]>
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" |
1cc8bd4b | 76 | #include "list.h" |
f3787121 | 77 | #include "log.h" |
f2d5a09d | 78 | #include "lxccontainer.h" |
f3787121 CB |
79 | #include "lxclock.h" |
80 | #include "lxcseccomp.h" | |
f3787121 | 81 | #include "mainloop.h" |
63376d7d | 82 | #include "monitor.h" |
f549edcc | 83 | #include "namespace.h" |
811ef482 | 84 | #include "network.h" |
f3787121 | 85 | #include "start.h" |
28d832c4 | 86 | #include "storage.h" |
f2d5a09d | 87 | #include "storage_utils.h" |
f3787121 CB |
88 | #include "sync.h" |
89 | #include "utils.h" | |
fe4de9a6 | 90 | #include "lsm/lsm.h" |
36eb9bde CLG |
91 | |
92 | lxc_log_define(lxc_start, lxc); | |
93 | ||
f01f7975 | 94 | extern void mod_all_rdeps(struct lxc_container *c, bool inc); |
790255cf | 95 | static bool do_destroy_container(struct lxc_handler *handler); |
28272964 CB |
96 | static int lxc_rmdir_onedev_wrapper(void *data); |
97 | static void lxc_destroy_container_on_signal(struct lxc_handler *handler, | |
98 | const char *name); | |
99 | ||
c8154066 SH |
100 | static void print_top_failing_dir(const char *path) |
101 | { | |
102 | size_t len = strlen(path); | |
f3787121 | 103 | char *copy = alloca(len + 1), *p, *e, saved; |
c8154066 SH |
104 | strcpy(copy, path); |
105 | ||
106 | p = copy; | |
107 | e = copy + len; | |
108 | while (p < e) { | |
f3787121 CB |
109 | while (p < e && *p == '/') |
110 | p++; | |
111 | while (p < e && *p != '/') | |
112 | p++; | |
c8154066 SH |
113 | saved = *p; |
114 | *p = '\0'; | |
115 | if (access(copy, X_OK)) { | |
408da065 CB |
116 | SYSERROR("Could not access %s. Please grant it x " |
117 | "access, or add an ACL for the container " | |
118 | "root.", copy); | |
c8154066 SH |
119 | return; |
120 | } | |
121 | *p = saved; | |
122 | } | |
123 | } | |
124 | ||
408da065 CB |
125 | static void close_ns(int ns_fd[LXC_NS_MAX]) |
126 | { | |
9f30a190 MM |
127 | int i; |
128 | ||
9f30a190 MM |
129 | for (i = 0; i < LXC_NS_MAX; i++) { |
130 | if (ns_fd[i] > -1) { | |
131 | close(ns_fd[i]); | |
132 | ns_fd[i] = -1; | |
133 | } | |
134 | } | |
9f30a190 MM |
135 | } |
136 | ||
4d8ac866 | 137 | /* preserve_ns: open /proc/@pid/ns/@ns for each namespace specified |
62d05d9b | 138 | * in clone_flags. |
4d8ac866 | 139 | * Return true on success, false on failure. |
62d05d9b | 140 | */ |
4d8ac866 | 141 | static bool preserve_ns(int ns_fd[LXC_NS_MAX], int clone_flags, pid_t pid) |
f3787121 | 142 | { |
62d05d9b | 143 | int i, ret; |
9f30a190 | 144 | |
9f30a190 MM |
145 | for (i = 0; i < LXC_NS_MAX; i++) |
146 | ns_fd[i] = -1; | |
147 | ||
4d8ac866 CB |
148 | ret = lxc_preserve_ns(pid, ""); |
149 | if (ret < 0) { | |
150 | SYSERROR("Kernel does not support attaching to namespaces."); | |
62d05d9b | 151 | return false; |
4d8ac866 CB |
152 | } else { |
153 | close(ret); | |
cd43d2d1 SH |
154 | } |
155 | ||
9f30a190 MM |
156 | for (i = 0; i < LXC_NS_MAX; i++) { |
157 | if ((clone_flags & ns_info[i].clone_flag) == 0) | |
158 | continue; | |
28d9e29e | 159 | |
4d8ac866 | 160 | ns_fd[i] = lxc_preserve_ns(pid, ns_info[i].proc_name); |
9f30a190 MM |
161 | if (ns_fd[i] < 0) |
162 | goto error; | |
fa3a5b22 CB |
163 | |
164 | DEBUG("Preserved %s namespace via fd %d", ns_info[i].proc_name, ns_fd[i]); | |
9f30a190 MM |
165 | } |
166 | ||
62d05d9b | 167 | return true; |
9f30a190 MM |
168 | |
169 | error: | |
4d8ac866 CB |
170 | if (errno == ENOENT) |
171 | SYSERROR("Kernel does not support attaching to %s namespaces.", ns_info[i].proc_name); | |
172 | else | |
173 | SYSERROR("Failed to open file descriptor for %s namespace: %s.", ns_info[i].proc_name, strerror(errno)); | |
9f30a190 | 174 | close_ns(ns_fd); |
62d05d9b | 175 | return false; |
9f30a190 MM |
176 | } |
177 | ||
80090207 CLG |
178 | static int match_fd(int fd) |
179 | { | |
180 | return (fd == 0 || fd == 1 || fd == 2); | |
181 | } | |
182 | ||
47a46cf1 CB |
183 | int lxc_check_inherited(struct lxc_conf *conf, bool closeall, |
184 | int *fds_to_ignore, size_t len_fds) | |
80090207 | 185 | { |
74f96976 | 186 | struct dirent *direntp; |
80090207 | 187 | int fd, fddir; |
47a46cf1 | 188 | size_t i; |
80090207 | 189 | DIR *dir; |
80090207 | 190 | |
d2cf4c37 SH |
191 | if (conf && conf->close_all_fds) |
192 | closeall = true; | |
193 | ||
b119f362 | 194 | restart: |
80090207 CLG |
195 | dir = opendir("/proc/self/fd"); |
196 | if (!dir) { | |
13277ec4 | 197 | WARN("Failed to open directory: %s.", strerror(errno)); |
80090207 CLG |
198 | return -1; |
199 | } | |
200 | ||
201 | fddir = dirfd(dir); | |
202 | ||
74f96976 | 203 | while ((direntp = readdir(dir))) { |
d39b10eb CB |
204 | struct lxc_list *cur; |
205 | bool matched = false; | |
206 | ||
80090207 CLG |
207 | if (!strcmp(direntp->d_name, ".")) |
208 | continue; | |
209 | ||
210 | if (!strcmp(direntp->d_name, "..")) | |
211 | continue; | |
212 | ||
d4cff0d2 CB |
213 | if (lxc_safe_int(direntp->d_name, &fd) < 0) { |
214 | INFO("Could not parse file descriptor for: %s", direntp->d_name); | |
215 | continue; | |
216 | } | |
80090207 | 217 | |
47a46cf1 CB |
218 | for (i = 0; i < len_fds; i++) |
219 | if (fds_to_ignore[i] == fd) | |
220 | break; | |
221 | ||
222 | if (fd == fddir || fd == lxc_log_fd || | |
223 | (i < len_fds && fd == fds_to_ignore[i])) | |
80090207 CLG |
224 | continue; |
225 | ||
d39b10eb | 226 | /* Keep state clients that wait on reboots. */ |
3ee9e4fb CB |
227 | if (conf) { |
228 | lxc_list_for_each(cur, &conf->state_clients) { | |
229 | struct lxc_state_client *client = cur->elem; | |
d39b10eb | 230 | |
3ee9e4fb CB |
231 | if (client->clientfd != fd) |
232 | continue; | |
d39b10eb | 233 | |
3ee9e4fb CB |
234 | matched = true; |
235 | break; | |
236 | } | |
d39b10eb CB |
237 | } |
238 | ||
239 | if (matched) | |
240 | continue; | |
241 | ||
858377e4 SH |
242 | if (current_config && fd == current_config->logfd) |
243 | continue; | |
244 | ||
80090207 CLG |
245 | if (match_fd(fd)) |
246 | continue; | |
80090207 | 247 | |
d2cf4c37 | 248 | if (closeall) { |
b119f362 SH |
249 | close(fd); |
250 | closedir(dir); | |
28d9e29e | 251 | INFO("Closed inherited fd %d", fd); |
b119f362 SH |
252 | goto restart; |
253 | } | |
28d9e29e | 254 | WARN("Inherited fd %d", fd); |
80090207 CLG |
255 | } |
256 | ||
408da065 | 257 | /* Only enable syslog at this point to avoid the above logging function |
64c57ea1 BD |
258 | * to open a new fd and make the check_inherited function enter an |
259 | * infinite loop. | |
260 | */ | |
261 | lxc_log_enable_syslog(); | |
262 | ||
92c7f629 GK |
263 | closedir(dir); /* cannot fail */ |
264 | return 0; | |
80090207 CLG |
265 | } |
266 | ||
83ee7875 | 267 | static int setup_signal_fd(sigset_t *oldmask) |
b0a33c1e | 268 | { |
269 | sigset_t mask; | |
270 | int fd; | |
271 | ||
408da065 | 272 | /* Block everything except serious error signals. */ |
f3304a29 FW |
273 | if (sigfillset(&mask) || |
274 | sigdelset(&mask, SIGILL) || | |
275 | sigdelset(&mask, SIGSEGV) || | |
276 | sigdelset(&mask, SIGBUS) || | |
b5159817 | 277 | sigdelset(&mask, SIGWINCH) || |
f3304a29 | 278 | sigprocmask(SIG_BLOCK, &mask, oldmask)) { |
408da065 | 279 | SYSERROR("Failed to set signal mask."); |
b0a33c1e | 280 | return -1; |
281 | } | |
282 | ||
283 | fd = signalfd(-1, &mask, 0); | |
284 | if (fd < 0) { | |
408da065 | 285 | SYSERROR("Failed to create signal file descriptor."); |
b0a33c1e | 286 | return -1; |
287 | } | |
288 | ||
289 | if (fcntl(fd, F_SETFD, FD_CLOEXEC)) { | |
408da065 | 290 | SYSERROR("Failed to set FD_CLOEXEC on the signal file descriptor: %d.", fd); |
b0a33c1e | 291 | close(fd); |
292 | return -1; | |
293 | } | |
294 | ||
408da065 | 295 | DEBUG("Set SIGCHLD handler with file descriptor: %d.", fd); |
1ac470c0 | 296 | |
b0a33c1e | 297 | return fd; |
298 | } | |
299 | ||
84c92abd | 300 | static int signal_handler(int fd, uint32_t events, void *data, |
f3787121 | 301 | struct lxc_epoll_descr *descr) |
b0a33c1e | 302 | { |
15cd25fd | 303 | int ret; |
3c319edb CB |
304 | siginfo_t info; |
305 | struct signalfd_siginfo siginfo; | |
306 | struct lxc_handler *hdlr = data; | |
15cd25fd DL |
307 | |
308 | ret = read(fd, &siginfo, sizeof(siginfo)); | |
309 | if (ret < 0) { | |
408da065 | 310 | ERROR("Failed to read signal info from signal file descriptor: %d.", fd); |
15cd25fd DL |
311 | return -1; |
312 | } | |
313 | ||
314 | if (ret != sizeof(siginfo)) { | |
408da065 | 315 | ERROR("Unexpected size for siginfo struct."); |
15cd25fd DL |
316 | return -1; |
317 | } | |
318 | ||
408da065 | 319 | /* Check whether init is running. */ |
80507ee8 | 320 | info.si_pid = 0; |
3c319edb CB |
321 | ret = waitid(P_PID, hdlr->pid, &info, WEXITED | WNOWAIT | WNOHANG); |
322 | if (ret == 0 && info.si_pid == hdlr->pid) | |
323 | hdlr->init_died = true; | |
80507ee8 | 324 | |
f3304a29 | 325 | if (siginfo.ssi_signo != SIGCHLD) { |
3c319edb CB |
326 | kill(hdlr->pid, siginfo.ssi_signo); |
327 | INFO("Forwarded signal %d to pid %d.", siginfo.ssi_signo, hdlr->pid); | |
328 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE : 0; | |
f3304a29 FW |
329 | } |
330 | ||
408da065 CB |
331 | if (siginfo.ssi_code == CLD_STOPPED) { |
332 | INFO("Container init process was stopped."); | |
3c319edb | 333 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE : 0; |
408da065 CB |
334 | } else if (siginfo.ssi_code == CLD_CONTINUED) { |
335 | INFO("Container init process was continued."); | |
3c319edb | 336 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE : 0; |
15cd25fd | 337 | } |
1ac470c0 | 338 | |
408da065 CB |
339 | /* More robustness, protect ourself from a SIGCHLD sent |
340 | * by a process different from the container init. | |
82d89dce | 341 | */ |
3c319edb CB |
342 | if (siginfo.ssi_pid != hdlr->pid) { |
343 | NOTICE("Received SIGCHLD from pid %d instead of container init %d.", siginfo.ssi_pid, hdlr->pid); | |
344 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE : 0; | |
82d89dce DL |
345 | } |
346 | ||
3c319edb CB |
347 | DEBUG("Container init process %d exited.", hdlr->pid); |
348 | return LXC_MAINLOOP_CLOSE; | |
b0a33c1e | 349 | } |
350 | ||
5e5576a4 CB |
351 | static int lxc_serve_state_clients(const char *name, |
352 | struct lxc_handler *handler, | |
353 | lxc_state_t state) | |
66aeffc7 | 354 | { |
dbc9832d CB |
355 | ssize_t ret; |
356 | struct lxc_list *cur, *next; | |
d39b10eb | 357 | struct lxc_state_client *client; |
dbc9832d CB |
358 | struct lxc_msg msg = {.type = lxc_msg_state, .value = state}; |
359 | ||
bc631984 | 360 | process_lock(); |
66aeffc7 | 361 | handler->state = state; |
9dfa4041 | 362 | TRACE("Set container state to %s", lxc_state2str(state)); |
dbc9832d | 363 | |
d39b10eb | 364 | if (lxc_list_empty(&handler->conf->state_clients)) { |
9dfa4041 | 365 | TRACE("No state clients registered"); |
dbc9832d | 366 | process_unlock(); |
65ee1673 | 367 | lxc_monitor_send_state(name, state, handler->lxcpath); |
dbc9832d CB |
368 | return 0; |
369 | } | |
370 | ||
371 | strncpy(msg.name, name, sizeof(msg.name)); | |
372 | msg.name[sizeof(msg.name) - 1] = 0; | |
373 | ||
d39b10eb | 374 | lxc_list_for_each_safe(cur, &handler->conf->state_clients, next) { |
dbc9832d CB |
375 | client = cur->elem; |
376 | ||
d39b10eb | 377 | if (client->states[state] == 0) { |
9dfa4041 | 378 | TRACE("State %s not registered for state client %d", |
dbc9832d CB |
379 | lxc_state2str(state), client->clientfd); |
380 | continue; | |
381 | } | |
382 | ||
9dfa4041 | 383 | TRACE("Sending state %s to state client %d", |
dbc9832d CB |
384 | lxc_state2str(state), client->clientfd); |
385 | ||
386 | again: | |
387 | ret = send(client->clientfd, &msg, sizeof(msg), 0); | |
ee8377bd CB |
388 | if (ret <= 0) { |
389 | if (errno == EINTR) { | |
390 | TRACE("Caught EINTR; retrying"); | |
dbc9832d | 391 | goto again; |
ee8377bd | 392 | } |
dbc9832d | 393 | |
9dfa4041 CB |
394 | ERROR("%s - Failed to send message to client", |
395 | strerror(errno)); | |
dbc9832d CB |
396 | } |
397 | ||
398 | /* kick client from list */ | |
399 | close(client->clientfd); | |
400 | lxc_list_del(cur); | |
401 | free(cur->elem); | |
402 | free(cur); | |
403 | } | |
404 | process_unlock(); | |
405 | ||
5e5576a4 CB |
406 | return 0; |
407 | } | |
408 | ||
409 | static int lxc_serve_state_socket_pair(const char *name, | |
410 | struct lxc_handler *handler, | |
411 | lxc_state_t state) | |
412 | { | |
413 | ssize_t ret; | |
414 | ||
415 | if (!handler->backgrounded || | |
416 | handler->state_socket_pair[1] < 0 || | |
417 | state == STARTING) | |
418 | return 0; | |
419 | ||
420 | /* Close read end of the socket pair. */ | |
421 | close(handler->state_socket_pair[0]); | |
422 | handler->state_socket_pair[0] = -1; | |
423 | ||
ee8377bd | 424 | again: |
5e5576a4 CB |
425 | ret = lxc_abstract_unix_send_credential(handler->state_socket_pair[1], |
426 | &(int){state}, sizeof(int)); | |
ee8377bd CB |
427 | if (ret != sizeof(int)) { |
428 | if (errno == EINTR) | |
429 | goto again; | |
5e5576a4 CB |
430 | SYSERROR("Failed to send state to %d", |
431 | handler->state_socket_pair[1]); | |
ee8377bd CB |
432 | return -1; |
433 | } | |
5e5576a4 CB |
434 | |
435 | TRACE("Sent container state \"%s\" to %d", lxc_state2str(state), | |
436 | handler->state_socket_pair[1]); | |
437 | ||
438 | /* Close write end of the socket pair. */ | |
439 | close(handler->state_socket_pair[1]); | |
440 | handler->state_socket_pair[1] = -1; | |
441 | ||
442 | return 0; | |
443 | } | |
444 | ||
445 | int lxc_set_state(const char *name, struct lxc_handler *handler, | |
446 | lxc_state_t state) | |
447 | { | |
448 | int ret; | |
449 | ||
450 | ret = lxc_serve_state_socket_pair(name, handler, state); | |
451 | if (ret < 0) { | |
452 | ERROR("Failed to synchronize via anonymous pair of unix sockets"); | |
453 | return -1; | |
454 | } | |
455 | ||
456 | ret = lxc_serve_state_clients(name, handler, state); | |
457 | if (ret < 0) | |
458 | return -1; | |
459 | ||
dbc9832d CB |
460 | /* This function will try to connect to the legacy lxc-monitord state |
461 | * server and only exists for backwards compatibility. | |
462 | */ | |
9123e471 | 463 | lxc_monitor_send_state(name, state, handler->lxcpath); |
dbc9832d | 464 | |
66aeffc7 DL |
465 | return 0; |
466 | } | |
467 | ||
735f2c6e | 468 | int lxc_poll(const char *name, struct lxc_handler *handler) |
b0a33c1e | 469 | { |
3c319edb CB |
470 | int ret; |
471 | struct lxc_epoll_descr descr, descr_console; | |
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 | ||
1cc8bd4b | 485 | ret = lxc_mainloop_add_handler(&descr, handler->sigfd, signal_handler, handler); |
3c319edb | 486 | if (ret < 0) { |
1cc8bd4b | 487 | ERROR("Failed to add signal handler for %d to mainloop", handler->sigfd); |
3c319edb | 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 | 511 | ret = lxc_mainloop(&descr, -1); |
1cc8bd4b CB |
512 | close(descr.epfd); |
513 | descr.epfd = -EBADF; | |
514 | if (ret < 0 || !handler->init_died) | |
515 | goto out_mainloop; | |
516 | ||
517 | ret = lxc_mainloop(&descr_console, 0); | |
3c319edb CB |
518 | |
519 | out_mainloop: | |
b0a33c1e | 520 | lxc_mainloop_close(&descr); |
1cc8bd4b | 521 | TRACE("Closed mainloop"); |
408da065 | 522 | |
3c319edb CB |
523 | out_mainloop_console: |
524 | lxc_mainloop_close(&descr_console); | |
1cc8bd4b | 525 | TRACE("Closed console mainloop"); |
3c319edb | 526 | |
b0a33c1e | 527 | out_sigfd: |
1cc8bd4b CB |
528 | close(handler->sigfd); |
529 | TRACE("Closed signal file descriptor %d", handler->sigfd); | |
530 | handler->sigfd = -EBADF; | |
408da065 | 531 | |
1cc8bd4b | 532 | return ret; |
b0a33c1e | 533 | } |
534 | ||
41784e4e CB |
535 | void lxc_zero_handler(struct lxc_handler *handler) |
536 | { | |
537 | int i; | |
538 | ||
539 | memset(handler, 0, sizeof(struct lxc_handler)); | |
540 | ||
541 | handler->clone_flags = -1; | |
542 | ||
543 | handler->pinfd = -1; | |
544 | ||
545 | handler->sigfd = -1; | |
546 | ||
547 | for (i = 0; i < LXC_NS_MAX; i++) | |
548 | handler->nsfd[i] = -1; | |
549 | ||
550 | handler->data_sock[0] = -1; | |
551 | handler->data_sock[1] = -1; | |
552 | ||
553 | handler->state_socket_pair[0] = -1; | |
554 | handler->state_socket_pair[1] = -1; | |
555 | ||
556 | handler->sync_sock[0] = -1; | |
557 | handler->sync_sock[1] = -1; | |
558 | } | |
559 | ||
1cc8bd4b CB |
560 | static void lxc_put_nsfds(struct lxc_handler *handler) |
561 | { | |
562 | int i; | |
563 | ||
564 | for (i = 0; i < LXC_NS_MAX; i++) { | |
565 | if (handler->nsfd[i] < 0) | |
566 | continue; | |
567 | ||
568 | close(handler->nsfd[i]); | |
569 | handler->nsfd[i] = -EBADF; | |
570 | } | |
571 | } | |
572 | ||
f2e07cb6 CB |
573 | void lxc_free_handler(struct lxc_handler *handler) |
574 | { | |
1cc8bd4b CB |
575 | if (handler->pinfd >= 0) |
576 | close(handler->pinfd); | |
577 | ||
578 | if (handler->sigfd >= 0) | |
579 | close(handler->sigfd); | |
580 | ||
581 | lxc_put_nsfds(handler); | |
582 | ||
583 | if (handler->conf && handler->conf->reboot == 0) | |
584 | if (handler->conf->maincmd_fd) | |
585 | close(handler->conf->maincmd_fd); | |
f2e07cb6 | 586 | |
5e5576a4 CB |
587 | if (handler->state_socket_pair[0] >= 0) |
588 | close(handler->state_socket_pair[0]); | |
589 | ||
590 | if (handler->state_socket_pair[1] >= 0) | |
591 | close(handler->state_socket_pair[1]); | |
592 | ||
f2e07cb6 CB |
593 | handler->conf = NULL; |
594 | free(handler); | |
1cc8bd4b | 595 | handler = NULL; |
f2e07cb6 CB |
596 | } |
597 | ||
aa460476 | 598 | struct lxc_handler *lxc_init_handler(const char *name, struct lxc_conf *conf, |
5e5576a4 | 599 | const char *lxcpath, bool daemonize) |
59eb99ba | 600 | { |
5e5576a4 | 601 | int i, ret; |
3a0f472d DL |
602 | struct lxc_handler *handler; |
603 | ||
604 | handler = malloc(sizeof(*handler)); | |
aa460476 CB |
605 | if (!handler) { |
606 | ERROR("failed to allocate memory"); | |
3a0f472d | 607 | return NULL; |
aa460476 | 608 | } |
59eb99ba DL |
609 | |
610 | memset(handler, 0, sizeof(*handler)); | |
611 | ||
790255cf CB |
612 | /* Note that am_unpriv() checks the effective uid. We probably don't |
613 | * care if we are real root only if we are running as root so this | |
614 | * should be fine. | |
615 | */ | |
d0fbc7ba | 616 | handler->am_root = !am_unpriv(); |
c6012571 | 617 | handler->data_sock[0] = handler->data_sock[1] = -1; |
fae349da | 618 | handler->conf = conf; |
9123e471 | 619 | handler->lxcpath = lxcpath; |
5c068da9 | 620 | handler->pinfd = -1; |
1cc8bd4b CB |
621 | handler->sigfd = -EBADF; |
622 | handler->init_died = false; | |
5e5576a4 | 623 | handler->state_socket_pair[0] = handler->state_socket_pair[1] = -1; |
d39b10eb CB |
624 | if (handler->conf->reboot == 0) |
625 | lxc_list_init(&handler->conf->state_clients); | |
fae349da | 626 | |
b6b2b194 WB |
627 | for (i = 0; i < LXC_NS_MAX; i++) |
628 | handler->nsfd[i] = -1; | |
629 | ||
f0ecc19d | 630 | handler->name = name; |
3bdf52d7 | 631 | |
5e5576a4 CB |
632 | if (daemonize && !handler->conf->reboot) { |
633 | /* Create socketpair() to synchronize on daemonized startup. | |
634 | * When the container reboots we don't need to synchronize again | |
635 | * currently so don't open another socketpair(). | |
636 | */ | |
637 | ret = socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0, | |
638 | handler->state_socket_pair); | |
639 | if (ret < 0) { | |
640 | ERROR("Failed to create anonymous pair of unix sockets"); | |
641 | goto on_error; | |
642 | } | |
643 | TRACE("Created anonymous pair {%d,%d} of unix sockets", | |
644 | handler->state_socket_pair[0], | |
645 | handler->state_socket_pair[1]); | |
646 | } | |
647 | ||
d39b10eb CB |
648 | if (handler->conf->reboot == 0) { |
649 | handler->conf->maincmd_fd = lxc_cmd_init(name, lxcpath, "command"); | |
650 | if (handler->conf->maincmd_fd < 0) { | |
651 | ERROR("Failed to set up command socket"); | |
652 | goto on_error; | |
653 | } | |
aa460476 | 654 | } |
9dfa4041 | 655 | TRACE("Unix domain socket %d for command server is ready", |
aa460476 CB |
656 | handler->conf->maincmd_fd); |
657 | ||
658 | return handler; | |
659 | ||
f2e07cb6 CB |
660 | on_error: |
661 | lxc_free_handler(handler); | |
aa460476 CB |
662 | |
663 | return NULL; | |
664 | } | |
665 | ||
666 | int lxc_init(const char *name, struct lxc_handler *handler) | |
667 | { | |
a2c09be0 | 668 | int ret; |
4a03ded4 | 669 | const char *loglevel; |
aa460476 CB |
670 | struct lxc_conf *conf = handler->conf; |
671 | ||
672 | lsm_init(); | |
673 | TRACE("initialized LSM"); | |
d2e30e99 | 674 | |
8f2c3a70 | 675 | if (lxc_read_seccomp_config(conf) != 0) { |
408da065 | 676 | ERROR("Failed loading seccomp policy."); |
d2e30e99 | 677 | goto out_close_maincmd_fd; |
8f2c3a70 | 678 | } |
dbc9832d | 679 | TRACE("read seccomp policy"); |
8f2c3a70 | 680 | |
408da065 | 681 | /* Begin by setting the state to STARTING. */ |
25c2aca5 | 682 | if (lxc_set_state(name, handler, STARTING)) { |
408da065 | 683 | ERROR("Failed to set state for container \"%s\" to \"%s\".", name, lxc_state2str(STARTING)); |
051151de | 684 | goto out_close_maincmd_fd; |
0ad19a3f | 685 | } |
dbc9832d | 686 | TRACE("set container state to \"STARTING\""); |
0ad19a3f | 687 | |
408da065 CB |
688 | /* Start of environment variable setup for hooks. */ |
689 | if (name && setenv("LXC_NAME", name, 1)) | |
690 | SYSERROR("Failed to set environment variable: LXC_NAME=%s.", name); | |
691 | ||
692 | if (conf->rcfile && setenv("LXC_CONFIG_FILE", conf->rcfile, 1)) | |
693 | SYSERROR("Failed to set environment variable: LXC_CONFIG_FILE=%s.", conf->rcfile); | |
694 | ||
695 | if (conf->rootfs.mount && setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1)) | |
696 | SYSERROR("Failed to set environment variable: LXC_ROOTFS_MOUNT=%s.", conf->rootfs.mount); | |
697 | ||
698 | if (conf->rootfs.path && setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1)) | |
699 | SYSERROR("Failed to set environment variable: LXC_ROOTFS_PATH=%s.", conf->rootfs.path); | |
700 | ||
701 | if (conf->console.path && setenv("LXC_CONSOLE", conf->console.path, 1)) | |
702 | SYSERROR("Failed to set environment variable: LXC_CONSOLE=%s.", conf->console.path); | |
703 | ||
704 | if (conf->console.log_path && setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1)) | |
705 | SYSERROR("Failed to set environment variable: LXC_CONSOLE_LOGPATH=%s.", conf->console.log_path); | |
706 | ||
707 | if (setenv("LXC_CGNS_AWARE", "1", 1)) | |
708 | SYSERROR("Failed to set environment variable LXC_CGNS_AWARE=1."); | |
b8f88d9b | 709 | |
4a03ded4 CB |
710 | loglevel = lxc_log_priority_to_string(lxc_log_get_level()); |
711 | if (setenv("LXC_LOG_LEVEL", loglevel, 1)) | |
712 | SYSERROR("Failed to set environment variable LXC_LOG_LEVEL=%s", loglevel); | |
a2c09be0 CB |
713 | |
714 | if (conf->hooks_version == 0) | |
715 | ret = setenv("LXC_HOOK_VERSION", "0", 1); | |
716 | else | |
717 | ret = setenv("LXC_HOOK_VERSION", "1", 1); | |
718 | if (ret < 0) | |
719 | SYSERROR("Failed to set environment variable LXC_HOOK_VERSION=%u", conf->hooks_version); | |
408da065 | 720 | /* End of environment variable setup for hooks. */ |
f7bee6c6 | 721 | |
dbc9832d CB |
722 | TRACE("set environment variables"); |
723 | ||
14a7b0f9 | 724 | if (run_lxc_hooks(name, "pre-start", conf, NULL)) { |
408da065 | 725 | ERROR("Failed to run lxc.hook.pre-start for container \"%s\".", name); |
773fb9ca SH |
726 | goto out_aborting; |
727 | } | |
dbc9832d | 728 | TRACE("ran pre-start hooks"); |
26ddeedd | 729 | |
408da065 CB |
730 | /* The signal fd has to be created before forking otherwise if the child |
731 | * process exits before we setup the signal fd, the event will be lost | |
732 | * and the command will be stuck. | |
733 | */ | |
83ee7875 | 734 | handler->sigfd = setup_signal_fd(&handler->oldmask); |
59eb99ba | 735 | if (handler->sigfd < 0) { |
408da065 | 736 | ERROR("Failed to setup SIGCHLD fd handler."); |
b5159817 DE |
737 | goto out_delete_tty; |
738 | } | |
dbc9832d | 739 | TRACE("set up signal fd"); |
b5159817 | 740 | |
408da065 | 741 | /* Do this after setting up signals since it might unblock SIGWINCH. */ |
4d1ffb0a CB |
742 | ret = lxc_console_create(conf); |
743 | if (ret < 0) { | |
744 | ERROR("Failed to create console"); | |
b5159817 | 745 | goto out_restore_sigmask; |
b0a33c1e | 746 | } |
4d1ffb0a | 747 | TRACE("Created console"); |
b0a33c1e | 748 | |
54117de5 | 749 | if (lxc_ttys_shift_ids(conf) < 0) { |
408da065 | 750 | ERROR("Failed to shift tty into container."); |
c4d10a05 SH |
751 | goto out_restore_sigmask; |
752 | } | |
dbc9832d | 753 | TRACE("shifted tty ids"); |
c4d10a05 | 754 | |
dbc9832d | 755 | INFO("container \"%s\" is initialized", name); |
aa460476 | 756 | return 0; |
59eb99ba | 757 | |
b5159817 DE |
758 | out_restore_sigmask: |
759 | sigprocmask(SIG_SETMASK, &handler->oldmask, NULL); | |
59eb99ba | 760 | out_delete_tty: |
fae349da | 761 | lxc_delete_tty(&conf->tty_info); |
59eb99ba | 762 | out_aborting: |
25c2aca5 | 763 | lxc_set_state(name, handler, ABORTING); |
d2e30e99 DE |
764 | out_close_maincmd_fd: |
765 | close(conf->maincmd_fd); | |
766 | conf->maincmd_fd = -1; | |
aa460476 | 767 | return -1; |
59eb99ba DL |
768 | } |
769 | ||
3b72c4a0 | 770 | void lxc_fini(const char *name, struct lxc_handler *handler) |
59eb99ba | 771 | { |
b3286b62 | 772 | int i, rc; |
18b3b9c1 | 773 | pid_t self; |
dbc9832d | 774 | struct lxc_list *cur, *next; |
dbc9832d | 775 | char *namespaces[LXC_NS_MAX + 1]; |
b3286b62 | 776 | size_t namespace_count = 0; |
b6b2b194 | 777 | |
408da065 CB |
778 | /* The STOPPING state is there for future cleanup code which can take |
779 | * awhile. | |
59eb99ba | 780 | */ |
25c2aca5 | 781 | lxc_set_state(name, handler, STOPPING); |
b6b2b194 | 782 | |
0059379f | 783 | self = lxc_raw_getpid(); |
b3286b62 | 784 | for (i = 0; i < LXC_NS_MAX; i++) { |
18b3b9c1 CB |
785 | if (handler->nsfd[i] < 0) |
786 | continue; | |
787 | ||
788 | if (handler->conf->hooks_version == 0) | |
789 | rc = asprintf(&namespaces[namespace_count], | |
790 | "%s:/proc/%d/fd/%d", ns_info[i].proc_name, | |
791 | self, handler->nsfd[i]); | |
792 | else | |
793 | rc = asprintf(&namespaces[namespace_count], | |
794 | "/proc/%d/fd/%d", self, handler->nsfd[i]); | |
795 | if (rc == -1) { | |
796 | SYSERROR("Failed to allocate memory."); | |
797 | break; | |
b3286b62 | 798 | } |
18b3b9c1 CB |
799 | |
800 | if (handler->conf->hooks_version == 0) { | |
801 | namespace_count++; | |
802 | continue; | |
803 | } | |
804 | ||
805 | rc = setenv(ns_info[i].env_name, namespaces[namespace_count], 1); | |
806 | if (rc < 0) | |
807 | SYSERROR("Failed to set environment variable %s=%s", | |
808 | ns_info[i].env_name, namespaces[namespace_count]); | |
809 | else | |
810 | TRACE("Set environment variable %s=%s", | |
811 | ns_info[i].env_name, namespaces[namespace_count]); | |
812 | ||
813 | namespace_count++; | |
b3286b62 WB |
814 | } |
815 | namespaces[namespace_count] = NULL; | |
c154af98 | 816 | |
408da065 CB |
817 | if (handler->conf->reboot && setenv("LXC_TARGET", "reboot", 1)) |
818 | SYSERROR("Failed to set environment variable: LXC_TARGET=reboot."); | |
819 | ||
820 | if (!handler->conf->reboot && setenv("LXC_TARGET", "stop", 1)) | |
821 | SYSERROR("Failed to set environment variable: LXC_TARGET=stop."); | |
c154af98 | 822 | |
18b3b9c1 | 823 | if (handler->conf->hooks_version == 0) |
14a7b0f9 | 824 | rc = run_lxc_hooks(name, "stop", handler->conf, namespaces); |
18b3b9c1 | 825 | else |
14a7b0f9 | 826 | rc = run_lxc_hooks(name, "stop", handler->conf, NULL); |
c154af98 | 827 | |
b3286b62 WB |
828 | while (namespace_count--) |
829 | free(namespaces[namespace_count]); | |
28d9e29e | 830 | |
4288b79f CB |
831 | cgroup_destroy(handler); |
832 | ||
d39b10eb | 833 | if (handler->conf->reboot == 0) { |
f893d898 CB |
834 | /* For all new state clients simply close the command socket. |
835 | * This will inform all state clients that the container is | |
836 | * STOPPED and also prevents a race between a open()/close() on | |
837 | * the command socket causing a new process to get ECONNREFUSED | |
838 | * because we haven't yet closed the command socket. | |
839 | */ | |
d39b10eb CB |
840 | close(handler->conf->maincmd_fd); |
841 | handler->conf->maincmd_fd = -1; | |
c3184275 CB |
842 | TRACE("Closed command socket"); |
843 | ||
844 | /* This function will try to connect to the legacy lxc-monitord | |
845 | * state server and only exists for backwards compatibility. | |
846 | */ | |
847 | lxc_monitor_send_state(name, STOPPED, handler->lxcpath); | |
848 | ||
849 | /* The command socket is closed so no one can acces the command | |
850 | * socket anymore so there's no need to lock it. | |
851 | */ | |
852 | handler->state = STOPPED; | |
853 | TRACE("Set container state to \"STOPPED\""); | |
f893d898 CB |
854 | } else { |
855 | lxc_set_state(name, handler, STOPPED); | |
d39b10eb | 856 | } |
4288b79f | 857 | |
14a7b0f9 | 858 | if (run_lxc_hooks(name, "post-stop", handler->conf, NULL)) { |
408da065 | 859 | ERROR("Failed to run lxc.hook.post-stop for container \"%s\".", name); |
f3787121 CB |
860 | if (handler->conf->reboot) { |
861 | WARN("Container will be stopped instead of rebooted."); | |
862 | handler->conf->reboot = 0; | |
408da065 CB |
863 | if (setenv("LXC_TARGET", "stop", 1)) |
864 | WARN("Failed to set environment variable: LXC_TARGET=stop."); | |
f3787121 CB |
865 | } |
866 | } | |
26ddeedd | 867 | |
408da065 | 868 | /* Reset mask set by setup_signal_fd. */ |
8f64a3f6 | 869 | if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) |
408da065 | 870 | WARN("Failed to restore signal mask."); |
8f64a3f6 | 871 | |
b5159817 | 872 | lxc_console_delete(&handler->conf->console); |
b2431939 | 873 | lxc_delete_tty(&handler->conf->tty_info); |
dbc9832d | 874 | |
dbc9832d CB |
875 | /* The command socket is now closed, no more state clients can register |
876 | * themselves from now on. So free the list of state clients. | |
877 | */ | |
d39b10eb CB |
878 | lxc_list_for_each_safe(cur, &handler->conf->state_clients, next) { |
879 | struct lxc_state_client *client = cur->elem; | |
880 | ||
881 | /* Keep state clients that want to be notified about reboots. */ | |
882 | if ((handler->conf->reboot > 0) && (client->states[RUNNING] == 2)) | |
883 | continue; | |
884 | ||
dbc9832d CB |
885 | /* close state client socket */ |
886 | close(client->clientfd); | |
887 | lxc_list_del(cur); | |
888 | free(cur->elem); | |
889 | free(cur); | |
890 | } | |
891 | ||
2c5f2ede | 892 | if (handler->conf->ephemeral == 1 && handler->conf->reboot != 1) |
28272964 | 893 | lxc_destroy_container_on_signal(handler, name); |
2c5f2ede | 894 | |
1cc8bd4b | 895 | lxc_free_handler(handler); |
59eb99ba DL |
896 | } |
897 | ||
735f2c6e | 898 | void lxc_abort(const char *name, struct lxc_handler *handler) |
59eb99ba | 899 | { |
73e608b2 SH |
900 | int ret, status; |
901 | ||
25c2aca5 | 902 | lxc_set_state(name, handler, ABORTING); |
7d9fb3e9 DL |
903 | if (handler->pid > 0) |
904 | kill(handler->pid, SIGKILL); | |
408da065 CB |
905 | while ((ret = waitpid(-1, &status, 0)) > 0) { |
906 | ; | |
907 | } | |
59eb99ba DL |
908 | } |
909 | ||
18225d19 CB |
910 | static int lxc_set_death_signal(int signal) |
911 | { | |
912 | int ret; | |
913 | pid_t ppid; | |
914 | ||
915 | ret = prctl(PR_SET_PDEATHSIG, signal, 0, 0, 0); | |
916 | ||
917 | /* Check whether we have been orphaned. */ | |
918 | ppid = (pid_t)syscall(SYS_getppid); | |
919 | if (ppid == 1) { | |
920 | pid_t self; | |
921 | ||
922 | self = lxc_raw_getpid(); | |
923 | ret = kill(self, SIGKILL); | |
924 | if (ret < 0) | |
925 | return -1; | |
926 | } | |
927 | ||
928 | if (ret < 0) { | |
929 | SYSERROR("Failed to set PR_SET_PDEATHSIG to %d", signal); | |
930 | return -1; | |
931 | } | |
932 | ||
933 | return 0; | |
934 | } | |
935 | ||
ffe1e01a | 936 | static int do_start(void *data) |
50e98013 | 937 | { |
8bf3abfb | 938 | int ret; |
7c661726 | 939 | struct lxc_list *iterator; |
7a55c157 | 940 | char path[PATH_MAX]; |
fa3a5b22 | 941 | struct lxc_handler *handler = data; |
928b1f04 YT |
942 | bool have_cap_setgid; |
943 | uid_t new_uid; | |
944 | gid_t new_gid; | |
5af9369b | 945 | int devnull_fd = -1; |
50e98013 | 946 | |
18225d19 | 947 | lxc_sync_fini_parent(handler); |
50e98013 | 948 | |
408da065 CB |
949 | /* This prctl must be before the synchro, so if the parent dies before |
950 | * we set the parent death signal, we will detect its death with the | |
951 | * synchro right after, otherwise we have a window where the parent can | |
952 | * exit before we set the pdeath signal leading to a unsupervized | |
953 | * container. | |
743ecd2e | 954 | */ |
18225d19 | 955 | ret = lxc_set_death_signal(SIGKILL); |
912314fc CB |
956 | if (ret < 0) { |
957 | SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL"); | |
18225d19 | 958 | goto out_warn_father; |
743ecd2e DL |
959 | } |
960 | ||
18225d19 CB |
961 | ret = sigprocmask(SIG_SETMASK, &handler->oldmask, NULL); |
962 | if (ret < 0) { | |
963 | SYSERROR("Failed to set signal mask"); | |
964 | goto out_warn_father; | |
965 | } | |
50e98013 | 966 | |
408da065 | 967 | /* Don't leak the pinfd to the container. */ |
d39b10eb | 968 | if (handler->pinfd >= 0) |
0d03360a | 969 | close(handler->pinfd); |
2b0e17e4 | 970 | |
18225d19 CB |
971 | ret = lxc_sync_wait_parent(handler, LXC_SYNC_STARTUP); |
972 | if (ret < 0) | |
973 | goto out_warn_father; | |
5b1e83cb | 974 | |
408da065 CB |
975 | /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See |
976 | * https://github.com/lxc/lxd/issues/1978. | |
977 | */ | |
5b1e83cb | 978 | if ((handler->clone_flags & (CLONE_NEWNET | CLONE_NEWUSER)) == |
408da065 | 979 | (CLONE_NEWNET | CLONE_NEWUSER)) { |
5b1e83cb SH |
980 | ret = unshare(CLONE_NEWNET); |
981 | if (ret < 0) { | |
408da065 | 982 | SYSERROR("Failed to unshare CLONE_NEWNET."); |
5b1e83cb SH |
983 | goto out_warn_father; |
984 | } | |
408da065 | 985 | INFO("Unshared CLONE_NEWNET."); |
5b1e83cb SH |
986 | } |
987 | ||
408da065 CB |
988 | /* Tell the parent task it can begin to configure the container and wait |
989 | * for it to finish. | |
3c22086f CLG |
990 | */ |
991 | if (lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE)) | |
9d7f9e52 | 992 | return -1; |
50e98013 | 993 | |
7ab1ba02 CB |
994 | if (lxc_network_recv_veth_names_from_parent(handler) < 0) { |
995 | ERROR("Failed to receive veth names from parent"); | |
658979c5 | 996 | goto out_warn_father; |
7ab1ba02 | 997 | } |
658979c5 | 998 | |
408da065 | 999 | /* If we are in a new user namespace, become root there to have |
d08f8d2f | 1000 | * privilege over our namespace. |
f6d3e3e4 SH |
1001 | */ |
1002 | if (!lxc_list_empty(&handler->conf->id_map)) { | |
4160c3a0 CB |
1003 | uid_t nsuid = (handler->conf->root_nsuid_map != NULL) |
1004 | ? 0 | |
1005 | : handler->conf->init_uid; | |
1006 | gid_t nsgid = (handler->conf->root_nsgid_map != NULL) | |
1007 | ? 0 | |
1008 | : handler->conf->init_gid; | |
1009 | ||
1010 | ret = lxc_switch_uid_gid(nsuid, nsgid); | |
4b826b1f | 1011 | if (ret < 0) |
f6d3e3e4 | 1012 | goto out_warn_father; |
d08f8d2f CB |
1013 | |
1014 | /* Drop groups only after we switched to a valid gid in the new | |
1015 | * user namespace. | |
1016 | */ | |
4b826b1f | 1017 | ret = lxc_setgroups(0, NULL); |
a19b974f | 1018 | if (ret < 0 && (handler->am_root || errno != EPERM)) |
c476bdce | 1019 | goto out_warn_father; |
4b826b1f CB |
1020 | |
1021 | if (!handler->am_root) { | |
1022 | ret = prctl(PR_SET_DUMPABLE, 1, 0, 0, 0); | |
1023 | if (ret < 0) | |
1024 | goto out_warn_father; | |
1025 | } | |
912314fc CB |
1026 | |
1027 | /* set{g,u}id() clears deathsignal */ | |
18225d19 | 1028 | ret = lxc_set_death_signal(SIGKILL); |
912314fc CB |
1029 | if (ret < 0) { |
1030 | SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL"); | |
1031 | goto out_warn_father; | |
1032 | } | |
f6d3e3e4 SH |
1033 | } |
1034 | ||
99b71824 | 1035 | if (access(handler->lxcpath, X_OK)) { |
c8154066 SH |
1036 | print_top_failing_dir(handler->lxcpath); |
1037 | goto out_warn_father; | |
1038 | } | |
1039 | ||
a91cde21 | 1040 | ret = snprintf(path, sizeof(path), "%s/dev/null", handler->conf->rootfs.mount); |
408da065 | 1041 | if (ret < 0 || ret >= sizeof(path)) |
7a55c157 | 1042 | goto out_warn_father; |
7a55c157 TA |
1043 | |
1044 | /* In order to checkpoint restore, we need to have everything in the | |
1045 | * same mount namespace. However, some containers may not have a | |
1046 | * reasonable /dev (in particular, they may not have /dev/null), so we | |
1047 | * can't set init's std fds to /dev/null by opening it from inside the | |
1048 | * container. | |
1049 | * | |
1050 | * If that's the case, fall back to using the host's /dev/null. This | |
1051 | * means that migration won't work, but at least we won't spew output | |
1052 | * where it isn't wanted. | |
1053 | */ | |
1054 | if (handler->backgrounded && !handler->conf->autodev && access(path, F_OK) < 0) { | |
c44de748 AM |
1055 | devnull_fd = open_devnull(); |
1056 | ||
1057 | if (devnull_fd < 0) | |
1058 | goto out_warn_father; | |
408da065 CB |
1059 | WARN("Using /dev/null from the host for container init's " |
1060 | "standard file descriptors. Migration will not work."); | |
c44de748 AM |
1061 | } |
1062 | ||
408da065 | 1063 | /* Ask father to setup cgroups and wait for him to finish. */ |
544a48a0 | 1064 | if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP)) |
c44de748 | 1065 | goto out_error; |
544a48a0 | 1066 | |
deefdf8a CB |
1067 | /* Unshare cgroup namespace after we have setup our cgroups. If we do it |
1068 | * earlier we end up with a wrong view of /proc/self/cgroup. For | |
1069 | * example, assume we unshare(CLONE_NEWCGROUP) first, and then create | |
1070 | * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then | |
1071 | * /proc/self/cgroup would show us: | |
1072 | * | |
1073 | * 8:cpuset:/lxc/c | |
1074 | * | |
1075 | * whereas it should actually show | |
1076 | * | |
1077 | * 8:cpuset:/ | |
1078 | */ | |
5af9369b | 1079 | if (handler->clone_flags & CLONE_NEWCGROUP) { |
deefdf8a CB |
1080 | if (unshare(CLONE_NEWCGROUP) < 0) { |
1081 | INFO("Failed to unshare CLONE_NEWCGROUP."); | |
1082 | goto out_warn_father; | |
1083 | } | |
1084 | INFO("Unshared CLONE_NEWCGROUP."); | |
1085 | } | |
1086 | ||
149857af CB |
1087 | /* Add the requested environment variables to the current environment to |
1088 | * allow them to be used by the various hooks, such as the start hook | |
1089 | * above. | |
98ff08ca | 1090 | */ |
98ff08ca CB |
1091 | lxc_list_for_each(iterator, &handler->conf->environment) { |
1092 | if (putenv((char *)iterator->elem)) { | |
1093 | SYSERROR("Failed to set environment variable: %s.", (char *)iterator->elem); | |
1094 | goto out_warn_father; | |
1095 | } | |
1096 | } | |
1097 | ||
f4152036 | 1098 | /* Setup the container, ip, names, utsname, ... */ |
3b988b33 | 1099 | ret = lxc_setup(handler); |
7729f8e5 | 1100 | close(handler->data_sock[1]); |
8bf3abfb | 1101 | close(handler->data_sock[0]); |
7729f8e5 | 1102 | if (ret < 0) { |
f4152036 CB |
1103 | ERROR("Failed to setup container \"%s\".", handler->name); |
1104 | goto out_warn_father; | |
1105 | } | |
1106 | ||
408da065 | 1107 | /* Set the label to change to when we exec(2) the container's init. */ |
7aff4f43 | 1108 | if (lsm_process_label_set(NULL, handler->conf, 1, 1) < 0) |
e075f5d9 | 1109 | goto out_warn_father; |
5112cd70 | 1110 | |
029cdff5 | 1111 | /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it |
408da065 CB |
1112 | * before we aren't allowed anymore. |
1113 | */ | |
029cdff5 CB |
1114 | if (handler->conf->no_new_privs) { |
1115 | if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) { | |
1116 | SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges."); | |
1117 | goto out_warn_father; | |
1118 | } | |
1119 | DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges."); | |
1120 | } | |
1121 | ||
0d9acb99 DE |
1122 | /* Some init's such as busybox will set sane tty settings on stdin, |
1123 | * stdout, stderr which it thinks is the console. We already set them | |
1124 | * the way we wanted on the real terminal, and we want init to do its | |
408da065 CB |
1125 | * setup on its console ie. the pty allocated in lxc_console_create() so |
1126 | * make sure that that pty is stdin,stdout,stderr. | |
0d9acb99 | 1127 | */ |
5d113f65 CB |
1128 | if (handler->conf->console.slave >= 0) { |
1129 | if (handler->backgrounded || handler->conf->is_execute == 0) | |
1130 | ret = set_stdfds(handler->conf->console.slave); | |
1131 | else | |
1132 | ret = lxc_console_set_stdfds(handler->conf->console.slave); | |
1133 | if (ret < 0) { | |
c5b93afb LF |
1134 | ERROR("Failed to redirect std{in,out,err} to pty file " |
1135 | "descriptor %d", | |
1136 | handler->conf->console.slave); | |
1137 | goto out_warn_father; | |
1138 | } | |
5d113f65 | 1139 | } |
0d9acb99 | 1140 | |
408da065 | 1141 | /* If we mounted a temporary proc, then unmount it now. */ |
5112cd70 | 1142 | tmp_proc_unmount(handler->conf); |
e075f5d9 | 1143 | |
8f2c3a70 SH |
1144 | if (lxc_seccomp_load(handler->conf) != 0) |
1145 | goto out_warn_father; | |
1146 | ||
14a7b0f9 | 1147 | if (run_lxc_hooks(handler->name, "start", handler->conf, NULL)) { |
408da065 | 1148 | ERROR("Failed to run lxc.hook.start for container \"%s\".", handler->name); |
773fb9ca SH |
1149 | goto out_warn_father; |
1150 | } | |
fc25b815 | 1151 | |
773fb9ca | 1152 | close(handler->sigfd); |
26ddeedd | 1153 | |
7a55c157 TA |
1154 | if (devnull_fd < 0) { |
1155 | devnull_fd = open_devnull(); | |
1156 | ||
1157 | if (devnull_fd < 0) | |
1158 | goto out_warn_father; | |
1159 | } | |
1160 | ||
c5b93afb LF |
1161 | if (handler->conf->console.slave < 0 && handler->backgrounded) |
1162 | if (set_stdfds(devnull_fd) < 0) { | |
1163 | ERROR("Failed to redirect std{in,out,err} to " | |
1164 | "\"/dev/null\""); | |
1165 | goto out_warn_father; | |
1166 | } | |
507cee36 | 1167 | |
c44de748 AM |
1168 | if (devnull_fd >= 0) { |
1169 | close(devnull_fd); | |
1170 | devnull_fd = -1; | |
1171 | } | |
1172 | ||
8c9a7665 TA |
1173 | setsid(); |
1174 | ||
3c491553 L |
1175 | if (handler->conf->init_cwd && chdir(handler->conf->init_cwd)) { |
1176 | SYSERROR("Could not change directory to \"%s\"", handler->conf->init_cwd); | |
1177 | goto out_warn_father; | |
1178 | } | |
1179 | ||
f4152036 CB |
1180 | if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP_LIMITS)) |
1181 | goto out_warn_father; | |
1182 | ||
149857af CB |
1183 | /* Reset the environment variables the user requested in a clear |
1184 | * environment. | |
1185 | */ | |
1186 | if (clearenv()) { | |
1187 | SYSERROR("Failed to clear environment."); | |
1188 | /* Don't error out though. */ | |
1189 | } | |
1190 | ||
1191 | lxc_list_for_each(iterator, &handler->conf->environment) { | |
1192 | if (putenv((char *)iterator->elem)) { | |
1193 | SYSERROR("Failed to set environment variable: %s.", (char *)iterator->elem); | |
1194 | goto out_warn_father; | |
1195 | } | |
1196 | } | |
1197 | ||
98ff08ca CB |
1198 | if (putenv("container=lxc")) { |
1199 | SYSERROR("Failed to set environment variable: container=lxc."); | |
1200 | goto out_warn_father; | |
1201 | } | |
1202 | ||
1203 | if (handler->conf->pty_names) { | |
1204 | if (putenv(handler->conf->pty_names)) { | |
1205 | SYSERROR("Failed to set environment variable for container ptys."); | |
1206 | goto out_warn_father; | |
1207 | } | |
1208 | } | |
1209 | ||
76bdf299 CB |
1210 | /* The container has been setup. We can now switch to an unprivileged |
1211 | * uid/gid. | |
1212 | */ | |
928b1f04 YT |
1213 | new_uid = handler->conf->init_uid; |
1214 | new_gid = handler->conf->init_gid; | |
76bdf299 | 1215 | |
928b1f04 YT |
1216 | /* If we are in a new user namespace we already dropped all |
1217 | * groups when we switched to root in the new user namespace | |
1218 | * further above. Only drop groups if we can, so ensure that we | |
1219 | * have necessary privilege. | |
1220 | */ | |
1221 | #if HAVE_LIBCAP | |
1222 | have_cap_setgid = lxc_proc_cap_is_set(CAP_SETGID, CAP_EFFECTIVE); | |
1223 | #else | |
1224 | have_cap_setgid = false; | |
1225 | #endif | |
1226 | if (lxc_list_empty(&handler->conf->id_map) && have_cap_setgid) { | |
1227 | if (lxc_setgroups(0, NULL) < 0) | |
76bdf299 CB |
1228 | goto out_warn_father; |
1229 | } | |
1230 | ||
928b1f04 YT |
1231 | if (lxc_switch_uid_gid(new_uid, new_gid) < 0) |
1232 | goto out_warn_father; | |
1233 | ||
408da065 CB |
1234 | /* After this call, we are in error because this ops should not return |
1235 | * as it execs. | |
1236 | */ | |
c4ea60df | 1237 | handler->ops->start(handler, handler->data); |
50e98013 DL |
1238 | |
1239 | out_warn_father: | |
408da065 CB |
1240 | /* We want the parent to know something went wrong, so we return a |
1241 | * special error code. | |
1242 | */ | |
d1ccb562 | 1243 | lxc_sync_wake_parent(handler, LXC_SYNC_ERROR); |
c44de748 AM |
1244 | |
1245 | out_error: | |
1246 | if (devnull_fd >= 0) | |
1247 | close(devnull_fd); | |
1248 | ||
50e98013 DL |
1249 | return -1; |
1250 | } | |
1251 | ||
ae467c54 | 1252 | static int lxc_recv_ttys_from_child(struct lxc_handler *handler) |
e8bd4e43 | 1253 | { |
ae467c54 | 1254 | int i; |
ae467c54 CB |
1255 | struct lxc_pty_info *pty_info; |
1256 | int ret = -1; | |
c6012571 | 1257 | int sock = handler->data_sock[1]; |
e8bd4e43 | 1258 | struct lxc_conf *conf = handler->conf; |
e8bd4e43 SH |
1259 | struct lxc_tty_info *tty_info = &conf->tty_info; |
1260 | ||
1261 | if (!conf->tty) | |
1262 | return 0; | |
1263 | ||
408da065 CB |
1264 | tty_info->pty_info = malloc(sizeof(*tty_info->pty_info) * conf->tty); |
1265 | if (!tty_info->pty_info) | |
e8bd4e43 | 1266 | return -1; |
e8bd4e43 | 1267 | |
672c1e58 CB |
1268 | for (i = 0; i < conf->tty; i++) { |
1269 | int ttyfds[2]; | |
1270 | ||
1271 | ret = lxc_abstract_unix_recv_fds(sock, ttyfds, 2, NULL, 0); | |
1272 | if (ret < 0) | |
1273 | break; | |
ae467c54 | 1274 | |
672c1e58 | 1275 | pty_info = &tty_info->pty_info[i]; |
e8bd4e43 | 1276 | pty_info->busy = 0; |
672c1e58 CB |
1277 | pty_info->master = ttyfds[0]; |
1278 | pty_info->slave = ttyfds[1]; | |
1279 | TRACE("Received pty with master fd %d and slave fd %d from " | |
ae467c54 | 1280 | "parent", pty_info->master, pty_info->slave); |
e8bd4e43 | 1281 | } |
ae467c54 | 1282 | if (ret < 0) |
672c1e58 | 1283 | ERROR("Failed to receive %d ttys from child: %s", conf->tty, |
ae467c54 CB |
1284 | strerror(errno)); |
1285 | else | |
672c1e58 CB |
1286 | TRACE("Received %d ttys from child", conf->tty); |
1287 | ||
1288 | tty_info->nbtty = conf->tty; | |
ae467c54 CB |
1289 | |
1290 | return ret; | |
e8bd4e43 SH |
1291 | } |
1292 | ||
5af9369b | 1293 | int resolve_clone_flags(struct lxc_handler *handler) |
f813849c | 1294 | { |
03df7ab5 | 1295 | handler->clone_flags = CLONE_NEWNS; |
f813849c | 1296 | |
fa3a5b22 CB |
1297 | if (!handler->conf->inherit_ns[LXC_NS_USER]) { |
1298 | if (!lxc_list_empty(&handler->conf->id_map)) | |
1299 | handler->clone_flags |= CLONE_NEWUSER; | |
1300 | } else { | |
1301 | INFO("Inheriting user namespace"); | |
1302 | } | |
f813849c | 1303 | |
28d9e29e | 1304 | if (!handler->conf->inherit_ns[LXC_NS_NET]) { |
408da065 | 1305 | if (!lxc_requests_empty_network(handler)) |
f813849c TA |
1306 | handler->clone_flags |= CLONE_NEWNET; |
1307 | } else { | |
28d9e29e | 1308 | INFO("Inheriting net namespace"); |
f813849c TA |
1309 | } |
1310 | ||
28d9e29e | 1311 | if (!handler->conf->inherit_ns[LXC_NS_IPC]) |
f813849c | 1312 | handler->clone_flags |= CLONE_NEWIPC; |
408da065 | 1313 | else |
28d9e29e | 1314 | INFO("Inheriting ipc namespace"); |
f813849c | 1315 | |
28d9e29e | 1316 | if (!handler->conf->inherit_ns[LXC_NS_UTS]) |
f813849c | 1317 | handler->clone_flags |= CLONE_NEWUTS; |
408da065 | 1318 | else |
28d9e29e | 1319 | INFO("Inheriting uts namespace"); |
03df7ab5 | 1320 | |
28d9e29e | 1321 | if (!handler->conf->inherit_ns[LXC_NS_PID]) |
03df7ab5 L |
1322 | handler->clone_flags |= CLONE_NEWPID; |
1323 | else | |
28d9e29e | 1324 | INFO("Inheriting pid namespace"); |
5af9369b CB |
1325 | |
1326 | if (cgns_supported()) { | |
1327 | if (!handler->conf->inherit_ns[LXC_NS_CGROUP]) | |
1328 | handler->clone_flags |= CLONE_NEWCGROUP; | |
1329 | else | |
1330 | INFO("Inheriting cgroup namespace"); | |
1331 | } else if (handler->conf->inherit_ns[LXC_NS_CGROUP]) { | |
1332 | return -EINVAL; | |
1333 | } | |
1334 | ||
1335 | return 0; | |
f813849c TA |
1336 | } |
1337 | ||
8deca6c9 CB |
1338 | /* Note that this function is used with clone(CLONE_VM). Some glibc versions |
1339 | * used to reset the pid/tid to -1 when CLONE_VM was used without CLONE_THREAD. | |
1340 | * But since the memory between parent and child is shared on CLONE_VM this | |
1341 | * would invalidate the getpid() cache that glibc used to maintain and so | |
1342 | * getpid() in the child would return the parent's pid. This is all fixed in | |
1343 | * newer glibc versions where the getpid() cache is removed and the pid/tid is | |
1344 | * not reset anymore. | |
1345 | * However, if for whatever reason you - dear commiter - somehow need to get the | |
1346 | * pid of the dummy intermediate process for do_share_ns() you need to call | |
0059379f CB |
1347 | * lxc_raw_getpid(). The next lxc_raw_clone() call does not employ CLONE_VM and |
1348 | * will be fine. | |
8deca6c9 CB |
1349 | */ |
1350 | static inline int do_share_ns(void *arg) | |
fa3a5b22 | 1351 | { |
8deca6c9 | 1352 | int i, flags, ret; |
fa3a5b22 CB |
1353 | struct lxc_handler *handler = arg; |
1354 | ||
8deca6c9 CB |
1355 | for (i = 0; i < LXC_NS_MAX; i++) { |
1356 | if (handler->nsfd[i] < 0) | |
1357 | continue; | |
fa3a5b22 | 1358 | |
8deca6c9 | 1359 | ret = setns(handler->nsfd[i], 0); |
fa3a5b22 | 1360 | if (ret < 0) |
8deca6c9 | 1361 | return -1; |
fa3a5b22 | 1362 | |
8deca6c9 | 1363 | DEBUG("Inherited %s namespace", ns_info[i].proc_name); |
fa3a5b22 CB |
1364 | } |
1365 | ||
8deca6c9 CB |
1366 | flags = handler->on_clone_flags; |
1367 | flags |= CLONE_PARENT; | |
ac2ba696 | 1368 | handler->pid = lxc_raw_clone_cb(do_start, handler, flags); |
8deca6c9 | 1369 | if (handler->pid < 0) |
fa3a5b22 CB |
1370 | return -1; |
1371 | ||
8deca6c9 | 1372 | return 0; |
fa3a5b22 CB |
1373 | } |
1374 | ||
480588e6 CB |
1375 | /* lxc_spawn() performs crucial setup tasks and clone()s the new process which |
1376 | * exec()s the requested container binary. | |
1377 | * Note that lxc_spawn() runs in the parent namespaces. Any operations performed | |
1378 | * right here should be double checked if they'd pose a security risk. (For | |
1379 | * example, any {u}mount() operations performed here will be reflected on the | |
1380 | * host!) | |
1381 | */ | |
74a3920a | 1382 | static int lxc_spawn(struct lxc_handler *handler) |
59eb99ba | 1383 | { |
8deca6c9 | 1384 | int i, ret; |
08dd2805 | 1385 | char pidstr[20]; |
57927bf2 CB |
1386 | bool wants_to_map_ids; |
1387 | struct lxc_list *id_map; | |
6e5fc7a5 | 1388 | const char *name = handler->name; |
28d9e29e | 1389 | const char *lxcpath = handler->lxcpath; |
8deca6c9 | 1390 | bool cgroups_connected = false, share_ns = false; |
28d9e29e | 1391 | struct lxc_conf *conf = handler->conf; |
9f30a190 | 1392 | |
28d9e29e | 1393 | id_map = &conf->id_map; |
57927bf2 | 1394 | wants_to_map_ids = !lxc_list_empty(id_map); |
2f2623ec | 1395 | |
28d9e29e CB |
1396 | for (i = 0; i < LXC_NS_MAX; i++) { |
1397 | if (!conf->inherit_ns[i]) | |
1398 | continue; | |
1399 | ||
28d9e29e CB |
1400 | handler->nsfd[i] = lxc_inherit_namespace(conf->inherit_ns[i], lxcpath, ns_info[i].proc_name); |
1401 | if (handler->nsfd[i] < 0) | |
1402 | return -1; | |
fa3a5b22 | 1403 | |
8deca6c9 | 1404 | share_ns = true; |
28d9e29e | 1405 | } |
50e98013 | 1406 | |
3c22086f | 1407 | if (lxc_sync_init(handler)) |
9d7f9e52 | 1408 | return -1; |
0ad19a3f | 1409 | |
b9f522c5 | 1410 | ret = socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0, |
790255cf | 1411 | handler->data_sock); |
c6012571 | 1412 | if (ret < 0) { |
e8bd4e43 SH |
1413 | lxc_sync_fini(handler); |
1414 | return -1; | |
1415 | } | |
1416 | ||
5af9369b CB |
1417 | ret = resolve_clone_flags(handler); |
1418 | if (ret < 0) { | |
1419 | lxc_sync_fini(handler); | |
1420 | return -1; | |
1421 | } | |
9f30a190 | 1422 | |
f813849c | 1423 | if (handler->clone_flags & CLONE_NEWNET) { |
28d9e29e | 1424 | if (!lxc_list_empty(&conf->network)) { |
26b797f3 | 1425 | |
9f30a190 MM |
1426 | /* Find gateway addresses from the link device, which is |
1427 | * no longer accessible inside the container. Do this | |
1428 | * before creating network interfaces, since goto | |
408da065 CB |
1429 | * out_delete_net does not work before lxc_clone. |
1430 | */ | |
9f30a190 | 1431 | if (lxc_find_gateway_addresses(handler)) { |
408da065 | 1432 | ERROR("Failed to find gateway addresses."); |
9f30a190 MM |
1433 | lxc_sync_fini(handler); |
1434 | return -1; | |
1435 | } | |
1436 | ||
408da065 CB |
1437 | /* That should be done before the clone because we will |
1438 | * fill the netdev index and use them in the child. | |
9f30a190 | 1439 | */ |
811ef482 | 1440 | if (lxc_create_network_priv(handler)) { |
408da065 | 1441 | ERROR("Failed to create the network."); |
9f30a190 MM |
1442 | lxc_sync_fini(handler); |
1443 | return -1; | |
1444 | } | |
19a26f82 | 1445 | } |
82d5ae15 DL |
1446 | } |
1447 | ||
d4ef7c50 | 1448 | if (!cgroup_init(handler)) { |
408da065 | 1449 | ERROR("Failed initializing cgroup support."); |
33ad9f1a CS |
1450 | goto out_delete_net; |
1451 | } | |
1452 | ||
7e4dfe0b SH |
1453 | cgroups_connected = true; |
1454 | ||
d4ef7c50 | 1455 | if (!cgroup_create(handler)) { |
408da065 | 1456 | ERROR("Failed creating cgroups."); |
47d8fb3b CS |
1457 | goto out_delete_net; |
1458 | } | |
1459 | ||
408da065 CB |
1460 | /* If the rootfs is not a blockdev, prevent the container from marking |
1461 | * it readonly. | |
1462 | * If the container is unprivileged then skip rootfs pinning. | |
0c547523 | 1463 | */ |
0ee35059 | 1464 | if (!wants_to_map_ids) { |
28d9e29e | 1465 | handler->pinfd = pin_rootfs(conf->rootfs.path); |
5e32a990 | 1466 | if (handler->pinfd == -1) |
408da065 | 1467 | INFO("Failed to pin the rootfs for container \"%s\".", handler->name); |
5e32a990 | 1468 | } |
0c547523 | 1469 | |
408da065 | 1470 | /* Create a process in a new set of namespaces. */ |
8deca6c9 | 1471 | handler->on_clone_flags = handler->clone_flags; |
408da065 CB |
1472 | if (handler->clone_flags & CLONE_NEWUSER) { |
1473 | /* If CLONE_NEWUSER and CLONE_NEWNET was requested, we need to | |
1474 | * clone a new user namespace first and only later unshare our | |
1475 | * network namespace to ensure that network devices ownership is | |
1476 | * set up correctly. | |
1477 | */ | |
8deca6c9 | 1478 | handler->on_clone_flags &= ~CLONE_NEWNET; |
408da065 | 1479 | } |
5af9369b CB |
1480 | /* The cgroup namespace gets unshare()ed not clone()ed. */ |
1481 | handler->on_clone_flags &= ~CLONE_NEWCGROUP; | |
732375f5 | 1482 | |
4e232466 CB |
1483 | if (share_ns) { |
1484 | pid_t attacher_pid; | |
1485 | ||
1486 | attacher_pid = lxc_clone(do_share_ns, handler, CLONE_VFORK | CLONE_VM | CLONE_FILES); | |
1487 | if (attacher_pid < 0) { | |
1488 | SYSERROR(LXC_CLONE_ERROR); | |
1489 | goto out_delete_net; | |
1490 | } | |
1491 | ||
1492 | ret = wait_for_pid(attacher_pid); | |
1493 | if (ret < 0) { | |
1494 | SYSERROR("Intermediate process failed"); | |
1495 | goto out_delete_net; | |
1496 | } | |
1497 | } else { | |
ac2ba696 | 1498 | handler->pid = lxc_raw_clone_cb(do_start, handler, handler->on_clone_flags); |
4e232466 CB |
1499 | } |
1500 | if (handler->pid < 0) { | |
1501 | SYSERROR(LXC_CLONE_ERROR); | |
7fef7a06 | 1502 | goto out_delete_net; |
0ad19a3f | 1503 | } |
fa3a5b22 | 1504 | TRACE("Cloned child process %d", handler->pid); |
1bd8d726 | 1505 | |
9662e444 | 1506 | for (i = 0; i < LXC_NS_MAX; i++) |
8deca6c9 | 1507 | if (handler->on_clone_flags & ns_info[i].clone_flag) |
28d9e29e | 1508 | INFO("Cloned %s", ns_info[i].flag_name); |
0ad19a3f | 1509 | |
8bf3abfb | 1510 | if (!preserve_ns(handler->nsfd, handler->on_clone_flags, handler->pid)) { |
28d9e29e CB |
1511 | ERROR("Failed to preserve cloned namespaces for lxc.hook.stop"); |
1512 | goto out_delete_net; | |
1513 | } | |
b6b2b194 | 1514 | |
3c22086f CLG |
1515 | lxc_sync_fini_child(handler); |
1516 | ||
408da065 CB |
1517 | /* Map the container uids. The container became an invalid userid the |
1518 | * moment it was cloned with CLONE_NEWUSER. This call doesn't change | |
1519 | * anything immediately, but allows the container to setuid(0) (0 being | |
1520 | * mapped to something else on the host.) later to become a valid uid | |
1521 | * again. | |
1522 | */ | |
fa3a5b22 CB |
1523 | if (wants_to_map_ids) { |
1524 | if (!handler->conf->inherit_ns[LXC_NS_USER]) { | |
1525 | ret = lxc_map_ids(id_map, handler->pid); | |
1526 | if (ret < 0) { | |
1527 | ERROR("Failed to set up id mapping."); | |
1528 | goto out_delete_net; | |
1529 | } | |
1530 | } | |
5b1e83cb SH |
1531 | } |
1532 | ||
9e2bdae4 | 1533 | if (lxc_sync_wake_child(handler, LXC_SYNC_STARTUP)) |
5b1e83cb | 1534 | goto out_delete_net; |
5b1e83cb | 1535 | |
9e2bdae4 | 1536 | if (lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE)) |
5b1e83cb | 1537 | goto out_delete_net; |
0ad19a3f | 1538 | |
d4ef7c50 | 1539 | if (!cgroup_create_legacy(handler)) { |
408da065 | 1540 | ERROR("Failed to setup legacy cgroups for container \"%s\".", name); |
ae5c8b8e | 1541 | goto out_delete_net; |
33ad9f1a | 1542 | } |
9daf6f5d | 1543 | if (!cgroup_setup_limits(handler, false)) { |
408da065 | 1544 | ERROR("Failed to setup cgroup limits for container \"%s\".", name); |
6031a6e5 DE |
1545 | goto out_delete_net; |
1546 | } | |
1547 | ||
d4ef7c50 | 1548 | if (!cgroup_enter(handler)) |
7fef7a06 | 1549 | goto out_delete_net; |
218d4250 | 1550 | |
0996e18a SH |
1551 | if (!cgroup_chown(handler)) |
1552 | goto out_delete_net; | |
1553 | ||
fa3a5b22 CB |
1554 | /* Now we're ready to preserve the network namespace */ |
1555 | ret = lxc_preserve_ns(handler->pid, "net"); | |
1556 | if (ret < 0) { | |
1557 | ERROR("%s - Failed to preserve net namespace", strerror(errno)); | |
1558 | goto out_delete_net; | |
1559 | } | |
1560 | handler->nsfd[LXC_NS_NET] = ret; | |
1561 | DEBUG("Preserved net namespace via fd %d", ret); | |
1562 | ||
408da065 | 1563 | /* Create the network configuration. */ |
d5088cf2 | 1564 | if (handler->clone_flags & CLONE_NEWNET) { |
74c6e2b0 CB |
1565 | if (lxc_network_move_created_netdev_priv(handler->lxcpath, |
1566 | handler->name, | |
28d9e29e | 1567 | &conf->network, |
74c6e2b0 | 1568 | handler->pid)) { |
408da065 | 1569 | ERROR("Failed to create the configured network."); |
7fef7a06 | 1570 | goto out_delete_net; |
82d5ae15 | 1571 | } |
74c6e2b0 CB |
1572 | |
1573 | if (lxc_create_network_unpriv(handler->lxcpath, handler->name, | |
28d9e29e | 1574 | &conf->network, |
74c6e2b0 CB |
1575 | handler->pid)) { |
1576 | ERROR("Failed to create the configured network."); | |
1577 | goto out_delete_net; | |
1578 | } | |
0ad19a3f | 1579 | } |
1580 | ||
7ab1ba02 CB |
1581 | if (lxc_network_send_veth_names_to_child(handler) < 0) { |
1582 | ERROR("Failed to send veth names to child"); | |
1583 | goto out_delete_net; | |
658979c5 SH |
1584 | } |
1585 | ||
61d7a733 YT |
1586 | if (!lxc_list_empty(&conf->procs)) { |
1587 | ret = setup_proc_filesystem(&conf->procs, handler->pid); | |
1588 | if (ret < 0) | |
1589 | goto out_delete_net; | |
1590 | } | |
1591 | ||
408da065 CB |
1592 | /* Tell the child to continue its initialization. We'll get |
1593 | * LXC_SYNC_CGROUP when it is ready for us to setup cgroups. | |
3c22086f CLG |
1594 | */ |
1595 | if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE)) | |
544a48a0 SH |
1596 | goto out_delete_net; |
1597 | ||
28d9e29e | 1598 | if (!lxc_list_empty(&conf->limits) && setup_resource_limits(&conf->limits, handler->pid)) { |
c6d09e15 | 1599 | ERROR("failed to setup resource limits for '%s'", name); |
84ff3af7 | 1600 | goto out_delete_net; |
c6d09e15 WB |
1601 | } |
1602 | ||
f4152036 CB |
1603 | if (lxc_sync_barrier_child(handler, LXC_SYNC_CGROUP_UNSHARE)) |
1604 | goto out_delete_net; | |
1605 | ||
9daf6f5d | 1606 | if (!cgroup_setup_limits(handler, true)) { |
408da065 | 1607 | ERROR("Failed to setup the devices cgroup for container \"%s\".", name); |
b98f7d6e | 1608 | goto out_delete_net; |
544a48a0 | 1609 | } |
f4152036 | 1610 | TRACE("Set up cgroup device limits"); |
544a48a0 | 1611 | |
73d28d42 | 1612 | cgroup_disconnect(); |
7e4dfe0b | 1613 | cgroups_connected = false; |
73d28d42 | 1614 | |
8bf3abfb CB |
1615 | if (handler->clone_flags & CLONE_NEWCGROUP) { |
1616 | /* Now we're ready to preserve the cgroup namespace */ | |
1617 | ret = lxc_preserve_ns(handler->pid, "cgroup"); | |
1618 | if (ret < 0) { | |
1619 | ERROR("%s - Failed to preserve cgroup namespace", strerror(errno)); | |
1620 | goto out_delete_net; | |
1621 | } | |
1622 | handler->nsfd[LXC_NS_CGROUP] = ret; | |
1623 | DEBUG("Preserved cgroup namespace via fd %d", ret); | |
1624 | } | |
1625 | ||
08dd2805 SH |
1626 | snprintf(pidstr, 20, "%d", handler->pid); |
1627 | if (setenv("LXC_PID", pidstr, 1)) | |
1628 | SYSERROR("Failed to set environment variable: LXC_PID=%s.", pidstr); | |
1629 | ||
1630 | /* Run any host-side start hooks */ | |
14a7b0f9 | 1631 | if (run_lxc_hooks(name, "start-host", conf, NULL)) { |
08dd2805 SH |
1632 | ERROR("Failed to run lxc.hook.start-host for container \"%s\".", name); |
1633 | return -1; | |
1634 | } | |
1635 | ||
408da065 CB |
1636 | /* Tell the child to complete its initialization and wait for it to exec |
1637 | * or return an error. (The child will never return | |
08dd2805 | 1638 | * LXC_SYNC_READY_START+1. It will either close the sync pipe, causing |
408da065 CB |
1639 | * lxc_sync_barrier_child to return success, or return a different |
1640 | * value, causing us to error out). | |
544a48a0 | 1641 | */ |
08dd2805 | 1642 | if (lxc_sync_barrier_child(handler, LXC_SYNC_READY_START)) |
3c22086f | 1643 | return -1; |
0ad19a3f | 1644 | |
790255cf CB |
1645 | if (lxc_network_recv_name_and_ifindex_from_child(handler) < 0) { |
1646 | ERROR("Failed to receive names and ifindices for network " | |
1647 | "devices from child"); | |
1648 | goto out_delete_net; | |
1649 | } | |
1650 | ||
1651 | /* Now all networks are created, network devices are moved into place, | |
1652 | * and the correct names and ifindeces in the respective namespaces have | |
1653 | * been recorded. The corresponding structs have now all been filled. So | |
1654 | * log them for debugging purposes. | |
1655 | */ | |
28d9e29e | 1656 | lxc_log_configured_netdevs(conf); |
790255cf | 1657 | |
f4152036 CB |
1658 | /* Read tty fds allocated by child. */ |
1659 | if (lxc_recv_ttys_from_child(handler) < 0) { | |
1660 | ERROR("Failed to receive tty info from child process."); | |
1661 | goto out_delete_net; | |
1662 | } | |
1663 | ||
23c53af9 | 1664 | if (handler->ops->post_start(handler, handler->data)) |
e6126dbe MN |
1665 | goto out_abort; |
1666 | ||
25c2aca5 | 1667 | if (lxc_set_state(name, handler, RUNNING)) { |
408da065 CB |
1668 | ERROR("Failed to set state for container \"%s\" to \"%s\".", name, |
1669 | lxc_state2str(RUNNING)); | |
59eb99ba | 1670 | goto out_abort; |
3f21c114 | 1671 | } |
22ebac19 | 1672 | |
3c22086f | 1673 | lxc_sync_fini(handler); |
0c547523 | 1674 | |
e6126dbe | 1675 | return 0; |
1ac470c0 | 1676 | |
7fef7a06 | 1677 | out_delete_net: |
7e4dfe0b SH |
1678 | if (cgroups_connected) |
1679 | cgroup_disconnect(); | |
1bd8d726 | 1680 | |
bb84beda CB |
1681 | if (handler->clone_flags & CLONE_NEWNET) |
1682 | lxc_delete_network(handler); | |
1bd8d726 | 1683 | |
59eb99ba DL |
1684 | out_abort: |
1685 | lxc_abort(name, handler); | |
3c22086f | 1686 | lxc_sync_fini(handler); |
5c068da9 SH |
1687 | if (handler->pinfd >= 0) { |
1688 | close(handler->pinfd); | |
1689 | handler->pinfd = -1; | |
1690 | } | |
1691 | ||
b79fcd86 | 1692 | return -1; |
59eb99ba | 1693 | } |
0ad19a3f | 1694 | |
aa460476 | 1695 | int __lxc_start(const char *name, struct lxc_handler *handler, |
507cee36 TA |
1696 | struct lxc_operations* ops, void *data, const char *lxcpath, |
1697 | bool backgrounded) | |
59eb99ba | 1698 | { |
59eb99ba | 1699 | int status; |
aa460476 | 1700 | int err = -1; |
aa460476 | 1701 | struct lxc_conf *conf = handler->conf; |
80090207 | 1702 | |
aa460476 | 1703 | if (lxc_init(name, handler) < 0) { |
408da065 | 1704 | ERROR("Failed to initialize container \"%s\".", name); |
66aeffc7 | 1705 | return -1; |
0ad19a3f | 1706 | } |
ee70bf78 CLG |
1707 | handler->ops = ops; |
1708 | handler->data = data; | |
507cee36 | 1709 | handler->backgrounded = backgrounded; |
e6126dbe | 1710 | |
76a26f55 | 1711 | if (!attach_block_device(handler->conf)) { |
408da065 | 1712 | ERROR("Failed to attach block device."); |
76a26f55 SH |
1713 | goto out_fini_nonet; |
1714 | } | |
1715 | ||
35120d9c | 1716 | if (geteuid() == 0 && !lxc_list_empty(&conf->id_map)) { |
408da065 | 1717 | /* If the backing store is a device, mount it here and now. */ |
35120d9c SH |
1718 | if (rootfs_is_blockdev(conf)) { |
1719 | if (unshare(CLONE_NEWNS) < 0) { | |
408da065 | 1720 | ERROR("Failed to unshare CLONE_NEWNS."); |
35120d9c SH |
1721 | goto out_fini_nonet; |
1722 | } | |
408da065 CB |
1723 | INFO("Unshared CLONE_NEWNS."); |
1724 | ||
6a0c909a | 1725 | remount_all_slave(); |
35120d9c | 1726 | if (do_rootfs_setup(conf, name, lxcpath) < 0) { |
408da065 | 1727 | ERROR("Error setting up rootfs mount as root before spawn."); |
35120d9c SH |
1728 | goto out_fini_nonet; |
1729 | } | |
408da065 | 1730 | INFO("Set up container rootfs as host root."); |
35120d9c SH |
1731 | } |
1732 | } | |
1733 | ||
23c53af9 | 1734 | err = lxc_spawn(handler); |
59eb99ba | 1735 | if (err) { |
408da065 | 1736 | ERROR("Failed to spawn container \"%s\".", name); |
76a26f55 | 1737 | goto out_detach_blockdev; |
0ad19a3f | 1738 | } |
a9e1109e CB |
1739 | /* close parent side of data socket */ |
1740 | close(handler->data_sock[0]); | |
1741 | handler->data_sock[0] = -1; | |
1742 | close(handler->data_sock[1]); | |
1743 | handler->data_sock[1] = -1; | |
0ad19a3f | 1744 | |
8bee8851 WB |
1745 | handler->conf->reboot = 0; |
1746 | ||
3a0f472d | 1747 | err = lxc_poll(name, handler); |
e043236e | 1748 | if (err) { |
408da065 | 1749 | ERROR("LXC mainloop exited with error: %d.", err); |
59eb99ba DL |
1750 | goto out_abort; |
1751 | } | |
0ad19a3f | 1752 | |
3a0f472d | 1753 | while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR) |
1bc5cc8c | 1754 | continue; |
e043236e | 1755 | |
408da065 CB |
1756 | /* If the child process exited but was not signaled, it didn't call |
1757 | * reboot. This should mean it was an lxc-execute which simply exited. | |
1758 | * In any case, treat it as a 'halt'. | |
8b004f07 | 1759 | */ |
d028235d | 1760 | if (WIFSIGNALED(status)) { |
8b004f07 SH |
1761 | switch(WTERMSIG(status)) { |
1762 | case SIGINT: /* halt */ | |
408da065 | 1763 | DEBUG("Container \"%s\" is halting.", name); |
8b004f07 SH |
1764 | break; |
1765 | case SIGHUP: /* reboot */ | |
408da065 | 1766 | DEBUG("Container \"%s\" is rebooting.", name); |
8b004f07 SH |
1767 | handler->conf->reboot = 1; |
1768 | break; | |
c2b9bd9e | 1769 | case SIGSYS: /* seccomp */ |
408da065 | 1770 | DEBUG("Container \"%s\" violated its seccomp policy.", name); |
c2b9bd9e | 1771 | break; |
8b004f07 | 1772 | default: |
408da065 | 1773 | DEBUG("Unknown exit status for container \"%s\" init %d.", name, WTERMSIG(status)); |
8b004f07 SH |
1774 | break; |
1775 | } | |
d028235d | 1776 | } |
828695d9 | 1777 | |
b809f232 CB |
1778 | err = lxc_restore_phys_nics_to_netns(handler); |
1779 | if (err < 0) | |
1780 | ERROR("Failed to move physical network devices back to parent " | |
1781 | "network namespace"); | |
ce5782df | 1782 | |
5c068da9 SH |
1783 | if (handler->pinfd >= 0) { |
1784 | close(handler->pinfd); | |
1785 | handler->pinfd = -1; | |
1786 | } | |
1787 | ||
1787abca | 1788 | lxc_monitor_send_exit_code(name, status, handler->lxcpath); |
3a0f472d | 1789 | err = lxc_error_set_and_log(handler->pid, status); |
1bd8d726 | 1790 | |
9d7f9e52 | 1791 | out_fini: |
bb84beda | 1792 | lxc_delete_network(handler); |
74a2b586 | 1793 | |
76a26f55 SH |
1794 | out_detach_blockdev: |
1795 | detach_block_device(handler->conf); | |
1796 | ||
74a2b586 | 1797 | out_fini_nonet: |
3a0f472d | 1798 | lxc_fini(name, handler); |
0ad19a3f | 1799 | return err; |
1800 | ||
59eb99ba | 1801 | out_abort: |
3a0f472d | 1802 | lxc_abort(name, handler); |
9d7f9e52 | 1803 | goto out_fini; |
0ad19a3f | 1804 | } |
ee70bf78 CLG |
1805 | |
1806 | struct start_args { | |
1807 | char *const *argv; | |
1808 | }; | |
1809 | ||
1810 | static int start(struct lxc_handler *handler, void* data) | |
1811 | { | |
1812 | struct start_args *arg = data; | |
1813 | ||
408da065 | 1814 | NOTICE("Exec'ing \"%s\".", arg->argv[0]); |
ee70bf78 CLG |
1815 | |
1816 | execvp(arg->argv[0], arg->argv); | |
408da065 | 1817 | SYSERROR("Failed to exec \"%s\".", arg->argv[0]); |
ee70bf78 CLG |
1818 | return 0; |
1819 | } | |
1820 | ||
1821 | static int post_start(struct lxc_handler *handler, void* data) | |
1822 | { | |
1823 | struct start_args *arg = data; | |
1824 | ||
408da065 | 1825 | NOTICE("Started \"%s\" with pid \"%d\".", arg->argv[0], handler->pid); |
ee70bf78 CLG |
1826 | return 0; |
1827 | } | |
1828 | ||
1829 | static struct lxc_operations start_ops = { | |
1830 | .start = start, | |
1831 | .post_start = post_start | |
1832 | }; | |
1833 | ||
aa460476 | 1834 | int lxc_start(const char *name, char *const argv[], struct lxc_handler *handler, |
507cee36 | 1835 | const char *lxcpath, bool backgrounded) |
ee70bf78 CLG |
1836 | { |
1837 | struct start_args start_arg = { | |
1838 | .argv = argv, | |
1839 | }; | |
1840 | ||
aa460476 | 1841 | return __lxc_start(name, handler, &start_ops, &start_arg, lxcpath, backgrounded); |
ee70bf78 | 1842 | } |
28272964 CB |
1843 | |
1844 | static void lxc_destroy_container_on_signal(struct lxc_handler *handler, | |
1845 | const char *name) | |
1846 | { | |
1847 | char destroy[MAXPATHLEN]; | |
1848 | bool bret = true; | |
1849 | int ret = 0; | |
f01f7975 | 1850 | struct lxc_container *c; |
df31363a | 1851 | if (handler->conf->rootfs.path && handler->conf->rootfs.mount) { |
790255cf | 1852 | bret = do_destroy_container(handler); |
28272964 | 1853 | if (!bret) { |
408da065 | 1854 | ERROR("Error destroying rootfs for container \"%s\".", name); |
28272964 CB |
1855 | return; |
1856 | } | |
1857 | } | |
408da065 | 1858 | INFO("Destroyed rootfs for container \"%s\".", name); |
28272964 CB |
1859 | |
1860 | ret = snprintf(destroy, MAXPATHLEN, "%s/%s", handler->lxcpath, name); | |
1861 | if (ret < 0 || ret >= MAXPATHLEN) { | |
408da065 | 1862 | ERROR("Error destroying directory for container \"%s\".", name); |
28272964 CB |
1863 | return; |
1864 | } | |
1865 | ||
f01f7975 CB |
1866 | c = lxc_container_new(name, handler->lxcpath); |
1867 | if (c) { | |
1868 | if (container_disk_lock(c)) { | |
408da065 | 1869 | INFO("Could not update lxc_snapshots file."); |
f01f7975 CB |
1870 | lxc_container_put(c); |
1871 | } else { | |
1872 | mod_all_rdeps(c, false); | |
1873 | container_disk_unlock(c); | |
1874 | lxc_container_put(c); | |
1875 | } | |
1876 | } | |
1877 | ||
d0fbc7ba | 1878 | if (!handler->am_root) |
94b0a3ac CB |
1879 | ret = userns_exec_full(handler->conf, lxc_rmdir_onedev_wrapper, |
1880 | destroy, "lxc_rmdir_onedev_wrapper"); | |
28272964 CB |
1881 | else |
1882 | ret = lxc_rmdir_onedev(destroy, NULL); | |
1883 | ||
1884 | if (ret < 0) { | |
408da065 | 1885 | ERROR("Error destroying directory for container \"%s\".", name); |
28272964 CB |
1886 | return; |
1887 | } | |
408da065 | 1888 | INFO("Destroyed directory for container \"%s\".", name); |
28272964 CB |
1889 | } |
1890 | ||
1891 | static int lxc_rmdir_onedev_wrapper(void *data) | |
1892 | { | |
1893 | char *arg = (char *) data; | |
1894 | return lxc_rmdir_onedev(arg, NULL); | |
1895 | } | |
1896 | ||
790255cf | 1897 | static bool do_destroy_container(struct lxc_handler *handler) { |
94b0a3ac CB |
1898 | int ret; |
1899 | ||
d0fbc7ba | 1900 | if (!handler->am_root) { |
94b0a3ac CB |
1901 | ret = userns_exec_full(handler->conf, storage_destroy_wrapper, |
1902 | handler->conf, "storage_destroy_wrapper"); | |
1903 | if (ret < 0) | |
d028235d | 1904 | return false; |
10bc1861 | 1905 | |
d028235d SG |
1906 | return true; |
1907 | } | |
10bc1861 | 1908 | |
790255cf | 1909 | return storage_destroy(handler->conf); |
28272964 | 1910 | } |