { |
15cd25fd | 340 | int ret; |
3c319edb CB |
341 | siginfo_t info; |
342 | struct signalfd_siginfo siginfo; | |
343 | struct lxc_handler *hdlr = data; | |
15cd25fd | 344 | |
489f39be | 345 | ret = lxc_read_nointr(fd, &siginfo, sizeof(siginfo)); |
15cd25fd | 346 | if (ret < 0) { |
6e94162a | 347 | ERROR("Failed to read signal info from signal file descriptor %d", fd); |
b2a48508 | 348 | return LXC_MAINLOOP_ERROR; |
15cd25fd DL |
349 | } |
350 | ||
351 | if (ret != sizeof(siginfo)) { | |
6e94162a CB |
352 | ERROR("Unexpected size for struct signalfd_siginfo"); |
353 | return -EINVAL; | |
15cd25fd DL |
354 | } |
355 | ||
408da065 | 356 | /* Check whether init is running. */ |
80507ee8 | 357 | info.si_pid = 0; |
3c319edb CB |
358 | ret = waitid(P_PID, hdlr->pid, &info, WEXITED | WNOWAIT | WNOHANG); |
359 | if (ret == 0 && info.si_pid == hdlr->pid) | |
360 | hdlr->init_died = true; | |
80507ee8 | 361 | |
cd5177e9 TA |
362 | /* Try to figure out a reasonable exit status to report. */ |
363 | if (hdlr->init_died) { | |
364 | switch (info.si_code) { | |
365 | case CLD_EXITED: | |
366 | hdlr->exit_status = info.si_status << 8; | |
367 | break; | |
368 | case CLD_KILLED: | |
369 | case CLD_DUMPED: | |
370 | case CLD_STOPPED: | |
371 | hdlr->exit_status = info.si_status << 8 | 0x7f; | |
372 | break; | |
373 | case CLD_CONTINUED: | |
374 | /* Huh? The waitid() told us it's dead *and* continued? */ | |
375 | WARN("Init %d dead and continued?", hdlr->pid); | |
376 | hdlr->exit_status = 1; | |
377 | break; | |
378 | default: | |
20993a97 | 379 | ERROR("Unknown si_code: %d", info.si_code); |
55e1311e | 380 | hdlr->exit_status = 1; |
cd5177e9 TA |
381 | } |
382 | } | |
383 | ||
d4b5d7a8 TA |
384 | if (siginfo.ssi_signo == SIGHUP) { |
385 | kill(hdlr->pid, SIGTERM); | |
386 | INFO("Killing %d since terminal hung up", hdlr->pid); | |
b2a48508 | 387 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE : LXC_MAINLOOP_CONTINUE; |
d4b5d7a8 TA |
388 | } |
389 | ||
9cb94384 TA |
390 | if (siginfo.ssi_signo != SIGCHLD) { |
391 | kill(hdlr->pid, siginfo.ssi_signo); | |
392 | INFO("Forwarded signal %d to pid %d", siginfo.ssi_signo, hdlr->pid); | |
b2a48508 | 393 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE : LXC_MAINLOOP_CONTINUE; |
9cb94384 TA |
394 | } |
395 | ||
3a9e949f TA |
396 | /* More robustness, protect ourself from a SIGCHLD sent |
397 | * by a process different from the container init. | |
398 | */ | |
399 | if (siginfo.ssi_pid != hdlr->pid) { | |
6e94162a CB |
400 | NOTICE("Received %d from pid %d instead of container init %d", |
401 | siginfo.ssi_signo, siginfo.ssi_pid, hdlr->pid); | |
b2a48508 | 402 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE : LXC_MAINLOOP_CONTINUE; |
3a9e949f TA |
403 | } |
404 | ||
408da065 | 405 | if (siginfo.ssi_code == CLD_STOPPED) { |
6e94162a | 406 | INFO("Container init process was stopped"); |
b2a48508 | 407 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE : LXC_MAINLOOP_CONTINUE; |
408da065 | 408 | } else if (siginfo.ssi_code == CLD_CONTINUED) { |
6e94162a | 409 | INFO("Container init process was continued"); |
b2a48508 | 410 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE : LXC_MAINLOOP_CONTINUE; |
15cd25fd | 411 | } |
1ac470c0 | 412 | |
6e94162a | 413 | DEBUG("Container init process %d exited", hdlr->pid); |
3c319edb | 414 | return LXC_MAINLOOP_CLOSE; |
b0a33c1e | 415 | } |
416 | ||
974a8aba CB |
417 | int lxc_serve_state_clients(const char *name, struct lxc_handler *handler, |
418 | lxc_state_t state) | |
66aeffc7 | 419 | { |
9de31d5a | 420 | size_t retlen; |
dbc9832d CB |
421 | ssize_t ret; |
422 | struct lxc_list *cur, *next; | |
d39b10eb | 423 | struct lxc_state_client *client; |
dbc9832d CB |
424 | struct lxc_msg msg = {.type = lxc_msg_state, .value = state}; |
425 | ||
974a8aba CB |
426 | if (state == THAWED) |
427 | handler->state = RUNNING; | |
428 | else | |
429 | handler->state = state; | |
430 | ||
9dfa4041 | 431 | TRACE("Set container state to %s", lxc_state2str(state)); |
dbc9832d | 432 | |
d39b10eb | 433 | if (lxc_list_empty(&handler->conf->state_clients)) { |
9dfa4041 | 434 | TRACE("No state clients registered"); |
dbc9832d CB |
435 | return 0; |
436 | } | |
437 | ||
9de31d5a CB |
438 | retlen = strlcpy(msg.name, name, sizeof(msg.name)); |
439 | if (retlen >= sizeof(msg.name)) | |
440 | return -E2BIG; | |
dbc9832d | 441 | |
d39b10eb | 442 | lxc_list_for_each_safe(cur, &handler->conf->state_clients, next) { |
dbc9832d CB |
443 | client = cur->elem; |
444 | ||
d39b10eb | 445 | if (client->states[state] == 0) { |
9dfa4041 | 446 | TRACE("State %s not registered for state client %d", |
dbc9832d CB |
447 | lxc_state2str(state), client->clientfd); |
448 | continue; | |
449 | } | |
450 | ||
9dfa4041 | 451 | TRACE("Sending state %s to state client %d", |
dbc9832d CB |
452 | lxc_state2str(state), client->clientfd); |
453 | ||
454 | again: | |
57b1ab60 | 455 | ret = send(client->clientfd, &msg, sizeof(msg), MSG_NOSIGNAL); |
ee8377bd CB |
456 | if (ret <= 0) { |
457 | if (errno == EINTR) { | |
458 | TRACE("Caught EINTR; retrying"); | |
dbc9832d | 459 | goto again; |
ee8377bd | 460 | } |
dbc9832d | 461 | |
6d1400b5 | 462 | SYSERROR("Failed to send message to client"); |
dbc9832d CB |
463 | } |
464 | ||
465 | /* kick client from list */ | |
dbc9832d | 466 | lxc_list_del(cur); |
300d1cb4 | 467 | close(client->clientfd); |
dbc9832d CB |
468 | free(cur->elem); |
469 | free(cur); | |
470 | } | |
dbc9832d | 471 | |
5e5576a4 CB |
472 | return 0; |
473 | } | |
474 | ||
475 | static int lxc_serve_state_socket_pair(const char *name, | |
476 | struct lxc_handler *handler, | |
477 | lxc_state_t state) | |
478 | { | |
479 | ssize_t ret; | |
480 | ||
481 | if (!handler->backgrounded || | |
482 | handler->state_socket_pair[1] < 0 || | |
483 | state == STARTING) | |
484 | return 0; | |
485 | ||
486 | /* Close read end of the socket pair. */ | |
487 | close(handler->state_socket_pair[0]); | |
488 | handler->state_socket_pair[0] = -1; | |
489 | ||
ee8377bd | 490 | again: |
5e5576a4 CB |
491 | ret = lxc_abstract_unix_send_credential(handler->state_socket_pair[1], |
492 | &(int){state}, sizeof(int)); | |
9044b79e | 493 | if (ret < 0) { |
494 | SYSERROR("Failed to send state to %d", handler->state_socket_pair[1]); | |
495 | ||
ee8377bd CB |
496 | if (errno == EINTR) |
497 | goto again; | |
9044b79e | 498 | |
499 | return -1; | |
500 | } | |
501 | ||
502 | if (ret != sizeof(int)) { | |
503 | ERROR("Message too long : %d", handler->state_socket_pair[1]); | |
ee8377bd CB |
504 | return -1; |
505 | } | |
5e5576a4 CB |
506 | |
507 | TRACE("Sent container state \"%s\" to %d", lxc_state2str(state), | |
508 | handler->state_socket_pair[1]); | |
509 | ||
510 | /* Close write end of the socket pair. */ | |
511 | close(handler->state_socket_pair[1]); | |
512 | handler->state_socket_pair[1] = -1; | |
513 | ||
514 | return 0; | |
515 | } | |
516 | ||
517 | int lxc_set_state(const char *name, struct lxc_handler *handler, | |
518 | lxc_state_t state) | |
519 | { | |
520 | int ret; | |
521 | ||
522 | ret = lxc_serve_state_socket_pair(name, handler, state); | |
523 | if (ret < 0) { | |
524 | ERROR("Failed to synchronize via anonymous pair of unix sockets"); | |
525 | return -1; | |
526 | } | |
527 | ||
528 | ret = lxc_serve_state_clients(name, handler, state); | |
529 | if (ret < 0) | |
530 | return -1; | |
531 | ||
dbc9832d CB |
532 | /* This function will try to connect to the legacy lxc-monitord state |
533 | * server and only exists for backwards compatibility. | |
534 | */ | |
9123e471 | 535 | lxc_monitor_send_state(name, state, handler->lxcpath); |
dbc9832d | 536 | |
66aeffc7 DL |
537 | return 0; |
538 | } | |
539 | ||
735f2c6e | 540 | int lxc_poll(const char *name, struct lxc_handler *handler) |
b0a33c1e | 541 | { |
3c319edb | 542 | int ret; |
86530b0a | 543 | bool has_console = true; |
3c319edb | 544 | struct lxc_epoll_descr descr, descr_console; |
b0a33c1e | 545 | |
03760215 CB |
546 | if (handler->conf->console.path && |
547 | strcmp(handler->conf->console.path, "none") == 0) | |
86530b0a L |
548 | has_console = false; |
549 | ||
3c319edb CB |
550 | ret = lxc_mainloop_open(&descr); |
551 | if (ret < 0) { | |
552 | ERROR("Failed to create mainloop"); | |
50c8bf05 | 553 | goto out_sigfd; |
b0a33c1e | 554 | } |
555 | ||
30a33fbd CB |
556 | if (has_console) { |
557 | ret = lxc_mainloop_open(&descr_console); | |
558 | if (ret < 0) { | |
559 | ERROR("Failed to create console mainloop"); | |
560 | goto out_mainloop; | |
561 | } | |
3c319edb CB |
562 | } |
563 | ||
1cc8bd4b | 564 | ret = lxc_mainloop_add_handler(&descr, handler->sigfd, signal_handler, handler); |
3c319edb | 565 | if (ret < 0) { |
1cc8bd4b | 566 | ERROR("Failed to add signal handler for %d to mainloop", handler->sigfd); |
3c319edb | 567 | goto out_mainloop_console; |
b0a33c1e | 568 | } |
569 | ||
30a33fbd | 570 | if (has_console) { |
dcad02f8 | 571 | struct lxc_terminal *console = &handler->conf->console; |
63376d7d | 572 | |
093bce5f | 573 | ret = lxc_terminal_mainloop_add(&descr, console); |
30a33fbd CB |
574 | if (ret < 0) { |
575 | ERROR("Failed to add console handlers to mainloop"); | |
576 | goto out_mainloop_console; | |
577 | } | |
578 | ||
093bce5f | 579 | ret = lxc_terminal_mainloop_add(&descr_console, console); |
30a33fbd CB |
580 | if (ret < 0) { |
581 | ERROR("Failed to add console handlers to console mainloop"); | |
582 | goto out_mainloop_console; | |
583 | } | |
a54585ad L |
584 | |
585 | handler->conf->console.descr = &descr; | |
563f2f2c DL |
586 | } |
587 | ||
3c319edb CB |
588 | ret = lxc_cmd_mainloop_add(name, &descr, handler); |
589 | if (ret < 0) { | |
590 | ERROR("Failed to add command handler to mainloop"); | |
591 | goto out_mainloop_console; | |
592 | } | |
b0a33c1e | 593 | |
3c319edb | 594 | TRACE("Mainloop is ready"); |
b0a33c1e | 595 | |
3c319edb | 596 | ret = lxc_mainloop(&descr, -1); |
1cc8bd4b CB |
597 | close(descr.epfd); |
598 | descr.epfd = -EBADF; | |
599 | if (ret < 0 || !handler->init_died) | |
42b09f94 | 600 | goto out_mainloop_console; |
1cc8bd4b | 601 | |
30a33fbd CB |
602 | if (has_console) |
603 | ret = lxc_mainloop(&descr_console, 0); | |
604 | ||
3c319edb | 605 | out_mainloop_console: |
30a33fbd CB |
606 | if (has_console) { |
607 | lxc_mainloop_close(&descr_console); | |
608 | TRACE("Closed console mainloop"); | |
609 | } | |
3c319edb | 610 | |
42b09f94 CB |
611 | out_mainloop: |
612 | lxc_mainloop_close(&descr); | |
613 | TRACE("Closed mainloop"); | |
614 | ||
b0a33c1e | 615 | out_sigfd: |
1cc8bd4b CB |
616 | close(handler->sigfd); |
617 | TRACE("Closed signal file descriptor %d", handler->sigfd); | |
618 | handler->sigfd = -EBADF; | |
408da065 | 619 | |
1cc8bd4b | 620 | return ret; |
b0a33c1e | 621 | } |
622 | ||
41784e4e CB |
623 | void lxc_zero_handler(struct lxc_handler *handler) |
624 | { | |
625 | int i; | |
626 | ||
627 | memset(handler, 0, sizeof(struct lxc_handler)); | |
628 | ||
41784e4e CB |
629 | handler->pinfd = -1; |
630 | ||
631 | handler->sigfd = -1; | |
632 | ||
633 | for (i = 0; i < LXC_NS_MAX; i++) | |
634 | handler->nsfd[i] = -1; | |
635 | ||
636 | handler->data_sock[0] = -1; | |
637 | handler->data_sock[1] = -1; | |
638 | ||
639 | handler->state_socket_pair[0] = -1; | |
640 | handler->state_socket_pair[1] = -1; | |
641 | ||
642 | handler->sync_sock[0] = -1; | |
643 | handler->sync_sock[1] = -1; | |
644 | } | |
645 | ||
f2e07cb6 CB |
646 | void lxc_free_handler(struct lxc_handler *handler) |
647 | { | |
1cc8bd4b CB |
648 | if (handler->pinfd >= 0) |
649 | close(handler->pinfd); | |
650 | ||
651 | if (handler->sigfd >= 0) | |
652 | close(handler->sigfd); | |
653 | ||
654 | lxc_put_nsfds(handler); | |
655 | ||
80308d07 | 656 | if (handler->conf && handler->conf->reboot == REBOOT_NONE) |
0d0d3655 | 657 | if (handler->conf->maincmd_fd >= 0) |
9044b79e | 658 | lxc_abstract_unix_close(handler->conf->maincmd_fd); |
f2e07cb6 | 659 | |
5e5576a4 CB |
660 | if (handler->state_socket_pair[0] >= 0) |
661 | close(handler->state_socket_pair[0]); | |
662 | ||
663 | if (handler->state_socket_pair[1] >= 0) | |
664 | close(handler->state_socket_pair[1]); | |
665 | ||
f2e07cb6 CB |
666 | handler->conf = NULL; |
667 | free(handler); | |
1cc8bd4b | 668 | handler = NULL; |
f2e07cb6 CB |
669 | } |
670 | ||
aa460476 | 671 | struct lxc_handler *lxc_init_handler(const char *name, struct lxc_conf *conf, |
5e5576a4 | 672 | const char *lxcpath, bool daemonize) |
59eb99ba | 673 | { |
5e5576a4 | 674 | int i, ret; |
3a0f472d DL |
675 | struct lxc_handler *handler; |
676 | ||
677 | handler = malloc(sizeof(*handler)); | |
fdecbc9c | 678 | if (!handler) |
3a0f472d | 679 | return NULL; |
9044b79e | 680 | |
59eb99ba DL |
681 | memset(handler, 0, sizeof(*handler)); |
682 | ||
fdecbc9c CB |
683 | /* Note that am_guest_unpriv() checks the effective uid. We |
684 | * probably don't care if we are real root only if we are running | |
685 | * as root so this should be fine. | |
790255cf | 686 | */ |
e0010464 | 687 | handler->am_root = !am_guest_unpriv(); |
c6012571 | 688 | handler->data_sock[0] = handler->data_sock[1] = -1; |
fae349da | 689 | handler->conf = conf; |
9123e471 | 690 | handler->lxcpath = lxcpath; |
5c068da9 | 691 | handler->pinfd = -1; |
1cc8bd4b CB |
692 | handler->sigfd = -EBADF; |
693 | handler->init_died = false; | |
5e5576a4 | 694 | handler->state_socket_pair[0] = handler->state_socket_pair[1] = -1; |
80308d07 | 695 | if (handler->conf->reboot == REBOOT_NONE) |
d39b10eb | 696 | lxc_list_init(&handler->conf->state_clients); |
fae349da | 697 | |
b6b2b194 WB |
698 | for (i = 0; i < LXC_NS_MAX; i++) |
699 | handler->nsfd[i] = -1; | |
700 | ||
f0ecc19d | 701 | handler->name = name; |
3bdf52d7 | 702 | |
80308d07 | 703 | if (daemonize && handler->conf->reboot == REBOOT_NONE) { |
5e5576a4 | 704 | /* Create socketpair() to synchronize on daemonized startup. |
fdecbc9c CB |
705 | * When the container reboots we don't need to synchronize |
706 | * again currently so don't open another socketpair(). | |
5e5576a4 CB |
707 | */ |
708 | ret = socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0, | |
709 | handler->state_socket_pair); | |
710 | if (ret < 0) { | |
711 | ERROR("Failed to create anonymous pair of unix sockets"); | |
712 | goto on_error; | |
713 | } | |
9044b79e | 714 | |
5e5576a4 CB |
715 | TRACE("Created anonymous pair {%d,%d} of unix sockets", |
716 | handler->state_socket_pair[0], | |
717 | handler->state_socket_pair[1]); | |
718 | } | |
719 | ||
80308d07 | 720 | if (handler->conf->reboot == REBOOT_NONE) { |
d39b10eb CB |
721 | handler->conf->maincmd_fd = lxc_cmd_init(name, lxcpath, "command"); |
722 | if (handler->conf->maincmd_fd < 0) { | |
723 | ERROR("Failed to set up command socket"); | |
724 | goto on_error; | |
725 | } | |
aa460476 | 726 | } |
9044b79e | 727 | |
9dfa4041 | 728 | TRACE("Unix domain socket %d for command server is ready", |
aa460476 CB |
729 | handler->conf->maincmd_fd); |
730 | ||
731 | return handler; | |
732 | ||
f2e07cb6 CB |
733 | on_error: |
734 | lxc_free_handler(handler); | |
aa460476 CB |
735 | |
736 | return NULL; | |
737 | } | |
738 | ||
739 | int lxc_init(const char *name, struct lxc_handler *handler) | |
740 | { | |
a2c09be0 | 741 | int ret; |
4a03ded4 | 742 | const char *loglevel; |
aa460476 CB |
743 | struct lxc_conf *conf = handler->conf; |
744 | ||
745 | lsm_init(); | |
2170b263 | 746 | TRACE("Initialized LSM"); |
d2e30e99 | 747 | |
2170b263 CB |
748 | ret = lxc_read_seccomp_config(conf); |
749 | if (ret < 0) { | |
750 | ERROR("Failed loading seccomp policy"); | |
d2e30e99 | 751 | goto out_close_maincmd_fd; |
8f2c3a70 | 752 | } |
2170b263 | 753 | TRACE("Read seccomp policy"); |
8f2c3a70 | 754 | |
408da065 | 755 | /* Begin by setting the state to STARTING. */ |
2170b263 CB |
756 | ret = lxc_set_state(name, handler, STARTING); |
757 | if (ret < 0) { | |
758 | ERROR("Failed to set state to \"%s\"", lxc_state2str(STARTING)); | |
051151de | 759 | goto out_close_maincmd_fd; |
0ad19a3f | 760 | } |
46800e77 | 761 | TRACE("Set container state to \"STARTING\""); |
0ad19a3f | 762 | |
408da065 | 763 | /* Start of environment variable setup for hooks. */ |
24c2a48d CB |
764 | ret = setenv("LXC_NAME", name, 1); |
765 | if (ret < 0) | |
766 | SYSERROR("Failed to set environment variable: LXC_NAME=%s", name); | |
408da065 | 767 | |
2170b263 CB |
768 | if (conf->rcfile) { |
769 | ret = setenv("LXC_CONFIG_FILE", conf->rcfile, 1); | |
770 | if (ret < 0) | |
771 | SYSERROR("Failed to set environment variable: " | |
772 | "LXC_CONFIG_FILE=%s", conf->rcfile); | |
773 | } | |
408da065 | 774 | |
2170b263 CB |
775 | if (conf->rootfs.mount) { |
776 | ret = setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1); | |
777 | if (ret < 0) | |
778 | SYSERROR("Failed to set environment variable: " | |
779 | "LXC_ROOTFS_MOUNT=%s", conf->rootfs.mount); | |
780 | } | |
408da065 | 781 | |
2170b263 CB |
782 | if (conf->rootfs.path) { |
783 | ret = setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1); | |
784 | if (ret < 0) | |
785 | SYSERROR("Failed to set environment variable: " | |
786 | "LXC_ROOTFS_PATH=%s", conf->rootfs.path); | |
787 | } | |
408da065 | 788 | |
2170b263 CB |
789 | if (conf->console.path) { |
790 | ret = setenv("LXC_CONSOLE", conf->console.path, 1); | |
791 | if (ret < 0) | |
792 | SYSERROR("Failed to set environment variable: " | |
793 | "LXC_CONSOLE=%s", conf->console.path); | |
794 | } | |
408da065 | 795 | |
2170b263 CB |
796 | if (conf->console.log_path) { |
797 | ret = setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1); | |
798 | if (ret < 0) | |
799 | SYSERROR("Failed to set environment variable: " | |
800 | "LXC_CONSOLE_LOGPATH=%s", conf->console.log_path); | |
801 | } | |
408da065 | 802 | |
2170b263 CB |
803 | if (cgns_supported()) { |
804 | ret = setenv("LXC_CGNS_AWARE", "1", 1); | |
805 | if (ret < 0) | |
806 | SYSERROR("Failed to set environment variable " | |
807 | "LXC_CGNS_AWARE=1"); | |
808 | } | |
b8f88d9b | 809 | |
4a03ded4 | 810 | loglevel = lxc_log_priority_to_string(lxc_log_get_level()); |
2170b263 CB |
811 | ret = setenv("LXC_LOG_LEVEL", loglevel, 1); |
812 | if (ret < 0) | |
813 | SYSERROR("Set environment variable LXC_LOG_LEVEL=%s", | |
814 | loglevel); | |
a2c09be0 CB |
815 | |
816 | if (conf->hooks_version == 0) | |
817 | ret = setenv("LXC_HOOK_VERSION", "0", 1); | |
818 | else | |
819 | ret = setenv("LXC_HOOK_VERSION", "1", 1); | |
820 | if (ret < 0) | |
821 | SYSERROR("Failed to set environment variable LXC_HOOK_VERSION=%u", conf->hooks_version); | |
408da065 | 822 | /* End of environment variable setup for hooks. */ |
f7bee6c6 | 823 | |
2170b263 | 824 | TRACE("Set environment variables"); |
dbc9832d | 825 | |
2170b263 CB |
826 | ret = run_lxc_hooks(name, "pre-start", conf, NULL); |
827 | if (ret < 0) { | |
828 | ERROR("Failed to run lxc.hook.pre-start for container \"%s\"", name); | |
773fb9ca SH |
829 | goto out_aborting; |
830 | } | |
2170b263 | 831 | TRACE("Ran pre-start hooks"); |
26ddeedd | 832 | |
408da065 CB |
833 | /* The signal fd has to be created before forking otherwise if the child |
834 | * process exits before we setup the signal fd, the event will be lost | |
835 | * and the command will be stuck. | |
836 | */ | |
83ee7875 | 837 | handler->sigfd = setup_signal_fd(&handler->oldmask); |
59eb99ba | 838 | if (handler->sigfd < 0) { |
408da065 | 839 | ERROR("Failed to setup SIGCHLD fd handler."); |
b5159817 DE |
840 | goto out_delete_tty; |
841 | } | |
2170b263 | 842 | TRACE("Set up signal fd"); |
b5159817 | 843 | |
408da065 | 844 | /* Do this after setting up signals since it might unblock SIGWINCH. */ |
564e31c4 | 845 | ret = lxc_terminal_setup(conf); |
4d1ffb0a CB |
846 | if (ret < 0) { |
847 | ERROR("Failed to create console"); | |
b5159817 | 848 | goto out_restore_sigmask; |
b0a33c1e | 849 | } |
4d1ffb0a | 850 | TRACE("Created console"); |
b0a33c1e | 851 | |
2f835b4b | 852 | ret = lxc_terminal_map_ids(conf, &conf->console); |
7cfeddd7 | 853 | if (ret < 0) { |
2170b263 | 854 | ERROR("Failed to chown console"); |
5b74eb3c | 855 | goto out_delete_terminal; |
c4d10a05 | 856 | } |
2170b263 | 857 | TRACE("Chowned console"); |
c4d10a05 | 858 | |
2202afc9 CB |
859 | handler->cgroup_ops = cgroup_init(handler); |
860 | if (!handler->cgroup_ops) { | |
861 | ERROR("Failed to initialize cgroup driver"); | |
5b74eb3c | 862 | goto out_delete_terminal; |
2202afc9 CB |
863 | } |
864 | TRACE("Initialized cgroup driver"); | |
865 | ||
1800f924 WB |
866 | ret = lsm_process_prepare(conf, handler->lxcpath); |
867 | if (ret < 0) { | |
868 | ERROR("Failed to initialize LSM"); | |
869 | goto out_destroy_cgroups; | |
870 | } | |
871 | TRACE("Initialized LSM"); | |
872 | ||
2170b263 | 873 | INFO("Container \"%s\" is initialized", name); |
aa460476 | 874 | return 0; |
59eb99ba | 875 | |
1800f924 WB |
876 | out_destroy_cgroups: |
877 | handler->cgroup_ops->destroy(handler->cgroup_ops, handler); | |
878 | ||
5b74eb3c CB |
879 | out_delete_terminal: |
880 | lxc_terminal_delete(&handler->conf->console); | |
ea918412 | 881 | |
b5159817 | 882 | out_restore_sigmask: |
4c496daa | 883 | (void)pthread_sigmask(SIG_SETMASK, &handler->oldmask, NULL); |
ea918412 | 884 | |
59eb99ba | 885 | out_delete_tty: |
0e4be3cf | 886 | lxc_delete_tty(&conf->ttys); |
ea918412 | 887 | |
59eb99ba | 888 | out_aborting: |
4c496daa | 889 | (void)lxc_set_state(name, handler, ABORTING); |
ea918412 | 890 | |
d2e30e99 | 891 | out_close_maincmd_fd: |
9044b79e | 892 | lxc_abstract_unix_close(conf->maincmd_fd); |
d2e30e99 | 893 | conf->maincmd_fd = -1; |
aa460476 | 894 | return -1; |
59eb99ba DL |
895 | } |
896 | ||
3b72c4a0 | 897 | void lxc_fini(const char *name, struct lxc_handler *handler) |
59eb99ba | 898 | { |
c097467f | 899 | int i, ret; |
18b3b9c1 | 900 | pid_t self; |
dbc9832d | 901 | struct lxc_list *cur, *next; |
dbc9832d | 902 | char *namespaces[LXC_NS_MAX + 1]; |
b3286b62 | 903 | size_t namespace_count = 0; |
2202afc9 | 904 | struct cgroup_ops *cgroup_ops = handler->cgroup_ops; |
b6b2b194 | 905 | |
408da065 CB |
906 | /* The STOPPING state is there for future cleanup code which can take |
907 | * awhile. | |
59eb99ba | 908 | */ |
25c2aca5 | 909 | lxc_set_state(name, handler, STOPPING); |
b6b2b194 | 910 | |
0059379f | 911 | self = lxc_raw_getpid(); |
b3286b62 | 912 | for (i = 0; i < LXC_NS_MAX; i++) { |
18b3b9c1 CB |
913 | if (handler->nsfd[i] < 0) |
914 | continue; | |
915 | ||
916 | if (handler->conf->hooks_version == 0) | |
c097467f | 917 | ret = asprintf(&namespaces[namespace_count], |
18b3b9c1 CB |
918 | "%s:/proc/%d/fd/%d", ns_info[i].proc_name, |
919 | self, handler->nsfd[i]); | |
920 | else | |
c097467f | 921 | ret = asprintf(&namespaces[namespace_count], |
18b3b9c1 | 922 | "/proc/%d/fd/%d", self, handler->nsfd[i]); |
c097467f CB |
923 | if (ret == -1) { |
924 | SYSERROR("Failed to allocate memory"); | |
18b3b9c1 | 925 | break; |
b3286b62 | 926 | } |
18b3b9c1 CB |
927 | |
928 | if (handler->conf->hooks_version == 0) { | |
929 | namespace_count++; | |
930 | continue; | |
931 | } | |
932 | ||
c097467f CB |
933 | ret = setenv(ns_info[i].env_name, namespaces[namespace_count], 1); |
934 | if (ret < 0) | |
18b3b9c1 CB |
935 | SYSERROR("Failed to set environment variable %s=%s", |
936 | ns_info[i].env_name, namespaces[namespace_count]); | |
937 | else | |
938 | TRACE("Set environment variable %s=%s", | |
939 | ns_info[i].env_name, namespaces[namespace_count]); | |
940 | ||
941 | namespace_count++; | |
b3286b62 WB |
942 | } |
943 | namespaces[namespace_count] = NULL; | |
c154af98 | 944 | |
80308d07 | 945 | if (handler->conf->reboot > REBOOT_NONE) { |
c097467f CB |
946 | ret = setenv("LXC_TARGET", "reboot", 1); |
947 | if (ret < 0) | |
948 | SYSERROR("Failed to set environment variable: " | |
949 | "LXC_TARGET=reboot"); | |
950 | } | |
408da065 | 951 | |
80308d07 | 952 | if (handler->conf->reboot == REBOOT_NONE) { |
c097467f CB |
953 | ret = setenv("LXC_TARGET", "stop", 1); |
954 | if (ret < 0) | |
955 | SYSERROR("Failed to set environment variable: " | |
956 | "LXC_TARGET=stop"); | |
957 | } | |
c154af98 | 958 | |
18b3b9c1 | 959 | if (handler->conf->hooks_version == 0) |
c097467f | 960 | ret = run_lxc_hooks(name, "stop", handler->conf, namespaces); |
18b3b9c1 | 961 | else |
c097467f CB |
962 | ret = run_lxc_hooks(name, "stop", handler->conf, NULL); |
963 | if (ret < 0) | |
964 | ERROR("Failed to run \"lxc.hook.stop\" hook"); | |
c154af98 | 965 | |
b3286b62 WB |
966 | while (namespace_count--) |
967 | free(namespaces[namespace_count]); | |
28d9e29e | 968 | |
1800f924 WB |
969 | lsm_process_cleanup(handler->conf, handler->lxcpath); |
970 | ||
2202afc9 CB |
971 | cgroup_ops->destroy(cgroup_ops, handler); |
972 | cgroup_exit(cgroup_ops); | |
4288b79f | 973 | |
80308d07 | 974 | if (handler->conf->reboot == REBOOT_NONE) { |
f893d898 CB |
975 | /* For all new state clients simply close the command socket. |
976 | * This will inform all state clients that the container is | |
977 | * STOPPED and also prevents a race between a open()/close() on | |
978 | * the command socket causing a new process to get ECONNREFUSED | |
979 | * because we haven't yet closed the command socket. | |
980 | */ | |
9044b79e | 981 | lxc_abstract_unix_close(handler->conf->maincmd_fd); |
d39b10eb | 982 | handler->conf->maincmd_fd = -1; |
c3184275 CB |
983 | TRACE("Closed command socket"); |
984 | ||
985 | /* This function will try to connect to the legacy lxc-monitord | |
986 | * state server and only exists for backwards compatibility. | |
987 | */ | |
988 | lxc_monitor_send_state(name, STOPPED, handler->lxcpath); | |
989 | ||
990 | /* The command socket is closed so no one can acces the command | |
991 | * socket anymore so there's no need to lock it. | |
992 | */ | |
993 | handler->state = STOPPED; | |
994 | TRACE("Set container state to \"STOPPED\""); | |
f893d898 CB |
995 | } else { |
996 | lxc_set_state(name, handler, STOPPED); | |
d39b10eb | 997 | } |
4288b79f | 998 | |
46800e77 CB |
999 | ret = run_lxc_hooks(name, "post-stop", handler->conf, NULL); |
1000 | if (ret < 0) { | |
c097467f | 1001 | ERROR("Failed to run lxc.hook.post-stop for container \"%s\"", name); |
80308d07 | 1002 | if (handler->conf->reboot > REBOOT_NONE) { |
c097467f | 1003 | WARN("Container will be stopped instead of rebooted"); |
80308d07 | 1004 | handler->conf->reboot = REBOOT_NONE; |
c097467f CB |
1005 | |
1006 | ret = setenv("LXC_TARGET", "stop", 1); | |
1007 | if (ret < 0) | |
1008 | WARN("Failed to set environment variable: " | |
1009 | "LXC_TARGET=stop"); | |
f3787121 CB |
1010 | } |
1011 | } | |
26ddeedd | 1012 | |
408da065 | 1013 | /* Reset mask set by setup_signal_fd. */ |
b467714b | 1014 | ret = pthread_sigmask(SIG_SETMASK, &handler->oldmask, NULL); |
c097467f | 1015 | if (ret < 0) |
a24c5678 | 1016 | SYSWARN("Failed to restore signal mask"); |
8f64a3f6 | 1017 | |
2aac071b | 1018 | lxc_terminal_delete(&handler->conf->console); |
0e4be3cf | 1019 | lxc_delete_tty(&handler->conf->ttys); |
dbc9832d | 1020 | |
dbc9832d CB |
1021 | /* The command socket is now closed, no more state clients can register |
1022 | * themselves from now on. So free the list of state clients. | |
1023 | */ | |
d39b10eb CB |
1024 | lxc_list_for_each_safe(cur, &handler->conf->state_clients, next) { |
1025 | struct lxc_state_client *client = cur->elem; | |
1026 | ||
1027 | /* Keep state clients that want to be notified about reboots. */ | |
80308d07 CB |
1028 | if ((handler->conf->reboot > REBOOT_NONE) && |
1029 | (client->states[RUNNING] == 2)) | |
d39b10eb CB |
1030 | continue; |
1031 | ||
dbc9832d | 1032 | /* close state client socket */ |
dbc9832d | 1033 | lxc_list_del(cur); |
c097467f | 1034 | close(client->clientfd); |
dbc9832d CB |
1035 | free(cur->elem); |
1036 | free(cur); | |
1037 | } | |
1038 | ||
80308d07 | 1039 | if (handler->conf->ephemeral == 1 && handler->conf->reboot != REBOOT_REQ) |
28272964 | 1040 | lxc_destroy_container_on_signal(handler, name); |
2c5f2ede | 1041 | |
1cc8bd4b | 1042 | lxc_free_handler(handler); |
59eb99ba DL |
1043 | } |
1044 | ||
735f2c6e | 1045 | void lxc_abort(const char *name, struct lxc_handler *handler) |
59eb99ba | 1046 | { |
73e608b2 SH |
1047 | int ret, status; |
1048 | ||
25c2aca5 | 1049 | lxc_set_state(name, handler, ABORTING); |
0e4f9d51 CB |
1050 | |
1051 | if (handler->pid > 0) { | |
1052 | ret = kill(handler->pid, SIGKILL); | |
1053 | if (ret < 0) | |
1054 | SYSERROR("Failed to send SIGKILL to %d", handler->pid); | |
1055 | } | |
1056 | ||
408da065 CB |
1057 | while ((ret = waitpid(-1, &status, 0)) > 0) { |
1058 | ; | |
1059 | } | |
59eb99ba DL |
1060 | } |
1061 | ||
ffe1e01a | 1062 | static int do_start(void *data) |
50e98013 | 1063 | { |
8bf3abfb | 1064 | int ret; |
7a55c157 | 1065 | char path[PATH_MAX]; |
928b1f04 YT |
1066 | bool have_cap_setgid; |
1067 | uid_t new_uid; | |
1068 | gid_t new_gid; | |
e96a536c | 1069 | struct lxc_list *iterator; |
5af9369b | 1070 | int devnull_fd = -1; |
e96a536c | 1071 | struct lxc_handler *handler = data; |
50e98013 | 1072 | |
18225d19 | 1073 | lxc_sync_fini_parent(handler); |
50e98013 | 1074 | |
408da065 CB |
1075 | /* This prctl must be before the synchro, so if the parent dies before |
1076 | * we set the parent death signal, we will detect its death with the | |
1077 | * synchro right after, otherwise we have a window where the parent can | |
1078 | * exit before we set the pdeath signal leading to a unsupervized | |
1079 | * container. | |
743ecd2e | 1080 | */ |
18225d19 | 1081 | ret = lxc_set_death_signal(SIGKILL); |
912314fc CB |
1082 | if (ret < 0) { |
1083 | SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL"); | |
18225d19 | 1084 | goto out_warn_father; |
743ecd2e DL |
1085 | } |
1086 | ||
611ddd34 CB |
1087 | ret = lxc_ambient_caps_up(); |
1088 | if (ret < 0) { | |
6f5e532f | 1089 | ERROR("Failed to raise ambient capabilities"); |
611ddd34 CB |
1090 | goto out_warn_father; |
1091 | } | |
1092 | ||
b467714b | 1093 | ret = pthread_sigmask(SIG_SETMASK, &handler->oldmask, NULL); |
18225d19 CB |
1094 | if (ret < 0) { |
1095 | SYSERROR("Failed to set signal mask"); | |
1096 | goto out_warn_father; | |
1097 | } | |
50e98013 | 1098 | |
408da065 | 1099 | /* Don't leak the pinfd to the container. */ |
d39b10eb | 1100 | if (handler->pinfd >= 0) |
0d03360a | 1101 | close(handler->pinfd); |
2b0e17e4 | 1102 | |
18225d19 CB |
1103 | ret = lxc_sync_wait_parent(handler, LXC_SYNC_STARTUP); |
1104 | if (ret < 0) | |
1105 | goto out_warn_father; | |
5b1e83cb | 1106 | |
408da065 CB |
1107 | /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See |
1108 | * https://github.com/lxc/lxd/issues/1978. | |
1109 | */ | |
becad0ec | 1110 | if ((handler->ns_clone_flags & (CLONE_NEWNET | CLONE_NEWUSER)) == |
408da065 | 1111 | (CLONE_NEWNET | CLONE_NEWUSER)) { |
5b1e83cb SH |
1112 | ret = unshare(CLONE_NEWNET); |
1113 | if (ret < 0) { | |
e96a536c | 1114 | SYSERROR("Failed to unshare CLONE_NEWNET"); |
5b1e83cb SH |
1115 | goto out_warn_father; |
1116 | } | |
e96a536c | 1117 | INFO("Unshared CLONE_NEWNET"); |
5b1e83cb SH |
1118 | } |
1119 | ||
408da065 CB |
1120 | /* Tell the parent task it can begin to configure the container and wait |
1121 | * for it to finish. | |
3c22086f | 1122 | */ |
e96a536c CB |
1123 | ret = lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE); |
1124 | if (ret < 0) | |
611ddd34 | 1125 | goto out_error; |
50e98013 | 1126 | |
e96a536c CB |
1127 | ret = lxc_network_recv_veth_names_from_parent(handler); |
1128 | if (ret < 0) { | |
7ab1ba02 | 1129 | ERROR("Failed to receive veth names from parent"); |
658979c5 | 1130 | goto out_warn_father; |
7ab1ba02 | 1131 | } |
658979c5 | 1132 | |
408da065 | 1133 | /* If we are in a new user namespace, become root there to have |
d08f8d2f | 1134 | * privilege over our namespace. |
f6d3e3e4 SH |
1135 | */ |
1136 | if (!lxc_list_empty(&handler->conf->id_map)) { | |
4160c3a0 CB |
1137 | uid_t nsuid = (handler->conf->root_nsuid_map != NULL) |
1138 | ? 0 | |
1139 | : handler->conf->init_uid; | |
1140 | gid_t nsgid = (handler->conf->root_nsgid_map != NULL) | |
1141 | ? 0 | |
1142 | : handler->conf->init_gid; | |
1143 | ||
1144 | ret = lxc_switch_uid_gid(nsuid, nsgid); | |
4b826b1f | 1145 | if (ret < 0) |
f6d3e3e4 | 1146 | goto out_warn_father; |
d08f8d2f CB |
1147 | |
1148 | /* Drop groups only after we switched to a valid gid in the new | |
1149 | * user namespace. | |
1150 | */ | |
4b826b1f | 1151 | ret = lxc_setgroups(0, NULL); |
a19b974f | 1152 | if (ret < 0 && (handler->am_root || errno != EPERM)) |
c476bdce | 1153 | goto out_warn_father; |
4b826b1f | 1154 | |
d7883725 CB |
1155 | ret = prctl(PR_SET_DUMPABLE, 1, 0, 0, 0); |
1156 | if (ret < 0) | |
1157 | goto out_warn_father; | |
912314fc CB |
1158 | |
1159 | /* set{g,u}id() clears deathsignal */ | |
18225d19 | 1160 | ret = lxc_set_death_signal(SIGKILL); |
912314fc CB |
1161 | if (ret < 0) { |
1162 | SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL"); | |
1163 | goto out_warn_father; | |
1164 | } | |
f6d3e3e4 SH |
1165 | } |
1166 | ||
e96a536c CB |
1167 | ret = access(handler->lxcpath, X_OK); |
1168 | if (ret != 0) { | |
c8154066 SH |
1169 | print_top_failing_dir(handler->lxcpath); |
1170 | goto out_warn_father; | |
1171 | } | |
1172 | ||
e96a536c CB |
1173 | ret = snprintf(path, sizeof(path), "%s/dev/null", |
1174 | handler->conf->rootfs.mount); | |
408da065 | 1175 | if (ret < 0 || ret >= sizeof(path)) |
7a55c157 | 1176 | goto out_warn_father; |
7a55c157 TA |
1177 | |
1178 | /* In order to checkpoint restore, we need to have everything in the | |
1179 | * same mount namespace. However, some containers may not have a | |
1180 | * reasonable /dev (in particular, they may not have /dev/null), so we | |
1181 | * can't set init's std fds to /dev/null by opening it from inside the | |
1182 | * container. | |
1183 | * | |
1184 | * If that's the case, fall back to using the host's /dev/null. This | |
1185 | * means that migration won't work, but at least we won't spew output | |
1186 | * where it isn't wanted. | |
1187 | */ | |
e96a536c CB |
1188 | if (handler->backgrounded && !handler->conf->autodev) { |
1189 | ret = access(path, F_OK); | |
1190 | if (ret != 0) { | |
1191 | devnull_fd = open_devnull(); | |
c44de748 | 1192 | |
e96a536c CB |
1193 | if (devnull_fd < 0) |
1194 | goto out_warn_father; | |
1195 | WARN("Using /dev/null from the host for container " | |
1196 | "init's standard file descriptors. Migration will " | |
1197 | "not work"); | |
1198 | } | |
c44de748 AM |
1199 | } |
1200 | ||
408da065 | 1201 | /* Ask father to setup cgroups and wait for him to finish. */ |
e96a536c CB |
1202 | ret = lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP); |
1203 | if (ret < 0) | |
c44de748 | 1204 | goto out_error; |
544a48a0 | 1205 | |
deefdf8a CB |
1206 | /* Unshare cgroup namespace after we have setup our cgroups. If we do it |
1207 | * earlier we end up with a wrong view of /proc/self/cgroup. For | |
1208 | * example, assume we unshare(CLONE_NEWCGROUP) first, and then create | |
1209 | * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then | |
1210 | * /proc/self/cgroup would show us: | |
1211 | * | |
1212 | * 8:cpuset:/lxc/c | |
1213 | * | |
1214 | * whereas it should actually show | |
1215 | * | |
1216 | * 8:cpuset:/ | |
1217 | */ | |
becad0ec | 1218 | if (handler->ns_clone_flags & CLONE_NEWCGROUP) { |
e96a536c CB |
1219 | ret = unshare(CLONE_NEWCGROUP); |
1220 | if (ret < 0) { | |
1221 | INFO("Failed to unshare CLONE_NEWCGROUP"); | |
deefdf8a CB |
1222 | goto out_warn_father; |
1223 | } | |
e96a536c | 1224 | INFO("Unshared CLONE_NEWCGROUP"); |
deefdf8a CB |
1225 | } |
1226 | ||
149857af CB |
1227 | /* Add the requested environment variables to the current environment to |
1228 | * allow them to be used by the various hooks, such as the start hook | |
1229 | * above. | |
98ff08ca | 1230 | */ |
98ff08ca | 1231 | lxc_list_for_each(iterator, &handler->conf->environment) { |
e96a536c CB |
1232 | ret = putenv((char *)iterator->elem); |
1233 | if (ret < 0) { | |
1234 | SYSERROR("Failed to set environment variable: %s", | |
1235 | (char *)iterator->elem); | |
98ff08ca CB |
1236 | goto out_warn_father; |
1237 | } | |
1238 | } | |
1239 | ||
f4152036 | 1240 | /* Setup the container, ip, names, utsname, ... */ |
3b988b33 | 1241 | ret = lxc_setup(handler); |
7729f8e5 | 1242 | close(handler->data_sock[1]); |
8bf3abfb | 1243 | close(handler->data_sock[0]); |
7729f8e5 | 1244 | if (ret < 0) { |
e96a536c | 1245 | ERROR("Failed to setup container \"%s\"", handler->name); |
f4152036 CB |
1246 | goto out_warn_father; |
1247 | } | |
1248 | ||
408da065 | 1249 | /* Set the label to change to when we exec(2) the container's init. */ |
1800f924 | 1250 | ret = lsm_process_label_set(NULL, handler->conf, true); |
e96a536c | 1251 | if (ret < 0) |
e075f5d9 | 1252 | goto out_warn_father; |
5112cd70 | 1253 | |
029cdff5 | 1254 | /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it |
408da065 CB |
1255 | * before we aren't allowed anymore. |
1256 | */ | |
029cdff5 | 1257 | if (handler->conf->no_new_privs) { |
e96a536c CB |
1258 | ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); |
1259 | if (ret < 0) { | |
1260 | SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block " | |
1261 | "execve() gainable privileges"); | |
029cdff5 CB |
1262 | goto out_warn_father; |
1263 | } | |
e96a536c CB |
1264 | DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable " |
1265 | "privileges"); | |
029cdff5 CB |
1266 | } |
1267 | ||
0d9acb99 DE |
1268 | /* Some init's such as busybox will set sane tty settings on stdin, |
1269 | * stdout, stderr which it thinks is the console. We already set them | |
1270 | * the way we wanted on the real terminal, and we want init to do its | |
564e31c4 | 1271 | * setup on its console ie. the pty allocated in lxc_terminal_setup() so |
408da065 | 1272 | * make sure that that pty is stdin,stdout,stderr. |
0d9acb99 | 1273 | */ |
5d113f65 | 1274 | if (handler->conf->console.slave >= 0) { |
07c4ea31 | 1275 | if (handler->backgrounded || !handler->conf->is_execute) |
5d113f65 CB |
1276 | ret = set_stdfds(handler->conf->console.slave); |
1277 | else | |
ae6d3913 | 1278 | ret = lxc_terminal_set_stdfds(handler->conf->console.slave); |
5d113f65 | 1279 | if (ret < 0) { |
c5b93afb | 1280 | ERROR("Failed to redirect std{in,out,err} to pty file " |
e96a536c | 1281 | "descriptor %d", handler->conf->console.slave); |
c5b93afb LF |
1282 | goto out_warn_father; |
1283 | } | |
5d113f65 | 1284 | } |
0d9acb99 | 1285 | |
408da065 | 1286 | /* If we mounted a temporary proc, then unmount it now. */ |
5112cd70 | 1287 | tmp_proc_unmount(handler->conf); |
e075f5d9 | 1288 | |
e96a536c CB |
1289 | ret = lxc_seccomp_load(handler->conf); |
1290 | if (ret < 0) | |
8f2c3a70 SH |
1291 | goto out_warn_father; |
1292 | ||
e96a536c CB |
1293 | ret = run_lxc_hooks(handler->name, "start", handler->conf, NULL); |
1294 | if (ret < 0) { | |
1295 | ERROR("Failed to run lxc.hook.start for container \"%s\"", | |
1296 | handler->name); | |
773fb9ca SH |
1297 | goto out_warn_father; |
1298 | } | |
fc25b815 | 1299 | |
773fb9ca | 1300 | close(handler->sigfd); |
26ddeedd | 1301 | |
e96a536c | 1302 | if (handler->conf->console.slave < 0 && handler->backgrounded) { |
f4c177c3 CB |
1303 | if (devnull_fd < 0) { |
1304 | devnull_fd = open_devnull(); | |
1305 | if (devnull_fd < 0) | |
1306 | goto out_warn_father; | |
1307 | } | |
1308 | ||
e96a536c CB |
1309 | ret = set_stdfds(devnull_fd); |
1310 | if (ret < 0) { | |
1311 | ERROR("Failed to redirect std{in,out,err} to \"/dev/null\""); | |
c5b93afb LF |
1312 | goto out_warn_father; |
1313 | } | |
e96a536c | 1314 | } |
507cee36 | 1315 | |
c44de748 AM |
1316 | if (devnull_fd >= 0) { |
1317 | close(devnull_fd); | |
1318 | devnull_fd = -1; | |
1319 | } | |
1320 | ||
8c9a7665 TA |
1321 | setsid(); |
1322 | ||
e96a536c CB |
1323 | if (handler->conf->init_cwd) { |
1324 | ret = chdir(handler->conf->init_cwd); | |
1325 | if (ret < 0) { | |
1326 | SYSERROR("Could not change directory to \"%s\"", | |
1327 | handler->conf->init_cwd); | |
1328 | goto out_warn_father; | |
1329 | } | |
3c491553 L |
1330 | } |
1331 | ||
e96a536c CB |
1332 | ret = lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP_LIMITS); |
1333 | if (ret < 0) | |
f4152036 CB |
1334 | goto out_warn_father; |
1335 | ||
149857af CB |
1336 | /* Reset the environment variables the user requested in a clear |
1337 | * environment. | |
1338 | */ | |
e96a536c CB |
1339 | ret = clearenv(); |
1340 | /* Don't error out though. */ | |
1341 | if (ret < 0) | |
149857af | 1342 | SYSERROR("Failed to clear environment."); |
149857af CB |
1343 | |
1344 | lxc_list_for_each(iterator, &handler->conf->environment) { | |
e96a536c CB |
1345 | ret = putenv((char *)iterator->elem); |
1346 | if (ret < 0) { | |
1347 | SYSERROR("Failed to set environment variable: %s", | |
1348 | (char *)iterator->elem); | |
149857af CB |
1349 | goto out_warn_father; |
1350 | } | |
1351 | } | |
1352 | ||
e96a536c CB |
1353 | ret = putenv("container=lxc"); |
1354 | if (ret < 0) { | |
1355 | SYSERROR("Failed to set environment variable: container=lxc"); | |
98ff08ca CB |
1356 | goto out_warn_father; |
1357 | } | |
1358 | ||
885766f5 CB |
1359 | if (handler->conf->ttys.tty_names) { |
1360 | ret = putenv(handler->conf->ttys.tty_names); | |
e96a536c CB |
1361 | if (ret < 0) { |
1362 | SYSERROR("Failed to set environment variable for container ptys"); | |
98ff08ca CB |
1363 | goto out_warn_father; |
1364 | } | |
1365 | } | |
1366 | ||
76bdf299 CB |
1367 | /* The container has been setup. We can now switch to an unprivileged |
1368 | * uid/gid. | |
1369 | */ | |
928b1f04 YT |
1370 | new_uid = handler->conf->init_uid; |
1371 | new_gid = handler->conf->init_gid; | |
76bdf299 | 1372 | |
e96a536c CB |
1373 | /* If we are in a new user namespace we already dropped all groups when |
1374 | * we switched to root in the new user namespace further above. Only | |
1375 | * drop groups if we can, so ensure that we have necessary privilege. | |
928b1f04 YT |
1376 | */ |
1377 | #if HAVE_LIBCAP | |
1378 | have_cap_setgid = lxc_proc_cap_is_set(CAP_SETGID, CAP_EFFECTIVE); | |
1379 | #else | |
1380 | have_cap_setgid = false; | |
1381 | #endif | |
1382 | if (lxc_list_empty(&handler->conf->id_map) && have_cap_setgid) { | |
e96a536c CB |
1383 | ret = lxc_setgroups(0, NULL); |
1384 | if (ret < 0) | |
76bdf299 CB |
1385 | goto out_warn_father; |
1386 | } | |
1387 | ||
e96a536c CB |
1388 | ret = lxc_switch_uid_gid(new_uid, new_gid); |
1389 | if (ret < 0) | |
928b1f04 YT |
1390 | goto out_warn_father; |
1391 | ||
611ddd34 CB |
1392 | ret = lxc_ambient_caps_down(); |
1393 | if (ret < 0) { | |
6f5e532f | 1394 | ERROR("Failed to clear ambient capabilities"); |
611ddd34 CB |
1395 | goto out_warn_father; |
1396 | } | |
1397 | ||
258f8051 CB |
1398 | if (handler->conf->monitor_signal_pdeath != SIGKILL) { |
1399 | ret = lxc_set_death_signal(handler->conf->monitor_signal_pdeath); | |
1400 | if (ret < 0) { | |
1401 | SYSERROR("Failed to set PR_SET_PDEATHSIG to %d", | |
1402 | handler->conf->monitor_signal_pdeath); | |
1403 | goto out_warn_father; | |
1404 | } | |
1405 | } | |
1406 | ||
408da065 CB |
1407 | /* After this call, we are in error because this ops should not return |
1408 | * as it execs. | |
1409 | */ | |
c4ea60df | 1410 | handler->ops->start(handler, handler->data); |
50e98013 DL |
1411 | |
1412 | out_warn_father: | |
408da065 CB |
1413 | /* We want the parent to know something went wrong, so we return a |
1414 | * special error code. | |
1415 | */ | |
d1ccb562 | 1416 | lxc_sync_wake_parent(handler, LXC_SYNC_ERROR); |
c44de748 AM |
1417 | |
1418 | out_error: | |
1419 | if (devnull_fd >= 0) | |
1420 | close(devnull_fd); | |
1421 | ||
50e98013 DL |
1422 | return -1; |
1423 | } | |
1424 | ||
ae467c54 | 1425 | static int lxc_recv_ttys_from_child(struct lxc_handler *handler) |
e8bd4e43 | 1426 | { |
ae467c54 | 1427 | int i; |
2520facd | 1428 | struct lxc_terminal_info *tty; |
ae467c54 | 1429 | int ret = -1; |
c6012571 | 1430 | int sock = handler->data_sock[1]; |
e8bd4e43 | 1431 | struct lxc_conf *conf = handler->conf; |
0e4be3cf | 1432 | struct lxc_tty_info *ttys = &conf->ttys; |
e8bd4e43 | 1433 | |
885766f5 | 1434 | if (!conf->ttys.max) |
e8bd4e43 SH |
1435 | return 0; |
1436 | ||
885766f5 | 1437 | ttys->tty = malloc(sizeof(*ttys->tty) * ttys->max); |
0e4be3cf | 1438 | if (!ttys->tty) |
e8bd4e43 | 1439 | return -1; |
e8bd4e43 | 1440 | |
885766f5 | 1441 | for (i = 0; i < conf->ttys.max; i++) { |
672c1e58 CB |
1442 | int ttyfds[2]; |
1443 | ||
1444 | ret = lxc_abstract_unix_recv_fds(sock, ttyfds, 2, NULL, 0); | |
1445 | if (ret < 0) | |
1446 | break; | |
ae467c54 | 1447 | |
0e4be3cf | 1448 | tty = &ttys->tty[i]; |
2520facd CB |
1449 | tty->busy = 0; |
1450 | tty->master = ttyfds[0]; | |
1451 | tty->slave = ttyfds[1]; | |
672c1e58 | 1452 | TRACE("Received pty with master fd %d and slave fd %d from " |
2520facd | 1453 | "parent", tty->master, tty->slave); |
e8bd4e43 | 1454 | } |
6d1400b5 | 1455 | |
ae467c54 | 1456 | if (ret < 0) |
6d1400b5 | 1457 | SYSERROR("Failed to receive %zu ttys from child", ttys->max); |
ae467c54 | 1458 | else |
885766f5 | 1459 | TRACE("Received %zu ttys from child", ttys->max); |
ae467c54 CB |
1460 | |
1461 | return ret; | |
e8bd4e43 SH |
1462 | } |
1463 | ||
5af9369b | 1464 | int resolve_clone_flags(struct lxc_handler *handler) |
f813849c | 1465 | { |
8bc8c715 CB |
1466 | int i; |
1467 | struct lxc_conf *conf = handler->conf; | |
f813849c | 1468 | |
8bc8c715 CB |
1469 | for (i = 0; i < LXC_NS_MAX; i++) { |
1470 | if (conf->ns_keep != 0) { | |
1471 | if ((conf->ns_keep & ns_info[i].clone_flag) == 0) | |
becad0ec | 1472 | handler->ns_clone_flags |= ns_info[i].clone_flag; |
8bc8c715 CB |
1473 | } else if (conf->ns_clone != 0) { |
1474 | if ((conf->ns_clone & ns_info[i].clone_flag) > 0) | |
becad0ec | 1475 | handler->ns_clone_flags |= ns_info[i].clone_flag; |
8bc8c715 CB |
1476 | } else { |
1477 | if (i == LXC_NS_USER && lxc_list_empty(&handler->conf->id_map)) | |
1478 | continue; | |
1479 | ||
1480 | if (i == LXC_NS_NET && lxc_requests_empty_network(handler)) | |
1481 | continue; | |
1482 | ||
7bd05339 CB |
1483 | if (i == LXC_NS_CGROUP && !cgns_supported()) |
1484 | continue; | |
1485 | ||
becad0ec | 1486 | handler->ns_clone_flags |= ns_info[i].clone_flag; |
8bc8c715 | 1487 | } |
03df7ab5 | 1488 | |
8bc8c715 CB |
1489 | if (!conf->ns_share[i]) |
1490 | continue; | |
5af9369b | 1491 | |
becad0ec | 1492 | handler->ns_clone_flags &= ~ns_info[i].clone_flag; |
8bc8c715 | 1493 | TRACE("Sharing %s namespace", ns_info[i].proc_name); |
5af9369b CB |
1494 | } |
1495 | ||
1496 | return 0; | |
f813849c TA |
1497 | } |
1498 | ||
8deca6c9 CB |
1499 | /* Note that this function is used with clone(CLONE_VM). Some glibc versions |
1500 | * used to reset the pid/tid to -1 when CLONE_VM was used without CLONE_THREAD. | |
1501 | * But since the memory between parent and child is shared on CLONE_VM this | |
1502 | * would invalidate the getpid() cache that glibc used to maintain and so | |
1503 | * getpid() in the child would return the parent's pid. This is all fixed in | |
1504 | * newer glibc versions where the getpid() cache is removed and the pid/tid is | |
1505 | * not reset anymore. | |
1506 | * However, if for whatever reason you - dear commiter - somehow need to get the | |
1507 | * pid of the dummy intermediate process for do_share_ns() you need to call | |
0059379f CB |
1508 | * lxc_raw_getpid(). The next lxc_raw_clone() call does not employ CLONE_VM and |
1509 | * will be fine. | |
8deca6c9 CB |
1510 | */ |
1511 | static inline int do_share_ns(void *arg) | |
fa3a5b22 | 1512 | { |
8deca6c9 | 1513 | int i, flags, ret; |
fa3a5b22 CB |
1514 | struct lxc_handler *handler = arg; |
1515 | ||
8deca6c9 CB |
1516 | for (i = 0; i < LXC_NS_MAX; i++) { |
1517 | if (handler->nsfd[i] < 0) | |
1518 | continue; | |
fa3a5b22 | 1519 | |
8deca6c9 | 1520 | ret = setns(handler->nsfd[i], 0); |
c0b48eff CB |
1521 | if (ret < 0) { |
1522 | /* | |
1523 | * Note that joining a user and/or mount namespace | |
1524 | * requires the process is not multithreaded otherwise | |
1525 | * setns() will fail here. | |
1526 | */ | |
1527 | SYSERROR("Failed to inherit %s namespace", | |
1528 | ns_info[i].proc_name); | |
8deca6c9 | 1529 | return -1; |
c0b48eff | 1530 | } |
fa3a5b22 | 1531 | |
8deca6c9 | 1532 | DEBUG("Inherited %s namespace", ns_info[i].proc_name); |
fa3a5b22 CB |
1533 | } |
1534 | ||
becad0ec | 1535 | flags = handler->ns_on_clone_flags; |
8deca6c9 | 1536 | flags |= CLONE_PARENT; |
ac2ba696 | 1537 | handler->pid = lxc_raw_clone_cb(do_start, handler, flags); |
8deca6c9 | 1538 | if (handler->pid < 0) |
fa3a5b22 CB |
1539 | return -1; |
1540 | ||
8deca6c9 | 1541 | return 0; |
fa3a5b22 CB |
1542 | } |
1543 | ||
ea0e06dd CB |
1544 | static int lxc_setup_shmount(struct lxc_conf *conf) |
1545 | { | |
4d6cc24c LT |
1546 | size_t len_cont; |
1547 | char *full_cont_path; | |
1548 | int ret = -1; | |
1549 | ||
ea0e06dd | 1550 | /* Construct the shmount path under the container root. */ |
7a41e857 | 1551 | len_cont = strlen(conf->rootfs.mount) + 1 + strlen(conf->shmount.path_cont); |
4d6cc24c LT |
1552 | /* +1 for the terminating '\0' */ |
1553 | full_cont_path = malloc(len_cont + 1); | |
ea0e06dd | 1554 | if (!full_cont_path) { |
4d6cc24c LT |
1555 | SYSERROR("Not enough memory"); |
1556 | return -ENOMEM; | |
1557 | } | |
ea0e06dd CB |
1558 | |
1559 | ret = snprintf(full_cont_path, len_cont + 1, "%s/%s", | |
1560 | conf->rootfs.mount, conf->shmount.path_cont); | |
4d6cc24c LT |
1561 | if (ret < 0 || ret >= len_cont + 1) { |
1562 | SYSERROR("Failed to create filename"); | |
1563 | free(full_cont_path); | |
1564 | return -1; | |
1565 | } | |
1566 | ||
ea0e06dd | 1567 | /* Check if shmount point is already set up. */ |
7a41e857 | 1568 | if (is_shared_mountpoint(conf->shmount.path_host)) { |
ea0e06dd CB |
1569 | INFO("Path \"%s\" is already MS_SHARED. Reusing", |
1570 | conf->shmount.path_host); | |
4d6cc24c LT |
1571 | free(full_cont_path); |
1572 | return 0; | |
1573 | } | |
1574 | ||
1575 | /* Create host and cont mount paths */ | |
7a41e857 | 1576 | ret = mkdir_p(conf->shmount.path_host, 0711); |
4d6cc24c | 1577 | if (ret < 0 && errno != EEXIST) { |
ea0e06dd CB |
1578 | SYSERROR("Failed to create directory \"%s\"", |
1579 | conf->shmount.path_host); | |
4d6cc24c LT |
1580 | free(full_cont_path); |
1581 | return ret; | |
1582 | } | |
1583 | ||
1584 | ret = mkdir_p(full_cont_path, 0711); | |
1585 | if (ret < 0 && errno != EEXIST) { | |
1586 | SYSERROR("Failed to create directory \"%s\"", full_cont_path); | |
1587 | free(full_cont_path); | |
1588 | return ret; | |
1589 | } | |
1590 | ||
1591 | /* Prepare host mountpoint */ | |
ea0e06dd CB |
1592 | ret = mount("tmpfs", conf->shmount.path_host, "tmpfs", 0, |
1593 | "size=100k,mode=0711"); | |
4d6cc24c | 1594 | if (ret < 0) { |
7a41e857 | 1595 | SYSERROR("Failed to mount \"%s\"", conf->shmount.path_host); |
4d6cc24c LT |
1596 | free(full_cont_path); |
1597 | return ret; | |
1598 | } | |
ea0e06dd | 1599 | |
7a41e857 | 1600 | ret = mount(conf->shmount.path_host, conf->shmount.path_host, "none", |
ea0e06dd | 1601 | MS_REC | MS_SHARED, ""); |
4d6cc24c | 1602 | if (ret < 0) { |
7a41e857 | 1603 | SYSERROR("Failed to make shared \"%s\"", conf->shmount.path_host); |
4d6cc24c LT |
1604 | free(full_cont_path); |
1605 | return ret; | |
1606 | } | |
1607 | ||
ea0e06dd | 1608 | INFO("Setup shared mount point \"%s\"", conf->shmount.path_host); |
4d6cc24c LT |
1609 | free(full_cont_path); |
1610 | return 0; | |
1611 | } | |
1612 | ||
480588e6 CB |
1613 | /* lxc_spawn() performs crucial setup tasks and clone()s the new process which |
1614 | * exec()s the requested container binary. | |
1615 | * Note that lxc_spawn() runs in the parent namespaces. Any operations performed | |
1616 | * right here should be double checked if they'd pose a security risk. (For | |
1617 | * example, any {u}mount() operations performed here will be reflected on the | |
1618 | * host!) | |
1619 | */ | |
74a3920a | 1620 | static int lxc_spawn(struct lxc_handler *handler) |
59eb99ba | 1621 | { |
8deca6c9 | 1622 | int i, ret; |
08dd2805 | 1623 | char pidstr[20]; |
57927bf2 CB |
1624 | bool wants_to_map_ids; |
1625 | struct lxc_list *id_map; | |
6e5fc7a5 | 1626 | const char *name = handler->name; |
28d9e29e | 1627 | const char *lxcpath = handler->lxcpath; |
2202afc9 | 1628 | bool share_ns = false; |
28d9e29e | 1629 | struct lxc_conf *conf = handler->conf; |
2202afc9 | 1630 | struct cgroup_ops *cgroup_ops = handler->cgroup_ops; |
9f30a190 | 1631 | |
28d9e29e | 1632 | id_map = &conf->id_map; |
57927bf2 | 1633 | wants_to_map_ids = !lxc_list_empty(id_map); |
2f2623ec | 1634 | |
28d9e29e | 1635 | for (i = 0; i < LXC_NS_MAX; i++) { |
b074bbf1 | 1636 | if (!conf->ns_share[i]) |
28d9e29e CB |
1637 | continue; |
1638 | ||
b074bbf1 | 1639 | handler->nsfd[i] = lxc_inherit_namespace(conf->ns_share[i], lxcpath, ns_info[i].proc_name); |
28d9e29e CB |
1640 | if (handler->nsfd[i] < 0) |
1641 | return -1; | |
fa3a5b22 | 1642 | |
8deca6c9 | 1643 | share_ns = true; |
28d9e29e | 1644 | } |
50e98013 | 1645 | |
cfc62c60 CB |
1646 | ret = lxc_sync_init(handler); |
1647 | if (ret < 0) | |
9d7f9e52 | 1648 | return -1; |
0ad19a3f | 1649 | |
b9f522c5 | 1650 | ret = socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0, |
790255cf | 1651 | handler->data_sock); |
c6012571 | 1652 | if (ret < 0) { |
e8bd4e43 SH |
1653 | lxc_sync_fini(handler); |
1654 | return -1; | |
1655 | } | |
1656 | ||
5af9369b CB |
1657 | ret = resolve_clone_flags(handler); |
1658 | if (ret < 0) { | |
1659 | lxc_sync_fini(handler); | |
1660 | return -1; | |
1661 | } | |
9f30a190 | 1662 | |
becad0ec | 1663 | if (handler->ns_clone_flags & CLONE_NEWNET) { |
28d9e29e | 1664 | if (!lxc_list_empty(&conf->network)) { |
26b797f3 | 1665 | |
9f30a190 MM |
1666 | /* Find gateway addresses from the link device, which is |
1667 | * no longer accessible inside the container. Do this | |
1668 | * before creating network interfaces, since goto | |
408da065 CB |
1669 | * out_delete_net does not work before lxc_clone. |
1670 | */ | |
cfc62c60 CB |
1671 | ret = lxc_find_gateway_addresses(handler); |
1672 | if (ret < 0) { | |
1673 | ERROR("Failed to find gateway addresses"); | |
9f30a190 MM |
1674 | lxc_sync_fini(handler); |
1675 | return -1; | |
1676 | } | |
1677 | ||
408da065 CB |
1678 | /* That should be done before the clone because we will |
1679 | * fill the netdev index and use them in the child. | |
9f30a190 | 1680 | */ |
cfc62c60 CB |
1681 | ret = lxc_create_network_priv(handler); |
1682 | if (ret < 0) { | |
1683 | ERROR("Failed to create the network"); | |
9f30a190 MM |
1684 | lxc_sync_fini(handler); |
1685 | return -1; | |
1686 | } | |
19a26f82 | 1687 | } |
82d5ae15 DL |
1688 | } |
1689 | ||
ea0e06dd CB |
1690 | if (conf->shmount.path_host) { |
1691 | if (!conf->shmount.path_cont) { | |
4d6cc24c LT |
1692 | lxc_sync_fini(handler); |
1693 | return -1; | |
ea0e06dd CB |
1694 | } |
1695 | ||
4d6cc24c LT |
1696 | ret = lxc_setup_shmount(conf); |
1697 | if (ret < 0) { | |
1698 | ERROR("Failed to setup shared mount point"); | |
1699 | lxc_sync_fini(handler); | |
1700 | return -1; | |
1701 | } | |
1702 | } | |
1703 | ||
1704 | if (!cgroup_init(handler)) { | |
1705 | ERROR("Failed initializing cgroup support"); | |
1706 | goto out_delete_net; | |
1707 | } | |
1708 | ||
2202afc9 | 1709 | if (!cgroup_ops->create(cgroup_ops, handler)) { |
cfc62c60 | 1710 | ERROR("Failed creating cgroups"); |
47d8fb3b CS |
1711 | goto out_delete_net; |
1712 | } | |
1713 | ||
408da065 CB |
1714 | /* If the rootfs is not a blockdev, prevent the container from marking |
1715 | * it readonly. | |
1716 | * If the container is unprivileged then skip rootfs pinning. | |
0c547523 | 1717 | */ |
0ee35059 | 1718 | if (!wants_to_map_ids) { |
28d9e29e | 1719 | handler->pinfd = pin_rootfs(conf->rootfs.path); |
5e32a990 | 1720 | if (handler->pinfd == -1) |
cfc62c60 | 1721 | INFO("Failed to pin the rootfs for container \"%s\"", handler->name); |
5e32a990 | 1722 | } |
0c547523 | 1723 | |
408da065 | 1724 | /* Create a process in a new set of namespaces. */ |
becad0ec CB |
1725 | handler->ns_on_clone_flags = handler->ns_clone_flags; |
1726 | if (handler->ns_clone_flags & CLONE_NEWUSER) { | |
408da065 CB |
1727 | /* If CLONE_NEWUSER and CLONE_NEWNET was requested, we need to |
1728 | * clone a new user namespace first and only later unshare our | |
1729 | * network namespace to ensure that network devices ownership is | |
1730 | * set up correctly. | |
1731 | */ | |
becad0ec | 1732 | handler->ns_on_clone_flags &= ~CLONE_NEWNET; |
408da065 | 1733 | } |
5af9369b | 1734 | /* The cgroup namespace gets unshare()ed not clone()ed. */ |
becad0ec | 1735 | handler->ns_on_clone_flags &= ~CLONE_NEWCGROUP; |
732375f5 | 1736 | |
4e232466 CB |
1737 | if (share_ns) { |
1738 | pid_t attacher_pid; | |
1739 | ||
cfc62c60 CB |
1740 | attacher_pid = lxc_clone(do_share_ns, handler, |
1741 | CLONE_VFORK | CLONE_VM | CLONE_FILES); | |
4e232466 CB |
1742 | if (attacher_pid < 0) { |
1743 | SYSERROR(LXC_CLONE_ERROR); | |
1744 | goto out_delete_net; | |
1745 | } | |
1746 | ||
1747 | ret = wait_for_pid(attacher_pid); | |
1748 | if (ret < 0) { | |
1749 | SYSERROR("Intermediate process failed"); | |
1750 | goto out_delete_net; | |
1751 | } | |
1752 | } else { | |
cfc62c60 | 1753 | handler->pid = lxc_raw_clone_cb(do_start, handler, |
becad0ec | 1754 | handler->ns_on_clone_flags); |
4e232466 CB |
1755 | } |
1756 | if (handler->pid < 0) { | |
1757 | SYSERROR(LXC_CLONE_ERROR); | |
7fef7a06 | 1758 | goto out_delete_net; |
0ad19a3f | 1759 | } |
fa3a5b22 | 1760 | TRACE("Cloned child process %d", handler->pid); |
1bd8d726 | 1761 | |
9662e444 | 1762 | for (i = 0; i < LXC_NS_MAX; i++) |
becad0ec | 1763 | if (handler->ns_on_clone_flags & ns_info[i].clone_flag) |
28d9e29e | 1764 | INFO("Cloned %s", ns_info[i].flag_name); |
0ad19a3f | 1765 | |
4cb53844 | 1766 | if (!lxc_try_preserve_namespaces(handler, handler->ns_on_clone_flags, handler->pid)) { |
28d9e29e CB |
1767 | ERROR("Failed to preserve cloned namespaces for lxc.hook.stop"); |
1768 | goto out_delete_net; | |
1769 | } | |
b6b2b194 | 1770 | |
3c22086f CLG |
1771 | lxc_sync_fini_child(handler); |
1772 | ||
408da065 CB |
1773 | /* Map the container uids. The container became an invalid userid the |
1774 | * moment it was cloned with CLONE_NEWUSER. This call doesn't change | |
1775 | * anything immediately, but allows the container to setuid(0) (0 being | |
1776 | * mapped to something else on the host.) later to become a valid uid | |
1777 | * again. | |
1778 | */ | |
fa3a5b22 | 1779 | if (wants_to_map_ids) { |
df3c5314 | 1780 | if (!handler->conf->ns_share[LXC_NS_USER] && |
46186acd | 1781 | (handler->conf->ns_keep & CLONE_NEWUSER) == 0) { |
fa3a5b22 CB |
1782 | ret = lxc_map_ids(id_map, handler->pid); |
1783 | if (ret < 0) { | |
1784 | ERROR("Failed to set up id mapping."); | |
1785 | goto out_delete_net; | |
1786 | } | |
1787 | } | |
5b1e83cb SH |
1788 | } |
1789 | ||
cfc62c60 CB |
1790 | ret = lxc_sync_wake_child(handler, LXC_SYNC_STARTUP); |
1791 | if (ret < 0) | |
5b1e83cb | 1792 | goto out_delete_net; |
5b1e83cb | 1793 | |
cfc62c60 CB |
1794 | ret = lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE); |
1795 | if (ret < 0) | |
5b1e83cb | 1796 | goto out_delete_net; |
0ad19a3f | 1797 | |
2202afc9 | 1798 | if (!cgroup_ops->setup_limits(cgroup_ops, handler->conf, false)) { |
cfc62c60 | 1799 | ERROR("Failed to setup cgroup limits for container \"%s\"", name); |
6031a6e5 DE |
1800 | goto out_delete_net; |
1801 | } | |
1802 | ||
2202afc9 | 1803 | if (!cgroup_ops->enter(cgroup_ops, handler->pid)) |
7fef7a06 | 1804 | goto out_delete_net; |
218d4250 | 1805 | |
2202afc9 | 1806 | if (!cgroup_ops->chown(cgroup_ops, handler->conf)) |
0996e18a SH |
1807 | goto out_delete_net; |
1808 | ||
fa3a5b22 | 1809 | /* Now we're ready to preserve the network namespace */ |
4cb53844 | 1810 | ret = lxc_try_preserve_ns(handler->pid, "net"); |
fa3a5b22 | 1811 | if (ret < 0) { |
4cb53844 | 1812 | if (ret != -EOPNOTSUPP) { |
6d1400b5 | 1813 | SYSERROR("Failed to preserve net namespace"); |
4cb53844 CB |
1814 | goto out_delete_net; |
1815 | } | |
1816 | } else { | |
1817 | handler->nsfd[LXC_NS_NET] = ret; | |
1818 | DEBUG("Preserved net namespace via fd %d", ret); | |
fa3a5b22 | 1819 | } |
fa3a5b22 | 1820 | |
408da065 | 1821 | /* Create the network configuration. */ |
becad0ec | 1822 | if (handler->ns_clone_flags & CLONE_NEWNET) { |
cfc62c60 CB |
1823 | ret = lxc_network_move_created_netdev_priv(handler->lxcpath, |
1824 | handler->name, | |
1825 | &conf->network, | |
1826 | handler->pid); | |
1827 | if (ret < 0) { | |
1828 | ERROR("Failed to create the configured network"); | |
7fef7a06 | 1829 | goto out_delete_net; |
82d5ae15 | 1830 | } |
74c6e2b0 | 1831 | |
cfc62c60 | 1832 | ret = lxc_create_network_unpriv(handler->lxcpath, handler->name, |
4d781681 | 1833 | &conf->network, handler->pid, conf->hooks_version); |
cfc62c60 CB |
1834 | if (ret < 0) { |
1835 | ERROR("Failed to create the configured network"); | |
74c6e2b0 CB |
1836 | goto out_delete_net; |
1837 | } | |
0ad19a3f | 1838 | } |
1839 | ||
cfc62c60 CB |
1840 | ret = lxc_network_send_veth_names_to_child(handler); |
1841 | if (ret < 0) { | |
7ab1ba02 CB |
1842 | ERROR("Failed to send veth names to child"); |
1843 | goto out_delete_net; | |
658979c5 SH |
1844 | } |
1845 | ||
61d7a733 YT |
1846 | if (!lxc_list_empty(&conf->procs)) { |
1847 | ret = setup_proc_filesystem(&conf->procs, handler->pid); | |
1848 | if (ret < 0) | |
1849 | goto out_delete_net; | |
1850 | } | |
1851 | ||
408da065 CB |
1852 | /* Tell the child to continue its initialization. We'll get |
1853 | * LXC_SYNC_CGROUP when it is ready for us to setup cgroups. | |
3c22086f | 1854 | */ |
cfc62c60 CB |
1855 | ret = lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE); |
1856 | if (ret < 0) | |
544a48a0 SH |
1857 | goto out_delete_net; |
1858 | ||
cfc62c60 CB |
1859 | if (!lxc_list_empty(&conf->limits)) { |
1860 | ret = setup_resource_limits(&conf->limits, handler->pid); | |
1861 | if (ret < 0) { | |
1862 | ERROR("Failed to setup resource limits"); | |
1863 | goto out_delete_net; | |
1864 | } | |
c6d09e15 WB |
1865 | } |
1866 | ||
cfc62c60 CB |
1867 | ret = lxc_sync_barrier_child(handler, LXC_SYNC_CGROUP_UNSHARE); |
1868 | if (ret < 0) | |
f4152036 CB |
1869 | goto out_delete_net; |
1870 | ||
2202afc9 | 1871 | if (!cgroup_ops->setup_limits(cgroup_ops, handler->conf, true)) { |
cfc62c60 | 1872 | ERROR("Failed to setup legacy device cgroup controller limits"); |
b98f7d6e | 1873 | goto out_delete_net; |
544a48a0 | 1874 | } |
cfc62c60 | 1875 | TRACE("Set up legacy device cgroup controller limits"); |
544a48a0 | 1876 | |
becad0ec | 1877 | if (handler->ns_clone_flags & CLONE_NEWCGROUP) { |
8bf3abfb | 1878 | /* Now we're ready to preserve the cgroup namespace */ |
4cb53844 | 1879 | ret = lxc_try_preserve_ns(handler->pid, "cgroup"); |
8bf3abfb | 1880 | if (ret < 0) { |
4cb53844 | 1881 | if (ret != -EOPNOTSUPP) { |
6d1400b5 | 1882 | SYSERROR("Failed to preserve cgroup namespace"); |
4cb53844 CB |
1883 | goto out_delete_net; |
1884 | } | |
1885 | } else { | |
1886 | handler->nsfd[LXC_NS_CGROUP] = ret; | |
1887 | DEBUG("Preserved cgroup namespace via fd %d", ret); | |
8bf3abfb | 1888 | } |
8bf3abfb CB |
1889 | } |
1890 | ||
5b7f756a CB |
1891 | ret = snprintf(pidstr, 20, "%d", handler->pid); |
1892 | if (ret < 0 || ret >= 20) | |
1893 | goto out_delete_net; | |
1894 | ||
cfc62c60 CB |
1895 | ret = setenv("LXC_PID", pidstr, 1); |
1896 | if (ret < 0) | |
1897 | SYSERROR("Failed to set environment variable: LXC_PID=%s", pidstr); | |
08dd2805 SH |
1898 | |
1899 | /* Run any host-side start hooks */ | |
cfc62c60 CB |
1900 | ret = run_lxc_hooks(name, "start-host", conf, NULL); |
1901 | if (ret < 0) { | |
1902 | ERROR("Failed to run lxc.hook.start-host"); | |
341ed84c | 1903 | goto out_delete_net; |
08dd2805 SH |
1904 | } |
1905 | ||
408da065 CB |
1906 | /* Tell the child to complete its initialization and wait for it to exec |
1907 | * or return an error. (The child will never return | |
08dd2805 | 1908 | * LXC_SYNC_READY_START+1. It will either close the sync pipe, causing |
408da065 CB |
1909 | * lxc_sync_barrier_child to return success, or return a different |
1910 | * value, causing us to error out). | |
544a48a0 | 1911 | */ |
cfc62c60 CB |
1912 | ret = lxc_sync_barrier_child(handler, LXC_SYNC_READY_START); |
1913 | if (ret < 0) | |
341ed84c | 1914 | goto out_delete_net; |
0ad19a3f | 1915 | |
cfc62c60 CB |
1916 | ret = lxc_network_recv_name_and_ifindex_from_child(handler); |
1917 | if (ret < 0) { | |
790255cf CB |
1918 | ERROR("Failed to receive names and ifindices for network " |
1919 | "devices from child"); | |
1920 | goto out_delete_net; | |
1921 | } | |
1922 | ||
1923 | /* Now all networks are created, network devices are moved into place, | |
1924 | * and the correct names and ifindeces in the respective namespaces have | |
1925 | * been recorded. The corresponding structs have now all been filled. So | |
1926 | * log them for debugging purposes. | |
1927 | */ | |
28d9e29e | 1928 | lxc_log_configured_netdevs(conf); |
790255cf | 1929 | |
f4152036 | 1930 | /* Read tty fds allocated by child. */ |
cfc62c60 CB |
1931 | ret = lxc_recv_ttys_from_child(handler); |
1932 | if (ret < 0) { | |
1933 | ERROR("Failed to receive tty info from child process"); | |
f4152036 CB |
1934 | goto out_delete_net; |
1935 | } | |
1936 | ||
cfc62c60 CB |
1937 | ret = handler->ops->post_start(handler, handler->data); |
1938 | if (ret < 0) | |
e6126dbe MN |
1939 | goto out_abort; |
1940 | ||
cfc62c60 CB |
1941 | ret = lxc_set_state(name, handler, RUNNING); |
1942 | if (ret < 0) { | |
1943 | ERROR("Failed to set state to \"%s\"", lxc_state2str(RUNNING)); | |
59eb99ba | 1944 | goto out_abort; |
3f21c114 | 1945 | } |
22ebac19 | 1946 | |
3c22086f | 1947 | lxc_sync_fini(handler); |
0c547523 | 1948 | |
e6126dbe | 1949 | return 0; |
1ac470c0 | 1950 | |
7fef7a06 | 1951 | out_delete_net: |
becad0ec | 1952 | if (handler->ns_clone_flags & CLONE_NEWNET) |
bb84beda | 1953 | lxc_delete_network(handler); |
1bd8d726 | 1954 | |
59eb99ba DL |
1955 | out_abort: |
1956 | lxc_abort(name, handler); | |
3c22086f | 1957 | lxc_sync_fini(handler); |
5c068da9 SH |
1958 | if (handler->pinfd >= 0) { |
1959 | close(handler->pinfd); | |
1960 | handler->pinfd = -1; | |
1961 | } | |
1962 | ||
b79fcd86 | 1963 | return -1; |
59eb99ba | 1964 | } |
0ad19a3f | 1965 | |
aa460476 | 1966 | int __lxc_start(const char *name, struct lxc_handler *handler, |
507cee36 | 1967 | struct lxc_operations* ops, void *data, const char *lxcpath, |
a3b4f3d6 | 1968 | bool backgrounded, int *error_num) |
59eb99ba | 1969 | { |
c30e9b19 | 1970 | int ret, status; |
aa460476 | 1971 | struct lxc_conf *conf = handler->conf; |
80090207 | 1972 | |
c30e9b19 CB |
1973 | ret = lxc_init(name, handler); |
1974 | if (ret < 0) { | |
1975 | ERROR("Failed to initialize container \"%s\"", name); | |
66aeffc7 | 1976 | return -1; |
0ad19a3f | 1977 | } |
ee70bf78 CLG |
1978 | handler->ops = ops; |
1979 | handler->data = data; | |
507cee36 | 1980 | handler->backgrounded = backgrounded; |
e6126dbe | 1981 | |
76a26f55 | 1982 | if (!attach_block_device(handler->conf)) { |
c30e9b19 | 1983 | ERROR("Failed to attach block device"); |
76a26f55 SH |
1984 | goto out_fini_nonet; |
1985 | } | |
1986 | ||
35120d9c | 1987 | if (geteuid() == 0 && !lxc_list_empty(&conf->id_map)) { |
408da065 | 1988 | /* If the backing store is a device, mount it here and now. */ |
35120d9c | 1989 | if (rootfs_is_blockdev(conf)) { |
c30e9b19 CB |
1990 | ret = unshare(CLONE_NEWNS); |
1991 | if (ret < 0) { | |
1992 | ERROR("Failed to unshare CLONE_NEWNS"); | |
35120d9c SH |
1993 | goto out_fini_nonet; |
1994 | } | |
c30e9b19 | 1995 | INFO("Unshared CLONE_NEWNS"); |
408da065 | 1996 | |
6a0c909a | 1997 | remount_all_slave(); |
8ce1abc2 | 1998 | ret = lxc_setup_rootfs_prepare_root(conf, name, lxcpath); |
c30e9b19 CB |
1999 | if (ret < 0) { |
2000 | ERROR("Error setting up rootfs mount as root before spawn"); | |
35120d9c SH |
2001 | goto out_fini_nonet; |
2002 | } | |
c30e9b19 | 2003 | INFO("Set up container rootfs as host root"); |
35120d9c SH |
2004 | } |
2005 | } | |
2006 | ||
c30e9b19 CB |
2007 | ret = lxc_spawn(handler); |
2008 | if (ret < 0) { | |
2009 | ERROR("Failed to spawn container \"%s\"", name); | |
76a26f55 | 2010 | goto out_detach_blockdev; |
0ad19a3f | 2011 | } |
a9e1109e CB |
2012 | /* close parent side of data socket */ |
2013 | close(handler->data_sock[0]); | |
2014 | handler->data_sock[0] = -1; | |
2015 | close(handler->data_sock[1]); | |
2016 | handler->data_sock[1] = -1; | |
0ad19a3f | 2017 | |
80308d07 | 2018 | handler->conf->reboot = REBOOT_NONE; |
8bee8851 | 2019 | |
c30e9b19 CB |
2020 | ret = lxc_poll(name, handler); |
2021 | if (ret) { | |
2022 | ERROR("LXC mainloop exited with error: %d", ret); | |
59eb99ba DL |
2023 | goto out_abort; |
2024 | } | |
0ad19a3f | 2025 | |
321db026 DJ |
2026 | if (!handler->init_died && handler->pid > 0) { |
2027 | ERROR("Child process is not killed"); | |
2028 | goto out_abort; | |
2029 | } | |
2030 | ||
c30e9b19 CB |
2031 | status = lxc_wait_for_pid_status(handler->pid); |
2032 | if (status < 0) | |
2033 | SYSERROR("Failed to retrieve status for %d", handler->pid); | |
e043236e | 2034 | |
408da065 CB |
2035 | /* If the child process exited but was not signaled, it didn't call |
2036 | * reboot. This should mean it was an lxc-execute which simply exited. | |
2037 | * In any case, treat it as a 'halt'. | |
8b004f07 | 2038 | */ |
d028235d | 2039 | if (WIFSIGNALED(status)) { |
8b004f07 SH |
2040 | switch(WTERMSIG(status)) { |
2041 | case SIGINT: /* halt */ | |
c30e9b19 | 2042 | DEBUG("Container \"%s\" is halting", name); |
8b004f07 SH |
2043 | break; |
2044 | case SIGHUP: /* reboot */ | |
c30e9b19 | 2045 | DEBUG("Container \"%s\" is rebooting", name); |
80308d07 | 2046 | handler->conf->reboot = REBOOT_REQ; |
8b004f07 | 2047 | break; |
c2b9bd9e | 2048 | case SIGSYS: /* seccomp */ |
c30e9b19 | 2049 | DEBUG("Container \"%s\" violated its seccomp policy", name); |
c2b9bd9e | 2050 | break; |
8b004f07 | 2051 | default: |
c30e9b19 | 2052 | DEBUG("Unknown exit status for container \"%s\" init %d", name, WTERMSIG(status)); |
8b004f07 SH |
2053 | break; |
2054 | } | |
d028235d | 2055 | } |
828695d9 | 2056 | |
c30e9b19 CB |
2057 | ret = lxc_restore_phys_nics_to_netns(handler); |
2058 | if (ret < 0) | |
b809f232 CB |
2059 | ERROR("Failed to move physical network devices back to parent " |
2060 | "network namespace"); | |
ce5782df | 2061 | |
5c068da9 SH |
2062 | if (handler->pinfd >= 0) { |
2063 | close(handler->pinfd); | |
2064 | handler->pinfd = -1; | |
2065 | } | |
2066 | ||
1787abca | 2067 | lxc_monitor_send_exit_code(name, status, handler->lxcpath); |
eb808539 | 2068 | lxc_error_set_and_log(handler->pid, status); |
a3b4f3d6 TA |
2069 | if (error_num) |
2070 | *error_num = handler->exit_status; | |
1bd8d726 | 2071 | |
9d7f9e52 | 2072 | out_fini: |
bb84beda | 2073 | lxc_delete_network(handler); |
74a2b586 | 2074 | |
76a26f55 SH |
2075 | out_detach_blockdev: |
2076 | detach_block_device(handler->conf); | |
2077 | ||
74a2b586 | 2078 | out_fini_nonet: |
3a0f472d | 2079 | lxc_fini(name, handler); |
c30e9b19 | 2080 | return ret; |
0ad19a3f | 2081 | |
59eb99ba | 2082 | out_abort: |
3a0f472d | 2083 | lxc_abort(name, handler); |
9d7f9e52 | 2084 | goto out_fini; |
0ad19a3f | 2085 | } |
ee70bf78 CLG |
2086 | |
2087 | struct start_args { | |
2088 | char *const *argv; | |
2089 | }; | |
2090 | ||
2091 | static int start(struct lxc_handler *handler, void* data) | |
2092 | { | |
2093 | struct start_args *arg = data; | |
2094 | ||
984984e4 | 2095 | NOTICE("Exec'ing \"%s\"", arg->argv[0]); |
ee70bf78 CLG |
2096 | |
2097 | execvp(arg->argv[0], arg->argv); | |
984984e4 | 2098 | SYSERROR("Failed to exec \"%s\"", arg->argv[0]); |
ee70bf78 CLG |
2099 | return 0; |
2100 | } | |
2101 | ||
2102 | static int post_start(struct lxc_handler *handler, void* data) | |
2103 | { | |
2104 | struct start_args *arg = data; | |
2105 | ||
4c8e880e | 2106 | NOTICE("Started \"%s\" with pid \"%d\"", arg->argv[0], handler->pid); |
ee70bf78 CLG |
2107 | return 0; |
2108 | } | |
2109 | ||
2110 | static struct lxc_operations start_ops = { | |
2111 | .start = start, | |
2112 | .post_start = post_start | |
2113 | }; | |
2114 | ||
aa460476 | 2115 | int lxc_start(const char *name, char *const argv[], struct lxc_handler *handler, |
a3b4f3d6 | 2116 | const char *lxcpath, bool backgrounded, int *error_num) |
ee70bf78 CLG |
2117 | { |
2118 | struct start_args start_arg = { | |
2119 | .argv = argv, | |
2120 | }; | |
2121 | ||
b2efeb0b | 2122 | TRACE("Doing lxc_start"); |
a3b4f3d6 | 2123 | return __lxc_start(name, handler, &start_ops, &start_arg, lxcpath, backgrounded, error_num); |
ee70bf78 | 2124 | } |
28272964 CB |
2125 | |
2126 | static void lxc_destroy_container_on_signal(struct lxc_handler *handler, | |
2127 | const char *name) | |
2128 | { | |
2129 | char destroy[MAXPATHLEN]; | |
f01f7975 | 2130 | struct lxc_container *c; |
ae3beac9 CB |
2131 | int ret = 0; |
2132 | bool bret = true; | |
2133 | ||
df31363a | 2134 | if (handler->conf->rootfs.path && handler->conf->rootfs.mount) { |
790255cf | 2135 | bret = do_destroy_container(handler); |
28272964 | 2136 | if (!bret) { |
ae3beac9 | 2137 | ERROR("Error destroying rootfs for container \"%s\"", name); |
28272964 CB |
2138 | return; |
2139 | } | |
2140 | } | |
ae3beac9 | 2141 | INFO("Destroyed rootfs for container \"%s\"", name); |
28272964 CB |
2142 | |
2143 | ret = snprintf(destroy, MAXPATHLEN, "%s/%s", handler->lxcpath, name); | |
2144 | if (ret < 0 || ret >= MAXPATHLEN) { | |
ae3beac9 | 2145 | ERROR("Error destroying directory for container \"%s\"", name); |
28272964 CB |
2146 | return; |
2147 | } | |
2148 | ||
f01f7975 CB |
2149 | c = lxc_container_new(name, handler->lxcpath); |
2150 | if (c) { | |
2151 | if (container_disk_lock(c)) { | |
ae3beac9 | 2152 | INFO("Could not update lxc_snapshots file"); |
f01f7975 CB |
2153 | lxc_container_put(c); |
2154 | } else { | |
2155 | mod_all_rdeps(c, false); | |
2156 | container_disk_unlock(c); | |
2157 | lxc_container_put(c); | |
2158 | } | |
2159 | } | |
2160 | ||
d0fbc7ba | 2161 | if (!handler->am_root) |
94b0a3ac CB |
2162 | ret = userns_exec_full(handler->conf, lxc_rmdir_onedev_wrapper, |
2163 | destroy, "lxc_rmdir_onedev_wrapper"); | |
28272964 CB |
2164 | else |
2165 | ret = lxc_rmdir_onedev(destroy, NULL); | |
2166 | ||
2167 | if (ret < 0) { | |
ae3beac9 | 2168 | ERROR("Error destroying directory for container \"%s\"", name); |
28272964 CB |
2169 | return; |
2170 | } | |
ae3beac9 | 2171 | INFO("Destroyed directory for container \"%s\"", name); |
28272964 CB |
2172 | } |
2173 | ||
2174 | static int lxc_rmdir_onedev_wrapper(void *data) | |
2175 | { | |
2176 | char *arg = (char *) data; | |
2177 | return lxc_rmdir_onedev(arg, NULL); | |
2178 | } | |
2179 | ||
134df645 CB |
2180 | static bool do_destroy_container(struct lxc_handler *handler) |
2181 | { | |
94b0a3ac CB |
2182 | int ret; |
2183 | ||
d0fbc7ba | 2184 | if (!handler->am_root) { |
94b0a3ac CB |
2185 | ret = userns_exec_full(handler->conf, storage_destroy_wrapper, |
2186 | handler->conf, "storage_destroy_wrapper"); | |
2187 | if (ret < 0) | |
d028235d | 2188 | return false; |
10bc1861 | 2189 | |
d028235d SG |
2190 | return true; |
2191 | } | |
10bc1861 | 2192 | |
790255cf | 2193 | return storage_destroy(handler->conf); |
28272964 | 2194 | } |