]>
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" | |
c988c8b1 CB |
30 | #include "cgroups/cgroup.h" |
31 | #include "cgroups/cgroup_utils.h" | |
f3787121 | 32 | #include "commands.h" |
bbf5cf35 | 33 | #include "commands_utils.h" |
59eac805 | 34 | #include "compiler.h" |
f3787121 | 35 | #include "conf.h" |
d38dd64a | 36 | #include "config.h" |
74c6e2b0 | 37 | #include "confile_utils.h" |
e2bcd7db | 38 | #include "error.h" |
7fbb15ec | 39 | #include "file_utils.h" |
1cc8bd4b | 40 | #include "list.h" |
f3787121 | 41 | #include "log.h" |
d38dd64a | 42 | #include "lsm/lsm.h" |
f2d5a09d | 43 | #include "lxccontainer.h" |
f3787121 CB |
44 | #include "lxclock.h" |
45 | #include "lxcseccomp.h" | |
3ef9b3d3 | 46 | #include "macro.h" |
f3787121 | 47 | #include "mainloop.h" |
4ffeaf27 | 48 | #include "memory_utils.h" |
63376d7d | 49 | #include "monitor.h" |
f549edcc | 50 | #include "namespace.h" |
811ef482 | 51 | #include "network.h" |
f40988c7 | 52 | #include "process_utils.h" |
f3787121 | 53 | #include "start.h" |
6be5397b CB |
54 | #include "storage/storage.h" |
55 | #include "storage/storage_utils.h" | |
0ed9b1bc | 56 | #include "sync.h" |
59524108 | 57 | #include "syscall_wrappers.h" |
0ed9b1bc CB |
58 | #include "terminal.h" |
59 | #include "utils.h" | |
36eb9bde | 60 | |
d38dd64a CB |
61 | #if HAVE_LIBCAP |
62 | #include <sys/capability.h> | |
63 | #endif | |
64 | ||
9de31d5a CB |
65 | #ifndef HAVE_STRLCPY |
66 | #include "include/strlcpy.h" | |
67 | #endif | |
68 | ||
ac2cecc4 | 69 | lxc_log_define(start, lxc); |
36eb9bde | 70 | |
f01f7975 | 71 | extern void mod_all_rdeps(struct lxc_container *c, bool inc); |
790255cf | 72 | static bool do_destroy_container(struct lxc_handler *handler); |
28272964 CB |
73 | static int lxc_rmdir_onedev_wrapper(void *data); |
74 | static void lxc_destroy_container_on_signal(struct lxc_handler *handler, | |
75 | const char *name); | |
76 | ||
c8154066 SH |
77 | static void print_top_failing_dir(const char *path) |
78 | { | |
dcd9a847 | 79 | __do_free char *copy = NULL; |
9e5f5f2f | 80 | int ret; |
4ffeaf27 | 81 | char *e, *p, saved; |
cbb9c7c7 | 82 | |
4ffeaf27 | 83 | copy = must_copy_string(path); |
c8154066 | 84 | p = copy; |
4ffeaf27 | 85 | e = copy + strlen(path); |
cbb9c7c7 | 86 | |
c8154066 | 87 | while (p < e) { |
f3787121 CB |
88 | while (p < e && *p == '/') |
89 | p++; | |
9e5f5f2f | 90 | |
f3787121 CB |
91 | while (p < e && *p != '/') |
92 | p++; | |
9e5f5f2f | 93 | |
c8154066 SH |
94 | saved = *p; |
95 | *p = '\0'; | |
9e5f5f2f CB |
96 | |
97 | ret = access(copy, X_OK); | |
98 | if (ret != 0) { | |
818a57fc | 99 | SYSERROR("Could not access %s. Please grant it x access, or add an ACL for the container " "root", copy); |
c8154066 SH |
100 | return; |
101 | } | |
102 | *p = saved; | |
103 | } | |
104 | } | |
105 | ||
5032bf39 | 106 | static void lxc_put_nsfds(struct lxc_handler *handler) |
408da065 | 107 | { |
818a57fc | 108 | for (int i = 0; i < LXC_NS_MAX; i++) { |
5032bf39 | 109 | if (handler->nsfd[i] < 0) |
39cd919c CB |
110 | continue; |
111 | ||
818a57fc | 112 | close_prot_errno_disarm(handler->nsfd[i]); |
9f30a190 | 113 | } |
9f30a190 MM |
114 | } |
115 | ||
8db6be1b CB |
116 | static int lxc_try_preserve_namespace(struct lxc_handler *handler, |
117 | lxc_namespace_t idx, const char *ns) | |
4cb53844 | 118 | { |
8db6be1b CB |
119 | __do_close int fd = -EBADF; |
120 | int ret; | |
4cb53844 | 121 | |
8db6be1b | 122 | fd = lxc_preserve_ns(handler->pid, ns); |
857ba1f0 CB |
123 | if (fd < 0) |
124 | return -errno; | |
4cb53844 | 125 | |
8db6be1b CB |
126 | ret = strnprintf(handler->nsfd_paths[idx], |
127 | sizeof(handler->nsfd_paths[idx]), "%s:/proc/%d/fd/%d", | |
128 | ns_info[idx].proc_name, handler->monitor_pid, fd); | |
cb3b010c CB |
129 | if (ret < 0) |
130 | return ret_errno(EIO); | |
8db6be1b | 131 | |
cb3b010c CB |
132 | /* |
133 | * In case LXC is configured for exposing information to hooks as | |
134 | * argv-style arguments prepare an argv array we can use. | |
135 | */ | |
8db6be1b CB |
136 | handler->hook_argv[handler->hook_argc] = handler->nsfd_paths[idx]; |
137 | handler->hook_argc++; | |
8db6be1b CB |
138 | |
139 | DEBUG("Preserved %s namespace via fd %d and stashed path as %s", | |
140 | ns_info[idx].proc_name, fd, handler->nsfd_paths[idx]); | |
141 | ||
142 | handler->nsfd[idx] = move_fd(fd); | |
143 | return 0; | |
4cb53844 CB |
144 | } |
145 | ||
146 | /* lxc_try_preserve_namespaces: open /proc/@pid/ns/@ns for each namespace | |
147 | * specified in ns_clone_flags. | |
4d8ac866 | 148 | * Return true on success, false on failure. |
62d05d9b | 149 | */ |
4cb53844 | 150 | static bool lxc_try_preserve_namespaces(struct lxc_handler *handler, |
8db6be1b | 151 | int ns_clone_flags) |
f3787121 | 152 | { |
8db6be1b CB |
153 | for (lxc_namespace_t ns_idx = 0; ns_idx < LXC_NS_MAX; ns_idx++) |
154 | handler->nsfd[ns_idx] = -EBADF; | |
9f30a190 | 155 | |
8db6be1b CB |
156 | for (lxc_namespace_t ns_idx = 0; ns_idx < LXC_NS_MAX; ns_idx++) { |
157 | int ret; | |
857ba1f0 | 158 | const char *ns = ns_info[ns_idx].proc_name; |
4cb53844 | 159 | |
8db6be1b | 160 | if ((ns_clone_flags & ns_info[ns_idx].clone_flag) == 0) |
9f30a190 | 161 | continue; |
28d9e29e | 162 | |
8db6be1b CB |
163 | ret = lxc_try_preserve_namespace(handler, ns_idx, |
164 | ns_info[ns_idx].proc_name); | |
165 | if (ret < 0) { | |
857ba1f0 CB |
166 | if (ret == -ENOENT) { |
167 | SYSERROR("Kernel does not support preserving %s namespaces", ns); | |
168 | continue; | |
169 | } | |
170 | ||
9beaca55 CB |
171 | /* |
172 | * Handle kernels that do not support interacting with | |
173 | * namespaces through procfs. | |
4cb53844 | 174 | */ |
4cb53844 | 175 | lxc_put_nsfds(handler); |
857ba1f0 | 176 | return log_error_errno(false, errno, "Failed to preserve %s namespace", ns); |
4cb53844 | 177 | } |
9f30a190 MM |
178 | } |
179 | ||
62d05d9b | 180 | return true; |
9f30a190 MM |
181 | } |
182 | ||
ec1dc633 | 183 | static inline bool match_stdfds(int fd) |
80090207 | 184 | { |
ec1dc633 | 185 | return (fd == STDIN_FILENO || fd == STDOUT_FILENO || fd == STDERR_FILENO); |
80090207 CLG |
186 | } |
187 | ||
4c03724e | 188 | #ifdef HAVE_DLOG |
189 | static bool match_dlog_fds(struct dirent *direntp) | |
190 | { | |
191 | char path[PATH_MAX] = {0}; | |
192 | char link[PATH_MAX] = {0}; | |
193 | ssize_t linklen; | |
194 | int ret; | |
195 | ||
fa60cd7b CB |
196 | ret = strnprintf(path, sizeof(path), "/proc/self/fd/%s", direntp->d_name); |
197 | if (ret < 0) | |
818a57fc | 198 | return log_error(false, "Failed to create file descriptor name"); |
4c03724e | 199 | |
200 | linklen = readlink(path, link, PATH_MAX); | |
818a57fc CB |
201 | if (linklen < 0) |
202 | return log_error(false, "Failed to read link path - \"%s\"", path); | |
203 | else if (linklen >= PATH_MAX) | |
204 | return log_error(false, "The name of link path is too long - \"%s\"", path); | |
205 | ||
d0269705 CB |
206 | if (strequal(link, "/dev/log_main") || |
207 | strequal(link, "/dev/log_system") || | |
208 | strequal(link, "/dev/log_radio")) | |
4c03724e | 209 | return true; |
210 | ||
211 | return false; | |
212 | } | |
213 | #endif | |
214 | ||
46abf219 RJ |
215 | /* Parses the LISTEN_FDS environment variable value. |
216 | * The returned value is the highest fd number up to which the | |
217 | * file descriptors must be passed to the container process. | |
218 | * | |
219 | * For example, if LISTEN_FDS=2 then 4 is returned and file descriptors 3 and 4 | |
220 | * MUST be passed to the container process (in addition to the standard streams) | |
221 | * to support [socket activation][systemd-listen-fds]. | |
222 | */ | |
223 | static unsigned int get_listen_fds_max(void) | |
224 | { | |
225 | int ret; | |
226 | unsigned int num_fds; | |
227 | const char *val; | |
228 | ||
229 | val = getenv("LISTEN_FDS"); | |
230 | if (!val) | |
231 | return 0; | |
232 | ||
233 | ret = lxc_safe_uint(val, &num_fds); | |
234 | if (ret < 0) | |
235 | return syserror_ret(0, "Failed to parse \"LISTEN_FDS=%s\" environment variable", val); | |
236 | ||
237 | return log_trace(num_fds, "Parsed \"LISTEN_FDS=%s\" environment variable", val); | |
238 | } | |
239 | ||
47a46cf1 CB |
240 | int lxc_check_inherited(struct lxc_conf *conf, bool closeall, |
241 | int *fds_to_ignore, size_t len_fds) | |
80090207 | 242 | { |
80090207 | 243 | int fd, fddir; |
47a46cf1 | 244 | size_t i; |
80090207 | 245 | DIR *dir; |
a5a70219 | 246 | struct dirent *direntp; |
46abf219 | 247 | unsigned int listen_fds_max; |
80090207 | 248 | |
d2cf4c37 SH |
249 | if (conf && conf->close_all_fds) |
250 | closeall = true; | |
251 | ||
46abf219 RJ |
252 | listen_fds_max = get_listen_fds_max(); |
253 | ||
4e2d6b9a CB |
254 | /* |
255 | * Disable syslog at this point to avoid the above logging | |
256 | * function to open a new fd and make the check_inherited function | |
257 | * enter an infinite loop. | |
258 | */ | |
259 | lxc_log_syslog_disable(); | |
260 | ||
b119f362 | 261 | restart: |
80090207 | 262 | dir = opendir("/proc/self/fd"); |
818a57fc CB |
263 | if (!dir) |
264 | return log_warn(-1, "Failed to open directory"); | |
80090207 CLG |
265 | |
266 | fddir = dirfd(dir); | |
267 | ||
74f96976 | 268 | while ((direntp = readdir(dir))) { |
a5a70219 | 269 | int ret; |
d39b10eb CB |
270 | struct lxc_list *cur; |
271 | bool matched = false; | |
272 | ||
d0269705 | 273 | if (strequal(direntp->d_name, ".")) |
80090207 CLG |
274 | continue; |
275 | ||
d0269705 | 276 | if (strequal(direntp->d_name, "..")) |
80090207 CLG |
277 | continue; |
278 | ||
a5a70219 CB |
279 | ret = lxc_safe_int(direntp->d_name, &fd); |
280 | if (ret < 0) { | |
281 | INFO("Could not parse file descriptor for \"%s\"", direntp->d_name); | |
d4cff0d2 CB |
282 | continue; |
283 | } | |
80090207 | 284 | |
47a46cf1 CB |
285 | for (i = 0; i < len_fds; i++) |
286 | if (fds_to_ignore[i] == fd) | |
287 | break; | |
288 | ||
289 | if (fd == fddir || fd == lxc_log_fd || | |
290 | (i < len_fds && fd == fds_to_ignore[i])) | |
80090207 CLG |
291 | continue; |
292 | ||
d39b10eb | 293 | /* Keep state clients that wait on reboots. */ |
3ee9e4fb CB |
294 | if (conf) { |
295 | lxc_list_for_each(cur, &conf->state_clients) { | |
296 | struct lxc_state_client *client = cur->elem; | |
d39b10eb | 297 | |
3ee9e4fb CB |
298 | if (client->clientfd != fd) |
299 | continue; | |
d39b10eb | 300 | |
3ee9e4fb CB |
301 | matched = true; |
302 | break; | |
303 | } | |
d39b10eb CB |
304 | } |
305 | ||
306 | if (matched) | |
307 | continue; | |
308 | ||
858377e4 SH |
309 | if (current_config && fd == current_config->logfd) |
310 | continue; | |
311 | ||
ec1dc633 | 312 | if (match_stdfds(fd)) |
80090207 | 313 | continue; |
80090207 | 314 | |
1d5e5f26 | 315 | #ifdef HAVE_DLOG |
316 | if (match_dlog_fds(direntp)) | |
317 | continue; | |
318 | ||
319 | #endif | |
46abf219 RJ |
320 | |
321 | if (fd <= listen_fds_max) { | |
322 | INFO("Inheriting fd %d (using the LISTEN_FDS environment variable)", fd); | |
323 | continue; | |
324 | } | |
325 | ||
d2cf4c37 | 326 | if (closeall) { |
4e2d6b9a CB |
327 | if (close(fd)) |
328 | SYSINFO("Closed inherited fd %d", fd); | |
329 | else | |
330 | INFO("Closed inherited fd %d", fd); | |
b119f362 | 331 | closedir(dir); |
b119f362 SH |
332 | goto restart; |
333 | } | |
28d9e29e | 334 | WARN("Inherited fd %d", fd); |
80090207 | 335 | } |
4e2d6b9a | 336 | closedir(dir); |
80090207 | 337 | |
4e2d6b9a CB |
338 | /* |
339 | * Only enable syslog at this point to avoid the above logging | |
340 | * function to open a new fd and make the check_inherited function | |
341 | * enter an infinite loop. | |
64c57ea1 | 342 | */ |
4e2d6b9a | 343 | lxc_log_syslog_enable(); |
64c57ea1 | 344 | |
92c7f629 | 345 | return 0; |
80090207 CLG |
346 | } |
347 | ||
83ee7875 | 348 | static int setup_signal_fd(sigset_t *oldmask) |
b0a33c1e | 349 | { |
7288dfb6 | 350 | int ret; |
b0a33c1e | 351 | sigset_t mask; |
7288dfb6 | 352 | const int signals[] = {SIGBUS, SIGILL, SIGSEGV, SIGWINCH}; |
b0a33c1e | 353 | |
408da065 | 354 | /* Block everything except serious error signals. */ |
df0795b1 CB |
355 | ret = sigfillset(&mask); |
356 | if (ret < 0) | |
357 | return -EBADF; | |
358 | ||
9b10bef5 | 359 | for (int sig = 0; sig < (sizeof(signals) / sizeof(signals[0])); sig++) { |
df0795b1 CB |
360 | ret = sigdelset(&mask, signals[sig]); |
361 | if (ret < 0) | |
362 | return -EBADF; | |
b0a33c1e | 363 | } |
364 | ||
b467714b | 365 | ret = pthread_sigmask(SIG_BLOCK, &mask, oldmask); |
818a57fc CB |
366 | if (ret < 0) |
367 | return log_error_errno(-EBADF, errno, | |
368 | "Failed to set signal mask"); | |
b0a33c1e | 369 | |
df0795b1 | 370 | ret = signalfd(-1, &mask, SFD_CLOEXEC); |
818a57fc CB |
371 | if (ret < 0) |
372 | return log_error_errno(-EBADF, | |
373 | errno, "Failed to create signal file descriptor"); | |
b0a33c1e | 374 | |
df0795b1 | 375 | TRACE("Created signal file descriptor %d", ret); |
1ac470c0 | 376 | |
df0795b1 | 377 | return ret; |
b0a33c1e | 378 | } |
379 | ||
84c92abd | 380 | static int signal_handler(int fd, uint32_t events, void *data, |
3298b37d | 381 | struct lxc_async_descr *descr) |
b0a33c1e | 382 | { |
15cd25fd | 383 | int ret; |
3c319edb CB |
384 | siginfo_t info; |
385 | struct signalfd_siginfo siginfo; | |
386 | struct lxc_handler *hdlr = data; | |
15cd25fd | 387 | |
489f39be | 388 | ret = lxc_read_nointr(fd, &siginfo, sizeof(siginfo)); |
818a57fc CB |
389 | if (ret < 0) |
390 | return log_error(LXC_MAINLOOP_ERROR, "Failed to read signal info from signal file descriptor %d", fd); | |
15cd25fd | 391 | |
818a57fc | 392 | if (ret != sizeof(siginfo)) |
f7a97743 | 393 | return log_error(LXC_MAINLOOP_ERROR, "Unexpected size for struct signalfd_siginfo"); |
15cd25fd | 394 | |
408da065 | 395 | /* Check whether init is running. */ |
80507ee8 | 396 | info.si_pid = 0; |
3c319edb CB |
397 | ret = waitid(P_PID, hdlr->pid, &info, WEXITED | WNOWAIT | WNOHANG); |
398 | if (ret == 0 && info.si_pid == hdlr->pid) | |
399 | hdlr->init_died = true; | |
80507ee8 | 400 | |
543d2f83 CB |
401 | TRACE("Received signal ssi_signo(%d) for ssi_pid(%d), si_signo(%d), si_pid(%d)", |
402 | siginfo.ssi_signo, siginfo.ssi_pid, info.si_signo, info.si_pid); | |
403 | ||
cd5177e9 TA |
404 | /* Try to figure out a reasonable exit status to report. */ |
405 | if (hdlr->init_died) { | |
406 | switch (info.si_code) { | |
407 | case CLD_EXITED: | |
408 | hdlr->exit_status = info.si_status << 8; | |
409 | break; | |
410 | case CLD_KILLED: | |
411 | case CLD_DUMPED: | |
412 | case CLD_STOPPED: | |
413 | hdlr->exit_status = info.si_status << 8 | 0x7f; | |
414 | break; | |
415 | case CLD_CONTINUED: | |
416 | /* Huh? The waitid() told us it's dead *and* continued? */ | |
417 | WARN("Init %d dead and continued?", hdlr->pid); | |
418 | hdlr->exit_status = 1; | |
419 | break; | |
420 | default: | |
20993a97 | 421 | ERROR("Unknown si_code: %d", info.si_code); |
55e1311e | 422 | hdlr->exit_status = 1; |
cd5177e9 TA |
423 | } |
424 | } | |
425 | ||
d4b5d7a8 | 426 | if (siginfo.ssi_signo == SIGHUP) { |
33942046 CB |
427 | if (hdlr->pidfd >= 0) |
428 | lxc_raw_pidfd_send_signal(hdlr->pidfd, SIGTERM, NULL, 0); | |
d9bb2fba CB |
429 | else |
430 | kill(hdlr->pid, SIGTERM); | |
d4b5d7a8 | 431 | INFO("Killing %d since terminal hung up", hdlr->pid); |
9b10bef5 CB |
432 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE |
433 | : LXC_MAINLOOP_CONTINUE; | |
d4b5d7a8 TA |
434 | } |
435 | ||
9cb94384 | 436 | if (siginfo.ssi_signo != SIGCHLD) { |
33942046 CB |
437 | if (hdlr->pidfd >= 0) |
438 | lxc_raw_pidfd_send_signal(hdlr->pidfd, | |
439 | siginfo.ssi_signo, NULL, 0); | |
d9bb2fba CB |
440 | else |
441 | kill(hdlr->pid, siginfo.ssi_signo); | |
9cb94384 | 442 | INFO("Forwarded signal %d to pid %d", siginfo.ssi_signo, hdlr->pid); |
9b10bef5 CB |
443 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE |
444 | : LXC_MAINLOOP_CONTINUE; | |
9cb94384 TA |
445 | } |
446 | ||
3a9e949f TA |
447 | /* More robustness, protect ourself from a SIGCHLD sent |
448 | * by a process different from the container init. | |
449 | */ | |
450 | if (siginfo.ssi_pid != hdlr->pid) { | |
6e94162a CB |
451 | NOTICE("Received %d from pid %d instead of container init %d", |
452 | siginfo.ssi_signo, siginfo.ssi_pid, hdlr->pid); | |
9b10bef5 CB |
453 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE |
454 | : LXC_MAINLOOP_CONTINUE; | |
3a9e949f TA |
455 | } |
456 | ||
408da065 | 457 | if (siginfo.ssi_code == CLD_STOPPED) { |
6e94162a | 458 | INFO("Container init process was stopped"); |
9b10bef5 CB |
459 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE |
460 | : LXC_MAINLOOP_CONTINUE; | |
461 | } | |
462 | ||
463 | if (siginfo.ssi_code == CLD_CONTINUED) { | |
6e94162a | 464 | INFO("Container init process was continued"); |
9b10bef5 CB |
465 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE |
466 | : LXC_MAINLOOP_CONTINUE; | |
15cd25fd | 467 | } |
1ac470c0 | 468 | |
818a57fc | 469 | return log_debug(LXC_MAINLOOP_CLOSE, "Container init process %d exited", hdlr->pid); |
b0a33c1e | 470 | } |
471 | ||
974a8aba CB |
472 | int lxc_serve_state_clients(const char *name, struct lxc_handler *handler, |
473 | lxc_state_t state) | |
66aeffc7 | 474 | { |
9de31d5a | 475 | size_t retlen; |
dbc9832d CB |
476 | ssize_t ret; |
477 | struct lxc_list *cur, *next; | |
dbc9832d CB |
478 | struct lxc_msg msg = {.type = lxc_msg_state, .value = state}; |
479 | ||
974a8aba CB |
480 | if (state == THAWED) |
481 | handler->state = RUNNING; | |
482 | else | |
483 | handler->state = state; | |
484 | ||
9dfa4041 | 485 | TRACE("Set container state to %s", lxc_state2str(state)); |
dbc9832d | 486 | |
818a57fc CB |
487 | if (lxc_list_empty(&handler->conf->state_clients)) |
488 | return log_trace(0, "No state clients registered"); | |
dbc9832d | 489 | |
9de31d5a CB |
490 | retlen = strlcpy(msg.name, name, sizeof(msg.name)); |
491 | if (retlen >= sizeof(msg.name)) | |
492 | return -E2BIG; | |
dbc9832d | 493 | |
d39b10eb | 494 | lxc_list_for_each_safe(cur, &handler->conf->state_clients, next) { |
d85617bc | 495 | struct lxc_state_client *client = cur->elem; |
dbc9832d | 496 | |
d39b10eb | 497 | if (client->states[state] == 0) { |
9dfa4041 | 498 | TRACE("State %s not registered for state client %d", |
dbc9832d CB |
499 | lxc_state2str(state), client->clientfd); |
500 | continue; | |
501 | } | |
502 | ||
9dfa4041 | 503 | TRACE("Sending state %s to state client %d", |
dbc9832d CB |
504 | lxc_state2str(state), client->clientfd); |
505 | ||
7fbb15ec CB |
506 | ret = lxc_send_nointr(client->clientfd, &msg, sizeof(msg), MSG_NOSIGNAL); |
507 | if (ret <= 0) | |
6d1400b5 | 508 | SYSERROR("Failed to send message to client"); |
dbc9832d CB |
509 | |
510 | /* kick client from list */ | |
dbc9832d | 511 | lxc_list_del(cur); |
300d1cb4 | 512 | close(client->clientfd); |
dbc9832d CB |
513 | free(cur->elem); |
514 | free(cur); | |
515 | } | |
dbc9832d | 516 | |
5e5576a4 CB |
517 | return 0; |
518 | } | |
519 | ||
520 | static int lxc_serve_state_socket_pair(const char *name, | |
521 | struct lxc_handler *handler, | |
522 | lxc_state_t state) | |
523 | { | |
524 | ssize_t ret; | |
525 | ||
bb955810 | 526 | if (!handler->daemonize || |
5e5576a4 CB |
527 | handler->state_socket_pair[1] < 0 || |
528 | state == STARTING) | |
529 | return 0; | |
530 | ||
531 | /* Close read end of the socket pair. */ | |
a783a414 | 532 | close_prot_errno_disarm(handler->state_socket_pair[0]); |
5e5576a4 | 533 | |
ee8377bd | 534 | again: |
5e5576a4 CB |
535 | ret = lxc_abstract_unix_send_credential(handler->state_socket_pair[1], |
536 | &(int){state}, sizeof(int)); | |
9044b79e | 537 | if (ret < 0) { |
538 | SYSERROR("Failed to send state to %d", handler->state_socket_pair[1]); | |
539 | ||
ee8377bd CB |
540 | if (errno == EINTR) |
541 | goto again; | |
9044b79e | 542 | |
543 | return -1; | |
544 | } | |
545 | ||
818a57fc CB |
546 | if (ret != sizeof(int)) |
547 | return log_error(-1, "Message too long : %d", handler->state_socket_pair[1]); | |
5e5576a4 CB |
548 | |
549 | TRACE("Sent container state \"%s\" to %d", lxc_state2str(state), | |
550 | handler->state_socket_pair[1]); | |
551 | ||
552 | /* Close write end of the socket pair. */ | |
818a57fc | 553 | close_prot_errno_disarm(handler->state_socket_pair[1]); |
5e5576a4 CB |
554 | |
555 | return 0; | |
556 | } | |
557 | ||
558 | int lxc_set_state(const char *name, struct lxc_handler *handler, | |
559 | lxc_state_t state) | |
560 | { | |
561 | int ret; | |
562 | ||
563 | ret = lxc_serve_state_socket_pair(name, handler, state); | |
818a57fc CB |
564 | if (ret < 0) |
565 | return log_error(-1, "Failed to synchronize via anonymous pair of unix sockets"); | |
5e5576a4 CB |
566 | |
567 | ret = lxc_serve_state_clients(name, handler, state); | |
568 | if (ret < 0) | |
569 | return -1; | |
570 | ||
dbc9832d CB |
571 | /* This function will try to connect to the legacy lxc-monitord state |
572 | * server and only exists for backwards compatibility. | |
573 | */ | |
9123e471 | 574 | lxc_monitor_send_state(name, state, handler->lxcpath); |
dbc9832d | 575 | |
66aeffc7 DL |
576 | return 0; |
577 | } | |
578 | ||
735f2c6e | 579 | int lxc_poll(const char *name, struct lxc_handler *handler) |
b0a33c1e | 580 | { |
3c319edb | 581 | int ret; |
543d2f83 | 582 | struct lxc_terminal *console = &handler->conf->console; |
3298b37d | 583 | struct lxc_async_descr descr, descr_console; |
b0a33c1e | 584 | |
543d2f83 CB |
585 | if (!wants_console(console)) |
586 | console = NULL; | |
86530b0a | 587 | |
3c319edb CB |
588 | ret = lxc_mainloop_open(&descr); |
589 | if (ret < 0) { | |
590 | ERROR("Failed to create mainloop"); | |
50c8bf05 | 591 | goto out_sigfd; |
b0a33c1e | 592 | } |
593 | ||
543d2f83 | 594 | if (console) { |
30a33fbd CB |
595 | ret = lxc_mainloop_open(&descr_console); |
596 | if (ret < 0) { | |
597 | ERROR("Failed to create console mainloop"); | |
598 | goto out_mainloop; | |
599 | } | |
3c319edb CB |
600 | } |
601 | ||
543d2f83 CB |
602 | ret = lxc_mainloop_add_handler(&descr, handler->sigfd, |
603 | signal_handler, | |
604 | default_cleanup_handler, | |
605 | handler, "signal_handler"); | |
3c319edb | 606 | if (ret < 0) { |
1cc8bd4b | 607 | ERROR("Failed to add signal handler for %d to mainloop", handler->sigfd); |
3c319edb | 608 | goto out_mainloop_console; |
b0a33c1e | 609 | } |
610 | ||
2ac0f627 CB |
611 | ret = lxc_seccomp_setup_proxy(&handler->conf->seccomp, &descr, handler); |
612 | if (ret < 0) { | |
613 | ERROR("Failed to setup seccomp proxy"); | |
c3e3c21a | 614 | goto out_mainloop_console; |
2ac0f627 | 615 | } |
cdb2a47f | 616 | |
543d2f83 | 617 | if (console) { |
093bce5f | 618 | ret = lxc_terminal_mainloop_add(&descr, console); |
30a33fbd CB |
619 | if (ret < 0) { |
620 | ERROR("Failed to add console handlers to mainloop"); | |
621 | goto out_mainloop_console; | |
622 | } | |
563f2f2c DL |
623 | } |
624 | ||
3c319edb CB |
625 | ret = lxc_cmd_mainloop_add(name, &descr, handler); |
626 | if (ret < 0) { | |
627 | ERROR("Failed to add command handler to mainloop"); | |
628 | goto out_mainloop_console; | |
629 | } | |
b0a33c1e | 630 | |
3c319edb | 631 | TRACE("Mainloop is ready"); |
b0a33c1e | 632 | |
3c319edb | 633 | ret = lxc_mainloop(&descr, -1); |
a783a414 | 634 | close_prot_errno_disarm(descr.epfd); |
1cc8bd4b | 635 | if (ret < 0 || !handler->init_died) |
42b09f94 | 636 | goto out_mainloop_console; |
1cc8bd4b | 637 | |
543d2f83 CB |
638 | if (console) { |
639 | ret = lxc_terminal_mainloop_add(&descr_console, console); | |
640 | if (ret == 0) | |
641 | ret = lxc_mainloop(&descr_console, 0); | |
642 | } | |
30a33fbd | 643 | |
3c319edb | 644 | out_mainloop_console: |
543d2f83 | 645 | if (console) { |
30a33fbd CB |
646 | lxc_mainloop_close(&descr_console); |
647 | TRACE("Closed console mainloop"); | |
648 | } | |
3c319edb | 649 | |
42b09f94 CB |
650 | out_mainloop: |
651 | lxc_mainloop_close(&descr); | |
652 | TRACE("Closed mainloop"); | |
653 | ||
b0a33c1e | 654 | out_sigfd: |
1cc8bd4b | 655 | TRACE("Closed signal file descriptor %d", handler->sigfd); |
a783a414 | 656 | close_prot_errno_disarm(handler->sigfd); |
408da065 | 657 | |
1cc8bd4b | 658 | return ret; |
b0a33c1e | 659 | } |
660 | ||
a42abcce | 661 | void lxc_put_handler(struct lxc_handler *handler) |
f2e07cb6 | 662 | { |
a783a414 CB |
663 | close_prot_errno_disarm(handler->pidfd); |
664 | close_prot_errno_disarm(handler->sigfd); | |
1cc8bd4b | 665 | lxc_put_nsfds(handler); |
80308d07 | 666 | if (handler->conf && handler->conf->reboot == REBOOT_NONE) |
a783a414 CB |
667 | close_prot_errno_disarm(handler->conf->maincmd_fd); |
668 | close_prot_errno_disarm(handler->monitor_status_fd); | |
669 | close_prot_errno_disarm(handler->state_socket_pair[0]); | |
670 | close_prot_errno_disarm(handler->state_socket_pair[1]); | |
671 | cgroup_exit(handler->cgroup_ops); | |
a42abcce CB |
672 | if (handler->conf && handler->conf->reboot == REBOOT_NONE) |
673 | free_disarm(handler); | |
674 | else | |
675 | handler->conf = NULL; | |
f2e07cb6 CB |
676 | } |
677 | ||
a42abcce CB |
678 | struct lxc_handler *lxc_init_handler(struct lxc_handler *old, |
679 | const char *name, struct lxc_conf *conf, | |
5e5576a4 | 680 | const char *lxcpath, bool daemonize) |
59eb99ba | 681 | { |
85c279bb | 682 | int nr_keep_fds = 0; |
818a57fc | 683 | int ret; |
3a0f472d DL |
684 | struct lxc_handler *handler; |
685 | ||
a42abcce CB |
686 | if (!old) |
687 | handler = zalloc(sizeof(*handler)); | |
688 | else | |
689 | handler = old; | |
fdecbc9c | 690 | if (!handler) |
3a0f472d | 691 | return NULL; |
9044b79e | 692 | |
fdecbc9c CB |
693 | /* Note that am_guest_unpriv() checks the effective uid. We |
694 | * probably don't care if we are real root only if we are running | |
695 | * as root so this should be fine. | |
790255cf | 696 | */ |
e0010464 | 697 | handler->am_root = !am_guest_unpriv(); |
fae349da | 698 | handler->conf = conf; |
9123e471 | 699 | handler->lxcpath = lxcpath; |
818a57fc CB |
700 | handler->init_died = false; |
701 | handler->data_sock[0] = -EBADF; | |
702 | handler->data_sock[1] = -EBADF; | |
4d8bdfa0 | 703 | handler->monitor_status_fd = -EBADF; |
33942046 | 704 | handler->pidfd = -EBADF; |
1cc8bd4b | 705 | handler->sigfd = -EBADF; |
818a57fc CB |
706 | handler->state_socket_pair[0] = -EBADF; |
707 | handler->state_socket_pair[1] = -EBADF; | |
80308d07 | 708 | if (handler->conf->reboot == REBOOT_NONE) |
d39b10eb | 709 | lxc_list_init(&handler->conf->state_clients); |
fae349da | 710 | |
8db6be1b CB |
711 | for (lxc_namespace_t idx = 0; idx < LXC_NS_MAX; idx++) { |
712 | handler->nsfd[idx] = -EBADF; | |
713 | ||
714 | if (handler->conf->reboot == REBOOT_NONE) | |
715 | continue; | |
716 | ||
717 | handler->nsfd_paths[idx][0] = '\0'; | |
718 | handler->hook_argv[idx] = NULL; | |
719 | ||
720 | if (handler->hook_argc != 0) | |
721 | handler->hook_argc = 0; | |
722 | } | |
b6b2b194 | 723 | |
f0ecc19d | 724 | handler->name = name; |
c581d2a6 CB |
725 | if (daemonize) |
726 | handler->transient_pid = lxc_raw_getpid(); | |
727 | else | |
728 | handler->transient_pid = -1; | |
3bdf52d7 | 729 | |
80308d07 | 730 | if (daemonize && handler->conf->reboot == REBOOT_NONE) { |
5e5576a4 | 731 | /* Create socketpair() to synchronize on daemonized startup. |
fdecbc9c CB |
732 | * When the container reboots we don't need to synchronize |
733 | * again currently so don't open another socketpair(). | |
5e5576a4 CB |
734 | */ |
735 | ret = socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0, | |
736 | handler->state_socket_pair); | |
737 | if (ret < 0) { | |
738 | ERROR("Failed to create anonymous pair of unix sockets"); | |
739 | goto on_error; | |
740 | } | |
9044b79e | 741 | |
5e5576a4 CB |
742 | TRACE("Created anonymous pair {%d,%d} of unix sockets", |
743 | handler->state_socket_pair[0], | |
744 | handler->state_socket_pair[1]); | |
85c279bb CB |
745 | handler->keep_fds[nr_keep_fds++] = handler->state_socket_pair[0]; |
746 | handler->keep_fds[nr_keep_fds++] = handler->state_socket_pair[1]; | |
5e5576a4 CB |
747 | } |
748 | ||
80308d07 | 749 | if (handler->conf->reboot == REBOOT_NONE) { |
6834acff | 750 | handler->conf->maincmd_fd = lxc_server_init(name, lxcpath, "command"); |
d39b10eb CB |
751 | if (handler->conf->maincmd_fd < 0) { |
752 | ERROR("Failed to set up command socket"); | |
753 | goto on_error; | |
754 | } | |
85c279bb | 755 | handler->keep_fds[nr_keep_fds++] = handler->conf->maincmd_fd; |
aa460476 | 756 | } |
9044b79e | 757 | |
9dfa4041 | 758 | TRACE("Unix domain socket %d for command server is ready", |
aa460476 CB |
759 | handler->conf->maincmd_fd); |
760 | ||
761 | return handler; | |
762 | ||
f2e07cb6 | 763 | on_error: |
a42abcce | 764 | lxc_put_handler(handler); |
aa460476 CB |
765 | |
766 | return NULL; | |
767 | } | |
768 | ||
769 | int lxc_init(const char *name, struct lxc_handler *handler) | |
770 | { | |
f62cf1d4 | 771 | __do_close int status_fd = -EBADF; |
a2c09be0 | 772 | int ret; |
4a03ded4 | 773 | const char *loglevel; |
aa460476 CB |
774 | struct lxc_conf *conf = handler->conf; |
775 | ||
434c8e15 | 776 | handler->monitor_pid = lxc_raw_getpid(); |
4d8bdfa0 | 777 | status_fd = open("/proc/self/status", O_RDONLY | O_CLOEXEC); |
59b1b67e CB |
778 | if (status_fd < 0) |
779 | return log_error_errno(-1, errno, "Failed to open monitor status fd"); | |
434c8e15 | 780 | |
4eb19ac0 | 781 | handler->lsm_ops = lsm_init_static(); |
2170b263 | 782 | TRACE("Initialized LSM"); |
d2e30e99 | 783 | |
408da065 | 784 | /* Begin by setting the state to STARTING. */ |
2170b263 | 785 | ret = lxc_set_state(name, handler, STARTING); |
59b1b67e CB |
786 | if (ret < 0) |
787 | return log_error(-1, "Failed to set state to \"%s\"", lxc_state2str(STARTING)); | |
46800e77 | 788 | TRACE("Set container state to \"STARTING\""); |
0ad19a3f | 789 | |
408da065 | 790 | /* Start of environment variable setup for hooks. */ |
24c2a48d CB |
791 | ret = setenv("LXC_NAME", name, 1); |
792 | if (ret < 0) | |
793 | SYSERROR("Failed to set environment variable: LXC_NAME=%s", name); | |
408da065 | 794 | |
2170b263 CB |
795 | if (conf->rcfile) { |
796 | ret = setenv("LXC_CONFIG_FILE", conf->rcfile, 1); | |
797 | if (ret < 0) | |
818a57fc | 798 | SYSERROR("Failed to set environment variable: LXC_CONFIG_FILE=%s", conf->rcfile); |
2170b263 | 799 | } |
408da065 | 800 | |
2170b263 CB |
801 | if (conf->rootfs.mount) { |
802 | ret = setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1); | |
803 | if (ret < 0) | |
818a57fc | 804 | SYSERROR("Failed to set environment variable: LXC_ROOTFS_MOUNT=%s", conf->rootfs.mount); |
2170b263 | 805 | } |
408da065 | 806 | |
2170b263 CB |
807 | if (conf->rootfs.path) { |
808 | ret = setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1); | |
809 | if (ret < 0) | |
818a57fc | 810 | SYSERROR("Failed to set environment variable: LXC_ROOTFS_PATH=%s", conf->rootfs.path); |
2170b263 | 811 | } |
408da065 | 812 | |
2170b263 CB |
813 | if (conf->console.path) { |
814 | ret = setenv("LXC_CONSOLE", conf->console.path, 1); | |
815 | if (ret < 0) | |
818a57fc | 816 | SYSERROR("Failed to set environment variable: LXC_CONSOLE=%s", conf->console.path); |
2170b263 | 817 | } |
408da065 | 818 | |
2170b263 CB |
819 | if (conf->console.log_path) { |
820 | ret = setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1); | |
821 | if (ret < 0) | |
818a57fc | 822 | SYSERROR("Failed to set environment variable: LXC_CONSOLE_LOGPATH=%s", conf->console.log_path); |
2170b263 | 823 | } |
408da065 | 824 | |
2170b263 CB |
825 | if (cgns_supported()) { |
826 | ret = setenv("LXC_CGNS_AWARE", "1", 1); | |
827 | if (ret < 0) | |
818a57fc | 828 | SYSERROR("Failed to set environment variable LXC_CGNS_AWARE=1"); |
2170b263 | 829 | } |
b8f88d9b | 830 | |
4a03ded4 | 831 | loglevel = lxc_log_priority_to_string(lxc_log_get_level()); |
2170b263 CB |
832 | ret = setenv("LXC_LOG_LEVEL", loglevel, 1); |
833 | if (ret < 0) | |
818a57fc | 834 | SYSERROR("Set environment variable LXC_LOG_LEVEL=%s", loglevel); |
a2c09be0 CB |
835 | |
836 | if (conf->hooks_version == 0) | |
837 | ret = setenv("LXC_HOOK_VERSION", "0", 1); | |
838 | else | |
839 | ret = setenv("LXC_HOOK_VERSION", "1", 1); | |
840 | if (ret < 0) | |
841 | SYSERROR("Failed to set environment variable LXC_HOOK_VERSION=%u", conf->hooks_version); | |
408da065 | 842 | /* End of environment variable setup for hooks. */ |
f7bee6c6 | 843 | |
2170b263 | 844 | TRACE("Set environment variables"); |
dbc9832d | 845 | |
2170b263 | 846 | ret = run_lxc_hooks(name, "pre-start", conf, NULL); |
65a2dc8b CB |
847 | if (ret < 0) |
848 | return log_error(-1, "Failed to run lxc.hook.pre-start for container \"%s\"", name); | |
2170b263 | 849 | TRACE("Ran pre-start hooks"); |
26ddeedd | 850 | |
9f77617b CB |
851 | ret = lxc_terminal_parent(conf); |
852 | if (ret < 0) | |
853 | return log_error(-1, "Failed to allocate terminal"); | |
854 | ||
408da065 CB |
855 | /* The signal fd has to be created before forking otherwise if the child |
856 | * process exits before we setup the signal fd, the event will be lost | |
857 | * and the command will be stuck. | |
858 | */ | |
83ee7875 | 859 | handler->sigfd = setup_signal_fd(&handler->oldmask); |
65a2dc8b CB |
860 | if (handler->sigfd < 0) |
861 | return log_error(-1, "Failed to setup SIGCHLD fd handler."); | |
2170b263 | 862 | TRACE("Set up signal fd"); |
b5159817 | 863 | |
5a087e05 | 864 | handler->cgroup_ops = cgroup_init(handler->conf); |
2202afc9 CB |
865 | if (!handler->cgroup_ops) { |
866 | ERROR("Failed to initialize cgroup driver"); | |
9f77617b | 867 | goto out_restore_sigmask; |
2202afc9 CB |
868 | } |
869 | TRACE("Initialized cgroup driver"); | |
870 | ||
2e1361a6 | 871 | ret = lxc_read_seccomp_config(conf); |
9f77617b CB |
872 | if (ret < 0) { |
873 | ERROR("Failed to read seccomp policy"); | |
874 | goto out_restore_sigmask; | |
875 | } | |
2e1361a6 CB |
876 | TRACE("Read seccomp policy"); |
877 | ||
af04d847 | 878 | ret = handler->lsm_ops->prepare(handler->lsm_ops, conf, handler->lxcpath); |
1800f924 WB |
879 | if (ret < 0) { |
880 | ERROR("Failed to initialize LSM"); | |
9f77617b | 881 | goto out_restore_sigmask; |
1800f924 WB |
882 | } |
883 | TRACE("Initialized LSM"); | |
884 | ||
2170b263 | 885 | INFO("Container \"%s\" is initialized", name); |
4d8bdfa0 | 886 | handler->monitor_status_fd = move_fd(status_fd); |
aa460476 | 887 | return 0; |
59eb99ba | 888 | |
b5159817 | 889 | out_restore_sigmask: |
4c496daa | 890 | (void)pthread_sigmask(SIG_SETMASK, &handler->oldmask, NULL); |
ea918412 | 891 | |
aa460476 | 892 | return -1; |
59eb99ba DL |
893 | } |
894 | ||
8db6be1b CB |
895 | void lxc_expose_namespace_environment(const struct lxc_handler *handler) |
896 | { | |
897 | for (lxc_namespace_t i = 0; i < LXC_NS_MAX; i++) { | |
898 | int ret; | |
899 | const char *fd_path; | |
900 | ||
901 | if (handler->nsfd[i] < 0) | |
902 | continue; | |
903 | ||
904 | fd_path = handler->nsfd_paths[i] + strcspn(handler->nsfd_paths[i], "/"); | |
905 | ret = setenv(ns_info[i].env_name, fd_path, 1); | |
906 | if (ret < 0) | |
907 | SYSERROR("Failed to set environment variable %s=%s", | |
908 | ns_info[i].env_name, fd_path); | |
909 | else | |
910 | TRACE("Set environment variable %s=%s", | |
911 | ns_info[i].env_name, fd_path); | |
912 | } | |
913 | } | |
914 | ||
fd5be714 | 915 | void lxc_end(struct lxc_handler *handler) |
59eb99ba | 916 | { |
818a57fc | 917 | int ret; |
dbc9832d | 918 | struct lxc_list *cur, *next; |
0c5859ff | 919 | const char *name = handler->name; |
2202afc9 | 920 | struct cgroup_ops *cgroup_ops = handler->cgroup_ops; |
b6b2b194 | 921 | |
408da065 CB |
922 | /* The STOPPING state is there for future cleanup code which can take |
923 | * awhile. | |
59eb99ba | 924 | */ |
25c2aca5 | 925 | lxc_set_state(name, handler, STOPPING); |
b6b2b194 | 926 | |
8db6be1b CB |
927 | /* Passing information to hooks via environment variables. */ |
928 | if (handler->conf->hooks_version > 0) | |
929 | lxc_expose_namespace_environment(handler); | |
c154af98 | 930 | |
80308d07 | 931 | if (handler->conf->reboot > REBOOT_NONE) { |
c097467f CB |
932 | ret = setenv("LXC_TARGET", "reboot", 1); |
933 | if (ret < 0) | |
818a57fc | 934 | SYSERROR("Failed to set environment variable: LXC_TARGET=reboot"); |
c097467f | 935 | } |
408da065 | 936 | |
80308d07 | 937 | if (handler->conf->reboot == REBOOT_NONE) { |
c097467f CB |
938 | ret = setenv("LXC_TARGET", "stop", 1); |
939 | if (ret < 0) | |
818a57fc | 940 | SYSERROR("Failed to set environment variable: LXC_TARGET=stop"); |
c097467f | 941 | } |
c154af98 | 942 | |
18b3b9c1 | 943 | if (handler->conf->hooks_version == 0) |
8db6be1b | 944 | ret = run_lxc_hooks(name, "stop", handler->conf, handler->hook_argv); |
18b3b9c1 | 945 | else |
c097467f CB |
946 | ret = run_lxc_hooks(name, "stop", handler->conf, NULL); |
947 | if (ret < 0) | |
948 | ERROR("Failed to run \"lxc.hook.stop\" hook"); | |
c154af98 | 949 | |
af04d847 | 950 | handler->lsm_ops->cleanup(handler->lsm_ops, handler->conf, handler->lxcpath); |
1800f924 | 951 | |
e2aed383 WB |
952 | if (cgroup_ops) { |
953 | cgroup_ops->payload_destroy(cgroup_ops, handler); | |
954 | cgroup_ops->monitor_destroy(cgroup_ops, handler); | |
955 | } | |
4288b79f | 956 | |
79ff643d CB |
957 | put_lxc_rootfs(&handler->conf->rootfs, true); |
958 | ||
80308d07 | 959 | if (handler->conf->reboot == REBOOT_NONE) { |
f893d898 CB |
960 | /* For all new state clients simply close the command socket. |
961 | * This will inform all state clients that the container is | |
962 | * STOPPED and also prevents a race between a open()/close() on | |
963 | * the command socket causing a new process to get ECONNREFUSED | |
964 | * because we haven't yet closed the command socket. | |
965 | */ | |
b87ee312 | 966 | close_prot_errno_disarm(handler->conf->maincmd_fd); |
c3184275 CB |
967 | TRACE("Closed command socket"); |
968 | ||
969 | /* This function will try to connect to the legacy lxc-monitord | |
970 | * state server and only exists for backwards compatibility. | |
971 | */ | |
972 | lxc_monitor_send_state(name, STOPPED, handler->lxcpath); | |
973 | ||
974 | /* The command socket is closed so no one can acces the command | |
975 | * socket anymore so there's no need to lock it. | |
976 | */ | |
977 | handler->state = STOPPED; | |
978 | TRACE("Set container state to \"STOPPED\""); | |
f893d898 CB |
979 | } else { |
980 | lxc_set_state(name, handler, STOPPED); | |
695d0e56 | 981 | TRACE("Set container state to \"STOPPED\""); |
d39b10eb | 982 | } |
4288b79f | 983 | |
048493a3 CB |
984 | /* Avoid lingering namespace references. */ |
985 | lxc_put_nsfds(handler); | |
986 | ||
46800e77 CB |
987 | ret = run_lxc_hooks(name, "post-stop", handler->conf, NULL); |
988 | if (ret < 0) { | |
c097467f | 989 | ERROR("Failed to run lxc.hook.post-stop for container \"%s\"", name); |
80308d07 | 990 | if (handler->conf->reboot > REBOOT_NONE) { |
c097467f | 991 | WARN("Container will be stopped instead of rebooted"); |
80308d07 | 992 | handler->conf->reboot = REBOOT_NONE; |
c097467f CB |
993 | |
994 | ret = setenv("LXC_TARGET", "stop", 1); | |
995 | if (ret < 0) | |
818a57fc | 996 | WARN("Failed to set environment variable: LXC_TARGET=stop"); |
f3787121 CB |
997 | } |
998 | } | |
26ddeedd | 999 | |
408da065 | 1000 | /* Reset mask set by setup_signal_fd. */ |
b467714b | 1001 | ret = pthread_sigmask(SIG_SETMASK, &handler->oldmask, NULL); |
c097467f | 1002 | if (ret < 0) |
a24c5678 | 1003 | SYSWARN("Failed to restore signal mask"); |
8f64a3f6 | 1004 | |
2aac071b | 1005 | lxc_terminal_delete(&handler->conf->console); |
0e4be3cf | 1006 | lxc_delete_tty(&handler->conf->ttys); |
f797f05e | 1007 | close_prot_errno_disarm(handler->conf->devpts_fd); |
dbc9832d | 1008 | |
dbc9832d CB |
1009 | /* The command socket is now closed, no more state clients can register |
1010 | * themselves from now on. So free the list of state clients. | |
1011 | */ | |
d39b10eb CB |
1012 | lxc_list_for_each_safe(cur, &handler->conf->state_clients, next) { |
1013 | struct lxc_state_client *client = cur->elem; | |
1014 | ||
1015 | /* Keep state clients that want to be notified about reboots. */ | |
80308d07 CB |
1016 | if ((handler->conf->reboot > REBOOT_NONE) && |
1017 | (client->states[RUNNING] == 2)) | |
d39b10eb CB |
1018 | continue; |
1019 | ||
dbc9832d | 1020 | /* close state client socket */ |
dbc9832d | 1021 | lxc_list_del(cur); |
c097467f | 1022 | close(client->clientfd); |
dbc9832d CB |
1023 | free(cur->elem); |
1024 | free(cur); | |
1025 | } | |
1026 | ||
80308d07 | 1027 | if (handler->conf->ephemeral == 1 && handler->conf->reboot != REBOOT_REQ) |
28272964 | 1028 | lxc_destroy_container_on_signal(handler, name); |
2c5f2ede | 1029 | |
a42abcce | 1030 | lxc_put_handler(handler); |
59eb99ba DL |
1031 | } |
1032 | ||
0c5859ff | 1033 | void lxc_abort(struct lxc_handler *handler) |
59eb99ba | 1034 | { |
33942046 CB |
1035 | int ret = 0; |
1036 | int status; | |
73e608b2 | 1037 | |
0c5859ff | 1038 | lxc_set_state(handler->name, handler, ABORTING); |
0e4f9d51 | 1039 | |
c718fac1 | 1040 | if (handler->pidfd >= 0) { |
33942046 | 1041 | ret = lxc_raw_pidfd_send_signal(handler->pidfd, SIGKILL, NULL, 0); |
c718fac1 | 1042 | if (ret) |
11c7d349 CB |
1043 | SYSWARN("Failed to send SIGKILL via pidfd %d for process %d", |
1044 | handler->pidfd, handler->pid); | |
c718fac1 CB |
1045 | } |
1046 | ||
65a2dc8b | 1047 | if ((!ret || errno != ESRCH) && handler->pid > 0) |
11c7d349 CB |
1048 | if (kill(handler->pid, SIGKILL)) |
1049 | SYSWARN("Failed to send SIGKILL to %d", handler->pid); | |
0e4f9d51 | 1050 | |
33942046 CB |
1051 | do { |
1052 | ret = waitpid(-1, &status, 0); | |
1053 | } while (ret > 0); | |
59eb99ba DL |
1054 | } |
1055 | ||
ffe1e01a | 1056 | static int do_start(void *data) |
50e98013 | 1057 | { |
cdb2a47f | 1058 | struct lxc_handler *handler = data; |
f62cf1d4 | 1059 | __lxc_unused __do_close int data_sock0 = handler->data_sock[0], |
9ff57a59 | 1060 | data_sock1 = handler->data_sock[1]; |
d3103162 | 1061 | __do_close int devnull_fd = -EBADF, status_fd = -EBADF; |
8bf3abfb | 1062 | int ret; |
928b1f04 YT |
1063 | uid_t new_uid; |
1064 | gid_t new_gid; | |
e96a536c | 1065 | struct lxc_list *iterator; |
964581c2 CB |
1066 | uid_t nsuid = 0; |
1067 | gid_t nsgid = 0; | |
50e98013 | 1068 | |
18225d19 | 1069 | lxc_sync_fini_parent(handler); |
50e98013 | 1070 | |
d17c815d | 1071 | if (lxc_abstract_unix_recv_one_fd(data_sock1, &status_fd, NULL, 0) < 0) { |
77093b7c | 1072 | ERROR("Failed to receive status file descriptor from parent process"); |
4d8bdfa0 CB |
1073 | goto out_warn_father; |
1074 | } | |
1075 | ||
408da065 CB |
1076 | /* This prctl must be before the synchro, so if the parent dies before |
1077 | * we set the parent death signal, we will detect its death with the | |
1078 | * synchro right after, otherwise we have a window where the parent can | |
1079 | * exit before we set the pdeath signal leading to a unsupervized | |
1080 | * container. | |
743ecd2e | 1081 | */ |
4d8bdfa0 | 1082 | ret = lxc_set_death_signal(SIGKILL, handler->monitor_pid, status_fd); |
912314fc CB |
1083 | if (ret < 0) { |
1084 | SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL"); | |
18225d19 | 1085 | goto out_warn_father; |
743ecd2e DL |
1086 | } |
1087 | ||
611ddd34 CB |
1088 | ret = lxc_ambient_caps_up(); |
1089 | if (ret < 0) { | |
6f5e532f | 1090 | ERROR("Failed to raise ambient capabilities"); |
611ddd34 CB |
1091 | goto out_warn_father; |
1092 | } | |
1093 | ||
b467714b | 1094 | ret = pthread_sigmask(SIG_SETMASK, &handler->oldmask, NULL); |
18225d19 CB |
1095 | if (ret < 0) { |
1096 | SYSERROR("Failed to set signal mask"); | |
1097 | goto out_warn_father; | |
1098 | } | |
50e98013 | 1099 | |
6e48e7c5 | 1100 | if (!lxc_sync_wait_parent(handler, START_SYNC_STARTUP)) |
18225d19 | 1101 | goto out_warn_father; |
5b1e83cb | 1102 | |
408da065 CB |
1103 | /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See |
1104 | * https://github.com/lxc/lxd/issues/1978. | |
1105 | */ | |
f7176c3e | 1106 | if (handler->ns_unshare_flags & CLONE_NEWNET) { |
5b1e83cb SH |
1107 | ret = unshare(CLONE_NEWNET); |
1108 | if (ret < 0) { | |
e96a536c | 1109 | SYSERROR("Failed to unshare CLONE_NEWNET"); |
5b1e83cb SH |
1110 | goto out_warn_father; |
1111 | } | |
e96a536c | 1112 | INFO("Unshared CLONE_NEWNET"); |
5b1e83cb SH |
1113 | } |
1114 | ||
408da065 | 1115 | /* If we are in a new user namespace, become root there to have |
d08f8d2f | 1116 | * privilege over our namespace. |
f6d3e3e4 SH |
1117 | */ |
1118 | if (!lxc_list_empty(&handler->conf->id_map)) { | |
964581c2 CB |
1119 | if (!handler->conf->root_nsuid_map) |
1120 | nsuid = handler->conf->init_uid; | |
1121 | ||
1122 | if (!handler->conf->root_nsgid_map) | |
1123 | nsgid = handler->conf->init_gid; | |
4160c3a0 | 1124 | |
d08f8d2f CB |
1125 | /* Drop groups only after we switched to a valid gid in the new |
1126 | * user namespace. | |
1127 | */ | |
8917c382 | 1128 | if (!lxc_drop_groups() && |
8af07f82 | 1129 | (handler->am_root || errno != EPERM)) |
c476bdce | 1130 | goto out_warn_father; |
4b826b1f | 1131 | |
b58214ac CB |
1132 | if (!lxc_switch_uid_gid(nsuid, nsgid)) |
1133 | goto out_warn_father; | |
1134 | ||
b81689a1 CB |
1135 | ret = prctl(PR_SET_DUMPABLE, prctl_arg(1), prctl_arg(0), |
1136 | prctl_arg(0), prctl_arg(0)); | |
d7883725 CB |
1137 | if (ret < 0) |
1138 | goto out_warn_father; | |
912314fc CB |
1139 | |
1140 | /* set{g,u}id() clears deathsignal */ | |
4d8bdfa0 | 1141 | ret = lxc_set_death_signal(SIGKILL, handler->monitor_pid, status_fd); |
912314fc CB |
1142 | if (ret < 0) { |
1143 | SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL"); | |
1144 | goto out_warn_father; | |
1145 | } | |
f6d3e3e4 SH |
1146 | } |
1147 | ||
e96a536c CB |
1148 | ret = access(handler->lxcpath, X_OK); |
1149 | if (ret != 0) { | |
c8154066 SH |
1150 | print_top_failing_dir(handler->lxcpath); |
1151 | goto out_warn_father; | |
1152 | } | |
1153 | ||
7a55c157 TA |
1154 | /* In order to checkpoint restore, we need to have everything in the |
1155 | * same mount namespace. However, some containers may not have a | |
1156 | * reasonable /dev (in particular, they may not have /dev/null), so we | |
1157 | * can't set init's std fds to /dev/null by opening it from inside the | |
1158 | * container. | |
1159 | * | |
1160 | * If that's the case, fall back to using the host's /dev/null. This | |
1161 | * means that migration won't work, but at least we won't spew output | |
1162 | * where it isn't wanted. | |
1163 | */ | |
bb955810 | 1164 | if (handler->daemonize && !handler->conf->autodev) { |
2806a87d | 1165 | char path[PATH_MAX]; |
3c09b97c | 1166 | |
fa60cd7b CB |
1167 | ret = strnprintf(path, sizeof(path), "%s/dev/null", |
1168 | handler->conf->rootfs.mount); | |
1169 | if (ret < 0) | |
2806a87d | 1170 | goto out_warn_father; |
3c09b97c | 1171 | |
e96a536c CB |
1172 | ret = access(path, F_OK); |
1173 | if (ret != 0) { | |
1174 | devnull_fd = open_devnull(); | |
c44de748 | 1175 | |
e96a536c CB |
1176 | if (devnull_fd < 0) |
1177 | goto out_warn_father; | |
818a57fc | 1178 | WARN("Using /dev/null from the host for container init's standard file descriptors. Migration will not work"); |
e96a536c | 1179 | } |
c44de748 AM |
1180 | } |
1181 | ||
6bc4165d CB |
1182 | /* |
1183 | * Tell the parent task it can begin to configure the container and wait | |
1184 | * for it to finish. | |
1185 | */ | |
1186 | if (!lxc_sync_wake_parent(handler, START_SYNC_CONFIGURE)) | |
c44de748 | 1187 | goto out_error; |
544a48a0 | 1188 | |
deefdf8a CB |
1189 | /* Unshare cgroup namespace after we have setup our cgroups. If we do it |
1190 | * earlier we end up with a wrong view of /proc/self/cgroup. For | |
1191 | * example, assume we unshare(CLONE_NEWCGROUP) first, and then create | |
1192 | * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then | |
1193 | * /proc/self/cgroup would show us: | |
1194 | * | |
1195 | * 8:cpuset:/lxc/c | |
1196 | * | |
1197 | * whereas it should actually show | |
1198 | * | |
1199 | * 8:cpuset:/ | |
1200 | */ | |
f7176c3e | 1201 | if (handler->ns_unshare_flags & CLONE_NEWCGROUP) { |
e96a536c CB |
1202 | ret = unshare(CLONE_NEWCGROUP); |
1203 | if (ret < 0) { | |
bca7c59c CB |
1204 | if (errno != EINVAL) { |
1205 | SYSERROR("Failed to unshare CLONE_NEWCGROUP"); | |
1206 | goto out_warn_father; | |
1207 | } | |
1208 | ||
1209 | handler->ns_clone_flags &= ~CLONE_NEWCGROUP; | |
1210 | SYSINFO("Kernel does not support CLONE_NEWCGROUP"); | |
1211 | } else { | |
1212 | INFO("Unshared CLONE_NEWCGROUP"); | |
deefdf8a | 1213 | } |
deefdf8a CB |
1214 | } |
1215 | ||
f7176c3e | 1216 | if (handler->ns_unshare_flags & CLONE_NEWTIME) { |
70fd7fc9 CB |
1217 | ret = unshare(CLONE_NEWTIME); |
1218 | if (ret < 0) { | |
1219 | if (errno != EINVAL) { | |
1220 | SYSERROR("Failed to unshare CLONE_NEWTIME"); | |
1221 | goto out_warn_father; | |
1222 | } | |
1223 | ||
1224 | handler->ns_clone_flags &= ~CLONE_NEWTIME; | |
1225 | SYSINFO("Kernel does not support CLONE_NEWTIME"); | |
1226 | } else { | |
1227 | __do_close int timens_fd = -EBADF; | |
1228 | ||
1229 | INFO("Unshared CLONE_NEWTIME"); | |
1230 | ||
1231 | if (handler->conf->timens.s_boot) | |
1232 | ret = timens_offset_write(CLOCK_BOOTTIME, handler->conf->timens.s_boot, 0); | |
1233 | else if (handler->conf->timens.ns_boot) | |
1234 | ret = timens_offset_write(CLOCK_BOOTTIME, 0, handler->conf->timens.ns_boot); | |
1235 | if (ret) { | |
1236 | SYSERROR("Failed to write CLONE_BOOTTIME offset"); | |
1237 | goto out_warn_father; | |
1238 | } | |
1239 | TRACE("Wrote CLOCK_BOOTTIME offset"); | |
1240 | ||
1241 | if (handler->conf->timens.s_monotonic) | |
1242 | ret = timens_offset_write(CLOCK_MONOTONIC, handler->conf->timens.s_monotonic, 0); | |
1243 | else if (handler->conf->timens.ns_monotonic) | |
1244 | ret = timens_offset_write(CLOCK_MONOTONIC, 0, handler->conf->timens.ns_monotonic); | |
1245 | if (ret) { | |
1246 | SYSERROR("Failed to write CLONE_MONOTONIC offset"); | |
1247 | goto out_warn_father; | |
1248 | } | |
1249 | TRACE("Wrote CLOCK_MONOTONIC offset"); | |
1250 | ||
1251 | timens_fd = open("/proc/self/ns/time_for_children", O_RDONLY | O_CLOEXEC); | |
1252 | if (timens_fd < 0) { | |
1253 | SYSERROR("Failed to open \"/proc/self/ns/time_for_children\""); | |
1254 | goto out_warn_father; | |
1255 | } | |
1256 | ||
1257 | ret = setns(timens_fd, CLONE_NEWTIME); | |
cfe6f435 | 1258 | if (ret) { |
70fd7fc9 CB |
1259 | SYSERROR("Failed to setns(%d(\"/proc/self/ns/time_for_children\"))", timens_fd); |
1260 | goto out_warn_father; | |
1261 | } | |
1262 | } | |
1263 | } | |
1264 | ||
149857af CB |
1265 | /* Add the requested environment variables to the current environment to |
1266 | * allow them to be used by the various hooks, such as the start hook | |
317494f1 | 1267 | * below. |
98ff08ca | 1268 | */ |
98ff08ca | 1269 | lxc_list_for_each(iterator, &handler->conf->environment) { |
e96a536c CB |
1270 | ret = putenv((char *)iterator->elem); |
1271 | if (ret < 0) { | |
1272 | SYSERROR("Failed to set environment variable: %s", | |
1273 | (char *)iterator->elem); | |
98ff08ca CB |
1274 | goto out_warn_father; |
1275 | } | |
1276 | } | |
1277 | ||
6bc4165d CB |
1278 | if (!lxc_sync_wait_parent(handler, START_SYNC_POST_CONFIGURE)) |
1279 | goto out_warn_father; | |
1280 | ||
f4152036 | 1281 | /* Setup the container, ip, names, utsname, ... */ |
3b988b33 | 1282 | ret = lxc_setup(handler); |
7729f8e5 | 1283 | if (ret < 0) { |
e96a536c | 1284 | ERROR("Failed to setup container \"%s\"", handler->name); |
f4152036 CB |
1285 | goto out_warn_father; |
1286 | } | |
1287 | ||
408da065 | 1288 | /* Set the label to change to when we exec(2) the container's init. */ |
af04d847 | 1289 | ret = handler->lsm_ops->process_label_set(handler->lsm_ops, NULL, handler->conf, true); |
e96a536c | 1290 | if (ret < 0) |
e075f5d9 | 1291 | goto out_warn_father; |
5112cd70 | 1292 | |
029cdff5 | 1293 | /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it |
408da065 CB |
1294 | * before we aren't allowed anymore. |
1295 | */ | |
029cdff5 | 1296 | if (handler->conf->no_new_privs) { |
b81689a1 CB |
1297 | ret = prctl(PR_SET_NO_NEW_PRIVS, prctl_arg(1), prctl_arg(0), |
1298 | prctl_arg(0), prctl_arg(0)); | |
e96a536c | 1299 | if (ret < 0) { |
818a57fc | 1300 | SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges"); |
029cdff5 CB |
1301 | goto out_warn_father; |
1302 | } | |
818a57fc | 1303 | DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges"); |
029cdff5 CB |
1304 | } |
1305 | ||
408da065 | 1306 | /* If we mounted a temporary proc, then unmount it now. */ |
5112cd70 | 1307 | tmp_proc_unmount(handler->conf); |
e075f5d9 | 1308 | |
e96a536c CB |
1309 | ret = lxc_seccomp_load(handler->conf); |
1310 | if (ret < 0) | |
8f2c3a70 SH |
1311 | goto out_warn_father; |
1312 | ||
e96a536c CB |
1313 | ret = run_lxc_hooks(handler->name, "start", handler->conf, NULL); |
1314 | if (ret < 0) { | |
1315 | ERROR("Failed to run lxc.hook.start for container \"%s\"", | |
1316 | handler->name); | |
773fb9ca SH |
1317 | goto out_warn_father; |
1318 | } | |
fc25b815 | 1319 | |
a783a414 | 1320 | close_prot_errno_disarm(handler->sigfd); |
26ddeedd | 1321 | |
41808e20 | 1322 | if (handler->conf->console.pty < 0 && handler->daemonize) { |
f4c177c3 CB |
1323 | if (devnull_fd < 0) { |
1324 | devnull_fd = open_devnull(); | |
1325 | if (devnull_fd < 0) | |
1326 | goto out_warn_father; | |
1327 | } | |
1328 | ||
e96a536c CB |
1329 | ret = set_stdfds(devnull_fd); |
1330 | if (ret < 0) { | |
1331 | ERROR("Failed to redirect std{in,out,err} to \"/dev/null\""); | |
c5b93afb LF |
1332 | goto out_warn_father; |
1333 | } | |
e96a536c | 1334 | } |
507cee36 | 1335 | |
a783a414 | 1336 | close_prot_errno_disarm(devnull_fd); |
c44de748 | 1337 | |
8c9a7665 TA |
1338 | setsid(); |
1339 | ||
e96a536c CB |
1340 | if (handler->conf->init_cwd) { |
1341 | ret = chdir(handler->conf->init_cwd); | |
1342 | if (ret < 0) { | |
1343 | SYSERROR("Could not change directory to \"%s\"", | |
1344 | handler->conf->init_cwd); | |
1345 | goto out_warn_father; | |
1346 | } | |
3c491553 L |
1347 | } |
1348 | ||
6e48e7c5 | 1349 | if (!lxc_sync_barrier_parent(handler, START_SYNC_CGROUP_LIMITS)) |
f4152036 CB |
1350 | goto out_warn_father; |
1351 | ||
111ed96e | 1352 | ret = lxc_sync_fds_child(handler); |
1b82d721 | 1353 | if (ret < 0) { |
111ed96e | 1354 | SYSERROR("Failed to sync file descriptors with parent"); |
1b82d721 CB |
1355 | goto out_warn_father; |
1356 | } | |
1357 | ||
1b82d721 CB |
1358 | if (!lxc_sync_wait_parent(handler, START_SYNC_READY_START)) |
1359 | goto out_warn_father; | |
1360 | ||
149857af CB |
1361 | /* Reset the environment variables the user requested in a clear |
1362 | * environment. | |
1363 | */ | |
e96a536c CB |
1364 | ret = clearenv(); |
1365 | /* Don't error out though. */ | |
1366 | if (ret < 0) | |
149857af | 1367 | SYSERROR("Failed to clear environment."); |
149857af CB |
1368 | |
1369 | lxc_list_for_each(iterator, &handler->conf->environment) { | |
e96a536c CB |
1370 | ret = putenv((char *)iterator->elem); |
1371 | if (ret < 0) { | |
1372 | SYSERROR("Failed to set environment variable: %s", | |
1373 | (char *)iterator->elem); | |
149857af CB |
1374 | goto out_warn_father; |
1375 | } | |
1376 | } | |
1377 | ||
e96a536c CB |
1378 | ret = putenv("container=lxc"); |
1379 | if (ret < 0) { | |
1380 | SYSERROR("Failed to set environment variable: container=lxc"); | |
98ff08ca CB |
1381 | goto out_warn_father; |
1382 | } | |
1383 | ||
885766f5 CB |
1384 | if (handler->conf->ttys.tty_names) { |
1385 | ret = putenv(handler->conf->ttys.tty_names); | |
e96a536c CB |
1386 | if (ret < 0) { |
1387 | SYSERROR("Failed to set environment variable for container ptys"); | |
98ff08ca CB |
1388 | goto out_warn_father; |
1389 | } | |
1390 | } | |
1391 | ||
76bdf299 CB |
1392 | /* The container has been setup. We can now switch to an unprivileged |
1393 | * uid/gid. | |
1394 | */ | |
928b1f04 YT |
1395 | new_uid = handler->conf->init_uid; |
1396 | new_gid = handler->conf->init_gid; | |
76bdf299 | 1397 | |
964581c2 CB |
1398 | /* Avoid unnecessary syscalls. */ |
1399 | if (new_uid == nsuid) | |
1400 | new_uid = LXC_INVALID_UID; | |
1401 | ||
1402 | if (new_gid == nsgid) | |
1403 | new_gid = LXC_INVALID_GID; | |
c353b0b9 | 1404 | |
6aff5157 | 1405 | /* Make sure that the processes STDIO is correctly owned by the user that we are switching to */ |
c353b0b9 CB |
1406 | ret = fix_stdio_permissions(new_uid); |
1407 | if (ret) | |
1408 | WARN("Failed to ajust stdio permissions"); | |
964581c2 | 1409 | |
8af07f82 CB |
1410 | /* If we are in a new user namespace we already dropped all groups when |
1411 | * we switched to root in the new user namespace further above. Only | |
1412 | * drop groups if we can, so ensure that we have necessary privilege. | |
1413 | */ | |
bf31b337 | 1414 | if (lxc_list_empty(&handler->conf->id_map)) { |
8af07f82 CB |
1415 | #if HAVE_LIBCAP |
1416 | if (lxc_proc_cap_is_set(CAP_SETGID, CAP_EFFECTIVE)) | |
1417 | #endif | |
bf31b337 RJ |
1418 | { |
1419 | if (handler->conf->init_groups.size > 0) { | |
1420 | if (!lxc_setgroups(handler->conf->init_groups.list, | |
1421 | handler->conf->init_groups.size)) | |
1422 | goto out_warn_father; | |
1423 | } else { | |
1424 | if (!lxc_drop_groups()) | |
1425 | goto out_warn_father; | |
1426 | } | |
1427 | } | |
1428 | } | |
8af07f82 | 1429 | |
b58214ac CB |
1430 | if (!lxc_switch_uid_gid(new_uid, new_gid)) |
1431 | goto out_warn_father; | |
1432 | ||
611ddd34 CB |
1433 | ret = lxc_ambient_caps_down(); |
1434 | if (ret < 0) { | |
6f5e532f | 1435 | ERROR("Failed to clear ambient capabilities"); |
611ddd34 CB |
1436 | goto out_warn_father; |
1437 | } | |
1438 | ||
258f8051 | 1439 | if (handler->conf->monitor_signal_pdeath != SIGKILL) { |
4d8bdfa0 CB |
1440 | ret = lxc_set_death_signal(handler->conf->monitor_signal_pdeath, |
1441 | handler->monitor_pid, status_fd); | |
258f8051 CB |
1442 | if (ret < 0) { |
1443 | SYSERROR("Failed to set PR_SET_PDEATHSIG to %d", | |
1444 | handler->conf->monitor_signal_pdeath); | |
1445 | goto out_warn_father; | |
1446 | } | |
1447 | } | |
1448 | ||
d3103162 CB |
1449 | /* |
1450 | * After this call, we are in error because this ops should not return | |
408da065 CB |
1451 | * as it execs. |
1452 | */ | |
c4ea60df | 1453 | handler->ops->start(handler, handler->data); |
50e98013 DL |
1454 | |
1455 | out_warn_father: | |
d3103162 CB |
1456 | /* |
1457 | * We want the parent to know something went wrong, so we return a | |
408da065 CB |
1458 | * special error code. |
1459 | */ | |
946a0c6d | 1460 | lxc_sync_wake_parent(handler, SYNC_ERROR); |
c44de748 AM |
1461 | |
1462 | out_error: | |
50e98013 DL |
1463 | return -1; |
1464 | } | |
1465 | ||
5af9369b | 1466 | int resolve_clone_flags(struct lxc_handler *handler) |
f813849c | 1467 | { |
8bc8c715 CB |
1468 | int i; |
1469 | struct lxc_conf *conf = handler->conf; | |
70fd7fc9 CB |
1470 | bool wants_timens = conf->timens.s_boot || conf->timens.ns_boot || |
1471 | conf->timens.s_monotonic || conf->timens.ns_monotonic; | |
f813849c | 1472 | |
8bc8c715 | 1473 | for (i = 0; i < LXC_NS_MAX; i++) { |
f7a0c6ee CB |
1474 | if (conf->ns_keep) { |
1475 | if (!(conf->ns_keep & ns_info[i].clone_flag)) | |
becad0ec | 1476 | handler->ns_clone_flags |= ns_info[i].clone_flag; |
f7a0c6ee CB |
1477 | } else if (conf->ns_clone) { |
1478 | if ((conf->ns_clone & ns_info[i].clone_flag)) | |
becad0ec | 1479 | handler->ns_clone_flags |= ns_info[i].clone_flag; |
8bc8c715 CB |
1480 | } else { |
1481 | if (i == LXC_NS_USER && lxc_list_empty(&handler->conf->id_map)) | |
1482 | continue; | |
1483 | ||
1484 | if (i == LXC_NS_NET && lxc_requests_empty_network(handler)) | |
1485 | continue; | |
1486 | ||
7bd05339 CB |
1487 | if (i == LXC_NS_CGROUP && !cgns_supported()) |
1488 | continue; | |
1489 | ||
70fd7fc9 CB |
1490 | if (i == LXC_NS_TIME && !wants_timens) |
1491 | continue; | |
1492 | ||
becad0ec | 1493 | handler->ns_clone_flags |= ns_info[i].clone_flag; |
8bc8c715 | 1494 | } |
03df7ab5 | 1495 | |
8bc8c715 CB |
1496 | if (!conf->ns_share[i]) |
1497 | continue; | |
5af9369b | 1498 | |
becad0ec | 1499 | handler->ns_clone_flags &= ~ns_info[i].clone_flag; |
8bc8c715 | 1500 | TRACE("Sharing %s namespace", ns_info[i].proc_name); |
5af9369b CB |
1501 | } |
1502 | ||
70fd7fc9 CB |
1503 | if (wants_timens && (conf->ns_keep & ns_info[LXC_NS_TIME].clone_flag)) |
1504 | return log_trace_errno(-1, EINVAL, "Requested to keep time namespace while also specifying offsets"); | |
1505 | ||
f7176c3e CB |
1506 | /* Deal with namespaces that are unshared. */ |
1507 | if (handler->ns_clone_flags & CLONE_NEWTIME) | |
1508 | handler->ns_unshare_flags |= CLONE_NEWTIME; | |
1509 | ||
1510 | if (!pure_unified_layout(handler->cgroup_ops) && handler->ns_clone_flags & CLONE_NEWCGROUP) | |
1511 | handler->ns_unshare_flags |= CLONE_NEWCGROUP; | |
1512 | ||
1513 | if ((handler->ns_clone_flags & (CLONE_NEWNET | CLONE_NEWUSER)) == | |
1514 | (CLONE_NEWNET | CLONE_NEWUSER)) | |
1515 | handler->ns_unshare_flags |= CLONE_NEWNET; | |
1516 | ||
1517 | /* Deal with namespaces that are spawned. */ | |
1518 | handler->ns_on_clone_flags = handler->ns_clone_flags & ~handler->ns_unshare_flags; | |
1519 | ||
1520 | handler->clone_flags = handler->ns_on_clone_flags | CLONE_PIDFD; | |
1521 | ||
5af9369b | 1522 | return 0; |
f813849c TA |
1523 | } |
1524 | ||
8deca6c9 CB |
1525 | /* Note that this function is used with clone(CLONE_VM). Some glibc versions |
1526 | * used to reset the pid/tid to -1 when CLONE_VM was used without CLONE_THREAD. | |
1527 | * But since the memory between parent and child is shared on CLONE_VM this | |
1528 | * would invalidate the getpid() cache that glibc used to maintain and so | |
1529 | * getpid() in the child would return the parent's pid. This is all fixed in | |
1530 | * newer glibc versions where the getpid() cache is removed and the pid/tid is | |
1531 | * not reset anymore. | |
b35ddc5b | 1532 | * However, if for whatever reason you - dear committer - somehow need to get the |
8de0119d CB |
1533 | * pid of the placeholder intermediate process for do_share_ns() you need to |
1534 | * call lxc_raw_getpid(). The next lxc_raw_clone() call does not employ | |
1535 | * CLONE_VM and will be fine. | |
8deca6c9 CB |
1536 | */ |
1537 | static inline int do_share_ns(void *arg) | |
fa3a5b22 | 1538 | { |
8deca6c9 | 1539 | int i, flags, ret; |
fa3a5b22 CB |
1540 | struct lxc_handler *handler = arg; |
1541 | ||
8deca6c9 CB |
1542 | for (i = 0; i < LXC_NS_MAX; i++) { |
1543 | if (handler->nsfd[i] < 0) | |
1544 | continue; | |
fa3a5b22 | 1545 | |
8deca6c9 | 1546 | ret = setns(handler->nsfd[i], 0); |
c0b48eff CB |
1547 | if (ret < 0) { |
1548 | /* | |
1549 | * Note that joining a user and/or mount namespace | |
1550 | * requires the process is not multithreaded otherwise | |
1551 | * setns() will fail here. | |
1552 | */ | |
1553 | SYSERROR("Failed to inherit %s namespace", | |
1554 | ns_info[i].proc_name); | |
8deca6c9 | 1555 | return -1; |
c0b48eff | 1556 | } |
fa3a5b22 | 1557 | |
8deca6c9 | 1558 | DEBUG("Inherited %s namespace", ns_info[i].proc_name); |
fa3a5b22 CB |
1559 | } |
1560 | ||
becad0ec | 1561 | flags = handler->ns_on_clone_flags; |
8deca6c9 | 1562 | flags |= CLONE_PARENT; |
33942046 CB |
1563 | handler->pid = lxc_raw_clone_cb(do_start, handler, CLONE_PIDFD | flags, |
1564 | &handler->pidfd); | |
8deca6c9 | 1565 | if (handler->pid < 0) |
fa3a5b22 CB |
1566 | return -1; |
1567 | ||
8deca6c9 | 1568 | return 0; |
fa3a5b22 CB |
1569 | } |
1570 | ||
480588e6 CB |
1571 | /* lxc_spawn() performs crucial setup tasks and clone()s the new process which |
1572 | * exec()s the requested container binary. | |
1573 | * Note that lxc_spawn() runs in the parent namespaces. Any operations performed | |
1574 | * right here should be double checked if they'd pose a security risk. (For | |
1575 | * example, any {u}mount() operations performed here will be reflected on the | |
1576 | * host!) | |
1577 | */ | |
74a3920a | 1578 | static int lxc_spawn(struct lxc_handler *handler) |
59eb99ba | 1579 | { |
f62cf1d4 | 1580 | __do_close int data_sock0 = -EBADF, data_sock1 = -EBADF; |
8deca6c9 | 1581 | int i, ret; |
08dd2805 | 1582 | char pidstr[20]; |
57927bf2 CB |
1583 | bool wants_to_map_ids; |
1584 | struct lxc_list *id_map; | |
6e5fc7a5 | 1585 | const char *name = handler->name; |
28d9e29e | 1586 | const char *lxcpath = handler->lxcpath; |
2202afc9 | 1587 | bool share_ns = false; |
28d9e29e | 1588 | struct lxc_conf *conf = handler->conf; |
2202afc9 | 1589 | struct cgroup_ops *cgroup_ops = handler->cgroup_ops; |
9f30a190 | 1590 | |
28d9e29e | 1591 | id_map = &conf->id_map; |
57927bf2 | 1592 | wants_to_map_ids = !lxc_list_empty(id_map); |
2f2623ec | 1593 | |
28d9e29e | 1594 | for (i = 0; i < LXC_NS_MAX; i++) { |
b074bbf1 | 1595 | if (!conf->ns_share[i]) |
28d9e29e CB |
1596 | continue; |
1597 | ||
b074bbf1 | 1598 | handler->nsfd[i] = lxc_inherit_namespace(conf->ns_share[i], lxcpath, ns_info[i].proc_name); |
28d9e29e CB |
1599 | if (handler->nsfd[i] < 0) |
1600 | return -1; | |
fa3a5b22 | 1601 | |
8deca6c9 | 1602 | share_ns = true; |
28d9e29e | 1603 | } |
50e98013 | 1604 | |
6e48e7c5 | 1605 | if (!lxc_sync_init(handler)) |
9d7f9e52 | 1606 | return -1; |
0ad19a3f | 1607 | |
b9f522c5 | 1608 | ret = socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0, |
790255cf | 1609 | handler->data_sock); |
95ef0d7c CB |
1610 | if (ret < 0) |
1611 | goto out_sync_fini; | |
cdb2a47f CB |
1612 | data_sock0 = handler->data_sock[0]; |
1613 | data_sock1 = handler->data_sock[1]; | |
e8bd4e43 | 1614 | |
5af9369b | 1615 | ret = resolve_clone_flags(handler); |
95ef0d7c CB |
1616 | if (ret < 0) |
1617 | goto out_sync_fini; | |
9f30a190 | 1618 | |
03ca4af8 TP |
1619 | if (handler->ns_clone_flags & CLONE_NEWNET) { |
1620 | ret = lxc_find_gateway_addresses(handler); | |
1621 | if (ret) { | |
1622 | ERROR("Failed to find gateway addresses"); | |
1623 | goto out_sync_fini; | |
1624 | } | |
1625 | } | |
1626 | ||
e8b181f5 | 1627 | if (!cgroup_ops->payload_create(cgroup_ops, handler)) { |
cfc62c60 | 1628 | ERROR("Failed creating cgroups"); |
47d8fb3b CS |
1629 | goto out_delete_net; |
1630 | } | |
1631 | ||
408da065 | 1632 | /* Create a process in a new set of namespaces. */ |
4e232466 CB |
1633 | if (share_ns) { |
1634 | pid_t attacher_pid; | |
1635 | ||
cfc62c60 | 1636 | attacher_pid = lxc_clone(do_share_ns, handler, |
33258b95 | 1637 | CLONE_VFORK | CLONE_VM | CLONE_FILES, NULL); |
4e232466 CB |
1638 | if (attacher_pid < 0) { |
1639 | SYSERROR(LXC_CLONE_ERROR); | |
1640 | goto out_delete_net; | |
1641 | } | |
1642 | ||
1643 | ret = wait_for_pid(attacher_pid); | |
1644 | if (ret < 0) { | |
1645 | SYSERROR("Intermediate process failed"); | |
1646 | goto out_delete_net; | |
1647 | } | |
f7176c3e CB |
1648 | |
1649 | if (handler->pid < 0) { | |
1650 | SYSERROR(LXC_CLONE_ERROR); | |
1651 | goto out_delete_net; | |
1652 | } | |
4e232466 | 1653 | } else { |
84b66ced | 1654 | int cgroup_fd = -EBADF; |
f7176c3e CB |
1655 | |
1656 | struct lxc_clone_args clone_args = { | |
1657 | .flags = handler->clone_flags, | |
1658 | .pidfd = ptr_to_u64(&handler->pidfd), | |
1659 | .exit_signal = SIGCHLD, | |
1660 | }; | |
1661 | ||
1662 | if (handler->ns_clone_flags & CLONE_NEWCGROUP) { | |
1663 | cgroup_fd = cgroup_unified_fd(cgroup_ops); | |
1664 | if (cgroup_fd >= 0) { | |
1665 | handler->clone_flags |= CLONE_INTO_CGROUP; | |
1666 | clone_args.flags |= CLONE_INTO_CGROUP; | |
1667 | clone_args.cgroup = cgroup_fd; | |
1668 | } | |
1669 | } | |
1670 | ||
1671 | /* Try to spawn directly into target cgroup. */ | |
1672 | handler->pid = lxc_clone3(&clone_args, CLONE_ARGS_SIZE_VER2); | |
1673 | if (handler->pid < 0) { | |
1674 | SYSTRACE("Failed to spawn container directly into target cgroup"); | |
1675 | ||
1676 | /* Kernel might simply be too old for CLONE_INTO_CGROUP. */ | |
1677 | handler->clone_flags &= ~(CLONE_INTO_CGROUP | CLONE_NEWCGROUP); | |
1678 | handler->ns_on_clone_flags &= ~CLONE_NEWCGROUP; | |
1679 | handler->ns_unshare_flags |= CLONE_NEWCGROUP; | |
1680 | ||
1681 | clone_args.flags = handler->clone_flags; | |
1682 | ||
1683 | handler->pid = lxc_clone3(&clone_args, CLONE_ARGS_SIZE_VER0); | |
1684 | } else if (cgroup_fd >= 0) { | |
1685 | TRACE("Spawned container directly into target cgroup via cgroup2 fd %d", cgroup_fd); | |
1686 | } | |
1687 | ||
1688 | /* Kernel might be too old for clone3(). */ | |
1689 | if (handler->pid < 0) { | |
1690 | SYSTRACE("Failed to spawn container via clone3()"); | |
abd80bde CB |
1691 | |
1692 | /* | |
1693 | * In contrast to all other architectures arm64 verifies that | |
1694 | * the argument we use to retrieve the pidfd with is | |
1695 | * initialized to 0. But we need to be able to initialize it to | |
1696 | * a negative value such as our customary -EBADF so we can | |
1697 | * detect whether this kernel supports pidfds. If the syscall | |
1698 | * returns and the pidfd variable is set to something >= 0 then | |
1699 | * we know this is a kernel supporting pidfds. But if we can't | |
1700 | * set it to -EBADF then this won't work since 0 is a valid | |
1701 | * file descriptor too. And since legacy clone silently ignores | |
1702 | * unknown flags we are left without any way to detect support | |
1703 | * for pidfds. So let's special-case arm64 to not fail starting | |
1704 | * containers. | |
1705 | */ | |
1706 | #if defined(__aarch64__) | |
1707 | handler->pid = lxc_raw_legacy_clone(handler->clone_flags & ~CLONE_PIDFD, NULL); | |
1708 | #else | |
f7176c3e | 1709 | handler->pid = lxc_raw_legacy_clone(handler->clone_flags, &handler->pidfd); |
abd80bde | 1710 | #endif |
f7176c3e CB |
1711 | } |
1712 | ||
1713 | if (handler->pid < 0) { | |
1714 | SYSERROR(LXC_CLONE_ERROR); | |
1715 | goto out_delete_net; | |
1716 | } | |
1717 | ||
1718 | if (handler->pid == 0) { | |
1719 | (void)do_start(handler); | |
1720 | _exit(EXIT_FAILURE); | |
1721 | } | |
0ad19a3f | 1722 | } |
f7176c3e CB |
1723 | if (handler->pidfd < 0) |
1724 | handler->clone_flags &= ~CLONE_PIDFD; | |
fa3a5b22 | 1725 | TRACE("Cloned child process %d", handler->pid); |
1bd8d726 | 1726 | |
39293f22 CB |
1727 | /* Verify that we can actually make use of pidfds. */ |
1728 | if (!lxc_can_use_pidfd(handler->pidfd)) | |
1729 | close_prot_errno_disarm(handler->pidfd); | |
1730 | ||
fa60cd7b CB |
1731 | ret = strnprintf(pidstr, 20, "%d", handler->pid); |
1732 | if (ret < 0) | |
1871e646 CB |
1733 | goto out_delete_net; |
1734 | ||
1735 | ret = setenv("LXC_PID", pidstr, 1); | |
1736 | if (ret < 0) | |
1737 | SYSERROR("Failed to set environment variable: LXC_PID=%s", pidstr); | |
1738 | ||
9662e444 | 1739 | for (i = 0; i < LXC_NS_MAX; i++) |
becad0ec | 1740 | if (handler->ns_on_clone_flags & ns_info[i].clone_flag) |
28d9e29e | 1741 | INFO("Cloned %s", ns_info[i].flag_name); |
0ad19a3f | 1742 | |
8db6be1b | 1743 | if (!lxc_try_preserve_namespaces(handler, handler->ns_on_clone_flags)) { |
28d9e29e CB |
1744 | ERROR("Failed to preserve cloned namespaces for lxc.hook.stop"); |
1745 | goto out_delete_net; | |
1746 | } | |
b6b2b194 | 1747 | |
3c22086f CLG |
1748 | lxc_sync_fini_child(handler); |
1749 | ||
4d8bdfa0 CB |
1750 | if (lxc_abstract_unix_send_fds(handler->data_sock[0], &handler->monitor_status_fd, 1, NULL, 0) < 0) { |
1751 | ERROR("Failed to send status file descriptor to child process"); | |
1752 | goto out_delete_net; | |
1753 | } | |
1754 | close_prot_errno_disarm(handler->monitor_status_fd); | |
1755 | ||
408da065 CB |
1756 | /* Map the container uids. The container became an invalid userid the |
1757 | * moment it was cloned with CLONE_NEWUSER. This call doesn't change | |
1758 | * anything immediately, but allows the container to setuid(0) (0 being | |
1759 | * mapped to something else on the host.) later to become a valid uid | |
1760 | * again. | |
1761 | */ | |
fa3a5b22 | 1762 | if (wants_to_map_ids) { |
df3c5314 | 1763 | if (!handler->conf->ns_share[LXC_NS_USER] && |
46186acd | 1764 | (handler->conf->ns_keep & CLONE_NEWUSER) == 0) { |
fa3a5b22 CB |
1765 | ret = lxc_map_ids(id_map, handler->pid); |
1766 | if (ret < 0) { | |
1767 | ERROR("Failed to set up id mapping."); | |
1768 | goto out_delete_net; | |
1769 | } | |
1770 | } | |
5b1e83cb SH |
1771 | } |
1772 | ||
c581d2a6 | 1773 | if (!cgroup_ops->setup_limits_legacy(cgroup_ops, handler->conf, false)) { |
cfc62c60 | 1774 | ERROR("Failed to setup cgroup limits for container \"%s\"", name); |
6031a6e5 DE |
1775 | goto out_delete_net; |
1776 | } | |
1777 | ||
838d1556 CB |
1778 | if (!cgroup_ops->payload_delegate_controllers(cgroup_ops)) { |
1779 | ERROR("Failed to delegate controllers to payload cgroup"); | |
7fef7a06 | 1780 | goto out_delete_net; |
fe70edee | 1781 | } |
218d4250 | 1782 | |
838d1556 CB |
1783 | if (!cgroup_ops->payload_enter(cgroup_ops, handler)) { |
1784 | ERROR("Failed to enter cgroups"); | |
c581d2a6 CB |
1785 | goto out_delete_net; |
1786 | } | |
1787 | ||
1788 | if (!cgroup_ops->setup_limits(cgroup_ops, handler)) { | |
1789 | ERROR("Failed to setup cgroup limits for container \"%s\"", name); | |
1790 | goto out_delete_net; | |
1791 | } | |
1792 | ||
2202afc9 | 1793 | if (!cgroup_ops->chown(cgroup_ops, handler->conf)) |
0996e18a SH |
1794 | goto out_delete_net; |
1795 | ||
6bc4165d CB |
1796 | if (!lxc_sync_barrier_child(handler, START_SYNC_STARTUP)) |
1797 | goto out_delete_net; | |
1798 | ||
aa0c0e7b RK |
1799 | /* If not done yet, we're now ready to preserve the network namespace */ |
1800 | if (handler->nsfd[LXC_NS_NET] < 0) { | |
8db6be1b | 1801 | ret = lxc_try_preserve_namespace(handler, LXC_NS_NET, "net"); |
aa0c0e7b | 1802 | if (ret < 0) { |
857ba1f0 | 1803 | if (ret != -ENOENT) { |
aa0c0e7b RK |
1804 | SYSERROR("Failed to preserve net namespace"); |
1805 | goto out_delete_net; | |
1806 | } | |
4cb53844 | 1807 | } |
fa3a5b22 | 1808 | } |
aa0c0e7b RK |
1809 | ret = lxc_netns_set_nsid(handler->nsfd[LXC_NS_NET]); |
1810 | if (ret < 0) | |
1811 | SYSWARN("Failed to allocate new network namespace id"); | |
1812 | else | |
1813 | TRACE("Allocated new network namespace id"); | |
fa3a5b22 | 1814 | |
408da065 | 1815 | /* Create the network configuration. */ |
becad0ec | 1816 | if (handler->ns_clone_flags & CLONE_NEWNET) { |
e389f2af | 1817 | ret = lxc_create_network(handler); |
cfc62c60 | 1818 | if (ret < 0) { |
e389f2af | 1819 | ERROR("Failed to create the network"); |
7fef7a06 | 1820 | goto out_delete_net; |
82d5ae15 | 1821 | } |
0ad19a3f | 1822 | } |
1823 | ||
6bc4165d CB |
1824 | if (!lxc_list_empty(&conf->procs)) { |
1825 | ret = setup_proc_filesystem(&conf->procs, handler->pid); | |
1826 | if (ret < 0) | |
1827 | goto out_delete_net; | |
1828 | } | |
1829 | ||
1830 | if (!lxc_list_empty(&conf->limits)) { | |
1831 | ret = setup_resource_limits(&conf->limits, handler->pid); | |
1832 | if (ret < 0) { | |
1833 | ERROR("Failed to setup resource limits"); | |
1834 | goto out_delete_net; | |
1835 | } | |
1836 | } | |
1837 | ||
8945dad0 CB |
1838 | /* Tell the child to continue its initialization. */ |
1839 | if (!lxc_sync_wake_child(handler, START_SYNC_POST_CONFIGURE)) | |
1840 | goto out_delete_net; | |
1841 | ||
e4564b7e CB |
1842 | ret = lxc_rootfs_prepare_parent(handler); |
1843 | if (ret) { | |
1844 | ERROR("Failed to prepare rootfs"); | |
1845 | goto out_delete_net; | |
1846 | } | |
1847 | ||
8945dad0 CB |
1848 | if (handler->ns_clone_flags & CLONE_NEWNET) { |
1849 | ret = lxc_network_send_to_child(handler); | |
1850 | if (ret < 0) { | |
6bc4165d | 1851 | SYSERROR("Failed to send veth names to child"); |
8945dad0 CB |
1852 | goto out_delete_net; |
1853 | } | |
1854 | } | |
1855 | ||
6bc4165d | 1856 | if (!lxc_sync_wait_child(handler, START_SYNC_IDMAPPED_MOUNTS)) |
f4152036 CB |
1857 | goto out_delete_net; |
1858 | ||
1b82d721 CB |
1859 | ret = lxc_idmapped_mounts_parent(handler); |
1860 | if (ret) { | |
1861 | ERROR("Failed to setup mount entries"); | |
1862 | goto out_delete_net; | |
1863 | } | |
1864 | ||
1865 | if (!lxc_sync_wait_child(handler, START_SYNC_CGROUP_LIMITS)) | |
1866 | goto out_delete_net; | |
1867 | ||
432faf20 | 1868 | /* |
3a89b0ab CB |
1869 | * With isolation the limiting devices cgroup was already setup, so |
1870 | * only setup devices here if we have no namespace directory. | |
432faf20 WB |
1871 | */ |
1872 | if (!handler->conf->cgroup_meta.namespace_dir && | |
1873 | !cgroup_ops->setup_limits_legacy(cgroup_ops, handler->conf, true)) { | |
cfc62c60 | 1874 | ERROR("Failed to setup legacy device cgroup controller limits"); |
b98f7d6e | 1875 | goto out_delete_net; |
544a48a0 | 1876 | } |
cfc62c60 | 1877 | TRACE("Set up legacy device cgroup controller limits"); |
544a48a0 | 1878 | |
bf651989 CB |
1879 | if (!cgroup_ops->devices_activate(cgroup_ops, handler)) { |
1880 | ERROR("Failed to setup cgroup2 device controller limits"); | |
1881 | goto out_delete_net; | |
1882 | } | |
1883 | TRACE("Set up cgroup2 device controller limits"); | |
1884 | ||
6bc4165d CB |
1885 | cgroup_ops->finalize(cgroup_ops); |
1886 | TRACE("Finished setting up cgroups"); | |
1887 | ||
1888 | /* Run any host-side start hooks */ | |
1889 | ret = run_lxc_hooks(name, "start-host", conf, NULL); | |
1890 | if (ret < 0) { | |
1891 | ERROR("Failed to run lxc.hook.start-host"); | |
1892 | goto out_delete_net; | |
1893 | } | |
1894 | ||
1895 | if (!lxc_sync_wake_child(handler, START_SYNC_FDS)) | |
1896 | goto out_delete_net; | |
1897 | ||
f7176c3e | 1898 | if (handler->ns_unshare_flags & CLONE_NEWCGROUP) { |
8bf3abfb | 1899 | /* Now we're ready to preserve the cgroup namespace */ |
8db6be1b | 1900 | ret = lxc_try_preserve_namespace(handler, LXC_NS_CGROUP, "cgroup"); |
8bf3abfb | 1901 | if (ret < 0) { |
857ba1f0 | 1902 | if (ret != -ENOENT) { |
6d1400b5 | 1903 | SYSERROR("Failed to preserve cgroup namespace"); |
4cb53844 CB |
1904 | goto out_delete_net; |
1905 | } | |
8bf3abfb | 1906 | } |
8bf3abfb CB |
1907 | } |
1908 | ||
f7176c3e | 1909 | if (handler->ns_unshare_flags & CLONE_NEWTIME) { |
245066af | 1910 | /* Now we're ready to preserve the time namespace */ |
8db6be1b | 1911 | ret = lxc_try_preserve_namespace(handler, LXC_NS_TIME, "time"); |
f1c43439 | 1912 | if (ret < 0) { |
857ba1f0 | 1913 | if (ret != -ENOENT) { |
f1c43439 CB |
1914 | SYSERROR("Failed to preserve time namespace"); |
1915 | goto out_delete_net; | |
1916 | } | |
f1c43439 CB |
1917 | } |
1918 | } | |
1919 | ||
493ae3fe | 1920 | ret = lxc_sync_fds_parent(handler); |
1b82d721 | 1921 | if (ret < 0) { |
493ae3fe | 1922 | SYSERROR("Failed to sync file descriptors with child"); |
1b82d721 | 1923 | goto out_delete_net; |
790255cf CB |
1924 | } |
1925 | ||
9f77617b CB |
1926 | ret = lxc_terminal_setup(conf); |
1927 | if (ret < 0) { | |
1928 | SYSERROR("Failed to create console"); | |
1929 | goto out_delete_net; | |
1930 | } | |
1931 | ||
1b82d721 CB |
1932 | /* |
1933 | * Tell the child to complete its initialization and wait for it to | |
1934 | * exec or return an error. (The child will never return | |
1935 | * START_SYNC_READY_START+1. It will either close the sync pipe, | |
1936 | * causing lxc_sync_barrier_child to return success, or return a | |
1937 | * different value, causing us to error out). | |
1938 | */ | |
1939 | if (!lxc_sync_barrier_child(handler, START_SYNC_READY_START)) | |
1940 | goto out_delete_net; | |
1941 | ||
1942 | /* Now all networks are created, network devices are moved into place, | |
1943 | * and the correct names and ifindices in the respective namespaces have | |
1944 | * been recorded. The corresponding structs have now all been filled. So | |
1945 | * log them for debugging purposes. | |
1946 | */ | |
1947 | lxc_log_configured_netdevs(conf); | |
1948 | ||
cfc62c60 CB |
1949 | ret = handler->ops->post_start(handler, handler->data); |
1950 | if (ret < 0) | |
e6126dbe MN |
1951 | goto out_abort; |
1952 | ||
cfc62c60 CB |
1953 | ret = lxc_set_state(name, handler, RUNNING); |
1954 | if (ret < 0) { | |
1955 | ERROR("Failed to set state to \"%s\"", lxc_state2str(RUNNING)); | |
59eb99ba | 1956 | goto out_abort; |
3f21c114 | 1957 | } |
22ebac19 | 1958 | |
3c22086f | 1959 | lxc_sync_fini(handler); |
0c547523 | 1960 | |
e6126dbe | 1961 | return 0; |
1ac470c0 | 1962 | |
7fef7a06 | 1963 | out_delete_net: |
becad0ec | 1964 | if (handler->ns_clone_flags & CLONE_NEWNET) |
bb84beda | 1965 | lxc_delete_network(handler); |
1bd8d726 | 1966 | |
59eb99ba | 1967 | out_abort: |
0c5859ff | 1968 | lxc_abort(handler); |
95ef0d7c CB |
1969 | |
1970 | out_sync_fini: | |
3c22086f | 1971 | lxc_sync_fini(handler); |
5c068da9 | 1972 | |
b79fcd86 | 1973 | return -1; |
59eb99ba | 1974 | } |
0ad19a3f | 1975 | |
7bdf97d2 CB |
1976 | int __lxc_start(struct lxc_handler *handler, struct lxc_operations *ops, |
1977 | void *data, const char *lxcpath, bool daemonize, int *error_num) | |
59eb99ba | 1978 | { |
c30e9b19 | 1979 | int ret, status; |
7bdf97d2 | 1980 | const char *name = handler->name; |
aa460476 | 1981 | struct lxc_conf *conf = handler->conf; |
72068e74 | 1982 | struct cgroup_ops *cgroup_ops; |
80090207 | 1983 | |
c30e9b19 CB |
1984 | ret = lxc_init(name, handler); |
1985 | if (ret < 0) { | |
1986 | ERROR("Failed to initialize container \"%s\"", name); | |
9ad66589 | 1987 | goto out_abort; |
0ad19a3f | 1988 | } |
ee70bf78 CLG |
1989 | handler->ops = ops; |
1990 | handler->data = data; | |
bb955810 | 1991 | handler->daemonize = daemonize; |
72068e74 | 1992 | cgroup_ops = handler->cgroup_ops; |
e6126dbe | 1993 | |
76a26f55 | 1994 | if (!attach_block_device(handler->conf)) { |
c30e9b19 | 1995 | ERROR("Failed to attach block device"); |
575ea467 | 1996 | ret = -1; |
9ad66589 | 1997 | goto out_abort; |
76a26f55 SH |
1998 | } |
1999 | ||
72068e74 CB |
2000 | if (!cgroup_ops->monitor_create(cgroup_ops, handler)) { |
2001 | ERROR("Failed to create monitor cgroup"); | |
575ea467 | 2002 | ret = -1; |
9ad66589 | 2003 | goto out_abort; |
eeef32bb CB |
2004 | } |
2005 | ||
838d1556 CB |
2006 | if (!cgroup_ops->monitor_delegate_controllers(cgroup_ops)) { |
2007 | ERROR("Failed to delegate controllers to monitor cgroup"); | |
575ea467 | 2008 | ret = -1; |
9ad66589 | 2009 | goto out_abort; |
72068e74 CB |
2010 | } |
2011 | ||
838d1556 CB |
2012 | if (!cgroup_ops->monitor_enter(cgroup_ops, handler)) { |
2013 | ERROR("Failed to enter monitor cgroup"); | |
c581d2a6 | 2014 | ret = -1; |
9ad66589 | 2015 | goto out_abort; |
c581d2a6 CB |
2016 | } |
2017 | ||
e26cf563 CB |
2018 | /* If the rootfs is not a blockdev, prevent the container from marking |
2019 | * it readonly. | |
2020 | * If the container is unprivileged then skip rootfs pinning. | |
2021 | */ | |
239f29c9 | 2022 | ret = lxc_rootfs_init(conf, !lxc_list_empty(&conf->id_map)); |
e26cf563 CB |
2023 | if (ret) { |
2024 | ERROR("Failed to handle rootfs pinning for container \"%s\"", handler->name); | |
2025 | ret = -1; | |
2026 | goto out_abort; | |
2027 | } | |
2028 | ||
35120d9c | 2029 | if (geteuid() == 0 && !lxc_list_empty(&conf->id_map)) { |
26ad2c6f | 2030 | /* |
4b875ef9 | 2031 | * Most filesystems can't be mounted inside a userns so handle them here. |
26ad2c6f | 2032 | */ |
4b875ef9 | 2033 | if (rootfs_is_blockdev(conf)) { |
c30e9b19 CB |
2034 | ret = unshare(CLONE_NEWNS); |
2035 | if (ret < 0) { | |
2036 | ERROR("Failed to unshare CLONE_NEWNS"); | |
9ad66589 | 2037 | goto out_abort; |
35120d9c | 2038 | } |
c30e9b19 | 2039 | INFO("Unshared CLONE_NEWNS"); |
408da065 | 2040 | |
8ce1abc2 | 2041 | ret = lxc_setup_rootfs_prepare_root(conf, name, lxcpath); |
c30e9b19 CB |
2042 | if (ret < 0) { |
2043 | ERROR("Error setting up rootfs mount as root before spawn"); | |
9ad66589 | 2044 | goto out_abort; |
35120d9c | 2045 | } |
c30e9b19 | 2046 | INFO("Set up container rootfs as host root"); |
35120d9c SH |
2047 | } |
2048 | } | |
2049 | ||
c30e9b19 CB |
2050 | ret = lxc_spawn(handler); |
2051 | if (ret < 0) { | |
2052 | ERROR("Failed to spawn container \"%s\"", name); | |
76a26f55 | 2053 | goto out_detach_blockdev; |
0ad19a3f | 2054 | } |
2055 | ||
80308d07 | 2056 | handler->conf->reboot = REBOOT_NONE; |
8bee8851 | 2057 | |
c30e9b19 CB |
2058 | ret = lxc_poll(name, handler); |
2059 | if (ret) { | |
2060 | ERROR("LXC mainloop exited with error: %d", ret); | |
9ad66589 | 2061 | goto out_delete_network; |
59eb99ba | 2062 | } |
0ad19a3f | 2063 | |
321db026 DJ |
2064 | if (!handler->init_died && handler->pid > 0) { |
2065 | ERROR("Child process is not killed"); | |
575ea467 | 2066 | ret = -1; |
9ad66589 | 2067 | goto out_delete_network; |
321db026 DJ |
2068 | } |
2069 | ||
c30e9b19 CB |
2070 | status = lxc_wait_for_pid_status(handler->pid); |
2071 | if (status < 0) | |
2072 | SYSERROR("Failed to retrieve status for %d", handler->pid); | |
e043236e | 2073 | |
408da065 CB |
2074 | /* If the child process exited but was not signaled, it didn't call |
2075 | * reboot. This should mean it was an lxc-execute which simply exited. | |
2076 | * In any case, treat it as a 'halt'. | |
8b004f07 | 2077 | */ |
d028235d | 2078 | if (WIFSIGNALED(status)) { |
8b004f07 SH |
2079 | switch(WTERMSIG(status)) { |
2080 | case SIGINT: /* halt */ | |
c30e9b19 | 2081 | DEBUG("Container \"%s\" is halting", name); |
8b004f07 SH |
2082 | break; |
2083 | case SIGHUP: /* reboot */ | |
c30e9b19 | 2084 | DEBUG("Container \"%s\" is rebooting", name); |
80308d07 | 2085 | handler->conf->reboot = REBOOT_REQ; |
8b004f07 | 2086 | break; |
c2b9bd9e | 2087 | case SIGSYS: /* seccomp */ |
c30e9b19 | 2088 | DEBUG("Container \"%s\" violated its seccomp policy", name); |
c2b9bd9e | 2089 | break; |
8b004f07 | 2090 | default: |
c30e9b19 | 2091 | DEBUG("Unknown exit status for container \"%s\" init %d", name, WTERMSIG(status)); |
8b004f07 SH |
2092 | break; |
2093 | } | |
d028235d | 2094 | } |
828695d9 | 2095 | |
c30e9b19 CB |
2096 | ret = lxc_restore_phys_nics_to_netns(handler); |
2097 | if (ret < 0) | |
818a57fc | 2098 | ERROR("Failed to move physical network devices back to parent network namespace"); |
ce5782df | 2099 | |
1787abca | 2100 | lxc_monitor_send_exit_code(name, status, handler->lxcpath); |
eb808539 | 2101 | lxc_error_set_and_log(handler->pid, status); |
a3b4f3d6 TA |
2102 | if (error_num) |
2103 | *error_num = handler->exit_status; | |
1bd8d726 | 2104 | |
bb84beda | 2105 | lxc_delete_network(handler); |
76a26f55 | 2106 | detach_block_device(handler->conf); |
fd5be714 | 2107 | lxc_end(handler); |
c30e9b19 | 2108 | return ret; |
0ad19a3f | 2109 | |
9ad66589 | 2110 | out_abort: |
0c5859ff | 2111 | lxc_abort(handler); |
945daa24 CB |
2112 | lxc_end(handler); |
2113 | return ret; | |
65a2dc8b CB |
2114 | |
2115 | out_detach_blockdev: | |
0c5859ff | 2116 | lxc_abort(handler); |
945daa24 CB |
2117 | detach_block_device(handler->conf); |
2118 | lxc_end(handler); | |
2119 | return ret; | |
65a2dc8b | 2120 | |
9ad66589 | 2121 | out_delete_network: |
0c5859ff | 2122 | lxc_abort(handler); |
abd953eb | 2123 | lxc_restore_phys_nics_to_netns(handler); |
945daa24 CB |
2124 | lxc_delete_network(handler); |
2125 | detach_block_device(handler->conf); | |
2126 | lxc_end(handler); | |
2127 | return ret; | |
0ad19a3f | 2128 | } |
ee70bf78 CLG |
2129 | |
2130 | struct start_args { | |
2131 | char *const *argv; | |
2132 | }; | |
2133 | ||
2134 | static int start(struct lxc_handler *handler, void* data) | |
2135 | { | |
2136 | struct start_args *arg = data; | |
2137 | ||
984984e4 | 2138 | NOTICE("Exec'ing \"%s\"", arg->argv[0]); |
ee70bf78 CLG |
2139 | |
2140 | execvp(arg->argv[0], arg->argv); | |
984984e4 | 2141 | SYSERROR("Failed to exec \"%s\"", arg->argv[0]); |
ee70bf78 CLG |
2142 | return 0; |
2143 | } | |
2144 | ||
2145 | static int post_start(struct lxc_handler *handler, void* data) | |
2146 | { | |
2147 | struct start_args *arg = data; | |
2148 | ||
4c8e880e | 2149 | NOTICE("Started \"%s\" with pid \"%d\"", arg->argv[0], handler->pid); |
ee70bf78 CLG |
2150 | return 0; |
2151 | } | |
2152 | ||
2153 | static struct lxc_operations start_ops = { | |
2154 | .start = start, | |
2155 | .post_start = post_start | |
2156 | }; | |
2157 | ||
7bdf97d2 | 2158 | int lxc_start(char *const argv[], struct lxc_handler *handler, |
bb955810 | 2159 | const char *lxcpath, bool daemonize, int *error_num) |
ee70bf78 CLG |
2160 | { |
2161 | struct start_args start_arg = { | |
2162 | .argv = argv, | |
2163 | }; | |
2164 | ||
b2efeb0b | 2165 | TRACE("Doing lxc_start"); |
7bdf97d2 | 2166 | return __lxc_start(handler, &start_ops, &start_arg, lxcpath, daemonize, error_num); |
ee70bf78 | 2167 | } |
28272964 CB |
2168 | |
2169 | static void lxc_destroy_container_on_signal(struct lxc_handler *handler, | |
2170 | const char *name) | |
2171 | { | |
245100a0 | 2172 | char destroy[PATH_MAX]; |
f01f7975 | 2173 | struct lxc_container *c; |
ae3beac9 CB |
2174 | int ret = 0; |
2175 | bool bret = true; | |
2176 | ||
df31363a | 2177 | if (handler->conf->rootfs.path && handler->conf->rootfs.mount) { |
790255cf | 2178 | bret = do_destroy_container(handler); |
28272964 | 2179 | if (!bret) { |
ae3beac9 | 2180 | ERROR("Error destroying rootfs for container \"%s\"", name); |
28272964 CB |
2181 | return; |
2182 | } | |
2183 | } | |
ae3beac9 | 2184 | INFO("Destroyed rootfs for container \"%s\"", name); |
28272964 | 2185 | |
fa60cd7b CB |
2186 | ret = strnprintf(destroy, sizeof(destroy), "%s/%s", handler->lxcpath, name); |
2187 | if (ret < 0) { | |
ae3beac9 | 2188 | ERROR("Error destroying directory for container \"%s\"", name); |
28272964 CB |
2189 | return; |
2190 | } | |
2191 | ||
f01f7975 CB |
2192 | c = lxc_container_new(name, handler->lxcpath); |
2193 | if (c) { | |
2194 | if (container_disk_lock(c)) { | |
ae3beac9 | 2195 | INFO("Could not update lxc_snapshots file"); |
f01f7975 CB |
2196 | lxc_container_put(c); |
2197 | } else { | |
2198 | mod_all_rdeps(c, false); | |
2199 | container_disk_unlock(c); | |
2200 | lxc_container_put(c); | |
2201 | } | |
2202 | } | |
2203 | ||
d0fbc7ba | 2204 | if (!handler->am_root) |
94b0a3ac CB |
2205 | ret = userns_exec_full(handler->conf, lxc_rmdir_onedev_wrapper, |
2206 | destroy, "lxc_rmdir_onedev_wrapper"); | |
28272964 CB |
2207 | else |
2208 | ret = lxc_rmdir_onedev(destroy, NULL); | |
2209 | ||
2210 | if (ret < 0) { | |
ae3beac9 | 2211 | ERROR("Error destroying directory for container \"%s\"", name); |
28272964 CB |
2212 | return; |
2213 | } | |
ae3beac9 | 2214 | INFO("Destroyed directory for container \"%s\"", name); |
28272964 CB |
2215 | } |
2216 | ||
2217 | static int lxc_rmdir_onedev_wrapper(void *data) | |
2218 | { | |
2219 | char *arg = (char *) data; | |
2220 | return lxc_rmdir_onedev(arg, NULL); | |
2221 | } | |
2222 | ||
134df645 CB |
2223 | static bool do_destroy_container(struct lxc_handler *handler) |
2224 | { | |
94b0a3ac CB |
2225 | int ret; |
2226 | ||
d0fbc7ba | 2227 | if (!handler->am_root) { |
94b0a3ac CB |
2228 | ret = userns_exec_full(handler->conf, storage_destroy_wrapper, |
2229 | handler->conf, "storage_destroy_wrapper"); | |
2230 | if (ret < 0) | |
d028235d | 2231 | return false; |
10bc1861 | 2232 | |
d028235d SG |
2233 | return true; |
2234 | } | |
10bc1861 | 2235 | |
790255cf | 2236 | return storage_destroy(handler->conf); |
28272964 | 2237 | } |