]>
Commit | Line | Data |
---|---|---|
cc73685d | 1 | /* SPDX-License-Identifier: LGPL-2.1+ */ |
0ad19a3f | 2 | |
d38dd64a CB |
3 | #ifndef _GNU_SOURCE |
4 | #define _GNU_SOURCE 1 | |
5 | #endif | |
0ad19a3f | 6 | #include <dirent.h> |
7 | #include <errno.h> | |
b0a33c1e | 8 | #include <fcntl.h> |
c476bdce | 9 | #include <grp.h> |
37515ebd | 10 | #include <poll.h> |
b467714b | 11 | #include <pthread.h> |
f3787121 CB |
12 | #include <signal.h> |
13 | #include <stdio.h> | |
14 | #include <stdlib.h> | |
15 | #include <string.h> | |
0ad19a3f | 16 | #include <sys/file.h> |
f4d507d5 | 17 | #include <sys/mount.h> |
f3787121 | 18 | #include <sys/param.h> |
0ad19a3f | 19 | #include <sys/prctl.h> |
f3787121 CB |
20 | #include <sys/socket.h> |
21 | #include <sys/stat.h> | |
22 | #include <sys/syscall.h> | |
ddceb1f9 | 23 | #include <sys/types.h> |
b0a33c1e | 24 | #include <sys/un.h> |
f3787121 | 25 | #include <sys/wait.h> |
d38dd64a | 26 | #include <unistd.h> |
495d2046 | 27 | |
f3787121 CB |
28 | #include "af_unix.h" |
29 | #include "caps.h" | |
563f2f2c | 30 | #include "cgroup.h" |
f3787121 | 31 | #include "commands.h" |
bbf5cf35 | 32 | #include "commands_utils.h" |
f3787121 | 33 | #include "conf.h" |
d38dd64a | 34 | #include "config.h" |
74c6e2b0 | 35 | #include "confile_utils.h" |
e2bcd7db | 36 | #include "error.h" |
7fbb15ec | 37 | #include "file_utils.h" |
1cc8bd4b | 38 | #include "list.h" |
f3787121 | 39 | #include "log.h" |
d38dd64a | 40 | #include "lsm/lsm.h" |
f2d5a09d | 41 | #include "lxccontainer.h" |
f3787121 CB |
42 | #include "lxclock.h" |
43 | #include "lxcseccomp.h" | |
3ef9b3d3 | 44 | #include "macro.h" |
f3787121 | 45 | #include "mainloop.h" |
4ffeaf27 | 46 | #include "memory_utils.h" |
63376d7d | 47 | #include "monitor.h" |
f549edcc | 48 | #include "namespace.h" |
811ef482 | 49 | #include "network.h" |
38e5c2db | 50 | #include "raw_syscalls.h" |
f3787121 | 51 | #include "start.h" |
6be5397b CB |
52 | #include "storage/storage.h" |
53 | #include "storage/storage_utils.h" | |
0ed9b1bc | 54 | #include "sync.h" |
59524108 | 55 | #include "syscall_wrappers.h" |
0ed9b1bc CB |
56 | #include "terminal.h" |
57 | #include "utils.h" | |
36eb9bde | 58 | |
d38dd64a CB |
59 | #if HAVE_LIBCAP |
60 | #include <sys/capability.h> | |
61 | #endif | |
62 | ||
9de31d5a CB |
63 | #ifndef HAVE_STRLCPY |
64 | #include "include/strlcpy.h" | |
65 | #endif | |
66 | ||
ac2cecc4 | 67 | lxc_log_define(start, lxc); |
36eb9bde | 68 | |
f01f7975 | 69 | extern void mod_all_rdeps(struct lxc_container *c, bool inc); |
790255cf | 70 | static bool do_destroy_container(struct lxc_handler *handler); |
28272964 CB |
71 | static int lxc_rmdir_onedev_wrapper(void *data); |
72 | static void lxc_destroy_container_on_signal(struct lxc_handler *handler, | |
73 | const char *name); | |
74 | ||
c8154066 SH |
75 | static void print_top_failing_dir(const char *path) |
76 | { | |
dcd9a847 | 77 | __do_free char *copy = NULL; |
9e5f5f2f | 78 | int ret; |
4ffeaf27 | 79 | char *e, *p, saved; |
cbb9c7c7 | 80 | |
4ffeaf27 | 81 | copy = must_copy_string(path); |
c8154066 | 82 | p = copy; |
4ffeaf27 | 83 | e = copy + strlen(path); |
cbb9c7c7 | 84 | |
c8154066 | 85 | while (p < e) { |
f3787121 CB |
86 | while (p < e && *p == '/') |
87 | p++; | |
9e5f5f2f | 88 | |
f3787121 CB |
89 | while (p < e && *p != '/') |
90 | p++; | |
9e5f5f2f | 91 | |
c8154066 SH |
92 | saved = *p; |
93 | *p = '\0'; | |
9e5f5f2f CB |
94 | |
95 | ret = access(copy, X_OK); | |
96 | if (ret != 0) { | |
818a57fc | 97 | SYSERROR("Could not access %s. Please grant it x access, or add an ACL for the container " "root", copy); |
c8154066 SH |
98 | return; |
99 | } | |
100 | *p = saved; | |
101 | } | |
102 | } | |
103 | ||
5032bf39 | 104 | static void lxc_put_nsfds(struct lxc_handler *handler) |
408da065 | 105 | { |
818a57fc | 106 | for (int i = 0; i < LXC_NS_MAX; i++) { |
5032bf39 | 107 | if (handler->nsfd[i] < 0) |
39cd919c CB |
108 | continue; |
109 | ||
818a57fc | 110 | close_prot_errno_disarm(handler->nsfd[i]); |
9f30a190 | 111 | } |
9f30a190 MM |
112 | } |
113 | ||
4cb53844 CB |
114 | static int lxc_try_preserve_ns(const int pid, const char *ns) |
115 | { | |
116 | int fd; | |
117 | ||
118 | fd = lxc_preserve_ns(pid, ns); | |
119 | if (fd < 0) { | |
818a57fc CB |
120 | if (errno != ENOENT) |
121 | return log_error_errno(-EINVAL, | |
122 | errno, "Failed to preserve %s namespace", | |
123 | ns); | |
4cb53844 | 124 | |
818a57fc CB |
125 | return log_error_errno(-EOPNOTSUPP, |
126 | errno, "Kernel does not support preserving %s namespaces", | |
127 | ns); | |
4cb53844 CB |
128 | } |
129 | ||
130 | return fd; | |
131 | } | |
132 | ||
133 | /* lxc_try_preserve_namespaces: open /proc/@pid/ns/@ns for each namespace | |
134 | * specified in ns_clone_flags. | |
4d8ac866 | 135 | * Return true on success, false on failure. |
62d05d9b | 136 | */ |
4cb53844 CB |
137 | static bool lxc_try_preserve_namespaces(struct lxc_handler *handler, |
138 | int ns_clone_flags, pid_t pid) | |
f3787121 | 139 | { |
9fef3355 | 140 | int i; |
9f30a190 | 141 | |
9f30a190 | 142 | for (i = 0; i < LXC_NS_MAX; i++) |
5032bf39 | 143 | handler->nsfd[i] = -EBADF; |
cd43d2d1 | 144 | |
9f30a190 | 145 | for (i = 0; i < LXC_NS_MAX; i++) { |
4cb53844 CB |
146 | int fd; |
147 | ||
becad0ec | 148 | if ((ns_clone_flags & ns_info[i].clone_flag) == 0) |
9f30a190 | 149 | continue; |
28d9e29e | 150 | |
4cb53844 CB |
151 | fd = lxc_try_preserve_ns(pid, ns_info[i].proc_name); |
152 | if (fd < 0) { | |
4cb53844 CB |
153 | /* Do not fail to start container on kernels that do |
154 | * not support interacting with namespaces through | |
155 | * /proc. | |
156 | */ | |
157 | if (fd == -EOPNOTSUPP) | |
158 | continue; | |
fa3a5b22 | 159 | |
4cb53844 CB |
160 | lxc_put_nsfds(handler); |
161 | return false; | |
162 | } | |
163 | ||
164 | handler->nsfd[i] = fd; | |
165 | DEBUG("Preserved %s namespace via fd %d", ns_info[i].proc_name, | |
166 | handler->nsfd[i]); | |
9f30a190 MM |
167 | } |
168 | ||
62d05d9b | 169 | return true; |
9f30a190 MM |
170 | } |
171 | ||
ec1dc633 | 172 | static inline bool match_stdfds(int fd) |
80090207 | 173 | { |
ec1dc633 | 174 | return (fd == STDIN_FILENO || fd == STDOUT_FILENO || fd == STDERR_FILENO); |
80090207 CLG |
175 | } |
176 | ||
4c03724e | 177 | #ifdef HAVE_DLOG |
178 | static bool match_dlog_fds(struct dirent *direntp) | |
179 | { | |
180 | char path[PATH_MAX] = {0}; | |
181 | char link[PATH_MAX] = {0}; | |
182 | ssize_t linklen; | |
183 | int ret; | |
184 | ||
185 | ret = snprintf(path, PATH_MAX, "/proc/self/fd/%s", direntp->d_name); | |
818a57fc CB |
186 | if (ret < 0 || ret >= PATH_MAX) |
187 | return log_error(false, "Failed to create file descriptor name"); | |
4c03724e | 188 | |
189 | linklen = readlink(path, link, PATH_MAX); | |
818a57fc CB |
190 | if (linklen < 0) |
191 | return log_error(false, "Failed to read link path - \"%s\"", path); | |
192 | else if (linklen >= PATH_MAX) | |
193 | return log_error(false, "The name of link path is too long - \"%s\"", path); | |
194 | ||
195 | if (strcmp(link, "/dev/log_main") == 0 || | |
196 | strcmp(link, "/dev/log_system") == 0 || | |
197 | strcmp(link, "/dev/log_radio") == 0) | |
4c03724e | 198 | return true; |
199 | ||
200 | return false; | |
201 | } | |
202 | #endif | |
203 | ||
47a46cf1 CB |
204 | int lxc_check_inherited(struct lxc_conf *conf, bool closeall, |
205 | int *fds_to_ignore, size_t len_fds) | |
80090207 | 206 | { |
80090207 | 207 | int fd, fddir; |
47a46cf1 | 208 | size_t i; |
80090207 | 209 | DIR *dir; |
a5a70219 | 210 | struct dirent *direntp; |
80090207 | 211 | |
d2cf4c37 SH |
212 | if (conf && conf->close_all_fds) |
213 | closeall = true; | |
214 | ||
b119f362 | 215 | restart: |
80090207 | 216 | dir = opendir("/proc/self/fd"); |
818a57fc CB |
217 | if (!dir) |
218 | return log_warn(-1, "Failed to open directory"); | |
80090207 CLG |
219 | |
220 | fddir = dirfd(dir); | |
221 | ||
74f96976 | 222 | while ((direntp = readdir(dir))) { |
a5a70219 | 223 | int ret; |
d39b10eb CB |
224 | struct lxc_list *cur; |
225 | bool matched = false; | |
226 | ||
a5a70219 | 227 | if (strcmp(direntp->d_name, ".") == 0) |
80090207 CLG |
228 | continue; |
229 | ||
a5a70219 | 230 | if (strcmp(direntp->d_name, "..") == 0) |
80090207 CLG |
231 | continue; |
232 | ||
a5a70219 CB |
233 | ret = lxc_safe_int(direntp->d_name, &fd); |
234 | if (ret < 0) { | |
235 | INFO("Could not parse file descriptor for \"%s\"", direntp->d_name); | |
d4cff0d2 CB |
236 | continue; |
237 | } | |
80090207 | 238 | |
47a46cf1 CB |
239 | for (i = 0; i < len_fds; i++) |
240 | if (fds_to_ignore[i] == fd) | |
241 | break; | |
242 | ||
243 | if (fd == fddir || fd == lxc_log_fd || | |
244 | (i < len_fds && fd == fds_to_ignore[i])) | |
80090207 CLG |
245 | continue; |
246 | ||
d39b10eb | 247 | /* Keep state clients that wait on reboots. */ |
3ee9e4fb CB |
248 | if (conf) { |
249 | lxc_list_for_each(cur, &conf->state_clients) { | |
250 | struct lxc_state_client *client = cur->elem; | |
d39b10eb | 251 | |
3ee9e4fb CB |
252 | if (client->clientfd != fd) |
253 | continue; | |
d39b10eb | 254 | |
3ee9e4fb CB |
255 | matched = true; |
256 | break; | |
257 | } | |
d39b10eb CB |
258 | } |
259 | ||
260 | if (matched) | |
261 | continue; | |
262 | ||
858377e4 SH |
263 | if (current_config && fd == current_config->logfd) |
264 | continue; | |
265 | ||
ec1dc633 | 266 | if (match_stdfds(fd)) |
80090207 | 267 | continue; |
80090207 | 268 | |
1d5e5f26 | 269 | #ifdef HAVE_DLOG |
270 | if (match_dlog_fds(direntp)) | |
271 | continue; | |
272 | ||
273 | #endif | |
d2cf4c37 | 274 | if (closeall) { |
b119f362 SH |
275 | close(fd); |
276 | closedir(dir); | |
28d9e29e | 277 | INFO("Closed inherited fd %d", fd); |
b119f362 SH |
278 | goto restart; |
279 | } | |
28d9e29e | 280 | WARN("Inherited fd %d", fd); |
80090207 CLG |
281 | } |
282 | ||
408da065 | 283 | /* Only enable syslog at this point to avoid the above logging function |
64c57ea1 BD |
284 | * to open a new fd and make the check_inherited function enter an |
285 | * infinite loop. | |
286 | */ | |
287 | lxc_log_enable_syslog(); | |
288 | ||
92c7f629 GK |
289 | closedir(dir); /* cannot fail */ |
290 | return 0; | |
80090207 CLG |
291 | } |
292 | ||
83ee7875 | 293 | static int setup_signal_fd(sigset_t *oldmask) |
b0a33c1e | 294 | { |
7288dfb6 | 295 | int ret; |
b0a33c1e | 296 | sigset_t mask; |
7288dfb6 | 297 | const int signals[] = {SIGBUS, SIGILL, SIGSEGV, SIGWINCH}; |
b0a33c1e | 298 | |
408da065 | 299 | /* Block everything except serious error signals. */ |
df0795b1 CB |
300 | ret = sigfillset(&mask); |
301 | if (ret < 0) | |
302 | return -EBADF; | |
303 | ||
9b10bef5 | 304 | for (int sig = 0; sig < (sizeof(signals) / sizeof(signals[0])); sig++) { |
df0795b1 CB |
305 | ret = sigdelset(&mask, signals[sig]); |
306 | if (ret < 0) | |
307 | return -EBADF; | |
b0a33c1e | 308 | } |
309 | ||
b467714b | 310 | ret = pthread_sigmask(SIG_BLOCK, &mask, oldmask); |
818a57fc CB |
311 | if (ret < 0) |
312 | return log_error_errno(-EBADF, errno, | |
313 | "Failed to set signal mask"); | |
b0a33c1e | 314 | |
df0795b1 | 315 | ret = signalfd(-1, &mask, SFD_CLOEXEC); |
818a57fc CB |
316 | if (ret < 0) |
317 | return log_error_errno(-EBADF, | |
318 | errno, "Failed to create signal file descriptor"); | |
b0a33c1e | 319 | |
df0795b1 | 320 | TRACE("Created signal file descriptor %d", ret); |
1ac470c0 | 321 | |
df0795b1 | 322 | return ret; |
b0a33c1e | 323 | } |
324 | ||
84c92abd | 325 | static int signal_handler(int fd, uint32_t events, void *data, |
f3787121 | 326 | struct lxc_epoll_descr *descr) |
b0a33c1e | 327 | { |
15cd25fd | 328 | int ret; |
3c319edb CB |
329 | siginfo_t info; |
330 | struct signalfd_siginfo siginfo; | |
331 | struct lxc_handler *hdlr = data; | |
15cd25fd | 332 | |
489f39be | 333 | ret = lxc_read_nointr(fd, &siginfo, sizeof(siginfo)); |
818a57fc CB |
334 | if (ret < 0) |
335 | return log_error(LXC_MAINLOOP_ERROR, "Failed to read signal info from signal file descriptor %d", fd); | |
15cd25fd | 336 | |
818a57fc CB |
337 | if (ret != sizeof(siginfo)) |
338 | return log_error(-EINVAL, "Unexpected size for struct signalfd_siginfo"); | |
15cd25fd | 339 | |
408da065 | 340 | /* Check whether init is running. */ |
80507ee8 | 341 | info.si_pid = 0; |
3c319edb CB |
342 | ret = waitid(P_PID, hdlr->pid, &info, WEXITED | WNOWAIT | WNOHANG); |
343 | if (ret == 0 && info.si_pid == hdlr->pid) | |
344 | hdlr->init_died = true; | |
80507ee8 | 345 | |
cd5177e9 TA |
346 | /* Try to figure out a reasonable exit status to report. */ |
347 | if (hdlr->init_died) { | |
348 | switch (info.si_code) { | |
349 | case CLD_EXITED: | |
350 | hdlr->exit_status = info.si_status << 8; | |
351 | break; | |
352 | case CLD_KILLED: | |
353 | case CLD_DUMPED: | |
354 | case CLD_STOPPED: | |
355 | hdlr->exit_status = info.si_status << 8 | 0x7f; | |
356 | break; | |
357 | case CLD_CONTINUED: | |
358 | /* Huh? The waitid() told us it's dead *and* continued? */ | |
359 | WARN("Init %d dead and continued?", hdlr->pid); | |
360 | hdlr->exit_status = 1; | |
361 | break; | |
362 | default: | |
20993a97 | 363 | ERROR("Unknown si_code: %d", info.si_code); |
55e1311e | 364 | hdlr->exit_status = 1; |
cd5177e9 TA |
365 | } |
366 | } | |
367 | ||
d4b5d7a8 | 368 | if (siginfo.ssi_signo == SIGHUP) { |
33942046 CB |
369 | if (hdlr->pidfd >= 0) |
370 | lxc_raw_pidfd_send_signal(hdlr->pidfd, SIGTERM, NULL, 0); | |
d9bb2fba CB |
371 | else |
372 | kill(hdlr->pid, SIGTERM); | |
d4b5d7a8 | 373 | INFO("Killing %d since terminal hung up", hdlr->pid); |
9b10bef5 CB |
374 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE |
375 | : LXC_MAINLOOP_CONTINUE; | |
d4b5d7a8 TA |
376 | } |
377 | ||
9cb94384 | 378 | if (siginfo.ssi_signo != SIGCHLD) { |
33942046 CB |
379 | if (hdlr->pidfd >= 0) |
380 | lxc_raw_pidfd_send_signal(hdlr->pidfd, | |
381 | siginfo.ssi_signo, NULL, 0); | |
d9bb2fba CB |
382 | else |
383 | kill(hdlr->pid, siginfo.ssi_signo); | |
9cb94384 | 384 | INFO("Forwarded signal %d to pid %d", siginfo.ssi_signo, hdlr->pid); |
9b10bef5 CB |
385 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE |
386 | : LXC_MAINLOOP_CONTINUE; | |
9cb94384 TA |
387 | } |
388 | ||
3a9e949f TA |
389 | /* More robustness, protect ourself from a SIGCHLD sent |
390 | * by a process different from the container init. | |
391 | */ | |
392 | if (siginfo.ssi_pid != hdlr->pid) { | |
6e94162a CB |
393 | NOTICE("Received %d from pid %d instead of container init %d", |
394 | siginfo.ssi_signo, siginfo.ssi_pid, hdlr->pid); | |
9b10bef5 CB |
395 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE |
396 | : LXC_MAINLOOP_CONTINUE; | |
3a9e949f TA |
397 | } |
398 | ||
408da065 | 399 | if (siginfo.ssi_code == CLD_STOPPED) { |
6e94162a | 400 | INFO("Container init process was stopped"); |
9b10bef5 CB |
401 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE |
402 | : LXC_MAINLOOP_CONTINUE; | |
403 | } | |
404 | ||
405 | if (siginfo.ssi_code == CLD_CONTINUED) { | |
6e94162a | 406 | INFO("Container init process was continued"); |
9b10bef5 CB |
407 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE |
408 | : LXC_MAINLOOP_CONTINUE; | |
15cd25fd | 409 | } |
1ac470c0 | 410 | |
818a57fc | 411 | return log_debug(LXC_MAINLOOP_CLOSE, "Container init process %d exited", hdlr->pid); |
b0a33c1e | 412 | } |
413 | ||
974a8aba CB |
414 | int lxc_serve_state_clients(const char *name, struct lxc_handler *handler, |
415 | lxc_state_t state) | |
66aeffc7 | 416 | { |
9de31d5a | 417 | size_t retlen; |
dbc9832d CB |
418 | ssize_t ret; |
419 | struct lxc_list *cur, *next; | |
dbc9832d CB |
420 | struct lxc_msg msg = {.type = lxc_msg_state, .value = state}; |
421 | ||
974a8aba CB |
422 | if (state == THAWED) |
423 | handler->state = RUNNING; | |
424 | else | |
425 | handler->state = state; | |
426 | ||
9dfa4041 | 427 | TRACE("Set container state to %s", lxc_state2str(state)); |
dbc9832d | 428 | |
818a57fc CB |
429 | if (lxc_list_empty(&handler->conf->state_clients)) |
430 | return log_trace(0, "No state clients registered"); | |
dbc9832d | 431 | |
9de31d5a CB |
432 | retlen = strlcpy(msg.name, name, sizeof(msg.name)); |
433 | if (retlen >= sizeof(msg.name)) | |
434 | return -E2BIG; | |
dbc9832d | 435 | |
d39b10eb | 436 | lxc_list_for_each_safe(cur, &handler->conf->state_clients, next) { |
d85617bc | 437 | struct lxc_state_client *client = cur->elem; |
dbc9832d | 438 | |
d39b10eb | 439 | if (client->states[state] == 0) { |
9dfa4041 | 440 | TRACE("State %s not registered for state client %d", |
dbc9832d CB |
441 | lxc_state2str(state), client->clientfd); |
442 | continue; | |
443 | } | |
444 | ||
9dfa4041 | 445 | TRACE("Sending state %s to state client %d", |
dbc9832d CB |
446 | lxc_state2str(state), client->clientfd); |
447 | ||
7fbb15ec CB |
448 | ret = lxc_send_nointr(client->clientfd, &msg, sizeof(msg), MSG_NOSIGNAL); |
449 | if (ret <= 0) | |
6d1400b5 | 450 | SYSERROR("Failed to send message to client"); |
dbc9832d CB |
451 | |
452 | /* kick client from list */ | |
dbc9832d | 453 | lxc_list_del(cur); |
300d1cb4 | 454 | close(client->clientfd); |
dbc9832d CB |
455 | free(cur->elem); |
456 | free(cur); | |
457 | } | |
dbc9832d | 458 | |
5e5576a4 CB |
459 | return 0; |
460 | } | |
461 | ||
462 | static int lxc_serve_state_socket_pair(const char *name, | |
463 | struct lxc_handler *handler, | |
464 | lxc_state_t state) | |
465 | { | |
466 | ssize_t ret; | |
467 | ||
bb955810 | 468 | if (!handler->daemonize || |
5e5576a4 CB |
469 | handler->state_socket_pair[1] < 0 || |
470 | state == STARTING) | |
471 | return 0; | |
472 | ||
473 | /* Close read end of the socket pair. */ | |
474 | close(handler->state_socket_pair[0]); | |
475 | handler->state_socket_pair[0] = -1; | |
476 | ||
ee8377bd | 477 | again: |
5e5576a4 CB |
478 | ret = lxc_abstract_unix_send_credential(handler->state_socket_pair[1], |
479 | &(int){state}, sizeof(int)); | |
9044b79e | 480 | if (ret < 0) { |
481 | SYSERROR("Failed to send state to %d", handler->state_socket_pair[1]); | |
482 | ||
ee8377bd CB |
483 | if (errno == EINTR) |
484 | goto again; | |
9044b79e | 485 | |
486 | return -1; | |
487 | } | |
488 | ||
818a57fc CB |
489 | if (ret != sizeof(int)) |
490 | return log_error(-1, "Message too long : %d", handler->state_socket_pair[1]); | |
5e5576a4 CB |
491 | |
492 | TRACE("Sent container state \"%s\" to %d", lxc_state2str(state), | |
493 | handler->state_socket_pair[1]); | |
494 | ||
495 | /* Close write end of the socket pair. */ | |
818a57fc | 496 | close_prot_errno_disarm(handler->state_socket_pair[1]); |
5e5576a4 CB |
497 | |
498 | return 0; | |
499 | } | |
500 | ||
501 | int lxc_set_state(const char *name, struct lxc_handler *handler, | |
502 | lxc_state_t state) | |
503 | { | |
504 | int ret; | |
505 | ||
506 | ret = lxc_serve_state_socket_pair(name, handler, state); | |
818a57fc CB |
507 | if (ret < 0) |
508 | return log_error(-1, "Failed to synchronize via anonymous pair of unix sockets"); | |
5e5576a4 CB |
509 | |
510 | ret = lxc_serve_state_clients(name, handler, state); | |
511 | if (ret < 0) | |
512 | return -1; | |
513 | ||
dbc9832d CB |
514 | /* This function will try to connect to the legacy lxc-monitord state |
515 | * server and only exists for backwards compatibility. | |
516 | */ | |
9123e471 | 517 | lxc_monitor_send_state(name, state, handler->lxcpath); |
dbc9832d | 518 | |
66aeffc7 DL |
519 | return 0; |
520 | } | |
521 | ||
735f2c6e | 522 | int lxc_poll(const char *name, struct lxc_handler *handler) |
b0a33c1e | 523 | { |
3c319edb | 524 | int ret; |
86530b0a | 525 | bool has_console = true; |
3c319edb | 526 | struct lxc_epoll_descr descr, descr_console; |
b0a33c1e | 527 | |
03760215 CB |
528 | if (handler->conf->console.path && |
529 | strcmp(handler->conf->console.path, "none") == 0) | |
86530b0a L |
530 | has_console = false; |
531 | ||
3c319edb CB |
532 | ret = lxc_mainloop_open(&descr); |
533 | if (ret < 0) { | |
534 | ERROR("Failed to create mainloop"); | |
50c8bf05 | 535 | goto out_sigfd; |
b0a33c1e | 536 | } |
537 | ||
30a33fbd CB |
538 | if (has_console) { |
539 | ret = lxc_mainloop_open(&descr_console); | |
540 | if (ret < 0) { | |
541 | ERROR("Failed to create console mainloop"); | |
542 | goto out_mainloop; | |
543 | } | |
3c319edb CB |
544 | } |
545 | ||
1cc8bd4b | 546 | ret = lxc_mainloop_add_handler(&descr, handler->sigfd, signal_handler, handler); |
3c319edb | 547 | if (ret < 0) { |
1cc8bd4b | 548 | ERROR("Failed to add signal handler for %d to mainloop", handler->sigfd); |
3c319edb | 549 | goto out_mainloop_console; |
b0a33c1e | 550 | } |
551 | ||
2ac0f627 CB |
552 | ret = lxc_seccomp_setup_proxy(&handler->conf->seccomp, &descr, handler); |
553 | if (ret < 0) { | |
554 | ERROR("Failed to setup seccomp proxy"); | |
c3e3c21a | 555 | goto out_mainloop_console; |
2ac0f627 | 556 | } |
cdb2a47f | 557 | |
30a33fbd | 558 | if (has_console) { |
dcad02f8 | 559 | struct lxc_terminal *console = &handler->conf->console; |
63376d7d | 560 | |
093bce5f | 561 | ret = lxc_terminal_mainloop_add(&descr, console); |
30a33fbd CB |
562 | if (ret < 0) { |
563 | ERROR("Failed to add console handlers to mainloop"); | |
564 | goto out_mainloop_console; | |
565 | } | |
566 | ||
093bce5f | 567 | ret = lxc_terminal_mainloop_add(&descr_console, console); |
30a33fbd CB |
568 | if (ret < 0) { |
569 | ERROR("Failed to add console handlers to console mainloop"); | |
570 | goto out_mainloop_console; | |
571 | } | |
a54585ad L |
572 | |
573 | handler->conf->console.descr = &descr; | |
563f2f2c DL |
574 | } |
575 | ||
3c319edb CB |
576 | ret = lxc_cmd_mainloop_add(name, &descr, handler); |
577 | if (ret < 0) { | |
578 | ERROR("Failed to add command handler to mainloop"); | |
579 | goto out_mainloop_console; | |
580 | } | |
b0a33c1e | 581 | |
3c319edb | 582 | TRACE("Mainloop is ready"); |
b0a33c1e | 583 | |
3c319edb | 584 | ret = lxc_mainloop(&descr, -1); |
1cc8bd4b CB |
585 | close(descr.epfd); |
586 | descr.epfd = -EBADF; | |
587 | if (ret < 0 || !handler->init_died) | |
42b09f94 | 588 | goto out_mainloop_console; |
1cc8bd4b | 589 | |
30a33fbd CB |
590 | if (has_console) |
591 | ret = lxc_mainloop(&descr_console, 0); | |
592 | ||
3c319edb | 593 | out_mainloop_console: |
30a33fbd CB |
594 | if (has_console) { |
595 | lxc_mainloop_close(&descr_console); | |
596 | TRACE("Closed console mainloop"); | |
597 | } | |
3c319edb | 598 | |
42b09f94 CB |
599 | out_mainloop: |
600 | lxc_mainloop_close(&descr); | |
601 | TRACE("Closed mainloop"); | |
602 | ||
b0a33c1e | 603 | out_sigfd: |
1cc8bd4b CB |
604 | close(handler->sigfd); |
605 | TRACE("Closed signal file descriptor %d", handler->sigfd); | |
606 | handler->sigfd = -EBADF; | |
408da065 | 607 | |
1cc8bd4b | 608 | return ret; |
b0a33c1e | 609 | } |
610 | ||
41784e4e CB |
611 | void lxc_zero_handler(struct lxc_handler *handler) |
612 | { | |
41784e4e CB |
613 | memset(handler, 0, sizeof(struct lxc_handler)); |
614 | ||
818a57fc | 615 | handler->pinfd = -EBADF; |
41784e4e | 616 | |
33942046 CB |
617 | handler->pidfd = -EBADF; |
618 | ||
818a57fc | 619 | handler->sigfd = -EBADF; |
41784e4e | 620 | |
818a57fc CB |
621 | for (int i = 0; i < LXC_NS_MAX; i++) |
622 | handler->nsfd[i] = -EBADF; | |
41784e4e | 623 | |
818a57fc CB |
624 | handler->data_sock[0] = -EBADF; |
625 | handler->data_sock[1] = -EBADF; | |
41784e4e | 626 | |
818a57fc CB |
627 | handler->state_socket_pair[0] = -EBADF; |
628 | handler->state_socket_pair[1] = -EBADF; | |
41784e4e | 629 | |
818a57fc CB |
630 | handler->sync_sock[0] = -EBADF; |
631 | handler->sync_sock[1] = -EBADF; | |
41784e4e CB |
632 | } |
633 | ||
f2e07cb6 CB |
634 | void lxc_free_handler(struct lxc_handler *handler) |
635 | { | |
1cc8bd4b | 636 | if (handler->pinfd >= 0) |
818a57fc | 637 | close_prot_errno_disarm(handler->pinfd); |
1cc8bd4b | 638 | |
33942046 | 639 | if (handler->pidfd >= 0) |
818a57fc | 640 | close_prot_errno_disarm(handler->pidfd); |
33942046 | 641 | |
1cc8bd4b | 642 | if (handler->sigfd >= 0) |
818a57fc | 643 | close_prot_errno_disarm(handler->sigfd); |
1cc8bd4b CB |
644 | |
645 | lxc_put_nsfds(handler); | |
646 | ||
80308d07 | 647 | if (handler->conf && handler->conf->reboot == REBOOT_NONE) |
0d0d3655 | 648 | if (handler->conf->maincmd_fd >= 0) |
9044b79e | 649 | lxc_abstract_unix_close(handler->conf->maincmd_fd); |
f2e07cb6 | 650 | |
4d8bdfa0 | 651 | if (handler->monitor_status_fd >= 0) |
818a57fc | 652 | close_prot_errno_disarm(handler->monitor_status_fd); |
4d8bdfa0 | 653 | |
5e5576a4 | 654 | if (handler->state_socket_pair[0] >= 0) |
818a57fc | 655 | close_prot_errno_disarm(handler->state_socket_pair[0]); |
5e5576a4 CB |
656 | |
657 | if (handler->state_socket_pair[1] >= 0) | |
818a57fc | 658 | close_prot_errno_disarm(handler->state_socket_pair[1]); |
5e5576a4 | 659 | |
9b10bef5 CB |
660 | if (handler->cgroup_ops) |
661 | cgroup_exit(handler->cgroup_ops); | |
662 | ||
f2e07cb6 | 663 | handler->conf = NULL; |
818a57fc | 664 | free_disarm(handler); |
f2e07cb6 CB |
665 | } |
666 | ||
aa460476 | 667 | struct lxc_handler *lxc_init_handler(const char *name, struct lxc_conf *conf, |
5e5576a4 | 668 | const char *lxcpath, bool daemonize) |
59eb99ba | 669 | { |
818a57fc | 670 | int ret; |
3a0f472d DL |
671 | struct lxc_handler *handler; |
672 | ||
673 | handler = malloc(sizeof(*handler)); | |
fdecbc9c | 674 | if (!handler) |
3a0f472d | 675 | return NULL; |
9044b79e | 676 | |
59eb99ba DL |
677 | memset(handler, 0, sizeof(*handler)); |
678 | ||
fdecbc9c CB |
679 | /* Note that am_guest_unpriv() checks the effective uid. We |
680 | * probably don't care if we are real root only if we are running | |
681 | * as root so this should be fine. | |
790255cf | 682 | */ |
e0010464 | 683 | handler->am_root = !am_guest_unpriv(); |
fae349da | 684 | handler->conf = conf; |
9123e471 | 685 | handler->lxcpath = lxcpath; |
818a57fc CB |
686 | handler->init_died = false; |
687 | handler->data_sock[0] = -EBADF; | |
688 | handler->data_sock[1] = -EBADF; | |
4d8bdfa0 | 689 | handler->monitor_status_fd = -EBADF; |
818a57fc | 690 | handler->pinfd = -EBADF; |
33942046 | 691 | handler->pidfd = -EBADF; |
1cc8bd4b | 692 | handler->sigfd = -EBADF; |
818a57fc CB |
693 | handler->state_socket_pair[0] = -EBADF; |
694 | handler->state_socket_pair[1] = -EBADF; | |
80308d07 | 695 | if (handler->conf->reboot == REBOOT_NONE) |
d39b10eb | 696 | lxc_list_init(&handler->conf->state_clients); |
fae349da | 697 | |
818a57fc CB |
698 | for (int i = 0; i < LXC_NS_MAX; i++) |
699 | handler->nsfd[i] = -EBADF; | |
b6b2b194 | 700 | |
f0ecc19d | 701 | handler->name = name; |
c581d2a6 CB |
702 | if (daemonize) |
703 | handler->transient_pid = lxc_raw_getpid(); | |
704 | else | |
705 | handler->transient_pid = -1; | |
3bdf52d7 | 706 | |
80308d07 | 707 | if (daemonize && handler->conf->reboot == REBOOT_NONE) { |
5e5576a4 | 708 | /* Create socketpair() to synchronize on daemonized startup. |
fdecbc9c CB |
709 | * When the container reboots we don't need to synchronize |
710 | * again currently so don't open another socketpair(). | |
5e5576a4 CB |
711 | */ |
712 | ret = socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0, | |
713 | handler->state_socket_pair); | |
714 | if (ret < 0) { | |
715 | ERROR("Failed to create anonymous pair of unix sockets"); | |
716 | goto on_error; | |
717 | } | |
9044b79e | 718 | |
5e5576a4 CB |
719 | TRACE("Created anonymous pair {%d,%d} of unix sockets", |
720 | handler->state_socket_pair[0], | |
721 | handler->state_socket_pair[1]); | |
722 | } | |
723 | ||
80308d07 | 724 | if (handler->conf->reboot == REBOOT_NONE) { |
d39b10eb CB |
725 | handler->conf->maincmd_fd = lxc_cmd_init(name, lxcpath, "command"); |
726 | if (handler->conf->maincmd_fd < 0) { | |
727 | ERROR("Failed to set up command socket"); | |
728 | goto on_error; | |
729 | } | |
aa460476 | 730 | } |
9044b79e | 731 | |
9dfa4041 | 732 | TRACE("Unix domain socket %d for command server is ready", |
aa460476 CB |
733 | handler->conf->maincmd_fd); |
734 | ||
735 | return handler; | |
736 | ||
f2e07cb6 CB |
737 | on_error: |
738 | lxc_free_handler(handler); | |
aa460476 CB |
739 | |
740 | return NULL; | |
741 | } | |
742 | ||
743 | int lxc_init(const char *name, struct lxc_handler *handler) | |
744 | { | |
4d8bdfa0 | 745 | __do_close_prot_errno int status_fd = -EBADF; |
a2c09be0 | 746 | int ret; |
4a03ded4 | 747 | const char *loglevel; |
aa460476 CB |
748 | struct lxc_conf *conf = handler->conf; |
749 | ||
434c8e15 | 750 | handler->monitor_pid = lxc_raw_getpid(); |
4d8bdfa0 CB |
751 | status_fd = open("/proc/self/status", O_RDONLY | O_CLOEXEC); |
752 | if (status_fd < 0) { | |
753 | SYSERROR("Failed to open monitor status fd"); | |
754 | goto out_close_maincmd_fd; | |
755 | } | |
434c8e15 | 756 | |
aa460476 | 757 | lsm_init(); |
2170b263 | 758 | TRACE("Initialized LSM"); |
d2e30e99 | 759 | |
2170b263 CB |
760 | ret = lxc_read_seccomp_config(conf); |
761 | if (ret < 0) { | |
762 | ERROR("Failed loading seccomp policy"); | |
d2e30e99 | 763 | goto out_close_maincmd_fd; |
8f2c3a70 | 764 | } |
2170b263 | 765 | TRACE("Read seccomp policy"); |
8f2c3a70 | 766 | |
408da065 | 767 | /* Begin by setting the state to STARTING. */ |
2170b263 CB |
768 | ret = lxc_set_state(name, handler, STARTING); |
769 | if (ret < 0) { | |
770 | ERROR("Failed to set state to \"%s\"", lxc_state2str(STARTING)); | |
051151de | 771 | goto out_close_maincmd_fd; |
0ad19a3f | 772 | } |
46800e77 | 773 | TRACE("Set container state to \"STARTING\""); |
0ad19a3f | 774 | |
408da065 | 775 | /* Start of environment variable setup for hooks. */ |
24c2a48d CB |
776 | ret = setenv("LXC_NAME", name, 1); |
777 | if (ret < 0) | |
778 | SYSERROR("Failed to set environment variable: LXC_NAME=%s", name); | |
408da065 | 779 | |
2170b263 CB |
780 | if (conf->rcfile) { |
781 | ret = setenv("LXC_CONFIG_FILE", conf->rcfile, 1); | |
782 | if (ret < 0) | |
818a57fc | 783 | SYSERROR("Failed to set environment variable: LXC_CONFIG_FILE=%s", conf->rcfile); |
2170b263 | 784 | } |
408da065 | 785 | |
2170b263 CB |
786 | if (conf->rootfs.mount) { |
787 | ret = setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1); | |
788 | if (ret < 0) | |
818a57fc | 789 | SYSERROR("Failed to set environment variable: LXC_ROOTFS_MOUNT=%s", conf->rootfs.mount); |
2170b263 | 790 | } |
408da065 | 791 | |
2170b263 CB |
792 | if (conf->rootfs.path) { |
793 | ret = setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1); | |
794 | if (ret < 0) | |
818a57fc | 795 | SYSERROR("Failed to set environment variable: LXC_ROOTFS_PATH=%s", conf->rootfs.path); |
2170b263 | 796 | } |
408da065 | 797 | |
2170b263 CB |
798 | if (conf->console.path) { |
799 | ret = setenv("LXC_CONSOLE", conf->console.path, 1); | |
800 | if (ret < 0) | |
818a57fc | 801 | SYSERROR("Failed to set environment variable: LXC_CONSOLE=%s", conf->console.path); |
2170b263 | 802 | } |
408da065 | 803 | |
2170b263 CB |
804 | if (conf->console.log_path) { |
805 | ret = setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1); | |
806 | if (ret < 0) | |
818a57fc | 807 | SYSERROR("Failed to set environment variable: LXC_CONSOLE_LOGPATH=%s", conf->console.log_path); |
2170b263 | 808 | } |
408da065 | 809 | |
2170b263 CB |
810 | if (cgns_supported()) { |
811 | ret = setenv("LXC_CGNS_AWARE", "1", 1); | |
812 | if (ret < 0) | |
818a57fc | 813 | SYSERROR("Failed to set environment variable LXC_CGNS_AWARE=1"); |
2170b263 | 814 | } |
b8f88d9b | 815 | |
4a03ded4 | 816 | loglevel = lxc_log_priority_to_string(lxc_log_get_level()); |
2170b263 CB |
817 | ret = setenv("LXC_LOG_LEVEL", loglevel, 1); |
818 | if (ret < 0) | |
818a57fc | 819 | SYSERROR("Set environment variable LXC_LOG_LEVEL=%s", loglevel); |
a2c09be0 CB |
820 | |
821 | if (conf->hooks_version == 0) | |
822 | ret = setenv("LXC_HOOK_VERSION", "0", 1); | |
823 | else | |
824 | ret = setenv("LXC_HOOK_VERSION", "1", 1); | |
825 | if (ret < 0) | |
826 | SYSERROR("Failed to set environment variable LXC_HOOK_VERSION=%u", conf->hooks_version); | |
408da065 | 827 | /* End of environment variable setup for hooks. */ |
f7bee6c6 | 828 | |
2170b263 | 829 | TRACE("Set environment variables"); |
dbc9832d | 830 | |
2170b263 CB |
831 | ret = run_lxc_hooks(name, "pre-start", conf, NULL); |
832 | if (ret < 0) { | |
833 | ERROR("Failed to run lxc.hook.pre-start for container \"%s\"", name); | |
773fb9ca SH |
834 | goto out_aborting; |
835 | } | |
2170b263 | 836 | TRACE("Ran pre-start hooks"); |
26ddeedd | 837 | |
408da065 CB |
838 | /* The signal fd has to be created before forking otherwise if the child |
839 | * process exits before we setup the signal fd, the event will be lost | |
840 | * and the command will be stuck. | |
841 | */ | |
83ee7875 | 842 | handler->sigfd = setup_signal_fd(&handler->oldmask); |
59eb99ba | 843 | if (handler->sigfd < 0) { |
408da065 | 844 | ERROR("Failed to setup SIGCHLD fd handler."); |
b5159817 DE |
845 | goto out_delete_tty; |
846 | } | |
2170b263 | 847 | TRACE("Set up signal fd"); |
b5159817 | 848 | |
408da065 | 849 | /* Do this after setting up signals since it might unblock SIGWINCH. */ |
564e31c4 | 850 | ret = lxc_terminal_setup(conf); |
4d1ffb0a CB |
851 | if (ret < 0) { |
852 | ERROR("Failed to create console"); | |
b5159817 | 853 | goto out_restore_sigmask; |
b0a33c1e | 854 | } |
4d1ffb0a | 855 | TRACE("Created console"); |
b0a33c1e | 856 | |
2f835b4b | 857 | ret = lxc_terminal_map_ids(conf, &conf->console); |
7cfeddd7 | 858 | if (ret < 0) { |
2170b263 | 859 | ERROR("Failed to chown console"); |
5b74eb3c | 860 | goto out_delete_terminal; |
c4d10a05 | 861 | } |
2170b263 | 862 | TRACE("Chowned console"); |
c4d10a05 | 863 | |
5a087e05 | 864 | handler->cgroup_ops = cgroup_init(handler->conf); |
2202afc9 CB |
865 | if (!handler->cgroup_ops) { |
866 | ERROR("Failed to initialize cgroup driver"); | |
5b74eb3c | 867 | goto out_delete_terminal; |
2202afc9 CB |
868 | } |
869 | TRACE("Initialized cgroup driver"); | |
870 | ||
1800f924 WB |
871 | ret = lsm_process_prepare(conf, handler->lxcpath); |
872 | if (ret < 0) { | |
873 | ERROR("Failed to initialize LSM"); | |
c581d2a6 | 874 | goto out_delete_terminal; |
1800f924 WB |
875 | } |
876 | TRACE("Initialized LSM"); | |
877 | ||
2170b263 | 878 | INFO("Container \"%s\" is initialized", name); |
4d8bdfa0 | 879 | handler->monitor_status_fd = move_fd(status_fd); |
aa460476 | 880 | return 0; |
59eb99ba | 881 | |
5b74eb3c CB |
882 | out_delete_terminal: |
883 | lxc_terminal_delete(&handler->conf->console); | |
ea918412 | 884 | |
b5159817 | 885 | out_restore_sigmask: |
4c496daa | 886 | (void)pthread_sigmask(SIG_SETMASK, &handler->oldmask, NULL); |
ea918412 | 887 | |
59eb99ba | 888 | out_delete_tty: |
0e4be3cf | 889 | lxc_delete_tty(&conf->ttys); |
ea918412 | 890 | |
59eb99ba | 891 | out_aborting: |
4c496daa | 892 | (void)lxc_set_state(name, handler, ABORTING); |
ea918412 | 893 | |
d2e30e99 | 894 | out_close_maincmd_fd: |
9044b79e | 895 | lxc_abstract_unix_close(conf->maincmd_fd); |
d2e30e99 | 896 | conf->maincmd_fd = -1; |
aa460476 | 897 | return -1; |
59eb99ba DL |
898 | } |
899 | ||
3b72c4a0 | 900 | void lxc_fini(const char *name, struct lxc_handler *handler) |
59eb99ba | 901 | { |
818a57fc | 902 | int ret; |
18b3b9c1 | 903 | pid_t self; |
dbc9832d | 904 | struct lxc_list *cur, *next; |
dbc9832d | 905 | char *namespaces[LXC_NS_MAX + 1]; |
b3286b62 | 906 | size_t namespace_count = 0; |
2202afc9 | 907 | struct cgroup_ops *cgroup_ops = handler->cgroup_ops; |
b6b2b194 | 908 | |
408da065 CB |
909 | /* The STOPPING state is there for future cleanup code which can take |
910 | * awhile. | |
59eb99ba | 911 | */ |
25c2aca5 | 912 | lxc_set_state(name, handler, STOPPING); |
b6b2b194 | 913 | |
0059379f | 914 | self = lxc_raw_getpid(); |
818a57fc | 915 | for (int i = 0; i < LXC_NS_MAX; i++) { |
18b3b9c1 CB |
916 | if (handler->nsfd[i] < 0) |
917 | continue; | |
918 | ||
919 | if (handler->conf->hooks_version == 0) | |
c097467f | 920 | ret = asprintf(&namespaces[namespace_count], |
18b3b9c1 CB |
921 | "%s:/proc/%d/fd/%d", ns_info[i].proc_name, |
922 | self, handler->nsfd[i]); | |
923 | else | |
c097467f | 924 | ret = asprintf(&namespaces[namespace_count], |
18b3b9c1 | 925 | "/proc/%d/fd/%d", self, handler->nsfd[i]); |
818a57fc | 926 | if (ret < 0) { |
c097467f | 927 | SYSERROR("Failed to allocate memory"); |
18b3b9c1 | 928 | break; |
b3286b62 | 929 | } |
18b3b9c1 CB |
930 | |
931 | if (handler->conf->hooks_version == 0) { | |
932 | namespace_count++; | |
933 | continue; | |
934 | } | |
935 | ||
c097467f CB |
936 | ret = setenv(ns_info[i].env_name, namespaces[namespace_count], 1); |
937 | if (ret < 0) | |
18b3b9c1 CB |
938 | SYSERROR("Failed to set environment variable %s=%s", |
939 | ns_info[i].env_name, namespaces[namespace_count]); | |
940 | else | |
941 | TRACE("Set environment variable %s=%s", | |
942 | ns_info[i].env_name, namespaces[namespace_count]); | |
943 | ||
944 | namespace_count++; | |
b3286b62 WB |
945 | } |
946 | namespaces[namespace_count] = NULL; | |
c154af98 | 947 | |
80308d07 | 948 | if (handler->conf->reboot > REBOOT_NONE) { |
c097467f CB |
949 | ret = setenv("LXC_TARGET", "reboot", 1); |
950 | if (ret < 0) | |
818a57fc | 951 | SYSERROR("Failed to set environment variable: LXC_TARGET=reboot"); |
c097467f | 952 | } |
408da065 | 953 | |
80308d07 | 954 | if (handler->conf->reboot == REBOOT_NONE) { |
c097467f CB |
955 | ret = setenv("LXC_TARGET", "stop", 1); |
956 | if (ret < 0) | |
818a57fc | 957 | SYSERROR("Failed to set environment variable: LXC_TARGET=stop"); |
c097467f | 958 | } |
c154af98 | 959 | |
18b3b9c1 | 960 | if (handler->conf->hooks_version == 0) |
c097467f | 961 | ret = run_lxc_hooks(name, "stop", handler->conf, namespaces); |
18b3b9c1 | 962 | else |
c097467f CB |
963 | ret = run_lxc_hooks(name, "stop", handler->conf, NULL); |
964 | if (ret < 0) | |
965 | ERROR("Failed to run \"lxc.hook.stop\" hook"); | |
c154af98 | 966 | |
b3286b62 WB |
967 | while (namespace_count--) |
968 | free(namespaces[namespace_count]); | |
28d9e29e | 969 | |
1800f924 WB |
970 | lsm_process_cleanup(handler->conf, handler->lxcpath); |
971 | ||
c581d2a6 CB |
972 | if (cgroup_ops) { |
973 | cgroup_ops->payload_destroy(cgroup_ops, handler); | |
974 | cgroup_ops->monitor_destroy(cgroup_ops, handler); | |
975 | } | |
4288b79f | 976 | |
80308d07 | 977 | if (handler->conf->reboot == REBOOT_NONE) { |
f893d898 CB |
978 | /* For all new state clients simply close the command socket. |
979 | * This will inform all state clients that the container is | |
980 | * STOPPED and also prevents a race between a open()/close() on | |
981 | * the command socket causing a new process to get ECONNREFUSED | |
982 | * because we haven't yet closed the command socket. | |
983 | */ | |
9044b79e | 984 | lxc_abstract_unix_close(handler->conf->maincmd_fd); |
818a57fc | 985 | handler->conf->maincmd_fd = -EBADF; |
c3184275 CB |
986 | TRACE("Closed command socket"); |
987 | ||
988 | /* This function will try to connect to the legacy lxc-monitord | |
989 | * state server and only exists for backwards compatibility. | |
990 | */ | |
991 | lxc_monitor_send_state(name, STOPPED, handler->lxcpath); | |
992 | ||
993 | /* The command socket is closed so no one can acces the command | |
994 | * socket anymore so there's no need to lock it. | |
995 | */ | |
996 | handler->state = STOPPED; | |
997 | TRACE("Set container state to \"STOPPED\""); | |
f893d898 CB |
998 | } else { |
999 | lxc_set_state(name, handler, STOPPED); | |
d39b10eb | 1000 | } |
4288b79f | 1001 | |
048493a3 CB |
1002 | /* Avoid lingering namespace references. */ |
1003 | lxc_put_nsfds(handler); | |
1004 | ||
46800e77 CB |
1005 | ret = run_lxc_hooks(name, "post-stop", handler->conf, NULL); |
1006 | if (ret < 0) { | |
c097467f | 1007 | ERROR("Failed to run lxc.hook.post-stop for container \"%s\"", name); |
80308d07 | 1008 | if (handler->conf->reboot > REBOOT_NONE) { |
c097467f | 1009 | WARN("Container will be stopped instead of rebooted"); |
80308d07 | 1010 | handler->conf->reboot = REBOOT_NONE; |
c097467f CB |
1011 | |
1012 | ret = setenv("LXC_TARGET", "stop", 1); | |
1013 | if (ret < 0) | |
818a57fc | 1014 | WARN("Failed to set environment variable: LXC_TARGET=stop"); |
f3787121 CB |
1015 | } |
1016 | } | |
26ddeedd | 1017 | |
408da065 | 1018 | /* Reset mask set by setup_signal_fd. */ |
b467714b | 1019 | ret = pthread_sigmask(SIG_SETMASK, &handler->oldmask, NULL); |
c097467f | 1020 | if (ret < 0) |
a24c5678 | 1021 | SYSWARN("Failed to restore signal mask"); |
8f64a3f6 | 1022 | |
2aac071b | 1023 | lxc_terminal_delete(&handler->conf->console); |
0e4be3cf | 1024 | lxc_delete_tty(&handler->conf->ttys); |
dbc9832d | 1025 | |
dbc9832d CB |
1026 | /* The command socket is now closed, no more state clients can register |
1027 | * themselves from now on. So free the list of state clients. | |
1028 | */ | |
d39b10eb CB |
1029 | lxc_list_for_each_safe(cur, &handler->conf->state_clients, next) { |
1030 | struct lxc_state_client *client = cur->elem; | |
1031 | ||
1032 | /* Keep state clients that want to be notified about reboots. */ | |
80308d07 CB |
1033 | if ((handler->conf->reboot > REBOOT_NONE) && |
1034 | (client->states[RUNNING] == 2)) | |
d39b10eb CB |
1035 | continue; |
1036 | ||
dbc9832d | 1037 | /* close state client socket */ |
dbc9832d | 1038 | lxc_list_del(cur); |
c097467f | 1039 | close(client->clientfd); |
dbc9832d CB |
1040 | free(cur->elem); |
1041 | free(cur); | |
1042 | } | |
1043 | ||
80308d07 | 1044 | if (handler->conf->ephemeral == 1 && handler->conf->reboot != REBOOT_REQ) |
28272964 | 1045 | lxc_destroy_container_on_signal(handler, name); |
2c5f2ede | 1046 | |
1cc8bd4b | 1047 | lxc_free_handler(handler); |
59eb99ba DL |
1048 | } |
1049 | ||
735f2c6e | 1050 | void lxc_abort(const char *name, struct lxc_handler *handler) |
59eb99ba | 1051 | { |
33942046 CB |
1052 | int ret = 0; |
1053 | int status; | |
73e608b2 | 1054 | |
25c2aca5 | 1055 | lxc_set_state(name, handler, ABORTING); |
0e4f9d51 | 1056 | |
c718fac1 | 1057 | if (handler->pidfd >= 0) { |
33942046 | 1058 | ret = lxc_raw_pidfd_send_signal(handler->pidfd, SIGKILL, NULL, 0); |
c718fac1 | 1059 | if (ret) |
11c7d349 CB |
1060 | SYSWARN("Failed to send SIGKILL via pidfd %d for process %d", |
1061 | handler->pidfd, handler->pid); | |
c718fac1 CB |
1062 | } |
1063 | ||
11c7d349 CB |
1064 | if (!ret || errno != ESRCH) |
1065 | if (kill(handler->pid, SIGKILL)) | |
1066 | SYSWARN("Failed to send SIGKILL to %d", handler->pid); | |
0e4f9d51 | 1067 | |
33942046 CB |
1068 | do { |
1069 | ret = waitpid(-1, &status, 0); | |
1070 | } while (ret > 0); | |
59eb99ba DL |
1071 | } |
1072 | ||
ffe1e01a | 1073 | static int do_start(void *data) |
50e98013 | 1074 | { |
cdb2a47f | 1075 | struct lxc_handler *handler = data; |
53675a8d | 1076 | __lxc_unused __do_close_prot_errno int data_sock0 = handler->data_sock[0], |
9ff57a59 | 1077 | data_sock1 = handler->data_sock[1]; |
4d8bdfa0 | 1078 | __do_close_prot_errno int status_fd = -EBADF; |
8bf3abfb | 1079 | int ret; |
928b1f04 YT |
1080 | uid_t new_uid; |
1081 | gid_t new_gid; | |
e96a536c | 1082 | struct lxc_list *iterator; |
964581c2 CB |
1083 | uid_t nsuid = 0; |
1084 | gid_t nsgid = 0; | |
5af9369b | 1085 | int devnull_fd = -1; |
50e98013 | 1086 | |
18225d19 | 1087 | lxc_sync_fini_parent(handler); |
50e98013 | 1088 | |
9ff57a59 | 1089 | if (lxc_abstract_unix_recv_fds(data_sock1, &status_fd, 1, NULL, 0) < 0) { |
4d8bdfa0 CB |
1090 | ERROR("Failed to receive status file descriptor to child process"); |
1091 | goto out_warn_father; | |
1092 | } | |
1093 | ||
408da065 CB |
1094 | /* This prctl must be before the synchro, so if the parent dies before |
1095 | * we set the parent death signal, we will detect its death with the | |
1096 | * synchro right after, otherwise we have a window where the parent can | |
1097 | * exit before we set the pdeath signal leading to a unsupervized | |
1098 | * container. | |
743ecd2e | 1099 | */ |
4d8bdfa0 | 1100 | ret = lxc_set_death_signal(SIGKILL, handler->monitor_pid, status_fd); |
912314fc CB |
1101 | if (ret < 0) { |
1102 | SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL"); | |
18225d19 | 1103 | goto out_warn_father; |
743ecd2e DL |
1104 | } |
1105 | ||
611ddd34 CB |
1106 | ret = lxc_ambient_caps_up(); |
1107 | if (ret < 0) { | |
6f5e532f | 1108 | ERROR("Failed to raise ambient capabilities"); |
611ddd34 CB |
1109 | goto out_warn_father; |
1110 | } | |
1111 | ||
b467714b | 1112 | ret = pthread_sigmask(SIG_SETMASK, &handler->oldmask, NULL); |
18225d19 CB |
1113 | if (ret < 0) { |
1114 | SYSERROR("Failed to set signal mask"); | |
1115 | goto out_warn_father; | |
1116 | } | |
50e98013 | 1117 | |
408da065 | 1118 | /* Don't leak the pinfd to the container. */ |
d39b10eb | 1119 | if (handler->pinfd >= 0) |
0d03360a | 1120 | close(handler->pinfd); |
2b0e17e4 | 1121 | |
18225d19 CB |
1122 | ret = lxc_sync_wait_parent(handler, LXC_SYNC_STARTUP); |
1123 | if (ret < 0) | |
1124 | goto out_warn_father; | |
5b1e83cb | 1125 | |
408da065 CB |
1126 | /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See |
1127 | * https://github.com/lxc/lxd/issues/1978. | |
1128 | */ | |
becad0ec | 1129 | if ((handler->ns_clone_flags & (CLONE_NEWNET | CLONE_NEWUSER)) == |
408da065 | 1130 | (CLONE_NEWNET | CLONE_NEWUSER)) { |
5b1e83cb SH |
1131 | ret = unshare(CLONE_NEWNET); |
1132 | if (ret < 0) { | |
e96a536c | 1133 | SYSERROR("Failed to unshare CLONE_NEWNET"); |
5b1e83cb SH |
1134 | goto out_warn_father; |
1135 | } | |
e96a536c | 1136 | INFO("Unshared CLONE_NEWNET"); |
5b1e83cb SH |
1137 | } |
1138 | ||
408da065 CB |
1139 | /* Tell the parent task it can begin to configure the container and wait |
1140 | * for it to finish. | |
3c22086f | 1141 | */ |
e96a536c CB |
1142 | ret = lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE); |
1143 | if (ret < 0) | |
611ddd34 | 1144 | goto out_error; |
50e98013 | 1145 | |
e389f2af | 1146 | if (handler->ns_clone_flags & CLONE_NEWNET) { |
3c09b97c | 1147 | ret = lxc_network_recv_from_parent(handler); |
e389f2af CB |
1148 | if (ret < 0) { |
1149 | ERROR("Failed to receive veth names from parent"); | |
1150 | goto out_warn_father; | |
1151 | } | |
7ab1ba02 | 1152 | } |
658979c5 | 1153 | |
408da065 | 1154 | /* If we are in a new user namespace, become root there to have |
d08f8d2f | 1155 | * privilege over our namespace. |
f6d3e3e4 SH |
1156 | */ |
1157 | if (!lxc_list_empty(&handler->conf->id_map)) { | |
964581c2 CB |
1158 | if (!handler->conf->root_nsuid_map) |
1159 | nsuid = handler->conf->init_uid; | |
1160 | ||
1161 | if (!handler->conf->root_nsgid_map) | |
1162 | nsgid = handler->conf->init_gid; | |
4160c3a0 | 1163 | |
d08f8d2f CB |
1164 | /* Drop groups only after we switched to a valid gid in the new |
1165 | * user namespace. | |
1166 | */ | |
8af07f82 CB |
1167 | if (!lxc_setgroups(0, NULL) && |
1168 | (handler->am_root || errno != EPERM)) | |
c476bdce | 1169 | goto out_warn_father; |
4b826b1f | 1170 | |
b58214ac CB |
1171 | if (!lxc_switch_uid_gid(nsuid, nsgid)) |
1172 | goto out_warn_father; | |
1173 | ||
b81689a1 CB |
1174 | ret = prctl(PR_SET_DUMPABLE, prctl_arg(1), prctl_arg(0), |
1175 | prctl_arg(0), prctl_arg(0)); | |
d7883725 CB |
1176 | if (ret < 0) |
1177 | goto out_warn_father; | |
912314fc CB |
1178 | |
1179 | /* set{g,u}id() clears deathsignal */ | |
4d8bdfa0 | 1180 | ret = lxc_set_death_signal(SIGKILL, handler->monitor_pid, status_fd); |
912314fc CB |
1181 | if (ret < 0) { |
1182 | SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL"); | |
1183 | goto out_warn_father; | |
1184 | } | |
f6d3e3e4 SH |
1185 | } |
1186 | ||
e96a536c CB |
1187 | ret = access(handler->lxcpath, X_OK); |
1188 | if (ret != 0) { | |
c8154066 SH |
1189 | print_top_failing_dir(handler->lxcpath); |
1190 | goto out_warn_father; | |
1191 | } | |
1192 | ||
7a55c157 TA |
1193 | /* In order to checkpoint restore, we need to have everything in the |
1194 | * same mount namespace. However, some containers may not have a | |
1195 | * reasonable /dev (in particular, they may not have /dev/null), so we | |
1196 | * can't set init's std fds to /dev/null by opening it from inside the | |
1197 | * container. | |
1198 | * | |
1199 | * If that's the case, fall back to using the host's /dev/null. This | |
1200 | * means that migration won't work, but at least we won't spew output | |
1201 | * where it isn't wanted. | |
1202 | */ | |
bb955810 | 1203 | if (handler->daemonize && !handler->conf->autodev) { |
2806a87d | 1204 | char path[PATH_MAX]; |
3c09b97c | 1205 | |
2806a87d RK |
1206 | ret = snprintf(path, sizeof(path), "%s/dev/null", |
1207 | handler->conf->rootfs.mount); | |
1208 | if (ret < 0 || ret >= sizeof(path)) | |
1209 | goto out_warn_father; | |
3c09b97c | 1210 | |
e96a536c CB |
1211 | ret = access(path, F_OK); |
1212 | if (ret != 0) { | |
1213 | devnull_fd = open_devnull(); | |
c44de748 | 1214 | |
e96a536c CB |
1215 | if (devnull_fd < 0) |
1216 | goto out_warn_father; | |
818a57fc | 1217 | WARN("Using /dev/null from the host for container init's standard file descriptors. Migration will not work"); |
e96a536c | 1218 | } |
c44de748 AM |
1219 | } |
1220 | ||
408da065 | 1221 | /* Ask father to setup cgroups and wait for him to finish. */ |
e96a536c CB |
1222 | ret = lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP); |
1223 | if (ret < 0) | |
c44de748 | 1224 | goto out_error; |
544a48a0 | 1225 | |
deefdf8a CB |
1226 | /* Unshare cgroup namespace after we have setup our cgroups. If we do it |
1227 | * earlier we end up with a wrong view of /proc/self/cgroup. For | |
1228 | * example, assume we unshare(CLONE_NEWCGROUP) first, and then create | |
1229 | * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then | |
1230 | * /proc/self/cgroup would show us: | |
1231 | * | |
1232 | * 8:cpuset:/lxc/c | |
1233 | * | |
1234 | * whereas it should actually show | |
1235 | * | |
1236 | * 8:cpuset:/ | |
1237 | */ | |
becad0ec | 1238 | if (handler->ns_clone_flags & CLONE_NEWCGROUP) { |
e96a536c CB |
1239 | ret = unshare(CLONE_NEWCGROUP); |
1240 | if (ret < 0) { | |
bca7c59c CB |
1241 | if (errno != EINVAL) { |
1242 | SYSERROR("Failed to unshare CLONE_NEWCGROUP"); | |
1243 | goto out_warn_father; | |
1244 | } | |
1245 | ||
1246 | handler->ns_clone_flags &= ~CLONE_NEWCGROUP; | |
1247 | SYSINFO("Kernel does not support CLONE_NEWCGROUP"); | |
1248 | } else { | |
1249 | INFO("Unshared CLONE_NEWCGROUP"); | |
deefdf8a | 1250 | } |
deefdf8a CB |
1251 | } |
1252 | ||
149857af CB |
1253 | /* Add the requested environment variables to the current environment to |
1254 | * allow them to be used by the various hooks, such as the start hook | |
317494f1 | 1255 | * below. |
98ff08ca | 1256 | */ |
98ff08ca | 1257 | lxc_list_for_each(iterator, &handler->conf->environment) { |
e96a536c CB |
1258 | ret = putenv((char *)iterator->elem); |
1259 | if (ret < 0) { | |
1260 | SYSERROR("Failed to set environment variable: %s", | |
1261 | (char *)iterator->elem); | |
98ff08ca CB |
1262 | goto out_warn_father; |
1263 | } | |
1264 | } | |
1265 | ||
f4152036 | 1266 | /* Setup the container, ip, names, utsname, ... */ |
3b988b33 | 1267 | ret = lxc_setup(handler); |
7729f8e5 | 1268 | if (ret < 0) { |
e96a536c | 1269 | ERROR("Failed to setup container \"%s\"", handler->name); |
f4152036 CB |
1270 | goto out_warn_father; |
1271 | } | |
1272 | ||
408da065 | 1273 | /* Set the label to change to when we exec(2) the container's init. */ |
1800f924 | 1274 | ret = lsm_process_label_set(NULL, handler->conf, true); |
e96a536c | 1275 | if (ret < 0) |
e075f5d9 | 1276 | goto out_warn_father; |
5112cd70 | 1277 | |
029cdff5 | 1278 | /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it |
408da065 CB |
1279 | * before we aren't allowed anymore. |
1280 | */ | |
029cdff5 | 1281 | if (handler->conf->no_new_privs) { |
b81689a1 CB |
1282 | ret = prctl(PR_SET_NO_NEW_PRIVS, prctl_arg(1), prctl_arg(0), |
1283 | prctl_arg(0), prctl_arg(0)); | |
e96a536c | 1284 | if (ret < 0) { |
818a57fc | 1285 | SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges"); |
029cdff5 CB |
1286 | goto out_warn_father; |
1287 | } | |
818a57fc | 1288 | DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges"); |
029cdff5 CB |
1289 | } |
1290 | ||
0d9acb99 DE |
1291 | /* Some init's such as busybox will set sane tty settings on stdin, |
1292 | * stdout, stderr which it thinks is the console. We already set them | |
1293 | * the way we wanted on the real terminal, and we want init to do its | |
564e31c4 | 1294 | * setup on its console ie. the pty allocated in lxc_terminal_setup() so |
408da065 | 1295 | * make sure that that pty is stdin,stdout,stderr. |
0d9acb99 | 1296 | */ |
5d113f65 | 1297 | if (handler->conf->console.slave >= 0) { |
bb955810 | 1298 | if (handler->daemonize || !handler->conf->is_execute) |
5d113f65 CB |
1299 | ret = set_stdfds(handler->conf->console.slave); |
1300 | else | |
ae6d3913 | 1301 | ret = lxc_terminal_set_stdfds(handler->conf->console.slave); |
5d113f65 | 1302 | if (ret < 0) { |
818a57fc CB |
1303 | ERROR("Failed to redirect std{in,out,err} to pty file descriptor %d", |
1304 | handler->conf->console.slave); | |
c5b93afb LF |
1305 | goto out_warn_father; |
1306 | } | |
5d113f65 | 1307 | } |
0d9acb99 | 1308 | |
408da065 | 1309 | /* If we mounted a temporary proc, then unmount it now. */ |
5112cd70 | 1310 | tmp_proc_unmount(handler->conf); |
e075f5d9 | 1311 | |
e96a536c CB |
1312 | ret = lxc_seccomp_load(handler->conf); |
1313 | if (ret < 0) | |
8f2c3a70 SH |
1314 | goto out_warn_father; |
1315 | ||
c3e3c21a CB |
1316 | ret = lxc_seccomp_send_notifier_fd(&handler->conf->seccomp, data_sock0); |
1317 | if (ret < 0) { | |
1318 | SYSERROR("Failed to send seccomp notify fd to parent"); | |
1319 | goto out_warn_father; | |
cdb2a47f | 1320 | } |
cdb2a47f | 1321 | |
e96a536c CB |
1322 | ret = run_lxc_hooks(handler->name, "start", handler->conf, NULL); |
1323 | if (ret < 0) { | |
1324 | ERROR("Failed to run lxc.hook.start for container \"%s\"", | |
1325 | handler->name); | |
773fb9ca SH |
1326 | goto out_warn_father; |
1327 | } | |
fc25b815 | 1328 | |
773fb9ca | 1329 | close(handler->sigfd); |
26ddeedd | 1330 | |
bb955810 | 1331 | if (handler->conf->console.slave < 0 && handler->daemonize) { |
f4c177c3 CB |
1332 | if (devnull_fd < 0) { |
1333 | devnull_fd = open_devnull(); | |
1334 | if (devnull_fd < 0) | |
1335 | goto out_warn_father; | |
1336 | } | |
1337 | ||
e96a536c CB |
1338 | ret = set_stdfds(devnull_fd); |
1339 | if (ret < 0) { | |
1340 | ERROR("Failed to redirect std{in,out,err} to \"/dev/null\""); | |
c5b93afb LF |
1341 | goto out_warn_father; |
1342 | } | |
e96a536c | 1343 | } |
507cee36 | 1344 | |
c44de748 AM |
1345 | if (devnull_fd >= 0) { |
1346 | close(devnull_fd); | |
1347 | devnull_fd = -1; | |
1348 | } | |
1349 | ||
8c9a7665 TA |
1350 | setsid(); |
1351 | ||
e96a536c CB |
1352 | if (handler->conf->init_cwd) { |
1353 | ret = chdir(handler->conf->init_cwd); | |
1354 | if (ret < 0) { | |
1355 | SYSERROR("Could not change directory to \"%s\"", | |
1356 | handler->conf->init_cwd); | |
1357 | goto out_warn_father; | |
1358 | } | |
3c491553 L |
1359 | } |
1360 | ||
e96a536c CB |
1361 | ret = lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP_LIMITS); |
1362 | if (ret < 0) | |
f4152036 CB |
1363 | goto out_warn_father; |
1364 | ||
149857af CB |
1365 | /* Reset the environment variables the user requested in a clear |
1366 | * environment. | |
1367 | */ | |
e96a536c CB |
1368 | ret = clearenv(); |
1369 | /* Don't error out though. */ | |
1370 | if (ret < 0) | |
149857af | 1371 | SYSERROR("Failed to clear environment."); |
149857af CB |
1372 | |
1373 | lxc_list_for_each(iterator, &handler->conf->environment) { | |
e96a536c CB |
1374 | ret = putenv((char *)iterator->elem); |
1375 | if (ret < 0) { | |
1376 | SYSERROR("Failed to set environment variable: %s", | |
1377 | (char *)iterator->elem); | |
149857af CB |
1378 | goto out_warn_father; |
1379 | } | |
1380 | } | |
1381 | ||
e96a536c CB |
1382 | ret = putenv("container=lxc"); |
1383 | if (ret < 0) { | |
1384 | SYSERROR("Failed to set environment variable: container=lxc"); | |
98ff08ca CB |
1385 | goto out_warn_father; |
1386 | } | |
1387 | ||
885766f5 CB |
1388 | if (handler->conf->ttys.tty_names) { |
1389 | ret = putenv(handler->conf->ttys.tty_names); | |
e96a536c CB |
1390 | if (ret < 0) { |
1391 | SYSERROR("Failed to set environment variable for container ptys"); | |
98ff08ca CB |
1392 | goto out_warn_father; |
1393 | } | |
1394 | } | |
1395 | ||
76bdf299 CB |
1396 | /* The container has been setup. We can now switch to an unprivileged |
1397 | * uid/gid. | |
1398 | */ | |
928b1f04 YT |
1399 | new_uid = handler->conf->init_uid; |
1400 | new_gid = handler->conf->init_gid; | |
76bdf299 | 1401 | |
964581c2 CB |
1402 | /* Avoid unnecessary syscalls. */ |
1403 | if (new_uid == nsuid) | |
1404 | new_uid = LXC_INVALID_UID; | |
1405 | ||
1406 | if (new_gid == nsgid) | |
1407 | new_gid = LXC_INVALID_GID; | |
1408 | ||
8af07f82 CB |
1409 | /* If we are in a new user namespace we already dropped all groups when |
1410 | * we switched to root in the new user namespace further above. Only | |
1411 | * drop groups if we can, so ensure that we have necessary privilege. | |
1412 | */ | |
1413 | if (lxc_list_empty(&handler->conf->id_map)) | |
1414 | #if HAVE_LIBCAP | |
1415 | if (lxc_proc_cap_is_set(CAP_SETGID, CAP_EFFECTIVE)) | |
1416 | #endif | |
1417 | if (!lxc_setgroups(0, NULL)) | |
1418 | goto out_warn_father; | |
1419 | ||
b58214ac CB |
1420 | if (!lxc_switch_uid_gid(new_uid, new_gid)) |
1421 | goto out_warn_father; | |
1422 | ||
611ddd34 CB |
1423 | ret = lxc_ambient_caps_down(); |
1424 | if (ret < 0) { | |
6f5e532f | 1425 | ERROR("Failed to clear ambient capabilities"); |
611ddd34 CB |
1426 | goto out_warn_father; |
1427 | } | |
1428 | ||
258f8051 | 1429 | if (handler->conf->monitor_signal_pdeath != SIGKILL) { |
4d8bdfa0 CB |
1430 | ret = lxc_set_death_signal(handler->conf->monitor_signal_pdeath, |
1431 | handler->monitor_pid, status_fd); | |
258f8051 CB |
1432 | if (ret < 0) { |
1433 | SYSERROR("Failed to set PR_SET_PDEATHSIG to %d", | |
1434 | handler->conf->monitor_signal_pdeath); | |
1435 | goto out_warn_father; | |
1436 | } | |
1437 | } | |
1438 | ||
408da065 CB |
1439 | /* After this call, we are in error because this ops should not return |
1440 | * as it execs. | |
1441 | */ | |
c4ea60df | 1442 | handler->ops->start(handler, handler->data); |
50e98013 DL |
1443 | |
1444 | out_warn_father: | |
408da065 CB |
1445 | /* We want the parent to know something went wrong, so we return a |
1446 | * special error code. | |
1447 | */ | |
d1ccb562 | 1448 | lxc_sync_wake_parent(handler, LXC_SYNC_ERROR); |
c44de748 AM |
1449 | |
1450 | out_error: | |
1451 | if (devnull_fd >= 0) | |
1452 | close(devnull_fd); | |
1453 | ||
50e98013 DL |
1454 | return -1; |
1455 | } | |
1456 | ||
ae467c54 | 1457 | static int lxc_recv_ttys_from_child(struct lxc_handler *handler) |
e8bd4e43 | 1458 | { |
ae467c54 | 1459 | int i; |
2520facd | 1460 | struct lxc_terminal_info *tty; |
ae467c54 | 1461 | int ret = -1; |
c6012571 | 1462 | int sock = handler->data_sock[1]; |
e8bd4e43 | 1463 | struct lxc_conf *conf = handler->conf; |
0e4be3cf | 1464 | struct lxc_tty_info *ttys = &conf->ttys; |
e8bd4e43 | 1465 | |
885766f5 | 1466 | if (!conf->ttys.max) |
e8bd4e43 SH |
1467 | return 0; |
1468 | ||
885766f5 | 1469 | ttys->tty = malloc(sizeof(*ttys->tty) * ttys->max); |
0e4be3cf | 1470 | if (!ttys->tty) |
e8bd4e43 | 1471 | return -1; |
e8bd4e43 | 1472 | |
885766f5 | 1473 | for (i = 0; i < conf->ttys.max; i++) { |
672c1e58 CB |
1474 | int ttyfds[2]; |
1475 | ||
1476 | ret = lxc_abstract_unix_recv_fds(sock, ttyfds, 2, NULL, 0); | |
1477 | if (ret < 0) | |
1478 | break; | |
ae467c54 | 1479 | |
0e4be3cf | 1480 | tty = &ttys->tty[i]; |
dd3de568 | 1481 | tty->busy = -1; |
2520facd CB |
1482 | tty->master = ttyfds[0]; |
1483 | tty->slave = ttyfds[1]; | |
818a57fc | 1484 | TRACE("Received pty with master fd %d and slave fd %d from child", tty->master, tty->slave); |
e8bd4e43 | 1485 | } |
6d1400b5 | 1486 | |
ae467c54 | 1487 | if (ret < 0) |
6d1400b5 | 1488 | SYSERROR("Failed to receive %zu ttys from child", ttys->max); |
ae467c54 | 1489 | else |
885766f5 | 1490 | TRACE("Received %zu ttys from child", ttys->max); |
ae467c54 CB |
1491 | |
1492 | return ret; | |
e8bd4e43 SH |
1493 | } |
1494 | ||
5af9369b | 1495 | int resolve_clone_flags(struct lxc_handler *handler) |
f813849c | 1496 | { |
8bc8c715 CB |
1497 | int i; |
1498 | struct lxc_conf *conf = handler->conf; | |
f813849c | 1499 | |
8bc8c715 | 1500 | for (i = 0; i < LXC_NS_MAX; i++) { |
f7a0c6ee CB |
1501 | if (conf->ns_keep) { |
1502 | if (!(conf->ns_keep & ns_info[i].clone_flag)) | |
becad0ec | 1503 | handler->ns_clone_flags |= ns_info[i].clone_flag; |
f7a0c6ee CB |
1504 | } else if (conf->ns_clone) { |
1505 | if ((conf->ns_clone & ns_info[i].clone_flag)) | |
becad0ec | 1506 | handler->ns_clone_flags |= ns_info[i].clone_flag; |
8bc8c715 CB |
1507 | } else { |
1508 | if (i == LXC_NS_USER && lxc_list_empty(&handler->conf->id_map)) | |
1509 | continue; | |
1510 | ||
1511 | if (i == LXC_NS_NET && lxc_requests_empty_network(handler)) | |
1512 | continue; | |
1513 | ||
7bd05339 CB |
1514 | if (i == LXC_NS_CGROUP && !cgns_supported()) |
1515 | continue; | |
1516 | ||
becad0ec | 1517 | handler->ns_clone_flags |= ns_info[i].clone_flag; |
8bc8c715 | 1518 | } |
03df7ab5 | 1519 | |
8bc8c715 CB |
1520 | if (!conf->ns_share[i]) |
1521 | continue; | |
5af9369b | 1522 | |
becad0ec | 1523 | handler->ns_clone_flags &= ~ns_info[i].clone_flag; |
8bc8c715 | 1524 | TRACE("Sharing %s namespace", ns_info[i].proc_name); |
5af9369b CB |
1525 | } |
1526 | ||
1527 | return 0; | |
f813849c TA |
1528 | } |
1529 | ||
8deca6c9 CB |
1530 | /* Note that this function is used with clone(CLONE_VM). Some glibc versions |
1531 | * used to reset the pid/tid to -1 when CLONE_VM was used without CLONE_THREAD. | |
1532 | * But since the memory between parent and child is shared on CLONE_VM this | |
1533 | * would invalidate the getpid() cache that glibc used to maintain and so | |
1534 | * getpid() in the child would return the parent's pid. This is all fixed in | |
1535 | * newer glibc versions where the getpid() cache is removed and the pid/tid is | |
1536 | * not reset anymore. | |
b35ddc5b | 1537 | * However, if for whatever reason you - dear committer - somehow need to get the |
8deca6c9 | 1538 | * pid of the dummy intermediate process for do_share_ns() you need to call |
0059379f CB |
1539 | * lxc_raw_getpid(). The next lxc_raw_clone() call does not employ CLONE_VM and |
1540 | * will be fine. | |
8deca6c9 CB |
1541 | */ |
1542 | static inline int do_share_ns(void *arg) | |
fa3a5b22 | 1543 | { |
8deca6c9 | 1544 | int i, flags, ret; |
fa3a5b22 CB |
1545 | struct lxc_handler *handler = arg; |
1546 | ||
8deca6c9 CB |
1547 | for (i = 0; i < LXC_NS_MAX; i++) { |
1548 | if (handler->nsfd[i] < 0) | |
1549 | continue; | |
fa3a5b22 | 1550 | |
8deca6c9 | 1551 | ret = setns(handler->nsfd[i], 0); |
c0b48eff CB |
1552 | if (ret < 0) { |
1553 | /* | |
1554 | * Note that joining a user and/or mount namespace | |
1555 | * requires the process is not multithreaded otherwise | |
1556 | * setns() will fail here. | |
1557 | */ | |
1558 | SYSERROR("Failed to inherit %s namespace", | |
1559 | ns_info[i].proc_name); | |
8deca6c9 | 1560 | return -1; |
c0b48eff | 1561 | } |
fa3a5b22 | 1562 | |
8deca6c9 | 1563 | DEBUG("Inherited %s namespace", ns_info[i].proc_name); |
fa3a5b22 CB |
1564 | } |
1565 | ||
becad0ec | 1566 | flags = handler->ns_on_clone_flags; |
8deca6c9 | 1567 | flags |= CLONE_PARENT; |
33942046 CB |
1568 | handler->pid = lxc_raw_clone_cb(do_start, handler, CLONE_PIDFD | flags, |
1569 | &handler->pidfd); | |
8deca6c9 | 1570 | if (handler->pid < 0) |
fa3a5b22 CB |
1571 | return -1; |
1572 | ||
8deca6c9 | 1573 | return 0; |
fa3a5b22 CB |
1574 | } |
1575 | ||
480588e6 CB |
1576 | /* lxc_spawn() performs crucial setup tasks and clone()s the new process which |
1577 | * exec()s the requested container binary. | |
1578 | * Note that lxc_spawn() runs in the parent namespaces. Any operations performed | |
1579 | * right here should be double checked if they'd pose a security risk. (For | |
1580 | * example, any {u}mount() operations performed here will be reflected on the | |
1581 | * host!) | |
1582 | */ | |
74a3920a | 1583 | static int lxc_spawn(struct lxc_handler *handler) |
59eb99ba | 1584 | { |
cdb2a47f | 1585 | __do_close_prot_errno int data_sock0 = -EBADF, data_sock1 = -EBADF; |
8deca6c9 | 1586 | int i, ret; |
08dd2805 | 1587 | char pidstr[20]; |
57927bf2 CB |
1588 | bool wants_to_map_ids; |
1589 | struct lxc_list *id_map; | |
6e5fc7a5 | 1590 | const char *name = handler->name; |
28d9e29e | 1591 | const char *lxcpath = handler->lxcpath; |
2202afc9 | 1592 | bool share_ns = false; |
28d9e29e | 1593 | struct lxc_conf *conf = handler->conf; |
2202afc9 | 1594 | struct cgroup_ops *cgroup_ops = handler->cgroup_ops; |
9f30a190 | 1595 | |
28d9e29e | 1596 | id_map = &conf->id_map; |
57927bf2 | 1597 | wants_to_map_ids = !lxc_list_empty(id_map); |
2f2623ec | 1598 | |
28d9e29e | 1599 | for (i = 0; i < LXC_NS_MAX; i++) { |
b074bbf1 | 1600 | if (!conf->ns_share[i]) |
28d9e29e CB |
1601 | continue; |
1602 | ||
b074bbf1 | 1603 | handler->nsfd[i] = lxc_inherit_namespace(conf->ns_share[i], lxcpath, ns_info[i].proc_name); |
28d9e29e CB |
1604 | if (handler->nsfd[i] < 0) |
1605 | return -1; | |
fa3a5b22 | 1606 | |
8deca6c9 | 1607 | share_ns = true; |
28d9e29e | 1608 | } |
50e98013 | 1609 | |
cfc62c60 CB |
1610 | ret = lxc_sync_init(handler); |
1611 | if (ret < 0) | |
9d7f9e52 | 1612 | return -1; |
0ad19a3f | 1613 | |
b9f522c5 | 1614 | ret = socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0, |
790255cf | 1615 | handler->data_sock); |
95ef0d7c CB |
1616 | if (ret < 0) |
1617 | goto out_sync_fini; | |
cdb2a47f CB |
1618 | data_sock0 = handler->data_sock[0]; |
1619 | data_sock1 = handler->data_sock[1]; | |
e8bd4e43 | 1620 | |
5af9369b | 1621 | ret = resolve_clone_flags(handler); |
95ef0d7c CB |
1622 | if (ret < 0) |
1623 | goto out_sync_fini; | |
9f30a190 | 1624 | |
03ca4af8 TP |
1625 | if (handler->ns_clone_flags & CLONE_NEWNET) { |
1626 | ret = lxc_find_gateway_addresses(handler); | |
1627 | if (ret) { | |
1628 | ERROR("Failed to find gateway addresses"); | |
1629 | goto out_sync_fini; | |
1630 | } | |
1631 | } | |
1632 | ||
e8b181f5 | 1633 | if (!cgroup_ops->payload_create(cgroup_ops, handler)) { |
cfc62c60 | 1634 | ERROR("Failed creating cgroups"); |
47d8fb3b CS |
1635 | goto out_delete_net; |
1636 | } | |
1637 | ||
408da065 CB |
1638 | /* If the rootfs is not a blockdev, prevent the container from marking |
1639 | * it readonly. | |
1640 | * If the container is unprivileged then skip rootfs pinning. | |
0c547523 | 1641 | */ |
0ee35059 | 1642 | if (!wants_to_map_ids) { |
28d9e29e | 1643 | handler->pinfd = pin_rootfs(conf->rootfs.path); |
5e32a990 | 1644 | if (handler->pinfd == -1) |
cfc62c60 | 1645 | INFO("Failed to pin the rootfs for container \"%s\"", handler->name); |
5e32a990 | 1646 | } |
0c547523 | 1647 | |
408da065 | 1648 | /* Create a process in a new set of namespaces. */ |
becad0ec CB |
1649 | handler->ns_on_clone_flags = handler->ns_clone_flags; |
1650 | if (handler->ns_clone_flags & CLONE_NEWUSER) { | |
408da065 CB |
1651 | /* If CLONE_NEWUSER and CLONE_NEWNET was requested, we need to |
1652 | * clone a new user namespace first and only later unshare our | |
1653 | * network namespace to ensure that network devices ownership is | |
1654 | * set up correctly. | |
1655 | */ | |
becad0ec | 1656 | handler->ns_on_clone_flags &= ~CLONE_NEWNET; |
408da065 | 1657 | } |
5af9369b | 1658 | /* The cgroup namespace gets unshare()ed not clone()ed. */ |
becad0ec | 1659 | handler->ns_on_clone_flags &= ~CLONE_NEWCGROUP; |
732375f5 | 1660 | |
4e232466 CB |
1661 | if (share_ns) { |
1662 | pid_t attacher_pid; | |
1663 | ||
cfc62c60 | 1664 | attacher_pid = lxc_clone(do_share_ns, handler, |
33258b95 | 1665 | CLONE_VFORK | CLONE_VM | CLONE_FILES, NULL); |
4e232466 CB |
1666 | if (attacher_pid < 0) { |
1667 | SYSERROR(LXC_CLONE_ERROR); | |
1668 | goto out_delete_net; | |
1669 | } | |
1670 | ||
1671 | ret = wait_for_pid(attacher_pid); | |
1672 | if (ret < 0) { | |
1673 | SYSERROR("Intermediate process failed"); | |
1674 | goto out_delete_net; | |
1675 | } | |
1676 | } else { | |
cfc62c60 | 1677 | handler->pid = lxc_raw_clone_cb(do_start, handler, |
33942046 CB |
1678 | CLONE_PIDFD | handler->ns_on_clone_flags, |
1679 | &handler->pidfd); | |
4e232466 CB |
1680 | } |
1681 | if (handler->pid < 0) { | |
1682 | SYSERROR(LXC_CLONE_ERROR); | |
7fef7a06 | 1683 | goto out_delete_net; |
0ad19a3f | 1684 | } |
fa3a5b22 | 1685 | TRACE("Cloned child process %d", handler->pid); |
1bd8d726 | 1686 | |
1871e646 CB |
1687 | ret = snprintf(pidstr, 20, "%d", handler->pid); |
1688 | if (ret < 0 || ret >= 20) | |
1689 | goto out_delete_net; | |
1690 | ||
1691 | ret = setenv("LXC_PID", pidstr, 1); | |
1692 | if (ret < 0) | |
1693 | SYSERROR("Failed to set environment variable: LXC_PID=%s", pidstr); | |
1694 | ||
9662e444 | 1695 | for (i = 0; i < LXC_NS_MAX; i++) |
becad0ec | 1696 | if (handler->ns_on_clone_flags & ns_info[i].clone_flag) |
28d9e29e | 1697 | INFO("Cloned %s", ns_info[i].flag_name); |
0ad19a3f | 1698 | |
4cb53844 | 1699 | if (!lxc_try_preserve_namespaces(handler, handler->ns_on_clone_flags, handler->pid)) { |
28d9e29e CB |
1700 | ERROR("Failed to preserve cloned namespaces for lxc.hook.stop"); |
1701 | goto out_delete_net; | |
1702 | } | |
b6b2b194 | 1703 | |
3c22086f CLG |
1704 | lxc_sync_fini_child(handler); |
1705 | ||
4d8bdfa0 CB |
1706 | if (lxc_abstract_unix_send_fds(handler->data_sock[0], &handler->monitor_status_fd, 1, NULL, 0) < 0) { |
1707 | ERROR("Failed to send status file descriptor to child process"); | |
1708 | goto out_delete_net; | |
1709 | } | |
1710 | close_prot_errno_disarm(handler->monitor_status_fd); | |
1711 | ||
408da065 CB |
1712 | /* Map the container uids. The container became an invalid userid the |
1713 | * moment it was cloned with CLONE_NEWUSER. This call doesn't change | |
1714 | * anything immediately, but allows the container to setuid(0) (0 being | |
1715 | * mapped to something else on the host.) later to become a valid uid | |
1716 | * again. | |
1717 | */ | |
fa3a5b22 | 1718 | if (wants_to_map_ids) { |
df3c5314 | 1719 | if (!handler->conf->ns_share[LXC_NS_USER] && |
46186acd | 1720 | (handler->conf->ns_keep & CLONE_NEWUSER) == 0) { |
fa3a5b22 CB |
1721 | ret = lxc_map_ids(id_map, handler->pid); |
1722 | if (ret < 0) { | |
1723 | ERROR("Failed to set up id mapping."); | |
1724 | goto out_delete_net; | |
1725 | } | |
1726 | } | |
5b1e83cb SH |
1727 | } |
1728 | ||
cfc62c60 CB |
1729 | ret = lxc_sync_wake_child(handler, LXC_SYNC_STARTUP); |
1730 | if (ret < 0) | |
5b1e83cb | 1731 | goto out_delete_net; |
5b1e83cb | 1732 | |
cfc62c60 CB |
1733 | ret = lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE); |
1734 | if (ret < 0) | |
5b1e83cb | 1735 | goto out_delete_net; |
0ad19a3f | 1736 | |
c581d2a6 | 1737 | if (!cgroup_ops->setup_limits_legacy(cgroup_ops, handler->conf, false)) { |
cfc62c60 | 1738 | ERROR("Failed to setup cgroup limits for container \"%s\"", name); |
6031a6e5 DE |
1739 | goto out_delete_net; |
1740 | } | |
1741 | ||
fe70edee | 1742 | if (!cgroup_ops->payload_enter(cgroup_ops, handler)) { |
7fef7a06 | 1743 | goto out_delete_net; |
fe70edee | 1744 | } |
218d4250 | 1745 | |
c581d2a6 CB |
1746 | if (!cgroup_ops->payload_delegate_controllers(cgroup_ops)) { |
1747 | ERROR("Failed to delegate controllers to payload cgroup"); | |
1748 | goto out_delete_net; | |
1749 | } | |
1750 | ||
1751 | if (!cgroup_ops->setup_limits(cgroup_ops, handler)) { | |
1752 | ERROR("Failed to setup cgroup limits for container \"%s\"", name); | |
1753 | goto out_delete_net; | |
1754 | } | |
1755 | ||
2202afc9 | 1756 | if (!cgroup_ops->chown(cgroup_ops, handler->conf)) |
0996e18a SH |
1757 | goto out_delete_net; |
1758 | ||
aa0c0e7b RK |
1759 | /* If not done yet, we're now ready to preserve the network namespace */ |
1760 | if (handler->nsfd[LXC_NS_NET] < 0) { | |
1761 | ret = lxc_try_preserve_ns(handler->pid, "net"); | |
1762 | if (ret < 0) { | |
1763 | if (ret != -EOPNOTSUPP) { | |
1764 | SYSERROR("Failed to preserve net namespace"); | |
1765 | goto out_delete_net; | |
1766 | } | |
1767 | } else { | |
1768 | handler->nsfd[LXC_NS_NET] = ret; | |
1769 | DEBUG("Preserved net namespace via fd %d", ret); | |
4cb53844 | 1770 | } |
fa3a5b22 | 1771 | } |
aa0c0e7b RK |
1772 | ret = lxc_netns_set_nsid(handler->nsfd[LXC_NS_NET]); |
1773 | if (ret < 0) | |
1774 | SYSWARN("Failed to allocate new network namespace id"); | |
1775 | else | |
1776 | TRACE("Allocated new network namespace id"); | |
fa3a5b22 | 1777 | |
408da065 | 1778 | /* Create the network configuration. */ |
becad0ec | 1779 | if (handler->ns_clone_flags & CLONE_NEWNET) { |
e389f2af | 1780 | ret = lxc_create_network(handler); |
cfc62c60 | 1781 | if (ret < 0) { |
e389f2af | 1782 | ERROR("Failed to create the network"); |
7fef7a06 | 1783 | goto out_delete_net; |
82d5ae15 | 1784 | } |
74c6e2b0 | 1785 | |
3c09b97c | 1786 | ret = lxc_network_send_to_child(handler); |
cfc62c60 | 1787 | if (ret < 0) { |
e389f2af | 1788 | ERROR("Failed to send veth names to child"); |
74c6e2b0 CB |
1789 | goto out_delete_net; |
1790 | } | |
0ad19a3f | 1791 | } |
1792 | ||
61d7a733 YT |
1793 | if (!lxc_list_empty(&conf->procs)) { |
1794 | ret = setup_proc_filesystem(&conf->procs, handler->pid); | |
1795 | if (ret < 0) | |
1796 | goto out_delete_net; | |
1797 | } | |
1798 | ||
408da065 CB |
1799 | /* Tell the child to continue its initialization. We'll get |
1800 | * LXC_SYNC_CGROUP when it is ready for us to setup cgroups. | |
3c22086f | 1801 | */ |
cfc62c60 CB |
1802 | ret = lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE); |
1803 | if (ret < 0) | |
544a48a0 SH |
1804 | goto out_delete_net; |
1805 | ||
cfc62c60 CB |
1806 | if (!lxc_list_empty(&conf->limits)) { |
1807 | ret = setup_resource_limits(&conf->limits, handler->pid); | |
1808 | if (ret < 0) { | |
1809 | ERROR("Failed to setup resource limits"); | |
1810 | goto out_delete_net; | |
1811 | } | |
c6d09e15 WB |
1812 | } |
1813 | ||
cfc62c60 CB |
1814 | ret = lxc_sync_barrier_child(handler, LXC_SYNC_CGROUP_UNSHARE); |
1815 | if (ret < 0) | |
f4152036 CB |
1816 | goto out_delete_net; |
1817 | ||
c581d2a6 | 1818 | if (!cgroup_ops->setup_limits_legacy(cgroup_ops, handler->conf, true)) { |
cfc62c60 | 1819 | ERROR("Failed to setup legacy device cgroup controller limits"); |
b98f7d6e | 1820 | goto out_delete_net; |
544a48a0 | 1821 | } |
cfc62c60 | 1822 | TRACE("Set up legacy device cgroup controller limits"); |
544a48a0 | 1823 | |
bf651989 CB |
1824 | if (!cgroup_ops->devices_activate(cgroup_ops, handler)) { |
1825 | ERROR("Failed to setup cgroup2 device controller limits"); | |
1826 | goto out_delete_net; | |
1827 | } | |
1828 | TRACE("Set up cgroup2 device controller limits"); | |
1829 | ||
becad0ec | 1830 | if (handler->ns_clone_flags & CLONE_NEWCGROUP) { |
8bf3abfb | 1831 | /* Now we're ready to preserve the cgroup namespace */ |
4cb53844 | 1832 | ret = lxc_try_preserve_ns(handler->pid, "cgroup"); |
8bf3abfb | 1833 | if (ret < 0) { |
4cb53844 | 1834 | if (ret != -EOPNOTSUPP) { |
6d1400b5 | 1835 | SYSERROR("Failed to preserve cgroup namespace"); |
4cb53844 CB |
1836 | goto out_delete_net; |
1837 | } | |
1838 | } else { | |
1839 | handler->nsfd[LXC_NS_CGROUP] = ret; | |
1840 | DEBUG("Preserved cgroup namespace via fd %d", ret); | |
8bf3abfb | 1841 | } |
8bf3abfb CB |
1842 | } |
1843 | ||
78eb6aa6 CB |
1844 | cgroup_ops->payload_finalize(cgroup_ops); |
1845 | TRACE("Finished setting up cgroups"); | |
1846 | ||
08dd2805 | 1847 | /* Run any host-side start hooks */ |
cfc62c60 CB |
1848 | ret = run_lxc_hooks(name, "start-host", conf, NULL); |
1849 | if (ret < 0) { | |
1850 | ERROR("Failed to run lxc.hook.start-host"); | |
341ed84c | 1851 | goto out_delete_net; |
08dd2805 SH |
1852 | } |
1853 | ||
408da065 CB |
1854 | /* Tell the child to complete its initialization and wait for it to exec |
1855 | * or return an error. (The child will never return | |
08dd2805 | 1856 | * LXC_SYNC_READY_START+1. It will either close the sync pipe, causing |
408da065 CB |
1857 | * lxc_sync_barrier_child to return success, or return a different |
1858 | * value, causing us to error out). | |
544a48a0 | 1859 | */ |
cfc62c60 CB |
1860 | ret = lxc_sync_barrier_child(handler, LXC_SYNC_READY_START); |
1861 | if (ret < 0) | |
341ed84c | 1862 | goto out_delete_net; |
0ad19a3f | 1863 | |
e389f2af CB |
1864 | if (handler->ns_clone_flags & CLONE_NEWNET) { |
1865 | ret = lxc_network_recv_name_and_ifindex_from_child(handler); | |
1866 | if (ret < 0) { | |
1867 | ERROR("Failed to receive names and ifindices for network devices from child"); | |
1868 | goto out_delete_net; | |
1869 | } | |
790255cf CB |
1870 | } |
1871 | ||
1872 | /* Now all networks are created, network devices are moved into place, | |
ad2ddfcd | 1873 | * and the correct names and ifindices in the respective namespaces have |
790255cf CB |
1874 | * been recorded. The corresponding structs have now all been filled. So |
1875 | * log them for debugging purposes. | |
1876 | */ | |
28d9e29e | 1877 | lxc_log_configured_netdevs(conf); |
790255cf | 1878 | |
f4152036 | 1879 | /* Read tty fds allocated by child. */ |
cfc62c60 CB |
1880 | ret = lxc_recv_ttys_from_child(handler); |
1881 | if (ret < 0) { | |
1882 | ERROR("Failed to receive tty info from child process"); | |
f4152036 CB |
1883 | goto out_delete_net; |
1884 | } | |
1885 | ||
c3e3c21a CB |
1886 | ret = lxc_seccomp_recv_notifier_fd(&handler->conf->seccomp, data_sock1); |
1887 | if (ret < 0) { | |
1888 | SYSERROR("Failed to receive seccomp notify fd from child"); | |
1889 | goto out_delete_net; | |
cdb2a47f | 1890 | } |
cdb2a47f | 1891 | |
cfc62c60 CB |
1892 | ret = handler->ops->post_start(handler, handler->data); |
1893 | if (ret < 0) | |
e6126dbe MN |
1894 | goto out_abort; |
1895 | ||
cfc62c60 CB |
1896 | ret = lxc_set_state(name, handler, RUNNING); |
1897 | if (ret < 0) { | |
1898 | ERROR("Failed to set state to \"%s\"", lxc_state2str(RUNNING)); | |
59eb99ba | 1899 | goto out_abort; |
3f21c114 | 1900 | } |
22ebac19 | 1901 | |
3c22086f | 1902 | lxc_sync_fini(handler); |
0c547523 | 1903 | |
e6126dbe | 1904 | return 0; |
1ac470c0 | 1905 | |
7fef7a06 | 1906 | out_delete_net: |
becad0ec | 1907 | if (handler->ns_clone_flags & CLONE_NEWNET) |
bb84beda | 1908 | lxc_delete_network(handler); |
1bd8d726 | 1909 | |
59eb99ba DL |
1910 | out_abort: |
1911 | lxc_abort(name, handler); | |
95ef0d7c CB |
1912 | |
1913 | out_sync_fini: | |
3c22086f | 1914 | lxc_sync_fini(handler); |
5c068da9 SH |
1915 | if (handler->pinfd >= 0) { |
1916 | close(handler->pinfd); | |
1917 | handler->pinfd = -1; | |
1918 | } | |
1919 | ||
b79fcd86 | 1920 | return -1; |
59eb99ba | 1921 | } |
0ad19a3f | 1922 | |
aa460476 | 1923 | int __lxc_start(const char *name, struct lxc_handler *handler, |
507cee36 | 1924 | struct lxc_operations* ops, void *data, const char *lxcpath, |
bb955810 | 1925 | bool daemonize, int *error_num) |
59eb99ba | 1926 | { |
c30e9b19 | 1927 | int ret, status; |
aa460476 | 1928 | struct lxc_conf *conf = handler->conf; |
72068e74 | 1929 | struct cgroup_ops *cgroup_ops; |
80090207 | 1930 | |
c30e9b19 CB |
1931 | ret = lxc_init(name, handler); |
1932 | if (ret < 0) { | |
1933 | ERROR("Failed to initialize container \"%s\"", name); | |
39cb2d9e | 1934 | goto out_fini_nonet; |
0ad19a3f | 1935 | } |
ee70bf78 CLG |
1936 | handler->ops = ops; |
1937 | handler->data = data; | |
bb955810 | 1938 | handler->daemonize = daemonize; |
72068e74 | 1939 | cgroup_ops = handler->cgroup_ops; |
e6126dbe | 1940 | |
76a26f55 | 1941 | if (!attach_block_device(handler->conf)) { |
c30e9b19 | 1942 | ERROR("Failed to attach block device"); |
575ea467 | 1943 | ret = -1; |
76a26f55 SH |
1944 | goto out_fini_nonet; |
1945 | } | |
1946 | ||
72068e74 CB |
1947 | if (!cgroup_ops->monitor_create(cgroup_ops, handler)) { |
1948 | ERROR("Failed to create monitor cgroup"); | |
575ea467 | 1949 | ret = -1; |
eeef32bb CB |
1950 | goto out_fini_nonet; |
1951 | } | |
1952 | ||
c581d2a6 | 1953 | if (!cgroup_ops->monitor_enter(cgroup_ops, handler)) { |
eeef32bb | 1954 | ERROR("Failed to enter monitor cgroup"); |
575ea467 | 1955 | ret = -1; |
72068e74 CB |
1956 | goto out_fini_nonet; |
1957 | } | |
1958 | ||
c581d2a6 CB |
1959 | if (!cgroup_ops->monitor_delegate_controllers(cgroup_ops)) { |
1960 | ERROR("Failed to delegate controllers to monitor cgroup"); | |
1961 | ret = -1; | |
1962 | goto out_fini_nonet; | |
1963 | } | |
1964 | ||
35120d9c | 1965 | if (geteuid() == 0 && !lxc_list_empty(&conf->id_map)) { |
408da065 | 1966 | /* If the backing store is a device, mount it here and now. */ |
35120d9c | 1967 | if (rootfs_is_blockdev(conf)) { |
c30e9b19 CB |
1968 | ret = unshare(CLONE_NEWNS); |
1969 | if (ret < 0) { | |
1970 | ERROR("Failed to unshare CLONE_NEWNS"); | |
35120d9c SH |
1971 | goto out_fini_nonet; |
1972 | } | |
c30e9b19 | 1973 | INFO("Unshared CLONE_NEWNS"); |
408da065 | 1974 | |
6a0c909a | 1975 | remount_all_slave(); |
8ce1abc2 | 1976 | ret = lxc_setup_rootfs_prepare_root(conf, name, lxcpath); |
c30e9b19 CB |
1977 | if (ret < 0) { |
1978 | ERROR("Error setting up rootfs mount as root before spawn"); | |
35120d9c SH |
1979 | goto out_fini_nonet; |
1980 | } | |
c30e9b19 | 1981 | INFO("Set up container rootfs as host root"); |
35120d9c SH |
1982 | } |
1983 | } | |
1984 | ||
c30e9b19 CB |
1985 | ret = lxc_spawn(handler); |
1986 | if (ret < 0) { | |
1987 | ERROR("Failed to spawn container \"%s\"", name); | |
76a26f55 | 1988 | goto out_detach_blockdev; |
0ad19a3f | 1989 | } |
1990 | ||
80308d07 | 1991 | handler->conf->reboot = REBOOT_NONE; |
8bee8851 | 1992 | |
c30e9b19 CB |
1993 | ret = lxc_poll(name, handler); |
1994 | if (ret) { | |
1995 | ERROR("LXC mainloop exited with error: %d", ret); | |
59eb99ba DL |
1996 | goto out_abort; |
1997 | } | |
0ad19a3f | 1998 | |
321db026 DJ |
1999 | if (!handler->init_died && handler->pid > 0) { |
2000 | ERROR("Child process is not killed"); | |
575ea467 | 2001 | ret = -1; |
321db026 DJ |
2002 | goto out_abort; |
2003 | } | |
2004 | ||
c30e9b19 CB |
2005 | status = lxc_wait_for_pid_status(handler->pid); |
2006 | if (status < 0) | |
2007 | SYSERROR("Failed to retrieve status for %d", handler->pid); | |
e043236e | 2008 | |
408da065 CB |
2009 | /* If the child process exited but was not signaled, it didn't call |
2010 | * reboot. This should mean it was an lxc-execute which simply exited. | |
2011 | * In any case, treat it as a 'halt'. | |
8b004f07 | 2012 | */ |
d028235d | 2013 | if (WIFSIGNALED(status)) { |
8b004f07 SH |
2014 | switch(WTERMSIG(status)) { |
2015 | case SIGINT: /* halt */ | |
c30e9b19 | 2016 | DEBUG("Container \"%s\" is halting", name); |
8b004f07 SH |
2017 | break; |
2018 | case SIGHUP: /* reboot */ | |
c30e9b19 | 2019 | DEBUG("Container \"%s\" is rebooting", name); |
80308d07 | 2020 | handler->conf->reboot = REBOOT_REQ; |
8b004f07 | 2021 | break; |
c2b9bd9e | 2022 | case SIGSYS: /* seccomp */ |
c30e9b19 | 2023 | DEBUG("Container \"%s\" violated its seccomp policy", name); |
c2b9bd9e | 2024 | break; |
8b004f07 | 2025 | default: |
c30e9b19 | 2026 | DEBUG("Unknown exit status for container \"%s\" init %d", name, WTERMSIG(status)); |
8b004f07 SH |
2027 | break; |
2028 | } | |
d028235d | 2029 | } |
828695d9 | 2030 | |
c30e9b19 CB |
2031 | ret = lxc_restore_phys_nics_to_netns(handler); |
2032 | if (ret < 0) | |
818a57fc | 2033 | ERROR("Failed to move physical network devices back to parent network namespace"); |
ce5782df | 2034 | |
5c068da9 SH |
2035 | if (handler->pinfd >= 0) { |
2036 | close(handler->pinfd); | |
2037 | handler->pinfd = -1; | |
2038 | } | |
2039 | ||
1787abca | 2040 | lxc_monitor_send_exit_code(name, status, handler->lxcpath); |
eb808539 | 2041 | lxc_error_set_and_log(handler->pid, status); |
a3b4f3d6 TA |
2042 | if (error_num) |
2043 | *error_num = handler->exit_status; | |
1bd8d726 | 2044 | |
9d7f9e52 | 2045 | out_fini: |
bb84beda | 2046 | lxc_delete_network(handler); |
74a2b586 | 2047 | |
76a26f55 SH |
2048 | out_detach_blockdev: |
2049 | detach_block_device(handler->conf); | |
2050 | ||
74a2b586 | 2051 | out_fini_nonet: |
3a0f472d | 2052 | lxc_fini(name, handler); |
c30e9b19 | 2053 | return ret; |
0ad19a3f | 2054 | |
59eb99ba | 2055 | out_abort: |
3a0f472d | 2056 | lxc_abort(name, handler); |
9d7f9e52 | 2057 | goto out_fini; |
0ad19a3f | 2058 | } |
ee70bf78 CLG |
2059 | |
2060 | struct start_args { | |
2061 | char *const *argv; | |
2062 | }; | |
2063 | ||
2064 | static int start(struct lxc_handler *handler, void* data) | |
2065 | { | |
2066 | struct start_args *arg = data; | |
2067 | ||
984984e4 | 2068 | NOTICE("Exec'ing \"%s\"", arg->argv[0]); |
ee70bf78 CLG |
2069 | |
2070 | execvp(arg->argv[0], arg->argv); | |
984984e4 | 2071 | SYSERROR("Failed to exec \"%s\"", arg->argv[0]); |
ee70bf78 CLG |
2072 | return 0; |
2073 | } | |
2074 | ||
2075 | static int post_start(struct lxc_handler *handler, void* data) | |
2076 | { | |
2077 | struct start_args *arg = data; | |
2078 | ||
4c8e880e | 2079 | NOTICE("Started \"%s\" with pid \"%d\"", arg->argv[0], handler->pid); |
ee70bf78 CLG |
2080 | return 0; |
2081 | } | |
2082 | ||
2083 | static struct lxc_operations start_ops = { | |
2084 | .start = start, | |
2085 | .post_start = post_start | |
2086 | }; | |
2087 | ||
aa460476 | 2088 | int lxc_start(const char *name, char *const argv[], struct lxc_handler *handler, |
bb955810 | 2089 | const char *lxcpath, bool daemonize, int *error_num) |
ee70bf78 CLG |
2090 | { |
2091 | struct start_args start_arg = { | |
2092 | .argv = argv, | |
2093 | }; | |
2094 | ||
b2efeb0b | 2095 | TRACE("Doing lxc_start"); |
bb955810 | 2096 | return __lxc_start(name, handler, &start_ops, &start_arg, lxcpath, daemonize, error_num); |
ee70bf78 | 2097 | } |
28272964 CB |
2098 | |
2099 | static void lxc_destroy_container_on_signal(struct lxc_handler *handler, | |
2100 | const char *name) | |
2101 | { | |
245100a0 | 2102 | char destroy[PATH_MAX]; |
f01f7975 | 2103 | struct lxc_container *c; |
ae3beac9 CB |
2104 | int ret = 0; |
2105 | bool bret = true; | |
2106 | ||
df31363a | 2107 | if (handler->conf->rootfs.path && handler->conf->rootfs.mount) { |
790255cf | 2108 | bret = do_destroy_container(handler); |
28272964 | 2109 | if (!bret) { |
ae3beac9 | 2110 | ERROR("Error destroying rootfs for container \"%s\"", name); |
28272964 CB |
2111 | return; |
2112 | } | |
2113 | } | |
ae3beac9 | 2114 | INFO("Destroyed rootfs for container \"%s\"", name); |
28272964 | 2115 | |
245100a0 CB |
2116 | ret = snprintf(destroy, PATH_MAX, "%s/%s", handler->lxcpath, name); |
2117 | if (ret < 0 || ret >= PATH_MAX) { | |
ae3beac9 | 2118 | ERROR("Error destroying directory for container \"%s\"", name); |
28272964 CB |
2119 | return; |
2120 | } | |
2121 | ||
f01f7975 CB |
2122 | c = lxc_container_new(name, handler->lxcpath); |
2123 | if (c) { | |
2124 | if (container_disk_lock(c)) { | |
ae3beac9 | 2125 | INFO("Could not update lxc_snapshots file"); |
f01f7975 CB |
2126 | lxc_container_put(c); |
2127 | } else { | |
2128 | mod_all_rdeps(c, false); | |
2129 | container_disk_unlock(c); | |
2130 | lxc_container_put(c); | |
2131 | } | |
2132 | } | |
2133 | ||
d0fbc7ba | 2134 | if (!handler->am_root) |
94b0a3ac CB |
2135 | ret = userns_exec_full(handler->conf, lxc_rmdir_onedev_wrapper, |
2136 | destroy, "lxc_rmdir_onedev_wrapper"); | |
28272964 CB |
2137 | else |
2138 | ret = lxc_rmdir_onedev(destroy, NULL); | |
2139 | ||
2140 | if (ret < 0) { | |
ae3beac9 | 2141 | ERROR("Error destroying directory for container \"%s\"", name); |
28272964 CB |
2142 | return; |
2143 | } | |
ae3beac9 | 2144 | INFO("Destroyed directory for container \"%s\"", name); |
28272964 CB |
2145 | } |
2146 | ||
2147 | static int lxc_rmdir_onedev_wrapper(void *data) | |
2148 | { | |
2149 | char *arg = (char *) data; | |
2150 | return lxc_rmdir_onedev(arg, NULL); | |
2151 | } | |
2152 | ||
134df645 CB |
2153 | static bool do_destroy_container(struct lxc_handler *handler) |
2154 | { | |
94b0a3ac CB |
2155 | int ret; |
2156 | ||
d0fbc7ba | 2157 | if (!handler->am_root) { |
94b0a3ac CB |
2158 | ret = userns_exec_full(handler->conf, storage_destroy_wrapper, |
2159 | handler->conf, "storage_destroy_wrapper"); | |
2160 | if (ret < 0) | |
d028235d | 2161 | return false; |
10bc1861 | 2162 | |
d028235d SG |
2163 | return true; |
2164 | } | |
10bc1861 | 2165 | |
790255cf | 2166 | return storage_destroy(handler->conf); |
28272964 | 2167 | } |