]>
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" |
59eac805 | 33 | #include "compiler.h" |
f3787121 | 34 | #include "conf.h" |
d38dd64a | 35 | #include "config.h" |
74c6e2b0 | 36 | #include "confile_utils.h" |
e2bcd7db | 37 | #include "error.h" |
7fbb15ec | 38 | #include "file_utils.h" |
1cc8bd4b | 39 | #include "list.h" |
f3787121 | 40 | #include "log.h" |
d38dd64a | 41 | #include "lsm/lsm.h" |
f2d5a09d | 42 | #include "lxccontainer.h" |
f3787121 CB |
43 | #include "lxclock.h" |
44 | #include "lxcseccomp.h" | |
3ef9b3d3 | 45 | #include "macro.h" |
f3787121 | 46 | #include "mainloop.h" |
4ffeaf27 | 47 | #include "memory_utils.h" |
63376d7d | 48 | #include "monitor.h" |
f549edcc | 49 | #include "namespace.h" |
811ef482 | 50 | #include "network.h" |
f40988c7 | 51 | #include "process_utils.h" |
f3787121 | 52 | #include "start.h" |
6be5397b CB |
53 | #include "storage/storage.h" |
54 | #include "storage/storage_utils.h" | |
0ed9b1bc | 55 | #include "sync.h" |
59524108 | 56 | #include "syscall_wrappers.h" |
0ed9b1bc CB |
57 | #include "terminal.h" |
58 | #include "utils.h" | |
36eb9bde | 59 | |
d38dd64a CB |
60 | #if HAVE_LIBCAP |
61 | #include <sys/capability.h> | |
62 | #endif | |
63 | ||
9de31d5a CB |
64 | #ifndef HAVE_STRLCPY |
65 | #include "include/strlcpy.h" | |
66 | #endif | |
67 | ||
ac2cecc4 | 68 | lxc_log_define(start, lxc); |
36eb9bde | 69 | |
f01f7975 | 70 | extern void mod_all_rdeps(struct lxc_container *c, bool inc); |
790255cf | 71 | static bool do_destroy_container(struct lxc_handler *handler); |
28272964 CB |
72 | static int lxc_rmdir_onedev_wrapper(void *data); |
73 | static void lxc_destroy_container_on_signal(struct lxc_handler *handler, | |
74 | const char *name); | |
75 | ||
c8154066 SH |
76 | static void print_top_failing_dir(const char *path) |
77 | { | |
dcd9a847 | 78 | __do_free char *copy = NULL; |
9e5f5f2f | 79 | int ret; |
4ffeaf27 | 80 | char *e, *p, saved; |
cbb9c7c7 | 81 | |
4ffeaf27 | 82 | copy = must_copy_string(path); |
c8154066 | 83 | p = copy; |
4ffeaf27 | 84 | e = copy + strlen(path); |
cbb9c7c7 | 85 | |
c8154066 | 86 | while (p < e) { |
f3787121 CB |
87 | while (p < e && *p == '/') |
88 | p++; | |
9e5f5f2f | 89 | |
f3787121 CB |
90 | while (p < e && *p != '/') |
91 | p++; | |
9e5f5f2f | 92 | |
c8154066 SH |
93 | saved = *p; |
94 | *p = '\0'; | |
9e5f5f2f CB |
95 | |
96 | ret = access(copy, X_OK); | |
97 | if (ret != 0) { | |
818a57fc | 98 | SYSERROR("Could not access %s. Please grant it x access, or add an ACL for the container " "root", copy); |
c8154066 SH |
99 | return; |
100 | } | |
101 | *p = saved; | |
102 | } | |
103 | } | |
104 | ||
5032bf39 | 105 | static void lxc_put_nsfds(struct lxc_handler *handler) |
408da065 | 106 | { |
818a57fc | 107 | for (int i = 0; i < LXC_NS_MAX; i++) { |
5032bf39 | 108 | if (handler->nsfd[i] < 0) |
39cd919c CB |
109 | continue; |
110 | ||
818a57fc | 111 | close_prot_errno_disarm(handler->nsfd[i]); |
9f30a190 | 112 | } |
9f30a190 MM |
113 | } |
114 | ||
8db6be1b CB |
115 | static int lxc_try_preserve_namespace(struct lxc_handler *handler, |
116 | lxc_namespace_t idx, const char *ns) | |
4cb53844 | 117 | { |
8db6be1b CB |
118 | __do_close int fd = -EBADF; |
119 | int ret; | |
4cb53844 | 120 | |
8db6be1b | 121 | fd = lxc_preserve_ns(handler->pid, ns); |
4cb53844 | 122 | if (fd < 0) { |
818a57fc | 123 | if (errno != ENOENT) |
8db6be1b CB |
124 | return log_error_errno(-EINVAL, errno, |
125 | "Failed to preserve %s namespace", ns); | |
4cb53844 | 126 | |
8db6be1b CB |
127 | return log_warn_errno(-EOPNOTSUPP, errno, |
128 | "Kernel does not support preserving %s namespaces", ns); | |
4cb53844 CB |
129 | } |
130 | ||
8db6be1b CB |
131 | ret = strnprintf(handler->nsfd_paths[idx], |
132 | sizeof(handler->nsfd_paths[idx]), "%s:/proc/%d/fd/%d", | |
133 | ns_info[idx].proc_name, handler->monitor_pid, fd); | |
134 | ||
135 | /* Legacy style argument passing as arguments to hooks. */ | |
136 | handler->hook_argv[handler->hook_argc] = handler->nsfd_paths[idx]; | |
137 | handler->hook_argc++; | |
138 | if (ret < 0) | |
139 | return ret_errno(EIO); | |
140 | ||
141 | DEBUG("Preserved %s namespace via fd %d and stashed path as %s", | |
142 | ns_info[idx].proc_name, fd, handler->nsfd_paths[idx]); | |
143 | ||
144 | handler->nsfd[idx] = move_fd(fd); | |
145 | return 0; | |
4cb53844 CB |
146 | } |
147 | ||
148 | /* lxc_try_preserve_namespaces: open /proc/@pid/ns/@ns for each namespace | |
149 | * specified in ns_clone_flags. | |
4d8ac866 | 150 | * Return true on success, false on failure. |
62d05d9b | 151 | */ |
4cb53844 | 152 | static bool lxc_try_preserve_namespaces(struct lxc_handler *handler, |
8db6be1b | 153 | int ns_clone_flags) |
f3787121 | 154 | { |
8db6be1b CB |
155 | for (lxc_namespace_t ns_idx = 0; ns_idx < LXC_NS_MAX; ns_idx++) |
156 | handler->nsfd[ns_idx] = -EBADF; | |
9f30a190 | 157 | |
8db6be1b CB |
158 | for (lxc_namespace_t ns_idx = 0; ns_idx < LXC_NS_MAX; ns_idx++) { |
159 | int ret; | |
4cb53844 | 160 | |
8db6be1b | 161 | if ((ns_clone_flags & ns_info[ns_idx].clone_flag) == 0) |
9f30a190 | 162 | continue; |
28d9e29e | 163 | |
8db6be1b CB |
164 | ret = lxc_try_preserve_namespace(handler, ns_idx, |
165 | ns_info[ns_idx].proc_name); | |
166 | if (ret < 0) { | |
4cb53844 CB |
167 | /* Do not fail to start container on kernels that do |
168 | * not support interacting with namespaces through | |
169 | * /proc. | |
170 | */ | |
8db6be1b | 171 | if (ret == -EOPNOTSUPP) |
4cb53844 | 172 | continue; |
fa3a5b22 | 173 | |
4cb53844 CB |
174 | lxc_put_nsfds(handler); |
175 | return false; | |
176 | } | |
9f30a190 MM |
177 | } |
178 | ||
62d05d9b | 179 | return true; |
9f30a190 MM |
180 | } |
181 | ||
ec1dc633 | 182 | static inline bool match_stdfds(int fd) |
80090207 | 183 | { |
ec1dc633 | 184 | return (fd == STDIN_FILENO || fd == STDOUT_FILENO || fd == STDERR_FILENO); |
80090207 CLG |
185 | } |
186 | ||
4c03724e | 187 | #ifdef HAVE_DLOG |
188 | static bool match_dlog_fds(struct dirent *direntp) | |
189 | { | |
190 | char path[PATH_MAX] = {0}; | |
191 | char link[PATH_MAX] = {0}; | |
192 | ssize_t linklen; | |
193 | int ret; | |
194 | ||
fa60cd7b CB |
195 | ret = strnprintf(path, sizeof(path), "/proc/self/fd/%s", direntp->d_name); |
196 | if (ret < 0) | |
818a57fc | 197 | return log_error(false, "Failed to create file descriptor name"); |
4c03724e | 198 | |
199 | linklen = readlink(path, link, PATH_MAX); | |
818a57fc CB |
200 | if (linklen < 0) |
201 | return log_error(false, "Failed to read link path - \"%s\"", path); | |
202 | else if (linklen >= PATH_MAX) | |
203 | return log_error(false, "The name of link path is too long - \"%s\"", path); | |
204 | ||
d0269705 CB |
205 | if (strequal(link, "/dev/log_main") || |
206 | strequal(link, "/dev/log_system") || | |
207 | strequal(link, "/dev/log_radio")) | |
4c03724e | 208 | return true; |
209 | ||
210 | return false; | |
211 | } | |
212 | #endif | |
213 | ||
47a46cf1 CB |
214 | int lxc_check_inherited(struct lxc_conf *conf, bool closeall, |
215 | int *fds_to_ignore, size_t len_fds) | |
80090207 | 216 | { |
80090207 | 217 | int fd, fddir; |
47a46cf1 | 218 | size_t i; |
80090207 | 219 | DIR *dir; |
a5a70219 | 220 | struct dirent *direntp; |
80090207 | 221 | |
d2cf4c37 SH |
222 | if (conf && conf->close_all_fds) |
223 | closeall = true; | |
224 | ||
4e2d6b9a CB |
225 | /* |
226 | * Disable syslog at this point to avoid the above logging | |
227 | * function to open a new fd and make the check_inherited function | |
228 | * enter an infinite loop. | |
229 | */ | |
230 | lxc_log_syslog_disable(); | |
231 | ||
b119f362 | 232 | restart: |
80090207 | 233 | dir = opendir("/proc/self/fd"); |
818a57fc CB |
234 | if (!dir) |
235 | return log_warn(-1, "Failed to open directory"); | |
80090207 CLG |
236 | |
237 | fddir = dirfd(dir); | |
238 | ||
74f96976 | 239 | while ((direntp = readdir(dir))) { |
a5a70219 | 240 | int ret; |
d39b10eb CB |
241 | struct lxc_list *cur; |
242 | bool matched = false; | |
243 | ||
d0269705 | 244 | if (strequal(direntp->d_name, ".")) |
80090207 CLG |
245 | continue; |
246 | ||
d0269705 | 247 | if (strequal(direntp->d_name, "..")) |
80090207 CLG |
248 | continue; |
249 | ||
a5a70219 CB |
250 | ret = lxc_safe_int(direntp->d_name, &fd); |
251 | if (ret < 0) { | |
252 | INFO("Could not parse file descriptor for \"%s\"", direntp->d_name); | |
d4cff0d2 CB |
253 | continue; |
254 | } | |
80090207 | 255 | |
47a46cf1 CB |
256 | for (i = 0; i < len_fds; i++) |
257 | if (fds_to_ignore[i] == fd) | |
258 | break; | |
259 | ||
260 | if (fd == fddir || fd == lxc_log_fd || | |
261 | (i < len_fds && fd == fds_to_ignore[i])) | |
80090207 CLG |
262 | continue; |
263 | ||
d39b10eb | 264 | /* Keep state clients that wait on reboots. */ |
3ee9e4fb CB |
265 | if (conf) { |
266 | lxc_list_for_each(cur, &conf->state_clients) { | |
267 | struct lxc_state_client *client = cur->elem; | |
d39b10eb | 268 | |
3ee9e4fb CB |
269 | if (client->clientfd != fd) |
270 | continue; | |
d39b10eb | 271 | |
3ee9e4fb CB |
272 | matched = true; |
273 | break; | |
274 | } | |
d39b10eb CB |
275 | } |
276 | ||
277 | if (matched) | |
278 | continue; | |
279 | ||
858377e4 SH |
280 | if (current_config && fd == current_config->logfd) |
281 | continue; | |
282 | ||
ec1dc633 | 283 | if (match_stdfds(fd)) |
80090207 | 284 | continue; |
80090207 | 285 | |
1d5e5f26 | 286 | #ifdef HAVE_DLOG |
287 | if (match_dlog_fds(direntp)) | |
288 | continue; | |
289 | ||
290 | #endif | |
d2cf4c37 | 291 | if (closeall) { |
4e2d6b9a CB |
292 | if (close(fd)) |
293 | SYSINFO("Closed inherited fd %d", fd); | |
294 | else | |
295 | INFO("Closed inherited fd %d", fd); | |
b119f362 | 296 | closedir(dir); |
b119f362 SH |
297 | goto restart; |
298 | } | |
28d9e29e | 299 | WARN("Inherited fd %d", fd); |
80090207 | 300 | } |
4e2d6b9a | 301 | closedir(dir); |
80090207 | 302 | |
4e2d6b9a CB |
303 | /* |
304 | * Only enable syslog at this point to avoid the above logging | |
305 | * function to open a new fd and make the check_inherited function | |
306 | * enter an infinite loop. | |
64c57ea1 | 307 | */ |
4e2d6b9a | 308 | lxc_log_syslog_enable(); |
64c57ea1 | 309 | |
92c7f629 | 310 | return 0; |
80090207 CLG |
311 | } |
312 | ||
83ee7875 | 313 | static int setup_signal_fd(sigset_t *oldmask) |
b0a33c1e | 314 | { |
7288dfb6 | 315 | int ret; |
b0a33c1e | 316 | sigset_t mask; |
7288dfb6 | 317 | const int signals[] = {SIGBUS, SIGILL, SIGSEGV, SIGWINCH}; |
b0a33c1e | 318 | |
408da065 | 319 | /* Block everything except serious error signals. */ |
df0795b1 CB |
320 | ret = sigfillset(&mask); |
321 | if (ret < 0) | |
322 | return -EBADF; | |
323 | ||
9b10bef5 | 324 | for (int sig = 0; sig < (sizeof(signals) / sizeof(signals[0])); sig++) { |
df0795b1 CB |
325 | ret = sigdelset(&mask, signals[sig]); |
326 | if (ret < 0) | |
327 | return -EBADF; | |
b0a33c1e | 328 | } |
329 | ||
b467714b | 330 | ret = pthread_sigmask(SIG_BLOCK, &mask, oldmask); |
818a57fc CB |
331 | if (ret < 0) |
332 | return log_error_errno(-EBADF, errno, | |
333 | "Failed to set signal mask"); | |
b0a33c1e | 334 | |
df0795b1 | 335 | ret = signalfd(-1, &mask, SFD_CLOEXEC); |
818a57fc CB |
336 | if (ret < 0) |
337 | return log_error_errno(-EBADF, | |
338 | errno, "Failed to create signal file descriptor"); | |
b0a33c1e | 339 | |
df0795b1 | 340 | TRACE("Created signal file descriptor %d", ret); |
1ac470c0 | 341 | |
df0795b1 | 342 | return ret; |
b0a33c1e | 343 | } |
344 | ||
84c92abd | 345 | static int signal_handler(int fd, uint32_t events, void *data, |
f3787121 | 346 | struct lxc_epoll_descr *descr) |
b0a33c1e | 347 | { |
15cd25fd | 348 | int ret; |
3c319edb CB |
349 | siginfo_t info; |
350 | struct signalfd_siginfo siginfo; | |
351 | struct lxc_handler *hdlr = data; | |
15cd25fd | 352 | |
489f39be | 353 | ret = lxc_read_nointr(fd, &siginfo, sizeof(siginfo)); |
818a57fc CB |
354 | if (ret < 0) |
355 | return log_error(LXC_MAINLOOP_ERROR, "Failed to read signal info from signal file descriptor %d", fd); | |
15cd25fd | 356 | |
818a57fc | 357 | if (ret != sizeof(siginfo)) |
f7a97743 | 358 | return log_error(LXC_MAINLOOP_ERROR, "Unexpected size for struct signalfd_siginfo"); |
15cd25fd | 359 | |
408da065 | 360 | /* Check whether init is running. */ |
80507ee8 | 361 | info.si_pid = 0; |
3c319edb CB |
362 | ret = waitid(P_PID, hdlr->pid, &info, WEXITED | WNOWAIT | WNOHANG); |
363 | if (ret == 0 && info.si_pid == hdlr->pid) | |
364 | hdlr->init_died = true; | |
80507ee8 | 365 | |
cd5177e9 TA |
366 | /* Try to figure out a reasonable exit status to report. */ |
367 | if (hdlr->init_died) { | |
368 | switch (info.si_code) { | |
369 | case CLD_EXITED: | |
370 | hdlr->exit_status = info.si_status << 8; | |
371 | break; | |
372 | case CLD_KILLED: | |
373 | case CLD_DUMPED: | |
374 | case CLD_STOPPED: | |
375 | hdlr->exit_status = info.si_status << 8 | 0x7f; | |
376 | break; | |
377 | case CLD_CONTINUED: | |
378 | /* Huh? The waitid() told us it's dead *and* continued? */ | |
379 | WARN("Init %d dead and continued?", hdlr->pid); | |
380 | hdlr->exit_status = 1; | |
381 | break; | |
382 | default: | |
20993a97 | 383 | ERROR("Unknown si_code: %d", info.si_code); |
55e1311e | 384 | hdlr->exit_status = 1; |
cd5177e9 TA |
385 | } |
386 | } | |
387 | ||
d4b5d7a8 | 388 | if (siginfo.ssi_signo == SIGHUP) { |
33942046 CB |
389 | if (hdlr->pidfd >= 0) |
390 | lxc_raw_pidfd_send_signal(hdlr->pidfd, SIGTERM, NULL, 0); | |
d9bb2fba CB |
391 | else |
392 | kill(hdlr->pid, SIGTERM); | |
d4b5d7a8 | 393 | INFO("Killing %d since terminal hung up", hdlr->pid); |
9b10bef5 CB |
394 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE |
395 | : LXC_MAINLOOP_CONTINUE; | |
d4b5d7a8 TA |
396 | } |
397 | ||
9cb94384 | 398 | if (siginfo.ssi_signo != SIGCHLD) { |
33942046 CB |
399 | if (hdlr->pidfd >= 0) |
400 | lxc_raw_pidfd_send_signal(hdlr->pidfd, | |
401 | siginfo.ssi_signo, NULL, 0); | |
d9bb2fba CB |
402 | else |
403 | kill(hdlr->pid, siginfo.ssi_signo); | |
9cb94384 | 404 | INFO("Forwarded signal %d to pid %d", siginfo.ssi_signo, hdlr->pid); |
9b10bef5 CB |
405 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE |
406 | : LXC_MAINLOOP_CONTINUE; | |
9cb94384 TA |
407 | } |
408 | ||
3a9e949f TA |
409 | /* More robustness, protect ourself from a SIGCHLD sent |
410 | * by a process different from the container init. | |
411 | */ | |
412 | if (siginfo.ssi_pid != hdlr->pid) { | |
6e94162a CB |
413 | NOTICE("Received %d from pid %d instead of container init %d", |
414 | siginfo.ssi_signo, siginfo.ssi_pid, hdlr->pid); | |
9b10bef5 CB |
415 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE |
416 | : LXC_MAINLOOP_CONTINUE; | |
3a9e949f TA |
417 | } |
418 | ||
408da065 | 419 | if (siginfo.ssi_code == CLD_STOPPED) { |
6e94162a | 420 | INFO("Container init process was stopped"); |
9b10bef5 CB |
421 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE |
422 | : LXC_MAINLOOP_CONTINUE; | |
423 | } | |
424 | ||
425 | if (siginfo.ssi_code == CLD_CONTINUED) { | |
6e94162a | 426 | INFO("Container init process was continued"); |
9b10bef5 CB |
427 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE |
428 | : LXC_MAINLOOP_CONTINUE; | |
15cd25fd | 429 | } |
1ac470c0 | 430 | |
818a57fc | 431 | return log_debug(LXC_MAINLOOP_CLOSE, "Container init process %d exited", hdlr->pid); |
b0a33c1e | 432 | } |
433 | ||
974a8aba CB |
434 | int lxc_serve_state_clients(const char *name, struct lxc_handler *handler, |
435 | lxc_state_t state) | |
66aeffc7 | 436 | { |
9de31d5a | 437 | size_t retlen; |
dbc9832d CB |
438 | ssize_t ret; |
439 | struct lxc_list *cur, *next; | |
dbc9832d CB |
440 | struct lxc_msg msg = {.type = lxc_msg_state, .value = state}; |
441 | ||
974a8aba CB |
442 | if (state == THAWED) |
443 | handler->state = RUNNING; | |
444 | else | |
445 | handler->state = state; | |
446 | ||
9dfa4041 | 447 | TRACE("Set container state to %s", lxc_state2str(state)); |
dbc9832d | 448 | |
818a57fc CB |
449 | if (lxc_list_empty(&handler->conf->state_clients)) |
450 | return log_trace(0, "No state clients registered"); | |
dbc9832d | 451 | |
9de31d5a CB |
452 | retlen = strlcpy(msg.name, name, sizeof(msg.name)); |
453 | if (retlen >= sizeof(msg.name)) | |
454 | return -E2BIG; | |
dbc9832d | 455 | |
d39b10eb | 456 | lxc_list_for_each_safe(cur, &handler->conf->state_clients, next) { |
d85617bc | 457 | struct lxc_state_client *client = cur->elem; |
dbc9832d | 458 | |
d39b10eb | 459 | if (client->states[state] == 0) { |
9dfa4041 | 460 | TRACE("State %s not registered for state client %d", |
dbc9832d CB |
461 | lxc_state2str(state), client->clientfd); |
462 | continue; | |
463 | } | |
464 | ||
9dfa4041 | 465 | TRACE("Sending state %s to state client %d", |
dbc9832d CB |
466 | lxc_state2str(state), client->clientfd); |
467 | ||
7fbb15ec CB |
468 | ret = lxc_send_nointr(client->clientfd, &msg, sizeof(msg), MSG_NOSIGNAL); |
469 | if (ret <= 0) | |
6d1400b5 | 470 | SYSERROR("Failed to send message to client"); |
dbc9832d CB |
471 | |
472 | /* kick client from list */ | |
dbc9832d | 473 | lxc_list_del(cur); |
300d1cb4 | 474 | close(client->clientfd); |
dbc9832d CB |
475 | free(cur->elem); |
476 | free(cur); | |
477 | } | |
dbc9832d | 478 | |
5e5576a4 CB |
479 | return 0; |
480 | } | |
481 | ||
482 | static int lxc_serve_state_socket_pair(const char *name, | |
483 | struct lxc_handler *handler, | |
484 | lxc_state_t state) | |
485 | { | |
486 | ssize_t ret; | |
487 | ||
bb955810 | 488 | if (!handler->daemonize || |
5e5576a4 CB |
489 | handler->state_socket_pair[1] < 0 || |
490 | state == STARTING) | |
491 | return 0; | |
492 | ||
493 | /* Close read end of the socket pair. */ | |
a783a414 | 494 | close_prot_errno_disarm(handler->state_socket_pair[0]); |
5e5576a4 | 495 | |
ee8377bd | 496 | again: |
5e5576a4 CB |
497 | ret = lxc_abstract_unix_send_credential(handler->state_socket_pair[1], |
498 | &(int){state}, sizeof(int)); | |
9044b79e | 499 | if (ret < 0) { |
500 | SYSERROR("Failed to send state to %d", handler->state_socket_pair[1]); | |
501 | ||
ee8377bd CB |
502 | if (errno == EINTR) |
503 | goto again; | |
9044b79e | 504 | |
505 | return -1; | |
506 | } | |
507 | ||
818a57fc CB |
508 | if (ret != sizeof(int)) |
509 | return log_error(-1, "Message too long : %d", handler->state_socket_pair[1]); | |
5e5576a4 CB |
510 | |
511 | TRACE("Sent container state \"%s\" to %d", lxc_state2str(state), | |
512 | handler->state_socket_pair[1]); | |
513 | ||
514 | /* Close write end of the socket pair. */ | |
818a57fc | 515 | close_prot_errno_disarm(handler->state_socket_pair[1]); |
5e5576a4 CB |
516 | |
517 | return 0; | |
518 | } | |
519 | ||
520 | int lxc_set_state(const char *name, struct lxc_handler *handler, | |
521 | lxc_state_t state) | |
522 | { | |
523 | int ret; | |
524 | ||
525 | ret = lxc_serve_state_socket_pair(name, handler, state); | |
818a57fc CB |
526 | if (ret < 0) |
527 | return log_error(-1, "Failed to synchronize via anonymous pair of unix sockets"); | |
5e5576a4 CB |
528 | |
529 | ret = lxc_serve_state_clients(name, handler, state); | |
530 | if (ret < 0) | |
531 | return -1; | |
532 | ||
dbc9832d CB |
533 | /* This function will try to connect to the legacy lxc-monitord state |
534 | * server and only exists for backwards compatibility. | |
535 | */ | |
9123e471 | 536 | lxc_monitor_send_state(name, state, handler->lxcpath); |
dbc9832d | 537 | |
66aeffc7 DL |
538 | return 0; |
539 | } | |
540 | ||
735f2c6e | 541 | int lxc_poll(const char *name, struct lxc_handler *handler) |
b0a33c1e | 542 | { |
3c319edb | 543 | int ret; |
86530b0a | 544 | bool has_console = true; |
3c319edb | 545 | struct lxc_epoll_descr descr, descr_console; |
b0a33c1e | 546 | |
03760215 | 547 | if (handler->conf->console.path && |
d0269705 | 548 | strequal(handler->conf->console.path, "none")) |
86530b0a L |
549 | has_console = false; |
550 | ||
3c319edb CB |
551 | ret = lxc_mainloop_open(&descr); |
552 | if (ret < 0) { | |
553 | ERROR("Failed to create mainloop"); | |
50c8bf05 | 554 | goto out_sigfd; |
b0a33c1e | 555 | } |
556 | ||
30a33fbd CB |
557 | if (has_console) { |
558 | ret = lxc_mainloop_open(&descr_console); | |
559 | if (ret < 0) { | |
560 | ERROR("Failed to create console mainloop"); | |
561 | goto out_mainloop; | |
562 | } | |
3c319edb CB |
563 | } |
564 | ||
1cc8bd4b | 565 | ret = lxc_mainloop_add_handler(&descr, handler->sigfd, signal_handler, handler); |
3c319edb | 566 | if (ret < 0) { |
1cc8bd4b | 567 | ERROR("Failed to add signal handler for %d to mainloop", handler->sigfd); |
3c319edb | 568 | goto out_mainloop_console; |
b0a33c1e | 569 | } |
570 | ||
2ac0f627 CB |
571 | ret = lxc_seccomp_setup_proxy(&handler->conf->seccomp, &descr, handler); |
572 | if (ret < 0) { | |
573 | ERROR("Failed to setup seccomp proxy"); | |
c3e3c21a | 574 | goto out_mainloop_console; |
2ac0f627 | 575 | } |
cdb2a47f | 576 | |
30a33fbd | 577 | if (has_console) { |
dcad02f8 | 578 | struct lxc_terminal *console = &handler->conf->console; |
63376d7d | 579 | |
093bce5f | 580 | ret = lxc_terminal_mainloop_add(&descr, console); |
30a33fbd CB |
581 | if (ret < 0) { |
582 | ERROR("Failed to add console handlers to mainloop"); | |
583 | goto out_mainloop_console; | |
584 | } | |
585 | ||
093bce5f | 586 | ret = lxc_terminal_mainloop_add(&descr_console, console); |
30a33fbd CB |
587 | if (ret < 0) { |
588 | ERROR("Failed to add console handlers to console mainloop"); | |
589 | goto out_mainloop_console; | |
590 | } | |
a54585ad L |
591 | |
592 | handler->conf->console.descr = &descr; | |
563f2f2c DL |
593 | } |
594 | ||
3c319edb CB |
595 | ret = lxc_cmd_mainloop_add(name, &descr, handler); |
596 | if (ret < 0) { | |
597 | ERROR("Failed to add command handler to mainloop"); | |
598 | goto out_mainloop_console; | |
599 | } | |
b0a33c1e | 600 | |
3c319edb | 601 | TRACE("Mainloop is ready"); |
b0a33c1e | 602 | |
3c319edb | 603 | ret = lxc_mainloop(&descr, -1); |
a783a414 | 604 | close_prot_errno_disarm(descr.epfd); |
1cc8bd4b | 605 | if (ret < 0 || !handler->init_died) |
42b09f94 | 606 | goto out_mainloop_console; |
1cc8bd4b | 607 | |
30a33fbd CB |
608 | if (has_console) |
609 | ret = lxc_mainloop(&descr_console, 0); | |
610 | ||
3c319edb | 611 | out_mainloop_console: |
30a33fbd CB |
612 | if (has_console) { |
613 | lxc_mainloop_close(&descr_console); | |
614 | TRACE("Closed console mainloop"); | |
615 | } | |
3c319edb | 616 | |
42b09f94 CB |
617 | out_mainloop: |
618 | lxc_mainloop_close(&descr); | |
619 | TRACE("Closed mainloop"); | |
620 | ||
b0a33c1e | 621 | out_sigfd: |
1cc8bd4b | 622 | TRACE("Closed signal file descriptor %d", handler->sigfd); |
a783a414 | 623 | close_prot_errno_disarm(handler->sigfd); |
408da065 | 624 | |
1cc8bd4b | 625 | return ret; |
b0a33c1e | 626 | } |
627 | ||
a42abcce | 628 | void lxc_put_handler(struct lxc_handler *handler) |
f2e07cb6 | 629 | { |
a783a414 CB |
630 | close_prot_errno_disarm(handler->pidfd); |
631 | close_prot_errno_disarm(handler->sigfd); | |
1cc8bd4b | 632 | lxc_put_nsfds(handler); |
80308d07 | 633 | if (handler->conf && handler->conf->reboot == REBOOT_NONE) |
a783a414 CB |
634 | close_prot_errno_disarm(handler->conf->maincmd_fd); |
635 | close_prot_errno_disarm(handler->monitor_status_fd); | |
636 | close_prot_errno_disarm(handler->state_socket_pair[0]); | |
637 | close_prot_errno_disarm(handler->state_socket_pair[1]); | |
638 | cgroup_exit(handler->cgroup_ops); | |
a42abcce CB |
639 | if (handler->conf && handler->conf->reboot == REBOOT_NONE) |
640 | free_disarm(handler); | |
641 | else | |
642 | handler->conf = NULL; | |
f2e07cb6 CB |
643 | } |
644 | ||
a42abcce CB |
645 | struct lxc_handler *lxc_init_handler(struct lxc_handler *old, |
646 | const char *name, struct lxc_conf *conf, | |
5e5576a4 | 647 | const char *lxcpath, bool daemonize) |
59eb99ba | 648 | { |
85c279bb | 649 | int nr_keep_fds = 0; |
818a57fc | 650 | int ret; |
3a0f472d DL |
651 | struct lxc_handler *handler; |
652 | ||
a42abcce CB |
653 | if (!old) |
654 | handler = zalloc(sizeof(*handler)); | |
655 | else | |
656 | handler = old; | |
fdecbc9c | 657 | if (!handler) |
3a0f472d | 658 | return NULL; |
9044b79e | 659 | |
fdecbc9c CB |
660 | /* Note that am_guest_unpriv() checks the effective uid. We |
661 | * probably don't care if we are real root only if we are running | |
662 | * as root so this should be fine. | |
790255cf | 663 | */ |
e0010464 | 664 | handler->am_root = !am_guest_unpriv(); |
fae349da | 665 | handler->conf = conf; |
9123e471 | 666 | handler->lxcpath = lxcpath; |
818a57fc CB |
667 | handler->init_died = false; |
668 | handler->data_sock[0] = -EBADF; | |
669 | handler->data_sock[1] = -EBADF; | |
4d8bdfa0 | 670 | handler->monitor_status_fd = -EBADF; |
33942046 | 671 | handler->pidfd = -EBADF; |
1cc8bd4b | 672 | handler->sigfd = -EBADF; |
818a57fc CB |
673 | handler->state_socket_pair[0] = -EBADF; |
674 | handler->state_socket_pair[1] = -EBADF; | |
80308d07 | 675 | if (handler->conf->reboot == REBOOT_NONE) |
d39b10eb | 676 | lxc_list_init(&handler->conf->state_clients); |
fae349da | 677 | |
8db6be1b CB |
678 | for (lxc_namespace_t idx = 0; idx < LXC_NS_MAX; idx++) { |
679 | handler->nsfd[idx] = -EBADF; | |
680 | ||
681 | if (handler->conf->reboot == REBOOT_NONE) | |
682 | continue; | |
683 | ||
684 | handler->nsfd_paths[idx][0] = '\0'; | |
685 | handler->hook_argv[idx] = NULL; | |
686 | ||
687 | if (handler->hook_argc != 0) | |
688 | handler->hook_argc = 0; | |
689 | } | |
b6b2b194 | 690 | |
f0ecc19d | 691 | handler->name = name; |
c581d2a6 CB |
692 | if (daemonize) |
693 | handler->transient_pid = lxc_raw_getpid(); | |
694 | else | |
695 | handler->transient_pid = -1; | |
3bdf52d7 | 696 | |
80308d07 | 697 | if (daemonize && handler->conf->reboot == REBOOT_NONE) { |
5e5576a4 | 698 | /* Create socketpair() to synchronize on daemonized startup. |
fdecbc9c CB |
699 | * When the container reboots we don't need to synchronize |
700 | * again currently so don't open another socketpair(). | |
5e5576a4 CB |
701 | */ |
702 | ret = socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0, | |
703 | handler->state_socket_pair); | |
704 | if (ret < 0) { | |
705 | ERROR("Failed to create anonymous pair of unix sockets"); | |
706 | goto on_error; | |
707 | } | |
9044b79e | 708 | |
5e5576a4 CB |
709 | TRACE("Created anonymous pair {%d,%d} of unix sockets", |
710 | handler->state_socket_pair[0], | |
711 | handler->state_socket_pair[1]); | |
85c279bb CB |
712 | handler->keep_fds[nr_keep_fds++] = handler->state_socket_pair[0]; |
713 | handler->keep_fds[nr_keep_fds++] = handler->state_socket_pair[1]; | |
5e5576a4 CB |
714 | } |
715 | ||
80308d07 | 716 | if (handler->conf->reboot == REBOOT_NONE) { |
d39b10eb CB |
717 | handler->conf->maincmd_fd = lxc_cmd_init(name, lxcpath, "command"); |
718 | if (handler->conf->maincmd_fd < 0) { | |
719 | ERROR("Failed to set up command socket"); | |
720 | goto on_error; | |
721 | } | |
85c279bb | 722 | handler->keep_fds[nr_keep_fds++] = handler->conf->maincmd_fd; |
aa460476 | 723 | } |
9044b79e | 724 | |
9dfa4041 | 725 | TRACE("Unix domain socket %d for command server is ready", |
aa460476 CB |
726 | handler->conf->maincmd_fd); |
727 | ||
728 | return handler; | |
729 | ||
f2e07cb6 | 730 | on_error: |
a42abcce | 731 | lxc_put_handler(handler); |
aa460476 CB |
732 | |
733 | return NULL; | |
734 | } | |
735 | ||
736 | int lxc_init(const char *name, struct lxc_handler *handler) | |
737 | { | |
f62cf1d4 | 738 | __do_close int status_fd = -EBADF; |
a2c09be0 | 739 | int ret; |
4a03ded4 | 740 | const char *loglevel; |
aa460476 CB |
741 | struct lxc_conf *conf = handler->conf; |
742 | ||
434c8e15 | 743 | handler->monitor_pid = lxc_raw_getpid(); |
4d8bdfa0 | 744 | status_fd = open("/proc/self/status", O_RDONLY | O_CLOEXEC); |
59b1b67e CB |
745 | if (status_fd < 0) |
746 | return log_error_errno(-1, errno, "Failed to open monitor status fd"); | |
434c8e15 | 747 | |
4eb19ac0 | 748 | handler->lsm_ops = lsm_init_static(); |
2170b263 | 749 | TRACE("Initialized LSM"); |
d2e30e99 | 750 | |
408da065 | 751 | /* Begin by setting the state to STARTING. */ |
2170b263 | 752 | ret = lxc_set_state(name, handler, STARTING); |
59b1b67e CB |
753 | if (ret < 0) |
754 | return log_error(-1, "Failed to set state to \"%s\"", lxc_state2str(STARTING)); | |
46800e77 | 755 | TRACE("Set container state to \"STARTING\""); |
0ad19a3f | 756 | |
408da065 | 757 | /* Start of environment variable setup for hooks. */ |
24c2a48d CB |
758 | ret = setenv("LXC_NAME", name, 1); |
759 | if (ret < 0) | |
760 | SYSERROR("Failed to set environment variable: LXC_NAME=%s", name); | |
408da065 | 761 | |
2170b263 CB |
762 | if (conf->rcfile) { |
763 | ret = setenv("LXC_CONFIG_FILE", conf->rcfile, 1); | |
764 | if (ret < 0) | |
818a57fc | 765 | SYSERROR("Failed to set environment variable: LXC_CONFIG_FILE=%s", conf->rcfile); |
2170b263 | 766 | } |
408da065 | 767 | |
2170b263 CB |
768 | if (conf->rootfs.mount) { |
769 | ret = setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1); | |
770 | if (ret < 0) | |
818a57fc | 771 | SYSERROR("Failed to set environment variable: LXC_ROOTFS_MOUNT=%s", conf->rootfs.mount); |
2170b263 | 772 | } |
408da065 | 773 | |
2170b263 CB |
774 | if (conf->rootfs.path) { |
775 | ret = setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1); | |
776 | if (ret < 0) | |
818a57fc | 777 | SYSERROR("Failed to set environment variable: LXC_ROOTFS_PATH=%s", conf->rootfs.path); |
2170b263 | 778 | } |
408da065 | 779 | |
2170b263 CB |
780 | if (conf->console.path) { |
781 | ret = setenv("LXC_CONSOLE", conf->console.path, 1); | |
782 | if (ret < 0) | |
818a57fc | 783 | SYSERROR("Failed to set environment variable: LXC_CONSOLE=%s", conf->console.path); |
2170b263 | 784 | } |
408da065 | 785 | |
2170b263 CB |
786 | if (conf->console.log_path) { |
787 | ret = setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1); | |
788 | if (ret < 0) | |
818a57fc | 789 | SYSERROR("Failed to set environment variable: LXC_CONSOLE_LOGPATH=%s", conf->console.log_path); |
2170b263 | 790 | } |
408da065 | 791 | |
2170b263 CB |
792 | if (cgns_supported()) { |
793 | ret = setenv("LXC_CGNS_AWARE", "1", 1); | |
794 | if (ret < 0) | |
818a57fc | 795 | SYSERROR("Failed to set environment variable LXC_CGNS_AWARE=1"); |
2170b263 | 796 | } |
b8f88d9b | 797 | |
4a03ded4 | 798 | loglevel = lxc_log_priority_to_string(lxc_log_get_level()); |
2170b263 CB |
799 | ret = setenv("LXC_LOG_LEVEL", loglevel, 1); |
800 | if (ret < 0) | |
818a57fc | 801 | SYSERROR("Set environment variable LXC_LOG_LEVEL=%s", 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 | 813 | ret = run_lxc_hooks(name, "pre-start", conf, NULL); |
65a2dc8b CB |
814 | if (ret < 0) |
815 | return log_error(-1, "Failed to run lxc.hook.pre-start for container \"%s\"", name); | |
2170b263 | 816 | TRACE("Ran pre-start hooks"); |
26ddeedd | 817 | |
408da065 CB |
818 | /* The signal fd has to be created before forking otherwise if the child |
819 | * process exits before we setup the signal fd, the event will be lost | |
820 | * and the command will be stuck. | |
821 | */ | |
83ee7875 | 822 | handler->sigfd = setup_signal_fd(&handler->oldmask); |
65a2dc8b CB |
823 | if (handler->sigfd < 0) |
824 | return log_error(-1, "Failed to setup SIGCHLD fd handler."); | |
2170b263 | 825 | TRACE("Set up signal fd"); |
b5159817 | 826 | |
408da065 | 827 | /* Do this after setting up signals since it might unblock SIGWINCH. */ |
564e31c4 | 828 | ret = lxc_terminal_setup(conf); |
4d1ffb0a CB |
829 | if (ret < 0) { |
830 | ERROR("Failed to create console"); | |
b5159817 | 831 | goto out_restore_sigmask; |
b0a33c1e | 832 | } |
4d1ffb0a | 833 | TRACE("Created console"); |
b0a33c1e | 834 | |
5a087e05 | 835 | handler->cgroup_ops = cgroup_init(handler->conf); |
2202afc9 CB |
836 | if (!handler->cgroup_ops) { |
837 | ERROR("Failed to initialize cgroup driver"); | |
5b74eb3c | 838 | goto out_delete_terminal; |
2202afc9 CB |
839 | } |
840 | TRACE("Initialized cgroup driver"); | |
841 | ||
2e1361a6 CB |
842 | ret = lxc_read_seccomp_config(conf); |
843 | if (ret < 0) | |
844 | return log_error(-1, "Failed loading seccomp policy"); | |
845 | TRACE("Read seccomp policy"); | |
846 | ||
af04d847 | 847 | ret = handler->lsm_ops->prepare(handler->lsm_ops, conf, handler->lxcpath); |
1800f924 WB |
848 | if (ret < 0) { |
849 | ERROR("Failed to initialize LSM"); | |
c581d2a6 | 850 | goto out_delete_terminal; |
1800f924 WB |
851 | } |
852 | TRACE("Initialized LSM"); | |
853 | ||
2170b263 | 854 | INFO("Container \"%s\" is initialized", name); |
4d8bdfa0 | 855 | handler->monitor_status_fd = move_fd(status_fd); |
aa460476 | 856 | return 0; |
59eb99ba | 857 | |
5b74eb3c CB |
858 | out_delete_terminal: |
859 | lxc_terminal_delete(&handler->conf->console); | |
ea918412 | 860 | |
b5159817 | 861 | out_restore_sigmask: |
4c496daa | 862 | (void)pthread_sigmask(SIG_SETMASK, &handler->oldmask, NULL); |
ea918412 | 863 | |
aa460476 | 864 | return -1; |
59eb99ba DL |
865 | } |
866 | ||
8db6be1b CB |
867 | void lxc_expose_namespace_environment(const struct lxc_handler *handler) |
868 | { | |
869 | for (lxc_namespace_t i = 0; i < LXC_NS_MAX; i++) { | |
870 | int ret; | |
871 | const char *fd_path; | |
872 | ||
873 | if (handler->nsfd[i] < 0) | |
874 | continue; | |
875 | ||
876 | fd_path = handler->nsfd_paths[i] + strcspn(handler->nsfd_paths[i], "/"); | |
877 | ret = setenv(ns_info[i].env_name, fd_path, 1); | |
878 | if (ret < 0) | |
879 | SYSERROR("Failed to set environment variable %s=%s", | |
880 | ns_info[i].env_name, fd_path); | |
881 | else | |
882 | TRACE("Set environment variable %s=%s", | |
883 | ns_info[i].env_name, fd_path); | |
884 | } | |
885 | } | |
886 | ||
fd5be714 | 887 | void lxc_end(struct lxc_handler *handler) |
59eb99ba | 888 | { |
818a57fc | 889 | int ret; |
dbc9832d | 890 | struct lxc_list *cur, *next; |
0c5859ff | 891 | const char *name = handler->name; |
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 | |
8db6be1b CB |
899 | /* Passing information to hooks via environment variables. */ |
900 | if (handler->conf->hooks_version > 0) | |
901 | lxc_expose_namespace_environment(handler); | |
c154af98 | 902 | |
80308d07 | 903 | if (handler->conf->reboot > REBOOT_NONE) { |
c097467f CB |
904 | ret = setenv("LXC_TARGET", "reboot", 1); |
905 | if (ret < 0) | |
818a57fc | 906 | SYSERROR("Failed to set environment variable: LXC_TARGET=reboot"); |
c097467f | 907 | } |
408da065 | 908 | |
80308d07 | 909 | if (handler->conf->reboot == REBOOT_NONE) { |
c097467f CB |
910 | ret = setenv("LXC_TARGET", "stop", 1); |
911 | if (ret < 0) | |
818a57fc | 912 | SYSERROR("Failed to set environment variable: LXC_TARGET=stop"); |
c097467f | 913 | } |
c154af98 | 914 | |
18b3b9c1 | 915 | if (handler->conf->hooks_version == 0) |
8db6be1b | 916 | ret = run_lxc_hooks(name, "stop", handler->conf, handler->hook_argv); |
18b3b9c1 | 917 | else |
c097467f CB |
918 | ret = run_lxc_hooks(name, "stop", handler->conf, NULL); |
919 | if (ret < 0) | |
920 | ERROR("Failed to run \"lxc.hook.stop\" hook"); | |
c154af98 | 921 | |
af04d847 | 922 | handler->lsm_ops->cleanup(handler->lsm_ops, handler->conf, handler->lxcpath); |
1800f924 | 923 | |
e2aed383 WB |
924 | if (cgroup_ops) { |
925 | cgroup_ops->payload_destroy(cgroup_ops, handler); | |
926 | cgroup_ops->monitor_destroy(cgroup_ops, handler); | |
927 | } | |
4288b79f | 928 | |
79ff643d CB |
929 | put_lxc_rootfs(&handler->conf->rootfs, true); |
930 | ||
80308d07 | 931 | if (handler->conf->reboot == REBOOT_NONE) { |
f893d898 CB |
932 | /* For all new state clients simply close the command socket. |
933 | * This will inform all state clients that the container is | |
934 | * STOPPED and also prevents a race between a open()/close() on | |
935 | * the command socket causing a new process to get ECONNREFUSED | |
936 | * because we haven't yet closed the command socket. | |
937 | */ | |
b87ee312 | 938 | close_prot_errno_disarm(handler->conf->maincmd_fd); |
c3184275 CB |
939 | TRACE("Closed command socket"); |
940 | ||
941 | /* This function will try to connect to the legacy lxc-monitord | |
942 | * state server and only exists for backwards compatibility. | |
943 | */ | |
944 | lxc_monitor_send_state(name, STOPPED, handler->lxcpath); | |
945 | ||
946 | /* The command socket is closed so no one can acces the command | |
947 | * socket anymore so there's no need to lock it. | |
948 | */ | |
949 | handler->state = STOPPED; | |
950 | TRACE("Set container state to \"STOPPED\""); | |
f893d898 CB |
951 | } else { |
952 | lxc_set_state(name, handler, STOPPED); | |
695d0e56 | 953 | TRACE("Set container state to \"STOPPED\""); |
d39b10eb | 954 | } |
4288b79f | 955 | |
048493a3 CB |
956 | /* Avoid lingering namespace references. */ |
957 | lxc_put_nsfds(handler); | |
958 | ||
46800e77 CB |
959 | ret = run_lxc_hooks(name, "post-stop", handler->conf, NULL); |
960 | if (ret < 0) { | |
c097467f | 961 | ERROR("Failed to run lxc.hook.post-stop for container \"%s\"", name); |
80308d07 | 962 | if (handler->conf->reboot > REBOOT_NONE) { |
c097467f | 963 | WARN("Container will be stopped instead of rebooted"); |
80308d07 | 964 | handler->conf->reboot = REBOOT_NONE; |
c097467f CB |
965 | |
966 | ret = setenv("LXC_TARGET", "stop", 1); | |
967 | if (ret < 0) | |
818a57fc | 968 | WARN("Failed to set environment variable: LXC_TARGET=stop"); |
f3787121 CB |
969 | } |
970 | } | |
26ddeedd | 971 | |
408da065 | 972 | /* Reset mask set by setup_signal_fd. */ |
b467714b | 973 | ret = pthread_sigmask(SIG_SETMASK, &handler->oldmask, NULL); |
c097467f | 974 | if (ret < 0) |
a24c5678 | 975 | SYSWARN("Failed to restore signal mask"); |
8f64a3f6 | 976 | |
2aac071b | 977 | lxc_terminal_delete(&handler->conf->console); |
0e4be3cf | 978 | lxc_delete_tty(&handler->conf->ttys); |
f797f05e | 979 | close_prot_errno_disarm(handler->conf->devpts_fd); |
dbc9832d | 980 | |
dbc9832d CB |
981 | /* The command socket is now closed, no more state clients can register |
982 | * themselves from now on. So free the list of state clients. | |
983 | */ | |
d39b10eb CB |
984 | lxc_list_for_each_safe(cur, &handler->conf->state_clients, next) { |
985 | struct lxc_state_client *client = cur->elem; | |
986 | ||
987 | /* Keep state clients that want to be notified about reboots. */ | |
80308d07 CB |
988 | if ((handler->conf->reboot > REBOOT_NONE) && |
989 | (client->states[RUNNING] == 2)) | |
d39b10eb CB |
990 | continue; |
991 | ||
dbc9832d | 992 | /* close state client socket */ |
dbc9832d | 993 | lxc_list_del(cur); |
c097467f | 994 | close(client->clientfd); |
dbc9832d CB |
995 | free(cur->elem); |
996 | free(cur); | |
997 | } | |
998 | ||
80308d07 | 999 | if (handler->conf->ephemeral == 1 && handler->conf->reboot != REBOOT_REQ) |
28272964 | 1000 | lxc_destroy_container_on_signal(handler, name); |
2c5f2ede | 1001 | |
a42abcce | 1002 | lxc_put_handler(handler); |
59eb99ba DL |
1003 | } |
1004 | ||
0c5859ff | 1005 | void lxc_abort(struct lxc_handler *handler) |
59eb99ba | 1006 | { |
33942046 CB |
1007 | int ret = 0; |
1008 | int status; | |
73e608b2 | 1009 | |
0c5859ff | 1010 | lxc_set_state(handler->name, handler, ABORTING); |
0e4f9d51 | 1011 | |
c718fac1 | 1012 | if (handler->pidfd >= 0) { |
33942046 | 1013 | ret = lxc_raw_pidfd_send_signal(handler->pidfd, SIGKILL, NULL, 0); |
c718fac1 | 1014 | if (ret) |
11c7d349 CB |
1015 | SYSWARN("Failed to send SIGKILL via pidfd %d for process %d", |
1016 | handler->pidfd, handler->pid); | |
c718fac1 CB |
1017 | } |
1018 | ||
65a2dc8b | 1019 | if ((!ret || errno != ESRCH) && handler->pid > 0) |
11c7d349 CB |
1020 | if (kill(handler->pid, SIGKILL)) |
1021 | SYSWARN("Failed to send SIGKILL to %d", handler->pid); | |
0e4f9d51 | 1022 | |
33942046 CB |
1023 | do { |
1024 | ret = waitpid(-1, &status, 0); | |
1025 | } while (ret > 0); | |
59eb99ba DL |
1026 | } |
1027 | ||
ffe1e01a | 1028 | static int do_start(void *data) |
50e98013 | 1029 | { |
cdb2a47f | 1030 | struct lxc_handler *handler = data; |
f62cf1d4 | 1031 | __lxc_unused __do_close int data_sock0 = handler->data_sock[0], |
9ff57a59 | 1032 | data_sock1 = handler->data_sock[1]; |
d3103162 | 1033 | __do_close int devnull_fd = -EBADF, status_fd = -EBADF; |
8bf3abfb | 1034 | int ret; |
928b1f04 YT |
1035 | uid_t new_uid; |
1036 | gid_t new_gid; | |
e96a536c | 1037 | struct lxc_list *iterator; |
964581c2 CB |
1038 | uid_t nsuid = 0; |
1039 | gid_t nsgid = 0; | |
50e98013 | 1040 | |
18225d19 | 1041 | lxc_sync_fini_parent(handler); |
50e98013 | 1042 | |
9ff57a59 | 1043 | if (lxc_abstract_unix_recv_fds(data_sock1, &status_fd, 1, NULL, 0) < 0) { |
4d8bdfa0 CB |
1044 | ERROR("Failed to receive status file descriptor to child process"); |
1045 | goto out_warn_father; | |
1046 | } | |
1047 | ||
408da065 CB |
1048 | /* This prctl must be before the synchro, so if the parent dies before |
1049 | * we set the parent death signal, we will detect its death with the | |
1050 | * synchro right after, otherwise we have a window where the parent can | |
1051 | * exit before we set the pdeath signal leading to a unsupervized | |
1052 | * container. | |
743ecd2e | 1053 | */ |
4d8bdfa0 | 1054 | ret = lxc_set_death_signal(SIGKILL, handler->monitor_pid, status_fd); |
912314fc CB |
1055 | if (ret < 0) { |
1056 | SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL"); | |
18225d19 | 1057 | goto out_warn_father; |
743ecd2e DL |
1058 | } |
1059 | ||
611ddd34 CB |
1060 | ret = lxc_ambient_caps_up(); |
1061 | if (ret < 0) { | |
6f5e532f | 1062 | ERROR("Failed to raise ambient capabilities"); |
611ddd34 CB |
1063 | goto out_warn_father; |
1064 | } | |
1065 | ||
b467714b | 1066 | ret = pthread_sigmask(SIG_SETMASK, &handler->oldmask, NULL); |
18225d19 CB |
1067 | if (ret < 0) { |
1068 | SYSERROR("Failed to set signal mask"); | |
1069 | goto out_warn_father; | |
1070 | } | |
50e98013 | 1071 | |
6e48e7c5 | 1072 | if (!lxc_sync_wait_parent(handler, START_SYNC_STARTUP)) |
18225d19 | 1073 | goto out_warn_father; |
5b1e83cb | 1074 | |
408da065 CB |
1075 | /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See |
1076 | * https://github.com/lxc/lxd/issues/1978. | |
1077 | */ | |
f7176c3e | 1078 | if (handler->ns_unshare_flags & CLONE_NEWNET) { |
5b1e83cb SH |
1079 | ret = unshare(CLONE_NEWNET); |
1080 | if (ret < 0) { | |
e96a536c | 1081 | SYSERROR("Failed to unshare CLONE_NEWNET"); |
5b1e83cb SH |
1082 | goto out_warn_father; |
1083 | } | |
e96a536c | 1084 | INFO("Unshared CLONE_NEWNET"); |
5b1e83cb SH |
1085 | } |
1086 | ||
408da065 CB |
1087 | /* Tell the parent task it can begin to configure the container and wait |
1088 | * for it to finish. | |
3c22086f | 1089 | */ |
6e48e7c5 | 1090 | if (!lxc_sync_barrier_parent(handler, START_SYNC_CONFIGURE)) |
611ddd34 | 1091 | goto out_error; |
50e98013 | 1092 | |
e389f2af | 1093 | if (handler->ns_clone_flags & CLONE_NEWNET) { |
3c09b97c | 1094 | ret = lxc_network_recv_from_parent(handler); |
e389f2af CB |
1095 | if (ret < 0) { |
1096 | ERROR("Failed to receive veth names from parent"); | |
1097 | goto out_warn_father; | |
1098 | } | |
7ab1ba02 | 1099 | } |
658979c5 | 1100 | |
408da065 | 1101 | /* If we are in a new user namespace, become root there to have |
d08f8d2f | 1102 | * privilege over our namespace. |
f6d3e3e4 SH |
1103 | */ |
1104 | if (!lxc_list_empty(&handler->conf->id_map)) { | |
964581c2 CB |
1105 | if (!handler->conf->root_nsuid_map) |
1106 | nsuid = handler->conf->init_uid; | |
1107 | ||
1108 | if (!handler->conf->root_nsgid_map) | |
1109 | nsgid = handler->conf->init_gid; | |
4160c3a0 | 1110 | |
d08f8d2f CB |
1111 | /* Drop groups only after we switched to a valid gid in the new |
1112 | * user namespace. | |
1113 | */ | |
8917c382 | 1114 | if (!lxc_drop_groups() && |
8af07f82 | 1115 | (handler->am_root || errno != EPERM)) |
c476bdce | 1116 | goto out_warn_father; |
4b826b1f | 1117 | |
b58214ac CB |
1118 | if (!lxc_switch_uid_gid(nsuid, nsgid)) |
1119 | goto out_warn_father; | |
1120 | ||
b81689a1 CB |
1121 | ret = prctl(PR_SET_DUMPABLE, prctl_arg(1), prctl_arg(0), |
1122 | prctl_arg(0), prctl_arg(0)); | |
d7883725 CB |
1123 | if (ret < 0) |
1124 | goto out_warn_father; | |
912314fc CB |
1125 | |
1126 | /* set{g,u}id() clears deathsignal */ | |
4d8bdfa0 | 1127 | ret = lxc_set_death_signal(SIGKILL, handler->monitor_pid, status_fd); |
912314fc CB |
1128 | if (ret < 0) { |
1129 | SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL"); | |
1130 | goto out_warn_father; | |
1131 | } | |
f6d3e3e4 SH |
1132 | } |
1133 | ||
e96a536c CB |
1134 | ret = access(handler->lxcpath, X_OK); |
1135 | if (ret != 0) { | |
c8154066 SH |
1136 | print_top_failing_dir(handler->lxcpath); |
1137 | goto out_warn_father; | |
1138 | } | |
1139 | ||
7a55c157 TA |
1140 | /* In order to checkpoint restore, we need to have everything in the |
1141 | * same mount namespace. However, some containers may not have a | |
1142 | * reasonable /dev (in particular, they may not have /dev/null), so we | |
1143 | * can't set init's std fds to /dev/null by opening it from inside the | |
1144 | * container. | |
1145 | * | |
1146 | * If that's the case, fall back to using the host's /dev/null. This | |
1147 | * means that migration won't work, but at least we won't spew output | |
1148 | * where it isn't wanted. | |
1149 | */ | |
bb955810 | 1150 | if (handler->daemonize && !handler->conf->autodev) { |
2806a87d | 1151 | char path[PATH_MAX]; |
3c09b97c | 1152 | |
fa60cd7b CB |
1153 | ret = strnprintf(path, sizeof(path), "%s/dev/null", |
1154 | handler->conf->rootfs.mount); | |
1155 | if (ret < 0) | |
2806a87d | 1156 | goto out_warn_father; |
3c09b97c | 1157 | |
e96a536c CB |
1158 | ret = access(path, F_OK); |
1159 | if (ret != 0) { | |
1160 | devnull_fd = open_devnull(); | |
c44de748 | 1161 | |
e96a536c CB |
1162 | if (devnull_fd < 0) |
1163 | goto out_warn_father; | |
818a57fc | 1164 | WARN("Using /dev/null from the host for container init's standard file descriptors. Migration will not work"); |
e96a536c | 1165 | } |
c44de748 AM |
1166 | } |
1167 | ||
408da065 | 1168 | /* Ask father to setup cgroups and wait for him to finish. */ |
6e48e7c5 | 1169 | if (!lxc_sync_barrier_parent(handler, START_SYNC_CGROUP)) |
c44de748 | 1170 | goto out_error; |
544a48a0 | 1171 | |
deefdf8a CB |
1172 | /* Unshare cgroup namespace after we have setup our cgroups. If we do it |
1173 | * earlier we end up with a wrong view of /proc/self/cgroup. For | |
1174 | * example, assume we unshare(CLONE_NEWCGROUP) first, and then create | |
1175 | * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then | |
1176 | * /proc/self/cgroup would show us: | |
1177 | * | |
1178 | * 8:cpuset:/lxc/c | |
1179 | * | |
1180 | * whereas it should actually show | |
1181 | * | |
1182 | * 8:cpuset:/ | |
1183 | */ | |
f7176c3e | 1184 | if (handler->ns_unshare_flags & CLONE_NEWCGROUP) { |
e96a536c CB |
1185 | ret = unshare(CLONE_NEWCGROUP); |
1186 | if (ret < 0) { | |
bca7c59c CB |
1187 | if (errno != EINVAL) { |
1188 | SYSERROR("Failed to unshare CLONE_NEWCGROUP"); | |
1189 | goto out_warn_father; | |
1190 | } | |
1191 | ||
1192 | handler->ns_clone_flags &= ~CLONE_NEWCGROUP; | |
1193 | SYSINFO("Kernel does not support CLONE_NEWCGROUP"); | |
1194 | } else { | |
1195 | INFO("Unshared CLONE_NEWCGROUP"); | |
deefdf8a | 1196 | } |
deefdf8a CB |
1197 | } |
1198 | ||
f7176c3e | 1199 | if (handler->ns_unshare_flags & CLONE_NEWTIME) { |
70fd7fc9 CB |
1200 | ret = unshare(CLONE_NEWTIME); |
1201 | if (ret < 0) { | |
1202 | if (errno != EINVAL) { | |
1203 | SYSERROR("Failed to unshare CLONE_NEWTIME"); | |
1204 | goto out_warn_father; | |
1205 | } | |
1206 | ||
1207 | handler->ns_clone_flags &= ~CLONE_NEWTIME; | |
1208 | SYSINFO("Kernel does not support CLONE_NEWTIME"); | |
1209 | } else { | |
1210 | __do_close int timens_fd = -EBADF; | |
1211 | ||
1212 | INFO("Unshared CLONE_NEWTIME"); | |
1213 | ||
1214 | if (handler->conf->timens.s_boot) | |
1215 | ret = timens_offset_write(CLOCK_BOOTTIME, handler->conf->timens.s_boot, 0); | |
1216 | else if (handler->conf->timens.ns_boot) | |
1217 | ret = timens_offset_write(CLOCK_BOOTTIME, 0, handler->conf->timens.ns_boot); | |
1218 | if (ret) { | |
1219 | SYSERROR("Failed to write CLONE_BOOTTIME offset"); | |
1220 | goto out_warn_father; | |
1221 | } | |
1222 | TRACE("Wrote CLOCK_BOOTTIME offset"); | |
1223 | ||
1224 | if (handler->conf->timens.s_monotonic) | |
1225 | ret = timens_offset_write(CLOCK_MONOTONIC, handler->conf->timens.s_monotonic, 0); | |
1226 | else if (handler->conf->timens.ns_monotonic) | |
1227 | ret = timens_offset_write(CLOCK_MONOTONIC, 0, handler->conf->timens.ns_monotonic); | |
1228 | if (ret) { | |
1229 | SYSERROR("Failed to write CLONE_MONOTONIC offset"); | |
1230 | goto out_warn_father; | |
1231 | } | |
1232 | TRACE("Wrote CLOCK_MONOTONIC offset"); | |
1233 | ||
1234 | timens_fd = open("/proc/self/ns/time_for_children", O_RDONLY | O_CLOEXEC); | |
1235 | if (timens_fd < 0) { | |
1236 | SYSERROR("Failed to open \"/proc/self/ns/time_for_children\""); | |
1237 | goto out_warn_father; | |
1238 | } | |
1239 | ||
1240 | ret = setns(timens_fd, CLONE_NEWTIME); | |
1241 | if (ret) { | |
1242 | SYSERROR("Failed to setns(%d(\"/proc/self/ns/time_for_children\"))", timens_fd); | |
1243 | goto out_warn_father; | |
1244 | } | |
1245 | } | |
1246 | } | |
1247 | ||
149857af CB |
1248 | /* Add the requested environment variables to the current environment to |
1249 | * allow them to be used by the various hooks, such as the start hook | |
317494f1 | 1250 | * below. |
98ff08ca | 1251 | */ |
98ff08ca | 1252 | lxc_list_for_each(iterator, &handler->conf->environment) { |
e96a536c CB |
1253 | ret = putenv((char *)iterator->elem); |
1254 | if (ret < 0) { | |
1255 | SYSERROR("Failed to set environment variable: %s", | |
1256 | (char *)iterator->elem); | |
98ff08ca CB |
1257 | goto out_warn_father; |
1258 | } | |
1259 | } | |
1260 | ||
f4152036 | 1261 | /* Setup the container, ip, names, utsname, ... */ |
3b988b33 | 1262 | ret = lxc_setup(handler); |
7729f8e5 | 1263 | if (ret < 0) { |
e96a536c | 1264 | ERROR("Failed to setup container \"%s\"", handler->name); |
f4152036 CB |
1265 | goto out_warn_father; |
1266 | } | |
1267 | ||
408da065 | 1268 | /* Set the label to change to when we exec(2) the container's init. */ |
af04d847 | 1269 | ret = handler->lsm_ops->process_label_set(handler->lsm_ops, NULL, handler->conf, true); |
e96a536c | 1270 | if (ret < 0) |
e075f5d9 | 1271 | goto out_warn_father; |
5112cd70 | 1272 | |
029cdff5 | 1273 | /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it |
408da065 CB |
1274 | * before we aren't allowed anymore. |
1275 | */ | |
029cdff5 | 1276 | if (handler->conf->no_new_privs) { |
b81689a1 CB |
1277 | ret = prctl(PR_SET_NO_NEW_PRIVS, prctl_arg(1), prctl_arg(0), |
1278 | prctl_arg(0), prctl_arg(0)); | |
e96a536c | 1279 | if (ret < 0) { |
818a57fc | 1280 | SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges"); |
029cdff5 CB |
1281 | goto out_warn_father; |
1282 | } | |
818a57fc | 1283 | DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges"); |
029cdff5 CB |
1284 | } |
1285 | ||
0d9acb99 DE |
1286 | /* Some init's such as busybox will set sane tty settings on stdin, |
1287 | * stdout, stderr which it thinks is the console. We already set them | |
1288 | * the way we wanted on the real terminal, and we want init to do its | |
564e31c4 | 1289 | * setup on its console ie. the pty allocated in lxc_terminal_setup() so |
408da065 | 1290 | * make sure that that pty is stdin,stdout,stderr. |
0d9acb99 | 1291 | */ |
41808e20 | 1292 | if (handler->conf->console.pty >= 0) { |
bb955810 | 1293 | if (handler->daemonize || !handler->conf->is_execute) |
41808e20 | 1294 | ret = set_stdfds(handler->conf->console.pty); |
5d113f65 | 1295 | else |
41808e20 | 1296 | ret = lxc_terminal_set_stdfds(handler->conf->console.pty); |
5d113f65 | 1297 | if (ret < 0) { |
818a57fc | 1298 | ERROR("Failed to redirect std{in,out,err} to pty file descriptor %d", |
41808e20 | 1299 | handler->conf->console.pty); |
c5b93afb LF |
1300 | goto out_warn_father; |
1301 | } | |
5d113f65 | 1302 | } |
0d9acb99 | 1303 | |
408da065 | 1304 | /* If we mounted a temporary proc, then unmount it now. */ |
5112cd70 | 1305 | tmp_proc_unmount(handler->conf); |
e075f5d9 | 1306 | |
e96a536c CB |
1307 | ret = lxc_seccomp_load(handler->conf); |
1308 | if (ret < 0) | |
8f2c3a70 SH |
1309 | goto out_warn_father; |
1310 | ||
c3e3c21a CB |
1311 | ret = lxc_seccomp_send_notifier_fd(&handler->conf->seccomp, data_sock0); |
1312 | if (ret < 0) { | |
1313 | SYSERROR("Failed to send seccomp notify fd to parent"); | |
1314 | goto out_warn_father; | |
cdb2a47f | 1315 | } |
cdb2a47f | 1316 | |
e96a536c CB |
1317 | ret = run_lxc_hooks(handler->name, "start", handler->conf, NULL); |
1318 | if (ret < 0) { | |
1319 | ERROR("Failed to run lxc.hook.start for container \"%s\"", | |
1320 | handler->name); | |
773fb9ca SH |
1321 | goto out_warn_father; |
1322 | } | |
fc25b815 | 1323 | |
a783a414 | 1324 | close_prot_errno_disarm(handler->sigfd); |
26ddeedd | 1325 | |
41808e20 | 1326 | if (handler->conf->console.pty < 0 && handler->daemonize) { |
f4c177c3 CB |
1327 | if (devnull_fd < 0) { |
1328 | devnull_fd = open_devnull(); | |
1329 | if (devnull_fd < 0) | |
1330 | goto out_warn_father; | |
1331 | } | |
1332 | ||
e96a536c CB |
1333 | ret = set_stdfds(devnull_fd); |
1334 | if (ret < 0) { | |
1335 | ERROR("Failed to redirect std{in,out,err} to \"/dev/null\""); | |
c5b93afb LF |
1336 | goto out_warn_father; |
1337 | } | |
e96a536c | 1338 | } |
507cee36 | 1339 | |
a783a414 | 1340 | close_prot_errno_disarm(devnull_fd); |
c44de748 | 1341 | |
8c9a7665 TA |
1342 | setsid(); |
1343 | ||
e96a536c CB |
1344 | if (handler->conf->init_cwd) { |
1345 | ret = chdir(handler->conf->init_cwd); | |
1346 | if (ret < 0) { | |
1347 | SYSERROR("Could not change directory to \"%s\"", | |
1348 | handler->conf->init_cwd); | |
1349 | goto out_warn_father; | |
1350 | } | |
3c491553 L |
1351 | } |
1352 | ||
6e48e7c5 | 1353 | if (!lxc_sync_barrier_parent(handler, START_SYNC_CGROUP_LIMITS)) |
f4152036 CB |
1354 | goto out_warn_father; |
1355 | ||
149857af CB |
1356 | /* Reset the environment variables the user requested in a clear |
1357 | * environment. | |
1358 | */ | |
e96a536c CB |
1359 | ret = clearenv(); |
1360 | /* Don't error out though. */ | |
1361 | if (ret < 0) | |
149857af | 1362 | SYSERROR("Failed to clear environment."); |
149857af CB |
1363 | |
1364 | lxc_list_for_each(iterator, &handler->conf->environment) { | |
e96a536c CB |
1365 | ret = putenv((char *)iterator->elem); |
1366 | if (ret < 0) { | |
1367 | SYSERROR("Failed to set environment variable: %s", | |
1368 | (char *)iterator->elem); | |
149857af CB |
1369 | goto out_warn_father; |
1370 | } | |
1371 | } | |
1372 | ||
e96a536c CB |
1373 | ret = putenv("container=lxc"); |
1374 | if (ret < 0) { | |
1375 | SYSERROR("Failed to set environment variable: container=lxc"); | |
98ff08ca CB |
1376 | goto out_warn_father; |
1377 | } | |
1378 | ||
885766f5 CB |
1379 | if (handler->conf->ttys.tty_names) { |
1380 | ret = putenv(handler->conf->ttys.tty_names); | |
e96a536c CB |
1381 | if (ret < 0) { |
1382 | SYSERROR("Failed to set environment variable for container ptys"); | |
98ff08ca CB |
1383 | goto out_warn_father; |
1384 | } | |
1385 | } | |
1386 | ||
76bdf299 CB |
1387 | /* The container has been setup. We can now switch to an unprivileged |
1388 | * uid/gid. | |
1389 | */ | |
928b1f04 YT |
1390 | new_uid = handler->conf->init_uid; |
1391 | new_gid = handler->conf->init_gid; | |
76bdf299 | 1392 | |
964581c2 CB |
1393 | /* Avoid unnecessary syscalls. */ |
1394 | if (new_uid == nsuid) | |
1395 | new_uid = LXC_INVALID_UID; | |
1396 | ||
1397 | if (new_gid == nsgid) | |
1398 | new_gid = LXC_INVALID_GID; | |
c353b0b9 | 1399 | |
6aff5157 | 1400 | /* Make sure that the processes STDIO is correctly owned by the user that we are switching to */ |
c353b0b9 CB |
1401 | ret = fix_stdio_permissions(new_uid); |
1402 | if (ret) | |
1403 | WARN("Failed to ajust stdio permissions"); | |
964581c2 | 1404 | |
8af07f82 CB |
1405 | /* If we are in a new user namespace we already dropped all groups when |
1406 | * we switched to root in the new user namespace further above. Only | |
1407 | * drop groups if we can, so ensure that we have necessary privilege. | |
1408 | */ | |
bf31b337 | 1409 | if (lxc_list_empty(&handler->conf->id_map)) { |
8af07f82 CB |
1410 | #if HAVE_LIBCAP |
1411 | if (lxc_proc_cap_is_set(CAP_SETGID, CAP_EFFECTIVE)) | |
1412 | #endif | |
bf31b337 RJ |
1413 | { |
1414 | if (handler->conf->init_groups.size > 0) { | |
1415 | if (!lxc_setgroups(handler->conf->init_groups.list, | |
1416 | handler->conf->init_groups.size)) | |
1417 | goto out_warn_father; | |
1418 | } else { | |
1419 | if (!lxc_drop_groups()) | |
1420 | goto out_warn_father; | |
1421 | } | |
1422 | } | |
1423 | } | |
8af07f82 | 1424 | |
b58214ac CB |
1425 | if (!lxc_switch_uid_gid(new_uid, new_gid)) |
1426 | goto out_warn_father; | |
1427 | ||
611ddd34 CB |
1428 | ret = lxc_ambient_caps_down(); |
1429 | if (ret < 0) { | |
6f5e532f | 1430 | ERROR("Failed to clear ambient capabilities"); |
611ddd34 CB |
1431 | goto out_warn_father; |
1432 | } | |
1433 | ||
258f8051 | 1434 | if (handler->conf->monitor_signal_pdeath != SIGKILL) { |
4d8bdfa0 CB |
1435 | ret = lxc_set_death_signal(handler->conf->monitor_signal_pdeath, |
1436 | handler->monitor_pid, status_fd); | |
258f8051 CB |
1437 | if (ret < 0) { |
1438 | SYSERROR("Failed to set PR_SET_PDEATHSIG to %d", | |
1439 | handler->conf->monitor_signal_pdeath); | |
1440 | goto out_warn_father; | |
1441 | } | |
1442 | } | |
1443 | ||
d3103162 CB |
1444 | /* |
1445 | * After this call, we are in error because this ops should not return | |
408da065 CB |
1446 | * as it execs. |
1447 | */ | |
c4ea60df | 1448 | handler->ops->start(handler, handler->data); |
50e98013 DL |
1449 | |
1450 | out_warn_father: | |
d3103162 CB |
1451 | /* |
1452 | * We want the parent to know something went wrong, so we return a | |
408da065 CB |
1453 | * special error code. |
1454 | */ | |
946a0c6d | 1455 | lxc_sync_wake_parent(handler, SYNC_ERROR); |
c44de748 AM |
1456 | |
1457 | out_error: | |
50e98013 DL |
1458 | return -1; |
1459 | } | |
1460 | ||
ae467c54 | 1461 | static int lxc_recv_ttys_from_child(struct lxc_handler *handler) |
e8bd4e43 | 1462 | { |
ae467c54 | 1463 | int i; |
2520facd | 1464 | struct lxc_terminal_info *tty; |
ae467c54 | 1465 | int ret = -1; |
c6012571 | 1466 | int sock = handler->data_sock[1]; |
e8bd4e43 | 1467 | struct lxc_conf *conf = handler->conf; |
0e4be3cf | 1468 | struct lxc_tty_info *ttys = &conf->ttys; |
e8bd4e43 | 1469 | |
885766f5 | 1470 | if (!conf->ttys.max) |
e8bd4e43 SH |
1471 | return 0; |
1472 | ||
885766f5 | 1473 | ttys->tty = malloc(sizeof(*ttys->tty) * ttys->max); |
0e4be3cf | 1474 | if (!ttys->tty) |
e8bd4e43 | 1475 | return -1; |
e8bd4e43 | 1476 | |
885766f5 | 1477 | for (i = 0; i < conf->ttys.max; i++) { |
672c1e58 CB |
1478 | int ttyfds[2]; |
1479 | ||
1480 | ret = lxc_abstract_unix_recv_fds(sock, ttyfds, 2, NULL, 0); | |
1481 | if (ret < 0) | |
1482 | break; | |
ae467c54 | 1483 | |
0e4be3cf | 1484 | tty = &ttys->tty[i]; |
dd3de568 | 1485 | tty->busy = -1; |
36a94ce8 | 1486 | tty->ptx = ttyfds[0]; |
41808e20 CB |
1487 | tty->pty = ttyfds[1]; |
1488 | TRACE("Received pty with ptx fd %d and pty fd %d from child", tty->ptx, tty->pty); | |
e8bd4e43 | 1489 | } |
6d1400b5 | 1490 | |
ae467c54 | 1491 | if (ret < 0) |
6d1400b5 | 1492 | SYSERROR("Failed to receive %zu ttys from child", ttys->max); |
ae467c54 | 1493 | else |
885766f5 | 1494 | TRACE("Received %zu ttys from child", ttys->max); |
ae467c54 CB |
1495 | |
1496 | return ret; | |
e8bd4e43 SH |
1497 | } |
1498 | ||
5af9369b | 1499 | int resolve_clone_flags(struct lxc_handler *handler) |
f813849c | 1500 | { |
8bc8c715 CB |
1501 | int i; |
1502 | struct lxc_conf *conf = handler->conf; | |
70fd7fc9 CB |
1503 | bool wants_timens = conf->timens.s_boot || conf->timens.ns_boot || |
1504 | conf->timens.s_monotonic || conf->timens.ns_monotonic; | |
f813849c | 1505 | |
8bc8c715 | 1506 | for (i = 0; i < LXC_NS_MAX; i++) { |
f7a0c6ee CB |
1507 | if (conf->ns_keep) { |
1508 | if (!(conf->ns_keep & ns_info[i].clone_flag)) | |
becad0ec | 1509 | handler->ns_clone_flags |= ns_info[i].clone_flag; |
f7a0c6ee CB |
1510 | } else if (conf->ns_clone) { |
1511 | if ((conf->ns_clone & ns_info[i].clone_flag)) | |
becad0ec | 1512 | handler->ns_clone_flags |= ns_info[i].clone_flag; |
8bc8c715 CB |
1513 | } else { |
1514 | if (i == LXC_NS_USER && lxc_list_empty(&handler->conf->id_map)) | |
1515 | continue; | |
1516 | ||
1517 | if (i == LXC_NS_NET && lxc_requests_empty_network(handler)) | |
1518 | continue; | |
1519 | ||
7bd05339 CB |
1520 | if (i == LXC_NS_CGROUP && !cgns_supported()) |
1521 | continue; | |
1522 | ||
70fd7fc9 CB |
1523 | if (i == LXC_NS_TIME && !wants_timens) |
1524 | continue; | |
1525 | ||
becad0ec | 1526 | handler->ns_clone_flags |= ns_info[i].clone_flag; |
8bc8c715 | 1527 | } |
03df7ab5 | 1528 | |
8bc8c715 CB |
1529 | if (!conf->ns_share[i]) |
1530 | continue; | |
5af9369b | 1531 | |
becad0ec | 1532 | handler->ns_clone_flags &= ~ns_info[i].clone_flag; |
8bc8c715 | 1533 | TRACE("Sharing %s namespace", ns_info[i].proc_name); |
5af9369b CB |
1534 | } |
1535 | ||
70fd7fc9 CB |
1536 | if (wants_timens && (conf->ns_keep & ns_info[LXC_NS_TIME].clone_flag)) |
1537 | return log_trace_errno(-1, EINVAL, "Requested to keep time namespace while also specifying offsets"); | |
1538 | ||
f7176c3e CB |
1539 | /* Deal with namespaces that are unshared. */ |
1540 | if (handler->ns_clone_flags & CLONE_NEWTIME) | |
1541 | handler->ns_unshare_flags |= CLONE_NEWTIME; | |
1542 | ||
1543 | if (!pure_unified_layout(handler->cgroup_ops) && handler->ns_clone_flags & CLONE_NEWCGROUP) | |
1544 | handler->ns_unshare_flags |= CLONE_NEWCGROUP; | |
1545 | ||
1546 | if ((handler->ns_clone_flags & (CLONE_NEWNET | CLONE_NEWUSER)) == | |
1547 | (CLONE_NEWNET | CLONE_NEWUSER)) | |
1548 | handler->ns_unshare_flags |= CLONE_NEWNET; | |
1549 | ||
1550 | /* Deal with namespaces that are spawned. */ | |
1551 | handler->ns_on_clone_flags = handler->ns_clone_flags & ~handler->ns_unshare_flags; | |
1552 | ||
1553 | handler->clone_flags = handler->ns_on_clone_flags | CLONE_PIDFD; | |
1554 | ||
5af9369b | 1555 | return 0; |
f813849c TA |
1556 | } |
1557 | ||
8deca6c9 CB |
1558 | /* Note that this function is used with clone(CLONE_VM). Some glibc versions |
1559 | * used to reset the pid/tid to -1 when CLONE_VM was used without CLONE_THREAD. | |
1560 | * But since the memory between parent and child is shared on CLONE_VM this | |
1561 | * would invalidate the getpid() cache that glibc used to maintain and so | |
1562 | * getpid() in the child would return the parent's pid. This is all fixed in | |
1563 | * newer glibc versions where the getpid() cache is removed and the pid/tid is | |
1564 | * not reset anymore. | |
b35ddc5b | 1565 | * However, if for whatever reason you - dear committer - somehow need to get the |
8deca6c9 | 1566 | * pid of the dummy intermediate process for do_share_ns() you need to call |
0059379f CB |
1567 | * lxc_raw_getpid(). The next lxc_raw_clone() call does not employ CLONE_VM and |
1568 | * will be fine. | |
8deca6c9 CB |
1569 | */ |
1570 | static inline int do_share_ns(void *arg) | |
fa3a5b22 | 1571 | { |
8deca6c9 | 1572 | int i, flags, ret; |
fa3a5b22 CB |
1573 | struct lxc_handler *handler = arg; |
1574 | ||
8deca6c9 CB |
1575 | for (i = 0; i < LXC_NS_MAX; i++) { |
1576 | if (handler->nsfd[i] < 0) | |
1577 | continue; | |
fa3a5b22 | 1578 | |
8deca6c9 | 1579 | ret = setns(handler->nsfd[i], 0); |
c0b48eff CB |
1580 | if (ret < 0) { |
1581 | /* | |
1582 | * Note that joining a user and/or mount namespace | |
1583 | * requires the process is not multithreaded otherwise | |
1584 | * setns() will fail here. | |
1585 | */ | |
1586 | SYSERROR("Failed to inherit %s namespace", | |
1587 | ns_info[i].proc_name); | |
8deca6c9 | 1588 | return -1; |
c0b48eff | 1589 | } |
fa3a5b22 | 1590 | |
8deca6c9 | 1591 | DEBUG("Inherited %s namespace", ns_info[i].proc_name); |
fa3a5b22 CB |
1592 | } |
1593 | ||
becad0ec | 1594 | flags = handler->ns_on_clone_flags; |
8deca6c9 | 1595 | flags |= CLONE_PARENT; |
33942046 CB |
1596 | handler->pid = lxc_raw_clone_cb(do_start, handler, CLONE_PIDFD | flags, |
1597 | &handler->pidfd); | |
8deca6c9 | 1598 | if (handler->pid < 0) |
fa3a5b22 CB |
1599 | return -1; |
1600 | ||
8deca6c9 | 1601 | return 0; |
fa3a5b22 CB |
1602 | } |
1603 | ||
480588e6 CB |
1604 | /* lxc_spawn() performs crucial setup tasks and clone()s the new process which |
1605 | * exec()s the requested container binary. | |
1606 | * Note that lxc_spawn() runs in the parent namespaces. Any operations performed | |
1607 | * right here should be double checked if they'd pose a security risk. (For | |
1608 | * example, any {u}mount() operations performed here will be reflected on the | |
1609 | * host!) | |
1610 | */ | |
74a3920a | 1611 | static int lxc_spawn(struct lxc_handler *handler) |
59eb99ba | 1612 | { |
f62cf1d4 | 1613 | __do_close int data_sock0 = -EBADF, data_sock1 = -EBADF; |
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 | |
6e48e7c5 | 1638 | if (!lxc_sync_init(handler)) |
9d7f9e52 | 1639 | return -1; |
0ad19a3f | 1640 | |
b9f522c5 | 1641 | ret = socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0, |
790255cf | 1642 | handler->data_sock); |
95ef0d7c CB |
1643 | if (ret < 0) |
1644 | goto out_sync_fini; | |
cdb2a47f CB |
1645 | data_sock0 = handler->data_sock[0]; |
1646 | data_sock1 = handler->data_sock[1]; | |
e8bd4e43 | 1647 | |
5af9369b | 1648 | ret = resolve_clone_flags(handler); |
95ef0d7c CB |
1649 | if (ret < 0) |
1650 | goto out_sync_fini; | |
9f30a190 | 1651 | |
03ca4af8 TP |
1652 | if (handler->ns_clone_flags & CLONE_NEWNET) { |
1653 | ret = lxc_find_gateway_addresses(handler); | |
1654 | if (ret) { | |
1655 | ERROR("Failed to find gateway addresses"); | |
1656 | goto out_sync_fini; | |
1657 | } | |
1658 | } | |
1659 | ||
e8b181f5 | 1660 | if (!cgroup_ops->payload_create(cgroup_ops, handler)) { |
cfc62c60 | 1661 | ERROR("Failed creating cgroups"); |
47d8fb3b CS |
1662 | goto out_delete_net; |
1663 | } | |
1664 | ||
408da065 CB |
1665 | /* If the rootfs is not a blockdev, prevent the container from marking |
1666 | * it readonly. | |
1667 | * If the container is unprivileged then skip rootfs pinning. | |
0c547523 | 1668 | */ |
79ff643d CB |
1669 | ret = lxc_rootfs_prepare(&conf->rootfs, wants_to_map_ids); |
1670 | if (ret) { | |
1671 | ERROR("Failed to handle rootfs pinning for container \"%s\"", handler->name); | |
1672 | goto out_delete_net; | |
5e32a990 | 1673 | } |
0c547523 | 1674 | |
408da065 | 1675 | /* Create a process in a new set of namespaces. */ |
4e232466 CB |
1676 | if (share_ns) { |
1677 | pid_t attacher_pid; | |
1678 | ||
cfc62c60 | 1679 | attacher_pid = lxc_clone(do_share_ns, handler, |
33258b95 | 1680 | CLONE_VFORK | CLONE_VM | CLONE_FILES, NULL); |
4e232466 CB |
1681 | if (attacher_pid < 0) { |
1682 | SYSERROR(LXC_CLONE_ERROR); | |
1683 | goto out_delete_net; | |
1684 | } | |
1685 | ||
1686 | ret = wait_for_pid(attacher_pid); | |
1687 | if (ret < 0) { | |
1688 | SYSERROR("Intermediate process failed"); | |
1689 | goto out_delete_net; | |
1690 | } | |
f7176c3e CB |
1691 | |
1692 | if (handler->pid < 0) { | |
1693 | SYSERROR(LXC_CLONE_ERROR); | |
1694 | goto out_delete_net; | |
1695 | } | |
4e232466 | 1696 | } else { |
84b66ced | 1697 | int cgroup_fd = -EBADF; |
f7176c3e CB |
1698 | |
1699 | struct lxc_clone_args clone_args = { | |
1700 | .flags = handler->clone_flags, | |
1701 | .pidfd = ptr_to_u64(&handler->pidfd), | |
1702 | .exit_signal = SIGCHLD, | |
1703 | }; | |
1704 | ||
1705 | if (handler->ns_clone_flags & CLONE_NEWCGROUP) { | |
1706 | cgroup_fd = cgroup_unified_fd(cgroup_ops); | |
1707 | if (cgroup_fd >= 0) { | |
1708 | handler->clone_flags |= CLONE_INTO_CGROUP; | |
1709 | clone_args.flags |= CLONE_INTO_CGROUP; | |
1710 | clone_args.cgroup = cgroup_fd; | |
1711 | } | |
1712 | } | |
1713 | ||
1714 | /* Try to spawn directly into target cgroup. */ | |
1715 | handler->pid = lxc_clone3(&clone_args, CLONE_ARGS_SIZE_VER2); | |
1716 | if (handler->pid < 0) { | |
1717 | SYSTRACE("Failed to spawn container directly into target cgroup"); | |
1718 | ||
1719 | /* Kernel might simply be too old for CLONE_INTO_CGROUP. */ | |
1720 | handler->clone_flags &= ~(CLONE_INTO_CGROUP | CLONE_NEWCGROUP); | |
1721 | handler->ns_on_clone_flags &= ~CLONE_NEWCGROUP; | |
1722 | handler->ns_unshare_flags |= CLONE_NEWCGROUP; | |
1723 | ||
1724 | clone_args.flags = handler->clone_flags; | |
1725 | ||
1726 | handler->pid = lxc_clone3(&clone_args, CLONE_ARGS_SIZE_VER0); | |
1727 | } else if (cgroup_fd >= 0) { | |
1728 | TRACE("Spawned container directly into target cgroup via cgroup2 fd %d", cgroup_fd); | |
1729 | } | |
1730 | ||
1731 | /* Kernel might be too old for clone3(). */ | |
1732 | if (handler->pid < 0) { | |
1733 | SYSTRACE("Failed to spawn container via clone3()"); | |
1734 | handler->pid = lxc_raw_legacy_clone(handler->clone_flags, &handler->pidfd); | |
1735 | } | |
1736 | ||
1737 | if (handler->pid < 0) { | |
1738 | SYSERROR(LXC_CLONE_ERROR); | |
1739 | goto out_delete_net; | |
1740 | } | |
1741 | ||
1742 | if (handler->pid == 0) { | |
1743 | (void)do_start(handler); | |
1744 | _exit(EXIT_FAILURE); | |
1745 | } | |
0ad19a3f | 1746 | } |
f7176c3e CB |
1747 | if (handler->pidfd < 0) |
1748 | handler->clone_flags &= ~CLONE_PIDFD; | |
fa3a5b22 | 1749 | TRACE("Cloned child process %d", handler->pid); |
1bd8d726 | 1750 | |
39293f22 CB |
1751 | /* Verify that we can actually make use of pidfds. */ |
1752 | if (!lxc_can_use_pidfd(handler->pidfd)) | |
1753 | close_prot_errno_disarm(handler->pidfd); | |
1754 | ||
fa60cd7b CB |
1755 | ret = strnprintf(pidstr, 20, "%d", handler->pid); |
1756 | if (ret < 0) | |
1871e646 CB |
1757 | goto out_delete_net; |
1758 | ||
1759 | ret = setenv("LXC_PID", pidstr, 1); | |
1760 | if (ret < 0) | |
1761 | SYSERROR("Failed to set environment variable: LXC_PID=%s", pidstr); | |
1762 | ||
9662e444 | 1763 | for (i = 0; i < LXC_NS_MAX; i++) |
becad0ec | 1764 | if (handler->ns_on_clone_flags & ns_info[i].clone_flag) |
28d9e29e | 1765 | INFO("Cloned %s", ns_info[i].flag_name); |
0ad19a3f | 1766 | |
8db6be1b | 1767 | if (!lxc_try_preserve_namespaces(handler, handler->ns_on_clone_flags)) { |
28d9e29e CB |
1768 | ERROR("Failed to preserve cloned namespaces for lxc.hook.stop"); |
1769 | goto out_delete_net; | |
1770 | } | |
b6b2b194 | 1771 | |
3c22086f CLG |
1772 | lxc_sync_fini_child(handler); |
1773 | ||
4d8bdfa0 CB |
1774 | if (lxc_abstract_unix_send_fds(handler->data_sock[0], &handler->monitor_status_fd, 1, NULL, 0) < 0) { |
1775 | ERROR("Failed to send status file descriptor to child process"); | |
1776 | goto out_delete_net; | |
1777 | } | |
1778 | close_prot_errno_disarm(handler->monitor_status_fd); | |
1779 | ||
408da065 CB |
1780 | /* Map the container uids. The container became an invalid userid the |
1781 | * moment it was cloned with CLONE_NEWUSER. This call doesn't change | |
1782 | * anything immediately, but allows the container to setuid(0) (0 being | |
1783 | * mapped to something else on the host.) later to become a valid uid | |
1784 | * again. | |
1785 | */ | |
fa3a5b22 | 1786 | if (wants_to_map_ids) { |
df3c5314 | 1787 | if (!handler->conf->ns_share[LXC_NS_USER] && |
46186acd | 1788 | (handler->conf->ns_keep & CLONE_NEWUSER) == 0) { |
fa3a5b22 CB |
1789 | ret = lxc_map_ids(id_map, handler->pid); |
1790 | if (ret < 0) { | |
1791 | ERROR("Failed to set up id mapping."); | |
1792 | goto out_delete_net; | |
1793 | } | |
1794 | } | |
5b1e83cb SH |
1795 | } |
1796 | ||
6e48e7c5 | 1797 | if (!lxc_sync_wake_child(handler, START_SYNC_STARTUP)) |
5b1e83cb | 1798 | goto out_delete_net; |
5b1e83cb | 1799 | |
6e48e7c5 | 1800 | if (!lxc_sync_wait_child(handler, START_SYNC_CONFIGURE)) |
5b1e83cb | 1801 | goto out_delete_net; |
0ad19a3f | 1802 | |
c581d2a6 | 1803 | if (!cgroup_ops->setup_limits_legacy(cgroup_ops, handler->conf, false)) { |
cfc62c60 | 1804 | ERROR("Failed to setup cgroup limits for container \"%s\"", name); |
6031a6e5 DE |
1805 | goto out_delete_net; |
1806 | } | |
1807 | ||
fe70edee | 1808 | if (!cgroup_ops->payload_enter(cgroup_ops, handler)) { |
b1ee458e | 1809 | ERROR("Failed to enter cgroups"); |
7fef7a06 | 1810 | goto out_delete_net; |
fe70edee | 1811 | } |
218d4250 | 1812 | |
c581d2a6 CB |
1813 | if (!cgroup_ops->payload_delegate_controllers(cgroup_ops)) { |
1814 | ERROR("Failed to delegate controllers to payload cgroup"); | |
1815 | goto out_delete_net; | |
1816 | } | |
1817 | ||
1818 | if (!cgroup_ops->setup_limits(cgroup_ops, handler)) { | |
1819 | ERROR("Failed to setup cgroup limits for container \"%s\"", name); | |
1820 | goto out_delete_net; | |
1821 | } | |
1822 | ||
2202afc9 | 1823 | if (!cgroup_ops->chown(cgroup_ops, handler->conf)) |
0996e18a SH |
1824 | goto out_delete_net; |
1825 | ||
aa0c0e7b RK |
1826 | /* If not done yet, we're now ready to preserve the network namespace */ |
1827 | if (handler->nsfd[LXC_NS_NET] < 0) { | |
8db6be1b | 1828 | ret = lxc_try_preserve_namespace(handler, LXC_NS_NET, "net"); |
aa0c0e7b RK |
1829 | if (ret < 0) { |
1830 | if (ret != -EOPNOTSUPP) { | |
1831 | SYSERROR("Failed to preserve net namespace"); | |
1832 | goto out_delete_net; | |
1833 | } | |
4cb53844 | 1834 | } |
fa3a5b22 | 1835 | } |
aa0c0e7b RK |
1836 | ret = lxc_netns_set_nsid(handler->nsfd[LXC_NS_NET]); |
1837 | if (ret < 0) | |
1838 | SYSWARN("Failed to allocate new network namespace id"); | |
1839 | else | |
1840 | TRACE("Allocated new network namespace id"); | |
fa3a5b22 | 1841 | |
408da065 | 1842 | /* Create the network configuration. */ |
becad0ec | 1843 | if (handler->ns_clone_flags & CLONE_NEWNET) { |
e389f2af | 1844 | ret = lxc_create_network(handler); |
cfc62c60 | 1845 | if (ret < 0) { |
e389f2af | 1846 | ERROR("Failed to create the network"); |
7fef7a06 | 1847 | goto out_delete_net; |
82d5ae15 | 1848 | } |
74c6e2b0 | 1849 | |
3c09b97c | 1850 | ret = lxc_network_send_to_child(handler); |
cfc62c60 | 1851 | if (ret < 0) { |
e389f2af | 1852 | ERROR("Failed to send veth names to child"); |
74c6e2b0 CB |
1853 | goto out_delete_net; |
1854 | } | |
0ad19a3f | 1855 | } |
1856 | ||
61d7a733 YT |
1857 | if (!lxc_list_empty(&conf->procs)) { |
1858 | ret = setup_proc_filesystem(&conf->procs, handler->pid); | |
1859 | if (ret < 0) | |
1860 | goto out_delete_net; | |
1861 | } | |
1862 | ||
408da065 | 1863 | /* Tell the child to continue its initialization. We'll get |
946a0c6d | 1864 | * START_SYNC_CGROUP when it is ready for us to setup cgroups. |
3c22086f | 1865 | */ |
6e48e7c5 | 1866 | if (!lxc_sync_barrier_child(handler, START_SYNC_POST_CONFIGURE)) |
544a48a0 SH |
1867 | goto out_delete_net; |
1868 | ||
cfc62c60 CB |
1869 | if (!lxc_list_empty(&conf->limits)) { |
1870 | ret = setup_resource_limits(&conf->limits, handler->pid); | |
1871 | if (ret < 0) { | |
1872 | ERROR("Failed to setup resource limits"); | |
1873 | goto out_delete_net; | |
1874 | } | |
c6d09e15 WB |
1875 | } |
1876 | ||
6e48e7c5 | 1877 | if (!lxc_sync_barrier_child(handler, START_SYNC_CGROUP_UNSHARE)) |
f4152036 CB |
1878 | goto out_delete_net; |
1879 | ||
432faf20 WB |
1880 | /* |
1881 | * with isolation the limiting devices cgroup was already setup, so | |
1882 | * only setup devices here if we have no namespace directory | |
1883 | */ | |
1884 | if (!handler->conf->cgroup_meta.namespace_dir && | |
1885 | !cgroup_ops->setup_limits_legacy(cgroup_ops, handler->conf, true)) { | |
cfc62c60 | 1886 | ERROR("Failed to setup legacy device cgroup controller limits"); |
b98f7d6e | 1887 | goto out_delete_net; |
544a48a0 | 1888 | } |
cfc62c60 | 1889 | TRACE("Set up legacy device cgroup controller limits"); |
544a48a0 | 1890 | |
bf651989 CB |
1891 | if (!cgroup_ops->devices_activate(cgroup_ops, handler)) { |
1892 | ERROR("Failed to setup cgroup2 device controller limits"); | |
1893 | goto out_delete_net; | |
1894 | } | |
1895 | TRACE("Set up cgroup2 device controller limits"); | |
1896 | ||
f7176c3e | 1897 | if (handler->ns_unshare_flags & CLONE_NEWCGROUP) { |
8bf3abfb | 1898 | /* Now we're ready to preserve the cgroup namespace */ |
8db6be1b | 1899 | ret = lxc_try_preserve_namespace(handler, LXC_NS_CGROUP, "cgroup"); |
8bf3abfb | 1900 | if (ret < 0) { |
4cb53844 | 1901 | if (ret != -EOPNOTSUPP) { |
6d1400b5 | 1902 | SYSERROR("Failed to preserve cgroup namespace"); |
4cb53844 CB |
1903 | goto out_delete_net; |
1904 | } | |
8bf3abfb | 1905 | } |
8bf3abfb CB |
1906 | } |
1907 | ||
78eb6aa6 CB |
1908 | cgroup_ops->payload_finalize(cgroup_ops); |
1909 | TRACE("Finished setting up cgroups"); | |
1910 | ||
f7176c3e | 1911 | if (handler->ns_unshare_flags & CLONE_NEWTIME) { |
f1c43439 | 1912 | /* Now we're ready to preserve the cgroup namespace */ |
8db6be1b | 1913 | ret = lxc_try_preserve_namespace(handler, LXC_NS_TIME, "time"); |
f1c43439 CB |
1914 | if (ret < 0) { |
1915 | if (ret != -EOPNOTSUPP) { | |
1916 | SYSERROR("Failed to preserve time namespace"); | |
1917 | goto out_delete_net; | |
1918 | } | |
f1c43439 CB |
1919 | } |
1920 | } | |
1921 | ||
08dd2805 | 1922 | /* Run any host-side start hooks */ |
cfc62c60 CB |
1923 | ret = run_lxc_hooks(name, "start-host", conf, NULL); |
1924 | if (ret < 0) { | |
1925 | ERROR("Failed to run lxc.hook.start-host"); | |
341ed84c | 1926 | goto out_delete_net; |
08dd2805 SH |
1927 | } |
1928 | ||
408da065 CB |
1929 | /* Tell the child to complete its initialization and wait for it to exec |
1930 | * or return an error. (The child will never return | |
946a0c6d | 1931 | * START_SYNC_READY_START+1. It will either close the sync pipe, causing |
408da065 CB |
1932 | * lxc_sync_barrier_child to return success, or return a different |
1933 | * value, causing us to error out). | |
544a48a0 | 1934 | */ |
6e48e7c5 | 1935 | if (!lxc_sync_barrier_child(handler, START_SYNC_READY_START)) |
341ed84c | 1936 | goto out_delete_net; |
0ad19a3f | 1937 | |
e389f2af CB |
1938 | if (handler->ns_clone_flags & CLONE_NEWNET) { |
1939 | ret = lxc_network_recv_name_and_ifindex_from_child(handler); | |
1940 | if (ret < 0) { | |
1941 | ERROR("Failed to receive names and ifindices for network devices from child"); | |
1942 | goto out_delete_net; | |
1943 | } | |
790255cf CB |
1944 | } |
1945 | ||
68f3899e | 1946 | ret = lxc_setup_devpts_parent(handler); |
f797f05e CB |
1947 | if (ret < 0) { |
1948 | SYSERROR("Failed to receive devpts fd from child"); | |
1949 | goto out_delete_net; | |
1950 | } | |
f797f05e | 1951 | |
790255cf | 1952 | /* Now all networks are created, network devices are moved into place, |
ad2ddfcd | 1953 | * and the correct names and ifindices in the respective namespaces have |
790255cf CB |
1954 | * been recorded. The corresponding structs have now all been filled. So |
1955 | * log them for debugging purposes. | |
1956 | */ | |
28d9e29e | 1957 | lxc_log_configured_netdevs(conf); |
790255cf | 1958 | |
f4152036 | 1959 | /* Read tty fds allocated by child. */ |
cfc62c60 CB |
1960 | ret = lxc_recv_ttys_from_child(handler); |
1961 | if (ret < 0) { | |
1962 | ERROR("Failed to receive tty info from child process"); | |
f4152036 CB |
1963 | goto out_delete_net; |
1964 | } | |
1965 | ||
c3e3c21a CB |
1966 | ret = lxc_seccomp_recv_notifier_fd(&handler->conf->seccomp, data_sock1); |
1967 | if (ret < 0) { | |
1968 | SYSERROR("Failed to receive seccomp notify fd from child"); | |
1969 | goto out_delete_net; | |
cdb2a47f | 1970 | } |
cdb2a47f | 1971 | |
cfc62c60 CB |
1972 | ret = handler->ops->post_start(handler, handler->data); |
1973 | if (ret < 0) | |
e6126dbe MN |
1974 | goto out_abort; |
1975 | ||
cfc62c60 CB |
1976 | ret = lxc_set_state(name, handler, RUNNING); |
1977 | if (ret < 0) { | |
1978 | ERROR("Failed to set state to \"%s\"", lxc_state2str(RUNNING)); | |
59eb99ba | 1979 | goto out_abort; |
3f21c114 | 1980 | } |
22ebac19 | 1981 | |
3c22086f | 1982 | lxc_sync_fini(handler); |
0c547523 | 1983 | |
e6126dbe | 1984 | return 0; |
1ac470c0 | 1985 | |
7fef7a06 | 1986 | out_delete_net: |
becad0ec | 1987 | if (handler->ns_clone_flags & CLONE_NEWNET) |
bb84beda | 1988 | lxc_delete_network(handler); |
1bd8d726 | 1989 | |
59eb99ba | 1990 | out_abort: |
0c5859ff | 1991 | lxc_abort(handler); |
95ef0d7c CB |
1992 | |
1993 | out_sync_fini: | |
3c22086f | 1994 | lxc_sync_fini(handler); |
5c068da9 | 1995 | |
b79fcd86 | 1996 | return -1; |
59eb99ba | 1997 | } |
0ad19a3f | 1998 | |
7bdf97d2 CB |
1999 | int __lxc_start(struct lxc_handler *handler, struct lxc_operations *ops, |
2000 | void *data, const char *lxcpath, bool daemonize, int *error_num) | |
59eb99ba | 2001 | { |
c30e9b19 | 2002 | int ret, status; |
7bdf97d2 | 2003 | const char *name = handler->name; |
aa460476 | 2004 | struct lxc_conf *conf = handler->conf; |
72068e74 | 2005 | struct cgroup_ops *cgroup_ops; |
80090207 | 2006 | |
c30e9b19 CB |
2007 | ret = lxc_init(name, handler); |
2008 | if (ret < 0) { | |
2009 | ERROR("Failed to initialize container \"%s\"", name); | |
9ad66589 | 2010 | goto out_abort; |
0ad19a3f | 2011 | } |
ee70bf78 CLG |
2012 | handler->ops = ops; |
2013 | handler->data = data; | |
bb955810 | 2014 | handler->daemonize = daemonize; |
72068e74 | 2015 | cgroup_ops = handler->cgroup_ops; |
e6126dbe | 2016 | |
76a26f55 | 2017 | if (!attach_block_device(handler->conf)) { |
c30e9b19 | 2018 | ERROR("Failed to attach block device"); |
575ea467 | 2019 | ret = -1; |
9ad66589 | 2020 | goto out_abort; |
76a26f55 SH |
2021 | } |
2022 | ||
72068e74 CB |
2023 | if (!cgroup_ops->monitor_create(cgroup_ops, handler)) { |
2024 | ERROR("Failed to create monitor cgroup"); | |
575ea467 | 2025 | ret = -1; |
9ad66589 | 2026 | goto out_abort; |
eeef32bb CB |
2027 | } |
2028 | ||
c581d2a6 | 2029 | if (!cgroup_ops->monitor_enter(cgroup_ops, handler)) { |
eeef32bb | 2030 | ERROR("Failed to enter monitor cgroup"); |
575ea467 | 2031 | ret = -1; |
9ad66589 | 2032 | goto out_abort; |
72068e74 CB |
2033 | } |
2034 | ||
c581d2a6 CB |
2035 | if (!cgroup_ops->monitor_delegate_controllers(cgroup_ops)) { |
2036 | ERROR("Failed to delegate controllers to monitor cgroup"); | |
2037 | ret = -1; | |
9ad66589 | 2038 | goto out_abort; |
c581d2a6 CB |
2039 | } |
2040 | ||
35120d9c | 2041 | if (geteuid() == 0 && !lxc_list_empty(&conf->id_map)) { |
408da065 | 2042 | /* If the backing store is a device, mount it here and now. */ |
35120d9c | 2043 | if (rootfs_is_blockdev(conf)) { |
c30e9b19 CB |
2044 | ret = unshare(CLONE_NEWNS); |
2045 | if (ret < 0) { | |
2046 | ERROR("Failed to unshare CLONE_NEWNS"); | |
9ad66589 | 2047 | goto out_abort; |
35120d9c | 2048 | } |
c30e9b19 | 2049 | INFO("Unshared CLONE_NEWNS"); |
408da065 | 2050 | |
9e61fb1f | 2051 | turn_into_dependent_mounts(); |
8ce1abc2 | 2052 | ret = lxc_setup_rootfs_prepare_root(conf, name, lxcpath); |
c30e9b19 CB |
2053 | if (ret < 0) { |
2054 | ERROR("Error setting up rootfs mount as root before spawn"); | |
9ad66589 | 2055 | goto out_abort; |
35120d9c | 2056 | } |
c30e9b19 | 2057 | INFO("Set up container rootfs as host root"); |
35120d9c SH |
2058 | } |
2059 | } | |
2060 | ||
c30e9b19 CB |
2061 | ret = lxc_spawn(handler); |
2062 | if (ret < 0) { | |
2063 | ERROR("Failed to spawn container \"%s\"", name); | |
76a26f55 | 2064 | goto out_detach_blockdev; |
0ad19a3f | 2065 | } |
2066 | ||
80308d07 | 2067 | handler->conf->reboot = REBOOT_NONE; |
8bee8851 | 2068 | |
c30e9b19 CB |
2069 | ret = lxc_poll(name, handler); |
2070 | if (ret) { | |
2071 | ERROR("LXC mainloop exited with error: %d", ret); | |
9ad66589 | 2072 | goto out_delete_network; |
59eb99ba | 2073 | } |
0ad19a3f | 2074 | |
321db026 DJ |
2075 | if (!handler->init_died && handler->pid > 0) { |
2076 | ERROR("Child process is not killed"); | |
575ea467 | 2077 | ret = -1; |
9ad66589 | 2078 | goto out_delete_network; |
321db026 DJ |
2079 | } |
2080 | ||
c30e9b19 CB |
2081 | status = lxc_wait_for_pid_status(handler->pid); |
2082 | if (status < 0) | |
2083 | SYSERROR("Failed to retrieve status for %d", handler->pid); | |
e043236e | 2084 | |
408da065 CB |
2085 | /* If the child process exited but was not signaled, it didn't call |
2086 | * reboot. This should mean it was an lxc-execute which simply exited. | |
2087 | * In any case, treat it as a 'halt'. | |
8b004f07 | 2088 | */ |
d028235d | 2089 | if (WIFSIGNALED(status)) { |
8b004f07 SH |
2090 | switch(WTERMSIG(status)) { |
2091 | case SIGINT: /* halt */ | |
c30e9b19 | 2092 | DEBUG("Container \"%s\" is halting", name); |
8b004f07 SH |
2093 | break; |
2094 | case SIGHUP: /* reboot */ | |
c30e9b19 | 2095 | DEBUG("Container \"%s\" is rebooting", name); |
80308d07 | 2096 | handler->conf->reboot = REBOOT_REQ; |
8b004f07 | 2097 | break; |
c2b9bd9e | 2098 | case SIGSYS: /* seccomp */ |
c30e9b19 | 2099 | DEBUG("Container \"%s\" violated its seccomp policy", name); |
c2b9bd9e | 2100 | break; |
8b004f07 | 2101 | default: |
c30e9b19 | 2102 | DEBUG("Unknown exit status for container \"%s\" init %d", name, WTERMSIG(status)); |
8b004f07 SH |
2103 | break; |
2104 | } | |
d028235d | 2105 | } |
828695d9 | 2106 | |
c30e9b19 CB |
2107 | ret = lxc_restore_phys_nics_to_netns(handler); |
2108 | if (ret < 0) | |
818a57fc | 2109 | ERROR("Failed to move physical network devices back to parent network namespace"); |
ce5782df | 2110 | |
1787abca | 2111 | lxc_monitor_send_exit_code(name, status, handler->lxcpath); |
eb808539 | 2112 | lxc_error_set_and_log(handler->pid, status); |
a3b4f3d6 TA |
2113 | if (error_num) |
2114 | *error_num = handler->exit_status; | |
1bd8d726 | 2115 | |
bb84beda | 2116 | lxc_delete_network(handler); |
76a26f55 | 2117 | detach_block_device(handler->conf); |
fd5be714 | 2118 | lxc_end(handler); |
c30e9b19 | 2119 | return ret; |
0ad19a3f | 2120 | |
9ad66589 | 2121 | out_abort: |
0c5859ff | 2122 | lxc_abort(handler); |
945daa24 CB |
2123 | lxc_end(handler); |
2124 | return ret; | |
65a2dc8b CB |
2125 | |
2126 | out_detach_blockdev: | |
0c5859ff | 2127 | lxc_abort(handler); |
945daa24 CB |
2128 | detach_block_device(handler->conf); |
2129 | lxc_end(handler); | |
2130 | return ret; | |
65a2dc8b | 2131 | |
9ad66589 | 2132 | out_delete_network: |
0c5859ff | 2133 | lxc_abort(handler); |
abd953eb | 2134 | lxc_restore_phys_nics_to_netns(handler); |
945daa24 CB |
2135 | lxc_delete_network(handler); |
2136 | detach_block_device(handler->conf); | |
2137 | lxc_end(handler); | |
2138 | return ret; | |
0ad19a3f | 2139 | } |
ee70bf78 CLG |
2140 | |
2141 | struct start_args { | |
2142 | char *const *argv; | |
2143 | }; | |
2144 | ||
2145 | static int start(struct lxc_handler *handler, void* data) | |
2146 | { | |
2147 | struct start_args *arg = data; | |
2148 | ||
984984e4 | 2149 | NOTICE("Exec'ing \"%s\"", arg->argv[0]); |
ee70bf78 CLG |
2150 | |
2151 | execvp(arg->argv[0], arg->argv); | |
984984e4 | 2152 | SYSERROR("Failed to exec \"%s\"", arg->argv[0]); |
ee70bf78 CLG |
2153 | return 0; |
2154 | } | |
2155 | ||
2156 | static int post_start(struct lxc_handler *handler, void* data) | |
2157 | { | |
2158 | struct start_args *arg = data; | |
2159 | ||
4c8e880e | 2160 | NOTICE("Started \"%s\" with pid \"%d\"", arg->argv[0], handler->pid); |
ee70bf78 CLG |
2161 | return 0; |
2162 | } | |
2163 | ||
2164 | static struct lxc_operations start_ops = { | |
2165 | .start = start, | |
2166 | .post_start = post_start | |
2167 | }; | |
2168 | ||
7bdf97d2 | 2169 | int lxc_start(char *const argv[], struct lxc_handler *handler, |
bb955810 | 2170 | const char *lxcpath, bool daemonize, int *error_num) |
ee70bf78 CLG |
2171 | { |
2172 | struct start_args start_arg = { | |
2173 | .argv = argv, | |
2174 | }; | |
2175 | ||
b2efeb0b | 2176 | TRACE("Doing lxc_start"); |
7bdf97d2 | 2177 | return __lxc_start(handler, &start_ops, &start_arg, lxcpath, daemonize, error_num); |
ee70bf78 | 2178 | } |
28272964 CB |
2179 | |
2180 | static void lxc_destroy_container_on_signal(struct lxc_handler *handler, | |
2181 | const char *name) | |
2182 | { | |
245100a0 | 2183 | char destroy[PATH_MAX]; |
f01f7975 | 2184 | struct lxc_container *c; |
ae3beac9 CB |
2185 | int ret = 0; |
2186 | bool bret = true; | |
2187 | ||
df31363a | 2188 | if (handler->conf->rootfs.path && handler->conf->rootfs.mount) { |
790255cf | 2189 | bret = do_destroy_container(handler); |
28272964 | 2190 | if (!bret) { |
ae3beac9 | 2191 | ERROR("Error destroying rootfs for container \"%s\"", name); |
28272964 CB |
2192 | return; |
2193 | } | |
2194 | } | |
ae3beac9 | 2195 | INFO("Destroyed rootfs for container \"%s\"", name); |
28272964 | 2196 | |
fa60cd7b CB |
2197 | ret = strnprintf(destroy, sizeof(destroy), "%s/%s", handler->lxcpath, name); |
2198 | if (ret < 0) { | |
ae3beac9 | 2199 | ERROR("Error destroying directory for container \"%s\"", name); |
28272964 CB |
2200 | return; |
2201 | } | |
2202 | ||
f01f7975 CB |
2203 | c = lxc_container_new(name, handler->lxcpath); |
2204 | if (c) { | |
2205 | if (container_disk_lock(c)) { | |
ae3beac9 | 2206 | INFO("Could not update lxc_snapshots file"); |
f01f7975 CB |
2207 | lxc_container_put(c); |
2208 | } else { | |
2209 | mod_all_rdeps(c, false); | |
2210 | container_disk_unlock(c); | |
2211 | lxc_container_put(c); | |
2212 | } | |
2213 | } | |
2214 | ||
d0fbc7ba | 2215 | if (!handler->am_root) |
94b0a3ac CB |
2216 | ret = userns_exec_full(handler->conf, lxc_rmdir_onedev_wrapper, |
2217 | destroy, "lxc_rmdir_onedev_wrapper"); | |
28272964 CB |
2218 | else |
2219 | ret = lxc_rmdir_onedev(destroy, NULL); | |
2220 | ||
2221 | if (ret < 0) { | |
ae3beac9 | 2222 | ERROR("Error destroying directory for container \"%s\"", name); |
28272964 CB |
2223 | return; |
2224 | } | |
ae3beac9 | 2225 | INFO("Destroyed directory for container \"%s\"", name); |
28272964 CB |
2226 | } |
2227 | ||
2228 | static int lxc_rmdir_onedev_wrapper(void *data) | |
2229 | { | |
2230 | char *arg = (char *) data; | |
2231 | return lxc_rmdir_onedev(arg, NULL); | |
2232 | } | |
2233 | ||
134df645 CB |
2234 | static bool do_destroy_container(struct lxc_handler *handler) |
2235 | { | |
94b0a3ac CB |
2236 | int ret; |
2237 | ||
d0fbc7ba | 2238 | if (!handler->am_root) { |
94b0a3ac CB |
2239 | ret = userns_exec_full(handler->conf, storage_destroy_wrapper, |
2240 | handler->conf, "storage_destroy_wrapper"); | |
2241 | if (ret < 0) | |
d028235d | 2242 | return false; |
10bc1861 | 2243 | |
d028235d SG |
2244 | return true; |
2245 | } | |
10bc1861 | 2246 | |
790255cf | 2247 | return storage_destroy(handler->conf); |
28272964 | 2248 | } |