]>
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 | int ret; |
3c319edb CB |
303 | siginfo_t info; |
304 | struct signalfd_siginfo siginfo; | |
305 | struct lxc_handler *hdlr = data; | |
15cd25fd DL |
306 | |
307 | ret = read(fd, &siginfo, sizeof(siginfo)); | |
308 | if (ret < 0) { | |
408da065 | 309 | ERROR("Failed to read signal info from signal file descriptor: %d.", fd); |
15cd25fd DL |
310 | return -1; |
311 | } | |
312 | ||
313 | if (ret != sizeof(siginfo)) { | |
408da065 | 314 | ERROR("Unexpected size for siginfo struct."); |
15cd25fd DL |
315 | return -1; |
316 | } | |
317 | ||
408da065 | 318 | /* Check whether init is running. */ |
80507ee8 | 319 | info.si_pid = 0; |
3c319edb CB |
320 | ret = waitid(P_PID, hdlr->pid, &info, WEXITED | WNOWAIT | WNOHANG); |
321 | if (ret == 0 && info.si_pid == hdlr->pid) | |
322 | hdlr->init_died = true; | |
80507ee8 | 323 | |
f3304a29 | 324 | if (siginfo.ssi_signo != SIGCHLD) { |
3c319edb CB |
325 | kill(hdlr->pid, siginfo.ssi_signo); |
326 | INFO("Forwarded signal %d to pid %d.", siginfo.ssi_signo, hdlr->pid); | |
327 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE : 0; | |
f3304a29 FW |
328 | } |
329 | ||
408da065 CB |
330 | if (siginfo.ssi_code == CLD_STOPPED) { |
331 | INFO("Container init process was stopped."); | |
3c319edb | 332 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE : 0; |
408da065 CB |
333 | } else if (siginfo.ssi_code == CLD_CONTINUED) { |
334 | INFO("Container init process was continued."); | |
3c319edb | 335 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE : 0; |
15cd25fd | 336 | } |
1ac470c0 | 337 | |
408da065 CB |
338 | /* More robustness, protect ourself from a SIGCHLD sent |
339 | * by a process different from the container init. | |
82d89dce | 340 | */ |
3c319edb CB |
341 | if (siginfo.ssi_pid != hdlr->pid) { |
342 | NOTICE("Received SIGCHLD from pid %d instead of container init %d.", siginfo.ssi_pid, hdlr->pid); | |
343 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE : 0; | |
82d89dce DL |
344 | } |
345 | ||
3c319edb CB |
346 | DEBUG("Container init process %d exited.", hdlr->pid); |
347 | return LXC_MAINLOOP_CLOSE; | |
b0a33c1e | 348 | } |
349 | ||
5e5576a4 CB |
350 | static int lxc_serve_state_clients(const char *name, |
351 | struct lxc_handler *handler, | |
352 | lxc_state_t state) | |
66aeffc7 | 353 | { |
dbc9832d CB |
354 | ssize_t ret; |
355 | struct lxc_list *cur, *next; | |
d39b10eb | 356 | struct lxc_state_client *client; |
dbc9832d CB |
357 | struct lxc_msg msg = {.type = lxc_msg_state, .value = state}; |
358 | ||
bc631984 | 359 | process_lock(); |
66aeffc7 | 360 | handler->state = state; |
9dfa4041 | 361 | TRACE("Set container state to %s", lxc_state2str(state)); |
dbc9832d | 362 | |
d39b10eb | 363 | if (lxc_list_empty(&handler->conf->state_clients)) { |
9dfa4041 | 364 | TRACE("No state clients registered"); |
dbc9832d | 365 | process_unlock(); |
65ee1673 | 366 | lxc_monitor_send_state(name, state, handler->lxcpath); |
dbc9832d CB |
367 | return 0; |
368 | } | |
369 | ||
370 | strncpy(msg.name, name, sizeof(msg.name)); | |
371 | msg.name[sizeof(msg.name) - 1] = 0; | |
372 | ||
d39b10eb | 373 | lxc_list_for_each_safe(cur, &handler->conf->state_clients, next) { |
dbc9832d CB |
374 | client = cur->elem; |
375 | ||
d39b10eb | 376 | if (client->states[state] == 0) { |
9dfa4041 | 377 | TRACE("State %s not registered for state client %d", |
dbc9832d CB |
378 | lxc_state2str(state), client->clientfd); |
379 | continue; | |
380 | } | |
381 | ||
9dfa4041 | 382 | TRACE("Sending state %s to state client %d", |
dbc9832d CB |
383 | lxc_state2str(state), client->clientfd); |
384 | ||
385 | again: | |
386 | ret = send(client->clientfd, &msg, sizeof(msg), 0); | |
ee8377bd CB |
387 | if (ret <= 0) { |
388 | if (errno == EINTR) { | |
389 | TRACE("Caught EINTR; retrying"); | |
dbc9832d | 390 | goto again; |
ee8377bd | 391 | } |
dbc9832d | 392 | |
9dfa4041 CB |
393 | ERROR("%s - Failed to send message to client", |
394 | strerror(errno)); | |
dbc9832d CB |
395 | } |
396 | ||
397 | /* kick client from list */ | |
398 | close(client->clientfd); | |
399 | lxc_list_del(cur); | |
400 | free(cur->elem); | |
401 | free(cur); | |
402 | } | |
403 | process_unlock(); | |
404 | ||
5e5576a4 CB |
405 | return 0; |
406 | } | |
407 | ||
408 | static int lxc_serve_state_socket_pair(const char *name, | |
409 | struct lxc_handler *handler, | |
410 | lxc_state_t state) | |
411 | { | |
412 | ssize_t ret; | |
413 | ||
414 | if (!handler->backgrounded || | |
415 | handler->state_socket_pair[1] < 0 || | |
416 | state == STARTING) | |
417 | return 0; | |
418 | ||
419 | /* Close read end of the socket pair. */ | |
420 | close(handler->state_socket_pair[0]); | |
421 | handler->state_socket_pair[0] = -1; | |
422 | ||
ee8377bd | 423 | again: |
5e5576a4 CB |
424 | ret = lxc_abstract_unix_send_credential(handler->state_socket_pair[1], |
425 | &(int){state}, sizeof(int)); | |
ee8377bd CB |
426 | if (ret != sizeof(int)) { |
427 | if (errno == EINTR) | |
428 | goto again; | |
5e5576a4 CB |
429 | SYSERROR("Failed to send state to %d", |
430 | handler->state_socket_pair[1]); | |
ee8377bd CB |
431 | return -1; |
432 | } | |
5e5576a4 CB |
433 | |
434 | TRACE("Sent container state \"%s\" to %d", lxc_state2str(state), | |
435 | handler->state_socket_pair[1]); | |
436 | ||
437 | /* Close write end of the socket pair. */ | |
438 | close(handler->state_socket_pair[1]); | |
439 | handler->state_socket_pair[1] = -1; | |
440 | ||
441 | return 0; | |
442 | } | |
443 | ||
444 | int lxc_set_state(const char *name, struct lxc_handler *handler, | |
445 | lxc_state_t state) | |
446 | { | |
447 | int ret; | |
448 | ||
449 | ret = lxc_serve_state_socket_pair(name, handler, state); | |
450 | if (ret < 0) { | |
451 | ERROR("Failed to synchronize via anonymous pair of unix sockets"); | |
452 | return -1; | |
453 | } | |
454 | ||
455 | ret = lxc_serve_state_clients(name, handler, state); | |
456 | if (ret < 0) | |
457 | return -1; | |
458 | ||
dbc9832d CB |
459 | /* This function will try to connect to the legacy lxc-monitord state |
460 | * server and only exists for backwards compatibility. | |
461 | */ | |
9123e471 | 462 | lxc_monitor_send_state(name, state, handler->lxcpath); |
dbc9832d | 463 | |
66aeffc7 DL |
464 | return 0; |
465 | } | |
466 | ||
735f2c6e | 467 | int lxc_poll(const char *name, struct lxc_handler *handler) |
b0a33c1e | 468 | { |
3c319edb CB |
469 | int ret; |
470 | struct lxc_epoll_descr descr, descr_console; | |
ca5f7926 | 471 | int sigfd = handler->sigfd; |
b0a33c1e | 472 | |
3c319edb CB |
473 | ret = lxc_mainloop_open(&descr); |
474 | if (ret < 0) { | |
475 | ERROR("Failed to create mainloop"); | |
50c8bf05 | 476 | goto out_sigfd; |
b0a33c1e | 477 | } |
478 | ||
3c319edb CB |
479 | ret = lxc_mainloop_open(&descr_console); |
480 | if (ret < 0) { | |
481 | ERROR("Failed to create console mainloop"); | |
482 | goto out_mainloop; | |
483 | } | |
484 | ||
485 | ret = lxc_mainloop_add_handler(&descr, sigfd, signal_handler, handler); | |
486 | if (ret < 0) { | |
487 | ERROR("Failed to add signal handler for %d to mainloop", sigfd); | |
488 | goto out_mainloop_console; | |
b0a33c1e | 489 | } |
490 | ||
3c319edb CB |
491 | ret = lxc_console_mainloop_add(&descr, handler->conf); |
492 | if (ret < 0) { | |
493 | ERROR("Failed to add console handlers to mainloop"); | |
494 | goto out_mainloop_console; | |
63376d7d DL |
495 | } |
496 | ||
3c319edb CB |
497 | ret = lxc_console_mainloop_add(&descr_console, handler->conf); |
498 | if (ret < 0) { | |
499 | ERROR("Failed to add console handlers to console mainloop"); | |
500 | goto out_mainloop_console; | |
563f2f2c DL |
501 | } |
502 | ||
3c319edb CB |
503 | ret = lxc_cmd_mainloop_add(name, &descr, handler); |
504 | if (ret < 0) { | |
505 | ERROR("Failed to add command handler to mainloop"); | |
506 | goto out_mainloop_console; | |
507 | } | |
b0a33c1e | 508 | |
3c319edb | 509 | TRACE("Mainloop is ready"); |
b0a33c1e | 510 | |
3c319edb CB |
511 | ret = lxc_mainloop(&descr, -1); |
512 | if (ret < 0) | |
513 | return -1; | |
514 | if (handler->init_died) | |
515 | return lxc_mainloop(&descr_console, 0); | |
516 | return ret; | |
517 | ||
518 | out_mainloop: | |
b0a33c1e | 519 | lxc_mainloop_close(&descr); |
408da065 | 520 | |
3c319edb CB |
521 | out_mainloop_console: |
522 | lxc_mainloop_close(&descr_console); | |
523 | ||
b0a33c1e | 524 | out_sigfd: |
525 | close(sigfd); | |
408da065 | 526 | |
c3e13372 | 527 | return -1; |
b0a33c1e | 528 | } |
529 | ||
41784e4e CB |
530 | void lxc_zero_handler(struct lxc_handler *handler) |
531 | { | |
532 | int i; | |
533 | ||
534 | memset(handler, 0, sizeof(struct lxc_handler)); | |
535 | ||
536 | handler->clone_flags = -1; | |
537 | ||
538 | handler->pinfd = -1; | |
539 | ||
540 | handler->sigfd = -1; | |
541 | ||
542 | for (i = 0; i < LXC_NS_MAX; i++) | |
543 | handler->nsfd[i] = -1; | |
544 | ||
545 | handler->data_sock[0] = -1; | |
546 | handler->data_sock[1] = -1; | |
547 | ||
548 | handler->state_socket_pair[0] = -1; | |
549 | handler->state_socket_pair[1] = -1; | |
550 | ||
551 | handler->sync_sock[0] = -1; | |
552 | handler->sync_sock[1] = -1; | |
553 | } | |
554 | ||
f2e07cb6 CB |
555 | void lxc_free_handler(struct lxc_handler *handler) |
556 | { | |
557 | if (handler->conf && handler->conf->maincmd_fd) | |
558 | close(handler->conf->maincmd_fd); | |
559 | ||
5e5576a4 CB |
560 | if (handler->state_socket_pair[0] >= 0) |
561 | close(handler->state_socket_pair[0]); | |
562 | ||
563 | if (handler->state_socket_pair[1] >= 0) | |
564 | close(handler->state_socket_pair[1]); | |
565 | ||
f2e07cb6 CB |
566 | handler->conf = NULL; |
567 | free(handler); | |
568 | } | |
569 | ||
aa460476 | 570 | struct lxc_handler *lxc_init_handler(const char *name, struct lxc_conf *conf, |
5e5576a4 | 571 | const char *lxcpath, bool daemonize) |
59eb99ba | 572 | { |
5e5576a4 | 573 | int i, ret; |
3a0f472d DL |
574 | struct lxc_handler *handler; |
575 | ||
576 | handler = malloc(sizeof(*handler)); | |
aa460476 CB |
577 | if (!handler) { |
578 | ERROR("failed to allocate memory"); | |
3a0f472d | 579 | return NULL; |
aa460476 | 580 | } |
59eb99ba DL |
581 | |
582 | memset(handler, 0, sizeof(*handler)); | |
583 | ||
790255cf CB |
584 | /* Note that am_unpriv() checks the effective uid. We probably don't |
585 | * care if we are real root only if we are running as root so this | |
586 | * should be fine. | |
587 | */ | |
d0fbc7ba | 588 | handler->am_root = !am_unpriv(); |
c6012571 | 589 | handler->data_sock[0] = handler->data_sock[1] = -1; |
fae349da | 590 | handler->conf = conf; |
9123e471 | 591 | handler->lxcpath = lxcpath; |
5c068da9 | 592 | handler->pinfd = -1; |
5e5576a4 | 593 | handler->state_socket_pair[0] = handler->state_socket_pair[1] = -1; |
d39b10eb CB |
594 | if (handler->conf->reboot == 0) |
595 | lxc_list_init(&handler->conf->state_clients); | |
fae349da | 596 | |
b6b2b194 WB |
597 | for (i = 0; i < LXC_NS_MAX; i++) |
598 | handler->nsfd[i] = -1; | |
599 | ||
f0ecc19d | 600 | handler->name = name; |
3bdf52d7 | 601 | |
5e5576a4 CB |
602 | if (daemonize && !handler->conf->reboot) { |
603 | /* Create socketpair() to synchronize on daemonized startup. | |
604 | * When the container reboots we don't need to synchronize again | |
605 | * currently so don't open another socketpair(). | |
606 | */ | |
607 | ret = socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0, | |
608 | handler->state_socket_pair); | |
609 | if (ret < 0) { | |
610 | ERROR("Failed to create anonymous pair of unix sockets"); | |
611 | goto on_error; | |
612 | } | |
613 | TRACE("Created anonymous pair {%d,%d} of unix sockets", | |
614 | handler->state_socket_pair[0], | |
615 | handler->state_socket_pair[1]); | |
616 | } | |
617 | ||
d39b10eb CB |
618 | if (handler->conf->reboot == 0) { |
619 | handler->conf->maincmd_fd = lxc_cmd_init(name, lxcpath, "command"); | |
620 | if (handler->conf->maincmd_fd < 0) { | |
621 | ERROR("Failed to set up command socket"); | |
622 | goto on_error; | |
623 | } | |
aa460476 | 624 | } |
9dfa4041 | 625 | TRACE("Unix domain socket %d for command server is ready", |
aa460476 CB |
626 | handler->conf->maincmd_fd); |
627 | ||
628 | return handler; | |
629 | ||
f2e07cb6 CB |
630 | on_error: |
631 | lxc_free_handler(handler); | |
aa460476 CB |
632 | |
633 | return NULL; | |
634 | } | |
635 | ||
636 | int lxc_init(const char *name, struct lxc_handler *handler) | |
637 | { | |
a2c09be0 | 638 | int ret; |
4a03ded4 | 639 | const char *loglevel; |
aa460476 CB |
640 | struct lxc_conf *conf = handler->conf; |
641 | ||
642 | lsm_init(); | |
643 | TRACE("initialized LSM"); | |
d2e30e99 | 644 | |
8f2c3a70 | 645 | if (lxc_read_seccomp_config(conf) != 0) { |
408da065 | 646 | ERROR("Failed loading seccomp policy."); |
d2e30e99 | 647 | goto out_close_maincmd_fd; |
8f2c3a70 | 648 | } |
dbc9832d | 649 | TRACE("read seccomp policy"); |
8f2c3a70 | 650 | |
408da065 | 651 | /* Begin by setting the state to STARTING. */ |
25c2aca5 | 652 | if (lxc_set_state(name, handler, STARTING)) { |
408da065 | 653 | ERROR("Failed to set state for container \"%s\" to \"%s\".", name, lxc_state2str(STARTING)); |
051151de | 654 | goto out_close_maincmd_fd; |
0ad19a3f | 655 | } |
dbc9832d | 656 | TRACE("set container state to \"STARTING\""); |
0ad19a3f | 657 | |
408da065 CB |
658 | /* Start of environment variable setup for hooks. */ |
659 | if (name && setenv("LXC_NAME", name, 1)) | |
660 | SYSERROR("Failed to set environment variable: LXC_NAME=%s.", name); | |
661 | ||
662 | if (conf->rcfile && setenv("LXC_CONFIG_FILE", conf->rcfile, 1)) | |
663 | SYSERROR("Failed to set environment variable: LXC_CONFIG_FILE=%s.", conf->rcfile); | |
664 | ||
665 | if (conf->rootfs.mount && setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1)) | |
666 | SYSERROR("Failed to set environment variable: LXC_ROOTFS_MOUNT=%s.", conf->rootfs.mount); | |
667 | ||
668 | if (conf->rootfs.path && setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1)) | |
669 | SYSERROR("Failed to set environment variable: LXC_ROOTFS_PATH=%s.", conf->rootfs.path); | |
670 | ||
671 | if (conf->console.path && setenv("LXC_CONSOLE", conf->console.path, 1)) | |
672 | SYSERROR("Failed to set environment variable: LXC_CONSOLE=%s.", conf->console.path); | |
673 | ||
674 | if (conf->console.log_path && setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1)) | |
675 | SYSERROR("Failed to set environment variable: LXC_CONSOLE_LOGPATH=%s.", conf->console.log_path); | |
676 | ||
677 | if (setenv("LXC_CGNS_AWARE", "1", 1)) | |
678 | SYSERROR("Failed to set environment variable LXC_CGNS_AWARE=1."); | |
b8f88d9b | 679 | |
4a03ded4 CB |
680 | loglevel = lxc_log_priority_to_string(lxc_log_get_level()); |
681 | if (setenv("LXC_LOG_LEVEL", loglevel, 1)) | |
682 | SYSERROR("Failed to set environment variable LXC_LOG_LEVEL=%s", loglevel); | |
a2c09be0 CB |
683 | |
684 | if (conf->hooks_version == 0) | |
685 | ret = setenv("LXC_HOOK_VERSION", "0", 1); | |
686 | else | |
687 | ret = setenv("LXC_HOOK_VERSION", "1", 1); | |
688 | if (ret < 0) | |
689 | SYSERROR("Failed to set environment variable LXC_HOOK_VERSION=%u", conf->hooks_version); | |
408da065 | 690 | /* End of environment variable setup for hooks. */ |
f7bee6c6 | 691 | |
dbc9832d CB |
692 | TRACE("set environment variables"); |
693 | ||
14a7b0f9 | 694 | if (run_lxc_hooks(name, "pre-start", conf, NULL)) { |
408da065 | 695 | ERROR("Failed to run lxc.hook.pre-start for container \"%s\".", name); |
773fb9ca SH |
696 | goto out_aborting; |
697 | } | |
dbc9832d | 698 | TRACE("ran pre-start hooks"); |
26ddeedd | 699 | |
408da065 CB |
700 | /* The signal fd has to be created before forking otherwise if the child |
701 | * process exits before we setup the signal fd, the event will be lost | |
702 | * and the command will be stuck. | |
703 | */ | |
83ee7875 | 704 | handler->sigfd = setup_signal_fd(&handler->oldmask); |
59eb99ba | 705 | if (handler->sigfd < 0) { |
408da065 | 706 | ERROR("Failed to setup SIGCHLD fd handler."); |
b5159817 DE |
707 | goto out_delete_tty; |
708 | } | |
dbc9832d | 709 | TRACE("set up signal fd"); |
b5159817 | 710 | |
408da065 | 711 | /* Do this after setting up signals since it might unblock SIGWINCH. */ |
b5159817 | 712 | if (lxc_console_create(conf)) { |
408da065 | 713 | ERROR("Failed to create console for container \"%s\".", name); |
b5159817 | 714 | goto out_restore_sigmask; |
b0a33c1e | 715 | } |
dbc9832d | 716 | TRACE("created console"); |
b0a33c1e | 717 | |
54117de5 | 718 | if (lxc_ttys_shift_ids(conf) < 0) { |
408da065 | 719 | ERROR("Failed to shift tty into container."); |
c4d10a05 SH |
720 | goto out_restore_sigmask; |
721 | } | |
dbc9832d | 722 | TRACE("shifted tty ids"); |
c4d10a05 | 723 | |
dbc9832d | 724 | INFO("container \"%s\" is initialized", name); |
aa460476 | 725 | return 0; |
59eb99ba | 726 | |
b5159817 DE |
727 | out_restore_sigmask: |
728 | sigprocmask(SIG_SETMASK, &handler->oldmask, NULL); | |
59eb99ba | 729 | out_delete_tty: |
fae349da | 730 | lxc_delete_tty(&conf->tty_info); |
59eb99ba | 731 | out_aborting: |
25c2aca5 | 732 | lxc_set_state(name, handler, ABORTING); |
d2e30e99 DE |
733 | out_close_maincmd_fd: |
734 | close(conf->maincmd_fd); | |
735 | conf->maincmd_fd = -1; | |
aa460476 | 736 | return -1; |
59eb99ba DL |
737 | } |
738 | ||
3b72c4a0 | 739 | void lxc_fini(const char *name, struct lxc_handler *handler) |
59eb99ba | 740 | { |
b3286b62 | 741 | int i, rc; |
18b3b9c1 | 742 | pid_t self; |
dbc9832d | 743 | struct lxc_list *cur, *next; |
dbc9832d | 744 | char *namespaces[LXC_NS_MAX + 1]; |
b3286b62 | 745 | size_t namespace_count = 0; |
b6b2b194 | 746 | |
408da065 CB |
747 | /* The STOPPING state is there for future cleanup code which can take |
748 | * awhile. | |
59eb99ba | 749 | */ |
25c2aca5 | 750 | lxc_set_state(name, handler, STOPPING); |
b6b2b194 | 751 | |
0059379f | 752 | self = lxc_raw_getpid(); |
b3286b62 | 753 | for (i = 0; i < LXC_NS_MAX; i++) { |
18b3b9c1 CB |
754 | if (handler->nsfd[i] < 0) |
755 | continue; | |
756 | ||
757 | if (handler->conf->hooks_version == 0) | |
758 | rc = asprintf(&namespaces[namespace_count], | |
759 | "%s:/proc/%d/fd/%d", ns_info[i].proc_name, | |
760 | self, handler->nsfd[i]); | |
761 | else | |
762 | rc = asprintf(&namespaces[namespace_count], | |
763 | "/proc/%d/fd/%d", self, handler->nsfd[i]); | |
764 | if (rc == -1) { | |
765 | SYSERROR("Failed to allocate memory."); | |
766 | break; | |
b3286b62 | 767 | } |
18b3b9c1 CB |
768 | |
769 | if (handler->conf->hooks_version == 0) { | |
770 | namespace_count++; | |
771 | continue; | |
772 | } | |
773 | ||
774 | rc = setenv(ns_info[i].env_name, namespaces[namespace_count], 1); | |
775 | if (rc < 0) | |
776 | SYSERROR("Failed to set environment variable %s=%s", | |
777 | ns_info[i].env_name, namespaces[namespace_count]); | |
778 | else | |
779 | TRACE("Set environment variable %s=%s", | |
780 | ns_info[i].env_name, namespaces[namespace_count]); | |
781 | ||
782 | namespace_count++; | |
b3286b62 WB |
783 | } |
784 | namespaces[namespace_count] = NULL; | |
c154af98 | 785 | |
408da065 CB |
786 | if (handler->conf->reboot && setenv("LXC_TARGET", "reboot", 1)) |
787 | SYSERROR("Failed to set environment variable: LXC_TARGET=reboot."); | |
788 | ||
789 | if (!handler->conf->reboot && setenv("LXC_TARGET", "stop", 1)) | |
790 | SYSERROR("Failed to set environment variable: LXC_TARGET=stop."); | |
c154af98 | 791 | |
18b3b9c1 | 792 | if (handler->conf->hooks_version == 0) |
14a7b0f9 | 793 | rc = run_lxc_hooks(name, "stop", handler->conf, namespaces); |
18b3b9c1 | 794 | else |
14a7b0f9 | 795 | rc = run_lxc_hooks(name, "stop", handler->conf, NULL); |
c154af98 | 796 | |
b3286b62 WB |
797 | while (namespace_count--) |
798 | free(namespaces[namespace_count]); | |
28d9e29e | 799 | |
b6b2b194 | 800 | for (i = 0; i < LXC_NS_MAX; i++) { |
28d9e29e CB |
801 | if (handler->nsfd[i] < 0) |
802 | continue; | |
738d0deb | 803 | |
28d9e29e CB |
804 | close(handler->nsfd[i]); |
805 | handler->nsfd[i] = -1; | |
738d0deb CB |
806 | } |
807 | ||
4288b79f CB |
808 | cgroup_destroy(handler); |
809 | ||
d39b10eb | 810 | if (handler->conf->reboot == 0) { |
f893d898 CB |
811 | /* For all new state clients simply close the command socket. |
812 | * This will inform all state clients that the container is | |
813 | * STOPPED and also prevents a race between a open()/close() on | |
814 | * the command socket causing a new process to get ECONNREFUSED | |
815 | * because we haven't yet closed the command socket. | |
816 | */ | |
d39b10eb CB |
817 | close(handler->conf->maincmd_fd); |
818 | handler->conf->maincmd_fd = -1; | |
c3184275 CB |
819 | TRACE("Closed command socket"); |
820 | ||
821 | /* This function will try to connect to the legacy lxc-monitord | |
822 | * state server and only exists for backwards compatibility. | |
823 | */ | |
824 | lxc_monitor_send_state(name, STOPPED, handler->lxcpath); | |
825 | ||
826 | /* The command socket is closed so no one can acces the command | |
827 | * socket anymore so there's no need to lock it. | |
828 | */ | |
829 | handler->state = STOPPED; | |
830 | TRACE("Set container state to \"STOPPED\""); | |
f893d898 CB |
831 | } else { |
832 | lxc_set_state(name, handler, STOPPED); | |
d39b10eb | 833 | } |
4288b79f | 834 | |
14a7b0f9 | 835 | if (run_lxc_hooks(name, "post-stop", handler->conf, NULL)) { |
408da065 | 836 | ERROR("Failed to run lxc.hook.post-stop for container \"%s\".", name); |
f3787121 CB |
837 | if (handler->conf->reboot) { |
838 | WARN("Container will be stopped instead of rebooted."); | |
839 | handler->conf->reboot = 0; | |
408da065 CB |
840 | if (setenv("LXC_TARGET", "stop", 1)) |
841 | WARN("Failed to set environment variable: LXC_TARGET=stop."); | |
f3787121 CB |
842 | } |
843 | } | |
26ddeedd | 844 | |
408da065 | 845 | /* Reset mask set by setup_signal_fd. */ |
8f64a3f6 | 846 | if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) |
408da065 | 847 | WARN("Failed to restore signal mask."); |
8f64a3f6 | 848 | |
b5159817 | 849 | lxc_console_delete(&handler->conf->console); |
b2431939 | 850 | lxc_delete_tty(&handler->conf->tty_info); |
dbc9832d | 851 | |
dbc9832d CB |
852 | /* The command socket is now closed, no more state clients can register |
853 | * themselves from now on. So free the list of state clients. | |
854 | */ | |
d39b10eb CB |
855 | lxc_list_for_each_safe(cur, &handler->conf->state_clients, next) { |
856 | struct lxc_state_client *client = cur->elem; | |
857 | ||
858 | /* Keep state clients that want to be notified about reboots. */ | |
859 | if ((handler->conf->reboot > 0) && (client->states[RUNNING] == 2)) | |
860 | continue; | |
861 | ||
dbc9832d CB |
862 | /* close state client socket */ |
863 | close(client->clientfd); | |
864 | lxc_list_del(cur); | |
865 | free(cur->elem); | |
866 | free(cur); | |
867 | } | |
868 | ||
c6012571 CB |
869 | if (handler->data_sock[0] != -1) { |
870 | close(handler->data_sock[0]); | |
871 | close(handler->data_sock[1]); | |
e8bd4e43 | 872 | } |
2c5f2ede CB |
873 | |
874 | if (handler->conf->ephemeral == 1 && handler->conf->reboot != 1) | |
28272964 | 875 | lxc_destroy_container_on_signal(handler, name); |
2c5f2ede | 876 | |
b2431939 | 877 | free(handler); |
59eb99ba DL |
878 | } |
879 | ||
735f2c6e | 880 | void lxc_abort(const char *name, struct lxc_handler *handler) |
59eb99ba | 881 | { |
73e608b2 SH |
882 | int ret, status; |
883 | ||
25c2aca5 | 884 | lxc_set_state(name, handler, ABORTING); |
7d9fb3e9 DL |
885 | if (handler->pid > 0) |
886 | kill(handler->pid, SIGKILL); | |
408da065 CB |
887 | while ((ret = waitpid(-1, &status, 0)) > 0) { |
888 | ; | |
889 | } | |
59eb99ba DL |
890 | } |
891 | ||
18225d19 CB |
892 | static int lxc_set_death_signal(int signal) |
893 | { | |
894 | int ret; | |
895 | pid_t ppid; | |
896 | ||
897 | ret = prctl(PR_SET_PDEATHSIG, signal, 0, 0, 0); | |
898 | ||
899 | /* Check whether we have been orphaned. */ | |
900 | ppid = (pid_t)syscall(SYS_getppid); | |
901 | if (ppid == 1) { | |
902 | pid_t self; | |
903 | ||
904 | self = lxc_raw_getpid(); | |
905 | ret = kill(self, SIGKILL); | |
906 | if (ret < 0) | |
907 | return -1; | |
908 | } | |
909 | ||
910 | if (ret < 0) { | |
911 | SYSERROR("Failed to set PR_SET_PDEATHSIG to %d", signal); | |
912 | return -1; | |
913 | } | |
914 | ||
915 | return 0; | |
916 | } | |
917 | ||
ffe1e01a | 918 | static int do_start(void *data) |
50e98013 | 919 | { |
8bf3abfb | 920 | int ret; |
7c661726 | 921 | struct lxc_list *iterator; |
7a55c157 | 922 | char path[PATH_MAX]; |
fa3a5b22 | 923 | struct lxc_handler *handler = data; |
928b1f04 YT |
924 | bool have_cap_setgid; |
925 | uid_t new_uid; | |
926 | gid_t new_gid; | |
5af9369b | 927 | int devnull_fd = -1; |
50e98013 | 928 | |
18225d19 | 929 | lxc_sync_fini_parent(handler); |
50e98013 | 930 | |
408da065 CB |
931 | /* This prctl must be before the synchro, so if the parent dies before |
932 | * we set the parent death signal, we will detect its death with the | |
933 | * synchro right after, otherwise we have a window where the parent can | |
934 | * exit before we set the pdeath signal leading to a unsupervized | |
935 | * container. | |
743ecd2e | 936 | */ |
18225d19 | 937 | ret = lxc_set_death_signal(SIGKILL); |
912314fc CB |
938 | if (ret < 0) { |
939 | SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL"); | |
18225d19 | 940 | goto out_warn_father; |
743ecd2e DL |
941 | } |
942 | ||
18225d19 CB |
943 | ret = sigprocmask(SIG_SETMASK, &handler->oldmask, NULL); |
944 | if (ret < 0) { | |
945 | SYSERROR("Failed to set signal mask"); | |
946 | goto out_warn_father; | |
947 | } | |
50e98013 | 948 | |
408da065 | 949 | /* Don't leak the pinfd to the container. */ |
d39b10eb | 950 | if (handler->pinfd >= 0) |
0d03360a | 951 | close(handler->pinfd); |
2b0e17e4 | 952 | |
18225d19 CB |
953 | ret = lxc_sync_wait_parent(handler, LXC_SYNC_STARTUP); |
954 | if (ret < 0) | |
955 | goto out_warn_father; | |
5b1e83cb | 956 | |
408da065 CB |
957 | /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See |
958 | * https://github.com/lxc/lxd/issues/1978. | |
959 | */ | |
5b1e83cb | 960 | if ((handler->clone_flags & (CLONE_NEWNET | CLONE_NEWUSER)) == |
408da065 | 961 | (CLONE_NEWNET | CLONE_NEWUSER)) { |
5b1e83cb SH |
962 | ret = unshare(CLONE_NEWNET); |
963 | if (ret < 0) { | |
408da065 | 964 | SYSERROR("Failed to unshare CLONE_NEWNET."); |
5b1e83cb SH |
965 | goto out_warn_father; |
966 | } | |
408da065 | 967 | INFO("Unshared CLONE_NEWNET."); |
5b1e83cb SH |
968 | } |
969 | ||
408da065 CB |
970 | /* Tell the parent task it can begin to configure the container and wait |
971 | * for it to finish. | |
3c22086f CLG |
972 | */ |
973 | if (lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE)) | |
9d7f9e52 | 974 | return -1; |
50e98013 | 975 | |
7ab1ba02 CB |
976 | if (lxc_network_recv_veth_names_from_parent(handler) < 0) { |
977 | ERROR("Failed to receive veth names from parent"); | |
658979c5 | 978 | goto out_warn_father; |
7ab1ba02 | 979 | } |
658979c5 | 980 | |
408da065 | 981 | /* If we are in a new user namespace, become root there to have |
d08f8d2f | 982 | * privilege over our namespace. |
f6d3e3e4 SH |
983 | */ |
984 | if (!lxc_list_empty(&handler->conf->id_map)) { | |
4b826b1f CB |
985 | ret = lxc_switch_uid_gid(0, 0); |
986 | if (ret < 0) | |
f6d3e3e4 | 987 | goto out_warn_father; |
d08f8d2f CB |
988 | |
989 | /* Drop groups only after we switched to a valid gid in the new | |
990 | * user namespace. | |
991 | */ | |
4b826b1f CB |
992 | ret = lxc_setgroups(0, NULL); |
993 | if (ret < 0) | |
c476bdce | 994 | goto out_warn_father; |
4b826b1f CB |
995 | |
996 | if (!handler->am_root) { | |
997 | ret = prctl(PR_SET_DUMPABLE, 1, 0, 0, 0); | |
998 | if (ret < 0) | |
999 | goto out_warn_father; | |
1000 | } | |
912314fc CB |
1001 | |
1002 | /* set{g,u}id() clears deathsignal */ | |
18225d19 | 1003 | ret = lxc_set_death_signal(SIGKILL); |
912314fc CB |
1004 | if (ret < 0) { |
1005 | SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL"); | |
1006 | goto out_warn_father; | |
1007 | } | |
f6d3e3e4 SH |
1008 | } |
1009 | ||
99b71824 | 1010 | if (access(handler->lxcpath, X_OK)) { |
c8154066 SH |
1011 | print_top_failing_dir(handler->lxcpath); |
1012 | goto out_warn_father; | |
1013 | } | |
1014 | ||
a91cde21 | 1015 | ret = snprintf(path, sizeof(path), "%s/dev/null", handler->conf->rootfs.mount); |
408da065 | 1016 | if (ret < 0 || ret >= sizeof(path)) |
7a55c157 | 1017 | goto out_warn_father; |
7a55c157 TA |
1018 | |
1019 | /* In order to checkpoint restore, we need to have everything in the | |
1020 | * same mount namespace. However, some containers may not have a | |
1021 | * reasonable /dev (in particular, they may not have /dev/null), so we | |
1022 | * can't set init's std fds to /dev/null by opening it from inside the | |
1023 | * container. | |
1024 | * | |
1025 | * If that's the case, fall back to using the host's /dev/null. This | |
1026 | * means that migration won't work, but at least we won't spew output | |
1027 | * where it isn't wanted. | |
1028 | */ | |
1029 | if (handler->backgrounded && !handler->conf->autodev && access(path, F_OK) < 0) { | |
c44de748 AM |
1030 | devnull_fd = open_devnull(); |
1031 | ||
1032 | if (devnull_fd < 0) | |
1033 | goto out_warn_father; | |
408da065 CB |
1034 | WARN("Using /dev/null from the host for container init's " |
1035 | "standard file descriptors. Migration will not work."); | |
c44de748 AM |
1036 | } |
1037 | ||
408da065 | 1038 | /* Ask father to setup cgroups and wait for him to finish. */ |
544a48a0 | 1039 | if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP)) |
c44de748 | 1040 | goto out_error; |
544a48a0 | 1041 | |
deefdf8a CB |
1042 | /* Unshare cgroup namespace after we have setup our cgroups. If we do it |
1043 | * earlier we end up with a wrong view of /proc/self/cgroup. For | |
1044 | * example, assume we unshare(CLONE_NEWCGROUP) first, and then create | |
1045 | * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then | |
1046 | * /proc/self/cgroup would show us: | |
1047 | * | |
1048 | * 8:cpuset:/lxc/c | |
1049 | * | |
1050 | * whereas it should actually show | |
1051 | * | |
1052 | * 8:cpuset:/ | |
1053 | */ | |
5af9369b | 1054 | if (handler->clone_flags & CLONE_NEWCGROUP) { |
deefdf8a CB |
1055 | if (unshare(CLONE_NEWCGROUP) < 0) { |
1056 | INFO("Failed to unshare CLONE_NEWCGROUP."); | |
1057 | goto out_warn_father; | |
1058 | } | |
1059 | INFO("Unshared CLONE_NEWCGROUP."); | |
1060 | } | |
1061 | ||
149857af CB |
1062 | /* Add the requested environment variables to the current environment to |
1063 | * allow them to be used by the various hooks, such as the start hook | |
1064 | * above. | |
98ff08ca | 1065 | */ |
98ff08ca CB |
1066 | lxc_list_for_each(iterator, &handler->conf->environment) { |
1067 | if (putenv((char *)iterator->elem)) { | |
1068 | SYSERROR("Failed to set environment variable: %s.", (char *)iterator->elem); | |
1069 | goto out_warn_father; | |
1070 | } | |
1071 | } | |
1072 | ||
f4152036 | 1073 | /* Setup the container, ip, names, utsname, ... */ |
3b988b33 | 1074 | ret = lxc_setup(handler); |
7729f8e5 | 1075 | close(handler->data_sock[1]); |
8bf3abfb | 1076 | close(handler->data_sock[0]); |
7729f8e5 | 1077 | if (ret < 0) { |
f4152036 CB |
1078 | ERROR("Failed to setup container \"%s\".", handler->name); |
1079 | goto out_warn_father; | |
1080 | } | |
1081 | ||
408da065 | 1082 | /* Set the label to change to when we exec(2) the container's init. */ |
7aff4f43 | 1083 | if (lsm_process_label_set(NULL, handler->conf, 1, 1) < 0) |
e075f5d9 | 1084 | goto out_warn_father; |
5112cd70 | 1085 | |
029cdff5 | 1086 | /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it |
408da065 CB |
1087 | * before we aren't allowed anymore. |
1088 | */ | |
029cdff5 CB |
1089 | if (handler->conf->no_new_privs) { |
1090 | if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) { | |
1091 | SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges."); | |
1092 | goto out_warn_father; | |
1093 | } | |
1094 | DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges."); | |
1095 | } | |
1096 | ||
0d9acb99 DE |
1097 | /* Some init's such as busybox will set sane tty settings on stdin, |
1098 | * stdout, stderr which it thinks is the console. We already set them | |
1099 | * the way we wanted on the real terminal, and we want init to do its | |
408da065 CB |
1100 | * setup on its console ie. the pty allocated in lxc_console_create() so |
1101 | * make sure that that pty is stdin,stdout,stderr. | |
0d9acb99 | 1102 | */ |
5d113f65 CB |
1103 | if (handler->conf->console.slave >= 0) { |
1104 | if (handler->backgrounded || handler->conf->is_execute == 0) | |
1105 | ret = set_stdfds(handler->conf->console.slave); | |
1106 | else | |
1107 | ret = lxc_console_set_stdfds(handler->conf->console.slave); | |
1108 | if (ret < 0) { | |
c5b93afb LF |
1109 | ERROR("Failed to redirect std{in,out,err} to pty file " |
1110 | "descriptor %d", | |
1111 | handler->conf->console.slave); | |
1112 | goto out_warn_father; | |
1113 | } | |
5d113f65 | 1114 | } |
0d9acb99 | 1115 | |
408da065 | 1116 | /* If we mounted a temporary proc, then unmount it now. */ |
5112cd70 | 1117 | tmp_proc_unmount(handler->conf); |
e075f5d9 | 1118 | |
8f2c3a70 SH |
1119 | if (lxc_seccomp_load(handler->conf) != 0) |
1120 | goto out_warn_father; | |
1121 | ||
14a7b0f9 | 1122 | if (run_lxc_hooks(handler->name, "start", handler->conf, NULL)) { |
408da065 | 1123 | ERROR("Failed to run lxc.hook.start for container \"%s\".", handler->name); |
773fb9ca SH |
1124 | goto out_warn_father; |
1125 | } | |
fc25b815 | 1126 | |
773fb9ca | 1127 | close(handler->sigfd); |
26ddeedd | 1128 | |
7a55c157 TA |
1129 | if (devnull_fd < 0) { |
1130 | devnull_fd = open_devnull(); | |
1131 | ||
1132 | if (devnull_fd < 0) | |
1133 | goto out_warn_father; | |
1134 | } | |
1135 | ||
c5b93afb LF |
1136 | if (handler->conf->console.slave < 0 && handler->backgrounded) |
1137 | if (set_stdfds(devnull_fd) < 0) { | |
1138 | ERROR("Failed to redirect std{in,out,err} to " | |
1139 | "\"/dev/null\""); | |
1140 | goto out_warn_father; | |
1141 | } | |
507cee36 | 1142 | |
c44de748 AM |
1143 | if (devnull_fd >= 0) { |
1144 | close(devnull_fd); | |
1145 | devnull_fd = -1; | |
1146 | } | |
1147 | ||
8c9a7665 TA |
1148 | setsid(); |
1149 | ||
3c491553 L |
1150 | if (handler->conf->init_cwd && chdir(handler->conf->init_cwd)) { |
1151 | SYSERROR("Could not change directory to \"%s\"", handler->conf->init_cwd); | |
1152 | goto out_warn_father; | |
1153 | } | |
1154 | ||
f4152036 CB |
1155 | if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP_LIMITS)) |
1156 | goto out_warn_father; | |
1157 | ||
149857af CB |
1158 | /* Reset the environment variables the user requested in a clear |
1159 | * environment. | |
1160 | */ | |
1161 | if (clearenv()) { | |
1162 | SYSERROR("Failed to clear environment."); | |
1163 | /* Don't error out though. */ | |
1164 | } | |
1165 | ||
1166 | lxc_list_for_each(iterator, &handler->conf->environment) { | |
1167 | if (putenv((char *)iterator->elem)) { | |
1168 | SYSERROR("Failed to set environment variable: %s.", (char *)iterator->elem); | |
1169 | goto out_warn_father; | |
1170 | } | |
1171 | } | |
1172 | ||
98ff08ca CB |
1173 | if (putenv("container=lxc")) { |
1174 | SYSERROR("Failed to set environment variable: container=lxc."); | |
1175 | goto out_warn_father; | |
1176 | } | |
1177 | ||
1178 | if (handler->conf->pty_names) { | |
1179 | if (putenv(handler->conf->pty_names)) { | |
1180 | SYSERROR("Failed to set environment variable for container ptys."); | |
1181 | goto out_warn_father; | |
1182 | } | |
1183 | } | |
1184 | ||
76bdf299 CB |
1185 | /* The container has been setup. We can now switch to an unprivileged |
1186 | * uid/gid. | |
1187 | */ | |
928b1f04 YT |
1188 | new_uid = handler->conf->init_uid; |
1189 | new_gid = handler->conf->init_gid; | |
76bdf299 | 1190 | |
928b1f04 YT |
1191 | /* If we are in a new user namespace we already dropped all |
1192 | * groups when we switched to root in the new user namespace | |
1193 | * further above. Only drop groups if we can, so ensure that we | |
1194 | * have necessary privilege. | |
1195 | */ | |
1196 | #if HAVE_LIBCAP | |
1197 | have_cap_setgid = lxc_proc_cap_is_set(CAP_SETGID, CAP_EFFECTIVE); | |
1198 | #else | |
1199 | have_cap_setgid = false; | |
1200 | #endif | |
1201 | if (lxc_list_empty(&handler->conf->id_map) && have_cap_setgid) { | |
1202 | if (lxc_setgroups(0, NULL) < 0) | |
76bdf299 CB |
1203 | goto out_warn_father; |
1204 | } | |
1205 | ||
928b1f04 YT |
1206 | if (lxc_switch_uid_gid(new_uid, new_gid) < 0) |
1207 | goto out_warn_father; | |
1208 | ||
408da065 CB |
1209 | /* After this call, we are in error because this ops should not return |
1210 | * as it execs. | |
1211 | */ | |
c4ea60df | 1212 | handler->ops->start(handler, handler->data); |
50e98013 DL |
1213 | |
1214 | out_warn_father: | |
408da065 CB |
1215 | /* We want the parent to know something went wrong, so we return a |
1216 | * special error code. | |
1217 | */ | |
d1ccb562 | 1218 | lxc_sync_wake_parent(handler, LXC_SYNC_ERROR); |
c44de748 AM |
1219 | |
1220 | out_error: | |
1221 | if (devnull_fd >= 0) | |
1222 | close(devnull_fd); | |
1223 | ||
50e98013 DL |
1224 | return -1; |
1225 | } | |
1226 | ||
ae467c54 | 1227 | static int lxc_recv_ttys_from_child(struct lxc_handler *handler) |
e8bd4e43 | 1228 | { |
ae467c54 | 1229 | int i; |
ae467c54 CB |
1230 | struct lxc_pty_info *pty_info; |
1231 | int ret = -1; | |
c6012571 | 1232 | int sock = handler->data_sock[1]; |
e8bd4e43 | 1233 | struct lxc_conf *conf = handler->conf; |
e8bd4e43 SH |
1234 | struct lxc_tty_info *tty_info = &conf->tty_info; |
1235 | ||
1236 | if (!conf->tty) | |
1237 | return 0; | |
1238 | ||
408da065 CB |
1239 | tty_info->pty_info = malloc(sizeof(*tty_info->pty_info) * conf->tty); |
1240 | if (!tty_info->pty_info) | |
e8bd4e43 | 1241 | return -1; |
e8bd4e43 | 1242 | |
672c1e58 CB |
1243 | for (i = 0; i < conf->tty; i++) { |
1244 | int ttyfds[2]; | |
1245 | ||
1246 | ret = lxc_abstract_unix_recv_fds(sock, ttyfds, 2, NULL, 0); | |
1247 | if (ret < 0) | |
1248 | break; | |
ae467c54 | 1249 | |
672c1e58 | 1250 | pty_info = &tty_info->pty_info[i]; |
e8bd4e43 | 1251 | pty_info->busy = 0; |
672c1e58 CB |
1252 | pty_info->master = ttyfds[0]; |
1253 | pty_info->slave = ttyfds[1]; | |
1254 | TRACE("Received pty with master fd %d and slave fd %d from " | |
ae467c54 | 1255 | "parent", pty_info->master, pty_info->slave); |
e8bd4e43 | 1256 | } |
ae467c54 | 1257 | if (ret < 0) |
672c1e58 | 1258 | ERROR("Failed to receive %d ttys from child: %s", conf->tty, |
ae467c54 CB |
1259 | strerror(errno)); |
1260 | else | |
672c1e58 CB |
1261 | TRACE("Received %d ttys from child", conf->tty); |
1262 | ||
1263 | tty_info->nbtty = conf->tty; | |
ae467c54 CB |
1264 | |
1265 | return ret; | |
e8bd4e43 SH |
1266 | } |
1267 | ||
5af9369b | 1268 | int resolve_clone_flags(struct lxc_handler *handler) |
f813849c | 1269 | { |
03df7ab5 | 1270 | handler->clone_flags = CLONE_NEWNS; |
f813849c | 1271 | |
fa3a5b22 CB |
1272 | if (!handler->conf->inherit_ns[LXC_NS_USER]) { |
1273 | if (!lxc_list_empty(&handler->conf->id_map)) | |
1274 | handler->clone_flags |= CLONE_NEWUSER; | |
1275 | } else { | |
1276 | INFO("Inheriting user namespace"); | |
1277 | } | |
f813849c | 1278 | |
28d9e29e | 1279 | if (!handler->conf->inherit_ns[LXC_NS_NET]) { |
408da065 | 1280 | if (!lxc_requests_empty_network(handler)) |
f813849c TA |
1281 | handler->clone_flags |= CLONE_NEWNET; |
1282 | } else { | |
28d9e29e | 1283 | INFO("Inheriting net namespace"); |
f813849c TA |
1284 | } |
1285 | ||
28d9e29e | 1286 | if (!handler->conf->inherit_ns[LXC_NS_IPC]) |
f813849c | 1287 | handler->clone_flags |= CLONE_NEWIPC; |
408da065 | 1288 | else |
28d9e29e | 1289 | INFO("Inheriting ipc namespace"); |
f813849c | 1290 | |
28d9e29e | 1291 | if (!handler->conf->inherit_ns[LXC_NS_UTS]) |
f813849c | 1292 | handler->clone_flags |= CLONE_NEWUTS; |
408da065 | 1293 | else |
28d9e29e | 1294 | INFO("Inheriting uts namespace"); |
03df7ab5 | 1295 | |
28d9e29e | 1296 | if (!handler->conf->inherit_ns[LXC_NS_PID]) |
03df7ab5 L |
1297 | handler->clone_flags |= CLONE_NEWPID; |
1298 | else | |
28d9e29e | 1299 | INFO("Inheriting pid namespace"); |
5af9369b CB |
1300 | |
1301 | if (cgns_supported()) { | |
1302 | if (!handler->conf->inherit_ns[LXC_NS_CGROUP]) | |
1303 | handler->clone_flags |= CLONE_NEWCGROUP; | |
1304 | else | |
1305 | INFO("Inheriting cgroup namespace"); | |
1306 | } else if (handler->conf->inherit_ns[LXC_NS_CGROUP]) { | |
1307 | return -EINVAL; | |
1308 | } | |
1309 | ||
1310 | return 0; | |
f813849c TA |
1311 | } |
1312 | ||
8deca6c9 CB |
1313 | /* Note that this function is used with clone(CLONE_VM). Some glibc versions |
1314 | * used to reset the pid/tid to -1 when CLONE_VM was used without CLONE_THREAD. | |
1315 | * But since the memory between parent and child is shared on CLONE_VM this | |
1316 | * would invalidate the getpid() cache that glibc used to maintain and so | |
1317 | * getpid() in the child would return the parent's pid. This is all fixed in | |
1318 | * newer glibc versions where the getpid() cache is removed and the pid/tid is | |
1319 | * not reset anymore. | |
1320 | * However, if for whatever reason you - dear commiter - somehow need to get the | |
1321 | * pid of the dummy intermediate process for do_share_ns() you need to call | |
0059379f CB |
1322 | * lxc_raw_getpid(). The next lxc_raw_clone() call does not employ CLONE_VM and |
1323 | * will be fine. | |
8deca6c9 CB |
1324 | */ |
1325 | static inline int do_share_ns(void *arg) | |
fa3a5b22 | 1326 | { |
8deca6c9 | 1327 | int i, flags, ret; |
fa3a5b22 CB |
1328 | struct lxc_handler *handler = arg; |
1329 | ||
8deca6c9 CB |
1330 | for (i = 0; i < LXC_NS_MAX; i++) { |
1331 | if (handler->nsfd[i] < 0) | |
1332 | continue; | |
fa3a5b22 | 1333 | |
8deca6c9 | 1334 | ret = setns(handler->nsfd[i], 0); |
fa3a5b22 | 1335 | if (ret < 0) |
8deca6c9 | 1336 | return -1; |
fa3a5b22 | 1337 | |
8deca6c9 | 1338 | DEBUG("Inherited %s namespace", ns_info[i].proc_name); |
fa3a5b22 CB |
1339 | } |
1340 | ||
8deca6c9 CB |
1341 | flags = handler->on_clone_flags; |
1342 | flags |= CLONE_PARENT; | |
ac2ba696 | 1343 | handler->pid = lxc_raw_clone_cb(do_start, handler, flags); |
8deca6c9 | 1344 | if (handler->pid < 0) |
fa3a5b22 CB |
1345 | return -1; |
1346 | ||
8deca6c9 | 1347 | return 0; |
fa3a5b22 CB |
1348 | } |
1349 | ||
480588e6 CB |
1350 | /* lxc_spawn() performs crucial setup tasks and clone()s the new process which |
1351 | * exec()s the requested container binary. | |
1352 | * Note that lxc_spawn() runs in the parent namespaces. Any operations performed | |
1353 | * right here should be double checked if they'd pose a security risk. (For | |
1354 | * example, any {u}mount() operations performed here will be reflected on the | |
1355 | * host!) | |
1356 | */ | |
74a3920a | 1357 | static int lxc_spawn(struct lxc_handler *handler) |
59eb99ba | 1358 | { |
8deca6c9 | 1359 | int i, ret; |
08dd2805 | 1360 | char pidstr[20]; |
57927bf2 CB |
1361 | bool wants_to_map_ids; |
1362 | struct lxc_list *id_map; | |
6e5fc7a5 | 1363 | const char *name = handler->name; |
28d9e29e | 1364 | const char *lxcpath = handler->lxcpath; |
8deca6c9 | 1365 | bool cgroups_connected = false, share_ns = false; |
28d9e29e | 1366 | struct lxc_conf *conf = handler->conf; |
9f30a190 | 1367 | |
28d9e29e | 1368 | id_map = &conf->id_map; |
57927bf2 | 1369 | wants_to_map_ids = !lxc_list_empty(id_map); |
2f2623ec | 1370 | |
28d9e29e CB |
1371 | for (i = 0; i < LXC_NS_MAX; i++) { |
1372 | if (!conf->inherit_ns[i]) | |
1373 | continue; | |
1374 | ||
28d9e29e CB |
1375 | handler->nsfd[i] = lxc_inherit_namespace(conf->inherit_ns[i], lxcpath, ns_info[i].proc_name); |
1376 | if (handler->nsfd[i] < 0) | |
1377 | return -1; | |
fa3a5b22 | 1378 | |
8deca6c9 | 1379 | share_ns = true; |
28d9e29e | 1380 | } |
50e98013 | 1381 | |
3c22086f | 1382 | if (lxc_sync_init(handler)) |
9d7f9e52 | 1383 | return -1; |
0ad19a3f | 1384 | |
b9f522c5 | 1385 | ret = socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0, |
790255cf | 1386 | handler->data_sock); |
c6012571 | 1387 | if (ret < 0) { |
e8bd4e43 SH |
1388 | lxc_sync_fini(handler); |
1389 | return -1; | |
1390 | } | |
1391 | ||
5af9369b CB |
1392 | ret = resolve_clone_flags(handler); |
1393 | if (ret < 0) { | |
1394 | lxc_sync_fini(handler); | |
1395 | return -1; | |
1396 | } | |
9f30a190 | 1397 | |
f813849c | 1398 | if (handler->clone_flags & CLONE_NEWNET) { |
28d9e29e | 1399 | if (!lxc_list_empty(&conf->network)) { |
26b797f3 | 1400 | |
9f30a190 MM |
1401 | /* Find gateway addresses from the link device, which is |
1402 | * no longer accessible inside the container. Do this | |
1403 | * before creating network interfaces, since goto | |
408da065 CB |
1404 | * out_delete_net does not work before lxc_clone. |
1405 | */ | |
9f30a190 | 1406 | if (lxc_find_gateway_addresses(handler)) { |
408da065 | 1407 | ERROR("Failed to find gateway addresses."); |
9f30a190 MM |
1408 | lxc_sync_fini(handler); |
1409 | return -1; | |
1410 | } | |
1411 | ||
408da065 CB |
1412 | /* That should be done before the clone because we will |
1413 | * fill the netdev index and use them in the child. | |
9f30a190 | 1414 | */ |
811ef482 | 1415 | if (lxc_create_network_priv(handler)) { |
408da065 | 1416 | ERROR("Failed to create the network."); |
9f30a190 MM |
1417 | lxc_sync_fini(handler); |
1418 | return -1; | |
1419 | } | |
19a26f82 | 1420 | } |
82d5ae15 DL |
1421 | } |
1422 | ||
d4ef7c50 | 1423 | if (!cgroup_init(handler)) { |
408da065 | 1424 | ERROR("Failed initializing cgroup support."); |
33ad9f1a CS |
1425 | goto out_delete_net; |
1426 | } | |
1427 | ||
7e4dfe0b SH |
1428 | cgroups_connected = true; |
1429 | ||
d4ef7c50 | 1430 | if (!cgroup_create(handler)) { |
408da065 | 1431 | ERROR("Failed creating cgroups."); |
47d8fb3b CS |
1432 | goto out_delete_net; |
1433 | } | |
1434 | ||
408da065 CB |
1435 | /* If the rootfs is not a blockdev, prevent the container from marking |
1436 | * it readonly. | |
1437 | * If the container is unprivileged then skip rootfs pinning. | |
0c547523 | 1438 | */ |
0ee35059 | 1439 | if (!wants_to_map_ids) { |
28d9e29e | 1440 | handler->pinfd = pin_rootfs(conf->rootfs.path); |
5e32a990 | 1441 | if (handler->pinfd == -1) |
408da065 | 1442 | INFO("Failed to pin the rootfs for container \"%s\".", handler->name); |
5e32a990 | 1443 | } |
0c547523 | 1444 | |
408da065 | 1445 | /* Create a process in a new set of namespaces. */ |
8deca6c9 | 1446 | handler->on_clone_flags = handler->clone_flags; |
408da065 CB |
1447 | if (handler->clone_flags & CLONE_NEWUSER) { |
1448 | /* If CLONE_NEWUSER and CLONE_NEWNET was requested, we need to | |
1449 | * clone a new user namespace first and only later unshare our | |
1450 | * network namespace to ensure that network devices ownership is | |
1451 | * set up correctly. | |
1452 | */ | |
8deca6c9 | 1453 | handler->on_clone_flags &= ~CLONE_NEWNET; |
408da065 | 1454 | } |
5af9369b CB |
1455 | /* The cgroup namespace gets unshare()ed not clone()ed. */ |
1456 | handler->on_clone_flags &= ~CLONE_NEWCGROUP; | |
732375f5 | 1457 | |
4e232466 CB |
1458 | if (share_ns) { |
1459 | pid_t attacher_pid; | |
1460 | ||
1461 | attacher_pid = lxc_clone(do_share_ns, handler, CLONE_VFORK | CLONE_VM | CLONE_FILES); | |
1462 | if (attacher_pid < 0) { | |
1463 | SYSERROR(LXC_CLONE_ERROR); | |
1464 | goto out_delete_net; | |
1465 | } | |
1466 | ||
1467 | ret = wait_for_pid(attacher_pid); | |
1468 | if (ret < 0) { | |
1469 | SYSERROR("Intermediate process failed"); | |
1470 | goto out_delete_net; | |
1471 | } | |
1472 | } else { | |
ac2ba696 | 1473 | handler->pid = lxc_raw_clone_cb(do_start, handler, handler->on_clone_flags); |
4e232466 CB |
1474 | } |
1475 | if (handler->pid < 0) { | |
1476 | SYSERROR(LXC_CLONE_ERROR); | |
7fef7a06 | 1477 | goto out_delete_net; |
0ad19a3f | 1478 | } |
fa3a5b22 | 1479 | TRACE("Cloned child process %d", handler->pid); |
1bd8d726 | 1480 | |
9662e444 | 1481 | for (i = 0; i < LXC_NS_MAX; i++) |
8deca6c9 | 1482 | if (handler->on_clone_flags & ns_info[i].clone_flag) |
28d9e29e | 1483 | INFO("Cloned %s", ns_info[i].flag_name); |
0ad19a3f | 1484 | |
8bf3abfb | 1485 | if (!preserve_ns(handler->nsfd, handler->on_clone_flags, handler->pid)) { |
28d9e29e CB |
1486 | ERROR("Failed to preserve cloned namespaces for lxc.hook.stop"); |
1487 | goto out_delete_net; | |
1488 | } | |
b6b2b194 | 1489 | |
3c22086f CLG |
1490 | lxc_sync_fini_child(handler); |
1491 | ||
408da065 CB |
1492 | /* Map the container uids. The container became an invalid userid the |
1493 | * moment it was cloned with CLONE_NEWUSER. This call doesn't change | |
1494 | * anything immediately, but allows the container to setuid(0) (0 being | |
1495 | * mapped to something else on the host.) later to become a valid uid | |
1496 | * again. | |
1497 | */ | |
fa3a5b22 CB |
1498 | if (wants_to_map_ids) { |
1499 | if (!handler->conf->inherit_ns[LXC_NS_USER]) { | |
1500 | ret = lxc_map_ids(id_map, handler->pid); | |
1501 | if (ret < 0) { | |
1502 | ERROR("Failed to set up id mapping."); | |
1503 | goto out_delete_net; | |
1504 | } | |
1505 | } | |
5b1e83cb SH |
1506 | } |
1507 | ||
9e2bdae4 | 1508 | if (lxc_sync_wake_child(handler, LXC_SYNC_STARTUP)) |
5b1e83cb | 1509 | goto out_delete_net; |
5b1e83cb | 1510 | |
9e2bdae4 | 1511 | if (lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE)) |
5b1e83cb | 1512 | goto out_delete_net; |
0ad19a3f | 1513 | |
d4ef7c50 | 1514 | if (!cgroup_create_legacy(handler)) { |
408da065 | 1515 | ERROR("Failed to setup legacy cgroups for container \"%s\".", name); |
ae5c8b8e | 1516 | goto out_delete_net; |
33ad9f1a | 1517 | } |
9daf6f5d | 1518 | if (!cgroup_setup_limits(handler, false)) { |
408da065 | 1519 | ERROR("Failed to setup cgroup limits for container \"%s\".", name); |
6031a6e5 DE |
1520 | goto out_delete_net; |
1521 | } | |
1522 | ||
d4ef7c50 | 1523 | if (!cgroup_enter(handler)) |
7fef7a06 | 1524 | goto out_delete_net; |
218d4250 | 1525 | |
0996e18a SH |
1526 | if (!cgroup_chown(handler)) |
1527 | goto out_delete_net; | |
1528 | ||
fa3a5b22 CB |
1529 | /* Now we're ready to preserve the network namespace */ |
1530 | ret = lxc_preserve_ns(handler->pid, "net"); | |
1531 | if (ret < 0) { | |
1532 | ERROR("%s - Failed to preserve net namespace", strerror(errno)); | |
1533 | goto out_delete_net; | |
1534 | } | |
1535 | handler->nsfd[LXC_NS_NET] = ret; | |
1536 | DEBUG("Preserved net namespace via fd %d", ret); | |
1537 | ||
408da065 | 1538 | /* Create the network configuration. */ |
d5088cf2 | 1539 | if (handler->clone_flags & CLONE_NEWNET) { |
74c6e2b0 CB |
1540 | if (lxc_network_move_created_netdev_priv(handler->lxcpath, |
1541 | handler->name, | |
28d9e29e | 1542 | &conf->network, |
74c6e2b0 | 1543 | handler->pid)) { |
408da065 | 1544 | ERROR("Failed to create the configured network."); |
7fef7a06 | 1545 | goto out_delete_net; |
82d5ae15 | 1546 | } |
74c6e2b0 CB |
1547 | |
1548 | if (lxc_create_network_unpriv(handler->lxcpath, handler->name, | |
28d9e29e | 1549 | &conf->network, |
74c6e2b0 CB |
1550 | handler->pid)) { |
1551 | ERROR("Failed to create the configured network."); | |
1552 | goto out_delete_net; | |
1553 | } | |
0ad19a3f | 1554 | } |
1555 | ||
7ab1ba02 CB |
1556 | if (lxc_network_send_veth_names_to_child(handler) < 0) { |
1557 | ERROR("Failed to send veth names to child"); | |
1558 | goto out_delete_net; | |
658979c5 SH |
1559 | } |
1560 | ||
61d7a733 YT |
1561 | if (!lxc_list_empty(&conf->procs)) { |
1562 | ret = setup_proc_filesystem(&conf->procs, handler->pid); | |
1563 | if (ret < 0) | |
1564 | goto out_delete_net; | |
1565 | } | |
1566 | ||
408da065 CB |
1567 | /* Tell the child to continue its initialization. We'll get |
1568 | * LXC_SYNC_CGROUP when it is ready for us to setup cgroups. | |
3c22086f CLG |
1569 | */ |
1570 | if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE)) | |
544a48a0 SH |
1571 | goto out_delete_net; |
1572 | ||
28d9e29e | 1573 | if (!lxc_list_empty(&conf->limits) && setup_resource_limits(&conf->limits, handler->pid)) { |
c6d09e15 | 1574 | ERROR("failed to setup resource limits for '%s'", name); |
84ff3af7 | 1575 | goto out_delete_net; |
c6d09e15 WB |
1576 | } |
1577 | ||
f4152036 CB |
1578 | if (lxc_sync_barrier_child(handler, LXC_SYNC_CGROUP_UNSHARE)) |
1579 | goto out_delete_net; | |
1580 | ||
9daf6f5d | 1581 | if (!cgroup_setup_limits(handler, true)) { |
408da065 | 1582 | ERROR("Failed to setup the devices cgroup for container \"%s\".", name); |
b98f7d6e | 1583 | goto out_delete_net; |
544a48a0 | 1584 | } |
f4152036 | 1585 | TRACE("Set up cgroup device limits"); |
544a48a0 | 1586 | |
73d28d42 | 1587 | cgroup_disconnect(); |
7e4dfe0b | 1588 | cgroups_connected = false; |
73d28d42 | 1589 | |
8bf3abfb CB |
1590 | if (handler->clone_flags & CLONE_NEWCGROUP) { |
1591 | /* Now we're ready to preserve the cgroup namespace */ | |
1592 | ret = lxc_preserve_ns(handler->pid, "cgroup"); | |
1593 | if (ret < 0) { | |
1594 | ERROR("%s - Failed to preserve cgroup namespace", strerror(errno)); | |
1595 | goto out_delete_net; | |
1596 | } | |
1597 | handler->nsfd[LXC_NS_CGROUP] = ret; | |
1598 | DEBUG("Preserved cgroup namespace via fd %d", ret); | |
1599 | } | |
1600 | ||
08dd2805 SH |
1601 | snprintf(pidstr, 20, "%d", handler->pid); |
1602 | if (setenv("LXC_PID", pidstr, 1)) | |
1603 | SYSERROR("Failed to set environment variable: LXC_PID=%s.", pidstr); | |
1604 | ||
1605 | /* Run any host-side start hooks */ | |
14a7b0f9 | 1606 | if (run_lxc_hooks(name, "start-host", conf, NULL)) { |
08dd2805 SH |
1607 | ERROR("Failed to run lxc.hook.start-host for container \"%s\".", name); |
1608 | return -1; | |
1609 | } | |
1610 | ||
408da065 CB |
1611 | /* Tell the child to complete its initialization and wait for it to exec |
1612 | * or return an error. (The child will never return | |
08dd2805 | 1613 | * LXC_SYNC_READY_START+1. It will either close the sync pipe, causing |
408da065 CB |
1614 | * lxc_sync_barrier_child to return success, or return a different |
1615 | * value, causing us to error out). | |
544a48a0 | 1616 | */ |
08dd2805 | 1617 | if (lxc_sync_barrier_child(handler, LXC_SYNC_READY_START)) |
3c22086f | 1618 | return -1; |
0ad19a3f | 1619 | |
790255cf CB |
1620 | if (lxc_network_recv_name_and_ifindex_from_child(handler) < 0) { |
1621 | ERROR("Failed to receive names and ifindices for network " | |
1622 | "devices from child"); | |
1623 | goto out_delete_net; | |
1624 | } | |
1625 | ||
1626 | /* Now all networks are created, network devices are moved into place, | |
1627 | * and the correct names and ifindeces in the respective namespaces have | |
1628 | * been recorded. The corresponding structs have now all been filled. So | |
1629 | * log them for debugging purposes. | |
1630 | */ | |
28d9e29e | 1631 | lxc_log_configured_netdevs(conf); |
790255cf | 1632 | |
f4152036 CB |
1633 | /* Read tty fds allocated by child. */ |
1634 | if (lxc_recv_ttys_from_child(handler) < 0) { | |
1635 | ERROR("Failed to receive tty info from child process."); | |
1636 | goto out_delete_net; | |
1637 | } | |
1638 | ||
23c53af9 | 1639 | if (handler->ops->post_start(handler, handler->data)) |
e6126dbe MN |
1640 | goto out_abort; |
1641 | ||
25c2aca5 | 1642 | if (lxc_set_state(name, handler, RUNNING)) { |
408da065 CB |
1643 | ERROR("Failed to set state for container \"%s\" to \"%s\".", name, |
1644 | lxc_state2str(RUNNING)); | |
59eb99ba | 1645 | goto out_abort; |
3f21c114 | 1646 | } |
22ebac19 | 1647 | |
3c22086f | 1648 | lxc_sync_fini(handler); |
0c547523 | 1649 | |
e6126dbe | 1650 | return 0; |
1ac470c0 | 1651 | |
7fef7a06 | 1652 | out_delete_net: |
7e4dfe0b SH |
1653 | if (cgroups_connected) |
1654 | cgroup_disconnect(); | |
1bd8d726 | 1655 | |
bb84beda CB |
1656 | if (handler->clone_flags & CLONE_NEWNET) |
1657 | lxc_delete_network(handler); | |
1bd8d726 | 1658 | |
59eb99ba DL |
1659 | out_abort: |
1660 | lxc_abort(name, handler); | |
3c22086f | 1661 | lxc_sync_fini(handler); |
5c068da9 SH |
1662 | if (handler->pinfd >= 0) { |
1663 | close(handler->pinfd); | |
1664 | handler->pinfd = -1; | |
1665 | } | |
1666 | ||
b79fcd86 | 1667 | return -1; |
59eb99ba | 1668 | } |
0ad19a3f | 1669 | |
aa460476 | 1670 | int __lxc_start(const char *name, struct lxc_handler *handler, |
507cee36 TA |
1671 | struct lxc_operations* ops, void *data, const char *lxcpath, |
1672 | bool backgrounded) | |
59eb99ba | 1673 | { |
59eb99ba | 1674 | int status; |
aa460476 | 1675 | int err = -1; |
aa460476 | 1676 | struct lxc_conf *conf = handler->conf; |
80090207 | 1677 | |
aa460476 | 1678 | if (lxc_init(name, handler) < 0) { |
408da065 | 1679 | ERROR("Failed to initialize container \"%s\".", name); |
66aeffc7 | 1680 | return -1; |
0ad19a3f | 1681 | } |
ee70bf78 CLG |
1682 | handler->ops = ops; |
1683 | handler->data = data; | |
507cee36 | 1684 | handler->backgrounded = backgrounded; |
e6126dbe | 1685 | |
76a26f55 | 1686 | if (!attach_block_device(handler->conf)) { |
408da065 | 1687 | ERROR("Failed to attach block device."); |
76a26f55 SH |
1688 | goto out_fini_nonet; |
1689 | } | |
1690 | ||
35120d9c | 1691 | if (geteuid() == 0 && !lxc_list_empty(&conf->id_map)) { |
408da065 | 1692 | /* If the backing store is a device, mount it here and now. */ |
35120d9c SH |
1693 | if (rootfs_is_blockdev(conf)) { |
1694 | if (unshare(CLONE_NEWNS) < 0) { | |
408da065 | 1695 | ERROR("Failed to unshare CLONE_NEWNS."); |
35120d9c SH |
1696 | goto out_fini_nonet; |
1697 | } | |
408da065 CB |
1698 | INFO("Unshared CLONE_NEWNS."); |
1699 | ||
6a0c909a | 1700 | remount_all_slave(); |
35120d9c | 1701 | if (do_rootfs_setup(conf, name, lxcpath) < 0) { |
408da065 | 1702 | ERROR("Error setting up rootfs mount as root before spawn."); |
35120d9c SH |
1703 | goto out_fini_nonet; |
1704 | } | |
408da065 | 1705 | INFO("Set up container rootfs as host root."); |
35120d9c SH |
1706 | } |
1707 | } | |
1708 | ||
23c53af9 | 1709 | err = lxc_spawn(handler); |
59eb99ba | 1710 | if (err) { |
408da065 | 1711 | ERROR("Failed to spawn container \"%s\".", name); |
76a26f55 | 1712 | goto out_detach_blockdev; |
0ad19a3f | 1713 | } |
a9e1109e CB |
1714 | /* close parent side of data socket */ |
1715 | close(handler->data_sock[0]); | |
1716 | handler->data_sock[0] = -1; | |
1717 | close(handler->data_sock[1]); | |
1718 | handler->data_sock[1] = -1; | |
0ad19a3f | 1719 | |
8bee8851 WB |
1720 | handler->conf->reboot = 0; |
1721 | ||
3a0f472d | 1722 | err = lxc_poll(name, handler); |
e043236e | 1723 | if (err) { |
408da065 | 1724 | ERROR("LXC mainloop exited with error: %d.", err); |
59eb99ba DL |
1725 | goto out_abort; |
1726 | } | |
0ad19a3f | 1727 | |
3a0f472d | 1728 | while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR) |
1bc5cc8c | 1729 | continue; |
e043236e | 1730 | |
408da065 CB |
1731 | /* If the child process exited but was not signaled, it didn't call |
1732 | * reboot. This should mean it was an lxc-execute which simply exited. | |
1733 | * In any case, treat it as a 'halt'. | |
8b004f07 | 1734 | */ |
d028235d | 1735 | if (WIFSIGNALED(status)) { |
8b004f07 SH |
1736 | switch(WTERMSIG(status)) { |
1737 | case SIGINT: /* halt */ | |
408da065 | 1738 | DEBUG("Container \"%s\" is halting.", name); |
8b004f07 SH |
1739 | break; |
1740 | case SIGHUP: /* reboot */ | |
408da065 | 1741 | DEBUG("Container \"%s\" is rebooting.", name); |
8b004f07 SH |
1742 | handler->conf->reboot = 1; |
1743 | break; | |
c2b9bd9e | 1744 | case SIGSYS: /* seccomp */ |
408da065 | 1745 | DEBUG("Container \"%s\" violated its seccomp policy.", name); |
c2b9bd9e | 1746 | break; |
8b004f07 | 1747 | default: |
408da065 | 1748 | DEBUG("Unknown exit status for container \"%s\" init %d.", name, WTERMSIG(status)); |
8b004f07 SH |
1749 | break; |
1750 | } | |
d028235d | 1751 | } |
828695d9 | 1752 | |
b809f232 CB |
1753 | err = lxc_restore_phys_nics_to_netns(handler); |
1754 | if (err < 0) | |
1755 | ERROR("Failed to move physical network devices back to parent " | |
1756 | "network namespace"); | |
ce5782df | 1757 | |
5c068da9 SH |
1758 | if (handler->pinfd >= 0) { |
1759 | close(handler->pinfd); | |
1760 | handler->pinfd = -1; | |
1761 | } | |
1762 | ||
1787abca | 1763 | lxc_monitor_send_exit_code(name, status, handler->lxcpath); |
3a0f472d | 1764 | err = lxc_error_set_and_log(handler->pid, status); |
1bd8d726 | 1765 | |
9d7f9e52 | 1766 | out_fini: |
bb84beda | 1767 | lxc_delete_network(handler); |
74a2b586 | 1768 | |
76a26f55 SH |
1769 | out_detach_blockdev: |
1770 | detach_block_device(handler->conf); | |
1771 | ||
74a2b586 | 1772 | out_fini_nonet: |
3a0f472d | 1773 | lxc_fini(name, handler); |
0ad19a3f | 1774 | return err; |
1775 | ||
59eb99ba | 1776 | out_abort: |
3a0f472d | 1777 | lxc_abort(name, handler); |
9d7f9e52 | 1778 | goto out_fini; |
0ad19a3f | 1779 | } |
ee70bf78 CLG |
1780 | |
1781 | struct start_args { | |
1782 | char *const *argv; | |
1783 | }; | |
1784 | ||
1785 | static int start(struct lxc_handler *handler, void* data) | |
1786 | { | |
1787 | struct start_args *arg = data; | |
1788 | ||
408da065 | 1789 | NOTICE("Exec'ing \"%s\".", arg->argv[0]); |
ee70bf78 CLG |
1790 | |
1791 | execvp(arg->argv[0], arg->argv); | |
408da065 | 1792 | SYSERROR("Failed to exec \"%s\".", arg->argv[0]); |
ee70bf78 CLG |
1793 | return 0; |
1794 | } | |
1795 | ||
1796 | static int post_start(struct lxc_handler *handler, void* data) | |
1797 | { | |
1798 | struct start_args *arg = data; | |
1799 | ||
408da065 | 1800 | NOTICE("Started \"%s\" with pid \"%d\".", arg->argv[0], handler->pid); |
ee70bf78 CLG |
1801 | return 0; |
1802 | } | |
1803 | ||
1804 | static struct lxc_operations start_ops = { | |
1805 | .start = start, | |
1806 | .post_start = post_start | |
1807 | }; | |
1808 | ||
aa460476 | 1809 | int lxc_start(const char *name, char *const argv[], struct lxc_handler *handler, |
507cee36 | 1810 | const char *lxcpath, bool backgrounded) |
ee70bf78 CLG |
1811 | { |
1812 | struct start_args start_arg = { | |
1813 | .argv = argv, | |
1814 | }; | |
1815 | ||
aa460476 | 1816 | return __lxc_start(name, handler, &start_ops, &start_arg, lxcpath, backgrounded); |
ee70bf78 | 1817 | } |
28272964 CB |
1818 | |
1819 | static void lxc_destroy_container_on_signal(struct lxc_handler *handler, | |
1820 | const char *name) | |
1821 | { | |
1822 | char destroy[MAXPATHLEN]; | |
1823 | bool bret = true; | |
1824 | int ret = 0; | |
f01f7975 | 1825 | struct lxc_container *c; |
df31363a | 1826 | if (handler->conf->rootfs.path && handler->conf->rootfs.mount) { |
790255cf | 1827 | bret = do_destroy_container(handler); |
28272964 | 1828 | if (!bret) { |
408da065 | 1829 | ERROR("Error destroying rootfs for container \"%s\".", name); |
28272964 CB |
1830 | return; |
1831 | } | |
1832 | } | |
408da065 | 1833 | INFO("Destroyed rootfs for container \"%s\".", name); |
28272964 CB |
1834 | |
1835 | ret = snprintf(destroy, MAXPATHLEN, "%s/%s", handler->lxcpath, name); | |
1836 | if (ret < 0 || ret >= MAXPATHLEN) { | |
408da065 | 1837 | ERROR("Error destroying directory for container \"%s\".", name); |
28272964 CB |
1838 | return; |
1839 | } | |
1840 | ||
f01f7975 CB |
1841 | c = lxc_container_new(name, handler->lxcpath); |
1842 | if (c) { | |
1843 | if (container_disk_lock(c)) { | |
408da065 | 1844 | INFO("Could not update lxc_snapshots file."); |
f01f7975 CB |
1845 | lxc_container_put(c); |
1846 | } else { | |
1847 | mod_all_rdeps(c, false); | |
1848 | container_disk_unlock(c); | |
1849 | lxc_container_put(c); | |
1850 | } | |
1851 | } | |
1852 | ||
d0fbc7ba | 1853 | if (!handler->am_root) |
94b0a3ac CB |
1854 | ret = userns_exec_full(handler->conf, lxc_rmdir_onedev_wrapper, |
1855 | destroy, "lxc_rmdir_onedev_wrapper"); | |
28272964 CB |
1856 | else |
1857 | ret = lxc_rmdir_onedev(destroy, NULL); | |
1858 | ||
1859 | if (ret < 0) { | |
408da065 | 1860 | ERROR("Error destroying directory for container \"%s\".", name); |
28272964 CB |
1861 | return; |
1862 | } | |
408da065 | 1863 | INFO("Destroyed directory for container \"%s\".", name); |
28272964 CB |
1864 | } |
1865 | ||
1866 | static int lxc_rmdir_onedev_wrapper(void *data) | |
1867 | { | |
1868 | char *arg = (char *) data; | |
1869 | return lxc_rmdir_onedev(arg, NULL); | |
1870 | } | |
1871 | ||
790255cf | 1872 | static bool do_destroy_container(struct lxc_handler *handler) { |
94b0a3ac CB |
1873 | int ret; |
1874 | ||
d0fbc7ba | 1875 | if (!handler->am_root) { |
94b0a3ac CB |
1876 | ret = userns_exec_full(handler->conf, storage_destroy_wrapper, |
1877 | handler->conf, "storage_destroy_wrapper"); | |
1878 | if (ret < 0) | |
d028235d | 1879 | return false; |
10bc1861 | 1880 | |
d028235d SG |
1881 | return true; |
1882 | } | |
10bc1861 | 1883 | |
790255cf | 1884 | return storage_destroy(handler->conf); |
28272964 | 1885 | } |