]>
Commit | Line | Data |
---|---|---|
cc73685d | 1 | /* SPDX-License-Identifier: LGPL-2.1+ */ |
0ad19a3f | 2 | |
d38dd64a CB |
3 | #ifndef _GNU_SOURCE |
4 | #define _GNU_SOURCE 1 | |
5 | #endif | |
0ad19a3f | 6 | #include <dirent.h> |
7 | #include <errno.h> | |
b0a33c1e | 8 | #include <fcntl.h> |
c476bdce | 9 | #include <grp.h> |
37515ebd | 10 | #include <poll.h> |
b467714b | 11 | #include <pthread.h> |
f3787121 CB |
12 | #include <signal.h> |
13 | #include <stdio.h> | |
14 | #include <stdlib.h> | |
15 | #include <string.h> | |
0ad19a3f | 16 | #include <sys/file.h> |
f4d507d5 | 17 | #include <sys/mount.h> |
f3787121 | 18 | #include <sys/param.h> |
0ad19a3f | 19 | #include <sys/prctl.h> |
f3787121 CB |
20 | #include <sys/socket.h> |
21 | #include <sys/stat.h> | |
22 | #include <sys/syscall.h> | |
ddceb1f9 | 23 | #include <sys/types.h> |
b0a33c1e | 24 | #include <sys/un.h> |
f3787121 | 25 | #include <sys/wait.h> |
d38dd64a | 26 | #include <unistd.h> |
495d2046 | 27 | |
f3787121 CB |
28 | #include "af_unix.h" |
29 | #include "caps.h" | |
563f2f2c | 30 | #include "cgroup.h" |
f3787121 | 31 | #include "commands.h" |
bbf5cf35 | 32 | #include "commands_utils.h" |
f3787121 | 33 | #include "conf.h" |
d38dd64a | 34 | #include "config.h" |
74c6e2b0 | 35 | #include "confile_utils.h" |
e2bcd7db | 36 | #include "error.h" |
7fbb15ec | 37 | #include "file_utils.h" |
1cc8bd4b | 38 | #include "list.h" |
f3787121 | 39 | #include "log.h" |
d38dd64a | 40 | #include "lsm/lsm.h" |
f2d5a09d | 41 | #include "lxccontainer.h" |
f3787121 CB |
42 | #include "lxclock.h" |
43 | #include "lxcseccomp.h" | |
3ef9b3d3 | 44 | #include "macro.h" |
f3787121 | 45 | #include "mainloop.h" |
4ffeaf27 | 46 | #include "memory_utils.h" |
63376d7d | 47 | #include "monitor.h" |
f549edcc | 48 | #include "namespace.h" |
811ef482 | 49 | #include "network.h" |
f40988c7 | 50 | #include "process_utils.h" |
f3787121 | 51 | #include "start.h" |
6be5397b CB |
52 | #include "storage/storage.h" |
53 | #include "storage/storage_utils.h" | |
0ed9b1bc | 54 | #include "sync.h" |
59524108 | 55 | #include "syscall_wrappers.h" |
0ed9b1bc CB |
56 | #include "terminal.h" |
57 | #include "utils.h" | |
36eb9bde | 58 | |
d38dd64a CB |
59 | #if HAVE_LIBCAP |
60 | #include <sys/capability.h> | |
61 | #endif | |
62 | ||
9de31d5a CB |
63 | #ifndef HAVE_STRLCPY |
64 | #include "include/strlcpy.h" | |
65 | #endif | |
66 | ||
ac2cecc4 | 67 | lxc_log_define(start, lxc); |
36eb9bde | 68 | |
f01f7975 | 69 | extern void mod_all_rdeps(struct lxc_container *c, bool inc); |
790255cf | 70 | static bool do_destroy_container(struct lxc_handler *handler); |
28272964 CB |
71 | static int lxc_rmdir_onedev_wrapper(void *data); |
72 | static void lxc_destroy_container_on_signal(struct lxc_handler *handler, | |
73 | const char *name); | |
74 | ||
c8154066 SH |
75 | static void print_top_failing_dir(const char *path) |
76 | { | |
dcd9a847 | 77 | __do_free char *copy = NULL; |
9e5f5f2f | 78 | int ret; |
4ffeaf27 | 79 | char *e, *p, saved; |
cbb9c7c7 | 80 | |
4ffeaf27 | 81 | copy = must_copy_string(path); |
c8154066 | 82 | p = copy; |
4ffeaf27 | 83 | e = copy + strlen(path); |
cbb9c7c7 | 84 | |
c8154066 | 85 | while (p < e) { |
f3787121 CB |
86 | while (p < e && *p == '/') |
87 | p++; | |
9e5f5f2f | 88 | |
f3787121 CB |
89 | while (p < e && *p != '/') |
90 | p++; | |
9e5f5f2f | 91 | |
c8154066 SH |
92 | saved = *p; |
93 | *p = '\0'; | |
9e5f5f2f CB |
94 | |
95 | ret = access(copy, X_OK); | |
96 | if (ret != 0) { | |
818a57fc | 97 | SYSERROR("Could not access %s. Please grant it x access, or add an ACL for the container " "root", copy); |
c8154066 SH |
98 | return; |
99 | } | |
100 | *p = saved; | |
101 | } | |
102 | } | |
103 | ||
5032bf39 | 104 | static void lxc_put_nsfds(struct lxc_handler *handler) |
408da065 | 105 | { |
818a57fc | 106 | for (int i = 0; i < LXC_NS_MAX; i++) { |
5032bf39 | 107 | if (handler->nsfd[i] < 0) |
39cd919c CB |
108 | continue; |
109 | ||
818a57fc | 110 | close_prot_errno_disarm(handler->nsfd[i]); |
9f30a190 | 111 | } |
9f30a190 MM |
112 | } |
113 | ||
4cb53844 CB |
114 | static int lxc_try_preserve_ns(const int pid, const char *ns) |
115 | { | |
116 | int fd; | |
117 | ||
118 | fd = lxc_preserve_ns(pid, ns); | |
119 | if (fd < 0) { | |
818a57fc CB |
120 | if (errno != ENOENT) |
121 | return log_error_errno(-EINVAL, | |
122 | errno, "Failed to preserve %s namespace", | |
123 | ns); | |
4cb53844 | 124 | |
b87ee312 CB |
125 | return log_warn_errno(-EOPNOTSUPP, |
126 | errno, "Kernel does not support preserving %s namespaces", | |
127 | ns); | |
4cb53844 CB |
128 | } |
129 | ||
130 | return fd; | |
131 | } | |
132 | ||
133 | /* lxc_try_preserve_namespaces: open /proc/@pid/ns/@ns for each namespace | |
134 | * specified in ns_clone_flags. | |
4d8ac866 | 135 | * Return true on success, false on failure. |
62d05d9b | 136 | */ |
4cb53844 CB |
137 | static bool lxc_try_preserve_namespaces(struct lxc_handler *handler, |
138 | int ns_clone_flags, pid_t pid) | |
f3787121 | 139 | { |
9fef3355 | 140 | int i; |
9f30a190 | 141 | |
9f30a190 | 142 | for (i = 0; i < LXC_NS_MAX; i++) |
5032bf39 | 143 | handler->nsfd[i] = -EBADF; |
cd43d2d1 | 144 | |
9f30a190 | 145 | for (i = 0; i < LXC_NS_MAX; i++) { |
4cb53844 CB |
146 | int fd; |
147 | ||
becad0ec | 148 | if ((ns_clone_flags & ns_info[i].clone_flag) == 0) |
9f30a190 | 149 | continue; |
28d9e29e | 150 | |
4cb53844 CB |
151 | fd = lxc_try_preserve_ns(pid, ns_info[i].proc_name); |
152 | if (fd < 0) { | |
4cb53844 CB |
153 | /* Do not fail to start container on kernels that do |
154 | * not support interacting with namespaces through | |
155 | * /proc. | |
156 | */ | |
157 | if (fd == -EOPNOTSUPP) | |
158 | continue; | |
fa3a5b22 | 159 | |
4cb53844 CB |
160 | lxc_put_nsfds(handler); |
161 | return false; | |
162 | } | |
163 | ||
164 | handler->nsfd[i] = fd; | |
165 | DEBUG("Preserved %s namespace via fd %d", ns_info[i].proc_name, | |
166 | handler->nsfd[i]); | |
9f30a190 MM |
167 | } |
168 | ||
62d05d9b | 169 | return true; |
9f30a190 MM |
170 | } |
171 | ||
ec1dc633 | 172 | static inline bool match_stdfds(int fd) |
80090207 | 173 | { |
ec1dc633 | 174 | return (fd == STDIN_FILENO || fd == STDOUT_FILENO || fd == STDERR_FILENO); |
80090207 CLG |
175 | } |
176 | ||
4c03724e | 177 | #ifdef HAVE_DLOG |
178 | static bool match_dlog_fds(struct dirent *direntp) | |
179 | { | |
180 | char path[PATH_MAX] = {0}; | |
181 | char link[PATH_MAX] = {0}; | |
182 | ssize_t linklen; | |
183 | int ret; | |
184 | ||
185 | ret = snprintf(path, PATH_MAX, "/proc/self/fd/%s", direntp->d_name); | |
818a57fc CB |
186 | if (ret < 0 || ret >= PATH_MAX) |
187 | return log_error(false, "Failed to create file descriptor name"); | |
4c03724e | 188 | |
189 | linklen = readlink(path, link, PATH_MAX); | |
818a57fc CB |
190 | if (linklen < 0) |
191 | return log_error(false, "Failed to read link path - \"%s\"", path); | |
192 | else if (linklen >= PATH_MAX) | |
193 | return log_error(false, "The name of link path is too long - \"%s\"", path); | |
194 | ||
195 | if (strcmp(link, "/dev/log_main") == 0 || | |
196 | strcmp(link, "/dev/log_system") == 0 || | |
197 | strcmp(link, "/dev/log_radio") == 0) | |
4c03724e | 198 | return true; |
199 | ||
200 | return false; | |
201 | } | |
202 | #endif | |
203 | ||
47a46cf1 CB |
204 | int lxc_check_inherited(struct lxc_conf *conf, bool closeall, |
205 | int *fds_to_ignore, size_t len_fds) | |
80090207 | 206 | { |
80090207 | 207 | int fd, fddir; |
47a46cf1 | 208 | size_t i; |
80090207 | 209 | DIR *dir; |
a5a70219 | 210 | struct dirent *direntp; |
80090207 | 211 | |
d2cf4c37 SH |
212 | if (conf && conf->close_all_fds) |
213 | closeall = true; | |
214 | ||
4e2d6b9a CB |
215 | /* |
216 | * Disable syslog at this point to avoid the above logging | |
217 | * function to open a new fd and make the check_inherited function | |
218 | * enter an infinite loop. | |
219 | */ | |
220 | lxc_log_syslog_disable(); | |
221 | ||
b119f362 | 222 | restart: |
80090207 | 223 | dir = opendir("/proc/self/fd"); |
818a57fc CB |
224 | if (!dir) |
225 | return log_warn(-1, "Failed to open directory"); | |
80090207 CLG |
226 | |
227 | fddir = dirfd(dir); | |
228 | ||
74f96976 | 229 | while ((direntp = readdir(dir))) { |
a5a70219 | 230 | int ret; |
d39b10eb CB |
231 | struct lxc_list *cur; |
232 | bool matched = false; | |
233 | ||
a5a70219 | 234 | if (strcmp(direntp->d_name, ".") == 0) |
80090207 CLG |
235 | continue; |
236 | ||
a5a70219 | 237 | if (strcmp(direntp->d_name, "..") == 0) |
80090207 CLG |
238 | continue; |
239 | ||
a5a70219 CB |
240 | ret = lxc_safe_int(direntp->d_name, &fd); |
241 | if (ret < 0) { | |
242 | INFO("Could not parse file descriptor for \"%s\"", direntp->d_name); | |
d4cff0d2 CB |
243 | continue; |
244 | } | |
80090207 | 245 | |
47a46cf1 CB |
246 | for (i = 0; i < len_fds; i++) |
247 | if (fds_to_ignore[i] == fd) | |
248 | break; | |
249 | ||
250 | if (fd == fddir || fd == lxc_log_fd || | |
251 | (i < len_fds && fd == fds_to_ignore[i])) | |
80090207 CLG |
252 | continue; |
253 | ||
d39b10eb | 254 | /* Keep state clients that wait on reboots. */ |
3ee9e4fb CB |
255 | if (conf) { |
256 | lxc_list_for_each(cur, &conf->state_clients) { | |
257 | struct lxc_state_client *client = cur->elem; | |
d39b10eb | 258 | |
3ee9e4fb CB |
259 | if (client->clientfd != fd) |
260 | continue; | |
d39b10eb | 261 | |
3ee9e4fb CB |
262 | matched = true; |
263 | break; | |
264 | } | |
d39b10eb CB |
265 | } |
266 | ||
267 | if (matched) | |
268 | continue; | |
269 | ||
858377e4 SH |
270 | if (current_config && fd == current_config->logfd) |
271 | continue; | |
272 | ||
ec1dc633 | 273 | if (match_stdfds(fd)) |
80090207 | 274 | continue; |
80090207 | 275 | |
1d5e5f26 | 276 | #ifdef HAVE_DLOG |
277 | if (match_dlog_fds(direntp)) | |
278 | continue; | |
279 | ||
280 | #endif | |
d2cf4c37 | 281 | if (closeall) { |
4e2d6b9a CB |
282 | if (close(fd)) |
283 | SYSINFO("Closed inherited fd %d", fd); | |
284 | else | |
285 | INFO("Closed inherited fd %d", fd); | |
b119f362 | 286 | closedir(dir); |
b119f362 SH |
287 | goto restart; |
288 | } | |
28d9e29e | 289 | WARN("Inherited fd %d", fd); |
80090207 | 290 | } |
4e2d6b9a | 291 | closedir(dir); |
80090207 | 292 | |
4e2d6b9a CB |
293 | /* |
294 | * Only enable syslog at this point to avoid the above logging | |
295 | * function to open a new fd and make the check_inherited function | |
296 | * enter an infinite loop. | |
64c57ea1 | 297 | */ |
4e2d6b9a | 298 | lxc_log_syslog_enable(); |
64c57ea1 | 299 | |
92c7f629 | 300 | return 0; |
80090207 CLG |
301 | } |
302 | ||
83ee7875 | 303 | static int setup_signal_fd(sigset_t *oldmask) |
b0a33c1e | 304 | { |
7288dfb6 | 305 | int ret; |
b0a33c1e | 306 | sigset_t mask; |
7288dfb6 | 307 | const int signals[] = {SIGBUS, SIGILL, SIGSEGV, SIGWINCH}; |
b0a33c1e | 308 | |
408da065 | 309 | /* Block everything except serious error signals. */ |
df0795b1 CB |
310 | ret = sigfillset(&mask); |
311 | if (ret < 0) | |
312 | return -EBADF; | |
313 | ||
9b10bef5 | 314 | for (int sig = 0; sig < (sizeof(signals) / sizeof(signals[0])); sig++) { |
df0795b1 CB |
315 | ret = sigdelset(&mask, signals[sig]); |
316 | if (ret < 0) | |
317 | return -EBADF; | |
b0a33c1e | 318 | } |
319 | ||
b467714b | 320 | ret = pthread_sigmask(SIG_BLOCK, &mask, oldmask); |
818a57fc CB |
321 | if (ret < 0) |
322 | return log_error_errno(-EBADF, errno, | |
323 | "Failed to set signal mask"); | |
b0a33c1e | 324 | |
df0795b1 | 325 | ret = signalfd(-1, &mask, SFD_CLOEXEC); |
818a57fc CB |
326 | if (ret < 0) |
327 | return log_error_errno(-EBADF, | |
328 | errno, "Failed to create signal file descriptor"); | |
b0a33c1e | 329 | |
df0795b1 | 330 | TRACE("Created signal file descriptor %d", ret); |
1ac470c0 | 331 | |
df0795b1 | 332 | return ret; |
b0a33c1e | 333 | } |
334 | ||
84c92abd | 335 | static int signal_handler(int fd, uint32_t events, void *data, |
f3787121 | 336 | struct lxc_epoll_descr *descr) |
b0a33c1e | 337 | { |
15cd25fd | 338 | int ret; |
3c319edb CB |
339 | siginfo_t info; |
340 | struct signalfd_siginfo siginfo; | |
341 | struct lxc_handler *hdlr = data; | |
15cd25fd | 342 | |
489f39be | 343 | ret = lxc_read_nointr(fd, &siginfo, sizeof(siginfo)); |
818a57fc CB |
344 | if (ret < 0) |
345 | return log_error(LXC_MAINLOOP_ERROR, "Failed to read signal info from signal file descriptor %d", fd); | |
15cd25fd | 346 | |
818a57fc | 347 | if (ret != sizeof(siginfo)) |
f7a97743 | 348 | return log_error(LXC_MAINLOOP_ERROR, "Unexpected size for struct signalfd_siginfo"); |
15cd25fd | 349 | |
408da065 | 350 | /* Check whether init is running. */ |
80507ee8 | 351 | info.si_pid = 0; |
3c319edb CB |
352 | ret = waitid(P_PID, hdlr->pid, &info, WEXITED | WNOWAIT | WNOHANG); |
353 | if (ret == 0 && info.si_pid == hdlr->pid) | |
354 | hdlr->init_died = true; | |
80507ee8 | 355 | |
cd5177e9 TA |
356 | /* Try to figure out a reasonable exit status to report. */ |
357 | if (hdlr->init_died) { | |
358 | switch (info.si_code) { | |
359 | case CLD_EXITED: | |
360 | hdlr->exit_status = info.si_status << 8; | |
361 | break; | |
362 | case CLD_KILLED: | |
363 | case CLD_DUMPED: | |
364 | case CLD_STOPPED: | |
365 | hdlr->exit_status = info.si_status << 8 | 0x7f; | |
366 | break; | |
367 | case CLD_CONTINUED: | |
368 | /* Huh? The waitid() told us it's dead *and* continued? */ | |
369 | WARN("Init %d dead and continued?", hdlr->pid); | |
370 | hdlr->exit_status = 1; | |
371 | break; | |
372 | default: | |
20993a97 | 373 | ERROR("Unknown si_code: %d", info.si_code); |
55e1311e | 374 | hdlr->exit_status = 1; |
cd5177e9 TA |
375 | } |
376 | } | |
377 | ||
d4b5d7a8 | 378 | if (siginfo.ssi_signo == SIGHUP) { |
33942046 CB |
379 | if (hdlr->pidfd >= 0) |
380 | lxc_raw_pidfd_send_signal(hdlr->pidfd, SIGTERM, NULL, 0); | |
d9bb2fba CB |
381 | else |
382 | kill(hdlr->pid, SIGTERM); | |
d4b5d7a8 | 383 | INFO("Killing %d since terminal hung up", hdlr->pid); |
9b10bef5 CB |
384 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE |
385 | : LXC_MAINLOOP_CONTINUE; | |
d4b5d7a8 TA |
386 | } |
387 | ||
9cb94384 | 388 | if (siginfo.ssi_signo != SIGCHLD) { |
33942046 CB |
389 | if (hdlr->pidfd >= 0) |
390 | lxc_raw_pidfd_send_signal(hdlr->pidfd, | |
391 | siginfo.ssi_signo, NULL, 0); | |
d9bb2fba CB |
392 | else |
393 | kill(hdlr->pid, siginfo.ssi_signo); | |
9cb94384 | 394 | INFO("Forwarded signal %d to pid %d", siginfo.ssi_signo, hdlr->pid); |
9b10bef5 CB |
395 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE |
396 | : LXC_MAINLOOP_CONTINUE; | |
9cb94384 TA |
397 | } |
398 | ||
3a9e949f TA |
399 | /* More robustness, protect ourself from a SIGCHLD sent |
400 | * by a process different from the container init. | |
401 | */ | |
402 | if (siginfo.ssi_pid != hdlr->pid) { | |
6e94162a CB |
403 | NOTICE("Received %d from pid %d instead of container init %d", |
404 | siginfo.ssi_signo, siginfo.ssi_pid, hdlr->pid); | |
9b10bef5 CB |
405 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE |
406 | : LXC_MAINLOOP_CONTINUE; | |
3a9e949f TA |
407 | } |
408 | ||
408da065 | 409 | if (siginfo.ssi_code == CLD_STOPPED) { |
6e94162a | 410 | INFO("Container init process was stopped"); |
9b10bef5 CB |
411 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE |
412 | : LXC_MAINLOOP_CONTINUE; | |
413 | } | |
414 | ||
415 | if (siginfo.ssi_code == CLD_CONTINUED) { | |
6e94162a | 416 | INFO("Container init process was continued"); |
9b10bef5 CB |
417 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE |
418 | : LXC_MAINLOOP_CONTINUE; | |
15cd25fd | 419 | } |
1ac470c0 | 420 | |
818a57fc | 421 | return log_debug(LXC_MAINLOOP_CLOSE, "Container init process %d exited", hdlr->pid); |
b0a33c1e | 422 | } |
423 | ||
974a8aba CB |
424 | int lxc_serve_state_clients(const char *name, struct lxc_handler *handler, |
425 | lxc_state_t state) | |
66aeffc7 | 426 | { |
9de31d5a | 427 | size_t retlen; |
dbc9832d CB |
428 | ssize_t ret; |
429 | struct lxc_list *cur, *next; | |
dbc9832d CB |
430 | struct lxc_msg msg = {.type = lxc_msg_state, .value = state}; |
431 | ||
974a8aba CB |
432 | if (state == THAWED) |
433 | handler->state = RUNNING; | |
434 | else | |
435 | handler->state = state; | |
436 | ||
9dfa4041 | 437 | TRACE("Set container state to %s", lxc_state2str(state)); |
dbc9832d | 438 | |
818a57fc CB |
439 | if (lxc_list_empty(&handler->conf->state_clients)) |
440 | return log_trace(0, "No state clients registered"); | |
dbc9832d | 441 | |
9de31d5a CB |
442 | retlen = strlcpy(msg.name, name, sizeof(msg.name)); |
443 | if (retlen >= sizeof(msg.name)) | |
444 | return -E2BIG; | |
dbc9832d | 445 | |
d39b10eb | 446 | lxc_list_for_each_safe(cur, &handler->conf->state_clients, next) { |
d85617bc | 447 | struct lxc_state_client *client = cur->elem; |
dbc9832d | 448 | |
d39b10eb | 449 | if (client->states[state] == 0) { |
9dfa4041 | 450 | TRACE("State %s not registered for state client %d", |
dbc9832d CB |
451 | lxc_state2str(state), client->clientfd); |
452 | continue; | |
453 | } | |
454 | ||
9dfa4041 | 455 | TRACE("Sending state %s to state client %d", |
dbc9832d CB |
456 | lxc_state2str(state), client->clientfd); |
457 | ||
7fbb15ec CB |
458 | ret = lxc_send_nointr(client->clientfd, &msg, sizeof(msg), MSG_NOSIGNAL); |
459 | if (ret <= 0) | |
6d1400b5 | 460 | SYSERROR("Failed to send message to client"); |
dbc9832d CB |
461 | |
462 | /* kick client from list */ | |
dbc9832d | 463 | lxc_list_del(cur); |
300d1cb4 | 464 | close(client->clientfd); |
dbc9832d CB |
465 | free(cur->elem); |
466 | free(cur); | |
467 | } | |
dbc9832d | 468 | |
5e5576a4 CB |
469 | return 0; |
470 | } | |
471 | ||
472 | static int lxc_serve_state_socket_pair(const char *name, | |
473 | struct lxc_handler *handler, | |
474 | lxc_state_t state) | |
475 | { | |
476 | ssize_t ret; | |
477 | ||
bb955810 | 478 | if (!handler->daemonize || |
5e5576a4 CB |
479 | handler->state_socket_pair[1] < 0 || |
480 | state == STARTING) | |
481 | return 0; | |
482 | ||
483 | /* Close read end of the socket pair. */ | |
a783a414 | 484 | close_prot_errno_disarm(handler->state_socket_pair[0]); |
5e5576a4 | 485 | |
ee8377bd | 486 | again: |
5e5576a4 CB |
487 | ret = lxc_abstract_unix_send_credential(handler->state_socket_pair[1], |
488 | &(int){state}, sizeof(int)); | |
9044b79e | 489 | if (ret < 0) { |
490 | SYSERROR("Failed to send state to %d", handler->state_socket_pair[1]); | |
491 | ||
ee8377bd CB |
492 | if (errno == EINTR) |
493 | goto again; | |
9044b79e | 494 | |
495 | return -1; | |
496 | } | |
497 | ||
818a57fc CB |
498 | if (ret != sizeof(int)) |
499 | return log_error(-1, "Message too long : %d", handler->state_socket_pair[1]); | |
5e5576a4 CB |
500 | |
501 | TRACE("Sent container state \"%s\" to %d", lxc_state2str(state), | |
502 | handler->state_socket_pair[1]); | |
503 | ||
504 | /* Close write end of the socket pair. */ | |
818a57fc | 505 | close_prot_errno_disarm(handler->state_socket_pair[1]); |
5e5576a4 CB |
506 | |
507 | return 0; | |
508 | } | |
509 | ||
510 | int lxc_set_state(const char *name, struct lxc_handler *handler, | |
511 | lxc_state_t state) | |
512 | { | |
513 | int ret; | |
514 | ||
515 | ret = lxc_serve_state_socket_pair(name, handler, state); | |
818a57fc CB |
516 | if (ret < 0) |
517 | return log_error(-1, "Failed to synchronize via anonymous pair of unix sockets"); | |
5e5576a4 CB |
518 | |
519 | ret = lxc_serve_state_clients(name, handler, state); | |
520 | if (ret < 0) | |
521 | return -1; | |
522 | ||
dbc9832d CB |
523 | /* This function will try to connect to the legacy lxc-monitord state |
524 | * server and only exists for backwards compatibility. | |
525 | */ | |
9123e471 | 526 | lxc_monitor_send_state(name, state, handler->lxcpath); |
dbc9832d | 527 | |
66aeffc7 DL |
528 | return 0; |
529 | } | |
530 | ||
735f2c6e | 531 | int lxc_poll(const char *name, struct lxc_handler *handler) |
b0a33c1e | 532 | { |
3c319edb | 533 | int ret; |
86530b0a | 534 | bool has_console = true; |
3c319edb | 535 | struct lxc_epoll_descr descr, descr_console; |
b0a33c1e | 536 | |
03760215 CB |
537 | if (handler->conf->console.path && |
538 | strcmp(handler->conf->console.path, "none") == 0) | |
86530b0a L |
539 | has_console = false; |
540 | ||
3c319edb CB |
541 | ret = lxc_mainloop_open(&descr); |
542 | if (ret < 0) { | |
543 | ERROR("Failed to create mainloop"); | |
50c8bf05 | 544 | goto out_sigfd; |
b0a33c1e | 545 | } |
546 | ||
30a33fbd CB |
547 | if (has_console) { |
548 | ret = lxc_mainloop_open(&descr_console); | |
549 | if (ret < 0) { | |
550 | ERROR("Failed to create console mainloop"); | |
551 | goto out_mainloop; | |
552 | } | |
3c319edb CB |
553 | } |
554 | ||
1cc8bd4b | 555 | ret = lxc_mainloop_add_handler(&descr, handler->sigfd, signal_handler, handler); |
3c319edb | 556 | if (ret < 0) { |
1cc8bd4b | 557 | ERROR("Failed to add signal handler for %d to mainloop", handler->sigfd); |
3c319edb | 558 | goto out_mainloop_console; |
b0a33c1e | 559 | } |
560 | ||
2ac0f627 CB |
561 | ret = lxc_seccomp_setup_proxy(&handler->conf->seccomp, &descr, handler); |
562 | if (ret < 0) { | |
563 | ERROR("Failed to setup seccomp proxy"); | |
c3e3c21a | 564 | goto out_mainloop_console; |
2ac0f627 | 565 | } |
cdb2a47f | 566 | |
30a33fbd | 567 | if (has_console) { |
dcad02f8 | 568 | struct lxc_terminal *console = &handler->conf->console; |
63376d7d | 569 | |
093bce5f | 570 | ret = lxc_terminal_mainloop_add(&descr, console); |
30a33fbd CB |
571 | if (ret < 0) { |
572 | ERROR("Failed to add console handlers to mainloop"); | |
573 | goto out_mainloop_console; | |
574 | } | |
575 | ||
093bce5f | 576 | ret = lxc_terminal_mainloop_add(&descr_console, console); |
30a33fbd CB |
577 | if (ret < 0) { |
578 | ERROR("Failed to add console handlers to console mainloop"); | |
579 | goto out_mainloop_console; | |
580 | } | |
a54585ad L |
581 | |
582 | handler->conf->console.descr = &descr; | |
563f2f2c DL |
583 | } |
584 | ||
3c319edb CB |
585 | ret = lxc_cmd_mainloop_add(name, &descr, handler); |
586 | if (ret < 0) { | |
587 | ERROR("Failed to add command handler to mainloop"); | |
588 | goto out_mainloop_console; | |
589 | } | |
b0a33c1e | 590 | |
3c319edb | 591 | TRACE("Mainloop is ready"); |
b0a33c1e | 592 | |
3c319edb | 593 | ret = lxc_mainloop(&descr, -1); |
a783a414 | 594 | close_prot_errno_disarm(descr.epfd); |
1cc8bd4b | 595 | if (ret < 0 || !handler->init_died) |
42b09f94 | 596 | goto out_mainloop_console; |
1cc8bd4b | 597 | |
30a33fbd CB |
598 | if (has_console) |
599 | ret = lxc_mainloop(&descr_console, 0); | |
600 | ||
3c319edb | 601 | out_mainloop_console: |
30a33fbd CB |
602 | if (has_console) { |
603 | lxc_mainloop_close(&descr_console); | |
604 | TRACE("Closed console mainloop"); | |
605 | } | |
3c319edb | 606 | |
42b09f94 CB |
607 | out_mainloop: |
608 | lxc_mainloop_close(&descr); | |
609 | TRACE("Closed mainloop"); | |
610 | ||
b0a33c1e | 611 | out_sigfd: |
1cc8bd4b | 612 | TRACE("Closed signal file descriptor %d", handler->sigfd); |
a783a414 | 613 | close_prot_errno_disarm(handler->sigfd); |
408da065 | 614 | |
1cc8bd4b | 615 | return ret; |
b0a33c1e | 616 | } |
617 | ||
a42abcce | 618 | void lxc_put_handler(struct lxc_handler *handler) |
f2e07cb6 | 619 | { |
a783a414 CB |
620 | close_prot_errno_disarm(handler->pinfd); |
621 | close_prot_errno_disarm(handler->pidfd); | |
622 | close_prot_errno_disarm(handler->sigfd); | |
1cc8bd4b | 623 | lxc_put_nsfds(handler); |
80308d07 | 624 | if (handler->conf && handler->conf->reboot == REBOOT_NONE) |
a783a414 CB |
625 | close_prot_errno_disarm(handler->conf->maincmd_fd); |
626 | close_prot_errno_disarm(handler->monitor_status_fd); | |
627 | close_prot_errno_disarm(handler->state_socket_pair[0]); | |
628 | close_prot_errno_disarm(handler->state_socket_pair[1]); | |
629 | cgroup_exit(handler->cgroup_ops); | |
a42abcce CB |
630 | if (handler->conf && handler->conf->reboot == REBOOT_NONE) |
631 | free_disarm(handler); | |
632 | else | |
633 | handler->conf = NULL; | |
f2e07cb6 CB |
634 | } |
635 | ||
a42abcce CB |
636 | struct lxc_handler *lxc_init_handler(struct lxc_handler *old, |
637 | const char *name, struct lxc_conf *conf, | |
5e5576a4 | 638 | const char *lxcpath, bool daemonize) |
59eb99ba | 639 | { |
85c279bb | 640 | int nr_keep_fds = 0; |
818a57fc | 641 | int ret; |
3a0f472d DL |
642 | struct lxc_handler *handler; |
643 | ||
a42abcce CB |
644 | if (!old) |
645 | handler = zalloc(sizeof(*handler)); | |
646 | else | |
647 | handler = old; | |
fdecbc9c | 648 | if (!handler) |
3a0f472d | 649 | return NULL; |
9044b79e | 650 | |
fdecbc9c CB |
651 | /* Note that am_guest_unpriv() checks the effective uid. We |
652 | * probably don't care if we are real root only if we are running | |
653 | * as root so this should be fine. | |
790255cf | 654 | */ |
e0010464 | 655 | handler->am_root = !am_guest_unpriv(); |
fae349da | 656 | handler->conf = conf; |
9123e471 | 657 | handler->lxcpath = lxcpath; |
818a57fc CB |
658 | handler->init_died = false; |
659 | handler->data_sock[0] = -EBADF; | |
660 | handler->data_sock[1] = -EBADF; | |
4d8bdfa0 | 661 | handler->monitor_status_fd = -EBADF; |
818a57fc | 662 | handler->pinfd = -EBADF; |
33942046 | 663 | handler->pidfd = -EBADF; |
1cc8bd4b | 664 | handler->sigfd = -EBADF; |
818a57fc CB |
665 | handler->state_socket_pair[0] = -EBADF; |
666 | handler->state_socket_pair[1] = -EBADF; | |
80308d07 | 667 | if (handler->conf->reboot == REBOOT_NONE) |
d39b10eb | 668 | lxc_list_init(&handler->conf->state_clients); |
fae349da | 669 | |
818a57fc CB |
670 | for (int i = 0; i < LXC_NS_MAX; i++) |
671 | handler->nsfd[i] = -EBADF; | |
b6b2b194 | 672 | |
f0ecc19d | 673 | handler->name = name; |
c581d2a6 CB |
674 | if (daemonize) |
675 | handler->transient_pid = lxc_raw_getpid(); | |
676 | else | |
677 | handler->transient_pid = -1; | |
3bdf52d7 | 678 | |
80308d07 | 679 | if (daemonize && handler->conf->reboot == REBOOT_NONE) { |
5e5576a4 | 680 | /* Create socketpair() to synchronize on daemonized startup. |
fdecbc9c CB |
681 | * When the container reboots we don't need to synchronize |
682 | * again currently so don't open another socketpair(). | |
5e5576a4 CB |
683 | */ |
684 | ret = socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0, | |
685 | handler->state_socket_pair); | |
686 | if (ret < 0) { | |
687 | ERROR("Failed to create anonymous pair of unix sockets"); | |
688 | goto on_error; | |
689 | } | |
9044b79e | 690 | |
5e5576a4 CB |
691 | TRACE("Created anonymous pair {%d,%d} of unix sockets", |
692 | handler->state_socket_pair[0], | |
693 | handler->state_socket_pair[1]); | |
85c279bb CB |
694 | handler->keep_fds[nr_keep_fds++] = handler->state_socket_pair[0]; |
695 | handler->keep_fds[nr_keep_fds++] = handler->state_socket_pair[1]; | |
5e5576a4 CB |
696 | } |
697 | ||
80308d07 | 698 | if (handler->conf->reboot == REBOOT_NONE) { |
d39b10eb CB |
699 | handler->conf->maincmd_fd = lxc_cmd_init(name, lxcpath, "command"); |
700 | if (handler->conf->maincmd_fd < 0) { | |
701 | ERROR("Failed to set up command socket"); | |
702 | goto on_error; | |
703 | } | |
85c279bb | 704 | handler->keep_fds[nr_keep_fds++] = handler->conf->maincmd_fd; |
aa460476 | 705 | } |
9044b79e | 706 | |
9dfa4041 | 707 | TRACE("Unix domain socket %d for command server is ready", |
aa460476 CB |
708 | handler->conf->maincmd_fd); |
709 | ||
710 | return handler; | |
711 | ||
f2e07cb6 | 712 | on_error: |
a42abcce | 713 | lxc_put_handler(handler); |
aa460476 CB |
714 | |
715 | return NULL; | |
716 | } | |
717 | ||
718 | int lxc_init(const char *name, struct lxc_handler *handler) | |
719 | { | |
f62cf1d4 | 720 | __do_close int status_fd = -EBADF; |
a2c09be0 | 721 | int ret; |
4a03ded4 | 722 | const char *loglevel; |
aa460476 CB |
723 | struct lxc_conf *conf = handler->conf; |
724 | ||
434c8e15 | 725 | handler->monitor_pid = lxc_raw_getpid(); |
4d8bdfa0 | 726 | status_fd = open("/proc/self/status", O_RDONLY | O_CLOEXEC); |
59b1b67e CB |
727 | if (status_fd < 0) |
728 | return log_error_errno(-1, errno, "Failed to open monitor status fd"); | |
434c8e15 | 729 | |
aa460476 | 730 | lsm_init(); |
2170b263 | 731 | TRACE("Initialized LSM"); |
d2e30e99 | 732 | |
408da065 | 733 | /* Begin by setting the state to STARTING. */ |
2170b263 | 734 | ret = lxc_set_state(name, handler, STARTING); |
59b1b67e CB |
735 | if (ret < 0) |
736 | return log_error(-1, "Failed to set state to \"%s\"", lxc_state2str(STARTING)); | |
46800e77 | 737 | TRACE("Set container state to \"STARTING\""); |
0ad19a3f | 738 | |
408da065 | 739 | /* Start of environment variable setup for hooks. */ |
24c2a48d CB |
740 | ret = setenv("LXC_NAME", name, 1); |
741 | if (ret < 0) | |
742 | SYSERROR("Failed to set environment variable: LXC_NAME=%s", name); | |
408da065 | 743 | |
2170b263 CB |
744 | if (conf->rcfile) { |
745 | ret = setenv("LXC_CONFIG_FILE", conf->rcfile, 1); | |
746 | if (ret < 0) | |
818a57fc | 747 | SYSERROR("Failed to set environment variable: LXC_CONFIG_FILE=%s", conf->rcfile); |
2170b263 | 748 | } |
408da065 | 749 | |
2170b263 CB |
750 | if (conf->rootfs.mount) { |
751 | ret = setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1); | |
752 | if (ret < 0) | |
818a57fc | 753 | SYSERROR("Failed to set environment variable: LXC_ROOTFS_MOUNT=%s", conf->rootfs.mount); |
2170b263 | 754 | } |
408da065 | 755 | |
2170b263 CB |
756 | if (conf->rootfs.path) { |
757 | ret = setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1); | |
758 | if (ret < 0) | |
818a57fc | 759 | SYSERROR("Failed to set environment variable: LXC_ROOTFS_PATH=%s", conf->rootfs.path); |
2170b263 | 760 | } |
408da065 | 761 | |
2170b263 CB |
762 | if (conf->console.path) { |
763 | ret = setenv("LXC_CONSOLE", conf->console.path, 1); | |
764 | if (ret < 0) | |
818a57fc | 765 | SYSERROR("Failed to set environment variable: LXC_CONSOLE=%s", conf->console.path); |
2170b263 | 766 | } |
408da065 | 767 | |
2170b263 CB |
768 | if (conf->console.log_path) { |
769 | ret = setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1); | |
770 | if (ret < 0) | |
818a57fc | 771 | SYSERROR("Failed to set environment variable: LXC_CONSOLE_LOGPATH=%s", conf->console.log_path); |
2170b263 | 772 | } |
408da065 | 773 | |
2170b263 CB |
774 | if (cgns_supported()) { |
775 | ret = setenv("LXC_CGNS_AWARE", "1", 1); | |
776 | if (ret < 0) | |
818a57fc | 777 | SYSERROR("Failed to set environment variable LXC_CGNS_AWARE=1"); |
2170b263 | 778 | } |
b8f88d9b | 779 | |
4a03ded4 | 780 | loglevel = lxc_log_priority_to_string(lxc_log_get_level()); |
2170b263 CB |
781 | ret = setenv("LXC_LOG_LEVEL", loglevel, 1); |
782 | if (ret < 0) | |
818a57fc | 783 | SYSERROR("Set environment variable LXC_LOG_LEVEL=%s", loglevel); |
a2c09be0 CB |
784 | |
785 | if (conf->hooks_version == 0) | |
786 | ret = setenv("LXC_HOOK_VERSION", "0", 1); | |
787 | else | |
788 | ret = setenv("LXC_HOOK_VERSION", "1", 1); | |
789 | if (ret < 0) | |
790 | SYSERROR("Failed to set environment variable LXC_HOOK_VERSION=%u", conf->hooks_version); | |
408da065 | 791 | /* End of environment variable setup for hooks. */ |
f7bee6c6 | 792 | |
2170b263 | 793 | TRACE("Set environment variables"); |
dbc9832d | 794 | |
2170b263 | 795 | ret = run_lxc_hooks(name, "pre-start", conf, NULL); |
65a2dc8b CB |
796 | if (ret < 0) |
797 | return log_error(-1, "Failed to run lxc.hook.pre-start for container \"%s\"", name); | |
2170b263 | 798 | TRACE("Ran pre-start hooks"); |
26ddeedd | 799 | |
408da065 CB |
800 | /* The signal fd has to be created before forking otherwise if the child |
801 | * process exits before we setup the signal fd, the event will be lost | |
802 | * and the command will be stuck. | |
803 | */ | |
83ee7875 | 804 | handler->sigfd = setup_signal_fd(&handler->oldmask); |
65a2dc8b CB |
805 | if (handler->sigfd < 0) |
806 | return log_error(-1, "Failed to setup SIGCHLD fd handler."); | |
2170b263 | 807 | TRACE("Set up signal fd"); |
b5159817 | 808 | |
408da065 | 809 | /* Do this after setting up signals since it might unblock SIGWINCH. */ |
564e31c4 | 810 | ret = lxc_terminal_setup(conf); |
4d1ffb0a CB |
811 | if (ret < 0) { |
812 | ERROR("Failed to create console"); | |
b5159817 | 813 | goto out_restore_sigmask; |
b0a33c1e | 814 | } |
4d1ffb0a | 815 | TRACE("Created console"); |
b0a33c1e | 816 | |
2f835b4b | 817 | ret = lxc_terminal_map_ids(conf, &conf->console); |
7cfeddd7 | 818 | if (ret < 0) { |
2170b263 | 819 | ERROR("Failed to chown console"); |
5b74eb3c | 820 | goto out_delete_terminal; |
c4d10a05 | 821 | } |
2170b263 | 822 | TRACE("Chowned console"); |
c4d10a05 | 823 | |
5a087e05 | 824 | handler->cgroup_ops = cgroup_init(handler->conf); |
2202afc9 CB |
825 | if (!handler->cgroup_ops) { |
826 | ERROR("Failed to initialize cgroup driver"); | |
5b74eb3c | 827 | goto out_delete_terminal; |
2202afc9 CB |
828 | } |
829 | TRACE("Initialized cgroup driver"); | |
830 | ||
2e1361a6 CB |
831 | ret = lxc_read_seccomp_config(conf); |
832 | if (ret < 0) | |
833 | return log_error(-1, "Failed loading seccomp policy"); | |
834 | TRACE("Read seccomp policy"); | |
835 | ||
1800f924 WB |
836 | ret = lsm_process_prepare(conf, handler->lxcpath); |
837 | if (ret < 0) { | |
838 | ERROR("Failed to initialize LSM"); | |
c581d2a6 | 839 | goto out_delete_terminal; |
1800f924 WB |
840 | } |
841 | TRACE("Initialized LSM"); | |
842 | ||
2170b263 | 843 | INFO("Container \"%s\" is initialized", name); |
4d8bdfa0 | 844 | handler->monitor_status_fd = move_fd(status_fd); |
aa460476 | 845 | return 0; |
59eb99ba | 846 | |
5b74eb3c CB |
847 | out_delete_terminal: |
848 | lxc_terminal_delete(&handler->conf->console); | |
ea918412 | 849 | |
b5159817 | 850 | out_restore_sigmask: |
4c496daa | 851 | (void)pthread_sigmask(SIG_SETMASK, &handler->oldmask, NULL); |
ea918412 | 852 | |
aa460476 | 853 | return -1; |
59eb99ba DL |
854 | } |
855 | ||
fd5be714 | 856 | void lxc_end(struct lxc_handler *handler) |
59eb99ba | 857 | { |
818a57fc | 858 | int ret; |
18b3b9c1 | 859 | pid_t self; |
dbc9832d | 860 | struct lxc_list *cur, *next; |
dbc9832d | 861 | char *namespaces[LXC_NS_MAX + 1]; |
b3286b62 | 862 | size_t namespace_count = 0; |
0c5859ff | 863 | const char *name = handler->name; |
2202afc9 | 864 | struct cgroup_ops *cgroup_ops = handler->cgroup_ops; |
b6b2b194 | 865 | |
408da065 CB |
866 | /* The STOPPING state is there for future cleanup code which can take |
867 | * awhile. | |
59eb99ba | 868 | */ |
25c2aca5 | 869 | lxc_set_state(name, handler, STOPPING); |
b6b2b194 | 870 | |
0059379f | 871 | self = lxc_raw_getpid(); |
818a57fc | 872 | for (int i = 0; i < LXC_NS_MAX; i++) { |
18b3b9c1 CB |
873 | if (handler->nsfd[i] < 0) |
874 | continue; | |
875 | ||
876 | if (handler->conf->hooks_version == 0) | |
c097467f | 877 | ret = asprintf(&namespaces[namespace_count], |
18b3b9c1 CB |
878 | "%s:/proc/%d/fd/%d", ns_info[i].proc_name, |
879 | self, handler->nsfd[i]); | |
880 | else | |
c097467f | 881 | ret = asprintf(&namespaces[namespace_count], |
18b3b9c1 | 882 | "/proc/%d/fd/%d", self, handler->nsfd[i]); |
818a57fc | 883 | if (ret < 0) { |
c097467f | 884 | SYSERROR("Failed to allocate memory"); |
18b3b9c1 | 885 | break; |
b3286b62 | 886 | } |
18b3b9c1 CB |
887 | |
888 | if (handler->conf->hooks_version == 0) { | |
889 | namespace_count++; | |
890 | continue; | |
891 | } | |
892 | ||
c097467f CB |
893 | ret = setenv(ns_info[i].env_name, namespaces[namespace_count], 1); |
894 | if (ret < 0) | |
18b3b9c1 CB |
895 | SYSERROR("Failed to set environment variable %s=%s", |
896 | ns_info[i].env_name, namespaces[namespace_count]); | |
897 | else | |
898 | TRACE("Set environment variable %s=%s", | |
899 | ns_info[i].env_name, namespaces[namespace_count]); | |
900 | ||
901 | namespace_count++; | |
b3286b62 WB |
902 | } |
903 | namespaces[namespace_count] = NULL; | |
c154af98 | 904 | |
80308d07 | 905 | if (handler->conf->reboot > REBOOT_NONE) { |
c097467f CB |
906 | ret = setenv("LXC_TARGET", "reboot", 1); |
907 | if (ret < 0) | |
818a57fc | 908 | SYSERROR("Failed to set environment variable: LXC_TARGET=reboot"); |
c097467f | 909 | } |
408da065 | 910 | |
80308d07 | 911 | if (handler->conf->reboot == REBOOT_NONE) { |
c097467f CB |
912 | ret = setenv("LXC_TARGET", "stop", 1); |
913 | if (ret < 0) | |
818a57fc | 914 | SYSERROR("Failed to set environment variable: LXC_TARGET=stop"); |
c097467f | 915 | } |
c154af98 | 916 | |
18b3b9c1 | 917 | if (handler->conf->hooks_version == 0) |
c097467f | 918 | ret = run_lxc_hooks(name, "stop", handler->conf, namespaces); |
18b3b9c1 | 919 | else |
c097467f CB |
920 | ret = run_lxc_hooks(name, "stop", handler->conf, NULL); |
921 | if (ret < 0) | |
922 | ERROR("Failed to run \"lxc.hook.stop\" hook"); | |
c154af98 | 923 | |
b3286b62 WB |
924 | while (namespace_count--) |
925 | free(namespaces[namespace_count]); | |
28d9e29e | 926 | |
1800f924 WB |
927 | lsm_process_cleanup(handler->conf, handler->lxcpath); |
928 | ||
e2aed383 WB |
929 | if (cgroup_ops) { |
930 | cgroup_ops->payload_destroy(cgroup_ops, handler); | |
931 | cgroup_ops->monitor_destroy(cgroup_ops, handler); | |
932 | } | |
4288b79f | 933 | |
80308d07 | 934 | if (handler->conf->reboot == REBOOT_NONE) { |
f893d898 CB |
935 | /* For all new state clients simply close the command socket. |
936 | * This will inform all state clients that the container is | |
937 | * STOPPED and also prevents a race between a open()/close() on | |
938 | * the command socket causing a new process to get ECONNREFUSED | |
939 | * because we haven't yet closed the command socket. | |
940 | */ | |
b87ee312 | 941 | close_prot_errno_disarm(handler->conf->maincmd_fd); |
c3184275 CB |
942 | TRACE("Closed command socket"); |
943 | ||
944 | /* This function will try to connect to the legacy lxc-monitord | |
945 | * state server and only exists for backwards compatibility. | |
946 | */ | |
947 | lxc_monitor_send_state(name, STOPPED, handler->lxcpath); | |
948 | ||
949 | /* The command socket is closed so no one can acces the command | |
950 | * socket anymore so there's no need to lock it. | |
951 | */ | |
952 | handler->state = STOPPED; | |
953 | TRACE("Set container state to \"STOPPED\""); | |
f893d898 CB |
954 | } else { |
955 | lxc_set_state(name, handler, STOPPED); | |
695d0e56 | 956 | TRACE("Set container state to \"STOPPED\""); |
d39b10eb | 957 | } |
4288b79f | 958 | |
048493a3 CB |
959 | /* Avoid lingering namespace references. */ |
960 | lxc_put_nsfds(handler); | |
961 | ||
46800e77 CB |
962 | ret = run_lxc_hooks(name, "post-stop", handler->conf, NULL); |
963 | if (ret < 0) { | |
c097467f | 964 | ERROR("Failed to run lxc.hook.post-stop for container \"%s\"", name); |
80308d07 | 965 | if (handler->conf->reboot > REBOOT_NONE) { |
c097467f | 966 | WARN("Container will be stopped instead of rebooted"); |
80308d07 | 967 | handler->conf->reboot = REBOOT_NONE; |
c097467f CB |
968 | |
969 | ret = setenv("LXC_TARGET", "stop", 1); | |
970 | if (ret < 0) | |
818a57fc | 971 | WARN("Failed to set environment variable: LXC_TARGET=stop"); |
f3787121 CB |
972 | } |
973 | } | |
26ddeedd | 974 | |
408da065 | 975 | /* Reset mask set by setup_signal_fd. */ |
b467714b | 976 | ret = pthread_sigmask(SIG_SETMASK, &handler->oldmask, NULL); |
c097467f | 977 | if (ret < 0) |
a24c5678 | 978 | SYSWARN("Failed to restore signal mask"); |
8f64a3f6 | 979 | |
2aac071b | 980 | lxc_terminal_delete(&handler->conf->console); |
0e4be3cf | 981 | lxc_delete_tty(&handler->conf->ttys); |
dbc9832d | 982 | |
dbc9832d CB |
983 | /* The command socket is now closed, no more state clients can register |
984 | * themselves from now on. So free the list of state clients. | |
985 | */ | |
d39b10eb CB |
986 | lxc_list_for_each_safe(cur, &handler->conf->state_clients, next) { |
987 | struct lxc_state_client *client = cur->elem; | |
988 | ||
989 | /* Keep state clients that want to be notified about reboots. */ | |
80308d07 CB |
990 | if ((handler->conf->reboot > REBOOT_NONE) && |
991 | (client->states[RUNNING] == 2)) | |
d39b10eb CB |
992 | continue; |
993 | ||
dbc9832d | 994 | /* close state client socket */ |
dbc9832d | 995 | lxc_list_del(cur); |
c097467f | 996 | close(client->clientfd); |
dbc9832d CB |
997 | free(cur->elem); |
998 | free(cur); | |
999 | } | |
1000 | ||
80308d07 | 1001 | if (handler->conf->ephemeral == 1 && handler->conf->reboot != REBOOT_REQ) |
28272964 | 1002 | lxc_destroy_container_on_signal(handler, name); |
2c5f2ede | 1003 | |
a42abcce | 1004 | lxc_put_handler(handler); |
59eb99ba DL |
1005 | } |
1006 | ||
0c5859ff | 1007 | void lxc_abort(struct lxc_handler *handler) |
59eb99ba | 1008 | { |
33942046 CB |
1009 | int ret = 0; |
1010 | int status; | |
73e608b2 | 1011 | |
0c5859ff | 1012 | lxc_set_state(handler->name, handler, ABORTING); |
0e4f9d51 | 1013 | |
c718fac1 | 1014 | if (handler->pidfd >= 0) { |
33942046 | 1015 | ret = lxc_raw_pidfd_send_signal(handler->pidfd, SIGKILL, NULL, 0); |
c718fac1 | 1016 | if (ret) |
11c7d349 CB |
1017 | SYSWARN("Failed to send SIGKILL via pidfd %d for process %d", |
1018 | handler->pidfd, handler->pid); | |
c718fac1 CB |
1019 | } |
1020 | ||
65a2dc8b | 1021 | if ((!ret || errno != ESRCH) && handler->pid > 0) |
11c7d349 CB |
1022 | if (kill(handler->pid, SIGKILL)) |
1023 | SYSWARN("Failed to send SIGKILL to %d", handler->pid); | |
0e4f9d51 | 1024 | |
33942046 CB |
1025 | do { |
1026 | ret = waitpid(-1, &status, 0); | |
1027 | } while (ret > 0); | |
59eb99ba DL |
1028 | } |
1029 | ||
ffe1e01a | 1030 | static int do_start(void *data) |
50e98013 | 1031 | { |
cdb2a47f | 1032 | struct lxc_handler *handler = data; |
f62cf1d4 | 1033 | __lxc_unused __do_close int data_sock0 = handler->data_sock[0], |
9ff57a59 | 1034 | data_sock1 = handler->data_sock[1]; |
d3103162 | 1035 | __do_close int devnull_fd = -EBADF, status_fd = -EBADF; |
8bf3abfb | 1036 | int ret; |
928b1f04 YT |
1037 | uid_t new_uid; |
1038 | gid_t new_gid; | |
e96a536c | 1039 | struct lxc_list *iterator; |
964581c2 CB |
1040 | uid_t nsuid = 0; |
1041 | gid_t nsgid = 0; | |
50e98013 | 1042 | |
18225d19 | 1043 | lxc_sync_fini_parent(handler); |
50e98013 | 1044 | |
9ff57a59 | 1045 | if (lxc_abstract_unix_recv_fds(data_sock1, &status_fd, 1, NULL, 0) < 0) { |
4d8bdfa0 CB |
1046 | ERROR("Failed to receive status file descriptor to child process"); |
1047 | goto out_warn_father; | |
1048 | } | |
1049 | ||
408da065 CB |
1050 | /* This prctl must be before the synchro, so if the parent dies before |
1051 | * we set the parent death signal, we will detect its death with the | |
1052 | * synchro right after, otherwise we have a window where the parent can | |
1053 | * exit before we set the pdeath signal leading to a unsupervized | |
1054 | * container. | |
743ecd2e | 1055 | */ |
4d8bdfa0 | 1056 | ret = lxc_set_death_signal(SIGKILL, handler->monitor_pid, status_fd); |
912314fc CB |
1057 | if (ret < 0) { |
1058 | SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL"); | |
18225d19 | 1059 | goto out_warn_father; |
743ecd2e DL |
1060 | } |
1061 | ||
611ddd34 CB |
1062 | ret = lxc_ambient_caps_up(); |
1063 | if (ret < 0) { | |
6f5e532f | 1064 | ERROR("Failed to raise ambient capabilities"); |
611ddd34 CB |
1065 | goto out_warn_father; |
1066 | } | |
1067 | ||
b467714b | 1068 | ret = pthread_sigmask(SIG_SETMASK, &handler->oldmask, NULL); |
18225d19 CB |
1069 | if (ret < 0) { |
1070 | SYSERROR("Failed to set signal mask"); | |
1071 | goto out_warn_father; | |
1072 | } | |
50e98013 | 1073 | |
408da065 | 1074 | /* Don't leak the pinfd to the container. */ |
a783a414 | 1075 | close_prot_errno_disarm(handler->pinfd); |
2b0e17e4 | 1076 | |
18225d19 CB |
1077 | ret = lxc_sync_wait_parent(handler, LXC_SYNC_STARTUP); |
1078 | if (ret < 0) | |
1079 | goto out_warn_father; | |
5b1e83cb | 1080 | |
408da065 CB |
1081 | /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See |
1082 | * https://github.com/lxc/lxd/issues/1978. | |
1083 | */ | |
becad0ec | 1084 | if ((handler->ns_clone_flags & (CLONE_NEWNET | CLONE_NEWUSER)) == |
408da065 | 1085 | (CLONE_NEWNET | CLONE_NEWUSER)) { |
5b1e83cb SH |
1086 | ret = unshare(CLONE_NEWNET); |
1087 | if (ret < 0) { | |
e96a536c | 1088 | SYSERROR("Failed to unshare CLONE_NEWNET"); |
5b1e83cb SH |
1089 | goto out_warn_father; |
1090 | } | |
e96a536c | 1091 | INFO("Unshared CLONE_NEWNET"); |
5b1e83cb SH |
1092 | } |
1093 | ||
408da065 CB |
1094 | /* Tell the parent task it can begin to configure the container and wait |
1095 | * for it to finish. | |
3c22086f | 1096 | */ |
e96a536c CB |
1097 | ret = lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE); |
1098 | if (ret < 0) | |
611ddd34 | 1099 | goto out_error; |
50e98013 | 1100 | |
e389f2af | 1101 | if (handler->ns_clone_flags & CLONE_NEWNET) { |
3c09b97c | 1102 | ret = lxc_network_recv_from_parent(handler); |
e389f2af CB |
1103 | if (ret < 0) { |
1104 | ERROR("Failed to receive veth names from parent"); | |
1105 | goto out_warn_father; | |
1106 | } | |
7ab1ba02 | 1107 | } |
658979c5 | 1108 | |
408da065 | 1109 | /* If we are in a new user namespace, become root there to have |
d08f8d2f | 1110 | * privilege over our namespace. |
f6d3e3e4 SH |
1111 | */ |
1112 | if (!lxc_list_empty(&handler->conf->id_map)) { | |
964581c2 CB |
1113 | if (!handler->conf->root_nsuid_map) |
1114 | nsuid = handler->conf->init_uid; | |
1115 | ||
1116 | if (!handler->conf->root_nsgid_map) | |
1117 | nsgid = handler->conf->init_gid; | |
4160c3a0 | 1118 | |
d08f8d2f CB |
1119 | /* Drop groups only after we switched to a valid gid in the new |
1120 | * user namespace. | |
1121 | */ | |
8af07f82 CB |
1122 | if (!lxc_setgroups(0, NULL) && |
1123 | (handler->am_root || errno != EPERM)) | |
c476bdce | 1124 | goto out_warn_father; |
4b826b1f | 1125 | |
b58214ac CB |
1126 | if (!lxc_switch_uid_gid(nsuid, nsgid)) |
1127 | goto out_warn_father; | |
1128 | ||
b81689a1 CB |
1129 | ret = prctl(PR_SET_DUMPABLE, prctl_arg(1), prctl_arg(0), |
1130 | prctl_arg(0), prctl_arg(0)); | |
d7883725 CB |
1131 | if (ret < 0) |
1132 | goto out_warn_father; | |
912314fc CB |
1133 | |
1134 | /* set{g,u}id() clears deathsignal */ | |
4d8bdfa0 | 1135 | ret = lxc_set_death_signal(SIGKILL, handler->monitor_pid, status_fd); |
912314fc CB |
1136 | if (ret < 0) { |
1137 | SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL"); | |
1138 | goto out_warn_father; | |
1139 | } | |
f6d3e3e4 SH |
1140 | } |
1141 | ||
e96a536c CB |
1142 | ret = access(handler->lxcpath, X_OK); |
1143 | if (ret != 0) { | |
c8154066 SH |
1144 | print_top_failing_dir(handler->lxcpath); |
1145 | goto out_warn_father; | |
1146 | } | |
1147 | ||
7a55c157 TA |
1148 | /* In order to checkpoint restore, we need to have everything in the |
1149 | * same mount namespace. However, some containers may not have a | |
1150 | * reasonable /dev (in particular, they may not have /dev/null), so we | |
1151 | * can't set init's std fds to /dev/null by opening it from inside the | |
1152 | * container. | |
1153 | * | |
1154 | * If that's the case, fall back to using the host's /dev/null. This | |
1155 | * means that migration won't work, but at least we won't spew output | |
1156 | * where it isn't wanted. | |
1157 | */ | |
bb955810 | 1158 | if (handler->daemonize && !handler->conf->autodev) { |
2806a87d | 1159 | char path[PATH_MAX]; |
3c09b97c | 1160 | |
2806a87d RK |
1161 | ret = snprintf(path, sizeof(path), "%s/dev/null", |
1162 | handler->conf->rootfs.mount); | |
1163 | if (ret < 0 || ret >= sizeof(path)) | |
1164 | goto out_warn_father; | |
3c09b97c | 1165 | |
e96a536c CB |
1166 | ret = access(path, F_OK); |
1167 | if (ret != 0) { | |
1168 | devnull_fd = open_devnull(); | |
c44de748 | 1169 | |
e96a536c CB |
1170 | if (devnull_fd < 0) |
1171 | goto out_warn_father; | |
818a57fc | 1172 | WARN("Using /dev/null from the host for container init's standard file descriptors. Migration will not work"); |
e96a536c | 1173 | } |
c44de748 AM |
1174 | } |
1175 | ||
408da065 | 1176 | /* Ask father to setup cgroups and wait for him to finish. */ |
e96a536c CB |
1177 | ret = lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP); |
1178 | if (ret < 0) | |
c44de748 | 1179 | goto out_error; |
544a48a0 | 1180 | |
deefdf8a CB |
1181 | /* Unshare cgroup namespace after we have setup our cgroups. If we do it |
1182 | * earlier we end up with a wrong view of /proc/self/cgroup. For | |
1183 | * example, assume we unshare(CLONE_NEWCGROUP) first, and then create | |
1184 | * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then | |
1185 | * /proc/self/cgroup would show us: | |
1186 | * | |
1187 | * 8:cpuset:/lxc/c | |
1188 | * | |
1189 | * whereas it should actually show | |
1190 | * | |
1191 | * 8:cpuset:/ | |
1192 | */ | |
becad0ec | 1193 | if (handler->ns_clone_flags & CLONE_NEWCGROUP) { |
e96a536c CB |
1194 | ret = unshare(CLONE_NEWCGROUP); |
1195 | if (ret < 0) { | |
bca7c59c CB |
1196 | if (errno != EINVAL) { |
1197 | SYSERROR("Failed to unshare CLONE_NEWCGROUP"); | |
1198 | goto out_warn_father; | |
1199 | } | |
1200 | ||
1201 | handler->ns_clone_flags &= ~CLONE_NEWCGROUP; | |
1202 | SYSINFO("Kernel does not support CLONE_NEWCGROUP"); | |
1203 | } else { | |
1204 | INFO("Unshared CLONE_NEWCGROUP"); | |
deefdf8a | 1205 | } |
deefdf8a CB |
1206 | } |
1207 | ||
70fd7fc9 CB |
1208 | if (handler->ns_clone_flags & CLONE_NEWTIME) { |
1209 | ret = unshare(CLONE_NEWTIME); | |
1210 | if (ret < 0) { | |
1211 | if (errno != EINVAL) { | |
1212 | SYSERROR("Failed to unshare CLONE_NEWTIME"); | |
1213 | goto out_warn_father; | |
1214 | } | |
1215 | ||
1216 | handler->ns_clone_flags &= ~CLONE_NEWTIME; | |
1217 | SYSINFO("Kernel does not support CLONE_NEWTIME"); | |
1218 | } else { | |
1219 | __do_close int timens_fd = -EBADF; | |
1220 | ||
1221 | INFO("Unshared CLONE_NEWTIME"); | |
1222 | ||
1223 | if (handler->conf->timens.s_boot) | |
1224 | ret = timens_offset_write(CLOCK_BOOTTIME, handler->conf->timens.s_boot, 0); | |
1225 | else if (handler->conf->timens.ns_boot) | |
1226 | ret = timens_offset_write(CLOCK_BOOTTIME, 0, handler->conf->timens.ns_boot); | |
1227 | if (ret) { | |
1228 | SYSERROR("Failed to write CLONE_BOOTTIME offset"); | |
1229 | goto out_warn_father; | |
1230 | } | |
1231 | TRACE("Wrote CLOCK_BOOTTIME offset"); | |
1232 | ||
1233 | if (handler->conf->timens.s_monotonic) | |
1234 | ret = timens_offset_write(CLOCK_MONOTONIC, handler->conf->timens.s_monotonic, 0); | |
1235 | else if (handler->conf->timens.ns_monotonic) | |
1236 | ret = timens_offset_write(CLOCK_MONOTONIC, 0, handler->conf->timens.ns_monotonic); | |
1237 | if (ret) { | |
1238 | SYSERROR("Failed to write CLONE_MONOTONIC offset"); | |
1239 | goto out_warn_father; | |
1240 | } | |
1241 | TRACE("Wrote CLOCK_MONOTONIC offset"); | |
1242 | ||
1243 | timens_fd = open("/proc/self/ns/time_for_children", O_RDONLY | O_CLOEXEC); | |
1244 | if (timens_fd < 0) { | |
1245 | SYSERROR("Failed to open \"/proc/self/ns/time_for_children\""); | |
1246 | goto out_warn_father; | |
1247 | } | |
1248 | ||
1249 | ret = setns(timens_fd, CLONE_NEWTIME); | |
1250 | if (ret) { | |
1251 | SYSERROR("Failed to setns(%d(\"/proc/self/ns/time_for_children\"))", timens_fd); | |
1252 | goto out_warn_father; | |
1253 | } | |
1254 | } | |
1255 | } | |
1256 | ||
149857af CB |
1257 | /* Add the requested environment variables to the current environment to |
1258 | * allow them to be used by the various hooks, such as the start hook | |
317494f1 | 1259 | * below. |
98ff08ca | 1260 | */ |
98ff08ca | 1261 | lxc_list_for_each(iterator, &handler->conf->environment) { |
e96a536c CB |
1262 | ret = putenv((char *)iterator->elem); |
1263 | if (ret < 0) { | |
1264 | SYSERROR("Failed to set environment variable: %s", | |
1265 | (char *)iterator->elem); | |
98ff08ca CB |
1266 | goto out_warn_father; |
1267 | } | |
1268 | } | |
1269 | ||
f4152036 | 1270 | /* Setup the container, ip, names, utsname, ... */ |
3b988b33 | 1271 | ret = lxc_setup(handler); |
7729f8e5 | 1272 | if (ret < 0) { |
e96a536c | 1273 | ERROR("Failed to setup container \"%s\"", handler->name); |
f4152036 CB |
1274 | goto out_warn_father; |
1275 | } | |
1276 | ||
408da065 | 1277 | /* Set the label to change to when we exec(2) the container's init. */ |
1800f924 | 1278 | ret = lsm_process_label_set(NULL, handler->conf, true); |
e96a536c | 1279 | if (ret < 0) |
e075f5d9 | 1280 | goto out_warn_father; |
5112cd70 | 1281 | |
029cdff5 | 1282 | /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it |
408da065 CB |
1283 | * before we aren't allowed anymore. |
1284 | */ | |
029cdff5 | 1285 | if (handler->conf->no_new_privs) { |
b81689a1 CB |
1286 | ret = prctl(PR_SET_NO_NEW_PRIVS, prctl_arg(1), prctl_arg(0), |
1287 | prctl_arg(0), prctl_arg(0)); | |
e96a536c | 1288 | if (ret < 0) { |
818a57fc | 1289 | SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges"); |
029cdff5 CB |
1290 | goto out_warn_father; |
1291 | } | |
818a57fc | 1292 | DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges"); |
029cdff5 CB |
1293 | } |
1294 | ||
0d9acb99 DE |
1295 | /* Some init's such as busybox will set sane tty settings on stdin, |
1296 | * stdout, stderr which it thinks is the console. We already set them | |
1297 | * the way we wanted on the real terminal, and we want init to do its | |
564e31c4 | 1298 | * setup on its console ie. the pty allocated in lxc_terminal_setup() so |
408da065 | 1299 | * make sure that that pty is stdin,stdout,stderr. |
0d9acb99 | 1300 | */ |
cecf3e83 | 1301 | if (handler->conf->console.pts >= 0) { |
bb955810 | 1302 | if (handler->daemonize || !handler->conf->is_execute) |
cecf3e83 | 1303 | ret = set_stdfds(handler->conf->console.pts); |
5d113f65 | 1304 | else |
cecf3e83 | 1305 | ret = lxc_terminal_set_stdfds(handler->conf->console.pts); |
5d113f65 | 1306 | if (ret < 0) { |
818a57fc | 1307 | ERROR("Failed to redirect std{in,out,err} to pty file descriptor %d", |
cecf3e83 | 1308 | handler->conf->console.pts); |
c5b93afb LF |
1309 | goto out_warn_father; |
1310 | } | |
5d113f65 | 1311 | } |
0d9acb99 | 1312 | |
408da065 | 1313 | /* If we mounted a temporary proc, then unmount it now. */ |
5112cd70 | 1314 | tmp_proc_unmount(handler->conf); |
e075f5d9 | 1315 | |
e96a536c CB |
1316 | ret = lxc_seccomp_load(handler->conf); |
1317 | if (ret < 0) | |
8f2c3a70 SH |
1318 | goto out_warn_father; |
1319 | ||
c3e3c21a CB |
1320 | ret = lxc_seccomp_send_notifier_fd(&handler->conf->seccomp, data_sock0); |
1321 | if (ret < 0) { | |
1322 | SYSERROR("Failed to send seccomp notify fd to parent"); | |
1323 | goto out_warn_father; | |
cdb2a47f | 1324 | } |
cdb2a47f | 1325 | |
e96a536c CB |
1326 | ret = run_lxc_hooks(handler->name, "start", handler->conf, NULL); |
1327 | if (ret < 0) { | |
1328 | ERROR("Failed to run lxc.hook.start for container \"%s\"", | |
1329 | handler->name); | |
773fb9ca SH |
1330 | goto out_warn_father; |
1331 | } | |
fc25b815 | 1332 | |
a783a414 | 1333 | close_prot_errno_disarm(handler->sigfd); |
26ddeedd | 1334 | |
cecf3e83 | 1335 | if (handler->conf->console.pts < 0 && handler->daemonize) { |
f4c177c3 CB |
1336 | if (devnull_fd < 0) { |
1337 | devnull_fd = open_devnull(); | |
1338 | if (devnull_fd < 0) | |
1339 | goto out_warn_father; | |
1340 | } | |
1341 | ||
e96a536c CB |
1342 | ret = set_stdfds(devnull_fd); |
1343 | if (ret < 0) { | |
1344 | ERROR("Failed to redirect std{in,out,err} to \"/dev/null\""); | |
c5b93afb LF |
1345 | goto out_warn_father; |
1346 | } | |
e96a536c | 1347 | } |
507cee36 | 1348 | |
a783a414 | 1349 | close_prot_errno_disarm(devnull_fd); |
c44de748 | 1350 | |
8c9a7665 TA |
1351 | setsid(); |
1352 | ||
e96a536c CB |
1353 | if (handler->conf->init_cwd) { |
1354 | ret = chdir(handler->conf->init_cwd); | |
1355 | if (ret < 0) { | |
1356 | SYSERROR("Could not change directory to \"%s\"", | |
1357 | handler->conf->init_cwd); | |
1358 | goto out_warn_father; | |
1359 | } | |
3c491553 L |
1360 | } |
1361 | ||
e96a536c CB |
1362 | ret = lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP_LIMITS); |
1363 | if (ret < 0) | |
f4152036 CB |
1364 | goto out_warn_father; |
1365 | ||
149857af CB |
1366 | /* Reset the environment variables the user requested in a clear |
1367 | * environment. | |
1368 | */ | |
e96a536c CB |
1369 | ret = clearenv(); |
1370 | /* Don't error out though. */ | |
1371 | if (ret < 0) | |
149857af | 1372 | SYSERROR("Failed to clear environment."); |
149857af CB |
1373 | |
1374 | lxc_list_for_each(iterator, &handler->conf->environment) { | |
e96a536c CB |
1375 | ret = putenv((char *)iterator->elem); |
1376 | if (ret < 0) { | |
1377 | SYSERROR("Failed to set environment variable: %s", | |
1378 | (char *)iterator->elem); | |
149857af CB |
1379 | goto out_warn_father; |
1380 | } | |
1381 | } | |
1382 | ||
e96a536c CB |
1383 | ret = putenv("container=lxc"); |
1384 | if (ret < 0) { | |
1385 | SYSERROR("Failed to set environment variable: container=lxc"); | |
98ff08ca CB |
1386 | goto out_warn_father; |
1387 | } | |
1388 | ||
885766f5 CB |
1389 | if (handler->conf->ttys.tty_names) { |
1390 | ret = putenv(handler->conf->ttys.tty_names); | |
e96a536c CB |
1391 | if (ret < 0) { |
1392 | SYSERROR("Failed to set environment variable for container ptys"); | |
98ff08ca CB |
1393 | goto out_warn_father; |
1394 | } | |
1395 | } | |
1396 | ||
76bdf299 CB |
1397 | /* The container has been setup. We can now switch to an unprivileged |
1398 | * uid/gid. | |
1399 | */ | |
928b1f04 YT |
1400 | new_uid = handler->conf->init_uid; |
1401 | new_gid = handler->conf->init_gid; | |
76bdf299 | 1402 | |
964581c2 CB |
1403 | /* Avoid unnecessary syscalls. */ |
1404 | if (new_uid == nsuid) | |
1405 | new_uid = LXC_INVALID_UID; | |
1406 | ||
1407 | if (new_gid == nsgid) | |
1408 | new_gid = LXC_INVALID_GID; | |
c353b0b9 | 1409 | |
6aff5157 | 1410 | /* Make sure that the processes STDIO is correctly owned by the user that we are switching to */ |
c353b0b9 CB |
1411 | ret = fix_stdio_permissions(new_uid); |
1412 | if (ret) | |
1413 | WARN("Failed to ajust stdio permissions"); | |
964581c2 | 1414 | |
8af07f82 CB |
1415 | /* If we are in a new user namespace we already dropped all groups when |
1416 | * we switched to root in the new user namespace further above. Only | |
1417 | * drop groups if we can, so ensure that we have necessary privilege. | |
1418 | */ | |
1419 | if (lxc_list_empty(&handler->conf->id_map)) | |
1420 | #if HAVE_LIBCAP | |
1421 | if (lxc_proc_cap_is_set(CAP_SETGID, CAP_EFFECTIVE)) | |
1422 | #endif | |
1423 | if (!lxc_setgroups(0, NULL)) | |
1424 | goto out_warn_father; | |
1425 | ||
b58214ac CB |
1426 | if (!lxc_switch_uid_gid(new_uid, new_gid)) |
1427 | goto out_warn_father; | |
1428 | ||
611ddd34 CB |
1429 | ret = lxc_ambient_caps_down(); |
1430 | if (ret < 0) { | |
6f5e532f | 1431 | ERROR("Failed to clear ambient capabilities"); |
611ddd34 CB |
1432 | goto out_warn_father; |
1433 | } | |
1434 | ||
258f8051 | 1435 | if (handler->conf->monitor_signal_pdeath != SIGKILL) { |
4d8bdfa0 CB |
1436 | ret = lxc_set_death_signal(handler->conf->monitor_signal_pdeath, |
1437 | handler->monitor_pid, status_fd); | |
258f8051 CB |
1438 | if (ret < 0) { |
1439 | SYSERROR("Failed to set PR_SET_PDEATHSIG to %d", | |
1440 | handler->conf->monitor_signal_pdeath); | |
1441 | goto out_warn_father; | |
1442 | } | |
1443 | } | |
1444 | ||
d3103162 CB |
1445 | /* |
1446 | * After this call, we are in error because this ops should not return | |
408da065 CB |
1447 | * as it execs. |
1448 | */ | |
c4ea60df | 1449 | handler->ops->start(handler, handler->data); |
50e98013 DL |
1450 | |
1451 | out_warn_father: | |
d3103162 CB |
1452 | /* |
1453 | * We want the parent to know something went wrong, so we return a | |
408da065 CB |
1454 | * special error code. |
1455 | */ | |
d1ccb562 | 1456 | lxc_sync_wake_parent(handler, LXC_SYNC_ERROR); |
c44de748 AM |
1457 | |
1458 | out_error: | |
50e98013 DL |
1459 | return -1; |
1460 | } | |
1461 | ||
ae467c54 | 1462 | static int lxc_recv_ttys_from_child(struct lxc_handler *handler) |
e8bd4e43 | 1463 | { |
ae467c54 | 1464 | int i; |
2520facd | 1465 | struct lxc_terminal_info *tty; |
ae467c54 | 1466 | int ret = -1; |
c6012571 | 1467 | int sock = handler->data_sock[1]; |
e8bd4e43 | 1468 | struct lxc_conf *conf = handler->conf; |
0e4be3cf | 1469 | struct lxc_tty_info *ttys = &conf->ttys; |
e8bd4e43 | 1470 | |
885766f5 | 1471 | if (!conf->ttys.max) |
e8bd4e43 SH |
1472 | return 0; |
1473 | ||
885766f5 | 1474 | ttys->tty = malloc(sizeof(*ttys->tty) * ttys->max); |
0e4be3cf | 1475 | if (!ttys->tty) |
e8bd4e43 | 1476 | return -1; |
e8bd4e43 | 1477 | |
885766f5 | 1478 | for (i = 0; i < conf->ttys.max; i++) { |
672c1e58 CB |
1479 | int ttyfds[2]; |
1480 | ||
1481 | ret = lxc_abstract_unix_recv_fds(sock, ttyfds, 2, NULL, 0); | |
1482 | if (ret < 0) | |
1483 | break; | |
ae467c54 | 1484 | |
0e4be3cf | 1485 | tty = &ttys->tty[i]; |
dd3de568 | 1486 | tty->busy = -1; |
cecf3e83 CB |
1487 | tty->ptmx = ttyfds[0]; |
1488 | tty->pts = ttyfds[1]; | |
1489 | TRACE("Received pty with ptmx fd %d and pts fd %d from child", tty->ptmx, tty->pts); | |
e8bd4e43 | 1490 | } |
6d1400b5 | 1491 | |
ae467c54 | 1492 | if (ret < 0) |
6d1400b5 | 1493 | SYSERROR("Failed to receive %zu ttys from child", ttys->max); |
ae467c54 | 1494 | else |
885766f5 | 1495 | TRACE("Received %zu ttys from child", ttys->max); |
ae467c54 CB |
1496 | |
1497 | return ret; | |
e8bd4e43 SH |
1498 | } |
1499 | ||
5af9369b | 1500 | int resolve_clone_flags(struct lxc_handler *handler) |
f813849c | 1501 | { |
8bc8c715 CB |
1502 | int i; |
1503 | struct lxc_conf *conf = handler->conf; | |
70fd7fc9 CB |
1504 | bool wants_timens = conf->timens.s_boot || conf->timens.ns_boot || |
1505 | conf->timens.s_monotonic || conf->timens.ns_monotonic; | |
f813849c | 1506 | |
8bc8c715 | 1507 | for (i = 0; i < LXC_NS_MAX; i++) { |
f7a0c6ee CB |
1508 | if (conf->ns_keep) { |
1509 | if (!(conf->ns_keep & ns_info[i].clone_flag)) | |
becad0ec | 1510 | handler->ns_clone_flags |= ns_info[i].clone_flag; |
f7a0c6ee CB |
1511 | } else if (conf->ns_clone) { |
1512 | if ((conf->ns_clone & ns_info[i].clone_flag)) | |
becad0ec | 1513 | handler->ns_clone_flags |= ns_info[i].clone_flag; |
8bc8c715 CB |
1514 | } else { |
1515 | if (i == LXC_NS_USER && lxc_list_empty(&handler->conf->id_map)) | |
1516 | continue; | |
1517 | ||
1518 | if (i == LXC_NS_NET && lxc_requests_empty_network(handler)) | |
1519 | continue; | |
1520 | ||
7bd05339 CB |
1521 | if (i == LXC_NS_CGROUP && !cgns_supported()) |
1522 | continue; | |
1523 | ||
70fd7fc9 CB |
1524 | if (i == LXC_NS_TIME && !wants_timens) |
1525 | continue; | |
1526 | ||
becad0ec | 1527 | handler->ns_clone_flags |= ns_info[i].clone_flag; |
8bc8c715 | 1528 | } |
03df7ab5 | 1529 | |
8bc8c715 CB |
1530 | if (!conf->ns_share[i]) |
1531 | continue; | |
5af9369b | 1532 | |
becad0ec | 1533 | handler->ns_clone_flags &= ~ns_info[i].clone_flag; |
8bc8c715 | 1534 | TRACE("Sharing %s namespace", ns_info[i].proc_name); |
5af9369b CB |
1535 | } |
1536 | ||
70fd7fc9 CB |
1537 | if (wants_timens && (conf->ns_keep & ns_info[LXC_NS_TIME].clone_flag)) |
1538 | return log_trace_errno(-1, EINVAL, "Requested to keep time namespace while also specifying offsets"); | |
1539 | ||
5af9369b | 1540 | return 0; |
f813849c TA |
1541 | } |
1542 | ||
8deca6c9 CB |
1543 | /* Note that this function is used with clone(CLONE_VM). Some glibc versions |
1544 | * used to reset the pid/tid to -1 when CLONE_VM was used without CLONE_THREAD. | |
1545 | * But since the memory between parent and child is shared on CLONE_VM this | |
1546 | * would invalidate the getpid() cache that glibc used to maintain and so | |
1547 | * getpid() in the child would return the parent's pid. This is all fixed in | |
1548 | * newer glibc versions where the getpid() cache is removed and the pid/tid is | |
1549 | * not reset anymore. | |
b35ddc5b | 1550 | * However, if for whatever reason you - dear committer - somehow need to get the |
8deca6c9 | 1551 | * pid of the dummy intermediate process for do_share_ns() you need to call |
0059379f CB |
1552 | * lxc_raw_getpid(). The next lxc_raw_clone() call does not employ CLONE_VM and |
1553 | * will be fine. | |
8deca6c9 CB |
1554 | */ |
1555 | static inline int do_share_ns(void *arg) | |
fa3a5b22 | 1556 | { |
8deca6c9 | 1557 | int i, flags, ret; |
fa3a5b22 CB |
1558 | struct lxc_handler *handler = arg; |
1559 | ||
8deca6c9 CB |
1560 | for (i = 0; i < LXC_NS_MAX; i++) { |
1561 | if (handler->nsfd[i] < 0) | |
1562 | continue; | |
fa3a5b22 | 1563 | |
8deca6c9 | 1564 | ret = setns(handler->nsfd[i], 0); |
c0b48eff CB |
1565 | if (ret < 0) { |
1566 | /* | |
1567 | * Note that joining a user and/or mount namespace | |
1568 | * requires the process is not multithreaded otherwise | |
1569 | * setns() will fail here. | |
1570 | */ | |
1571 | SYSERROR("Failed to inherit %s namespace", | |
1572 | ns_info[i].proc_name); | |
8deca6c9 | 1573 | return -1; |
c0b48eff | 1574 | } |
fa3a5b22 | 1575 | |
8deca6c9 | 1576 | DEBUG("Inherited %s namespace", ns_info[i].proc_name); |
fa3a5b22 CB |
1577 | } |
1578 | ||
becad0ec | 1579 | flags = handler->ns_on_clone_flags; |
8deca6c9 | 1580 | flags |= CLONE_PARENT; |
33942046 CB |
1581 | handler->pid = lxc_raw_clone_cb(do_start, handler, CLONE_PIDFD | flags, |
1582 | &handler->pidfd); | |
8deca6c9 | 1583 | if (handler->pid < 0) |
fa3a5b22 CB |
1584 | return -1; |
1585 | ||
8deca6c9 | 1586 | return 0; |
fa3a5b22 CB |
1587 | } |
1588 | ||
480588e6 CB |
1589 | /* lxc_spawn() performs crucial setup tasks and clone()s the new process which |
1590 | * exec()s the requested container binary. | |
1591 | * Note that lxc_spawn() runs in the parent namespaces. Any operations performed | |
1592 | * right here should be double checked if they'd pose a security risk. (For | |
1593 | * example, any {u}mount() operations performed here will be reflected on the | |
1594 | * host!) | |
1595 | */ | |
74a3920a | 1596 | static int lxc_spawn(struct lxc_handler *handler) |
59eb99ba | 1597 | { |
f62cf1d4 | 1598 | __do_close int data_sock0 = -EBADF, data_sock1 = -EBADF; |
8deca6c9 | 1599 | int i, ret; |
08dd2805 | 1600 | char pidstr[20]; |
57927bf2 CB |
1601 | bool wants_to_map_ids; |
1602 | struct lxc_list *id_map; | |
6e5fc7a5 | 1603 | const char *name = handler->name; |
28d9e29e | 1604 | const char *lxcpath = handler->lxcpath; |
2202afc9 | 1605 | bool share_ns = false; |
28d9e29e | 1606 | struct lxc_conf *conf = handler->conf; |
2202afc9 | 1607 | struct cgroup_ops *cgroup_ops = handler->cgroup_ops; |
9f30a190 | 1608 | |
28d9e29e | 1609 | id_map = &conf->id_map; |
57927bf2 | 1610 | wants_to_map_ids = !lxc_list_empty(id_map); |
2f2623ec | 1611 | |
28d9e29e | 1612 | for (i = 0; i < LXC_NS_MAX; i++) { |
b074bbf1 | 1613 | if (!conf->ns_share[i]) |
28d9e29e CB |
1614 | continue; |
1615 | ||
b074bbf1 | 1616 | handler->nsfd[i] = lxc_inherit_namespace(conf->ns_share[i], lxcpath, ns_info[i].proc_name); |
28d9e29e CB |
1617 | if (handler->nsfd[i] < 0) |
1618 | return -1; | |
fa3a5b22 | 1619 | |
8deca6c9 | 1620 | share_ns = true; |
28d9e29e | 1621 | } |
50e98013 | 1622 | |
cfc62c60 CB |
1623 | ret = lxc_sync_init(handler); |
1624 | if (ret < 0) | |
9d7f9e52 | 1625 | return -1; |
0ad19a3f | 1626 | |
b9f522c5 | 1627 | ret = socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0, |
790255cf | 1628 | handler->data_sock); |
95ef0d7c CB |
1629 | if (ret < 0) |
1630 | goto out_sync_fini; | |
cdb2a47f CB |
1631 | data_sock0 = handler->data_sock[0]; |
1632 | data_sock1 = handler->data_sock[1]; | |
e8bd4e43 | 1633 | |
5af9369b | 1634 | ret = resolve_clone_flags(handler); |
95ef0d7c CB |
1635 | if (ret < 0) |
1636 | goto out_sync_fini; | |
9f30a190 | 1637 | |
03ca4af8 TP |
1638 | if (handler->ns_clone_flags & CLONE_NEWNET) { |
1639 | ret = lxc_find_gateway_addresses(handler); | |
1640 | if (ret) { | |
1641 | ERROR("Failed to find gateway addresses"); | |
1642 | goto out_sync_fini; | |
1643 | } | |
1644 | } | |
1645 | ||
e8b181f5 | 1646 | if (!cgroup_ops->payload_create(cgroup_ops, handler)) { |
cfc62c60 | 1647 | ERROR("Failed creating cgroups"); |
47d8fb3b CS |
1648 | goto out_delete_net; |
1649 | } | |
1650 | ||
408da065 CB |
1651 | /* If the rootfs is not a blockdev, prevent the container from marking |
1652 | * it readonly. | |
1653 | * If the container is unprivileged then skip rootfs pinning. | |
0c547523 | 1654 | */ |
0ee35059 | 1655 | if (!wants_to_map_ids) { |
28d9e29e | 1656 | handler->pinfd = pin_rootfs(conf->rootfs.path); |
b87ee312 | 1657 | if (handler->pinfd == -EBADF) |
cfc62c60 | 1658 | INFO("Failed to pin the rootfs for container \"%s\"", handler->name); |
5e32a990 | 1659 | } |
0c547523 | 1660 | |
408da065 | 1661 | /* Create a process in a new set of namespaces. */ |
becad0ec CB |
1662 | handler->ns_on_clone_flags = handler->ns_clone_flags; |
1663 | if (handler->ns_clone_flags & CLONE_NEWUSER) { | |
408da065 CB |
1664 | /* If CLONE_NEWUSER and CLONE_NEWNET was requested, we need to |
1665 | * clone a new user namespace first and only later unshare our | |
1666 | * network namespace to ensure that network devices ownership is | |
1667 | * set up correctly. | |
1668 | */ | |
becad0ec | 1669 | handler->ns_on_clone_flags &= ~CLONE_NEWNET; |
408da065 | 1670 | } |
5af9369b | 1671 | /* The cgroup namespace gets unshare()ed not clone()ed. */ |
becad0ec | 1672 | handler->ns_on_clone_flags &= ~CLONE_NEWCGROUP; |
732375f5 | 1673 | |
70fd7fc9 CB |
1674 | /* The time namespace (currently) gets unshare()ed not clone()ed. */ |
1675 | handler->ns_on_clone_flags &= ~CLONE_NEWTIME; | |
1676 | ||
4e232466 CB |
1677 | if (share_ns) { |
1678 | pid_t attacher_pid; | |
1679 | ||
cfc62c60 | 1680 | attacher_pid = lxc_clone(do_share_ns, handler, |
33258b95 | 1681 | CLONE_VFORK | CLONE_VM | CLONE_FILES, NULL); |
4e232466 CB |
1682 | if (attacher_pid < 0) { |
1683 | SYSERROR(LXC_CLONE_ERROR); | |
1684 | goto out_delete_net; | |
1685 | } | |
1686 | ||
1687 | ret = wait_for_pid(attacher_pid); | |
1688 | if (ret < 0) { | |
1689 | SYSERROR("Intermediate process failed"); | |
1690 | goto out_delete_net; | |
1691 | } | |
1692 | } else { | |
cfc62c60 | 1693 | handler->pid = lxc_raw_clone_cb(do_start, handler, |
33942046 CB |
1694 | CLONE_PIDFD | handler->ns_on_clone_flags, |
1695 | &handler->pidfd); | |
4e232466 CB |
1696 | } |
1697 | if (handler->pid < 0) { | |
1698 | SYSERROR(LXC_CLONE_ERROR); | |
7fef7a06 | 1699 | goto out_delete_net; |
0ad19a3f | 1700 | } |
fa3a5b22 | 1701 | TRACE("Cloned child process %d", handler->pid); |
1bd8d726 | 1702 | |
39293f22 CB |
1703 | /* Verify that we can actually make use of pidfds. */ |
1704 | if (!lxc_can_use_pidfd(handler->pidfd)) | |
1705 | close_prot_errno_disarm(handler->pidfd); | |
1706 | ||
1871e646 CB |
1707 | ret = snprintf(pidstr, 20, "%d", handler->pid); |
1708 | if (ret < 0 || ret >= 20) | |
1709 | goto out_delete_net; | |
1710 | ||
1711 | ret = setenv("LXC_PID", pidstr, 1); | |
1712 | if (ret < 0) | |
1713 | SYSERROR("Failed to set environment variable: LXC_PID=%s", pidstr); | |
1714 | ||
9662e444 | 1715 | for (i = 0; i < LXC_NS_MAX; i++) |
becad0ec | 1716 | if (handler->ns_on_clone_flags & ns_info[i].clone_flag) |
28d9e29e | 1717 | INFO("Cloned %s", ns_info[i].flag_name); |
0ad19a3f | 1718 | |
4cb53844 | 1719 | if (!lxc_try_preserve_namespaces(handler, handler->ns_on_clone_flags, handler->pid)) { |
28d9e29e CB |
1720 | ERROR("Failed to preserve cloned namespaces for lxc.hook.stop"); |
1721 | goto out_delete_net; | |
1722 | } | |
b6b2b194 | 1723 | |
3c22086f CLG |
1724 | lxc_sync_fini_child(handler); |
1725 | ||
4d8bdfa0 CB |
1726 | if (lxc_abstract_unix_send_fds(handler->data_sock[0], &handler->monitor_status_fd, 1, NULL, 0) < 0) { |
1727 | ERROR("Failed to send status file descriptor to child process"); | |
1728 | goto out_delete_net; | |
1729 | } | |
1730 | close_prot_errno_disarm(handler->monitor_status_fd); | |
1731 | ||
408da065 CB |
1732 | /* Map the container uids. The container became an invalid userid the |
1733 | * moment it was cloned with CLONE_NEWUSER. This call doesn't change | |
1734 | * anything immediately, but allows the container to setuid(0) (0 being | |
1735 | * mapped to something else on the host.) later to become a valid uid | |
1736 | * again. | |
1737 | */ | |
fa3a5b22 | 1738 | if (wants_to_map_ids) { |
df3c5314 | 1739 | if (!handler->conf->ns_share[LXC_NS_USER] && |
46186acd | 1740 | (handler->conf->ns_keep & CLONE_NEWUSER) == 0) { |
fa3a5b22 CB |
1741 | ret = lxc_map_ids(id_map, handler->pid); |
1742 | if (ret < 0) { | |
1743 | ERROR("Failed to set up id mapping."); | |
1744 | goto out_delete_net; | |
1745 | } | |
1746 | } | |
5b1e83cb SH |
1747 | } |
1748 | ||
cfc62c60 CB |
1749 | ret = lxc_sync_wake_child(handler, LXC_SYNC_STARTUP); |
1750 | if (ret < 0) | |
5b1e83cb | 1751 | goto out_delete_net; |
5b1e83cb | 1752 | |
cfc62c60 CB |
1753 | ret = lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE); |
1754 | if (ret < 0) | |
5b1e83cb | 1755 | goto out_delete_net; |
0ad19a3f | 1756 | |
c581d2a6 | 1757 | if (!cgroup_ops->setup_limits_legacy(cgroup_ops, handler->conf, false)) { |
cfc62c60 | 1758 | ERROR("Failed to setup cgroup limits for container \"%s\"", name); |
6031a6e5 DE |
1759 | goto out_delete_net; |
1760 | } | |
1761 | ||
fe70edee | 1762 | if (!cgroup_ops->payload_enter(cgroup_ops, handler)) { |
b1ee458e | 1763 | ERROR("Failed to enter cgroups"); |
7fef7a06 | 1764 | goto out_delete_net; |
fe70edee | 1765 | } |
218d4250 | 1766 | |
c581d2a6 CB |
1767 | if (!cgroup_ops->payload_delegate_controllers(cgroup_ops)) { |
1768 | ERROR("Failed to delegate controllers to payload cgroup"); | |
1769 | goto out_delete_net; | |
1770 | } | |
1771 | ||
1772 | if (!cgroup_ops->setup_limits(cgroup_ops, handler)) { | |
1773 | ERROR("Failed to setup cgroup limits for container \"%s\"", name); | |
1774 | goto out_delete_net; | |
1775 | } | |
1776 | ||
2202afc9 | 1777 | if (!cgroup_ops->chown(cgroup_ops, handler->conf)) |
0996e18a SH |
1778 | goto out_delete_net; |
1779 | ||
aa0c0e7b RK |
1780 | /* If not done yet, we're now ready to preserve the network namespace */ |
1781 | if (handler->nsfd[LXC_NS_NET] < 0) { | |
1782 | ret = lxc_try_preserve_ns(handler->pid, "net"); | |
1783 | if (ret < 0) { | |
1784 | if (ret != -EOPNOTSUPP) { | |
1785 | SYSERROR("Failed to preserve net namespace"); | |
1786 | goto out_delete_net; | |
1787 | } | |
1788 | } else { | |
1789 | handler->nsfd[LXC_NS_NET] = ret; | |
1790 | DEBUG("Preserved net namespace via fd %d", ret); | |
4cb53844 | 1791 | } |
fa3a5b22 | 1792 | } |
aa0c0e7b RK |
1793 | ret = lxc_netns_set_nsid(handler->nsfd[LXC_NS_NET]); |
1794 | if (ret < 0) | |
1795 | SYSWARN("Failed to allocate new network namespace id"); | |
1796 | else | |
1797 | TRACE("Allocated new network namespace id"); | |
fa3a5b22 | 1798 | |
408da065 | 1799 | /* Create the network configuration. */ |
becad0ec | 1800 | if (handler->ns_clone_flags & CLONE_NEWNET) { |
e389f2af | 1801 | ret = lxc_create_network(handler); |
cfc62c60 | 1802 | if (ret < 0) { |
e389f2af | 1803 | ERROR("Failed to create the network"); |
7fef7a06 | 1804 | goto out_delete_net; |
82d5ae15 | 1805 | } |
74c6e2b0 | 1806 | |
3c09b97c | 1807 | ret = lxc_network_send_to_child(handler); |
cfc62c60 | 1808 | if (ret < 0) { |
e389f2af | 1809 | ERROR("Failed to send veth names to child"); |
74c6e2b0 CB |
1810 | goto out_delete_net; |
1811 | } | |
0ad19a3f | 1812 | } |
1813 | ||
61d7a733 YT |
1814 | if (!lxc_list_empty(&conf->procs)) { |
1815 | ret = setup_proc_filesystem(&conf->procs, handler->pid); | |
1816 | if (ret < 0) | |
1817 | goto out_delete_net; | |
1818 | } | |
1819 | ||
408da065 CB |
1820 | /* Tell the child to continue its initialization. We'll get |
1821 | * LXC_SYNC_CGROUP when it is ready for us to setup cgroups. | |
3c22086f | 1822 | */ |
cfc62c60 CB |
1823 | ret = lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE); |
1824 | if (ret < 0) | |
544a48a0 SH |
1825 | goto out_delete_net; |
1826 | ||
cfc62c60 CB |
1827 | if (!lxc_list_empty(&conf->limits)) { |
1828 | ret = setup_resource_limits(&conf->limits, handler->pid); | |
1829 | if (ret < 0) { | |
1830 | ERROR("Failed to setup resource limits"); | |
1831 | goto out_delete_net; | |
1832 | } | |
c6d09e15 WB |
1833 | } |
1834 | ||
cfc62c60 CB |
1835 | ret = lxc_sync_barrier_child(handler, LXC_SYNC_CGROUP_UNSHARE); |
1836 | if (ret < 0) | |
f4152036 CB |
1837 | goto out_delete_net; |
1838 | ||
432faf20 WB |
1839 | /* |
1840 | * with isolation the limiting devices cgroup was already setup, so | |
1841 | * only setup devices here if we have no namespace directory | |
1842 | */ | |
1843 | if (!handler->conf->cgroup_meta.namespace_dir && | |
1844 | !cgroup_ops->setup_limits_legacy(cgroup_ops, handler->conf, true)) { | |
cfc62c60 | 1845 | ERROR("Failed to setup legacy device cgroup controller limits"); |
b98f7d6e | 1846 | goto out_delete_net; |
544a48a0 | 1847 | } |
cfc62c60 | 1848 | TRACE("Set up legacy device cgroup controller limits"); |
544a48a0 | 1849 | |
bf651989 CB |
1850 | if (!cgroup_ops->devices_activate(cgroup_ops, handler)) { |
1851 | ERROR("Failed to setup cgroup2 device controller limits"); | |
1852 | goto out_delete_net; | |
1853 | } | |
1854 | TRACE("Set up cgroup2 device controller limits"); | |
1855 | ||
becad0ec | 1856 | if (handler->ns_clone_flags & CLONE_NEWCGROUP) { |
8bf3abfb | 1857 | /* Now we're ready to preserve the cgroup namespace */ |
4cb53844 | 1858 | ret = lxc_try_preserve_ns(handler->pid, "cgroup"); |
8bf3abfb | 1859 | if (ret < 0) { |
4cb53844 | 1860 | if (ret != -EOPNOTSUPP) { |
6d1400b5 | 1861 | SYSERROR("Failed to preserve cgroup namespace"); |
4cb53844 CB |
1862 | goto out_delete_net; |
1863 | } | |
1864 | } else { | |
1865 | handler->nsfd[LXC_NS_CGROUP] = ret; | |
1866 | DEBUG("Preserved cgroup namespace via fd %d", ret); | |
8bf3abfb | 1867 | } |
8bf3abfb CB |
1868 | } |
1869 | ||
78eb6aa6 CB |
1870 | cgroup_ops->payload_finalize(cgroup_ops); |
1871 | TRACE("Finished setting up cgroups"); | |
1872 | ||
08dd2805 | 1873 | /* Run any host-side start hooks */ |
cfc62c60 CB |
1874 | ret = run_lxc_hooks(name, "start-host", conf, NULL); |
1875 | if (ret < 0) { | |
1876 | ERROR("Failed to run lxc.hook.start-host"); | |
341ed84c | 1877 | goto out_delete_net; |
08dd2805 SH |
1878 | } |
1879 | ||
408da065 CB |
1880 | /* Tell the child to complete its initialization and wait for it to exec |
1881 | * or return an error. (The child will never return | |
08dd2805 | 1882 | * LXC_SYNC_READY_START+1. It will either close the sync pipe, causing |
408da065 CB |
1883 | * lxc_sync_barrier_child to return success, or return a different |
1884 | * value, causing us to error out). | |
544a48a0 | 1885 | */ |
cfc62c60 CB |
1886 | ret = lxc_sync_barrier_child(handler, LXC_SYNC_READY_START); |
1887 | if (ret < 0) | |
341ed84c | 1888 | goto out_delete_net; |
0ad19a3f | 1889 | |
e389f2af CB |
1890 | if (handler->ns_clone_flags & CLONE_NEWNET) { |
1891 | ret = lxc_network_recv_name_and_ifindex_from_child(handler); | |
1892 | if (ret < 0) { | |
1893 | ERROR("Failed to receive names and ifindices for network devices from child"); | |
1894 | goto out_delete_net; | |
1895 | } | |
790255cf CB |
1896 | } |
1897 | ||
1898 | /* Now all networks are created, network devices are moved into place, | |
ad2ddfcd | 1899 | * and the correct names and ifindices in the respective namespaces have |
790255cf CB |
1900 | * been recorded. The corresponding structs have now all been filled. So |
1901 | * log them for debugging purposes. | |
1902 | */ | |
28d9e29e | 1903 | lxc_log_configured_netdevs(conf); |
790255cf | 1904 | |
f4152036 | 1905 | /* Read tty fds allocated by child. */ |
cfc62c60 CB |
1906 | ret = lxc_recv_ttys_from_child(handler); |
1907 | if (ret < 0) { | |
1908 | ERROR("Failed to receive tty info from child process"); | |
f4152036 CB |
1909 | goto out_delete_net; |
1910 | } | |
1911 | ||
c3e3c21a CB |
1912 | ret = lxc_seccomp_recv_notifier_fd(&handler->conf->seccomp, data_sock1); |
1913 | if (ret < 0) { | |
1914 | SYSERROR("Failed to receive seccomp notify fd from child"); | |
1915 | goto out_delete_net; | |
cdb2a47f | 1916 | } |
cdb2a47f | 1917 | |
cfc62c60 CB |
1918 | ret = handler->ops->post_start(handler, handler->data); |
1919 | if (ret < 0) | |
e6126dbe MN |
1920 | goto out_abort; |
1921 | ||
cfc62c60 CB |
1922 | ret = lxc_set_state(name, handler, RUNNING); |
1923 | if (ret < 0) { | |
1924 | ERROR("Failed to set state to \"%s\"", lxc_state2str(RUNNING)); | |
59eb99ba | 1925 | goto out_abort; |
3f21c114 | 1926 | } |
22ebac19 | 1927 | |
3c22086f | 1928 | lxc_sync_fini(handler); |
0c547523 | 1929 | |
e6126dbe | 1930 | return 0; |
1ac470c0 | 1931 | |
7fef7a06 | 1932 | out_delete_net: |
becad0ec | 1933 | if (handler->ns_clone_flags & CLONE_NEWNET) |
bb84beda | 1934 | lxc_delete_network(handler); |
1bd8d726 | 1935 | |
59eb99ba | 1936 | out_abort: |
0c5859ff | 1937 | lxc_abort(handler); |
95ef0d7c CB |
1938 | |
1939 | out_sync_fini: | |
3c22086f | 1940 | lxc_sync_fini(handler); |
a783a414 | 1941 | close_prot_errno_disarm(handler->pinfd); |
5c068da9 | 1942 | |
b79fcd86 | 1943 | return -1; |
59eb99ba | 1944 | } |
0ad19a3f | 1945 | |
7bdf97d2 CB |
1946 | int __lxc_start(struct lxc_handler *handler, struct lxc_operations *ops, |
1947 | void *data, const char *lxcpath, bool daemonize, int *error_num) | |
59eb99ba | 1948 | { |
c30e9b19 | 1949 | int ret, status; |
7bdf97d2 | 1950 | const char *name = handler->name; |
aa460476 | 1951 | struct lxc_conf *conf = handler->conf; |
72068e74 | 1952 | struct cgroup_ops *cgroup_ops; |
80090207 | 1953 | |
c30e9b19 CB |
1954 | ret = lxc_init(name, handler); |
1955 | if (ret < 0) { | |
1956 | ERROR("Failed to initialize container \"%s\"", name); | |
9ad66589 | 1957 | goto out_abort; |
0ad19a3f | 1958 | } |
ee70bf78 CLG |
1959 | handler->ops = ops; |
1960 | handler->data = data; | |
bb955810 | 1961 | handler->daemonize = daemonize; |
72068e74 | 1962 | cgroup_ops = handler->cgroup_ops; |
e6126dbe | 1963 | |
76a26f55 | 1964 | if (!attach_block_device(handler->conf)) { |
c30e9b19 | 1965 | ERROR("Failed to attach block device"); |
575ea467 | 1966 | ret = -1; |
9ad66589 | 1967 | goto out_abort; |
76a26f55 SH |
1968 | } |
1969 | ||
72068e74 CB |
1970 | if (!cgroup_ops->monitor_create(cgroup_ops, handler)) { |
1971 | ERROR("Failed to create monitor cgroup"); | |
575ea467 | 1972 | ret = -1; |
9ad66589 | 1973 | goto out_abort; |
eeef32bb CB |
1974 | } |
1975 | ||
c581d2a6 | 1976 | if (!cgroup_ops->monitor_enter(cgroup_ops, handler)) { |
eeef32bb | 1977 | ERROR("Failed to enter monitor cgroup"); |
575ea467 | 1978 | ret = -1; |
9ad66589 | 1979 | goto out_abort; |
72068e74 CB |
1980 | } |
1981 | ||
c581d2a6 CB |
1982 | if (!cgroup_ops->monitor_delegate_controllers(cgroup_ops)) { |
1983 | ERROR("Failed to delegate controllers to monitor cgroup"); | |
1984 | ret = -1; | |
9ad66589 | 1985 | goto out_abort; |
c581d2a6 CB |
1986 | } |
1987 | ||
35120d9c | 1988 | if (geteuid() == 0 && !lxc_list_empty(&conf->id_map)) { |
408da065 | 1989 | /* If the backing store is a device, mount it here and now. */ |
35120d9c | 1990 | if (rootfs_is_blockdev(conf)) { |
c30e9b19 CB |
1991 | ret = unshare(CLONE_NEWNS); |
1992 | if (ret < 0) { | |
1993 | ERROR("Failed to unshare CLONE_NEWNS"); | |
9ad66589 | 1994 | goto out_abort; |
35120d9c | 1995 | } |
c30e9b19 | 1996 | INFO("Unshared CLONE_NEWNS"); |
408da065 | 1997 | |
9e61fb1f | 1998 | turn_into_dependent_mounts(); |
8ce1abc2 | 1999 | ret = lxc_setup_rootfs_prepare_root(conf, name, lxcpath); |
c30e9b19 CB |
2000 | if (ret < 0) { |
2001 | ERROR("Error setting up rootfs mount as root before spawn"); | |
9ad66589 | 2002 | goto out_abort; |
35120d9c | 2003 | } |
c30e9b19 | 2004 | INFO("Set up container rootfs as host root"); |
35120d9c SH |
2005 | } |
2006 | } | |
2007 | ||
c30e9b19 CB |
2008 | ret = lxc_spawn(handler); |
2009 | if (ret < 0) { | |
2010 | ERROR("Failed to spawn container \"%s\"", name); | |
76a26f55 | 2011 | goto out_detach_blockdev; |
0ad19a3f | 2012 | } |
2013 | ||
80308d07 | 2014 | handler->conf->reboot = REBOOT_NONE; |
8bee8851 | 2015 | |
c30e9b19 CB |
2016 | ret = lxc_poll(name, handler); |
2017 | if (ret) { | |
2018 | ERROR("LXC mainloop exited with error: %d", ret); | |
9ad66589 | 2019 | goto out_delete_network; |
59eb99ba | 2020 | } |
0ad19a3f | 2021 | |
321db026 DJ |
2022 | if (!handler->init_died && handler->pid > 0) { |
2023 | ERROR("Child process is not killed"); | |
575ea467 | 2024 | ret = -1; |
9ad66589 | 2025 | goto out_delete_network; |
321db026 DJ |
2026 | } |
2027 | ||
c30e9b19 CB |
2028 | status = lxc_wait_for_pid_status(handler->pid); |
2029 | if (status < 0) | |
2030 | SYSERROR("Failed to retrieve status for %d", handler->pid); | |
e043236e | 2031 | |
408da065 CB |
2032 | /* If the child process exited but was not signaled, it didn't call |
2033 | * reboot. This should mean it was an lxc-execute which simply exited. | |
2034 | * In any case, treat it as a 'halt'. | |
8b004f07 | 2035 | */ |
d028235d | 2036 | if (WIFSIGNALED(status)) { |
8b004f07 SH |
2037 | switch(WTERMSIG(status)) { |
2038 | case SIGINT: /* halt */ | |
c30e9b19 | 2039 | DEBUG("Container \"%s\" is halting", name); |
8b004f07 SH |
2040 | break; |
2041 | case SIGHUP: /* reboot */ | |
c30e9b19 | 2042 | DEBUG("Container \"%s\" is rebooting", name); |
80308d07 | 2043 | handler->conf->reboot = REBOOT_REQ; |
8b004f07 | 2044 | break; |
c2b9bd9e | 2045 | case SIGSYS: /* seccomp */ |
c30e9b19 | 2046 | DEBUG("Container \"%s\" violated its seccomp policy", name); |
c2b9bd9e | 2047 | break; |
8b004f07 | 2048 | default: |
c30e9b19 | 2049 | DEBUG("Unknown exit status for container \"%s\" init %d", name, WTERMSIG(status)); |
8b004f07 SH |
2050 | break; |
2051 | } | |
d028235d | 2052 | } |
828695d9 | 2053 | |
c30e9b19 CB |
2054 | ret = lxc_restore_phys_nics_to_netns(handler); |
2055 | if (ret < 0) | |
818a57fc | 2056 | ERROR("Failed to move physical network devices back to parent network namespace"); |
ce5782df | 2057 | |
a783a414 | 2058 | close_prot_errno_disarm(handler->pinfd); |
5c068da9 | 2059 | |
1787abca | 2060 | lxc_monitor_send_exit_code(name, status, handler->lxcpath); |
eb808539 | 2061 | lxc_error_set_and_log(handler->pid, status); |
a3b4f3d6 TA |
2062 | if (error_num) |
2063 | *error_num = handler->exit_status; | |
1bd8d726 | 2064 | |
9ad66589 CB |
2065 | /* These are not the droids you are looking for. */ |
2066 | __private_goto1: | |
bb84beda | 2067 | lxc_delete_network(handler); |
74a2b586 | 2068 | |
9ad66589 | 2069 | __private_goto2: |
76a26f55 SH |
2070 | detach_block_device(handler->conf); |
2071 | ||
9ad66589 | 2072 | __private_goto3: |
fd5be714 | 2073 | lxc_end(handler); |
65a2dc8b | 2074 | |
c30e9b19 | 2075 | return ret; |
0ad19a3f | 2076 | |
9ad66589 CB |
2077 | /* These are the droids you are looking for. */ |
2078 | out_abort: | |
0c5859ff | 2079 | lxc_abort(handler); |
9ad66589 | 2080 | goto __private_goto3; |
65a2dc8b CB |
2081 | |
2082 | out_detach_blockdev: | |
0c5859ff | 2083 | lxc_abort(handler); |
9ad66589 | 2084 | goto __private_goto2; |
65a2dc8b | 2085 | |
9ad66589 | 2086 | out_delete_network: |
0c5859ff | 2087 | lxc_abort(handler); |
9ad66589 | 2088 | goto __private_goto1; |
0ad19a3f | 2089 | } |
ee70bf78 CLG |
2090 | |
2091 | struct start_args { | |
2092 | char *const *argv; | |
2093 | }; | |
2094 | ||
2095 | static int start(struct lxc_handler *handler, void* data) | |
2096 | { | |
2097 | struct start_args *arg = data; | |
2098 | ||
984984e4 | 2099 | NOTICE("Exec'ing \"%s\"", arg->argv[0]); |
ee70bf78 CLG |
2100 | |
2101 | execvp(arg->argv[0], arg->argv); | |
984984e4 | 2102 | SYSERROR("Failed to exec \"%s\"", arg->argv[0]); |
ee70bf78 CLG |
2103 | return 0; |
2104 | } | |
2105 | ||
2106 | static int post_start(struct lxc_handler *handler, void* data) | |
2107 | { | |
2108 | struct start_args *arg = data; | |
2109 | ||
4c8e880e | 2110 | NOTICE("Started \"%s\" with pid \"%d\"", arg->argv[0], handler->pid); |
ee70bf78 CLG |
2111 | return 0; |
2112 | } | |
2113 | ||
2114 | static struct lxc_operations start_ops = { | |
2115 | .start = start, | |
2116 | .post_start = post_start | |
2117 | }; | |
2118 | ||
7bdf97d2 | 2119 | int lxc_start(char *const argv[], struct lxc_handler *handler, |
bb955810 | 2120 | const char *lxcpath, bool daemonize, int *error_num) |
ee70bf78 CLG |
2121 | { |
2122 | struct start_args start_arg = { | |
2123 | .argv = argv, | |
2124 | }; | |
2125 | ||
b2efeb0b | 2126 | TRACE("Doing lxc_start"); |
7bdf97d2 | 2127 | return __lxc_start(handler, &start_ops, &start_arg, lxcpath, daemonize, error_num); |
ee70bf78 | 2128 | } |
28272964 CB |
2129 | |
2130 | static void lxc_destroy_container_on_signal(struct lxc_handler *handler, | |
2131 | const char *name) | |
2132 | { | |
245100a0 | 2133 | char destroy[PATH_MAX]; |
f01f7975 | 2134 | struct lxc_container *c; |
ae3beac9 CB |
2135 | int ret = 0; |
2136 | bool bret = true; | |
2137 | ||
df31363a | 2138 | if (handler->conf->rootfs.path && handler->conf->rootfs.mount) { |
790255cf | 2139 | bret = do_destroy_container(handler); |
28272964 | 2140 | if (!bret) { |
ae3beac9 | 2141 | ERROR("Error destroying rootfs for container \"%s\"", name); |
28272964 CB |
2142 | return; |
2143 | } | |
2144 | } | |
ae3beac9 | 2145 | INFO("Destroyed rootfs for container \"%s\"", name); |
28272964 | 2146 | |
245100a0 CB |
2147 | ret = snprintf(destroy, PATH_MAX, "%s/%s", handler->lxcpath, name); |
2148 | if (ret < 0 || ret >= PATH_MAX) { | |
ae3beac9 | 2149 | ERROR("Error destroying directory for container \"%s\"", name); |
28272964 CB |
2150 | return; |
2151 | } | |
2152 | ||
f01f7975 CB |
2153 | c = lxc_container_new(name, handler->lxcpath); |
2154 | if (c) { | |
2155 | if (container_disk_lock(c)) { | |
ae3beac9 | 2156 | INFO("Could not update lxc_snapshots file"); |
f01f7975 CB |
2157 | lxc_container_put(c); |
2158 | } else { | |
2159 | mod_all_rdeps(c, false); | |
2160 | container_disk_unlock(c); | |
2161 | lxc_container_put(c); | |
2162 | } | |
2163 | } | |
2164 | ||
d0fbc7ba | 2165 | if (!handler->am_root) |
94b0a3ac CB |
2166 | ret = userns_exec_full(handler->conf, lxc_rmdir_onedev_wrapper, |
2167 | destroy, "lxc_rmdir_onedev_wrapper"); | |
28272964 CB |
2168 | else |
2169 | ret = lxc_rmdir_onedev(destroy, NULL); | |
2170 | ||
2171 | if (ret < 0) { | |
ae3beac9 | 2172 | ERROR("Error destroying directory for container \"%s\"", name); |
28272964 CB |
2173 | return; |
2174 | } | |
ae3beac9 | 2175 | INFO("Destroyed directory for container \"%s\"", name); |
28272964 CB |
2176 | } |
2177 | ||
2178 | static int lxc_rmdir_onedev_wrapper(void *data) | |
2179 | { | |
2180 | char *arg = (char *) data; | |
2181 | return lxc_rmdir_onedev(arg, NULL); | |
2182 | } | |
2183 | ||
134df645 CB |
2184 | static bool do_destroy_container(struct lxc_handler *handler) |
2185 | { | |
94b0a3ac CB |
2186 | int ret; |
2187 | ||
d0fbc7ba | 2188 | if (!handler->am_root) { |
94b0a3ac CB |
2189 | ret = userns_exec_full(handler->conf, storage_destroy_wrapper, |
2190 | handler->conf, "storage_destroy_wrapper"); | |
2191 | if (ret < 0) | |
d028235d | 2192 | return false; |
10bc1861 | 2193 | |
d028235d SG |
2194 | return true; |
2195 | } | |
10bc1861 | 2196 | |
790255cf | 2197 | return storage_destroy(handler->conf); |
28272964 | 2198 | } |