]>
Commit | Line | Data |
---|---|---|
cc73685d | 1 | /* SPDX-License-Identifier: LGPL-2.1+ */ |
0ad19a3f | 2 | |
1160ce89 CB |
3 | #include "config.h" |
4 | ||
0ad19a3f | 5 | #include <dirent.h> |
6 | #include <errno.h> | |
b0a33c1e | 7 | #include <fcntl.h> |
c476bdce | 8 | #include <grp.h> |
37515ebd | 9 | #include <poll.h> |
b467714b | 10 | #include <pthread.h> |
f3787121 CB |
11 | #include <signal.h> |
12 | #include <stdio.h> | |
13 | #include <stdlib.h> | |
14 | #include <string.h> | |
0ad19a3f | 15 | #include <sys/file.h> |
f4d507d5 | 16 | #include <sys/mount.h> |
f3787121 | 17 | #include <sys/param.h> |
0ad19a3f | 18 | #include <sys/prctl.h> |
f3787121 CB |
19 | #include <sys/socket.h> |
20 | #include <sys/stat.h> | |
21 | #include <sys/syscall.h> | |
ddceb1f9 | 22 | #include <sys/types.h> |
b0a33c1e | 23 | #include <sys/un.h> |
f3787121 | 24 | #include <sys/wait.h> |
d38dd64a | 25 | #include <unistd.h> |
495d2046 | 26 | |
12ae2a33 CB |
27 | #include "lxc.h" |
28 | ||
f3787121 | 29 | #include "af_unix.h" |
5eac34d5 | 30 | #include "attach_options.h" |
f3787121 | 31 | #include "caps.h" |
c988c8b1 CB |
32 | #include "cgroups/cgroup.h" |
33 | #include "cgroups/cgroup_utils.h" | |
f3787121 | 34 | #include "commands.h" |
bbf5cf35 | 35 | #include "commands_utils.h" |
59eac805 | 36 | #include "compiler.h" |
f3787121 | 37 | #include "conf.h" |
74c6e2b0 | 38 | #include "confile_utils.h" |
e2bcd7db | 39 | #include "error.h" |
7fbb15ec | 40 | #include "file_utils.h" |
1cc8bd4b | 41 | #include "list.h" |
f3787121 | 42 | #include "log.h" |
d38dd64a | 43 | #include "lsm/lsm.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 | ||
db4af8c5 | 65 | #if !HAVE_STRLCPY |
58db1a61 | 66 | #include "strlcpy.h" |
9de31d5a CB |
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 | int ret; | |
857ba1f0 | 155 | const char *ns = ns_info[ns_idx].proc_name; |
4cb53844 | 156 | |
8db6be1b | 157 | if ((ns_clone_flags & ns_info[ns_idx].clone_flag) == 0) |
9f30a190 | 158 | continue; |
28d9e29e | 159 | |
8db6be1b CB |
160 | ret = lxc_try_preserve_namespace(handler, ns_idx, |
161 | ns_info[ns_idx].proc_name); | |
162 | if (ret < 0) { | |
857ba1f0 CB |
163 | if (ret == -ENOENT) { |
164 | SYSERROR("Kernel does not support preserving %s namespaces", ns); | |
165 | continue; | |
166 | } | |
167 | ||
9beaca55 CB |
168 | /* |
169 | * Handle kernels that do not support interacting with | |
170 | * namespaces through procfs. | |
4cb53844 | 171 | */ |
4cb53844 | 172 | lxc_put_nsfds(handler); |
857ba1f0 | 173 | return log_error_errno(false, errno, "Failed to preserve %s namespace", ns); |
4cb53844 | 174 | } |
9f30a190 MM |
175 | } |
176 | ||
62d05d9b | 177 | return true; |
9f30a190 MM |
178 | } |
179 | ||
ec1dc633 | 180 | static inline bool match_stdfds(int fd) |
80090207 | 181 | { |
ec1dc633 | 182 | return (fd == STDIN_FILENO || fd == STDOUT_FILENO || fd == STDERR_FILENO); |
80090207 CLG |
183 | } |
184 | ||
4c03724e | 185 | #ifdef HAVE_DLOG |
186 | static bool match_dlog_fds(struct dirent *direntp) | |
187 | { | |
188 | char path[PATH_MAX] = {0}; | |
189 | char link[PATH_MAX] = {0}; | |
190 | ssize_t linklen; | |
191 | int ret; | |
192 | ||
fa60cd7b CB |
193 | ret = strnprintf(path, sizeof(path), "/proc/self/fd/%s", direntp->d_name); |
194 | if (ret < 0) | |
818a57fc | 195 | return log_error(false, "Failed to create file descriptor name"); |
4c03724e | 196 | |
197 | linklen = readlink(path, link, PATH_MAX); | |
818a57fc CB |
198 | if (linklen < 0) |
199 | return log_error(false, "Failed to read link path - \"%s\"", path); | |
200 | else if (linklen >= PATH_MAX) | |
201 | return log_error(false, "The name of link path is too long - \"%s\"", path); | |
202 | ||
d0269705 CB |
203 | if (strequal(link, "/dev/log_main") || |
204 | strequal(link, "/dev/log_system") || | |
205 | strequal(link, "/dev/log_radio")) | |
4c03724e | 206 | return true; |
207 | ||
208 | return false; | |
209 | } | |
210 | #endif | |
211 | ||
46abf219 RJ |
212 | /* Parses the LISTEN_FDS environment variable value. |
213 | * The returned value is the highest fd number up to which the | |
214 | * file descriptors must be passed to the container process. | |
215 | * | |
216 | * For example, if LISTEN_FDS=2 then 4 is returned and file descriptors 3 and 4 | |
217 | * MUST be passed to the container process (in addition to the standard streams) | |
218 | * to support [socket activation][systemd-listen-fds]. | |
219 | */ | |
220 | static unsigned int get_listen_fds_max(void) | |
221 | { | |
222 | int ret; | |
223 | unsigned int num_fds; | |
224 | const char *val; | |
225 | ||
226 | val = getenv("LISTEN_FDS"); | |
227 | if (!val) | |
228 | return 0; | |
229 | ||
230 | ret = lxc_safe_uint(val, &num_fds); | |
231 | if (ret < 0) | |
232 | return syserror_ret(0, "Failed to parse \"LISTEN_FDS=%s\" environment variable", val); | |
233 | ||
234 | return log_trace(num_fds, "Parsed \"LISTEN_FDS=%s\" environment variable", val); | |
235 | } | |
236 | ||
47a46cf1 CB |
237 | int lxc_check_inherited(struct lxc_conf *conf, bool closeall, |
238 | int *fds_to_ignore, size_t len_fds) | |
80090207 | 239 | { |
80090207 | 240 | int fd, fddir; |
47a46cf1 | 241 | size_t i; |
80090207 | 242 | DIR *dir; |
a5a70219 | 243 | struct dirent *direntp; |
46abf219 | 244 | unsigned int listen_fds_max; |
d2e75eba | 245 | struct lxc_state_client *client, *nclient; |
80090207 | 246 | |
d2cf4c37 SH |
247 | if (conf && conf->close_all_fds) |
248 | closeall = true; | |
249 | ||
46abf219 RJ |
250 | listen_fds_max = get_listen_fds_max(); |
251 | ||
4e2d6b9a CB |
252 | /* |
253 | * Disable syslog at this point to avoid the above logging | |
254 | * function to open a new fd and make the check_inherited function | |
255 | * enter an infinite loop. | |
256 | */ | |
257 | lxc_log_syslog_disable(); | |
258 | ||
b119f362 | 259 | restart: |
80090207 | 260 | dir = opendir("/proc/self/fd"); |
818a57fc CB |
261 | if (!dir) |
262 | return log_warn(-1, "Failed to open directory"); | |
80090207 CLG |
263 | |
264 | fddir = dirfd(dir); | |
265 | ||
74f96976 | 266 | while ((direntp = readdir(dir))) { |
a5a70219 | 267 | int ret; |
d39b10eb CB |
268 | bool matched = false; |
269 | ||
d0269705 | 270 | if (strequal(direntp->d_name, ".")) |
80090207 CLG |
271 | continue; |
272 | ||
d0269705 | 273 | if (strequal(direntp->d_name, "..")) |
80090207 CLG |
274 | continue; |
275 | ||
a5a70219 CB |
276 | ret = lxc_safe_int(direntp->d_name, &fd); |
277 | if (ret < 0) { | |
278 | INFO("Could not parse file descriptor for \"%s\"", direntp->d_name); | |
d4cff0d2 CB |
279 | continue; |
280 | } | |
80090207 | 281 | |
47a46cf1 CB |
282 | for (i = 0; i < len_fds; i++) |
283 | if (fds_to_ignore[i] == fd) | |
284 | break; | |
285 | ||
286 | if (fd == fddir || fd == lxc_log_fd || | |
287 | (i < len_fds && fd == fds_to_ignore[i])) | |
80090207 CLG |
288 | continue; |
289 | ||
d39b10eb | 290 | /* Keep state clients that wait on reboots. */ |
3ee9e4fb | 291 | if (conf) { |
d2e75eba | 292 | list_for_each_entry_safe(client, nclient, &conf->state_clients, head) { |
3ee9e4fb CB |
293 | if (client->clientfd != fd) |
294 | continue; | |
d39b10eb | 295 | |
3ee9e4fb CB |
296 | matched = true; |
297 | break; | |
298 | } | |
d39b10eb CB |
299 | } |
300 | ||
301 | if (matched) | |
302 | continue; | |
303 | ||
858377e4 SH |
304 | if (current_config && fd == current_config->logfd) |
305 | continue; | |
306 | ||
ec1dc633 | 307 | if (match_stdfds(fd)) |
80090207 | 308 | continue; |
80090207 | 309 | |
1d5e5f26 | 310 | #ifdef HAVE_DLOG |
311 | if (match_dlog_fds(direntp)) | |
312 | continue; | |
313 | ||
314 | #endif | |
46abf219 | 315 | |
a15c3609 | 316 | if ((size_t)fd <= listen_fds_max) { |
46abf219 RJ |
317 | INFO("Inheriting fd %d (using the LISTEN_FDS environment variable)", fd); |
318 | continue; | |
319 | } | |
320 | ||
d2cf4c37 | 321 | if (closeall) { |
4e2d6b9a CB |
322 | if (close(fd)) |
323 | SYSINFO("Closed inherited fd %d", fd); | |
324 | else | |
325 | INFO("Closed inherited fd %d", fd); | |
b119f362 | 326 | closedir(dir); |
b119f362 SH |
327 | goto restart; |
328 | } | |
28d9e29e | 329 | WARN("Inherited fd %d", fd); |
80090207 | 330 | } |
4e2d6b9a | 331 | closedir(dir); |
80090207 | 332 | |
4e2d6b9a CB |
333 | /* |
334 | * Only enable syslog at this point to avoid the above logging | |
335 | * function to open a new fd and make the check_inherited function | |
336 | * enter an infinite loop. | |
64c57ea1 | 337 | */ |
4e2d6b9a | 338 | lxc_log_syslog_enable(); |
64c57ea1 | 339 | |
92c7f629 | 340 | return 0; |
80090207 CLG |
341 | } |
342 | ||
83ee7875 | 343 | static int setup_signal_fd(sigset_t *oldmask) |
b0a33c1e | 344 | { |
7288dfb6 | 345 | int ret; |
b0a33c1e | 346 | sigset_t mask; |
7288dfb6 | 347 | const int signals[] = {SIGBUS, SIGILL, SIGSEGV, SIGWINCH}; |
b0a33c1e | 348 | |
408da065 | 349 | /* Block everything except serious error signals. */ |
df0795b1 CB |
350 | ret = sigfillset(&mask); |
351 | if (ret < 0) | |
352 | return -EBADF; | |
353 | ||
a15c3609 | 354 | for (size_t sig = 0; sig < (sizeof(signals) / sizeof(signals[0])); sig++) { |
df0795b1 CB |
355 | ret = sigdelset(&mask, signals[sig]); |
356 | if (ret < 0) | |
357 | return -EBADF; | |
b0a33c1e | 358 | } |
359 | ||
b467714b | 360 | ret = pthread_sigmask(SIG_BLOCK, &mask, oldmask); |
818a57fc CB |
361 | if (ret < 0) |
362 | return log_error_errno(-EBADF, errno, | |
363 | "Failed to set signal mask"); | |
b0a33c1e | 364 | |
df0795b1 | 365 | ret = signalfd(-1, &mask, SFD_CLOEXEC); |
818a57fc CB |
366 | if (ret < 0) |
367 | return log_error_errno(-EBADF, | |
368 | errno, "Failed to create signal file descriptor"); | |
b0a33c1e | 369 | |
df0795b1 | 370 | TRACE("Created signal file descriptor %d", ret); |
1ac470c0 | 371 | |
df0795b1 | 372 | return ret; |
b0a33c1e | 373 | } |
374 | ||
84c92abd | 375 | static int signal_handler(int fd, uint32_t events, void *data, |
3298b37d | 376 | struct lxc_async_descr *descr) |
b0a33c1e | 377 | { |
15cd25fd | 378 | int ret; |
3c319edb CB |
379 | siginfo_t info; |
380 | struct signalfd_siginfo siginfo; | |
381 | struct lxc_handler *hdlr = data; | |
15cd25fd | 382 | |
489f39be | 383 | ret = lxc_read_nointr(fd, &siginfo, sizeof(siginfo)); |
818a57fc CB |
384 | if (ret < 0) |
385 | return log_error(LXC_MAINLOOP_ERROR, "Failed to read signal info from signal file descriptor %d", fd); | |
15cd25fd | 386 | |
818a57fc | 387 | if (ret != sizeof(siginfo)) |
f7a97743 | 388 | return log_error(LXC_MAINLOOP_ERROR, "Unexpected size for struct signalfd_siginfo"); |
15cd25fd | 389 | |
408da065 | 390 | /* Check whether init is running. */ |
80507ee8 | 391 | info.si_pid = 0; |
3c319edb CB |
392 | ret = waitid(P_PID, hdlr->pid, &info, WEXITED | WNOWAIT | WNOHANG); |
393 | if (ret == 0 && info.si_pid == hdlr->pid) | |
394 | hdlr->init_died = true; | |
80507ee8 | 395 | |
543d2f83 CB |
396 | TRACE("Received signal ssi_signo(%d) for ssi_pid(%d), si_signo(%d), si_pid(%d)", |
397 | siginfo.ssi_signo, siginfo.ssi_pid, info.si_signo, info.si_pid); | |
398 | ||
cd5177e9 TA |
399 | /* Try to figure out a reasonable exit status to report. */ |
400 | if (hdlr->init_died) { | |
401 | switch (info.si_code) { | |
402 | case CLD_EXITED: | |
403 | hdlr->exit_status = info.si_status << 8; | |
404 | break; | |
405 | case CLD_KILLED: | |
406 | case CLD_DUMPED: | |
407 | case CLD_STOPPED: | |
408 | hdlr->exit_status = info.si_status << 8 | 0x7f; | |
409 | break; | |
410 | case CLD_CONTINUED: | |
411 | /* Huh? The waitid() told us it's dead *and* continued? */ | |
412 | WARN("Init %d dead and continued?", hdlr->pid); | |
413 | hdlr->exit_status = 1; | |
414 | break; | |
415 | default: | |
20993a97 | 416 | ERROR("Unknown si_code: %d", info.si_code); |
55e1311e | 417 | hdlr->exit_status = 1; |
cd5177e9 TA |
418 | } |
419 | } | |
420 | ||
d4b5d7a8 | 421 | if (siginfo.ssi_signo == SIGHUP) { |
33942046 CB |
422 | if (hdlr->pidfd >= 0) |
423 | lxc_raw_pidfd_send_signal(hdlr->pidfd, SIGTERM, NULL, 0); | |
d9bb2fba CB |
424 | else |
425 | kill(hdlr->pid, SIGTERM); | |
d4b5d7a8 | 426 | INFO("Killing %d since terminal hung up", hdlr->pid); |
9b10bef5 CB |
427 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE |
428 | : LXC_MAINLOOP_CONTINUE; | |
d4b5d7a8 TA |
429 | } |
430 | ||
9cb94384 | 431 | if (siginfo.ssi_signo != SIGCHLD) { |
33942046 CB |
432 | if (hdlr->pidfd >= 0) |
433 | lxc_raw_pidfd_send_signal(hdlr->pidfd, | |
434 | siginfo.ssi_signo, NULL, 0); | |
d9bb2fba CB |
435 | else |
436 | kill(hdlr->pid, siginfo.ssi_signo); | |
9cb94384 | 437 | INFO("Forwarded signal %d to pid %d", siginfo.ssi_signo, hdlr->pid); |
9b10bef5 CB |
438 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE |
439 | : LXC_MAINLOOP_CONTINUE; | |
9cb94384 TA |
440 | } |
441 | ||
3a9e949f TA |
442 | /* More robustness, protect ourself from a SIGCHLD sent |
443 | * by a process different from the container init. | |
444 | */ | |
a15c3609 | 445 | if ((__u64)siginfo.ssi_pid != (__u64)hdlr->pid) { |
6e94162a CB |
446 | NOTICE("Received %d from pid %d instead of container init %d", |
447 | siginfo.ssi_signo, siginfo.ssi_pid, hdlr->pid); | |
9b10bef5 CB |
448 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE |
449 | : LXC_MAINLOOP_CONTINUE; | |
3a9e949f TA |
450 | } |
451 | ||
408da065 | 452 | if (siginfo.ssi_code == CLD_STOPPED) { |
6e94162a | 453 | INFO("Container init process was stopped"); |
9b10bef5 CB |
454 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE |
455 | : LXC_MAINLOOP_CONTINUE; | |
456 | } | |
457 | ||
458 | if (siginfo.ssi_code == CLD_CONTINUED) { | |
6e94162a | 459 | INFO("Container init process was continued"); |
9b10bef5 CB |
460 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE |
461 | : LXC_MAINLOOP_CONTINUE; | |
15cd25fd | 462 | } |
1ac470c0 | 463 | |
818a57fc | 464 | return log_debug(LXC_MAINLOOP_CLOSE, "Container init process %d exited", hdlr->pid); |
b0a33c1e | 465 | } |
466 | ||
974a8aba CB |
467 | int lxc_serve_state_clients(const char *name, struct lxc_handler *handler, |
468 | lxc_state_t state) | |
66aeffc7 | 469 | { |
d2e75eba CB |
470 | struct lxc_msg msg = { |
471 | .type = lxc_msg_state, | |
472 | .value = state, | |
473 | }; | |
9de31d5a | 474 | size_t retlen; |
dbc9832d | 475 | ssize_t ret; |
d2e75eba | 476 | struct lxc_state_client *client, *nclient; |
dbc9832d | 477 | |
974a8aba CB |
478 | if (state == THAWED) |
479 | handler->state = RUNNING; | |
480 | else | |
481 | handler->state = state; | |
482 | ||
9dfa4041 | 483 | TRACE("Set container state to %s", lxc_state2str(state)); |
dbc9832d | 484 | |
d2e75eba | 485 | if (list_empty(&handler->conf->state_clients)) |
818a57fc | 486 | return log_trace(0, "No state clients registered"); |
dbc9832d | 487 | |
9de31d5a CB |
488 | retlen = strlcpy(msg.name, name, sizeof(msg.name)); |
489 | if (retlen >= sizeof(msg.name)) | |
490 | return -E2BIG; | |
dbc9832d | 491 | |
d2e75eba | 492 | list_for_each_entry_safe(client, nclient, &handler->conf->state_clients, head) { |
d39b10eb | 493 | if (client->states[state] == 0) { |
9dfa4041 | 494 | TRACE("State %s not registered for state client %d", |
dbc9832d CB |
495 | lxc_state2str(state), client->clientfd); |
496 | continue; | |
497 | } | |
498 | ||
9dfa4041 | 499 | TRACE("Sending state %s to state client %d", |
dbc9832d CB |
500 | lxc_state2str(state), client->clientfd); |
501 | ||
7fbb15ec CB |
502 | ret = lxc_send_nointr(client->clientfd, &msg, sizeof(msg), MSG_NOSIGNAL); |
503 | if (ret <= 0) | |
6d1400b5 | 504 | SYSERROR("Failed to send message to client"); |
dbc9832d CB |
505 | |
506 | /* kick client from list */ | |
d2e75eba | 507 | list_del(&client->head); |
300d1cb4 | 508 | close(client->clientfd); |
d2e75eba | 509 | free(client); |
dbc9832d | 510 | } |
dbc9832d | 511 | |
5e5576a4 CB |
512 | return 0; |
513 | } | |
514 | ||
515 | static int lxc_serve_state_socket_pair(const char *name, | |
516 | struct lxc_handler *handler, | |
517 | lxc_state_t state) | |
518 | { | |
519 | ssize_t ret; | |
520 | ||
bb955810 | 521 | if (!handler->daemonize || |
5e5576a4 CB |
522 | handler->state_socket_pair[1] < 0 || |
523 | state == STARTING) | |
524 | return 0; | |
525 | ||
526 | /* Close read end of the socket pair. */ | |
a783a414 | 527 | close_prot_errno_disarm(handler->state_socket_pair[0]); |
5e5576a4 | 528 | |
ee8377bd | 529 | again: |
5e5576a4 CB |
530 | ret = lxc_abstract_unix_send_credential(handler->state_socket_pair[1], |
531 | &(int){state}, sizeof(int)); | |
9044b79e | 532 | if (ret < 0) { |
533 | SYSERROR("Failed to send state to %d", handler->state_socket_pair[1]); | |
534 | ||
ee8377bd CB |
535 | if (errno == EINTR) |
536 | goto again; | |
9044b79e | 537 | |
538 | return -1; | |
539 | } | |
540 | ||
818a57fc CB |
541 | if (ret != sizeof(int)) |
542 | return log_error(-1, "Message too long : %d", handler->state_socket_pair[1]); | |
5e5576a4 CB |
543 | |
544 | TRACE("Sent container state \"%s\" to %d", lxc_state2str(state), | |
545 | handler->state_socket_pair[1]); | |
546 | ||
547 | /* Close write end of the socket pair. */ | |
818a57fc | 548 | close_prot_errno_disarm(handler->state_socket_pair[1]); |
5e5576a4 CB |
549 | |
550 | return 0; | |
551 | } | |
552 | ||
553 | int lxc_set_state(const char *name, struct lxc_handler *handler, | |
554 | lxc_state_t state) | |
555 | { | |
556 | int ret; | |
557 | ||
558 | ret = lxc_serve_state_socket_pair(name, handler, state); | |
818a57fc CB |
559 | if (ret < 0) |
560 | return log_error(-1, "Failed to synchronize via anonymous pair of unix sockets"); | |
5e5576a4 CB |
561 | |
562 | ret = lxc_serve_state_clients(name, handler, state); | |
563 | if (ret < 0) | |
564 | return -1; | |
565 | ||
dbc9832d CB |
566 | /* This function will try to connect to the legacy lxc-monitord state |
567 | * server and only exists for backwards compatibility. | |
568 | */ | |
9123e471 | 569 | lxc_monitor_send_state(name, state, handler->lxcpath); |
dbc9832d | 570 | |
66aeffc7 DL |
571 | return 0; |
572 | } | |
573 | ||
735f2c6e | 574 | int lxc_poll(const char *name, struct lxc_handler *handler) |
b0a33c1e | 575 | { |
3c319edb | 576 | int ret; |
543d2f83 | 577 | struct lxc_terminal *console = &handler->conf->console; |
3298b37d | 578 | struct lxc_async_descr descr, descr_console; |
b0a33c1e | 579 | |
543d2f83 CB |
580 | if (!wants_console(console)) |
581 | console = NULL; | |
86530b0a | 582 | |
3c319edb CB |
583 | ret = lxc_mainloop_open(&descr); |
584 | if (ret < 0) { | |
585 | ERROR("Failed to create mainloop"); | |
50c8bf05 | 586 | goto out_sigfd; |
b0a33c1e | 587 | } |
588 | ||
543d2f83 | 589 | if (console) { |
30a33fbd CB |
590 | ret = lxc_mainloop_open(&descr_console); |
591 | if (ret < 0) { | |
592 | ERROR("Failed to create console mainloop"); | |
593 | goto out_mainloop; | |
594 | } | |
3c319edb CB |
595 | } |
596 | ||
543d2f83 CB |
597 | ret = lxc_mainloop_add_handler(&descr, handler->sigfd, |
598 | signal_handler, | |
599 | default_cleanup_handler, | |
600 | handler, "signal_handler"); | |
3c319edb | 601 | if (ret < 0) { |
1cc8bd4b | 602 | ERROR("Failed to add signal handler for %d to mainloop", handler->sigfd); |
3c319edb | 603 | goto out_mainloop_console; |
b0a33c1e | 604 | } |
605 | ||
2ac0f627 CB |
606 | ret = lxc_seccomp_setup_proxy(&handler->conf->seccomp, &descr, handler); |
607 | if (ret < 0) { | |
608 | ERROR("Failed to setup seccomp proxy"); | |
c3e3c21a | 609 | goto out_mainloop_console; |
2ac0f627 | 610 | } |
cdb2a47f | 611 | |
543d2f83 | 612 | if (console) { |
093bce5f | 613 | ret = lxc_terminal_mainloop_add(&descr, console); |
30a33fbd CB |
614 | if (ret < 0) { |
615 | ERROR("Failed to add console handlers to mainloop"); | |
616 | goto out_mainloop_console; | |
617 | } | |
563f2f2c DL |
618 | } |
619 | ||
3c319edb CB |
620 | ret = lxc_cmd_mainloop_add(name, &descr, handler); |
621 | if (ret < 0) { | |
622 | ERROR("Failed to add command handler to mainloop"); | |
623 | goto out_mainloop_console; | |
624 | } | |
b0a33c1e | 625 | |
3c319edb | 626 | TRACE("Mainloop is ready"); |
b0a33c1e | 627 | |
3c319edb | 628 | ret = lxc_mainloop(&descr, -1); |
a585382b CB |
629 | if (descr.type == LXC_MAINLOOP_EPOLL) |
630 | close_prot_errno_disarm(descr.epfd); | |
1cc8bd4b | 631 | if (ret < 0 || !handler->init_died) |
42b09f94 | 632 | goto out_mainloop_console; |
1cc8bd4b | 633 | |
543d2f83 CB |
634 | if (console) { |
635 | ret = lxc_terminal_mainloop_add(&descr_console, console); | |
636 | if (ret == 0) | |
637 | ret = lxc_mainloop(&descr_console, 0); | |
638 | } | |
30a33fbd | 639 | |
3c319edb | 640 | out_mainloop_console: |
543d2f83 | 641 | if (console) { |
30a33fbd CB |
642 | lxc_mainloop_close(&descr_console); |
643 | TRACE("Closed console mainloop"); | |
644 | } | |
3c319edb | 645 | |
42b09f94 CB |
646 | out_mainloop: |
647 | lxc_mainloop_close(&descr); | |
648 | TRACE("Closed mainloop"); | |
649 | ||
b0a33c1e | 650 | out_sigfd: |
1cc8bd4b | 651 | TRACE("Closed signal file descriptor %d", handler->sigfd); |
a783a414 | 652 | close_prot_errno_disarm(handler->sigfd); |
408da065 | 653 | |
1cc8bd4b | 654 | return ret; |
b0a33c1e | 655 | } |
656 | ||
a42abcce | 657 | void lxc_put_handler(struct lxc_handler *handler) |
f2e07cb6 | 658 | { |
a783a414 CB |
659 | close_prot_errno_disarm(handler->pidfd); |
660 | close_prot_errno_disarm(handler->sigfd); | |
1cc8bd4b | 661 | lxc_put_nsfds(handler); |
80308d07 | 662 | if (handler->conf && handler->conf->reboot == REBOOT_NONE) |
a783a414 CB |
663 | close_prot_errno_disarm(handler->conf->maincmd_fd); |
664 | close_prot_errno_disarm(handler->monitor_status_fd); | |
665 | close_prot_errno_disarm(handler->state_socket_pair[0]); | |
666 | close_prot_errno_disarm(handler->state_socket_pair[1]); | |
667 | cgroup_exit(handler->cgroup_ops); | |
a42abcce CB |
668 | if (handler->conf && handler->conf->reboot == REBOOT_NONE) |
669 | free_disarm(handler); | |
670 | else | |
671 | handler->conf = NULL; | |
f2e07cb6 CB |
672 | } |
673 | ||
a42abcce CB |
674 | struct lxc_handler *lxc_init_handler(struct lxc_handler *old, |
675 | const char *name, struct lxc_conf *conf, | |
5e5576a4 | 676 | const char *lxcpath, bool daemonize) |
59eb99ba | 677 | { |
85c279bb | 678 | int nr_keep_fds = 0; |
818a57fc | 679 | int ret; |
3a0f472d DL |
680 | struct lxc_handler *handler; |
681 | ||
a42abcce CB |
682 | if (!old) |
683 | handler = zalloc(sizeof(*handler)); | |
684 | else | |
685 | handler = old; | |
fdecbc9c | 686 | if (!handler) |
3a0f472d | 687 | return NULL; |
9044b79e | 688 | |
fdecbc9c CB |
689 | /* Note that am_guest_unpriv() checks the effective uid. We |
690 | * probably don't care if we are real root only if we are running | |
691 | * as root so this should be fine. | |
790255cf | 692 | */ |
e0010464 | 693 | handler->am_root = !am_guest_unpriv(); |
fae349da | 694 | handler->conf = conf; |
9123e471 | 695 | handler->lxcpath = lxcpath; |
818a57fc CB |
696 | handler->init_died = false; |
697 | handler->data_sock[0] = -EBADF; | |
698 | handler->data_sock[1] = -EBADF; | |
4d8bdfa0 | 699 | handler->monitor_status_fd = -EBADF; |
33942046 | 700 | handler->pidfd = -EBADF; |
1cc8bd4b | 701 | handler->sigfd = -EBADF; |
818a57fc CB |
702 | handler->state_socket_pair[0] = -EBADF; |
703 | handler->state_socket_pair[1] = -EBADF; | |
80308d07 | 704 | if (handler->conf->reboot == REBOOT_NONE) |
d2e75eba | 705 | INIT_LIST_HEAD(&handler->conf->state_clients); |
fae349da | 706 | |
8db6be1b CB |
707 | for (lxc_namespace_t idx = 0; idx < LXC_NS_MAX; idx++) { |
708 | handler->nsfd[idx] = -EBADF; | |
709 | ||
710 | if (handler->conf->reboot == REBOOT_NONE) | |
711 | continue; | |
712 | ||
713 | handler->nsfd_paths[idx][0] = '\0'; | |
714 | handler->hook_argv[idx] = NULL; | |
715 | ||
716 | if (handler->hook_argc != 0) | |
717 | handler->hook_argc = 0; | |
718 | } | |
b6b2b194 | 719 | |
f0ecc19d | 720 | handler->name = name; |
c581d2a6 CB |
721 | if (daemonize) |
722 | handler->transient_pid = lxc_raw_getpid(); | |
723 | else | |
724 | handler->transient_pid = -1; | |
3bdf52d7 | 725 | |
80308d07 | 726 | if (daemonize && handler->conf->reboot == REBOOT_NONE) { |
5e5576a4 | 727 | /* Create socketpair() to synchronize on daemonized startup. |
fdecbc9c CB |
728 | * When the container reboots we don't need to synchronize |
729 | * again currently so don't open another socketpair(). | |
5e5576a4 CB |
730 | */ |
731 | ret = socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0, | |
732 | handler->state_socket_pair); | |
733 | if (ret < 0) { | |
734 | ERROR("Failed to create anonymous pair of unix sockets"); | |
735 | goto on_error; | |
736 | } | |
9044b79e | 737 | |
5e5576a4 CB |
738 | TRACE("Created anonymous pair {%d,%d} of unix sockets", |
739 | handler->state_socket_pair[0], | |
740 | handler->state_socket_pair[1]); | |
85c279bb CB |
741 | handler->keep_fds[nr_keep_fds++] = handler->state_socket_pair[0]; |
742 | handler->keep_fds[nr_keep_fds++] = handler->state_socket_pair[1]; | |
5e5576a4 CB |
743 | } |
744 | ||
80308d07 | 745 | if (handler->conf->reboot == REBOOT_NONE) { |
6834acff | 746 | handler->conf->maincmd_fd = lxc_server_init(name, lxcpath, "command"); |
d39b10eb CB |
747 | if (handler->conf->maincmd_fd < 0) { |
748 | ERROR("Failed to set up command socket"); | |
749 | goto on_error; | |
750 | } | |
85c279bb | 751 | handler->keep_fds[nr_keep_fds++] = handler->conf->maincmd_fd; |
aa460476 | 752 | } |
9044b79e | 753 | |
9dfa4041 | 754 | TRACE("Unix domain socket %d for command server is ready", |
aa460476 CB |
755 | handler->conf->maincmd_fd); |
756 | ||
757 | return handler; | |
758 | ||
f2e07cb6 | 759 | on_error: |
a42abcce | 760 | lxc_put_handler(handler); |
aa460476 CB |
761 | |
762 | return NULL; | |
763 | } | |
764 | ||
765 | int lxc_init(const char *name, struct lxc_handler *handler) | |
766 | { | |
f62cf1d4 | 767 | __do_close int status_fd = -EBADF; |
a2c09be0 | 768 | int ret; |
4a03ded4 | 769 | const char *loglevel; |
aa460476 CB |
770 | struct lxc_conf *conf = handler->conf; |
771 | ||
434c8e15 | 772 | handler->monitor_pid = lxc_raw_getpid(); |
4d8bdfa0 | 773 | status_fd = open("/proc/self/status", O_RDONLY | O_CLOEXEC); |
59b1b67e CB |
774 | if (status_fd < 0) |
775 | return log_error_errno(-1, errno, "Failed to open monitor status fd"); | |
434c8e15 | 776 | |
4eb19ac0 | 777 | handler->lsm_ops = lsm_init_static(); |
2170b263 | 778 | TRACE("Initialized LSM"); |
d2e30e99 | 779 | |
408da065 | 780 | /* Begin by setting the state to STARTING. */ |
2170b263 | 781 | ret = lxc_set_state(name, handler, STARTING); |
59b1b67e CB |
782 | if (ret < 0) |
783 | return log_error(-1, "Failed to set state to \"%s\"", lxc_state2str(STARTING)); | |
46800e77 | 784 | TRACE("Set container state to \"STARTING\""); |
0ad19a3f | 785 | |
408da065 | 786 | /* Start of environment variable setup for hooks. */ |
24c2a48d CB |
787 | ret = setenv("LXC_NAME", name, 1); |
788 | if (ret < 0) | |
789 | SYSERROR("Failed to set environment variable: LXC_NAME=%s", name); | |
408da065 | 790 | |
2170b263 CB |
791 | if (conf->rcfile) { |
792 | ret = setenv("LXC_CONFIG_FILE", conf->rcfile, 1); | |
793 | if (ret < 0) | |
818a57fc | 794 | SYSERROR("Failed to set environment variable: LXC_CONFIG_FILE=%s", conf->rcfile); |
2170b263 | 795 | } |
408da065 | 796 | |
2170b263 CB |
797 | if (conf->rootfs.mount) { |
798 | ret = setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1); | |
799 | if (ret < 0) | |
818a57fc | 800 | SYSERROR("Failed to set environment variable: LXC_ROOTFS_MOUNT=%s", conf->rootfs.mount); |
2170b263 | 801 | } |
408da065 | 802 | |
2170b263 CB |
803 | if (conf->rootfs.path) { |
804 | ret = setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1); | |
805 | if (ret < 0) | |
818a57fc | 806 | SYSERROR("Failed to set environment variable: LXC_ROOTFS_PATH=%s", conf->rootfs.path); |
2170b263 | 807 | } |
408da065 | 808 | |
2170b263 CB |
809 | if (conf->console.path) { |
810 | ret = setenv("LXC_CONSOLE", conf->console.path, 1); | |
811 | if (ret < 0) | |
818a57fc | 812 | SYSERROR("Failed to set environment variable: LXC_CONSOLE=%s", conf->console.path); |
2170b263 | 813 | } |
408da065 | 814 | |
2170b263 CB |
815 | if (conf->console.log_path) { |
816 | ret = setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1); | |
817 | if (ret < 0) | |
818a57fc | 818 | SYSERROR("Failed to set environment variable: LXC_CONSOLE_LOGPATH=%s", conf->console.log_path); |
2170b263 | 819 | } |
408da065 | 820 | |
2170b263 CB |
821 | if (cgns_supported()) { |
822 | ret = setenv("LXC_CGNS_AWARE", "1", 1); | |
823 | if (ret < 0) | |
818a57fc | 824 | SYSERROR("Failed to set environment variable LXC_CGNS_AWARE=1"); |
2170b263 | 825 | } |
b8f88d9b | 826 | |
4a03ded4 | 827 | loglevel = lxc_log_priority_to_string(lxc_log_get_level()); |
2170b263 CB |
828 | ret = setenv("LXC_LOG_LEVEL", loglevel, 1); |
829 | if (ret < 0) | |
818a57fc | 830 | SYSERROR("Set environment variable LXC_LOG_LEVEL=%s", loglevel); |
a2c09be0 CB |
831 | |
832 | if (conf->hooks_version == 0) | |
833 | ret = setenv("LXC_HOOK_VERSION", "0", 1); | |
834 | else | |
835 | ret = setenv("LXC_HOOK_VERSION", "1", 1); | |
836 | if (ret < 0) | |
837 | SYSERROR("Failed to set environment variable LXC_HOOK_VERSION=%u", conf->hooks_version); | |
408da065 | 838 | /* End of environment variable setup for hooks. */ |
f7bee6c6 | 839 | |
2170b263 | 840 | TRACE("Set environment variables"); |
dbc9832d | 841 | |
2170b263 | 842 | ret = run_lxc_hooks(name, "pre-start", conf, NULL); |
65a2dc8b CB |
843 | if (ret < 0) |
844 | return log_error(-1, "Failed to run lxc.hook.pre-start for container \"%s\"", name); | |
2170b263 | 845 | TRACE("Ran pre-start hooks"); |
26ddeedd | 846 | |
9f77617b CB |
847 | ret = lxc_terminal_parent(conf); |
848 | if (ret < 0) | |
849 | return log_error(-1, "Failed to allocate terminal"); | |
850 | ||
408da065 CB |
851 | /* The signal fd has to be created before forking otherwise if the child |
852 | * process exits before we setup the signal fd, the event will be lost | |
853 | * and the command will be stuck. | |
854 | */ | |
83ee7875 | 855 | handler->sigfd = setup_signal_fd(&handler->oldmask); |
65a2dc8b CB |
856 | if (handler->sigfd < 0) |
857 | return log_error(-1, "Failed to setup SIGCHLD fd handler."); | |
2170b263 | 858 | TRACE("Set up signal fd"); |
b5159817 | 859 | |
5a087e05 | 860 | handler->cgroup_ops = cgroup_init(handler->conf); |
2202afc9 CB |
861 | if (!handler->cgroup_ops) { |
862 | ERROR("Failed to initialize cgroup driver"); | |
9f77617b | 863 | goto out_restore_sigmask; |
2202afc9 CB |
864 | } |
865 | TRACE("Initialized cgroup driver"); | |
866 | ||
2e1361a6 | 867 | ret = lxc_read_seccomp_config(conf); |
9f77617b CB |
868 | if (ret < 0) { |
869 | ERROR("Failed to read seccomp policy"); | |
870 | goto out_restore_sigmask; | |
871 | } | |
2e1361a6 CB |
872 | TRACE("Read seccomp policy"); |
873 | ||
af04d847 | 874 | ret = handler->lsm_ops->prepare(handler->lsm_ops, conf, handler->lxcpath); |
1800f924 WB |
875 | if (ret < 0) { |
876 | ERROR("Failed to initialize LSM"); | |
9f77617b | 877 | goto out_restore_sigmask; |
1800f924 WB |
878 | } |
879 | TRACE("Initialized LSM"); | |
880 | ||
2170b263 | 881 | INFO("Container \"%s\" is initialized", name); |
4d8bdfa0 | 882 | handler->monitor_status_fd = move_fd(status_fd); |
aa460476 | 883 | return 0; |
59eb99ba | 884 | |
b5159817 | 885 | out_restore_sigmask: |
4c496daa | 886 | (void)pthread_sigmask(SIG_SETMASK, &handler->oldmask, NULL); |
ea918412 | 887 | |
aa460476 | 888 | return -1; |
59eb99ba DL |
889 | } |
890 | ||
8db6be1b CB |
891 | void lxc_expose_namespace_environment(const struct lxc_handler *handler) |
892 | { | |
893 | for (lxc_namespace_t i = 0; i < LXC_NS_MAX; i++) { | |
894 | int ret; | |
895 | const char *fd_path; | |
896 | ||
897 | if (handler->nsfd[i] < 0) | |
898 | continue; | |
899 | ||
900 | fd_path = handler->nsfd_paths[i] + strcspn(handler->nsfd_paths[i], "/"); | |
901 | ret = setenv(ns_info[i].env_name, fd_path, 1); | |
902 | if (ret < 0) | |
903 | SYSERROR("Failed to set environment variable %s=%s", | |
904 | ns_info[i].env_name, fd_path); | |
905 | else | |
906 | TRACE("Set environment variable %s=%s", | |
907 | ns_info[i].env_name, fd_path); | |
908 | } | |
909 | } | |
910 | ||
fd5be714 | 911 | void lxc_end(struct lxc_handler *handler) |
59eb99ba | 912 | { |
818a57fc | 913 | int ret; |
0c5859ff | 914 | const char *name = handler->name; |
2202afc9 | 915 | struct cgroup_ops *cgroup_ops = handler->cgroup_ops; |
d2e75eba | 916 | struct lxc_state_client *client, *nclient; |
b6b2b194 | 917 | |
408da065 CB |
918 | /* The STOPPING state is there for future cleanup code which can take |
919 | * awhile. | |
59eb99ba | 920 | */ |
25c2aca5 | 921 | lxc_set_state(name, handler, STOPPING); |
b6b2b194 | 922 | |
8db6be1b CB |
923 | /* Passing information to hooks via environment variables. */ |
924 | if (handler->conf->hooks_version > 0) | |
925 | lxc_expose_namespace_environment(handler); | |
c154af98 | 926 | |
80308d07 | 927 | if (handler->conf->reboot > REBOOT_NONE) { |
c097467f CB |
928 | ret = setenv("LXC_TARGET", "reboot", 1); |
929 | if (ret < 0) | |
818a57fc | 930 | SYSERROR("Failed to set environment variable: LXC_TARGET=reboot"); |
c097467f | 931 | } |
408da065 | 932 | |
80308d07 | 933 | if (handler->conf->reboot == REBOOT_NONE) { |
c097467f CB |
934 | ret = setenv("LXC_TARGET", "stop", 1); |
935 | if (ret < 0) | |
818a57fc | 936 | SYSERROR("Failed to set environment variable: LXC_TARGET=stop"); |
c097467f | 937 | } |
c154af98 | 938 | |
18b3b9c1 | 939 | if (handler->conf->hooks_version == 0) |
8db6be1b | 940 | ret = run_lxc_hooks(name, "stop", handler->conf, handler->hook_argv); |
18b3b9c1 | 941 | else |
c097467f CB |
942 | ret = run_lxc_hooks(name, "stop", handler->conf, NULL); |
943 | if (ret < 0) | |
944 | ERROR("Failed to run \"lxc.hook.stop\" hook"); | |
c154af98 | 945 | |
af04d847 | 946 | handler->lsm_ops->cleanup(handler->lsm_ops, handler->conf, handler->lxcpath); |
1800f924 | 947 | |
e2aed383 WB |
948 | if (cgroup_ops) { |
949 | cgroup_ops->payload_destroy(cgroup_ops, handler); | |
950 | cgroup_ops->monitor_destroy(cgroup_ops, handler); | |
951 | } | |
4288b79f | 952 | |
79ff643d CB |
953 | put_lxc_rootfs(&handler->conf->rootfs, true); |
954 | ||
80308d07 | 955 | if (handler->conf->reboot == REBOOT_NONE) { |
f893d898 CB |
956 | /* For all new state clients simply close the command socket. |
957 | * This will inform all state clients that the container is | |
958 | * STOPPED and also prevents a race between a open()/close() on | |
959 | * the command socket causing a new process to get ECONNREFUSED | |
960 | * because we haven't yet closed the command socket. | |
961 | */ | |
b87ee312 | 962 | close_prot_errno_disarm(handler->conf->maincmd_fd); |
c3184275 CB |
963 | TRACE("Closed command socket"); |
964 | ||
965 | /* This function will try to connect to the legacy lxc-monitord | |
966 | * state server and only exists for backwards compatibility. | |
967 | */ | |
968 | lxc_monitor_send_state(name, STOPPED, handler->lxcpath); | |
969 | ||
970 | /* The command socket is closed so no one can acces the command | |
971 | * socket anymore so there's no need to lock it. | |
972 | */ | |
973 | handler->state = STOPPED; | |
974 | TRACE("Set container state to \"STOPPED\""); | |
f893d898 CB |
975 | } else { |
976 | lxc_set_state(name, handler, STOPPED); | |
695d0e56 | 977 | TRACE("Set container state to \"STOPPED\""); |
d39b10eb | 978 | } |
4288b79f | 979 | |
048493a3 CB |
980 | /* Avoid lingering namespace references. */ |
981 | lxc_put_nsfds(handler); | |
982 | ||
46800e77 CB |
983 | ret = run_lxc_hooks(name, "post-stop", handler->conf, NULL); |
984 | if (ret < 0) { | |
c097467f | 985 | ERROR("Failed to run lxc.hook.post-stop for container \"%s\"", name); |
80308d07 | 986 | if (handler->conf->reboot > REBOOT_NONE) { |
c097467f | 987 | WARN("Container will be stopped instead of rebooted"); |
80308d07 | 988 | handler->conf->reboot = REBOOT_NONE; |
c097467f CB |
989 | |
990 | ret = setenv("LXC_TARGET", "stop", 1); | |
991 | if (ret < 0) | |
818a57fc | 992 | WARN("Failed to set environment variable: LXC_TARGET=stop"); |
f3787121 CB |
993 | } |
994 | } | |
26ddeedd | 995 | |
408da065 | 996 | /* Reset mask set by setup_signal_fd. */ |
b467714b | 997 | ret = pthread_sigmask(SIG_SETMASK, &handler->oldmask, NULL); |
c097467f | 998 | if (ret < 0) |
a24c5678 | 999 | SYSWARN("Failed to restore signal mask"); |
8f64a3f6 | 1000 | |
2aac071b | 1001 | lxc_terminal_delete(&handler->conf->console); |
0e4be3cf | 1002 | lxc_delete_tty(&handler->conf->ttys); |
f797f05e | 1003 | close_prot_errno_disarm(handler->conf->devpts_fd); |
dbc9832d | 1004 | |
dbc9832d CB |
1005 | /* The command socket is now closed, no more state clients can register |
1006 | * themselves from now on. So free the list of state clients. | |
1007 | */ | |
d2e75eba | 1008 | list_for_each_entry_safe(client, nclient, &handler->conf->state_clients, head) { |
d39b10eb | 1009 | /* Keep state clients that want to be notified about reboots. */ |
80308d07 CB |
1010 | if ((handler->conf->reboot > REBOOT_NONE) && |
1011 | (client->states[RUNNING] == 2)) | |
d39b10eb CB |
1012 | continue; |
1013 | ||
dbc9832d | 1014 | /* close state client socket */ |
d2e75eba | 1015 | list_del(&client->head); |
c097467f | 1016 | close(client->clientfd); |
d2e75eba | 1017 | free(client); |
dbc9832d CB |
1018 | } |
1019 | ||
80308d07 | 1020 | if (handler->conf->ephemeral == 1 && handler->conf->reboot != REBOOT_REQ) |
28272964 | 1021 | lxc_destroy_container_on_signal(handler, name); |
2c5f2ede | 1022 | |
a42abcce | 1023 | lxc_put_handler(handler); |
59eb99ba DL |
1024 | } |
1025 | ||
0c5859ff | 1026 | void lxc_abort(struct lxc_handler *handler) |
59eb99ba | 1027 | { |
33942046 CB |
1028 | int ret = 0; |
1029 | int status; | |
73e608b2 | 1030 | |
0c5859ff | 1031 | lxc_set_state(handler->name, handler, ABORTING); |
0e4f9d51 | 1032 | |
c718fac1 | 1033 | if (handler->pidfd >= 0) { |
33942046 | 1034 | ret = lxc_raw_pidfd_send_signal(handler->pidfd, SIGKILL, NULL, 0); |
c718fac1 | 1035 | if (ret) |
11c7d349 CB |
1036 | SYSWARN("Failed to send SIGKILL via pidfd %d for process %d", |
1037 | handler->pidfd, handler->pid); | |
c718fac1 CB |
1038 | } |
1039 | ||
65a2dc8b | 1040 | if ((!ret || errno != ESRCH) && handler->pid > 0) |
11c7d349 CB |
1041 | if (kill(handler->pid, SIGKILL)) |
1042 | SYSWARN("Failed to send SIGKILL to %d", handler->pid); | |
0e4f9d51 | 1043 | |
33942046 CB |
1044 | do { |
1045 | ret = waitpid(-1, &status, 0); | |
1046 | } while (ret > 0); | |
59eb99ba DL |
1047 | } |
1048 | ||
ffe1e01a | 1049 | static int do_start(void *data) |
50e98013 | 1050 | { |
cdb2a47f | 1051 | struct lxc_handler *handler = data; |
f62cf1d4 | 1052 | __lxc_unused __do_close int data_sock0 = handler->data_sock[0], |
c294a68d | 1053 | data_sock1 = handler->data_sock[1]; |
d3103162 | 1054 | __do_close int devnull_fd = -EBADF, status_fd = -EBADF; |
8bf3abfb | 1055 | int ret; |
928b1f04 YT |
1056 | uid_t new_uid; |
1057 | gid_t new_gid; | |
964581c2 CB |
1058 | uid_t nsuid = 0; |
1059 | gid_t nsgid = 0; | |
50e98013 | 1060 | |
18225d19 | 1061 | lxc_sync_fini_parent(handler); |
50e98013 | 1062 | |
d17c815d | 1063 | if (lxc_abstract_unix_recv_one_fd(data_sock1, &status_fd, NULL, 0) < 0) { |
77093b7c | 1064 | ERROR("Failed to receive status file descriptor from parent process"); |
4d8bdfa0 CB |
1065 | goto out_warn_father; |
1066 | } | |
1067 | ||
408da065 CB |
1068 | /* This prctl must be before the synchro, so if the parent dies before |
1069 | * we set the parent death signal, we will detect its death with the | |
1070 | * synchro right after, otherwise we have a window where the parent can | |
1071 | * exit before we set the pdeath signal leading to a unsupervized | |
1072 | * container. | |
743ecd2e | 1073 | */ |
4d8bdfa0 | 1074 | ret = lxc_set_death_signal(SIGKILL, handler->monitor_pid, status_fd); |
912314fc CB |
1075 | if (ret < 0) { |
1076 | SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL"); | |
18225d19 | 1077 | goto out_warn_father; |
743ecd2e DL |
1078 | } |
1079 | ||
611ddd34 CB |
1080 | ret = lxc_ambient_caps_up(); |
1081 | if (ret < 0) { | |
6f5e532f | 1082 | ERROR("Failed to raise ambient capabilities"); |
611ddd34 CB |
1083 | goto out_warn_father; |
1084 | } | |
1085 | ||
b467714b | 1086 | ret = pthread_sigmask(SIG_SETMASK, &handler->oldmask, NULL); |
18225d19 CB |
1087 | if (ret < 0) { |
1088 | SYSERROR("Failed to set signal mask"); | |
1089 | goto out_warn_father; | |
1090 | } | |
50e98013 | 1091 | |
6e48e7c5 | 1092 | if (!lxc_sync_wait_parent(handler, START_SYNC_STARTUP)) |
18225d19 | 1093 | goto out_warn_father; |
5b1e83cb | 1094 | |
408da065 CB |
1095 | /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See |
1096 | * https://github.com/lxc/lxd/issues/1978. | |
1097 | */ | |
f7176c3e | 1098 | if (handler->ns_unshare_flags & CLONE_NEWNET) { |
5b1e83cb SH |
1099 | ret = unshare(CLONE_NEWNET); |
1100 | if (ret < 0) { | |
e96a536c | 1101 | SYSERROR("Failed to unshare CLONE_NEWNET"); |
5b1e83cb SH |
1102 | goto out_warn_father; |
1103 | } | |
e96a536c | 1104 | INFO("Unshared CLONE_NEWNET"); |
5b1e83cb SH |
1105 | } |
1106 | ||
408da065 | 1107 | /* If we are in a new user namespace, become root there to have |
d08f8d2f | 1108 | * privilege over our namespace. |
f6d3e3e4 | 1109 | */ |
0589d744 | 1110 | if (!list_empty(&handler->conf->id_map)) { |
964581c2 CB |
1111 | if (!handler->conf->root_nsuid_map) |
1112 | nsuid = handler->conf->init_uid; | |
1113 | ||
1114 | if (!handler->conf->root_nsgid_map) | |
1115 | nsgid = handler->conf->init_gid; | |
4160c3a0 | 1116 | |
d08f8d2f CB |
1117 | /* Drop groups only after we switched to a valid gid in the new |
1118 | * user namespace. | |
1119 | */ | |
8917c382 | 1120 | if (!lxc_drop_groups() && |
8af07f82 | 1121 | (handler->am_root || errno != EPERM)) |
c476bdce | 1122 | goto out_warn_father; |
4b826b1f | 1123 | |
b58214ac CB |
1124 | if (!lxc_switch_uid_gid(nsuid, nsgid)) |
1125 | goto out_warn_father; | |
1126 | ||
b81689a1 CB |
1127 | ret = prctl(PR_SET_DUMPABLE, prctl_arg(1), prctl_arg(0), |
1128 | prctl_arg(0), prctl_arg(0)); | |
d7883725 CB |
1129 | if (ret < 0) |
1130 | goto out_warn_father; | |
912314fc CB |
1131 | |
1132 | /* set{g,u}id() clears deathsignal */ | |
4d8bdfa0 | 1133 | ret = lxc_set_death_signal(SIGKILL, handler->monitor_pid, status_fd); |
912314fc CB |
1134 | if (ret < 0) { |
1135 | SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL"); | |
1136 | goto out_warn_father; | |
1137 | } | |
f6d3e3e4 SH |
1138 | } |
1139 | ||
e96a536c CB |
1140 | ret = access(handler->lxcpath, X_OK); |
1141 | if (ret != 0) { | |
c8154066 SH |
1142 | print_top_failing_dir(handler->lxcpath); |
1143 | goto out_warn_father; | |
1144 | } | |
1145 | ||
7a55c157 TA |
1146 | /* In order to checkpoint restore, we need to have everything in the |
1147 | * same mount namespace. However, some containers may not have a | |
1148 | * reasonable /dev (in particular, they may not have /dev/null), so we | |
1149 | * can't set init's std fds to /dev/null by opening it from inside the | |
1150 | * container. | |
1151 | * | |
1152 | * If that's the case, fall back to using the host's /dev/null. This | |
1153 | * means that migration won't work, but at least we won't spew output | |
1154 | * where it isn't wanted. | |
1155 | */ | |
bb955810 | 1156 | if (handler->daemonize && !handler->conf->autodev) { |
2806a87d | 1157 | char path[PATH_MAX]; |
3c09b97c | 1158 | |
fa60cd7b CB |
1159 | ret = strnprintf(path, sizeof(path), "%s/dev/null", |
1160 | handler->conf->rootfs.mount); | |
1161 | if (ret < 0) | |
2806a87d | 1162 | goto out_warn_father; |
3c09b97c | 1163 | |
e96a536c CB |
1164 | ret = access(path, F_OK); |
1165 | if (ret != 0) { | |
1166 | devnull_fd = open_devnull(); | |
c44de748 | 1167 | |
e96a536c CB |
1168 | if (devnull_fd < 0) |
1169 | goto out_warn_father; | |
818a57fc | 1170 | WARN("Using /dev/null from the host for container init's standard file descriptors. Migration will not work"); |
e96a536c | 1171 | } |
c44de748 AM |
1172 | } |
1173 | ||
6bc4165d CB |
1174 | /* |
1175 | * Tell the parent task it can begin to configure the container and wait | |
1176 | * for it to finish. | |
1177 | */ | |
1178 | if (!lxc_sync_wake_parent(handler, START_SYNC_CONFIGURE)) | |
c44de748 | 1179 | goto out_error; |
544a48a0 | 1180 | |
deefdf8a CB |
1181 | /* Unshare cgroup namespace after we have setup our cgroups. If we do it |
1182 | * earlier we end up with a wrong view of /proc/self/cgroup. For | |
1183 | * example, assume we unshare(CLONE_NEWCGROUP) first, and then create | |
1184 | * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then | |
1185 | * /proc/self/cgroup would show us: | |
1186 | * | |
1187 | * 8:cpuset:/lxc/c | |
1188 | * | |
1189 | * whereas it should actually show | |
1190 | * | |
1191 | * 8:cpuset:/ | |
1192 | */ | |
f7176c3e | 1193 | if (handler->ns_unshare_flags & CLONE_NEWCGROUP) { |
e96a536c CB |
1194 | ret = unshare(CLONE_NEWCGROUP); |
1195 | if (ret < 0) { | |
bca7c59c CB |
1196 | if (errno != EINVAL) { |
1197 | SYSERROR("Failed to unshare CLONE_NEWCGROUP"); | |
1198 | goto out_warn_father; | |
1199 | } | |
1200 | ||
1201 | handler->ns_clone_flags &= ~CLONE_NEWCGROUP; | |
1202 | SYSINFO("Kernel does not support CLONE_NEWCGROUP"); | |
1203 | } else { | |
1204 | INFO("Unshared CLONE_NEWCGROUP"); | |
deefdf8a | 1205 | } |
deefdf8a CB |
1206 | } |
1207 | ||
f7176c3e | 1208 | if (handler->ns_unshare_flags & CLONE_NEWTIME) { |
70fd7fc9 CB |
1209 | ret = unshare(CLONE_NEWTIME); |
1210 | if (ret < 0) { | |
1211 | if (errno != EINVAL) { | |
1212 | SYSERROR("Failed to unshare CLONE_NEWTIME"); | |
1213 | goto out_warn_father; | |
1214 | } | |
1215 | ||
1216 | handler->ns_clone_flags &= ~CLONE_NEWTIME; | |
1217 | SYSINFO("Kernel does not support CLONE_NEWTIME"); | |
1218 | } else { | |
1219 | __do_close int timens_fd = -EBADF; | |
1220 | ||
1221 | INFO("Unshared CLONE_NEWTIME"); | |
1222 | ||
1223 | if (handler->conf->timens.s_boot) | |
1224 | ret = timens_offset_write(CLOCK_BOOTTIME, handler->conf->timens.s_boot, 0); | |
1225 | else if (handler->conf->timens.ns_boot) | |
1226 | ret = timens_offset_write(CLOCK_BOOTTIME, 0, handler->conf->timens.ns_boot); | |
1227 | if (ret) { | |
1228 | SYSERROR("Failed to write CLONE_BOOTTIME offset"); | |
1229 | goto out_warn_father; | |
1230 | } | |
1231 | TRACE("Wrote CLOCK_BOOTTIME offset"); | |
1232 | ||
1233 | if (handler->conf->timens.s_monotonic) | |
1234 | ret = timens_offset_write(CLOCK_MONOTONIC, handler->conf->timens.s_monotonic, 0); | |
1235 | else if (handler->conf->timens.ns_monotonic) | |
1236 | ret = timens_offset_write(CLOCK_MONOTONIC, 0, handler->conf->timens.ns_monotonic); | |
1237 | if (ret) { | |
1238 | SYSERROR("Failed to write CLONE_MONOTONIC offset"); | |
1239 | goto out_warn_father; | |
1240 | } | |
1241 | TRACE("Wrote CLOCK_MONOTONIC offset"); | |
1242 | ||
1243 | timens_fd = open("/proc/self/ns/time_for_children", O_RDONLY | O_CLOEXEC); | |
1244 | if (timens_fd < 0) { | |
1245 | SYSERROR("Failed to open \"/proc/self/ns/time_for_children\""); | |
1246 | goto out_warn_father; | |
1247 | } | |
1248 | ||
1249 | ret = setns(timens_fd, CLONE_NEWTIME); | |
cfe6f435 | 1250 | if (ret) { |
70fd7fc9 CB |
1251 | SYSERROR("Failed to setns(%d(\"/proc/self/ns/time_for_children\"))", timens_fd); |
1252 | goto out_warn_father; | |
1253 | } | |
1254 | } | |
1255 | } | |
1256 | ||
c294a68d CB |
1257 | /* |
1258 | * Add the requested environment variables to the current environment | |
1259 | * to allow them to be used by the various hooks, such as the start | |
1260 | * hook below. | |
98ff08ca | 1261 | */ |
c294a68d CB |
1262 | ret = lxc_set_environment(handler->conf); |
1263 | if (ret < 0) | |
1264 | goto out_warn_father; | |
98ff08ca | 1265 | |
6bc4165d CB |
1266 | if (!lxc_sync_wait_parent(handler, START_SYNC_POST_CONFIGURE)) |
1267 | goto out_warn_father; | |
1268 | ||
f4152036 | 1269 | /* Setup the container, ip, names, utsname, ... */ |
3b988b33 | 1270 | ret = lxc_setup(handler); |
7729f8e5 | 1271 | if (ret < 0) { |
e96a536c | 1272 | ERROR("Failed to setup container \"%s\"", handler->name); |
f4152036 CB |
1273 | goto out_warn_father; |
1274 | } | |
1275 | ||
408da065 | 1276 | /* Set the label to change to when we exec(2) the container's init. */ |
af04d847 | 1277 | ret = handler->lsm_ops->process_label_set(handler->lsm_ops, NULL, handler->conf, true); |
e96a536c | 1278 | if (ret < 0) |
e075f5d9 | 1279 | goto out_warn_father; |
5112cd70 | 1280 | |
029cdff5 | 1281 | /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it |
408da065 CB |
1282 | * before we aren't allowed anymore. |
1283 | */ | |
029cdff5 | 1284 | if (handler->conf->no_new_privs) { |
b81689a1 CB |
1285 | ret = prctl(PR_SET_NO_NEW_PRIVS, prctl_arg(1), prctl_arg(0), |
1286 | prctl_arg(0), prctl_arg(0)); | |
e96a536c | 1287 | if (ret < 0) { |
818a57fc | 1288 | SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges"); |
029cdff5 CB |
1289 | goto out_warn_father; |
1290 | } | |
818a57fc | 1291 | DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges"); |
029cdff5 CB |
1292 | } |
1293 | ||
408da065 | 1294 | /* If we mounted a temporary proc, then unmount it now. */ |
5112cd70 | 1295 | tmp_proc_unmount(handler->conf); |
e075f5d9 | 1296 | |
e96a536c CB |
1297 | ret = lxc_seccomp_load(handler->conf); |
1298 | if (ret < 0) | |
8f2c3a70 SH |
1299 | goto out_warn_father; |
1300 | ||
e96a536c CB |
1301 | ret = run_lxc_hooks(handler->name, "start", handler->conf, NULL); |
1302 | if (ret < 0) { | |
1303 | ERROR("Failed to run lxc.hook.start for container \"%s\"", | |
1304 | handler->name); | |
773fb9ca SH |
1305 | goto out_warn_father; |
1306 | } | |
fc25b815 | 1307 | |
a783a414 | 1308 | close_prot_errno_disarm(handler->sigfd); |
26ddeedd | 1309 | |
41808e20 | 1310 | if (handler->conf->console.pty < 0 && handler->daemonize) { |
f4c177c3 CB |
1311 | if (devnull_fd < 0) { |
1312 | devnull_fd = open_devnull(); | |
1313 | if (devnull_fd < 0) | |
1314 | goto out_warn_father; | |
1315 | } | |
1316 | ||
e96a536c CB |
1317 | ret = set_stdfds(devnull_fd); |
1318 | if (ret < 0) { | |
1319 | ERROR("Failed to redirect std{in,out,err} to \"/dev/null\""); | |
c5b93afb LF |
1320 | goto out_warn_father; |
1321 | } | |
e96a536c | 1322 | } |
507cee36 | 1323 | |
a783a414 | 1324 | close_prot_errno_disarm(devnull_fd); |
c44de748 | 1325 | |
8c9a7665 TA |
1326 | setsid(); |
1327 | ||
e96a536c CB |
1328 | if (handler->conf->init_cwd) { |
1329 | ret = chdir(handler->conf->init_cwd); | |
1330 | if (ret < 0) { | |
1331 | SYSERROR("Could not change directory to \"%s\"", | |
1332 | handler->conf->init_cwd); | |
1333 | goto out_warn_father; | |
1334 | } | |
3c491553 L |
1335 | } |
1336 | ||
6e48e7c5 | 1337 | if (!lxc_sync_barrier_parent(handler, START_SYNC_CGROUP_LIMITS)) |
f4152036 CB |
1338 | goto out_warn_father; |
1339 | ||
111ed96e | 1340 | ret = lxc_sync_fds_child(handler); |
1b82d721 | 1341 | if (ret < 0) { |
111ed96e | 1342 | SYSERROR("Failed to sync file descriptors with parent"); |
1b82d721 CB |
1343 | goto out_warn_father; |
1344 | } | |
1345 | ||
1b82d721 CB |
1346 | if (!lxc_sync_wait_parent(handler, START_SYNC_READY_START)) |
1347 | goto out_warn_father; | |
1348 | ||
149857af CB |
1349 | /* Reset the environment variables the user requested in a clear |
1350 | * environment. | |
1351 | */ | |
e96a536c CB |
1352 | ret = clearenv(); |
1353 | /* Don't error out though. */ | |
1354 | if (ret < 0) | |
149857af | 1355 | SYSERROR("Failed to clear environment."); |
149857af | 1356 | |
c294a68d CB |
1357 | ret = lxc_set_environment(handler->conf); |
1358 | if (ret < 0) | |
1359 | goto out_warn_father; | |
149857af | 1360 | |
e96a536c CB |
1361 | ret = putenv("container=lxc"); |
1362 | if (ret < 0) { | |
1363 | SYSERROR("Failed to set environment variable: container=lxc"); | |
98ff08ca CB |
1364 | goto out_warn_father; |
1365 | } | |
1366 | ||
885766f5 CB |
1367 | if (handler->conf->ttys.tty_names) { |
1368 | ret = putenv(handler->conf->ttys.tty_names); | |
e96a536c CB |
1369 | if (ret < 0) { |
1370 | SYSERROR("Failed to set environment variable for container ptys"); | |
98ff08ca CB |
1371 | goto out_warn_father; |
1372 | } | |
1373 | } | |
1374 | ||
76bdf299 CB |
1375 | /* The container has been setup. We can now switch to an unprivileged |
1376 | * uid/gid. | |
1377 | */ | |
928b1f04 YT |
1378 | new_uid = handler->conf->init_uid; |
1379 | new_gid = handler->conf->init_gid; | |
76bdf299 | 1380 | |
964581c2 CB |
1381 | /* Avoid unnecessary syscalls. */ |
1382 | if (new_uid == nsuid) | |
1383 | new_uid = LXC_INVALID_UID; | |
1384 | ||
1385 | if (new_gid == nsgid) | |
1386 | new_gid = LXC_INVALID_GID; | |
c353b0b9 | 1387 | |
6aff5157 | 1388 | /* Make sure that the processes STDIO is correctly owned by the user that we are switching to */ |
c353b0b9 CB |
1389 | ret = fix_stdio_permissions(new_uid); |
1390 | if (ret) | |
1391 | WARN("Failed to ajust stdio permissions"); | |
964581c2 | 1392 | |
8af07f82 CB |
1393 | /* If we are in a new user namespace we already dropped all groups when |
1394 | * we switched to root in the new user namespace further above. Only | |
1395 | * drop groups if we can, so ensure that we have necessary privilege. | |
1396 | */ | |
0589d744 | 1397 | if (list_empty(&handler->conf->id_map)) { |
8af07f82 CB |
1398 | #if HAVE_LIBCAP |
1399 | if (lxc_proc_cap_is_set(CAP_SETGID, CAP_EFFECTIVE)) | |
1400 | #endif | |
bf31b337 RJ |
1401 | { |
1402 | if (handler->conf->init_groups.size > 0) { | |
1403 | if (!lxc_setgroups(handler->conf->init_groups.list, | |
1404 | handler->conf->init_groups.size)) | |
1405 | goto out_warn_father; | |
1406 | } else { | |
1407 | if (!lxc_drop_groups()) | |
1408 | goto out_warn_father; | |
1409 | } | |
1410 | } | |
1411 | } | |
8af07f82 | 1412 | |
b58214ac CB |
1413 | if (!lxc_switch_uid_gid(new_uid, new_gid)) |
1414 | goto out_warn_father; | |
1415 | ||
611ddd34 CB |
1416 | ret = lxc_ambient_caps_down(); |
1417 | if (ret < 0) { | |
6f5e532f | 1418 | ERROR("Failed to clear ambient capabilities"); |
611ddd34 CB |
1419 | goto out_warn_father; |
1420 | } | |
1421 | ||
258f8051 | 1422 | if (handler->conf->monitor_signal_pdeath != SIGKILL) { |
4d8bdfa0 CB |
1423 | ret = lxc_set_death_signal(handler->conf->monitor_signal_pdeath, |
1424 | handler->monitor_pid, status_fd); | |
258f8051 CB |
1425 | if (ret < 0) { |
1426 | SYSERROR("Failed to set PR_SET_PDEATHSIG to %d", | |
1427 | handler->conf->monitor_signal_pdeath); | |
1428 | goto out_warn_father; | |
1429 | } | |
1430 | } | |
1431 | ||
d3103162 CB |
1432 | /* |
1433 | * After this call, we are in error because this ops should not return | |
408da065 CB |
1434 | * as it execs. |
1435 | */ | |
c4ea60df | 1436 | handler->ops->start(handler, handler->data); |
50e98013 DL |
1437 | |
1438 | out_warn_father: | |
d3103162 CB |
1439 | /* |
1440 | * We want the parent to know something went wrong, so we return a | |
408da065 CB |
1441 | * special error code. |
1442 | */ | |
946a0c6d | 1443 | lxc_sync_wake_parent(handler, SYNC_ERROR); |
c44de748 AM |
1444 | |
1445 | out_error: | |
50e98013 DL |
1446 | return -1; |
1447 | } | |
1448 | ||
5af9369b | 1449 | int resolve_clone_flags(struct lxc_handler *handler) |
f813849c | 1450 | { |
8bc8c715 CB |
1451 | int i; |
1452 | struct lxc_conf *conf = handler->conf; | |
70fd7fc9 CB |
1453 | bool wants_timens = conf->timens.s_boot || conf->timens.ns_boot || |
1454 | conf->timens.s_monotonic || conf->timens.ns_monotonic; | |
f813849c | 1455 | |
8bc8c715 | 1456 | for (i = 0; i < LXC_NS_MAX; i++) { |
f7a0c6ee CB |
1457 | if (conf->ns_keep) { |
1458 | if (!(conf->ns_keep & ns_info[i].clone_flag)) | |
becad0ec | 1459 | handler->ns_clone_flags |= ns_info[i].clone_flag; |
f7a0c6ee CB |
1460 | } else if (conf->ns_clone) { |
1461 | if ((conf->ns_clone & ns_info[i].clone_flag)) | |
becad0ec | 1462 | handler->ns_clone_flags |= ns_info[i].clone_flag; |
8bc8c715 | 1463 | } else { |
0589d744 | 1464 | if (i == LXC_NS_USER && list_empty(&handler->conf->id_map)) |
8bc8c715 CB |
1465 | continue; |
1466 | ||
1467 | if (i == LXC_NS_NET && lxc_requests_empty_network(handler)) | |
1468 | continue; | |
1469 | ||
7bd05339 CB |
1470 | if (i == LXC_NS_CGROUP && !cgns_supported()) |
1471 | continue; | |
1472 | ||
70fd7fc9 CB |
1473 | if (i == LXC_NS_TIME && !wants_timens) |
1474 | continue; | |
1475 | ||
becad0ec | 1476 | handler->ns_clone_flags |= ns_info[i].clone_flag; |
8bc8c715 | 1477 | } |
03df7ab5 | 1478 | |
8bc8c715 CB |
1479 | if (!conf->ns_share[i]) |
1480 | continue; | |
5af9369b | 1481 | |
becad0ec | 1482 | handler->ns_clone_flags &= ~ns_info[i].clone_flag; |
8bc8c715 | 1483 | TRACE("Sharing %s namespace", ns_info[i].proc_name); |
5af9369b CB |
1484 | } |
1485 | ||
70fd7fc9 CB |
1486 | if (wants_timens && (conf->ns_keep & ns_info[LXC_NS_TIME].clone_flag)) |
1487 | return log_trace_errno(-1, EINVAL, "Requested to keep time namespace while also specifying offsets"); | |
1488 | ||
f7176c3e CB |
1489 | /* Deal with namespaces that are unshared. */ |
1490 | if (handler->ns_clone_flags & CLONE_NEWTIME) | |
1491 | handler->ns_unshare_flags |= CLONE_NEWTIME; | |
1492 | ||
1493 | if (!pure_unified_layout(handler->cgroup_ops) && handler->ns_clone_flags & CLONE_NEWCGROUP) | |
1494 | handler->ns_unshare_flags |= CLONE_NEWCGROUP; | |
1495 | ||
1496 | if ((handler->ns_clone_flags & (CLONE_NEWNET | CLONE_NEWUSER)) == | |
1497 | (CLONE_NEWNET | CLONE_NEWUSER)) | |
1498 | handler->ns_unshare_flags |= CLONE_NEWNET; | |
1499 | ||
1500 | /* Deal with namespaces that are spawned. */ | |
1501 | handler->ns_on_clone_flags = handler->ns_clone_flags & ~handler->ns_unshare_flags; | |
1502 | ||
1503 | handler->clone_flags = handler->ns_on_clone_flags | CLONE_PIDFD; | |
1504 | ||
5af9369b | 1505 | return 0; |
f813849c TA |
1506 | } |
1507 | ||
8deca6c9 CB |
1508 | /* Note that this function is used with clone(CLONE_VM). Some glibc versions |
1509 | * used to reset the pid/tid to -1 when CLONE_VM was used without CLONE_THREAD. | |
1510 | * But since the memory between parent and child is shared on CLONE_VM this | |
1511 | * would invalidate the getpid() cache that glibc used to maintain and so | |
1512 | * getpid() in the child would return the parent's pid. This is all fixed in | |
1513 | * newer glibc versions where the getpid() cache is removed and the pid/tid is | |
1514 | * not reset anymore. | |
b35ddc5b | 1515 | * However, if for whatever reason you - dear committer - somehow need to get the |
8de0119d CB |
1516 | * pid of the placeholder intermediate process for do_share_ns() you need to |
1517 | * call lxc_raw_getpid(). The next lxc_raw_clone() call does not employ | |
1518 | * CLONE_VM and will be fine. | |
8deca6c9 CB |
1519 | */ |
1520 | static inline int do_share_ns(void *arg) | |
fa3a5b22 | 1521 | { |
8deca6c9 | 1522 | int i, flags, ret; |
fa3a5b22 CB |
1523 | struct lxc_handler *handler = arg; |
1524 | ||
8deca6c9 CB |
1525 | for (i = 0; i < LXC_NS_MAX; i++) { |
1526 | if (handler->nsfd[i] < 0) | |
1527 | continue; | |
fa3a5b22 | 1528 | |
8deca6c9 | 1529 | ret = setns(handler->nsfd[i], 0); |
c0b48eff CB |
1530 | if (ret < 0) { |
1531 | /* | |
1532 | * Note that joining a user and/or mount namespace | |
1533 | * requires the process is not multithreaded otherwise | |
1534 | * setns() will fail here. | |
1535 | */ | |
1536 | SYSERROR("Failed to inherit %s namespace", | |
1537 | ns_info[i].proc_name); | |
8deca6c9 | 1538 | return -1; |
c0b48eff | 1539 | } |
fa3a5b22 | 1540 | |
8deca6c9 | 1541 | DEBUG("Inherited %s namespace", ns_info[i].proc_name); |
fa3a5b22 CB |
1542 | } |
1543 | ||
becad0ec | 1544 | flags = handler->ns_on_clone_flags; |
8deca6c9 | 1545 | flags |= CLONE_PARENT; |
33942046 CB |
1546 | handler->pid = lxc_raw_clone_cb(do_start, handler, CLONE_PIDFD | flags, |
1547 | &handler->pidfd); | |
8deca6c9 | 1548 | if (handler->pid < 0) |
fa3a5b22 CB |
1549 | return -1; |
1550 | ||
8deca6c9 | 1551 | return 0; |
fa3a5b22 CB |
1552 | } |
1553 | ||
09996a48 CB |
1554 | static int core_scheduling(struct lxc_handler *handler) |
1555 | { | |
1556 | struct lxc_conf *conf = handler->conf; | |
1557 | int ret; | |
1558 | ||
1559 | if (!conf->sched_core) | |
1560 | return log_trace(0, "No new core scheduling domain requested"); | |
1561 | ||
c958a332 CB |
1562 | if (!(handler->ns_clone_flags & CLONE_NEWPID)) |
1563 | return syserror_set(-EINVAL, "Core scheduling currently requires a separate pid namespace"); | |
1564 | ||
47bfecf5 | 1565 | ret = core_scheduling_cookie_create_threadgroup(handler->pid); |
09996a48 | 1566 | if (ret < 0) { |
f3d90fca CB |
1567 | if (ret == -ENODEV) { |
1568 | INFO("The kernel doesn't support or doesn't use simultaneous multithreading (SMT)"); | |
1569 | conf->sched_core = false; | |
1570 | return 0; | |
1571 | } | |
09996a48 | 1572 | if (ret == -EINVAL) |
ffd98e8c | 1573 | return syserror("The kernel does not support core scheduling"); |
09996a48 CB |
1574 | |
1575 | return syserror("Failed to create new core scheduling domain"); | |
1576 | } | |
1577 | ||
f3d90fca CB |
1578 | ret = core_scheduling_cookie_get(handler->pid, &conf->sched_core_cookie); |
1579 | if (ret || !core_scheduling_cookie_valid(conf->sched_core_cookie)) | |
09996a48 CB |
1580 | return syserror("Failed to retrieve core scheduling domain cookie"); |
1581 | ||
1582 | TRACE("Created new core scheduling domain with cookie %llu", | |
2e9aa146 | 1583 | (llu)conf->sched_core_cookie); |
09996a48 CB |
1584 | |
1585 | return 0; | |
1586 | } | |
1587 | ||
6c50e09f CB |
1588 | static bool inherits_namespaces(const struct lxc_handler *handler) |
1589 | { | |
1590 | struct lxc_conf *conf = handler->conf; | |
1591 | ||
1592 | for (lxc_namespace_t i = 0; i < LXC_NS_MAX; i++) { | |
1593 | if (conf->ns_share[i]) | |
1594 | return true; | |
1595 | } | |
1596 | ||
1597 | return false; | |
1598 | } | |
1599 | ||
07a00b78 CB |
1600 | static inline void resolve_cgroup_clone_flags(struct lxc_handler *handler) |
1601 | { | |
1602 | handler->clone_flags &= ~(CLONE_INTO_CGROUP | CLONE_NEWCGROUP); | |
1603 | handler->ns_on_clone_flags &= ~(CLONE_INTO_CGROUP | CLONE_NEWCGROUP); | |
1604 | handler->ns_unshare_flags |= CLONE_NEWCGROUP; | |
1605 | } | |
1606 | ||
480588e6 CB |
1607 | /* lxc_spawn() performs crucial setup tasks and clone()s the new process which |
1608 | * exec()s the requested container binary. | |
1609 | * Note that lxc_spawn() runs in the parent namespaces. Any operations performed | |
1610 | * right here should be double checked if they'd pose a security risk. (For | |
1611 | * example, any {u}mount() operations performed here will be reflected on the | |
1612 | * host!) | |
1613 | */ | |
74a3920a | 1614 | static int lxc_spawn(struct lxc_handler *handler) |
59eb99ba | 1615 | { |
f62cf1d4 | 1616 | __do_close int data_sock0 = -EBADF, data_sock1 = -EBADF; |
8deca6c9 | 1617 | int i, ret; |
08dd2805 | 1618 | char pidstr[20]; |
57927bf2 | 1619 | bool wants_to_map_ids; |
0589d744 | 1620 | struct list_head *id_map; |
6e5fc7a5 | 1621 | const char *name = handler->name; |
28d9e29e | 1622 | struct lxc_conf *conf = handler->conf; |
2202afc9 | 1623 | struct cgroup_ops *cgroup_ops = handler->cgroup_ops; |
9f30a190 | 1624 | |
28d9e29e | 1625 | id_map = &conf->id_map; |
0589d744 | 1626 | wants_to_map_ids = !list_empty(id_map); |
2f2623ec | 1627 | |
6e48e7c5 | 1628 | if (!lxc_sync_init(handler)) |
9d7f9e52 | 1629 | return -1; |
0ad19a3f | 1630 | |
b9f522c5 | 1631 | ret = socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0, |
790255cf | 1632 | handler->data_sock); |
95ef0d7c CB |
1633 | if (ret < 0) |
1634 | goto out_sync_fini; | |
cdb2a47f CB |
1635 | data_sock0 = handler->data_sock[0]; |
1636 | data_sock1 = handler->data_sock[1]; | |
e8bd4e43 | 1637 | |
03ca4af8 TP |
1638 | if (handler->ns_clone_flags & CLONE_NEWNET) { |
1639 | ret = lxc_find_gateway_addresses(handler); | |
1640 | if (ret) { | |
1641 | ERROR("Failed to find gateway addresses"); | |
1642 | goto out_sync_fini; | |
1643 | } | |
1644 | } | |
1645 | ||
e8b181f5 | 1646 | if (!cgroup_ops->payload_create(cgroup_ops, handler)) { |
cfc62c60 | 1647 | ERROR("Failed creating cgroups"); |
47d8fb3b CS |
1648 | goto out_delete_net; |
1649 | } | |
1650 | ||
408da065 | 1651 | /* Create a process in a new set of namespaces. */ |
6c50e09f | 1652 | if (inherits_namespaces(handler)) { |
4e232466 CB |
1653 | pid_t attacher_pid; |
1654 | ||
07a00b78 | 1655 | resolve_cgroup_clone_flags(handler); |
cfc62c60 | 1656 | attacher_pid = lxc_clone(do_share_ns, handler, |
33258b95 | 1657 | CLONE_VFORK | CLONE_VM | CLONE_FILES, NULL); |
4e232466 CB |
1658 | if (attacher_pid < 0) { |
1659 | SYSERROR(LXC_CLONE_ERROR); | |
1660 | goto out_delete_net; | |
1661 | } | |
1662 | ||
1663 | ret = wait_for_pid(attacher_pid); | |
1664 | if (ret < 0) { | |
1665 | SYSERROR("Intermediate process failed"); | |
1666 | goto out_delete_net; | |
1667 | } | |
f7176c3e CB |
1668 | |
1669 | if (handler->pid < 0) { | |
1670 | SYSERROR(LXC_CLONE_ERROR); | |
1671 | goto out_delete_net; | |
1672 | } | |
4e232466 | 1673 | } else { |
84b66ced | 1674 | int cgroup_fd = -EBADF; |
f7176c3e | 1675 | |
63468abd | 1676 | struct clone_args clone_args = { |
f7176c3e CB |
1677 | .flags = handler->clone_flags, |
1678 | .pidfd = ptr_to_u64(&handler->pidfd), | |
1679 | .exit_signal = SIGCHLD, | |
1680 | }; | |
1681 | ||
1682 | if (handler->ns_clone_flags & CLONE_NEWCGROUP) { | |
1683 | cgroup_fd = cgroup_unified_fd(cgroup_ops); | |
1684 | if (cgroup_fd >= 0) { | |
1685 | handler->clone_flags |= CLONE_INTO_CGROUP; | |
1686 | clone_args.flags |= CLONE_INTO_CGROUP; | |
1687 | clone_args.cgroup = cgroup_fd; | |
1688 | } | |
1689 | } | |
1690 | ||
1691 | /* Try to spawn directly into target cgroup. */ | |
1692 | handler->pid = lxc_clone3(&clone_args, CLONE_ARGS_SIZE_VER2); | |
1693 | if (handler->pid < 0) { | |
1694 | SYSTRACE("Failed to spawn container directly into target cgroup"); | |
1695 | ||
1696 | /* Kernel might simply be too old for CLONE_INTO_CGROUP. */ | |
07a00b78 CB |
1697 | resolve_cgroup_clone_flags(handler); |
1698 | clone_args.flags = handler->clone_flags; | |
f7176c3e CB |
1699 | |
1700 | handler->pid = lxc_clone3(&clone_args, CLONE_ARGS_SIZE_VER0); | |
1701 | } else if (cgroup_fd >= 0) { | |
1702 | TRACE("Spawned container directly into target cgroup via cgroup2 fd %d", cgroup_fd); | |
1703 | } | |
1704 | ||
1705 | /* Kernel might be too old for clone3(). */ | |
1706 | if (handler->pid < 0) { | |
1707 | SYSTRACE("Failed to spawn container via clone3()"); | |
abd80bde CB |
1708 | |
1709 | /* | |
1710 | * In contrast to all other architectures arm64 verifies that | |
1711 | * the argument we use to retrieve the pidfd with is | |
1712 | * initialized to 0. But we need to be able to initialize it to | |
1713 | * a negative value such as our customary -EBADF so we can | |
1714 | * detect whether this kernel supports pidfds. If the syscall | |
1715 | * returns and the pidfd variable is set to something >= 0 then | |
1716 | * we know this is a kernel supporting pidfds. But if we can't | |
1717 | * set it to -EBADF then this won't work since 0 is a valid | |
1718 | * file descriptor too. And since legacy clone silently ignores | |
1719 | * unknown flags we are left without any way to detect support | |
1720 | * for pidfds. So let's special-case arm64 to not fail starting | |
1721 | * containers. | |
1722 | */ | |
1723 | #if defined(__aarch64__) | |
1724 | handler->pid = lxc_raw_legacy_clone(handler->clone_flags & ~CLONE_PIDFD, NULL); | |
1725 | #else | |
f7176c3e | 1726 | handler->pid = lxc_raw_legacy_clone(handler->clone_flags, &handler->pidfd); |
abd80bde | 1727 | #endif |
f7176c3e CB |
1728 | } |
1729 | ||
1730 | if (handler->pid < 0) { | |
1731 | SYSERROR(LXC_CLONE_ERROR); | |
1732 | goto out_delete_net; | |
1733 | } | |
1734 | ||
1735 | if (handler->pid == 0) { | |
1736 | (void)do_start(handler); | |
1737 | _exit(EXIT_FAILURE); | |
1738 | } | |
0ad19a3f | 1739 | } |
f7176c3e CB |
1740 | if (handler->pidfd < 0) |
1741 | handler->clone_flags &= ~CLONE_PIDFD; | |
fa3a5b22 | 1742 | TRACE("Cloned child process %d", handler->pid); |
1bd8d726 | 1743 | |
09996a48 CB |
1744 | ret = core_scheduling(handler); |
1745 | if (ret < 0) | |
1746 | goto out_delete_net; | |
1747 | ||
39293f22 CB |
1748 | /* Verify that we can actually make use of pidfds. */ |
1749 | if (!lxc_can_use_pidfd(handler->pidfd)) | |
1750 | close_prot_errno_disarm(handler->pidfd); | |
1751 | ||
fa60cd7b CB |
1752 | ret = strnprintf(pidstr, 20, "%d", handler->pid); |
1753 | if (ret < 0) | |
1871e646 CB |
1754 | goto out_delete_net; |
1755 | ||
1756 | ret = setenv("LXC_PID", pidstr, 1); | |
1757 | if (ret < 0) | |
1758 | SYSERROR("Failed to set environment variable: LXC_PID=%s", pidstr); | |
1759 | ||
9662e444 | 1760 | for (i = 0; i < LXC_NS_MAX; i++) |
becad0ec | 1761 | if (handler->ns_on_clone_flags & ns_info[i].clone_flag) |
28d9e29e | 1762 | INFO("Cloned %s", ns_info[i].flag_name); |
0ad19a3f | 1763 | |
8db6be1b | 1764 | if (!lxc_try_preserve_namespaces(handler, handler->ns_on_clone_flags)) { |
28d9e29e CB |
1765 | ERROR("Failed to preserve cloned namespaces for lxc.hook.stop"); |
1766 | goto out_delete_net; | |
1767 | } | |
b6b2b194 | 1768 | |
3c22086f CLG |
1769 | lxc_sync_fini_child(handler); |
1770 | ||
4d8bdfa0 CB |
1771 | if (lxc_abstract_unix_send_fds(handler->data_sock[0], &handler->monitor_status_fd, 1, NULL, 0) < 0) { |
1772 | ERROR("Failed to send status file descriptor to child process"); | |
1773 | goto out_delete_net; | |
1774 | } | |
1775 | close_prot_errno_disarm(handler->monitor_status_fd); | |
1776 | ||
408da065 CB |
1777 | /* Map the container uids. The container became an invalid userid the |
1778 | * moment it was cloned with CLONE_NEWUSER. This call doesn't change | |
1779 | * anything immediately, but allows the container to setuid(0) (0 being | |
1780 | * mapped to something else on the host.) later to become a valid uid | |
1781 | * again. | |
1782 | */ | |
fa3a5b22 | 1783 | if (wants_to_map_ids) { |
df3c5314 | 1784 | if (!handler->conf->ns_share[LXC_NS_USER] && |
46186acd | 1785 | (handler->conf->ns_keep & CLONE_NEWUSER) == 0) { |
fa3a5b22 CB |
1786 | ret = lxc_map_ids(id_map, handler->pid); |
1787 | if (ret < 0) { | |
1788 | ERROR("Failed to set up id mapping."); | |
1789 | goto out_delete_net; | |
1790 | } | |
1791 | } | |
5b1e83cb SH |
1792 | } |
1793 | ||
c581d2a6 | 1794 | if (!cgroup_ops->setup_limits_legacy(cgroup_ops, handler->conf, false)) { |
cfc62c60 | 1795 | ERROR("Failed to setup cgroup limits for container \"%s\"", name); |
6031a6e5 DE |
1796 | goto out_delete_net; |
1797 | } | |
1798 | ||
838d1556 CB |
1799 | if (!cgroup_ops->payload_delegate_controllers(cgroup_ops)) { |
1800 | ERROR("Failed to delegate controllers to payload cgroup"); | |
7fef7a06 | 1801 | goto out_delete_net; |
fe70edee | 1802 | } |
218d4250 | 1803 | |
838d1556 CB |
1804 | if (!cgroup_ops->payload_enter(cgroup_ops, handler)) { |
1805 | ERROR("Failed to enter cgroups"); | |
c581d2a6 CB |
1806 | goto out_delete_net; |
1807 | } | |
1808 | ||
1809 | if (!cgroup_ops->setup_limits(cgroup_ops, handler)) { | |
1810 | ERROR("Failed to setup cgroup limits for container \"%s\"", name); | |
1811 | goto out_delete_net; | |
1812 | } | |
1813 | ||
2202afc9 | 1814 | if (!cgroup_ops->chown(cgroup_ops, handler->conf)) |
0996e18a SH |
1815 | goto out_delete_net; |
1816 | ||
6bc4165d CB |
1817 | if (!lxc_sync_barrier_child(handler, START_SYNC_STARTUP)) |
1818 | goto out_delete_net; | |
1819 | ||
aa0c0e7b RK |
1820 | /* If not done yet, we're now ready to preserve the network namespace */ |
1821 | if (handler->nsfd[LXC_NS_NET] < 0) { | |
8db6be1b | 1822 | ret = lxc_try_preserve_namespace(handler, LXC_NS_NET, "net"); |
aa0c0e7b | 1823 | if (ret < 0) { |
857ba1f0 | 1824 | if (ret != -ENOENT) { |
aa0c0e7b RK |
1825 | SYSERROR("Failed to preserve net namespace"); |
1826 | goto out_delete_net; | |
1827 | } | |
4cb53844 | 1828 | } |
fa3a5b22 | 1829 | } |
aa0c0e7b RK |
1830 | ret = lxc_netns_set_nsid(handler->nsfd[LXC_NS_NET]); |
1831 | if (ret < 0) | |
1832 | SYSWARN("Failed to allocate new network namespace id"); | |
1833 | else | |
1834 | TRACE("Allocated new network namespace id"); | |
fa3a5b22 | 1835 | |
408da065 | 1836 | /* Create the network configuration. */ |
becad0ec | 1837 | if (handler->ns_clone_flags & CLONE_NEWNET) { |
e389f2af | 1838 | ret = lxc_create_network(handler); |
cfc62c60 | 1839 | if (ret < 0) { |
e389f2af | 1840 | ERROR("Failed to create the network"); |
7fef7a06 | 1841 | goto out_delete_net; |
82d5ae15 | 1842 | } |
0ad19a3f | 1843 | } |
1844 | ||
91d04bf9 CB |
1845 | ret = setup_proc_filesystem(conf, handler->pid); |
1846 | if (ret < 0) { | |
1847 | ERROR("Failed to setup procfs limits"); | |
1848 | goto out_delete_net; | |
6bc4165d CB |
1849 | } |
1850 | ||
223797c3 CB |
1851 | ret = setup_resource_limits(conf, handler->pid); |
1852 | if (ret < 0) { | |
1853 | ERROR("Failed to setup resource limits"); | |
1854 | goto out_delete_net; | |
6bc4165d CB |
1855 | } |
1856 | ||
8945dad0 CB |
1857 | /* Tell the child to continue its initialization. */ |
1858 | if (!lxc_sync_wake_child(handler, START_SYNC_POST_CONFIGURE)) | |
1859 | goto out_delete_net; | |
1860 | ||
e4564b7e CB |
1861 | ret = lxc_rootfs_prepare_parent(handler); |
1862 | if (ret) { | |
1863 | ERROR("Failed to prepare rootfs"); | |
1864 | goto out_delete_net; | |
1865 | } | |
1866 | ||
8945dad0 CB |
1867 | if (handler->ns_clone_flags & CLONE_NEWNET) { |
1868 | ret = lxc_network_send_to_child(handler); | |
1869 | if (ret < 0) { | |
6bc4165d | 1870 | SYSERROR("Failed to send veth names to child"); |
8945dad0 CB |
1871 | goto out_delete_net; |
1872 | } | |
1873 | } | |
1874 | ||
6bc4165d | 1875 | if (!lxc_sync_wait_child(handler, START_SYNC_IDMAPPED_MOUNTS)) |
f4152036 CB |
1876 | goto out_delete_net; |
1877 | ||
1b82d721 CB |
1878 | ret = lxc_idmapped_mounts_parent(handler); |
1879 | if (ret) { | |
1880 | ERROR("Failed to setup mount entries"); | |
1881 | goto out_delete_net; | |
1882 | } | |
1883 | ||
1884 | if (!lxc_sync_wait_child(handler, START_SYNC_CGROUP_LIMITS)) | |
1885 | goto out_delete_net; | |
1886 | ||
432faf20 | 1887 | /* |
3a89b0ab CB |
1888 | * With isolation the limiting devices cgroup was already setup, so |
1889 | * only setup devices here if we have no namespace directory. | |
432faf20 WB |
1890 | */ |
1891 | if (!handler->conf->cgroup_meta.namespace_dir && | |
1892 | !cgroup_ops->setup_limits_legacy(cgroup_ops, handler->conf, true)) { | |
cfc62c60 | 1893 | ERROR("Failed to setup legacy device cgroup controller limits"); |
b98f7d6e | 1894 | goto out_delete_net; |
544a48a0 | 1895 | } |
cfc62c60 | 1896 | TRACE("Set up legacy device cgroup controller limits"); |
544a48a0 | 1897 | |
bf651989 CB |
1898 | if (!cgroup_ops->devices_activate(cgroup_ops, handler)) { |
1899 | ERROR("Failed to setup cgroup2 device controller limits"); | |
1900 | goto out_delete_net; | |
1901 | } | |
1902 | TRACE("Set up cgroup2 device controller limits"); | |
1903 | ||
6bc4165d CB |
1904 | cgroup_ops->finalize(cgroup_ops); |
1905 | TRACE("Finished setting up cgroups"); | |
1906 | ||
1907 | /* Run any host-side start hooks */ | |
1908 | ret = run_lxc_hooks(name, "start-host", conf, NULL); | |
1909 | if (ret < 0) { | |
1910 | ERROR("Failed to run lxc.hook.start-host"); | |
1911 | goto out_delete_net; | |
1912 | } | |
1913 | ||
1914 | if (!lxc_sync_wake_child(handler, START_SYNC_FDS)) | |
1915 | goto out_delete_net; | |
1916 | ||
f7176c3e | 1917 | if (handler->ns_unshare_flags & CLONE_NEWCGROUP) { |
8bf3abfb | 1918 | /* Now we're ready to preserve the cgroup namespace */ |
8db6be1b | 1919 | ret = lxc_try_preserve_namespace(handler, LXC_NS_CGROUP, "cgroup"); |
8bf3abfb | 1920 | if (ret < 0) { |
857ba1f0 | 1921 | if (ret != -ENOENT) { |
6d1400b5 | 1922 | SYSERROR("Failed to preserve cgroup namespace"); |
4cb53844 CB |
1923 | goto out_delete_net; |
1924 | } | |
8bf3abfb | 1925 | } |
8bf3abfb CB |
1926 | } |
1927 | ||
f7176c3e | 1928 | if (handler->ns_unshare_flags & CLONE_NEWTIME) { |
245066af | 1929 | /* Now we're ready to preserve the time namespace */ |
8db6be1b | 1930 | ret = lxc_try_preserve_namespace(handler, LXC_NS_TIME, "time"); |
f1c43439 | 1931 | if (ret < 0) { |
857ba1f0 | 1932 | if (ret != -ENOENT) { |
f1c43439 CB |
1933 | SYSERROR("Failed to preserve time namespace"); |
1934 | goto out_delete_net; | |
1935 | } | |
f1c43439 CB |
1936 | } |
1937 | } | |
1938 | ||
493ae3fe | 1939 | ret = lxc_sync_fds_parent(handler); |
1b82d721 | 1940 | if (ret < 0) { |
493ae3fe | 1941 | SYSERROR("Failed to sync file descriptors with child"); |
1b82d721 | 1942 | goto out_delete_net; |
790255cf CB |
1943 | } |
1944 | ||
9f77617b CB |
1945 | ret = lxc_terminal_setup(conf); |
1946 | if (ret < 0) { | |
1947 | SYSERROR("Failed to create console"); | |
1948 | goto out_delete_net; | |
1949 | } | |
1950 | ||
1b82d721 CB |
1951 | /* |
1952 | * Tell the child to complete its initialization and wait for it to | |
1953 | * exec or return an error. (The child will never return | |
1954 | * START_SYNC_READY_START+1. It will either close the sync pipe, | |
1955 | * causing lxc_sync_barrier_child to return success, or return a | |
1956 | * different value, causing us to error out). | |
1957 | */ | |
1958 | if (!lxc_sync_barrier_child(handler, START_SYNC_READY_START)) | |
1959 | goto out_delete_net; | |
1960 | ||
1961 | /* Now all networks are created, network devices are moved into place, | |
1962 | * and the correct names and ifindices in the respective namespaces have | |
1963 | * been recorded. The corresponding structs have now all been filled. So | |
1964 | * log them for debugging purposes. | |
1965 | */ | |
1966 | lxc_log_configured_netdevs(conf); | |
1967 | ||
cfc62c60 CB |
1968 | ret = handler->ops->post_start(handler, handler->data); |
1969 | if (ret < 0) | |
e6126dbe MN |
1970 | goto out_abort; |
1971 | ||
cfc62c60 CB |
1972 | ret = lxc_set_state(name, handler, RUNNING); |
1973 | if (ret < 0) { | |
1974 | ERROR("Failed to set state to \"%s\"", lxc_state2str(RUNNING)); | |
59eb99ba | 1975 | goto out_abort; |
3f21c114 | 1976 | } |
22ebac19 | 1977 | |
3c22086f | 1978 | lxc_sync_fini(handler); |
0c547523 | 1979 | |
e6126dbe | 1980 | return 0; |
1ac470c0 | 1981 | |
7fef7a06 | 1982 | out_delete_net: |
becad0ec | 1983 | if (handler->ns_clone_flags & CLONE_NEWNET) |
bb84beda | 1984 | lxc_delete_network(handler); |
1bd8d726 | 1985 | |
59eb99ba | 1986 | out_abort: |
0c5859ff | 1987 | lxc_abort(handler); |
95ef0d7c CB |
1988 | |
1989 | out_sync_fini: | |
3c22086f | 1990 | lxc_sync_fini(handler); |
5c068da9 | 1991 | |
b79fcd86 | 1992 | return -1; |
59eb99ba | 1993 | } |
0ad19a3f | 1994 | |
6c50e09f CB |
1995 | static int lxc_inherit_namespaces(struct lxc_handler *handler) |
1996 | { | |
1997 | const char *lxcpath = handler->lxcpath; | |
1998 | struct lxc_conf *conf = handler->conf; | |
1999 | ||
2000 | for (lxc_namespace_t i = 0; i < LXC_NS_MAX; i++) { | |
2001 | if (!conf->ns_share[i]) | |
2002 | continue; | |
2003 | ||
2004 | handler->nsfd[i] = lxc_inherit_namespace(conf->ns_share[i], | |
2005 | lxcpath, | |
2006 | ns_info[i].proc_name); | |
2007 | if (handler->nsfd[i] < 0) | |
2008 | return -1; | |
2009 | ||
2010 | TRACE("Recording inherited %s namespace with fd %d", | |
2011 | ns_info[i].proc_name, handler->nsfd[i]); | |
2012 | } | |
2013 | ||
2014 | return 0; | |
2015 | } | |
2016 | ||
7bdf97d2 CB |
2017 | int __lxc_start(struct lxc_handler *handler, struct lxc_operations *ops, |
2018 | void *data, const char *lxcpath, bool daemonize, int *error_num) | |
59eb99ba | 2019 | { |
c30e9b19 | 2020 | int ret, status; |
7bdf97d2 | 2021 | const char *name = handler->name; |
aa460476 | 2022 | struct lxc_conf *conf = handler->conf; |
72068e74 | 2023 | struct cgroup_ops *cgroup_ops; |
80090207 | 2024 | |
c30e9b19 CB |
2025 | ret = lxc_init(name, handler); |
2026 | if (ret < 0) { | |
2027 | ERROR("Failed to initialize container \"%s\"", name); | |
9ad66589 | 2028 | goto out_abort; |
0ad19a3f | 2029 | } |
ee70bf78 CLG |
2030 | handler->ops = ops; |
2031 | handler->data = data; | |
bb955810 | 2032 | handler->daemonize = daemonize; |
72068e74 | 2033 | cgroup_ops = handler->cgroup_ops; |
e6126dbe | 2034 | |
76a26f55 | 2035 | if (!attach_block_device(handler->conf)) { |
c30e9b19 | 2036 | ERROR("Failed to attach block device"); |
575ea467 | 2037 | ret = -1; |
9ad66589 | 2038 | goto out_abort; |
76a26f55 SH |
2039 | } |
2040 | ||
72068e74 CB |
2041 | if (!cgroup_ops->monitor_create(cgroup_ops, handler)) { |
2042 | ERROR("Failed to create monitor cgroup"); | |
575ea467 | 2043 | ret = -1; |
9ad66589 | 2044 | goto out_abort; |
eeef32bb CB |
2045 | } |
2046 | ||
838d1556 CB |
2047 | if (!cgroup_ops->monitor_delegate_controllers(cgroup_ops)) { |
2048 | ERROR("Failed to delegate controllers to monitor cgroup"); | |
575ea467 | 2049 | ret = -1; |
9ad66589 | 2050 | goto out_abort; |
72068e74 CB |
2051 | } |
2052 | ||
838d1556 CB |
2053 | if (!cgroup_ops->monitor_enter(cgroup_ops, handler)) { |
2054 | ERROR("Failed to enter monitor cgroup"); | |
c581d2a6 | 2055 | ret = -1; |
9ad66589 | 2056 | goto out_abort; |
c581d2a6 CB |
2057 | } |
2058 | ||
6c50e09f CB |
2059 | ret = resolve_clone_flags(handler); |
2060 | if (ret < 0) { | |
2061 | ERROR("Failed to resolve clone flags"); | |
2062 | ret = -1; | |
2063 | goto out_abort; | |
2064 | } | |
2065 | ||
2066 | ret = lxc_inherit_namespaces(handler); | |
2067 | if (ret) { | |
2068 | SYSERROR("Failed to record inherited namespaces"); | |
2069 | ret = -1; | |
2070 | goto out_abort; | |
2071 | } | |
2072 | ||
e26cf563 CB |
2073 | /* If the rootfs is not a blockdev, prevent the container from marking |
2074 | * it readonly. | |
2075 | * If the container is unprivileged then skip rootfs pinning. | |
2076 | */ | |
0589d744 | 2077 | ret = lxc_rootfs_init(conf, !list_empty(&conf->id_map)); |
e26cf563 CB |
2078 | if (ret) { |
2079 | ERROR("Failed to handle rootfs pinning for container \"%s\"", handler->name); | |
2080 | ret = -1; | |
2081 | goto out_abort; | |
2082 | } | |
2083 | ||
0589d744 | 2084 | if (geteuid() == 0 && !list_empty(&conf->id_map)) { |
26ad2c6f | 2085 | /* |
4b875ef9 | 2086 | * Most filesystems can't be mounted inside a userns so handle them here. |
26ad2c6f | 2087 | */ |
4b875ef9 | 2088 | if (rootfs_is_blockdev(conf)) { |
c30e9b19 CB |
2089 | ret = unshare(CLONE_NEWNS); |
2090 | if (ret < 0) { | |
2091 | ERROR("Failed to unshare CLONE_NEWNS"); | |
9ad66589 | 2092 | goto out_abort; |
35120d9c | 2093 | } |
c30e9b19 | 2094 | INFO("Unshared CLONE_NEWNS"); |
408da065 | 2095 | |
8ce1abc2 | 2096 | ret = lxc_setup_rootfs_prepare_root(conf, name, lxcpath); |
c30e9b19 CB |
2097 | if (ret < 0) { |
2098 | ERROR("Error setting up rootfs mount as root before spawn"); | |
9ad66589 | 2099 | goto out_abort; |
35120d9c | 2100 | } |
c30e9b19 | 2101 | INFO("Set up container rootfs as host root"); |
35120d9c SH |
2102 | } |
2103 | } | |
2104 | ||
c30e9b19 CB |
2105 | ret = lxc_spawn(handler); |
2106 | if (ret < 0) { | |
2107 | ERROR("Failed to spawn container \"%s\"", name); | |
76a26f55 | 2108 | goto out_detach_blockdev; |
0ad19a3f | 2109 | } |
2110 | ||
80308d07 | 2111 | handler->conf->reboot = REBOOT_NONE; |
8bee8851 | 2112 | |
c30e9b19 CB |
2113 | ret = lxc_poll(name, handler); |
2114 | if (ret) { | |
2115 | ERROR("LXC mainloop exited with error: %d", ret); | |
9ad66589 | 2116 | goto out_delete_network; |
59eb99ba | 2117 | } |
0ad19a3f | 2118 | |
321db026 DJ |
2119 | if (!handler->init_died && handler->pid > 0) { |
2120 | ERROR("Child process is not killed"); | |
575ea467 | 2121 | ret = -1; |
9ad66589 | 2122 | goto out_delete_network; |
321db026 DJ |
2123 | } |
2124 | ||
c30e9b19 CB |
2125 | status = lxc_wait_for_pid_status(handler->pid); |
2126 | if (status < 0) | |
2127 | SYSERROR("Failed to retrieve status for %d", handler->pid); | |
e043236e | 2128 | |
408da065 CB |
2129 | /* If the child process exited but was not signaled, it didn't call |
2130 | * reboot. This should mean it was an lxc-execute which simply exited. | |
2131 | * In any case, treat it as a 'halt'. | |
8b004f07 | 2132 | */ |
d028235d | 2133 | if (WIFSIGNALED(status)) { |
b0dec2ce CB |
2134 | int signal_nr = WTERMSIG(status); |
2135 | switch(signal_nr) { | |
8b004f07 | 2136 | case SIGINT: /* halt */ |
b0dec2ce | 2137 | DEBUG("%s(%d) - Container \"%s\" is halting", signal_name(signal_nr), signal_nr, name); |
8b004f07 SH |
2138 | break; |
2139 | case SIGHUP: /* reboot */ | |
b0dec2ce | 2140 | DEBUG("%s(%d) - Container \"%s\" is rebooting", signal_name(signal_nr), signal_nr, name); |
80308d07 | 2141 | handler->conf->reboot = REBOOT_REQ; |
8b004f07 | 2142 | break; |
c2b9bd9e | 2143 | case SIGSYS: /* seccomp */ |
b0dec2ce | 2144 | DEBUG("%s(%d) - Container \"%s\" violated its seccomp policy", signal_name(signal_nr), signal_nr, name); |
c2b9bd9e | 2145 | break; |
8b004f07 | 2146 | default: |
b0dec2ce | 2147 | DEBUG("%s(%d) - Container \"%s\" init exited", signal_name(signal_nr), signal_nr, name); |
8b004f07 SH |
2148 | break; |
2149 | } | |
d028235d | 2150 | } |
828695d9 | 2151 | |
c30e9b19 CB |
2152 | ret = lxc_restore_phys_nics_to_netns(handler); |
2153 | if (ret < 0) | |
818a57fc | 2154 | ERROR("Failed to move physical network devices back to parent network namespace"); |
ce5782df | 2155 | |
1787abca | 2156 | lxc_monitor_send_exit_code(name, status, handler->lxcpath); |
eb808539 | 2157 | lxc_error_set_and_log(handler->pid, status); |
a3b4f3d6 TA |
2158 | if (error_num) |
2159 | *error_num = handler->exit_status; | |
1bd8d726 | 2160 | |
bb84beda | 2161 | lxc_delete_network(handler); |
76a26f55 | 2162 | detach_block_device(handler->conf); |
fd5be714 | 2163 | lxc_end(handler); |
c30e9b19 | 2164 | return ret; |
0ad19a3f | 2165 | |
9ad66589 | 2166 | out_abort: |
0c5859ff | 2167 | lxc_abort(handler); |
945daa24 CB |
2168 | lxc_end(handler); |
2169 | return ret; | |
65a2dc8b CB |
2170 | |
2171 | out_detach_blockdev: | |
0c5859ff | 2172 | lxc_abort(handler); |
945daa24 CB |
2173 | detach_block_device(handler->conf); |
2174 | lxc_end(handler); | |
2175 | return ret; | |
65a2dc8b | 2176 | |
9ad66589 | 2177 | out_delete_network: |
0c5859ff | 2178 | lxc_abort(handler); |
abd953eb | 2179 | lxc_restore_phys_nics_to_netns(handler); |
945daa24 CB |
2180 | lxc_delete_network(handler); |
2181 | detach_block_device(handler->conf); | |
2182 | lxc_end(handler); | |
2183 | return ret; | |
0ad19a3f | 2184 | } |
ee70bf78 CLG |
2185 | |
2186 | struct start_args { | |
2187 | char *const *argv; | |
2188 | }; | |
2189 | ||
2190 | static int start(struct lxc_handler *handler, void* data) | |
2191 | { | |
2192 | struct start_args *arg = data; | |
2193 | ||
984984e4 | 2194 | NOTICE("Exec'ing \"%s\"", arg->argv[0]); |
ee70bf78 CLG |
2195 | |
2196 | execvp(arg->argv[0], arg->argv); | |
984984e4 | 2197 | SYSERROR("Failed to exec \"%s\"", arg->argv[0]); |
ee70bf78 CLG |
2198 | return 0; |
2199 | } | |
2200 | ||
2201 | static int post_start(struct lxc_handler *handler, void* data) | |
2202 | { | |
2203 | struct start_args *arg = data; | |
2204 | ||
4c8e880e | 2205 | NOTICE("Started \"%s\" with pid \"%d\"", arg->argv[0], handler->pid); |
ee70bf78 CLG |
2206 | return 0; |
2207 | } | |
2208 | ||
2209 | static struct lxc_operations start_ops = { | |
2210 | .start = start, | |
2211 | .post_start = post_start | |
2212 | }; | |
2213 | ||
7bdf97d2 | 2214 | int lxc_start(char *const argv[], struct lxc_handler *handler, |
bb955810 | 2215 | const char *lxcpath, bool daemonize, int *error_num) |
ee70bf78 CLG |
2216 | { |
2217 | struct start_args start_arg = { | |
2218 | .argv = argv, | |
2219 | }; | |
2220 | ||
b2efeb0b | 2221 | TRACE("Doing lxc_start"); |
7bdf97d2 | 2222 | return __lxc_start(handler, &start_ops, &start_arg, lxcpath, daemonize, error_num); |
ee70bf78 | 2223 | } |
28272964 CB |
2224 | |
2225 | static void lxc_destroy_container_on_signal(struct lxc_handler *handler, | |
2226 | const char *name) | |
2227 | { | |
245100a0 | 2228 | char destroy[PATH_MAX]; |
f01f7975 | 2229 | struct lxc_container *c; |
ae3beac9 CB |
2230 | int ret = 0; |
2231 | bool bret = true; | |
2232 | ||
df31363a | 2233 | if (handler->conf->rootfs.path && handler->conf->rootfs.mount) { |
790255cf | 2234 | bret = do_destroy_container(handler); |
28272964 | 2235 | if (!bret) { |
ae3beac9 | 2236 | ERROR("Error destroying rootfs for container \"%s\"", name); |
28272964 CB |
2237 | return; |
2238 | } | |
2239 | } | |
ae3beac9 | 2240 | INFO("Destroyed rootfs for container \"%s\"", name); |
28272964 | 2241 | |
fa60cd7b CB |
2242 | ret = strnprintf(destroy, sizeof(destroy), "%s/%s", handler->lxcpath, name); |
2243 | if (ret < 0) { | |
ae3beac9 | 2244 | ERROR("Error destroying directory for container \"%s\"", name); |
28272964 CB |
2245 | return; |
2246 | } | |
2247 | ||
f01f7975 CB |
2248 | c = lxc_container_new(name, handler->lxcpath); |
2249 | if (c) { | |
2250 | if (container_disk_lock(c)) { | |
ae3beac9 | 2251 | INFO("Could not update lxc_snapshots file"); |
f01f7975 CB |
2252 | lxc_container_put(c); |
2253 | } else { | |
2254 | mod_all_rdeps(c, false); | |
2255 | container_disk_unlock(c); | |
2256 | lxc_container_put(c); | |
2257 | } | |
2258 | } | |
2259 | ||
d0fbc7ba | 2260 | if (!handler->am_root) |
94b0a3ac CB |
2261 | ret = userns_exec_full(handler->conf, lxc_rmdir_onedev_wrapper, |
2262 | destroy, "lxc_rmdir_onedev_wrapper"); | |
28272964 CB |
2263 | else |
2264 | ret = lxc_rmdir_onedev(destroy, NULL); | |
2265 | ||
2266 | if (ret < 0) { | |
ae3beac9 | 2267 | ERROR("Error destroying directory for container \"%s\"", name); |
28272964 CB |
2268 | return; |
2269 | } | |
ae3beac9 | 2270 | INFO("Destroyed directory for container \"%s\"", name); |
28272964 CB |
2271 | } |
2272 | ||
2273 | static int lxc_rmdir_onedev_wrapper(void *data) | |
2274 | { | |
2275 | char *arg = (char *) data; | |
2276 | return lxc_rmdir_onedev(arg, NULL); | |
2277 | } | |
2278 | ||
134df645 CB |
2279 | static bool do_destroy_container(struct lxc_handler *handler) |
2280 | { | |
94b0a3ac CB |
2281 | int ret; |
2282 | ||
d0fbc7ba | 2283 | if (!handler->am_root) { |
94b0a3ac CB |
2284 | ret = userns_exec_full(handler->conf, storage_destroy_wrapper, |
2285 | handler->conf, "storage_destroy_wrapper"); | |
2286 | if (ret < 0) | |
d028235d | 2287 | return false; |
10bc1861 | 2288 | |
d028235d SG |
2289 | return true; |
2290 | } | |
10bc1861 | 2291 | |
790255cf | 2292 | return storage_destroy(handler->conf); |
28272964 | 2293 | } |