]>
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 | ||
d38dd64a CB |
26 | #ifndef _GNU_SOURCE |
27 | #define _GNU_SOURCE 1 | |
28 | #endif | |
f3787121 | 29 | #include <alloca.h> |
0ad19a3f | 30 | #include <dirent.h> |
31 | #include <errno.h> | |
b0a33c1e | 32 | #include <fcntl.h> |
c476bdce | 33 | #include <grp.h> |
37515ebd | 34 | #include <poll.h> |
b467714b | 35 | #include <pthread.h> |
f3787121 CB |
36 | #include <signal.h> |
37 | #include <stdio.h> | |
38 | #include <stdlib.h> | |
39 | #include <string.h> | |
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> |
d38dd64a | 50 | #include <unistd.h> |
495d2046 | 51 | |
f3787121 CB |
52 | #include "af_unix.h" |
53 | #include "caps.h" | |
563f2f2c | 54 | #include "cgroup.h" |
f3787121 | 55 | #include "commands.h" |
bbf5cf35 | 56 | #include "commands_utils.h" |
f3787121 | 57 | #include "conf.h" |
d38dd64a | 58 | #include "config.h" |
74c6e2b0 | 59 | #include "confile_utils.h" |
e2bcd7db | 60 | #include "error.h" |
7fbb15ec | 61 | #include "file_utils.h" |
1cc8bd4b | 62 | #include "list.h" |
f3787121 | 63 | #include "log.h" |
d38dd64a | 64 | #include "lsm/lsm.h" |
f2d5a09d | 65 | #include "lxccontainer.h" |
f3787121 CB |
66 | #include "lxclock.h" |
67 | #include "lxcseccomp.h" | |
3ef9b3d3 | 68 | #include "macro.h" |
f3787121 | 69 | #include "mainloop.h" |
63376d7d | 70 | #include "monitor.h" |
f549edcc | 71 | #include "namespace.h" |
811ef482 | 72 | #include "network.h" |
38e5c2db | 73 | #include "raw_syscalls.h" |
f3787121 | 74 | #include "start.h" |
6be5397b CB |
75 | #include "storage/storage.h" |
76 | #include "storage/storage_utils.h" | |
0ed9b1bc CB |
77 | #include "sync.h" |
78 | #include "terminal.h" | |
79 | #include "utils.h" | |
36eb9bde | 80 | |
d38dd64a CB |
81 | #if HAVE_LIBCAP |
82 | #include <sys/capability.h> | |
83 | #endif | |
84 | ||
9de31d5a CB |
85 | #ifndef HAVE_STRLCPY |
86 | #include "include/strlcpy.h" | |
87 | #endif | |
88 | ||
ac2cecc4 | 89 | lxc_log_define(start, lxc); |
36eb9bde | 90 | |
f01f7975 | 91 | extern void mod_all_rdeps(struct lxc_container *c, bool inc); |
790255cf | 92 | static bool do_destroy_container(struct lxc_handler *handler); |
28272964 CB |
93 | static int lxc_rmdir_onedev_wrapper(void *data); |
94 | static void lxc_destroy_container_on_signal(struct lxc_handler *handler, | |
95 | const char *name); | |
96 | ||
c8154066 SH |
97 | static void print_top_failing_dir(const char *path) |
98 | { | |
9e5f5f2f CB |
99 | int ret; |
100 | size_t len; | |
101 | char *copy, *e, *p, saved; | |
c8154066 | 102 | |
9e5f5f2f CB |
103 | len = strlen(path); |
104 | copy = alloca(len + 1); | |
cbb9c7c7 DJ |
105 | (void)strlcpy(copy, path, len + 1); |
106 | ||
c8154066 SH |
107 | p = copy; |
108 | e = copy + len; | |
cbb9c7c7 | 109 | |
c8154066 | 110 | while (p < e) { |
f3787121 CB |
111 | while (p < e && *p == '/') |
112 | p++; | |
9e5f5f2f | 113 | |
f3787121 CB |
114 | while (p < e && *p != '/') |
115 | p++; | |
9e5f5f2f | 116 | |
c8154066 SH |
117 | saved = *p; |
118 | *p = '\0'; | |
9e5f5f2f CB |
119 | |
120 | ret = access(copy, X_OK); | |
121 | if (ret != 0) { | |
408da065 CB |
122 | SYSERROR("Could not access %s. Please grant it x " |
123 | "access, or add an ACL for the container " | |
9e5f5f2f | 124 | "root", copy); |
c8154066 SH |
125 | return; |
126 | } | |
127 | *p = saved; | |
128 | } | |
129 | } | |
130 | ||
5032bf39 | 131 | static void lxc_put_nsfds(struct lxc_handler *handler) |
408da065 | 132 | { |
9f30a190 MM |
133 | int i; |
134 | ||
9f30a190 | 135 | for (i = 0; i < LXC_NS_MAX; i++) { |
5032bf39 | 136 | if (handler->nsfd[i] < 0) |
39cd919c CB |
137 | continue; |
138 | ||
5032bf39 CB |
139 | close(handler->nsfd[i]); |
140 | handler->nsfd[i] = -EBADF; | |
9f30a190 | 141 | } |
9f30a190 MM |
142 | } |
143 | ||
4cb53844 CB |
144 | static int lxc_try_preserve_ns(const int pid, const char *ns) |
145 | { | |
146 | int fd; | |
147 | ||
148 | fd = lxc_preserve_ns(pid, ns); | |
149 | if (fd < 0) { | |
150 | if (errno != ENOENT) { | |
151 | SYSERROR("Failed to preserve %s namespace", ns); | |
152 | return -EINVAL; | |
153 | } | |
154 | ||
a24c5678 | 155 | SYSWARN("Kernel does not support preserving %s namespaces", ns); |
4cb53844 CB |
156 | return -EOPNOTSUPP; |
157 | } | |
158 | ||
159 | return fd; | |
160 | } | |
161 | ||
162 | /* lxc_try_preserve_namespaces: open /proc/@pid/ns/@ns for each namespace | |
163 | * specified in ns_clone_flags. | |
4d8ac866 | 164 | * Return true on success, false on failure. |
62d05d9b | 165 | */ |
4cb53844 CB |
166 | static bool lxc_try_preserve_namespaces(struct lxc_handler *handler, |
167 | int ns_clone_flags, pid_t pid) | |
f3787121 | 168 | { |
9fef3355 | 169 | int i; |
9f30a190 | 170 | |
9f30a190 | 171 | for (i = 0; i < LXC_NS_MAX; i++) |
5032bf39 | 172 | handler->nsfd[i] = -EBADF; |
cd43d2d1 | 173 | |
9f30a190 | 174 | for (i = 0; i < LXC_NS_MAX; i++) { |
4cb53844 CB |
175 | int fd; |
176 | ||
becad0ec | 177 | if ((ns_clone_flags & ns_info[i].clone_flag) == 0) |
9f30a190 | 178 | continue; |
28d9e29e | 179 | |
4cb53844 CB |
180 | fd = lxc_try_preserve_ns(pid, ns_info[i].proc_name); |
181 | if (fd < 0) { | |
182 | handler->nsfd[i] = -EBADF; | |
183 | ||
184 | /* Do not fail to start container on kernels that do | |
185 | * not support interacting with namespaces through | |
186 | * /proc. | |
187 | */ | |
188 | if (fd == -EOPNOTSUPP) | |
189 | continue; | |
fa3a5b22 | 190 | |
4cb53844 CB |
191 | lxc_put_nsfds(handler); |
192 | return false; | |
193 | } | |
194 | ||
195 | handler->nsfd[i] = fd; | |
196 | DEBUG("Preserved %s namespace via fd %d", ns_info[i].proc_name, | |
197 | handler->nsfd[i]); | |
9f30a190 MM |
198 | } |
199 | ||
62d05d9b | 200 | return true; |
9f30a190 MM |
201 | } |
202 | ||
ec1dc633 | 203 | static inline bool match_stdfds(int fd) |
80090207 | 204 | { |
ec1dc633 | 205 | return (fd == STDIN_FILENO || fd == STDOUT_FILENO || fd == STDERR_FILENO); |
80090207 CLG |
206 | } |
207 | ||
47a46cf1 CB |
208 | int lxc_check_inherited(struct lxc_conf *conf, bool closeall, |
209 | int *fds_to_ignore, size_t len_fds) | |
80090207 | 210 | { |
80090207 | 211 | int fd, fddir; |
47a46cf1 | 212 | size_t i; |
80090207 | 213 | DIR *dir; |
a5a70219 | 214 | struct dirent *direntp; |
80090207 | 215 | |
d2cf4c37 SH |
216 | if (conf && conf->close_all_fds) |
217 | closeall = true; | |
218 | ||
b119f362 | 219 | restart: |
80090207 CLG |
220 | dir = opendir("/proc/self/fd"); |
221 | if (!dir) { | |
a24c5678 | 222 | SYSWARN("Failed to open directory"); |
80090207 CLG |
223 | return -1; |
224 | } | |
225 | ||
226 | fddir = dirfd(dir); | |
227 | ||
74f96976 | 228 | while ((direntp = readdir(dir))) { |
a5a70219 | 229 | int ret; |
d39b10eb CB |
230 | struct lxc_list *cur; |
231 | bool matched = false; | |
232 | ||
a5a70219 | 233 | if (strcmp(direntp->d_name, ".") == 0) |
80090207 CLG |
234 | continue; |
235 | ||
a5a70219 | 236 | if (strcmp(direntp->d_name, "..") == 0) |
80090207 CLG |
237 | continue; |
238 | ||
a5a70219 CB |
239 | ret = lxc_safe_int(direntp->d_name, &fd); |
240 | if (ret < 0) { | |
241 | INFO("Could not parse file descriptor for \"%s\"", direntp->d_name); | |
d4cff0d2 CB |
242 | continue; |
243 | } | |
80090207 | 244 | |
47a46cf1 CB |
245 | for (i = 0; i < len_fds; i++) |
246 | if (fds_to_ignore[i] == fd) | |
247 | break; | |
248 | ||
249 | if (fd == fddir || fd == lxc_log_fd || | |
250 | (i < len_fds && fd == fds_to_ignore[i])) | |
80090207 CLG |
251 | continue; |
252 | ||
d39b10eb | 253 | /* Keep state clients that wait on reboots. */ |
3ee9e4fb CB |
254 | if (conf) { |
255 | lxc_list_for_each(cur, &conf->state_clients) { | |
256 | struct lxc_state_client *client = cur->elem; | |
d39b10eb | 257 | |
3ee9e4fb CB |
258 | if (client->clientfd != fd) |
259 | continue; | |
d39b10eb | 260 | |
3ee9e4fb CB |
261 | matched = true; |
262 | break; | |
263 | } | |
d39b10eb CB |
264 | } |
265 | ||
266 | if (matched) | |
267 | continue; | |
268 | ||
858377e4 SH |
269 | if (current_config && fd == current_config->logfd) |
270 | continue; | |
271 | ||
ec1dc633 | 272 | if (match_stdfds(fd)) |
80090207 | 273 | continue; |
80090207 | 274 | |
d2cf4c37 | 275 | if (closeall) { |
b119f362 SH |
276 | close(fd); |
277 | closedir(dir); | |
28d9e29e | 278 | INFO("Closed inherited fd %d", fd); |
b119f362 SH |
279 | goto restart; |
280 | } | |
28d9e29e | 281 | WARN("Inherited fd %d", fd); |
80090207 CLG |
282 | } |
283 | ||
408da065 | 284 | /* Only enable syslog at this point to avoid the above logging function |
64c57ea1 BD |
285 | * to open a new fd and make the check_inherited function enter an |
286 | * infinite loop. | |
287 | */ | |
288 | lxc_log_enable_syslog(); | |
289 | ||
92c7f629 GK |
290 | closedir(dir); /* cannot fail */ |
291 | return 0; | |
80090207 CLG |
292 | } |
293 | ||
83ee7875 | 294 | static int setup_signal_fd(sigset_t *oldmask) |
b0a33c1e | 295 | { |
7288dfb6 CB |
296 | int ret; |
297 | int sig; | |
b0a33c1e | 298 | sigset_t mask; |
7288dfb6 | 299 | const int signals[] = {SIGBUS, SIGILL, SIGSEGV, SIGWINCH}; |
b0a33c1e | 300 | |
408da065 | 301 | /* Block everything except serious error signals. */ |
df0795b1 CB |
302 | ret = sigfillset(&mask); |
303 | if (ret < 0) | |
304 | return -EBADF; | |
305 | ||
306 | for (sig = 0; sig < (sizeof(signals) / sizeof(signals[0])); sig++) { | |
307 | ret = sigdelset(&mask, signals[sig]); | |
308 | if (ret < 0) | |
309 | return -EBADF; | |
b0a33c1e | 310 | } |
311 | ||
b467714b | 312 | ret = pthread_sigmask(SIG_BLOCK, &mask, oldmask); |
df0795b1 CB |
313 | if (ret < 0) { |
314 | SYSERROR("Failed to set signal mask"); | |
315 | return -EBADF; | |
b0a33c1e | 316 | } |
317 | ||
df0795b1 CB |
318 | ret = signalfd(-1, &mask, SFD_CLOEXEC); |
319 | if (ret < 0) { | |
320 | SYSERROR("Failed to create signal file descriptor"); | |
321 | return -EBADF; | |
b0a33c1e | 322 | } |
323 | ||
df0795b1 | 324 | TRACE("Created signal file descriptor %d", ret); |
1ac470c0 | 325 | |
df0795b1 | 326 | return ret; |
b0a33c1e | 327 | } |
328 | ||
84c92abd | 329 | static int signal_handler(int fd, uint32_t events, void *data, |
f3787121 | 330 | struct lxc_epoll_descr *descr) |
b0a33c1e | 331 | { |
15cd25fd | 332 | int ret; |
3c319edb CB |
333 | siginfo_t info; |
334 | struct signalfd_siginfo siginfo; | |
335 | struct lxc_handler *hdlr = data; | |
15cd25fd | 336 | |
489f39be | 337 | ret = lxc_read_nointr(fd, &siginfo, sizeof(siginfo)); |
15cd25fd | 338 | if (ret < 0) { |
6e94162a | 339 | ERROR("Failed to read signal info from signal file descriptor %d", fd); |
b2a48508 | 340 | return LXC_MAINLOOP_ERROR; |
15cd25fd DL |
341 | } |
342 | ||
343 | if (ret != sizeof(siginfo)) { | |
6e94162a CB |
344 | ERROR("Unexpected size for struct signalfd_siginfo"); |
345 | return -EINVAL; | |
15cd25fd DL |
346 | } |
347 | ||
408da065 | 348 | /* Check whether init is running. */ |
80507ee8 | 349 | info.si_pid = 0; |
3c319edb CB |
350 | ret = waitid(P_PID, hdlr->pid, &info, WEXITED | WNOWAIT | WNOHANG); |
351 | if (ret == 0 && info.si_pid == hdlr->pid) | |
352 | hdlr->init_died = true; | |
80507ee8 | 353 | |
cd5177e9 TA |
354 | /* Try to figure out a reasonable exit status to report. */ |
355 | if (hdlr->init_died) { | |
356 | switch (info.si_code) { | |
357 | case CLD_EXITED: | |
358 | hdlr->exit_status = info.si_status << 8; | |
359 | break; | |
360 | case CLD_KILLED: | |
361 | case CLD_DUMPED: | |
362 | case CLD_STOPPED: | |
363 | hdlr->exit_status = info.si_status << 8 | 0x7f; | |
364 | break; | |
365 | case CLD_CONTINUED: | |
366 | /* Huh? The waitid() told us it's dead *and* continued? */ | |
367 | WARN("Init %d dead and continued?", hdlr->pid); | |
368 | hdlr->exit_status = 1; | |
369 | break; | |
370 | default: | |
20993a97 | 371 | ERROR("Unknown si_code: %d", info.si_code); |
55e1311e | 372 | hdlr->exit_status = 1; |
cd5177e9 TA |
373 | } |
374 | } | |
375 | ||
d4b5d7a8 TA |
376 | if (siginfo.ssi_signo == SIGHUP) { |
377 | kill(hdlr->pid, SIGTERM); | |
378 | INFO("Killing %d since terminal hung up", hdlr->pid); | |
b2a48508 | 379 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE : LXC_MAINLOOP_CONTINUE; |
d4b5d7a8 TA |
380 | } |
381 | ||
9cb94384 TA |
382 | if (siginfo.ssi_signo != SIGCHLD) { |
383 | kill(hdlr->pid, siginfo.ssi_signo); | |
384 | INFO("Forwarded signal %d to pid %d", siginfo.ssi_signo, hdlr->pid); | |
b2a48508 | 385 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE : LXC_MAINLOOP_CONTINUE; |
9cb94384 TA |
386 | } |
387 | ||
3a9e949f TA |
388 | /* More robustness, protect ourself from a SIGCHLD sent |
389 | * by a process different from the container init. | |
390 | */ | |
391 | if (siginfo.ssi_pid != hdlr->pid) { | |
6e94162a CB |
392 | NOTICE("Received %d from pid %d instead of container init %d", |
393 | siginfo.ssi_signo, siginfo.ssi_pid, hdlr->pid); | |
b2a48508 | 394 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE : LXC_MAINLOOP_CONTINUE; |
3a9e949f TA |
395 | } |
396 | ||
408da065 | 397 | if (siginfo.ssi_code == CLD_STOPPED) { |
6e94162a | 398 | INFO("Container init process was stopped"); |
b2a48508 | 399 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE : LXC_MAINLOOP_CONTINUE; |
408da065 | 400 | } else if (siginfo.ssi_code == CLD_CONTINUED) { |
6e94162a | 401 | INFO("Container init process was continued"); |
b2a48508 | 402 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE : LXC_MAINLOOP_CONTINUE; |
15cd25fd | 403 | } |
1ac470c0 | 404 | |
6e94162a | 405 | DEBUG("Container init process %d exited", hdlr->pid); |
3c319edb | 406 | return LXC_MAINLOOP_CLOSE; |
b0a33c1e | 407 | } |
408 | ||
974a8aba CB |
409 | int lxc_serve_state_clients(const char *name, struct lxc_handler *handler, |
410 | lxc_state_t state) | |
66aeffc7 | 411 | { |
9de31d5a | 412 | size_t retlen; |
dbc9832d CB |
413 | ssize_t ret; |
414 | struct lxc_list *cur, *next; | |
d39b10eb | 415 | struct lxc_state_client *client; |
dbc9832d CB |
416 | struct lxc_msg msg = {.type = lxc_msg_state, .value = state}; |
417 | ||
974a8aba CB |
418 | if (state == THAWED) |
419 | handler->state = RUNNING; | |
420 | else | |
421 | handler->state = state; | |
422 | ||
9dfa4041 | 423 | TRACE("Set container state to %s", lxc_state2str(state)); |
dbc9832d | 424 | |
d39b10eb | 425 | if (lxc_list_empty(&handler->conf->state_clients)) { |
9dfa4041 | 426 | TRACE("No state clients registered"); |
dbc9832d CB |
427 | return 0; |
428 | } | |
429 | ||
9de31d5a CB |
430 | retlen = strlcpy(msg.name, name, sizeof(msg.name)); |
431 | if (retlen >= sizeof(msg.name)) | |
432 | return -E2BIG; | |
dbc9832d | 433 | |
d39b10eb | 434 | lxc_list_for_each_safe(cur, &handler->conf->state_clients, next) { |
dbc9832d CB |
435 | client = cur->elem; |
436 | ||
d39b10eb | 437 | if (client->states[state] == 0) { |
9dfa4041 | 438 | TRACE("State %s not registered for state client %d", |
dbc9832d CB |
439 | lxc_state2str(state), client->clientfd); |
440 | continue; | |
441 | } | |
442 | ||
9dfa4041 | 443 | TRACE("Sending state %s to state client %d", |
dbc9832d CB |
444 | lxc_state2str(state), client->clientfd); |
445 | ||
7fbb15ec CB |
446 | ret = lxc_send_nointr(client->clientfd, &msg, sizeof(msg), MSG_NOSIGNAL); |
447 | if (ret <= 0) | |
6d1400b5 | 448 | SYSERROR("Failed to send message to client"); |
dbc9832d CB |
449 | |
450 | /* kick client from list */ | |
dbc9832d | 451 | lxc_list_del(cur); |
300d1cb4 | 452 | close(client->clientfd); |
dbc9832d CB |
453 | free(cur->elem); |
454 | free(cur); | |
455 | } | |
dbc9832d | 456 | |
5e5576a4 CB |
457 | return 0; |
458 | } | |
459 | ||
460 | static int lxc_serve_state_socket_pair(const char *name, | |
461 | struct lxc_handler *handler, | |
462 | lxc_state_t state) | |
463 | { | |
464 | ssize_t ret; | |
465 | ||
bb955810 | 466 | if (!handler->daemonize || |
5e5576a4 CB |
467 | handler->state_socket_pair[1] < 0 || |
468 | state == STARTING) | |
469 | return 0; | |
470 | ||
471 | /* Close read end of the socket pair. */ | |
472 | close(handler->state_socket_pair[0]); | |
473 | handler->state_socket_pair[0] = -1; | |
474 | ||
ee8377bd | 475 | again: |
5e5576a4 CB |
476 | ret = lxc_abstract_unix_send_credential(handler->state_socket_pair[1], |
477 | &(int){state}, sizeof(int)); | |
9044b79e | 478 | if (ret < 0) { |
479 | SYSERROR("Failed to send state to %d", handler->state_socket_pair[1]); | |
480 | ||
ee8377bd CB |
481 | if (errno == EINTR) |
482 | goto again; | |
9044b79e | 483 | |
484 | return -1; | |
485 | } | |
486 | ||
487 | if (ret != sizeof(int)) { | |
488 | ERROR("Message too long : %d", handler->state_socket_pair[1]); | |
ee8377bd CB |
489 | return -1; |
490 | } | |
5e5576a4 CB |
491 | |
492 | TRACE("Sent container state \"%s\" to %d", lxc_state2str(state), | |
493 | handler->state_socket_pair[1]); | |
494 | ||
495 | /* Close write end of the socket pair. */ | |
496 | close(handler->state_socket_pair[1]); | |
497 | handler->state_socket_pair[1] = -1; | |
498 | ||
499 | return 0; | |
500 | } | |
501 | ||
502 | int lxc_set_state(const char *name, struct lxc_handler *handler, | |
503 | lxc_state_t state) | |
504 | { | |
505 | int ret; | |
506 | ||
507 | ret = lxc_serve_state_socket_pair(name, handler, state); | |
508 | if (ret < 0) { | |
509 | ERROR("Failed to synchronize via anonymous pair of unix sockets"); | |
510 | return -1; | |
511 | } | |
512 | ||
513 | ret = lxc_serve_state_clients(name, handler, state); | |
514 | if (ret < 0) | |
515 | return -1; | |
516 | ||
dbc9832d CB |
517 | /* This function will try to connect to the legacy lxc-monitord state |
518 | * server and only exists for backwards compatibility. | |
519 | */ | |
9123e471 | 520 | lxc_monitor_send_state(name, state, handler->lxcpath); |
dbc9832d | 521 | |
66aeffc7 DL |
522 | return 0; |
523 | } | |
524 | ||
735f2c6e | 525 | int lxc_poll(const char *name, struct lxc_handler *handler) |
b0a33c1e | 526 | { |
3c319edb | 527 | int ret; |
86530b0a | 528 | bool has_console = true; |
3c319edb | 529 | struct lxc_epoll_descr descr, descr_console; |
b0a33c1e | 530 | |
03760215 CB |
531 | if (handler->conf->console.path && |
532 | strcmp(handler->conf->console.path, "none") == 0) | |
86530b0a L |
533 | has_console = false; |
534 | ||
3c319edb CB |
535 | ret = lxc_mainloop_open(&descr); |
536 | if (ret < 0) { | |
537 | ERROR("Failed to create mainloop"); | |
50c8bf05 | 538 | goto out_sigfd; |
b0a33c1e | 539 | } |
540 | ||
30a33fbd CB |
541 | if (has_console) { |
542 | ret = lxc_mainloop_open(&descr_console); | |
543 | if (ret < 0) { | |
544 | ERROR("Failed to create console mainloop"); | |
545 | goto out_mainloop; | |
546 | } | |
3c319edb CB |
547 | } |
548 | ||
1cc8bd4b | 549 | ret = lxc_mainloop_add_handler(&descr, handler->sigfd, signal_handler, handler); |
3c319edb | 550 | if (ret < 0) { |
1cc8bd4b | 551 | ERROR("Failed to add signal handler for %d to mainloop", handler->sigfd); |
3c319edb | 552 | goto out_mainloop_console; |
b0a33c1e | 553 | } |
554 | ||
30a33fbd | 555 | if (has_console) { |
dcad02f8 | 556 | struct lxc_terminal *console = &handler->conf->console; |
63376d7d | 557 | |
093bce5f | 558 | ret = lxc_terminal_mainloop_add(&descr, console); |
30a33fbd CB |
559 | if (ret < 0) { |
560 | ERROR("Failed to add console handlers to mainloop"); | |
561 | goto out_mainloop_console; | |
562 | } | |
563 | ||
093bce5f | 564 | ret = lxc_terminal_mainloop_add(&descr_console, console); |
30a33fbd CB |
565 | if (ret < 0) { |
566 | ERROR("Failed to add console handlers to console mainloop"); | |
567 | goto out_mainloop_console; | |
568 | } | |
a54585ad L |
569 | |
570 | handler->conf->console.descr = &descr; | |
563f2f2c DL |
571 | } |
572 | ||
3c319edb CB |
573 | ret = lxc_cmd_mainloop_add(name, &descr, handler); |
574 | if (ret < 0) { | |
575 | ERROR("Failed to add command handler to mainloop"); | |
576 | goto out_mainloop_console; | |
577 | } | |
b0a33c1e | 578 | |
3c319edb | 579 | TRACE("Mainloop is ready"); |
b0a33c1e | 580 | |
3c319edb | 581 | ret = lxc_mainloop(&descr, -1); |
1cc8bd4b CB |
582 | close(descr.epfd); |
583 | descr.epfd = -EBADF; | |
584 | if (ret < 0 || !handler->init_died) | |
42b09f94 | 585 | goto out_mainloop_console; |
1cc8bd4b | 586 | |
30a33fbd CB |
587 | if (has_console) |
588 | ret = lxc_mainloop(&descr_console, 0); | |
589 | ||
3c319edb | 590 | out_mainloop_console: |
30a33fbd CB |
591 | if (has_console) { |
592 | lxc_mainloop_close(&descr_console); | |
593 | TRACE("Closed console mainloop"); | |
594 | } | |
3c319edb | 595 | |
42b09f94 CB |
596 | out_mainloop: |
597 | lxc_mainloop_close(&descr); | |
598 | TRACE("Closed mainloop"); | |
599 | ||
b0a33c1e | 600 | out_sigfd: |
1cc8bd4b CB |
601 | close(handler->sigfd); |
602 | TRACE("Closed signal file descriptor %d", handler->sigfd); | |
603 | handler->sigfd = -EBADF; | |
408da065 | 604 | |
1cc8bd4b | 605 | return ret; |
b0a33c1e | 606 | } |
607 | ||
41784e4e CB |
608 | void lxc_zero_handler(struct lxc_handler *handler) |
609 | { | |
610 | int i; | |
611 | ||
612 | memset(handler, 0, sizeof(struct lxc_handler)); | |
613 | ||
41784e4e CB |
614 | handler->pinfd = -1; |
615 | ||
616 | handler->sigfd = -1; | |
617 | ||
618 | for (i = 0; i < LXC_NS_MAX; i++) | |
619 | handler->nsfd[i] = -1; | |
620 | ||
621 | handler->data_sock[0] = -1; | |
622 | handler->data_sock[1] = -1; | |
623 | ||
624 | handler->state_socket_pair[0] = -1; | |
625 | handler->state_socket_pair[1] = -1; | |
626 | ||
627 | handler->sync_sock[0] = -1; | |
628 | handler->sync_sock[1] = -1; | |
629 | } | |
630 | ||
f2e07cb6 CB |
631 | void lxc_free_handler(struct lxc_handler *handler) |
632 | { | |
1cc8bd4b CB |
633 | if (handler->pinfd >= 0) |
634 | close(handler->pinfd); | |
635 | ||
636 | if (handler->sigfd >= 0) | |
637 | close(handler->sigfd); | |
638 | ||
639 | lxc_put_nsfds(handler); | |
640 | ||
80308d07 | 641 | if (handler->conf && handler->conf->reboot == REBOOT_NONE) |
0d0d3655 | 642 | if (handler->conf->maincmd_fd >= 0) |
9044b79e | 643 | lxc_abstract_unix_close(handler->conf->maincmd_fd); |
f2e07cb6 | 644 | |
5e5576a4 CB |
645 | if (handler->state_socket_pair[0] >= 0) |
646 | close(handler->state_socket_pair[0]); | |
647 | ||
648 | if (handler->state_socket_pair[1] >= 0) | |
649 | close(handler->state_socket_pair[1]); | |
650 | ||
f2e07cb6 CB |
651 | handler->conf = NULL; |
652 | free(handler); | |
1cc8bd4b | 653 | handler = NULL; |
f2e07cb6 CB |
654 | } |
655 | ||
aa460476 | 656 | struct lxc_handler *lxc_init_handler(const char *name, struct lxc_conf *conf, |
5e5576a4 | 657 | const char *lxcpath, bool daemonize) |
59eb99ba | 658 | { |
5e5576a4 | 659 | int i, ret; |
3a0f472d DL |
660 | struct lxc_handler *handler; |
661 | ||
662 | handler = malloc(sizeof(*handler)); | |
fdecbc9c | 663 | if (!handler) |
3a0f472d | 664 | return NULL; |
9044b79e | 665 | |
59eb99ba DL |
666 | memset(handler, 0, sizeof(*handler)); |
667 | ||
fdecbc9c CB |
668 | /* Note that am_guest_unpriv() checks the effective uid. We |
669 | * probably don't care if we are real root only if we are running | |
670 | * as root so this should be fine. | |
790255cf | 671 | */ |
e0010464 | 672 | handler->am_root = !am_guest_unpriv(); |
c6012571 | 673 | handler->data_sock[0] = handler->data_sock[1] = -1; |
fae349da | 674 | handler->conf = conf; |
9123e471 | 675 | handler->lxcpath = lxcpath; |
5c068da9 | 676 | handler->pinfd = -1; |
1cc8bd4b CB |
677 | handler->sigfd = -EBADF; |
678 | handler->init_died = false; | |
5e5576a4 | 679 | handler->state_socket_pair[0] = handler->state_socket_pair[1] = -1; |
80308d07 | 680 | if (handler->conf->reboot == REBOOT_NONE) |
d39b10eb | 681 | lxc_list_init(&handler->conf->state_clients); |
fae349da | 682 | |
b6b2b194 WB |
683 | for (i = 0; i < LXC_NS_MAX; i++) |
684 | handler->nsfd[i] = -1; | |
685 | ||
f0ecc19d | 686 | handler->name = name; |
3bdf52d7 | 687 | |
80308d07 | 688 | if (daemonize && handler->conf->reboot == REBOOT_NONE) { |
5e5576a4 | 689 | /* Create socketpair() to synchronize on daemonized startup. |
fdecbc9c CB |
690 | * When the container reboots we don't need to synchronize |
691 | * again currently so don't open another socketpair(). | |
5e5576a4 CB |
692 | */ |
693 | ret = socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0, | |
694 | handler->state_socket_pair); | |
695 | if (ret < 0) { | |
696 | ERROR("Failed to create anonymous pair of unix sockets"); | |
697 | goto on_error; | |
698 | } | |
9044b79e | 699 | |
5e5576a4 CB |
700 | TRACE("Created anonymous pair {%d,%d} of unix sockets", |
701 | handler->state_socket_pair[0], | |
702 | handler->state_socket_pair[1]); | |
703 | } | |
704 | ||
80308d07 | 705 | if (handler->conf->reboot == REBOOT_NONE) { |
d39b10eb CB |
706 | handler->conf->maincmd_fd = lxc_cmd_init(name, lxcpath, "command"); |
707 | if (handler->conf->maincmd_fd < 0) { | |
708 | ERROR("Failed to set up command socket"); | |
709 | goto on_error; | |
710 | } | |
aa460476 | 711 | } |
9044b79e | 712 | |
9dfa4041 | 713 | TRACE("Unix domain socket %d for command server is ready", |
aa460476 CB |
714 | handler->conf->maincmd_fd); |
715 | ||
716 | return handler; | |
717 | ||
f2e07cb6 CB |
718 | on_error: |
719 | lxc_free_handler(handler); | |
aa460476 CB |
720 | |
721 | return NULL; | |
722 | } | |
723 | ||
724 | int lxc_init(const char *name, struct lxc_handler *handler) | |
725 | { | |
a2c09be0 | 726 | int ret; |
4a03ded4 | 727 | const char *loglevel; |
aa460476 CB |
728 | struct lxc_conf *conf = handler->conf; |
729 | ||
434c8e15 CB |
730 | handler->monitor_pid = lxc_raw_getpid(); |
731 | ||
aa460476 | 732 | lsm_init(); |
2170b263 | 733 | TRACE("Initialized LSM"); |
d2e30e99 | 734 | |
2170b263 CB |
735 | ret = lxc_read_seccomp_config(conf); |
736 | if (ret < 0) { | |
737 | ERROR("Failed loading seccomp policy"); | |
d2e30e99 | 738 | goto out_close_maincmd_fd; |
8f2c3a70 | 739 | } |
2170b263 | 740 | TRACE("Read seccomp policy"); |
8f2c3a70 | 741 | |
408da065 | 742 | /* Begin by setting the state to STARTING. */ |
2170b263 CB |
743 | ret = lxc_set_state(name, handler, STARTING); |
744 | if (ret < 0) { | |
745 | ERROR("Failed to set state to \"%s\"", lxc_state2str(STARTING)); | |
051151de | 746 | goto out_close_maincmd_fd; |
0ad19a3f | 747 | } |
46800e77 | 748 | TRACE("Set container state to \"STARTING\""); |
0ad19a3f | 749 | |
408da065 | 750 | /* Start of environment variable setup for hooks. */ |
24c2a48d CB |
751 | ret = setenv("LXC_NAME", name, 1); |
752 | if (ret < 0) | |
753 | SYSERROR("Failed to set environment variable: LXC_NAME=%s", name); | |
408da065 | 754 | |
2170b263 CB |
755 | if (conf->rcfile) { |
756 | ret = setenv("LXC_CONFIG_FILE", conf->rcfile, 1); | |
757 | if (ret < 0) | |
758 | SYSERROR("Failed to set environment variable: " | |
759 | "LXC_CONFIG_FILE=%s", conf->rcfile); | |
760 | } | |
408da065 | 761 | |
2170b263 CB |
762 | if (conf->rootfs.mount) { |
763 | ret = setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1); | |
764 | if (ret < 0) | |
765 | SYSERROR("Failed to set environment variable: " | |
766 | "LXC_ROOTFS_MOUNT=%s", conf->rootfs.mount); | |
767 | } | |
408da065 | 768 | |
2170b263 CB |
769 | if (conf->rootfs.path) { |
770 | ret = setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1); | |
771 | if (ret < 0) | |
772 | SYSERROR("Failed to set environment variable: " | |
773 | "LXC_ROOTFS_PATH=%s", conf->rootfs.path); | |
774 | } | |
408da065 | 775 | |
2170b263 CB |
776 | if (conf->console.path) { |
777 | ret = setenv("LXC_CONSOLE", conf->console.path, 1); | |
778 | if (ret < 0) | |
779 | SYSERROR("Failed to set environment variable: " | |
780 | "LXC_CONSOLE=%s", conf->console.path); | |
781 | } | |
408da065 | 782 | |
2170b263 CB |
783 | if (conf->console.log_path) { |
784 | ret = setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1); | |
785 | if (ret < 0) | |
786 | SYSERROR("Failed to set environment variable: " | |
787 | "LXC_CONSOLE_LOGPATH=%s", conf->console.log_path); | |
788 | } | |
408da065 | 789 | |
2170b263 CB |
790 | if (cgns_supported()) { |
791 | ret = setenv("LXC_CGNS_AWARE", "1", 1); | |
792 | if (ret < 0) | |
793 | SYSERROR("Failed to set environment variable " | |
794 | "LXC_CGNS_AWARE=1"); | |
795 | } | |
b8f88d9b | 796 | |
4a03ded4 | 797 | loglevel = lxc_log_priority_to_string(lxc_log_get_level()); |
2170b263 CB |
798 | ret = setenv("LXC_LOG_LEVEL", loglevel, 1); |
799 | if (ret < 0) | |
800 | SYSERROR("Set environment variable LXC_LOG_LEVEL=%s", | |
801 | loglevel); | |
a2c09be0 CB |
802 | |
803 | if (conf->hooks_version == 0) | |
804 | ret = setenv("LXC_HOOK_VERSION", "0", 1); | |
805 | else | |
806 | ret = setenv("LXC_HOOK_VERSION", "1", 1); | |
807 | if (ret < 0) | |
808 | SYSERROR("Failed to set environment variable LXC_HOOK_VERSION=%u", conf->hooks_version); | |
408da065 | 809 | /* End of environment variable setup for hooks. */ |
f7bee6c6 | 810 | |
2170b263 | 811 | TRACE("Set environment variables"); |
dbc9832d | 812 | |
2170b263 CB |
813 | ret = run_lxc_hooks(name, "pre-start", conf, NULL); |
814 | if (ret < 0) { | |
815 | ERROR("Failed to run lxc.hook.pre-start for container \"%s\"", name); | |
773fb9ca SH |
816 | goto out_aborting; |
817 | } | |
2170b263 | 818 | TRACE("Ran pre-start hooks"); |
26ddeedd | 819 | |
408da065 CB |
820 | /* The signal fd has to be created before forking otherwise if the child |
821 | * process exits before we setup the signal fd, the event will be lost | |
822 | * and the command will be stuck. | |
823 | */ | |
83ee7875 | 824 | handler->sigfd = setup_signal_fd(&handler->oldmask); |
59eb99ba | 825 | if (handler->sigfd < 0) { |
408da065 | 826 | ERROR("Failed to setup SIGCHLD fd handler."); |
b5159817 DE |
827 | goto out_delete_tty; |
828 | } | |
2170b263 | 829 | TRACE("Set up signal fd"); |
b5159817 | 830 | |
408da065 | 831 | /* Do this after setting up signals since it might unblock SIGWINCH. */ |
564e31c4 | 832 | ret = lxc_terminal_setup(conf); |
4d1ffb0a CB |
833 | if (ret < 0) { |
834 | ERROR("Failed to create console"); | |
b5159817 | 835 | goto out_restore_sigmask; |
b0a33c1e | 836 | } |
4d1ffb0a | 837 | TRACE("Created console"); |
b0a33c1e | 838 | |
2f835b4b | 839 | ret = lxc_terminal_map_ids(conf, &conf->console); |
7cfeddd7 | 840 | if (ret < 0) { |
2170b263 | 841 | ERROR("Failed to chown console"); |
5b74eb3c | 842 | goto out_delete_terminal; |
c4d10a05 | 843 | } |
2170b263 | 844 | TRACE("Chowned console"); |
c4d10a05 | 845 | |
5a087e05 | 846 | handler->cgroup_ops = cgroup_init(handler->conf); |
2202afc9 CB |
847 | if (!handler->cgroup_ops) { |
848 | ERROR("Failed to initialize cgroup driver"); | |
5b74eb3c | 849 | goto out_delete_terminal; |
2202afc9 CB |
850 | } |
851 | TRACE("Initialized cgroup driver"); | |
852 | ||
1800f924 WB |
853 | ret = lsm_process_prepare(conf, handler->lxcpath); |
854 | if (ret < 0) { | |
855 | ERROR("Failed to initialize LSM"); | |
856 | goto out_destroy_cgroups; | |
857 | } | |
858 | TRACE("Initialized LSM"); | |
859 | ||
2170b263 | 860 | INFO("Container \"%s\" is initialized", name); |
aa460476 | 861 | return 0; |
59eb99ba | 862 | |
1800f924 | 863 | out_destroy_cgroups: |
434c8e15 CB |
864 | handler->cgroup_ops->payload_destroy(handler->cgroup_ops, handler); |
865 | handler->cgroup_ops->monitor_destroy(handler->cgroup_ops, handler); | |
1800f924 | 866 | |
5b74eb3c CB |
867 | out_delete_terminal: |
868 | lxc_terminal_delete(&handler->conf->console); | |
ea918412 | 869 | |
b5159817 | 870 | out_restore_sigmask: |
4c496daa | 871 | (void)pthread_sigmask(SIG_SETMASK, &handler->oldmask, NULL); |
ea918412 | 872 | |
59eb99ba | 873 | out_delete_tty: |
0e4be3cf | 874 | lxc_delete_tty(&conf->ttys); |
ea918412 | 875 | |
59eb99ba | 876 | out_aborting: |
4c496daa | 877 | (void)lxc_set_state(name, handler, ABORTING); |
ea918412 | 878 | |
d2e30e99 | 879 | out_close_maincmd_fd: |
9044b79e | 880 | lxc_abstract_unix_close(conf->maincmd_fd); |
d2e30e99 | 881 | conf->maincmd_fd = -1; |
aa460476 | 882 | return -1; |
59eb99ba DL |
883 | } |
884 | ||
3b72c4a0 | 885 | void lxc_fini(const char *name, struct lxc_handler *handler) |
59eb99ba | 886 | { |
c097467f | 887 | int i, ret; |
18b3b9c1 | 888 | pid_t self; |
dbc9832d | 889 | struct lxc_list *cur, *next; |
dbc9832d | 890 | char *namespaces[LXC_NS_MAX + 1]; |
b3286b62 | 891 | size_t namespace_count = 0; |
2202afc9 | 892 | struct cgroup_ops *cgroup_ops = handler->cgroup_ops; |
b6b2b194 | 893 | |
408da065 CB |
894 | /* The STOPPING state is there for future cleanup code which can take |
895 | * awhile. | |
59eb99ba | 896 | */ |
25c2aca5 | 897 | lxc_set_state(name, handler, STOPPING); |
b6b2b194 | 898 | |
0059379f | 899 | self = lxc_raw_getpid(); |
b3286b62 | 900 | for (i = 0; i < LXC_NS_MAX; i++) { |
18b3b9c1 CB |
901 | if (handler->nsfd[i] < 0) |
902 | continue; | |
903 | ||
904 | if (handler->conf->hooks_version == 0) | |
c097467f | 905 | ret = asprintf(&namespaces[namespace_count], |
18b3b9c1 CB |
906 | "%s:/proc/%d/fd/%d", ns_info[i].proc_name, |
907 | self, handler->nsfd[i]); | |
908 | else | |
c097467f | 909 | ret = asprintf(&namespaces[namespace_count], |
18b3b9c1 | 910 | "/proc/%d/fd/%d", self, handler->nsfd[i]); |
c097467f CB |
911 | if (ret == -1) { |
912 | SYSERROR("Failed to allocate memory"); | |
18b3b9c1 | 913 | break; |
b3286b62 | 914 | } |
18b3b9c1 CB |
915 | |
916 | if (handler->conf->hooks_version == 0) { | |
917 | namespace_count++; | |
918 | continue; | |
919 | } | |
920 | ||
c097467f CB |
921 | ret = setenv(ns_info[i].env_name, namespaces[namespace_count], 1); |
922 | if (ret < 0) | |
18b3b9c1 CB |
923 | SYSERROR("Failed to set environment variable %s=%s", |
924 | ns_info[i].env_name, namespaces[namespace_count]); | |
925 | else | |
926 | TRACE("Set environment variable %s=%s", | |
927 | ns_info[i].env_name, namespaces[namespace_count]); | |
928 | ||
929 | namespace_count++; | |
b3286b62 WB |
930 | } |
931 | namespaces[namespace_count] = NULL; | |
c154af98 | 932 | |
80308d07 | 933 | if (handler->conf->reboot > REBOOT_NONE) { |
c097467f CB |
934 | ret = setenv("LXC_TARGET", "reboot", 1); |
935 | if (ret < 0) | |
936 | SYSERROR("Failed to set environment variable: " | |
937 | "LXC_TARGET=reboot"); | |
938 | } | |
408da065 | 939 | |
80308d07 | 940 | if (handler->conf->reboot == REBOOT_NONE) { |
c097467f CB |
941 | ret = setenv("LXC_TARGET", "stop", 1); |
942 | if (ret < 0) | |
943 | SYSERROR("Failed to set environment variable: " | |
944 | "LXC_TARGET=stop"); | |
945 | } | |
c154af98 | 946 | |
18b3b9c1 | 947 | if (handler->conf->hooks_version == 0) |
c097467f | 948 | ret = run_lxc_hooks(name, "stop", handler->conf, namespaces); |
18b3b9c1 | 949 | else |
c097467f CB |
950 | ret = run_lxc_hooks(name, "stop", handler->conf, NULL); |
951 | if (ret < 0) | |
952 | ERROR("Failed to run \"lxc.hook.stop\" hook"); | |
c154af98 | 953 | |
b3286b62 WB |
954 | while (namespace_count--) |
955 | free(namespaces[namespace_count]); | |
28d9e29e | 956 | |
1800f924 WB |
957 | lsm_process_cleanup(handler->conf, handler->lxcpath); |
958 | ||
434c8e15 CB |
959 | cgroup_ops->payload_destroy(cgroup_ops, handler); |
960 | cgroup_ops->monitor_destroy(cgroup_ops, handler); | |
2202afc9 | 961 | cgroup_exit(cgroup_ops); |
4288b79f | 962 | |
80308d07 | 963 | if (handler->conf->reboot == REBOOT_NONE) { |
f893d898 CB |
964 | /* For all new state clients simply close the command socket. |
965 | * This will inform all state clients that the container is | |
966 | * STOPPED and also prevents a race between a open()/close() on | |
967 | * the command socket causing a new process to get ECONNREFUSED | |
968 | * because we haven't yet closed the command socket. | |
969 | */ | |
9044b79e | 970 | lxc_abstract_unix_close(handler->conf->maincmd_fd); |
d39b10eb | 971 | handler->conf->maincmd_fd = -1; |
c3184275 CB |
972 | TRACE("Closed command socket"); |
973 | ||
974 | /* This function will try to connect to the legacy lxc-monitord | |
975 | * state server and only exists for backwards compatibility. | |
976 | */ | |
977 | lxc_monitor_send_state(name, STOPPED, handler->lxcpath); | |
978 | ||
979 | /* The command socket is closed so no one can acces the command | |
980 | * socket anymore so there's no need to lock it. | |
981 | */ | |
982 | handler->state = STOPPED; | |
983 | TRACE("Set container state to \"STOPPED\""); | |
f893d898 CB |
984 | } else { |
985 | lxc_set_state(name, handler, STOPPED); | |
d39b10eb | 986 | } |
4288b79f | 987 | |
46800e77 CB |
988 | ret = run_lxc_hooks(name, "post-stop", handler->conf, NULL); |
989 | if (ret < 0) { | |
c097467f | 990 | ERROR("Failed to run lxc.hook.post-stop for container \"%s\"", name); |
80308d07 | 991 | if (handler->conf->reboot > REBOOT_NONE) { |
c097467f | 992 | WARN("Container will be stopped instead of rebooted"); |
80308d07 | 993 | handler->conf->reboot = REBOOT_NONE; |
c097467f CB |
994 | |
995 | ret = setenv("LXC_TARGET", "stop", 1); | |
996 | if (ret < 0) | |
997 | WARN("Failed to set environment variable: " | |
998 | "LXC_TARGET=stop"); | |
f3787121 CB |
999 | } |
1000 | } | |
26ddeedd | 1001 | |
408da065 | 1002 | /* Reset mask set by setup_signal_fd. */ |
b467714b | 1003 | ret = pthread_sigmask(SIG_SETMASK, &handler->oldmask, NULL); |
c097467f | 1004 | if (ret < 0) |
a24c5678 | 1005 | SYSWARN("Failed to restore signal mask"); |
8f64a3f6 | 1006 | |
2aac071b | 1007 | lxc_terminal_delete(&handler->conf->console); |
0e4be3cf | 1008 | lxc_delete_tty(&handler->conf->ttys); |
dbc9832d | 1009 | |
dbc9832d CB |
1010 | /* The command socket is now closed, no more state clients can register |
1011 | * themselves from now on. So free the list of state clients. | |
1012 | */ | |
d39b10eb CB |
1013 | lxc_list_for_each_safe(cur, &handler->conf->state_clients, next) { |
1014 | struct lxc_state_client *client = cur->elem; | |
1015 | ||
1016 | /* Keep state clients that want to be notified about reboots. */ | |
80308d07 CB |
1017 | if ((handler->conf->reboot > REBOOT_NONE) && |
1018 | (client->states[RUNNING] == 2)) | |
d39b10eb CB |
1019 | continue; |
1020 | ||
dbc9832d | 1021 | /* close state client socket */ |
dbc9832d | 1022 | lxc_list_del(cur); |
c097467f | 1023 | close(client->clientfd); |
dbc9832d CB |
1024 | free(cur->elem); |
1025 | free(cur); | |
1026 | } | |
1027 | ||
80308d07 | 1028 | if (handler->conf->ephemeral == 1 && handler->conf->reboot != REBOOT_REQ) |
28272964 | 1029 | lxc_destroy_container_on_signal(handler, name); |
2c5f2ede | 1030 | |
1cc8bd4b | 1031 | lxc_free_handler(handler); |
59eb99ba DL |
1032 | } |
1033 | ||
735f2c6e | 1034 | void lxc_abort(const char *name, struct lxc_handler *handler) |
59eb99ba | 1035 | { |
73e608b2 SH |
1036 | int ret, status; |
1037 | ||
25c2aca5 | 1038 | lxc_set_state(name, handler, ABORTING); |
0e4f9d51 CB |
1039 | |
1040 | if (handler->pid > 0) { | |
1041 | ret = kill(handler->pid, SIGKILL); | |
1042 | if (ret < 0) | |
1043 | SYSERROR("Failed to send SIGKILL to %d", handler->pid); | |
1044 | } | |
1045 | ||
408da065 CB |
1046 | while ((ret = waitpid(-1, &status, 0)) > 0) { |
1047 | ; | |
1048 | } | |
59eb99ba DL |
1049 | } |
1050 | ||
ffe1e01a | 1051 | static int do_start(void *data) |
50e98013 | 1052 | { |
8bf3abfb | 1053 | int ret; |
7a55c157 | 1054 | char path[PATH_MAX]; |
928b1f04 YT |
1055 | uid_t new_uid; |
1056 | gid_t new_gid; | |
e96a536c | 1057 | struct lxc_list *iterator; |
964581c2 CB |
1058 | uid_t nsuid = 0; |
1059 | gid_t nsgid = 0; | |
5af9369b | 1060 | int devnull_fd = -1; |
e96a536c | 1061 | struct lxc_handler *handler = data; |
50e98013 | 1062 | |
18225d19 | 1063 | lxc_sync_fini_parent(handler); |
50e98013 | 1064 | |
408da065 CB |
1065 | /* This prctl must be before the synchro, so if the parent dies before |
1066 | * we set the parent death signal, we will detect its death with the | |
1067 | * synchro right after, otherwise we have a window where the parent can | |
1068 | * exit before we set the pdeath signal leading to a unsupervized | |
1069 | * container. | |
743ecd2e | 1070 | */ |
c7f493ae | 1071 | ret = lxc_set_death_signal(SIGKILL, 0); |
912314fc CB |
1072 | if (ret < 0) { |
1073 | SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL"); | |
18225d19 | 1074 | goto out_warn_father; |
743ecd2e DL |
1075 | } |
1076 | ||
611ddd34 CB |
1077 | ret = lxc_ambient_caps_up(); |
1078 | if (ret < 0) { | |
6f5e532f | 1079 | ERROR("Failed to raise ambient capabilities"); |
611ddd34 CB |
1080 | goto out_warn_father; |
1081 | } | |
1082 | ||
b467714b | 1083 | ret = pthread_sigmask(SIG_SETMASK, &handler->oldmask, NULL); |
18225d19 CB |
1084 | if (ret < 0) { |
1085 | SYSERROR("Failed to set signal mask"); | |
1086 | goto out_warn_father; | |
1087 | } | |
50e98013 | 1088 | |
408da065 | 1089 | /* Don't leak the pinfd to the container. */ |
d39b10eb | 1090 | if (handler->pinfd >= 0) |
0d03360a | 1091 | close(handler->pinfd); |
2b0e17e4 | 1092 | |
18225d19 CB |
1093 | ret = lxc_sync_wait_parent(handler, LXC_SYNC_STARTUP); |
1094 | if (ret < 0) | |
1095 | goto out_warn_father; | |
5b1e83cb | 1096 | |
408da065 CB |
1097 | /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See |
1098 | * https://github.com/lxc/lxd/issues/1978. | |
1099 | */ | |
becad0ec | 1100 | if ((handler->ns_clone_flags & (CLONE_NEWNET | CLONE_NEWUSER)) == |
408da065 | 1101 | (CLONE_NEWNET | CLONE_NEWUSER)) { |
5b1e83cb SH |
1102 | ret = unshare(CLONE_NEWNET); |
1103 | if (ret < 0) { | |
e96a536c | 1104 | SYSERROR("Failed to unshare CLONE_NEWNET"); |
5b1e83cb SH |
1105 | goto out_warn_father; |
1106 | } | |
e96a536c | 1107 | INFO("Unshared CLONE_NEWNET"); |
5b1e83cb SH |
1108 | } |
1109 | ||
408da065 CB |
1110 | /* Tell the parent task it can begin to configure the container and wait |
1111 | * for it to finish. | |
3c22086f | 1112 | */ |
e96a536c CB |
1113 | ret = lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE); |
1114 | if (ret < 0) | |
611ddd34 | 1115 | goto out_error; |
50e98013 | 1116 | |
e96a536c CB |
1117 | ret = lxc_network_recv_veth_names_from_parent(handler); |
1118 | if (ret < 0) { | |
7ab1ba02 | 1119 | ERROR("Failed to receive veth names from parent"); |
658979c5 | 1120 | goto out_warn_father; |
7ab1ba02 | 1121 | } |
658979c5 | 1122 | |
408da065 | 1123 | /* If we are in a new user namespace, become root there to have |
d08f8d2f | 1124 | * privilege over our namespace. |
f6d3e3e4 SH |
1125 | */ |
1126 | if (!lxc_list_empty(&handler->conf->id_map)) { | |
964581c2 CB |
1127 | if (!handler->conf->root_nsuid_map) |
1128 | nsuid = handler->conf->init_uid; | |
1129 | ||
1130 | if (!handler->conf->root_nsgid_map) | |
1131 | nsgid = handler->conf->init_gid; | |
4160c3a0 | 1132 | |
464c4611 | 1133 | if (!lxc_switch_uid_gid(nsuid, nsgid)) |
f6d3e3e4 | 1134 | goto out_warn_father; |
d08f8d2f CB |
1135 | |
1136 | /* Drop groups only after we switched to a valid gid in the new | |
1137 | * user namespace. | |
1138 | */ | |
8af07f82 CB |
1139 | if (!lxc_setgroups(0, NULL) && |
1140 | (handler->am_root || errno != EPERM)) | |
c476bdce | 1141 | goto out_warn_father; |
4b826b1f | 1142 | |
b81689a1 CB |
1143 | ret = prctl(PR_SET_DUMPABLE, prctl_arg(1), prctl_arg(0), |
1144 | prctl_arg(0), prctl_arg(0)); | |
d7883725 CB |
1145 | if (ret < 0) |
1146 | goto out_warn_father; | |
912314fc CB |
1147 | |
1148 | /* set{g,u}id() clears deathsignal */ | |
c7f493ae | 1149 | ret = lxc_set_death_signal(SIGKILL, 0); |
912314fc CB |
1150 | if (ret < 0) { |
1151 | SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL"); | |
1152 | goto out_warn_father; | |
1153 | } | |
f6d3e3e4 SH |
1154 | } |
1155 | ||
e96a536c CB |
1156 | ret = access(handler->lxcpath, X_OK); |
1157 | if (ret != 0) { | |
c8154066 SH |
1158 | print_top_failing_dir(handler->lxcpath); |
1159 | goto out_warn_father; | |
1160 | } | |
1161 | ||
e96a536c CB |
1162 | ret = snprintf(path, sizeof(path), "%s/dev/null", |
1163 | handler->conf->rootfs.mount); | |
408da065 | 1164 | if (ret < 0 || ret >= sizeof(path)) |
7a55c157 | 1165 | goto out_warn_father; |
7a55c157 TA |
1166 | |
1167 | /* In order to checkpoint restore, we need to have everything in the | |
1168 | * same mount namespace. However, some containers may not have a | |
1169 | * reasonable /dev (in particular, they may not have /dev/null), so we | |
1170 | * can't set init's std fds to /dev/null by opening it from inside the | |
1171 | * container. | |
1172 | * | |
1173 | * If that's the case, fall back to using the host's /dev/null. This | |
1174 | * means that migration won't work, but at least we won't spew output | |
1175 | * where it isn't wanted. | |
1176 | */ | |
bb955810 | 1177 | if (handler->daemonize && !handler->conf->autodev) { |
e96a536c CB |
1178 | ret = access(path, F_OK); |
1179 | if (ret != 0) { | |
1180 | devnull_fd = open_devnull(); | |
c44de748 | 1181 | |
e96a536c CB |
1182 | if (devnull_fd < 0) |
1183 | goto out_warn_father; | |
1184 | WARN("Using /dev/null from the host for container " | |
1185 | "init's standard file descriptors. Migration will " | |
1186 | "not work"); | |
1187 | } | |
c44de748 AM |
1188 | } |
1189 | ||
408da065 | 1190 | /* Ask father to setup cgroups and wait for him to finish. */ |
e96a536c CB |
1191 | ret = lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP); |
1192 | if (ret < 0) | |
c44de748 | 1193 | goto out_error; |
544a48a0 | 1194 | |
deefdf8a CB |
1195 | /* Unshare cgroup namespace after we have setup our cgroups. If we do it |
1196 | * earlier we end up with a wrong view of /proc/self/cgroup. For | |
1197 | * example, assume we unshare(CLONE_NEWCGROUP) first, and then create | |
1198 | * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then | |
1199 | * /proc/self/cgroup would show us: | |
1200 | * | |
1201 | * 8:cpuset:/lxc/c | |
1202 | * | |
1203 | * whereas it should actually show | |
1204 | * | |
1205 | * 8:cpuset:/ | |
1206 | */ | |
becad0ec | 1207 | if (handler->ns_clone_flags & CLONE_NEWCGROUP) { |
e96a536c CB |
1208 | ret = unshare(CLONE_NEWCGROUP); |
1209 | if (ret < 0) { | |
1210 | INFO("Failed to unshare CLONE_NEWCGROUP"); | |
deefdf8a CB |
1211 | goto out_warn_father; |
1212 | } | |
e96a536c | 1213 | INFO("Unshared CLONE_NEWCGROUP"); |
deefdf8a CB |
1214 | } |
1215 | ||
149857af CB |
1216 | /* Add the requested environment variables to the current environment to |
1217 | * allow them to be used by the various hooks, such as the start hook | |
1218 | * above. | |
98ff08ca | 1219 | */ |
98ff08ca | 1220 | lxc_list_for_each(iterator, &handler->conf->environment) { |
e96a536c CB |
1221 | ret = putenv((char *)iterator->elem); |
1222 | if (ret < 0) { | |
1223 | SYSERROR("Failed to set environment variable: %s", | |
1224 | (char *)iterator->elem); | |
98ff08ca CB |
1225 | goto out_warn_father; |
1226 | } | |
1227 | } | |
1228 | ||
f4152036 | 1229 | /* Setup the container, ip, names, utsname, ... */ |
3b988b33 | 1230 | ret = lxc_setup(handler); |
7729f8e5 | 1231 | close(handler->data_sock[1]); |
8bf3abfb | 1232 | close(handler->data_sock[0]); |
7729f8e5 | 1233 | if (ret < 0) { |
e96a536c | 1234 | ERROR("Failed to setup container \"%s\"", handler->name); |
f4152036 CB |
1235 | goto out_warn_father; |
1236 | } | |
1237 | ||
408da065 | 1238 | /* Set the label to change to when we exec(2) the container's init. */ |
1800f924 | 1239 | ret = lsm_process_label_set(NULL, handler->conf, true); |
e96a536c | 1240 | if (ret < 0) |
e075f5d9 | 1241 | goto out_warn_father; |
5112cd70 | 1242 | |
029cdff5 | 1243 | /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it |
408da065 CB |
1244 | * before we aren't allowed anymore. |
1245 | */ | |
029cdff5 | 1246 | if (handler->conf->no_new_privs) { |
b81689a1 CB |
1247 | ret = prctl(PR_SET_NO_NEW_PRIVS, prctl_arg(1), prctl_arg(0), |
1248 | prctl_arg(0), prctl_arg(0)); | |
e96a536c CB |
1249 | if (ret < 0) { |
1250 | SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block " | |
1251 | "execve() gainable privileges"); | |
029cdff5 CB |
1252 | goto out_warn_father; |
1253 | } | |
e96a536c CB |
1254 | DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable " |
1255 | "privileges"); | |
029cdff5 CB |
1256 | } |
1257 | ||
0d9acb99 DE |
1258 | /* Some init's such as busybox will set sane tty settings on stdin, |
1259 | * stdout, stderr which it thinks is the console. We already set them | |
1260 | * the way we wanted on the real terminal, and we want init to do its | |
564e31c4 | 1261 | * setup on its console ie. the pty allocated in lxc_terminal_setup() so |
408da065 | 1262 | * make sure that that pty is stdin,stdout,stderr. |
0d9acb99 | 1263 | */ |
5d113f65 | 1264 | if (handler->conf->console.slave >= 0) { |
bb955810 | 1265 | if (handler->daemonize || !handler->conf->is_execute) |
5d113f65 CB |
1266 | ret = set_stdfds(handler->conf->console.slave); |
1267 | else | |
ae6d3913 | 1268 | ret = lxc_terminal_set_stdfds(handler->conf->console.slave); |
5d113f65 | 1269 | if (ret < 0) { |
c5b93afb | 1270 | ERROR("Failed to redirect std{in,out,err} to pty file " |
e96a536c | 1271 | "descriptor %d", handler->conf->console.slave); |
c5b93afb LF |
1272 | goto out_warn_father; |
1273 | } | |
5d113f65 | 1274 | } |
0d9acb99 | 1275 | |
408da065 | 1276 | /* If we mounted a temporary proc, then unmount it now. */ |
5112cd70 | 1277 | tmp_proc_unmount(handler->conf); |
e075f5d9 | 1278 | |
e96a536c CB |
1279 | ret = lxc_seccomp_load(handler->conf); |
1280 | if (ret < 0) | |
8f2c3a70 SH |
1281 | goto out_warn_father; |
1282 | ||
e96a536c CB |
1283 | ret = run_lxc_hooks(handler->name, "start", handler->conf, NULL); |
1284 | if (ret < 0) { | |
1285 | ERROR("Failed to run lxc.hook.start for container \"%s\"", | |
1286 | handler->name); | |
773fb9ca SH |
1287 | goto out_warn_father; |
1288 | } | |
fc25b815 | 1289 | |
773fb9ca | 1290 | close(handler->sigfd); |
26ddeedd | 1291 | |
bb955810 | 1292 | if (handler->conf->console.slave < 0 && handler->daemonize) { |
f4c177c3 CB |
1293 | if (devnull_fd < 0) { |
1294 | devnull_fd = open_devnull(); | |
1295 | if (devnull_fd < 0) | |
1296 | goto out_warn_father; | |
1297 | } | |
1298 | ||
e96a536c CB |
1299 | ret = set_stdfds(devnull_fd); |
1300 | if (ret < 0) { | |
1301 | ERROR("Failed to redirect std{in,out,err} to \"/dev/null\""); | |
c5b93afb LF |
1302 | goto out_warn_father; |
1303 | } | |
e96a536c | 1304 | } |
507cee36 | 1305 | |
c44de748 AM |
1306 | if (devnull_fd >= 0) { |
1307 | close(devnull_fd); | |
1308 | devnull_fd = -1; | |
1309 | } | |
1310 | ||
8c9a7665 TA |
1311 | setsid(); |
1312 | ||
e96a536c CB |
1313 | if (handler->conf->init_cwd) { |
1314 | ret = chdir(handler->conf->init_cwd); | |
1315 | if (ret < 0) { | |
1316 | SYSERROR("Could not change directory to \"%s\"", | |
1317 | handler->conf->init_cwd); | |
1318 | goto out_warn_father; | |
1319 | } | |
3c491553 L |
1320 | } |
1321 | ||
e96a536c CB |
1322 | ret = lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP_LIMITS); |
1323 | if (ret < 0) | |
f4152036 CB |
1324 | goto out_warn_father; |
1325 | ||
149857af CB |
1326 | /* Reset the environment variables the user requested in a clear |
1327 | * environment. | |
1328 | */ | |
e96a536c CB |
1329 | ret = clearenv(); |
1330 | /* Don't error out though. */ | |
1331 | if (ret < 0) | |
149857af | 1332 | SYSERROR("Failed to clear environment."); |
149857af CB |
1333 | |
1334 | lxc_list_for_each(iterator, &handler->conf->environment) { | |
e96a536c CB |
1335 | ret = putenv((char *)iterator->elem); |
1336 | if (ret < 0) { | |
1337 | SYSERROR("Failed to set environment variable: %s", | |
1338 | (char *)iterator->elem); | |
149857af CB |
1339 | goto out_warn_father; |
1340 | } | |
1341 | } | |
1342 | ||
e96a536c CB |
1343 | ret = putenv("container=lxc"); |
1344 | if (ret < 0) { | |
1345 | SYSERROR("Failed to set environment variable: container=lxc"); | |
98ff08ca CB |
1346 | goto out_warn_father; |
1347 | } | |
1348 | ||
885766f5 CB |
1349 | if (handler->conf->ttys.tty_names) { |
1350 | ret = putenv(handler->conf->ttys.tty_names); | |
e96a536c CB |
1351 | if (ret < 0) { |
1352 | SYSERROR("Failed to set environment variable for container ptys"); | |
98ff08ca CB |
1353 | goto out_warn_father; |
1354 | } | |
1355 | } | |
1356 | ||
76bdf299 CB |
1357 | /* The container has been setup. We can now switch to an unprivileged |
1358 | * uid/gid. | |
1359 | */ | |
928b1f04 YT |
1360 | new_uid = handler->conf->init_uid; |
1361 | new_gid = handler->conf->init_gid; | |
76bdf299 | 1362 | |
964581c2 CB |
1363 | /* Avoid unnecessary syscalls. */ |
1364 | if (new_uid == nsuid) | |
1365 | new_uid = LXC_INVALID_UID; | |
1366 | ||
1367 | if (new_gid == nsgid) | |
1368 | new_gid = LXC_INVALID_GID; | |
1369 | ||
464c4611 | 1370 | if (!lxc_switch_uid_gid(new_uid, new_gid)) |
928b1f04 YT |
1371 | goto out_warn_father; |
1372 | ||
8af07f82 CB |
1373 | /* If we are in a new user namespace we already dropped all groups when |
1374 | * we switched to root in the new user namespace further above. Only | |
1375 | * drop groups if we can, so ensure that we have necessary privilege. | |
1376 | */ | |
1377 | if (lxc_list_empty(&handler->conf->id_map)) | |
1378 | #if HAVE_LIBCAP | |
1379 | if (lxc_proc_cap_is_set(CAP_SETGID, CAP_EFFECTIVE)) | |
1380 | #endif | |
1381 | if (!lxc_setgroups(0, NULL)) | |
1382 | goto out_warn_father; | |
1383 | ||
611ddd34 CB |
1384 | ret = lxc_ambient_caps_down(); |
1385 | if (ret < 0) { | |
6f5e532f | 1386 | ERROR("Failed to clear ambient capabilities"); |
611ddd34 CB |
1387 | goto out_warn_father; |
1388 | } | |
1389 | ||
258f8051 | 1390 | if (handler->conf->monitor_signal_pdeath != SIGKILL) { |
c7f493ae | 1391 | ret = lxc_set_death_signal(handler->conf->monitor_signal_pdeath, 0); |
258f8051 CB |
1392 | if (ret < 0) { |
1393 | SYSERROR("Failed to set PR_SET_PDEATHSIG to %d", | |
1394 | handler->conf->monitor_signal_pdeath); | |
1395 | goto out_warn_father; | |
1396 | } | |
1397 | } | |
1398 | ||
408da065 CB |
1399 | /* After this call, we are in error because this ops should not return |
1400 | * as it execs. | |
1401 | */ | |
c4ea60df | 1402 | handler->ops->start(handler, handler->data); |
50e98013 DL |
1403 | |
1404 | out_warn_father: | |
408da065 CB |
1405 | /* We want the parent to know something went wrong, so we return a |
1406 | * special error code. | |
1407 | */ | |
d1ccb562 | 1408 | lxc_sync_wake_parent(handler, LXC_SYNC_ERROR); |
c44de748 AM |
1409 | |
1410 | out_error: | |
1411 | if (devnull_fd >= 0) | |
1412 | close(devnull_fd); | |
1413 | ||
50e98013 DL |
1414 | return -1; |
1415 | } | |
1416 | ||
ae467c54 | 1417 | static int lxc_recv_ttys_from_child(struct lxc_handler *handler) |
e8bd4e43 | 1418 | { |
ae467c54 | 1419 | int i; |
2520facd | 1420 | struct lxc_terminal_info *tty; |
ae467c54 | 1421 | int ret = -1; |
c6012571 | 1422 | int sock = handler->data_sock[1]; |
e8bd4e43 | 1423 | struct lxc_conf *conf = handler->conf; |
0e4be3cf | 1424 | struct lxc_tty_info *ttys = &conf->ttys; |
e8bd4e43 | 1425 | |
885766f5 | 1426 | if (!conf->ttys.max) |
e8bd4e43 SH |
1427 | return 0; |
1428 | ||
885766f5 | 1429 | ttys->tty = malloc(sizeof(*ttys->tty) * ttys->max); |
0e4be3cf | 1430 | if (!ttys->tty) |
e8bd4e43 | 1431 | return -1; |
e8bd4e43 | 1432 | |
885766f5 | 1433 | for (i = 0; i < conf->ttys.max; i++) { |
672c1e58 CB |
1434 | int ttyfds[2]; |
1435 | ||
1436 | ret = lxc_abstract_unix_recv_fds(sock, ttyfds, 2, NULL, 0); | |
1437 | if (ret < 0) | |
1438 | break; | |
ae467c54 | 1439 | |
0e4be3cf | 1440 | tty = &ttys->tty[i]; |
2520facd CB |
1441 | tty->busy = 0; |
1442 | tty->master = ttyfds[0]; | |
1443 | tty->slave = ttyfds[1]; | |
672c1e58 | 1444 | TRACE("Received pty with master fd %d and slave fd %d from " |
2520facd | 1445 | "parent", tty->master, tty->slave); |
e8bd4e43 | 1446 | } |
6d1400b5 | 1447 | |
ae467c54 | 1448 | if (ret < 0) |
6d1400b5 | 1449 | SYSERROR("Failed to receive %zu ttys from child", ttys->max); |
ae467c54 | 1450 | else |
885766f5 | 1451 | TRACE("Received %zu ttys from child", ttys->max); |
ae467c54 CB |
1452 | |
1453 | return ret; | |
e8bd4e43 SH |
1454 | } |
1455 | ||
5af9369b | 1456 | int resolve_clone_flags(struct lxc_handler *handler) |
f813849c | 1457 | { |
8bc8c715 CB |
1458 | int i; |
1459 | struct lxc_conf *conf = handler->conf; | |
f813849c | 1460 | |
8bc8c715 CB |
1461 | for (i = 0; i < LXC_NS_MAX; i++) { |
1462 | if (conf->ns_keep != 0) { | |
1463 | if ((conf->ns_keep & ns_info[i].clone_flag) == 0) | |
becad0ec | 1464 | handler->ns_clone_flags |= ns_info[i].clone_flag; |
8bc8c715 CB |
1465 | } else if (conf->ns_clone != 0) { |
1466 | if ((conf->ns_clone & ns_info[i].clone_flag) > 0) | |
becad0ec | 1467 | handler->ns_clone_flags |= ns_info[i].clone_flag; |
8bc8c715 CB |
1468 | } else { |
1469 | if (i == LXC_NS_USER && lxc_list_empty(&handler->conf->id_map)) | |
1470 | continue; | |
1471 | ||
1472 | if (i == LXC_NS_NET && lxc_requests_empty_network(handler)) | |
1473 | continue; | |
1474 | ||
7bd05339 CB |
1475 | if (i == LXC_NS_CGROUP && !cgns_supported()) |
1476 | continue; | |
1477 | ||
becad0ec | 1478 | handler->ns_clone_flags |= ns_info[i].clone_flag; |
8bc8c715 | 1479 | } |
03df7ab5 | 1480 | |
8bc8c715 CB |
1481 | if (!conf->ns_share[i]) |
1482 | continue; | |
5af9369b | 1483 | |
becad0ec | 1484 | handler->ns_clone_flags &= ~ns_info[i].clone_flag; |
8bc8c715 | 1485 | TRACE("Sharing %s namespace", ns_info[i].proc_name); |
5af9369b CB |
1486 | } |
1487 | ||
1488 | return 0; | |
f813849c TA |
1489 | } |
1490 | ||
8deca6c9 CB |
1491 | /* Note that this function is used with clone(CLONE_VM). Some glibc versions |
1492 | * used to reset the pid/tid to -1 when CLONE_VM was used without CLONE_THREAD. | |
1493 | * But since the memory between parent and child is shared on CLONE_VM this | |
1494 | * would invalidate the getpid() cache that glibc used to maintain and so | |
1495 | * getpid() in the child would return the parent's pid. This is all fixed in | |
1496 | * newer glibc versions where the getpid() cache is removed and the pid/tid is | |
1497 | * not reset anymore. | |
1498 | * However, if for whatever reason you - dear commiter - somehow need to get the | |
1499 | * pid of the dummy intermediate process for do_share_ns() you need to call | |
0059379f CB |
1500 | * lxc_raw_getpid(). The next lxc_raw_clone() call does not employ CLONE_VM and |
1501 | * will be fine. | |
8deca6c9 CB |
1502 | */ |
1503 | static inline int do_share_ns(void *arg) | |
fa3a5b22 | 1504 | { |
8deca6c9 | 1505 | int i, flags, ret; |
fa3a5b22 CB |
1506 | struct lxc_handler *handler = arg; |
1507 | ||
8deca6c9 CB |
1508 | for (i = 0; i < LXC_NS_MAX; i++) { |
1509 | if (handler->nsfd[i] < 0) | |
1510 | continue; | |
fa3a5b22 | 1511 | |
8deca6c9 | 1512 | ret = setns(handler->nsfd[i], 0); |
c0b48eff CB |
1513 | if (ret < 0) { |
1514 | /* | |
1515 | * Note that joining a user and/or mount namespace | |
1516 | * requires the process is not multithreaded otherwise | |
1517 | * setns() will fail here. | |
1518 | */ | |
1519 | SYSERROR("Failed to inherit %s namespace", | |
1520 | ns_info[i].proc_name); | |
8deca6c9 | 1521 | return -1; |
c0b48eff | 1522 | } |
fa3a5b22 | 1523 | |
8deca6c9 | 1524 | DEBUG("Inherited %s namespace", ns_info[i].proc_name); |
fa3a5b22 CB |
1525 | } |
1526 | ||
becad0ec | 1527 | flags = handler->ns_on_clone_flags; |
8deca6c9 | 1528 | flags |= CLONE_PARENT; |
ac2ba696 | 1529 | handler->pid = lxc_raw_clone_cb(do_start, handler, flags); |
8deca6c9 | 1530 | if (handler->pid < 0) |
fa3a5b22 CB |
1531 | return -1; |
1532 | ||
8deca6c9 | 1533 | return 0; |
fa3a5b22 CB |
1534 | } |
1535 | ||
ea0e06dd CB |
1536 | static int lxc_setup_shmount(struct lxc_conf *conf) |
1537 | { | |
4d6cc24c LT |
1538 | size_t len_cont; |
1539 | char *full_cont_path; | |
1540 | int ret = -1; | |
1541 | ||
ea0e06dd | 1542 | /* Construct the shmount path under the container root. */ |
7a41e857 | 1543 | len_cont = strlen(conf->rootfs.mount) + 1 + strlen(conf->shmount.path_cont); |
4d6cc24c LT |
1544 | /* +1 for the terminating '\0' */ |
1545 | full_cont_path = malloc(len_cont + 1); | |
ea0e06dd | 1546 | if (!full_cont_path) { |
4d6cc24c LT |
1547 | SYSERROR("Not enough memory"); |
1548 | return -ENOMEM; | |
1549 | } | |
ea0e06dd CB |
1550 | |
1551 | ret = snprintf(full_cont_path, len_cont + 1, "%s/%s", | |
1552 | conf->rootfs.mount, conf->shmount.path_cont); | |
4d6cc24c LT |
1553 | if (ret < 0 || ret >= len_cont + 1) { |
1554 | SYSERROR("Failed to create filename"); | |
1555 | free(full_cont_path); | |
1556 | return -1; | |
1557 | } | |
1558 | ||
ea0e06dd | 1559 | /* Check if shmount point is already set up. */ |
7a41e857 | 1560 | if (is_shared_mountpoint(conf->shmount.path_host)) { |
ea0e06dd CB |
1561 | INFO("Path \"%s\" is already MS_SHARED. Reusing", |
1562 | conf->shmount.path_host); | |
4d6cc24c LT |
1563 | free(full_cont_path); |
1564 | return 0; | |
1565 | } | |
1566 | ||
1567 | /* Create host and cont mount paths */ | |
7a41e857 | 1568 | ret = mkdir_p(conf->shmount.path_host, 0711); |
4d6cc24c | 1569 | if (ret < 0 && errno != EEXIST) { |
ea0e06dd CB |
1570 | SYSERROR("Failed to create directory \"%s\"", |
1571 | conf->shmount.path_host); | |
4d6cc24c LT |
1572 | free(full_cont_path); |
1573 | return ret; | |
1574 | } | |
1575 | ||
1576 | ret = mkdir_p(full_cont_path, 0711); | |
1577 | if (ret < 0 && errno != EEXIST) { | |
1578 | SYSERROR("Failed to create directory \"%s\"", full_cont_path); | |
1579 | free(full_cont_path); | |
1580 | return ret; | |
1581 | } | |
1582 | ||
1583 | /* Prepare host mountpoint */ | |
ea0e06dd CB |
1584 | ret = mount("tmpfs", conf->shmount.path_host, "tmpfs", 0, |
1585 | "size=100k,mode=0711"); | |
4d6cc24c | 1586 | if (ret < 0) { |
7a41e857 | 1587 | SYSERROR("Failed to mount \"%s\"", conf->shmount.path_host); |
4d6cc24c LT |
1588 | free(full_cont_path); |
1589 | return ret; | |
1590 | } | |
ea0e06dd | 1591 | |
7a41e857 | 1592 | ret = mount(conf->shmount.path_host, conf->shmount.path_host, "none", |
ea0e06dd | 1593 | MS_REC | MS_SHARED, ""); |
4d6cc24c | 1594 | if (ret < 0) { |
7a41e857 | 1595 | SYSERROR("Failed to make shared \"%s\"", conf->shmount.path_host); |
4d6cc24c LT |
1596 | free(full_cont_path); |
1597 | return ret; | |
1598 | } | |
1599 | ||
ea0e06dd | 1600 | INFO("Setup shared mount point \"%s\"", conf->shmount.path_host); |
4d6cc24c LT |
1601 | free(full_cont_path); |
1602 | return 0; | |
1603 | } | |
1604 | ||
480588e6 CB |
1605 | /* lxc_spawn() performs crucial setup tasks and clone()s the new process which |
1606 | * exec()s the requested container binary. | |
1607 | * Note that lxc_spawn() runs in the parent namespaces. Any operations performed | |
1608 | * right here should be double checked if they'd pose a security risk. (For | |
1609 | * example, any {u}mount() operations performed here will be reflected on the | |
1610 | * host!) | |
1611 | */ | |
74a3920a | 1612 | static int lxc_spawn(struct lxc_handler *handler) |
59eb99ba | 1613 | { |
8deca6c9 | 1614 | int i, ret; |
08dd2805 | 1615 | char pidstr[20]; |
57927bf2 CB |
1616 | bool wants_to_map_ids; |
1617 | struct lxc_list *id_map; | |
6e5fc7a5 | 1618 | const char *name = handler->name; |
28d9e29e | 1619 | const char *lxcpath = handler->lxcpath; |
2202afc9 | 1620 | bool share_ns = false; |
28d9e29e | 1621 | struct lxc_conf *conf = handler->conf; |
2202afc9 | 1622 | struct cgroup_ops *cgroup_ops = handler->cgroup_ops; |
9f30a190 | 1623 | |
28d9e29e | 1624 | id_map = &conf->id_map; |
57927bf2 | 1625 | wants_to_map_ids = !lxc_list_empty(id_map); |
2f2623ec | 1626 | |
28d9e29e | 1627 | for (i = 0; i < LXC_NS_MAX; i++) { |
b074bbf1 | 1628 | if (!conf->ns_share[i]) |
28d9e29e CB |
1629 | continue; |
1630 | ||
b074bbf1 | 1631 | handler->nsfd[i] = lxc_inherit_namespace(conf->ns_share[i], lxcpath, ns_info[i].proc_name); |
28d9e29e CB |
1632 | if (handler->nsfd[i] < 0) |
1633 | return -1; | |
fa3a5b22 | 1634 | |
8deca6c9 | 1635 | share_ns = true; |
28d9e29e | 1636 | } |
50e98013 | 1637 | |
cfc62c60 CB |
1638 | ret = lxc_sync_init(handler); |
1639 | if (ret < 0) | |
9d7f9e52 | 1640 | return -1; |
0ad19a3f | 1641 | |
b9f522c5 | 1642 | ret = socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0, |
790255cf | 1643 | handler->data_sock); |
95ef0d7c CB |
1644 | if (ret < 0) |
1645 | goto out_sync_fini; | |
e8bd4e43 | 1646 | |
5af9369b | 1647 | ret = resolve_clone_flags(handler); |
95ef0d7c CB |
1648 | if (ret < 0) |
1649 | goto out_sync_fini; | |
9f30a190 | 1650 | |
78852a0c | 1651 | if (conf->shmount.path_host) { |
95ef0d7c CB |
1652 | if (!conf->shmount.path_cont) |
1653 | goto out_sync_fini; | |
78852a0c JS |
1654 | |
1655 | ret = lxc_setup_shmount(conf); | |
1656 | if (ret < 0) { | |
1657 | ERROR("Failed to setup shared mount point"); | |
95ef0d7c | 1658 | goto out_sync_fini; |
78852a0c JS |
1659 | } |
1660 | } | |
1661 | ||
becad0ec | 1662 | if (handler->ns_clone_flags & CLONE_NEWNET) { |
28d9e29e | 1663 | if (!lxc_list_empty(&conf->network)) { |
26b797f3 | 1664 | |
9f30a190 MM |
1665 | /* Find gateway addresses from the link device, which is |
1666 | * no longer accessible inside the container. Do this | |
1667 | * before creating network interfaces, since goto | |
408da065 CB |
1668 | * out_delete_net does not work before lxc_clone. |
1669 | */ | |
cfc62c60 CB |
1670 | ret = lxc_find_gateway_addresses(handler); |
1671 | if (ret < 0) { | |
1672 | ERROR("Failed to find gateway addresses"); | |
95ef0d7c | 1673 | goto out_sync_fini; |
9f30a190 MM |
1674 | } |
1675 | ||
408da065 CB |
1676 | /* That should be done before the clone because we will |
1677 | * fill the netdev index and use them in the child. | |
9f30a190 | 1678 | */ |
cfc62c60 CB |
1679 | ret = lxc_create_network_priv(handler); |
1680 | if (ret < 0) { | |
1681 | ERROR("Failed to create the network"); | |
78852a0c | 1682 | goto out_delete_net; |
9f30a190 | 1683 | } |
19a26f82 | 1684 | } |
82d5ae15 DL |
1685 | } |
1686 | ||
e8b181f5 | 1687 | if (!cgroup_ops->payload_create(cgroup_ops, handler)) { |
cfc62c60 | 1688 | ERROR("Failed creating cgroups"); |
47d8fb3b CS |
1689 | goto out_delete_net; |
1690 | } | |
1691 | ||
408da065 CB |
1692 | /* If the rootfs is not a blockdev, prevent the container from marking |
1693 | * it readonly. | |
1694 | * If the container is unprivileged then skip rootfs pinning. | |
0c547523 | 1695 | */ |
0ee35059 | 1696 | if (!wants_to_map_ids) { |
28d9e29e | 1697 | handler->pinfd = pin_rootfs(conf->rootfs.path); |
5e32a990 | 1698 | if (handler->pinfd == -1) |
cfc62c60 | 1699 | INFO("Failed to pin the rootfs for container \"%s\"", handler->name); |
5e32a990 | 1700 | } |
0c547523 | 1701 | |
408da065 | 1702 | /* Create a process in a new set of namespaces. */ |
becad0ec CB |
1703 | handler->ns_on_clone_flags = handler->ns_clone_flags; |
1704 | if (handler->ns_clone_flags & CLONE_NEWUSER) { | |
408da065 CB |
1705 | /* If CLONE_NEWUSER and CLONE_NEWNET was requested, we need to |
1706 | * clone a new user namespace first and only later unshare our | |
1707 | * network namespace to ensure that network devices ownership is | |
1708 | * set up correctly. | |
1709 | */ | |
becad0ec | 1710 | handler->ns_on_clone_flags &= ~CLONE_NEWNET; |
408da065 | 1711 | } |
5af9369b | 1712 | /* The cgroup namespace gets unshare()ed not clone()ed. */ |
becad0ec | 1713 | handler->ns_on_clone_flags &= ~CLONE_NEWCGROUP; |
732375f5 | 1714 | |
4e232466 CB |
1715 | if (share_ns) { |
1716 | pid_t attacher_pid; | |
1717 | ||
cfc62c60 CB |
1718 | attacher_pid = lxc_clone(do_share_ns, handler, |
1719 | CLONE_VFORK | CLONE_VM | CLONE_FILES); | |
4e232466 CB |
1720 | if (attacher_pid < 0) { |
1721 | SYSERROR(LXC_CLONE_ERROR); | |
1722 | goto out_delete_net; | |
1723 | } | |
1724 | ||
1725 | ret = wait_for_pid(attacher_pid); | |
1726 | if (ret < 0) { | |
1727 | SYSERROR("Intermediate process failed"); | |
1728 | goto out_delete_net; | |
1729 | } | |
1730 | } else { | |
cfc62c60 | 1731 | handler->pid = lxc_raw_clone_cb(do_start, handler, |
becad0ec | 1732 | handler->ns_on_clone_flags); |
4e232466 CB |
1733 | } |
1734 | if (handler->pid < 0) { | |
1735 | SYSERROR(LXC_CLONE_ERROR); | |
7fef7a06 | 1736 | goto out_delete_net; |
0ad19a3f | 1737 | } |
fa3a5b22 | 1738 | TRACE("Cloned child process %d", handler->pid); |
1bd8d726 | 1739 | |
9662e444 | 1740 | for (i = 0; i < LXC_NS_MAX; i++) |
becad0ec | 1741 | if (handler->ns_on_clone_flags & ns_info[i].clone_flag) |
28d9e29e | 1742 | INFO("Cloned %s", ns_info[i].flag_name); |
0ad19a3f | 1743 | |
4cb53844 | 1744 | if (!lxc_try_preserve_namespaces(handler, handler->ns_on_clone_flags, handler->pid)) { |
28d9e29e CB |
1745 | ERROR("Failed to preserve cloned namespaces for lxc.hook.stop"); |
1746 | goto out_delete_net; | |
1747 | } | |
b6b2b194 | 1748 | |
3c22086f CLG |
1749 | lxc_sync_fini_child(handler); |
1750 | ||
408da065 CB |
1751 | /* Map the container uids. The container became an invalid userid the |
1752 | * moment it was cloned with CLONE_NEWUSER. This call doesn't change | |
1753 | * anything immediately, but allows the container to setuid(0) (0 being | |
1754 | * mapped to something else on the host.) later to become a valid uid | |
1755 | * again. | |
1756 | */ | |
fa3a5b22 | 1757 | if (wants_to_map_ids) { |
df3c5314 | 1758 | if (!handler->conf->ns_share[LXC_NS_USER] && |
46186acd | 1759 | (handler->conf->ns_keep & CLONE_NEWUSER) == 0) { |
fa3a5b22 CB |
1760 | ret = lxc_map_ids(id_map, handler->pid); |
1761 | if (ret < 0) { | |
1762 | ERROR("Failed to set up id mapping."); | |
1763 | goto out_delete_net; | |
1764 | } | |
1765 | } | |
5b1e83cb SH |
1766 | } |
1767 | ||
cfc62c60 CB |
1768 | ret = lxc_sync_wake_child(handler, LXC_SYNC_STARTUP); |
1769 | if (ret < 0) | |
5b1e83cb | 1770 | goto out_delete_net; |
5b1e83cb | 1771 | |
cfc62c60 CB |
1772 | ret = lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE); |
1773 | if (ret < 0) | |
5b1e83cb | 1774 | goto out_delete_net; |
0ad19a3f | 1775 | |
2202afc9 | 1776 | if (!cgroup_ops->setup_limits(cgroup_ops, handler->conf, false)) { |
cfc62c60 | 1777 | ERROR("Failed to setup cgroup limits for container \"%s\"", name); |
6031a6e5 DE |
1778 | goto out_delete_net; |
1779 | } | |
1780 | ||
e8b181f5 | 1781 | if (!cgroup_ops->payload_enter(cgroup_ops, handler->pid)) |
7fef7a06 | 1782 | goto out_delete_net; |
218d4250 | 1783 | |
2202afc9 | 1784 | if (!cgroup_ops->chown(cgroup_ops, handler->conf)) |
0996e18a SH |
1785 | goto out_delete_net; |
1786 | ||
fa3a5b22 | 1787 | /* Now we're ready to preserve the network namespace */ |
4cb53844 | 1788 | ret = lxc_try_preserve_ns(handler->pid, "net"); |
fa3a5b22 | 1789 | if (ret < 0) { |
4cb53844 | 1790 | if (ret != -EOPNOTSUPP) { |
6d1400b5 | 1791 | SYSERROR("Failed to preserve net namespace"); |
4cb53844 CB |
1792 | goto out_delete_net; |
1793 | } | |
1794 | } else { | |
1795 | handler->nsfd[LXC_NS_NET] = ret; | |
1796 | DEBUG("Preserved net namespace via fd %d", ret); | |
1cd95214 CB |
1797 | |
1798 | ret = lxc_netns_set_nsid(handler->nsfd[LXC_NS_NET]); | |
bfcedc7e | 1799 | if (ret < 0) |
b7986d0d | 1800 | SYSERROR("Failed to allocate new network namespace id"); |
bfcedc7e | 1801 | else |
b7986d0d | 1802 | TRACE("Allocated new network namespace id"); |
fa3a5b22 | 1803 | } |
fa3a5b22 | 1804 | |
408da065 | 1805 | /* Create the network configuration. */ |
becad0ec | 1806 | if (handler->ns_clone_flags & CLONE_NEWNET) { |
cfc62c60 CB |
1807 | ret = lxc_network_move_created_netdev_priv(handler->lxcpath, |
1808 | handler->name, | |
1809 | &conf->network, | |
1810 | handler->pid); | |
1811 | if (ret < 0) { | |
1812 | ERROR("Failed to create the configured network"); | |
7fef7a06 | 1813 | goto out_delete_net; |
82d5ae15 | 1814 | } |
74c6e2b0 | 1815 | |
cfc62c60 | 1816 | ret = lxc_create_network_unpriv(handler->lxcpath, handler->name, |
4d781681 | 1817 | &conf->network, handler->pid, conf->hooks_version); |
cfc62c60 CB |
1818 | if (ret < 0) { |
1819 | ERROR("Failed to create the configured network"); | |
74c6e2b0 CB |
1820 | goto out_delete_net; |
1821 | } | |
0ad19a3f | 1822 | } |
1823 | ||
cfc62c60 CB |
1824 | ret = lxc_network_send_veth_names_to_child(handler); |
1825 | if (ret < 0) { | |
7ab1ba02 CB |
1826 | ERROR("Failed to send veth names to child"); |
1827 | goto out_delete_net; | |
658979c5 SH |
1828 | } |
1829 | ||
61d7a733 YT |
1830 | if (!lxc_list_empty(&conf->procs)) { |
1831 | ret = setup_proc_filesystem(&conf->procs, handler->pid); | |
1832 | if (ret < 0) | |
1833 | goto out_delete_net; | |
1834 | } | |
1835 | ||
408da065 CB |
1836 | /* Tell the child to continue its initialization. We'll get |
1837 | * LXC_SYNC_CGROUP when it is ready for us to setup cgroups. | |
3c22086f | 1838 | */ |
cfc62c60 CB |
1839 | ret = lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE); |
1840 | if (ret < 0) | |
544a48a0 SH |
1841 | goto out_delete_net; |
1842 | ||
cfc62c60 CB |
1843 | if (!lxc_list_empty(&conf->limits)) { |
1844 | ret = setup_resource_limits(&conf->limits, handler->pid); | |
1845 | if (ret < 0) { | |
1846 | ERROR("Failed to setup resource limits"); | |
1847 | goto out_delete_net; | |
1848 | } | |
c6d09e15 WB |
1849 | } |
1850 | ||
cfc62c60 CB |
1851 | ret = lxc_sync_barrier_child(handler, LXC_SYNC_CGROUP_UNSHARE); |
1852 | if (ret < 0) | |
f4152036 CB |
1853 | goto out_delete_net; |
1854 | ||
2202afc9 | 1855 | if (!cgroup_ops->setup_limits(cgroup_ops, handler->conf, true)) { |
cfc62c60 | 1856 | ERROR("Failed to setup legacy device cgroup controller limits"); |
b98f7d6e | 1857 | goto out_delete_net; |
544a48a0 | 1858 | } |
cfc62c60 | 1859 | TRACE("Set up legacy device cgroup controller limits"); |
544a48a0 | 1860 | |
becad0ec | 1861 | if (handler->ns_clone_flags & CLONE_NEWCGROUP) { |
8bf3abfb | 1862 | /* Now we're ready to preserve the cgroup namespace */ |
4cb53844 | 1863 | ret = lxc_try_preserve_ns(handler->pid, "cgroup"); |
8bf3abfb | 1864 | if (ret < 0) { |
4cb53844 | 1865 | if (ret != -EOPNOTSUPP) { |
6d1400b5 | 1866 | SYSERROR("Failed to preserve cgroup namespace"); |
4cb53844 CB |
1867 | goto out_delete_net; |
1868 | } | |
1869 | } else { | |
1870 | handler->nsfd[LXC_NS_CGROUP] = ret; | |
1871 | DEBUG("Preserved cgroup namespace via fd %d", ret); | |
8bf3abfb | 1872 | } |
8bf3abfb CB |
1873 | } |
1874 | ||
5b7f756a CB |
1875 | ret = snprintf(pidstr, 20, "%d", handler->pid); |
1876 | if (ret < 0 || ret >= 20) | |
1877 | goto out_delete_net; | |
1878 | ||
cfc62c60 CB |
1879 | ret = setenv("LXC_PID", pidstr, 1); |
1880 | if (ret < 0) | |
1881 | SYSERROR("Failed to set environment variable: LXC_PID=%s", pidstr); | |
08dd2805 SH |
1882 | |
1883 | /* Run any host-side start hooks */ | |
cfc62c60 CB |
1884 | ret = run_lxc_hooks(name, "start-host", conf, NULL); |
1885 | if (ret < 0) { | |
1886 | ERROR("Failed to run lxc.hook.start-host"); | |
341ed84c | 1887 | goto out_delete_net; |
08dd2805 SH |
1888 | } |
1889 | ||
408da065 CB |
1890 | /* Tell the child to complete its initialization and wait for it to exec |
1891 | * or return an error. (The child will never return | |
08dd2805 | 1892 | * LXC_SYNC_READY_START+1. It will either close the sync pipe, causing |
408da065 CB |
1893 | * lxc_sync_barrier_child to return success, or return a different |
1894 | * value, causing us to error out). | |
544a48a0 | 1895 | */ |
cfc62c60 CB |
1896 | ret = lxc_sync_barrier_child(handler, LXC_SYNC_READY_START); |
1897 | if (ret < 0) | |
341ed84c | 1898 | goto out_delete_net; |
0ad19a3f | 1899 | |
cfc62c60 CB |
1900 | ret = lxc_network_recv_name_and_ifindex_from_child(handler); |
1901 | if (ret < 0) { | |
790255cf CB |
1902 | ERROR("Failed to receive names and ifindices for network " |
1903 | "devices from child"); | |
1904 | goto out_delete_net; | |
1905 | } | |
1906 | ||
1907 | /* Now all networks are created, network devices are moved into place, | |
1908 | * and the correct names and ifindeces in the respective namespaces have | |
1909 | * been recorded. The corresponding structs have now all been filled. So | |
1910 | * log them for debugging purposes. | |
1911 | */ | |
28d9e29e | 1912 | lxc_log_configured_netdevs(conf); |
790255cf | 1913 | |
f4152036 | 1914 | /* Read tty fds allocated by child. */ |
cfc62c60 CB |
1915 | ret = lxc_recv_ttys_from_child(handler); |
1916 | if (ret < 0) { | |
1917 | ERROR("Failed to receive tty info from child process"); | |
f4152036 CB |
1918 | goto out_delete_net; |
1919 | } | |
1920 | ||
cfc62c60 CB |
1921 | ret = handler->ops->post_start(handler, handler->data); |
1922 | if (ret < 0) | |
e6126dbe MN |
1923 | goto out_abort; |
1924 | ||
cfc62c60 CB |
1925 | ret = lxc_set_state(name, handler, RUNNING); |
1926 | if (ret < 0) { | |
1927 | ERROR("Failed to set state to \"%s\"", lxc_state2str(RUNNING)); | |
59eb99ba | 1928 | goto out_abort; |
3f21c114 | 1929 | } |
22ebac19 | 1930 | |
3c22086f | 1931 | lxc_sync_fini(handler); |
0c547523 | 1932 | |
e6126dbe | 1933 | return 0; |
1ac470c0 | 1934 | |
7fef7a06 | 1935 | out_delete_net: |
becad0ec | 1936 | if (handler->ns_clone_flags & CLONE_NEWNET) |
bb84beda | 1937 | lxc_delete_network(handler); |
1bd8d726 | 1938 | |
59eb99ba DL |
1939 | out_abort: |
1940 | lxc_abort(name, handler); | |
95ef0d7c CB |
1941 | |
1942 | out_sync_fini: | |
3c22086f | 1943 | lxc_sync_fini(handler); |
5c068da9 SH |
1944 | if (handler->pinfd >= 0) { |
1945 | close(handler->pinfd); | |
1946 | handler->pinfd = -1; | |
1947 | } | |
1948 | ||
b79fcd86 | 1949 | return -1; |
59eb99ba | 1950 | } |
0ad19a3f | 1951 | |
aa460476 | 1952 | int __lxc_start(const char *name, struct lxc_handler *handler, |
507cee36 | 1953 | struct lxc_operations* ops, void *data, const char *lxcpath, |
bb955810 | 1954 | bool daemonize, int *error_num) |
59eb99ba | 1955 | { |
c30e9b19 | 1956 | int ret, status; |
aa460476 | 1957 | struct lxc_conf *conf = handler->conf; |
72068e74 | 1958 | struct cgroup_ops *cgroup_ops; |
80090207 | 1959 | |
c30e9b19 CB |
1960 | ret = lxc_init(name, handler); |
1961 | if (ret < 0) { | |
1962 | ERROR("Failed to initialize container \"%s\"", name); | |
66aeffc7 | 1963 | return -1; |
0ad19a3f | 1964 | } |
ee70bf78 CLG |
1965 | handler->ops = ops; |
1966 | handler->data = data; | |
bb955810 | 1967 | handler->daemonize = daemonize; |
72068e74 | 1968 | cgroup_ops = handler->cgroup_ops; |
e6126dbe | 1969 | |
76a26f55 | 1970 | if (!attach_block_device(handler->conf)) { |
c30e9b19 | 1971 | ERROR("Failed to attach block device"); |
76a26f55 SH |
1972 | goto out_fini_nonet; |
1973 | } | |
1974 | ||
72068e74 CB |
1975 | if (!cgroup_ops->monitor_create(cgroup_ops, handler)) { |
1976 | ERROR("Failed to create monitor cgroup"); | |
eeef32bb CB |
1977 | goto out_fini_nonet; |
1978 | } | |
1979 | ||
434c8e15 | 1980 | if (!cgroup_ops->monitor_enter(cgroup_ops, handler->monitor_pid)) { |
eeef32bb | 1981 | ERROR("Failed to enter monitor cgroup"); |
72068e74 CB |
1982 | goto out_fini_nonet; |
1983 | } | |
1984 | ||
35120d9c | 1985 | if (geteuid() == 0 && !lxc_list_empty(&conf->id_map)) { |
408da065 | 1986 | /* If the backing store is a device, mount it here and now. */ |
35120d9c | 1987 | if (rootfs_is_blockdev(conf)) { |
c30e9b19 CB |
1988 | ret = unshare(CLONE_NEWNS); |
1989 | if (ret < 0) { | |
1990 | ERROR("Failed to unshare CLONE_NEWNS"); | |
35120d9c SH |
1991 | goto out_fini_nonet; |
1992 | } | |
c30e9b19 | 1993 | INFO("Unshared CLONE_NEWNS"); |
408da065 | 1994 | |
6a0c909a | 1995 | remount_all_slave(); |
8ce1abc2 | 1996 | ret = lxc_setup_rootfs_prepare_root(conf, name, lxcpath); |
c30e9b19 CB |
1997 | if (ret < 0) { |
1998 | ERROR("Error setting up rootfs mount as root before spawn"); | |
35120d9c SH |
1999 | goto out_fini_nonet; |
2000 | } | |
c30e9b19 | 2001 | INFO("Set up container rootfs as host root"); |
35120d9c SH |
2002 | } |
2003 | } | |
2004 | ||
c30e9b19 CB |
2005 | ret = lxc_spawn(handler); |
2006 | if (ret < 0) { | |
2007 | ERROR("Failed to spawn container \"%s\"", name); | |
76a26f55 | 2008 | goto out_detach_blockdev; |
0ad19a3f | 2009 | } |
a9e1109e CB |
2010 | /* close parent side of data socket */ |
2011 | close(handler->data_sock[0]); | |
2012 | handler->data_sock[0] = -1; | |
2013 | close(handler->data_sock[1]); | |
2014 | handler->data_sock[1] = -1; | |
0ad19a3f | 2015 | |
80308d07 | 2016 | handler->conf->reboot = REBOOT_NONE; |
8bee8851 | 2017 | |
c30e9b19 CB |
2018 | ret = lxc_poll(name, handler); |
2019 | if (ret) { | |
2020 | ERROR("LXC mainloop exited with error: %d", ret); | |
59eb99ba DL |
2021 | goto out_abort; |
2022 | } | |
0ad19a3f | 2023 | |
321db026 DJ |
2024 | if (!handler->init_died && handler->pid > 0) { |
2025 | ERROR("Child process is not killed"); | |
2026 | goto out_abort; | |
2027 | } | |
2028 | ||
c30e9b19 CB |
2029 | status = lxc_wait_for_pid_status(handler->pid); |
2030 | if (status < 0) | |
2031 | SYSERROR("Failed to retrieve status for %d", handler->pid); | |
e043236e | 2032 | |
408da065 CB |
2033 | /* If the child process exited but was not signaled, it didn't call |
2034 | * reboot. This should mean it was an lxc-execute which simply exited. | |
2035 | * In any case, treat it as a 'halt'. | |
8b004f07 | 2036 | */ |
d028235d | 2037 | if (WIFSIGNALED(status)) { |
8b004f07 SH |
2038 | switch(WTERMSIG(status)) { |
2039 | case SIGINT: /* halt */ | |
c30e9b19 | 2040 | DEBUG("Container \"%s\" is halting", name); |
8b004f07 SH |
2041 | break; |
2042 | case SIGHUP: /* reboot */ | |
c30e9b19 | 2043 | DEBUG("Container \"%s\" is rebooting", name); |
80308d07 | 2044 | handler->conf->reboot = REBOOT_REQ; |
8b004f07 | 2045 | break; |
c2b9bd9e | 2046 | case SIGSYS: /* seccomp */ |
c30e9b19 | 2047 | DEBUG("Container \"%s\" violated its seccomp policy", name); |
c2b9bd9e | 2048 | break; |
8b004f07 | 2049 | default: |
c30e9b19 | 2050 | DEBUG("Unknown exit status for container \"%s\" init %d", name, WTERMSIG(status)); |
8b004f07 SH |
2051 | break; |
2052 | } | |
d028235d | 2053 | } |
828695d9 | 2054 | |
c30e9b19 CB |
2055 | ret = lxc_restore_phys_nics_to_netns(handler); |
2056 | if (ret < 0) | |
b809f232 CB |
2057 | ERROR("Failed to move physical network devices back to parent " |
2058 | "network namespace"); | |
ce5782df | 2059 | |
5c068da9 SH |
2060 | if (handler->pinfd >= 0) { |
2061 | close(handler->pinfd); | |
2062 | handler->pinfd = -1; | |
2063 | } | |
2064 | ||
1787abca | 2065 | lxc_monitor_send_exit_code(name, status, handler->lxcpath); |
eb808539 | 2066 | lxc_error_set_and_log(handler->pid, status); |
a3b4f3d6 TA |
2067 | if (error_num) |
2068 | *error_num = handler->exit_status; | |
1bd8d726 | 2069 | |
9d7f9e52 | 2070 | out_fini: |
bb84beda | 2071 | lxc_delete_network(handler); |
74a2b586 | 2072 | |
76a26f55 SH |
2073 | out_detach_blockdev: |
2074 | detach_block_device(handler->conf); | |
2075 | ||
74a2b586 | 2076 | out_fini_nonet: |
3a0f472d | 2077 | lxc_fini(name, handler); |
c30e9b19 | 2078 | return ret; |
0ad19a3f | 2079 | |
59eb99ba | 2080 | out_abort: |
3a0f472d | 2081 | lxc_abort(name, handler); |
9d7f9e52 | 2082 | goto out_fini; |
0ad19a3f | 2083 | } |
ee70bf78 CLG |
2084 | |
2085 | struct start_args { | |
2086 | char *const *argv; | |
2087 | }; | |
2088 | ||
2089 | static int start(struct lxc_handler *handler, void* data) | |
2090 | { | |
2091 | struct start_args *arg = data; | |
2092 | ||
984984e4 | 2093 | NOTICE("Exec'ing \"%s\"", arg->argv[0]); |
ee70bf78 CLG |
2094 | |
2095 | execvp(arg->argv[0], arg->argv); | |
984984e4 | 2096 | SYSERROR("Failed to exec \"%s\"", arg->argv[0]); |
ee70bf78 CLG |
2097 | return 0; |
2098 | } | |
2099 | ||
2100 | static int post_start(struct lxc_handler *handler, void* data) | |
2101 | { | |
2102 | struct start_args *arg = data; | |
2103 | ||
4c8e880e | 2104 | NOTICE("Started \"%s\" with pid \"%d\"", arg->argv[0], handler->pid); |
ee70bf78 CLG |
2105 | return 0; |
2106 | } | |
2107 | ||
2108 | static struct lxc_operations start_ops = { | |
2109 | .start = start, | |
2110 | .post_start = post_start | |
2111 | }; | |
2112 | ||
aa460476 | 2113 | int lxc_start(const char *name, char *const argv[], struct lxc_handler *handler, |
bb955810 | 2114 | const char *lxcpath, bool daemonize, int *error_num) |
ee70bf78 CLG |
2115 | { |
2116 | struct start_args start_arg = { | |
2117 | .argv = argv, | |
2118 | }; | |
2119 | ||
b2efeb0b | 2120 | TRACE("Doing lxc_start"); |
bb955810 | 2121 | return __lxc_start(name, handler, &start_ops, &start_arg, lxcpath, daemonize, error_num); |
ee70bf78 | 2122 | } |
28272964 CB |
2123 | |
2124 | static void lxc_destroy_container_on_signal(struct lxc_handler *handler, | |
2125 | const char *name) | |
2126 | { | |
2127 | char destroy[MAXPATHLEN]; | |
f01f7975 | 2128 | struct lxc_container *c; |
ae3beac9 CB |
2129 | int ret = 0; |
2130 | bool bret = true; | |
2131 | ||
df31363a | 2132 | if (handler->conf->rootfs.path && handler->conf->rootfs.mount) { |
790255cf | 2133 | bret = do_destroy_container(handler); |
28272964 | 2134 | if (!bret) { |
ae3beac9 | 2135 | ERROR("Error destroying rootfs for container \"%s\"", name); |
28272964 CB |
2136 | return; |
2137 | } | |
2138 | } | |
ae3beac9 | 2139 | INFO("Destroyed rootfs for container \"%s\"", name); |
28272964 CB |
2140 | |
2141 | ret = snprintf(destroy, MAXPATHLEN, "%s/%s", handler->lxcpath, name); | |
2142 | if (ret < 0 || ret >= MAXPATHLEN) { | |
ae3beac9 | 2143 | ERROR("Error destroying directory for container \"%s\"", name); |
28272964 CB |
2144 | return; |
2145 | } | |
2146 | ||
f01f7975 CB |
2147 | c = lxc_container_new(name, handler->lxcpath); |
2148 | if (c) { | |
2149 | if (container_disk_lock(c)) { | |
ae3beac9 | 2150 | INFO("Could not update lxc_snapshots file"); |
f01f7975 CB |
2151 | lxc_container_put(c); |
2152 | } else { | |
2153 | mod_all_rdeps(c, false); | |
2154 | container_disk_unlock(c); | |
2155 | lxc_container_put(c); | |
2156 | } | |
2157 | } | |
2158 | ||
d0fbc7ba | 2159 | if (!handler->am_root) |
94b0a3ac CB |
2160 | ret = userns_exec_full(handler->conf, lxc_rmdir_onedev_wrapper, |
2161 | destroy, "lxc_rmdir_onedev_wrapper"); | |
28272964 CB |
2162 | else |
2163 | ret = lxc_rmdir_onedev(destroy, NULL); | |
2164 | ||
2165 | if (ret < 0) { | |
ae3beac9 | 2166 | ERROR("Error destroying directory for container \"%s\"", name); |
28272964 CB |
2167 | return; |
2168 | } | |
ae3beac9 | 2169 | INFO("Destroyed directory for container \"%s\"", name); |
28272964 CB |
2170 | } |
2171 | ||
2172 | static int lxc_rmdir_onedev_wrapper(void *data) | |
2173 | { | |
2174 | char *arg = (char *) data; | |
2175 | return lxc_rmdir_onedev(arg, NULL); | |
2176 | } | |
2177 | ||
134df645 CB |
2178 | static bool do_destroy_container(struct lxc_handler *handler) |
2179 | { | |
94b0a3ac CB |
2180 | int ret; |
2181 | ||
d0fbc7ba | 2182 | if (!handler->am_root) { |
94b0a3ac CB |
2183 | ret = userns_exec_full(handler->conf, storage_destroy_wrapper, |
2184 | handler->conf, "storage_destroy_wrapper"); | |
2185 | if (ret < 0) | |
d028235d | 2186 | return false; |
10bc1861 | 2187 | |
d028235d SG |
2188 | return true; |
2189 | } | |
10bc1861 | 2190 | |
790255cf | 2191 | return storage_destroy(handler->conf); |
28272964 | 2192 | } |