]>
Commit | Line | Data |
---|---|---|
cc73685d | 1 | /* SPDX-License-Identifier: LGPL-2.1+ */ |
0ad19a3f | 2 | |
d38dd64a CB |
3 | #ifndef _GNU_SOURCE |
4 | #define _GNU_SOURCE 1 | |
5 | #endif | |
0ad19a3f | 6 | #include <dirent.h> |
7 | #include <errno.h> | |
b0a33c1e | 8 | #include <fcntl.h> |
c476bdce | 9 | #include <grp.h> |
37515ebd | 10 | #include <poll.h> |
b467714b | 11 | #include <pthread.h> |
f3787121 CB |
12 | #include <signal.h> |
13 | #include <stdio.h> | |
14 | #include <stdlib.h> | |
15 | #include <string.h> | |
0ad19a3f | 16 | #include <sys/file.h> |
f4d507d5 | 17 | #include <sys/mount.h> |
f3787121 | 18 | #include <sys/param.h> |
0ad19a3f | 19 | #include <sys/prctl.h> |
f3787121 CB |
20 | #include <sys/socket.h> |
21 | #include <sys/stat.h> | |
22 | #include <sys/syscall.h> | |
ddceb1f9 | 23 | #include <sys/types.h> |
b0a33c1e | 24 | #include <sys/un.h> |
f3787121 | 25 | #include <sys/wait.h> |
d38dd64a | 26 | #include <unistd.h> |
495d2046 | 27 | |
f3787121 CB |
28 | #include "af_unix.h" |
29 | #include "caps.h" | |
563f2f2c | 30 | #include "cgroup.h" |
f3787121 | 31 | #include "commands.h" |
bbf5cf35 | 32 | #include "commands_utils.h" |
59eac805 | 33 | #include "compiler.h" |
f3787121 | 34 | #include "conf.h" |
d38dd64a | 35 | #include "config.h" |
74c6e2b0 | 36 | #include "confile_utils.h" |
e2bcd7db | 37 | #include "error.h" |
7fbb15ec | 38 | #include "file_utils.h" |
1cc8bd4b | 39 | #include "list.h" |
f3787121 | 40 | #include "log.h" |
d38dd64a | 41 | #include "lsm/lsm.h" |
f2d5a09d | 42 | #include "lxccontainer.h" |
f3787121 CB |
43 | #include "lxclock.h" |
44 | #include "lxcseccomp.h" | |
3ef9b3d3 | 45 | #include "macro.h" |
f3787121 | 46 | #include "mainloop.h" |
4ffeaf27 | 47 | #include "memory_utils.h" |
63376d7d | 48 | #include "monitor.h" |
f549edcc | 49 | #include "namespace.h" |
811ef482 | 50 | #include "network.h" |
f40988c7 | 51 | #include "process_utils.h" |
f3787121 | 52 | #include "start.h" |
6be5397b CB |
53 | #include "storage/storage.h" |
54 | #include "storage/storage_utils.h" | |
0ed9b1bc | 55 | #include "sync.h" |
59524108 | 56 | #include "syscall_wrappers.h" |
0ed9b1bc CB |
57 | #include "terminal.h" |
58 | #include "utils.h" | |
36eb9bde | 59 | |
d38dd64a CB |
60 | #if HAVE_LIBCAP |
61 | #include <sys/capability.h> | |
62 | #endif | |
63 | ||
9de31d5a CB |
64 | #ifndef HAVE_STRLCPY |
65 | #include "include/strlcpy.h" | |
66 | #endif | |
67 | ||
ac2cecc4 | 68 | lxc_log_define(start, lxc); |
36eb9bde | 69 | |
f01f7975 | 70 | extern void mod_all_rdeps(struct lxc_container *c, bool inc); |
790255cf | 71 | static bool do_destroy_container(struct lxc_handler *handler); |
28272964 CB |
72 | static int lxc_rmdir_onedev_wrapper(void *data); |
73 | static void lxc_destroy_container_on_signal(struct lxc_handler *handler, | |
74 | const char *name); | |
75 | ||
c8154066 SH |
76 | static void print_top_failing_dir(const char *path) |
77 | { | |
dcd9a847 | 78 | __do_free char *copy = NULL; |
9e5f5f2f | 79 | int ret; |
4ffeaf27 | 80 | char *e, *p, saved; |
cbb9c7c7 | 81 | |
4ffeaf27 | 82 | copy = must_copy_string(path); |
c8154066 | 83 | p = copy; |
4ffeaf27 | 84 | e = copy + strlen(path); |
cbb9c7c7 | 85 | |
c8154066 | 86 | while (p < e) { |
f3787121 CB |
87 | while (p < e && *p == '/') |
88 | p++; | |
9e5f5f2f | 89 | |
f3787121 CB |
90 | while (p < e && *p != '/') |
91 | p++; | |
9e5f5f2f | 92 | |
c8154066 SH |
93 | saved = *p; |
94 | *p = '\0'; | |
9e5f5f2f CB |
95 | |
96 | ret = access(copy, X_OK); | |
97 | if (ret != 0) { | |
818a57fc | 98 | SYSERROR("Could not access %s. Please grant it x access, or add an ACL for the container " "root", copy); |
c8154066 SH |
99 | return; |
100 | } | |
101 | *p = saved; | |
102 | } | |
103 | } | |
104 | ||
5032bf39 | 105 | static void lxc_put_nsfds(struct lxc_handler *handler) |
408da065 | 106 | { |
818a57fc | 107 | for (int i = 0; i < LXC_NS_MAX; i++) { |
5032bf39 | 108 | if (handler->nsfd[i] < 0) |
39cd919c CB |
109 | continue; |
110 | ||
818a57fc | 111 | close_prot_errno_disarm(handler->nsfd[i]); |
9f30a190 | 112 | } |
9f30a190 MM |
113 | } |
114 | ||
4cb53844 CB |
115 | static int lxc_try_preserve_ns(const int pid, const char *ns) |
116 | { | |
117 | int fd; | |
118 | ||
119 | fd = lxc_preserve_ns(pid, ns); | |
120 | if (fd < 0) { | |
818a57fc CB |
121 | if (errno != ENOENT) |
122 | return log_error_errno(-EINVAL, | |
123 | errno, "Failed to preserve %s namespace", | |
124 | ns); | |
4cb53844 | 125 | |
b87ee312 CB |
126 | return log_warn_errno(-EOPNOTSUPP, |
127 | errno, "Kernel does not support preserving %s namespaces", | |
128 | ns); | |
4cb53844 CB |
129 | } |
130 | ||
131 | return fd; | |
132 | } | |
133 | ||
134 | /* lxc_try_preserve_namespaces: open /proc/@pid/ns/@ns for each namespace | |
135 | * specified in ns_clone_flags. | |
4d8ac866 | 136 | * Return true on success, false on failure. |
62d05d9b | 137 | */ |
4cb53844 CB |
138 | static bool lxc_try_preserve_namespaces(struct lxc_handler *handler, |
139 | int ns_clone_flags, pid_t pid) | |
f3787121 | 140 | { |
9fef3355 | 141 | int i; |
9f30a190 | 142 | |
9f30a190 | 143 | for (i = 0; i < LXC_NS_MAX; i++) |
5032bf39 | 144 | handler->nsfd[i] = -EBADF; |
cd43d2d1 | 145 | |
9f30a190 | 146 | for (i = 0; i < LXC_NS_MAX; i++) { |
4cb53844 CB |
147 | int fd; |
148 | ||
becad0ec | 149 | if ((ns_clone_flags & ns_info[i].clone_flag) == 0) |
9f30a190 | 150 | continue; |
28d9e29e | 151 | |
4cb53844 CB |
152 | fd = lxc_try_preserve_ns(pid, ns_info[i].proc_name); |
153 | if (fd < 0) { | |
4cb53844 CB |
154 | /* Do not fail to start container on kernels that do |
155 | * not support interacting with namespaces through | |
156 | * /proc. | |
157 | */ | |
158 | if (fd == -EOPNOTSUPP) | |
159 | continue; | |
fa3a5b22 | 160 | |
4cb53844 CB |
161 | lxc_put_nsfds(handler); |
162 | return false; | |
163 | } | |
164 | ||
165 | handler->nsfd[i] = fd; | |
166 | DEBUG("Preserved %s namespace via fd %d", ns_info[i].proc_name, | |
167 | handler->nsfd[i]); | |
9f30a190 MM |
168 | } |
169 | ||
62d05d9b | 170 | return true; |
9f30a190 MM |
171 | } |
172 | ||
ec1dc633 | 173 | static inline bool match_stdfds(int fd) |
80090207 | 174 | { |
ec1dc633 | 175 | return (fd == STDIN_FILENO || fd == STDOUT_FILENO || fd == STDERR_FILENO); |
80090207 CLG |
176 | } |
177 | ||
4c03724e | 178 | #ifdef HAVE_DLOG |
179 | static bool match_dlog_fds(struct dirent *direntp) | |
180 | { | |
181 | char path[PATH_MAX] = {0}; | |
182 | char link[PATH_MAX] = {0}; | |
183 | ssize_t linklen; | |
184 | int ret; | |
185 | ||
fa60cd7b CB |
186 | ret = strnprintf(path, sizeof(path), "/proc/self/fd/%s", direntp->d_name); |
187 | if (ret < 0) | |
818a57fc | 188 | return log_error(false, "Failed to create file descriptor name"); |
4c03724e | 189 | |
190 | linklen = readlink(path, link, PATH_MAX); | |
818a57fc CB |
191 | if (linklen < 0) |
192 | return log_error(false, "Failed to read link path - \"%s\"", path); | |
193 | else if (linklen >= PATH_MAX) | |
194 | return log_error(false, "The name of link path is too long - \"%s\"", path); | |
195 | ||
196 | if (strcmp(link, "/dev/log_main") == 0 || | |
197 | strcmp(link, "/dev/log_system") == 0 || | |
198 | strcmp(link, "/dev/log_radio") == 0) | |
4c03724e | 199 | return true; |
200 | ||
201 | return false; | |
202 | } | |
203 | #endif | |
204 | ||
47a46cf1 CB |
205 | int lxc_check_inherited(struct lxc_conf *conf, bool closeall, |
206 | int *fds_to_ignore, size_t len_fds) | |
80090207 | 207 | { |
80090207 | 208 | int fd, fddir; |
47a46cf1 | 209 | size_t i; |
80090207 | 210 | DIR *dir; |
a5a70219 | 211 | struct dirent *direntp; |
80090207 | 212 | |
d2cf4c37 SH |
213 | if (conf && conf->close_all_fds) |
214 | closeall = true; | |
215 | ||
4e2d6b9a CB |
216 | /* |
217 | * Disable syslog at this point to avoid the above logging | |
218 | * function to open a new fd and make the check_inherited function | |
219 | * enter an infinite loop. | |
220 | */ | |
221 | lxc_log_syslog_disable(); | |
222 | ||
b119f362 | 223 | restart: |
80090207 | 224 | dir = opendir("/proc/self/fd"); |
818a57fc CB |
225 | if (!dir) |
226 | return log_warn(-1, "Failed to open directory"); | |
80090207 CLG |
227 | |
228 | fddir = dirfd(dir); | |
229 | ||
74f96976 | 230 | while ((direntp = readdir(dir))) { |
a5a70219 | 231 | int ret; |
d39b10eb CB |
232 | struct lxc_list *cur; |
233 | bool matched = false; | |
234 | ||
a5a70219 | 235 | if (strcmp(direntp->d_name, ".") == 0) |
80090207 CLG |
236 | continue; |
237 | ||
a5a70219 | 238 | if (strcmp(direntp->d_name, "..") == 0) |
80090207 CLG |
239 | continue; |
240 | ||
a5a70219 CB |
241 | ret = lxc_safe_int(direntp->d_name, &fd); |
242 | if (ret < 0) { | |
243 | INFO("Could not parse file descriptor for \"%s\"", direntp->d_name); | |
d4cff0d2 CB |
244 | continue; |
245 | } | |
80090207 | 246 | |
47a46cf1 CB |
247 | for (i = 0; i < len_fds; i++) |
248 | if (fds_to_ignore[i] == fd) | |
249 | break; | |
250 | ||
251 | if (fd == fddir || fd == lxc_log_fd || | |
252 | (i < len_fds && fd == fds_to_ignore[i])) | |
80090207 CLG |
253 | continue; |
254 | ||
d39b10eb | 255 | /* Keep state clients that wait on reboots. */ |
3ee9e4fb CB |
256 | if (conf) { |
257 | lxc_list_for_each(cur, &conf->state_clients) { | |
258 | struct lxc_state_client *client = cur->elem; | |
d39b10eb | 259 | |
3ee9e4fb CB |
260 | if (client->clientfd != fd) |
261 | continue; | |
d39b10eb | 262 | |
3ee9e4fb CB |
263 | matched = true; |
264 | break; | |
265 | } | |
d39b10eb CB |
266 | } |
267 | ||
268 | if (matched) | |
269 | continue; | |
270 | ||
858377e4 SH |
271 | if (current_config && fd == current_config->logfd) |
272 | continue; | |
273 | ||
ec1dc633 | 274 | if (match_stdfds(fd)) |
80090207 | 275 | continue; |
80090207 | 276 | |
1d5e5f26 | 277 | #ifdef HAVE_DLOG |
278 | if (match_dlog_fds(direntp)) | |
279 | continue; | |
280 | ||
281 | #endif | |
d2cf4c37 | 282 | if (closeall) { |
4e2d6b9a CB |
283 | if (close(fd)) |
284 | SYSINFO("Closed inherited fd %d", fd); | |
285 | else | |
286 | INFO("Closed inherited fd %d", fd); | |
b119f362 | 287 | closedir(dir); |
b119f362 SH |
288 | goto restart; |
289 | } | |
28d9e29e | 290 | WARN("Inherited fd %d", fd); |
80090207 | 291 | } |
4e2d6b9a | 292 | closedir(dir); |
80090207 | 293 | |
4e2d6b9a CB |
294 | /* |
295 | * Only enable syslog at this point to avoid the above logging | |
296 | * function to open a new fd and make the check_inherited function | |
297 | * enter an infinite loop. | |
64c57ea1 | 298 | */ |
4e2d6b9a | 299 | lxc_log_syslog_enable(); |
64c57ea1 | 300 | |
92c7f629 | 301 | return 0; |
80090207 CLG |
302 | } |
303 | ||
83ee7875 | 304 | static int setup_signal_fd(sigset_t *oldmask) |
b0a33c1e | 305 | { |
7288dfb6 | 306 | int ret; |
b0a33c1e | 307 | sigset_t mask; |
7288dfb6 | 308 | const int signals[] = {SIGBUS, SIGILL, SIGSEGV, SIGWINCH}; |
b0a33c1e | 309 | |
408da065 | 310 | /* Block everything except serious error signals. */ |
df0795b1 CB |
311 | ret = sigfillset(&mask); |
312 | if (ret < 0) | |
313 | return -EBADF; | |
314 | ||
9b10bef5 | 315 | for (int sig = 0; sig < (sizeof(signals) / sizeof(signals[0])); sig++) { |
df0795b1 CB |
316 | ret = sigdelset(&mask, signals[sig]); |
317 | if (ret < 0) | |
318 | return -EBADF; | |
b0a33c1e | 319 | } |
320 | ||
b467714b | 321 | ret = pthread_sigmask(SIG_BLOCK, &mask, oldmask); |
818a57fc CB |
322 | if (ret < 0) |
323 | return log_error_errno(-EBADF, errno, | |
324 | "Failed to set signal mask"); | |
b0a33c1e | 325 | |
df0795b1 | 326 | ret = signalfd(-1, &mask, SFD_CLOEXEC); |
818a57fc CB |
327 | if (ret < 0) |
328 | return log_error_errno(-EBADF, | |
329 | errno, "Failed to create signal file descriptor"); | |
b0a33c1e | 330 | |
df0795b1 | 331 | TRACE("Created signal file descriptor %d", ret); |
1ac470c0 | 332 | |
df0795b1 | 333 | return ret; |
b0a33c1e | 334 | } |
335 | ||
84c92abd | 336 | static int signal_handler(int fd, uint32_t events, void *data, |
f3787121 | 337 | struct lxc_epoll_descr *descr) |
b0a33c1e | 338 | { |
15cd25fd | 339 | int ret; |
3c319edb CB |
340 | siginfo_t info; |
341 | struct signalfd_siginfo siginfo; | |
342 | struct lxc_handler *hdlr = data; | |
15cd25fd | 343 | |
489f39be | 344 | ret = lxc_read_nointr(fd, &siginfo, sizeof(siginfo)); |
818a57fc CB |
345 | if (ret < 0) |
346 | return log_error(LXC_MAINLOOP_ERROR, "Failed to read signal info from signal file descriptor %d", fd); | |
15cd25fd | 347 | |
818a57fc | 348 | if (ret != sizeof(siginfo)) |
f7a97743 | 349 | return log_error(LXC_MAINLOOP_ERROR, "Unexpected size for struct signalfd_siginfo"); |
15cd25fd | 350 | |
408da065 | 351 | /* Check whether init is running. */ |
80507ee8 | 352 | info.si_pid = 0; |
3c319edb CB |
353 | ret = waitid(P_PID, hdlr->pid, &info, WEXITED | WNOWAIT | WNOHANG); |
354 | if (ret == 0 && info.si_pid == hdlr->pid) | |
355 | hdlr->init_died = true; | |
80507ee8 | 356 | |
cd5177e9 TA |
357 | /* Try to figure out a reasonable exit status to report. */ |
358 | if (hdlr->init_died) { | |
359 | switch (info.si_code) { | |
360 | case CLD_EXITED: | |
361 | hdlr->exit_status = info.si_status << 8; | |
362 | break; | |
363 | case CLD_KILLED: | |
364 | case CLD_DUMPED: | |
365 | case CLD_STOPPED: | |
366 | hdlr->exit_status = info.si_status << 8 | 0x7f; | |
367 | break; | |
368 | case CLD_CONTINUED: | |
369 | /* Huh? The waitid() told us it's dead *and* continued? */ | |
370 | WARN("Init %d dead and continued?", hdlr->pid); | |
371 | hdlr->exit_status = 1; | |
372 | break; | |
373 | default: | |
20993a97 | 374 | ERROR("Unknown si_code: %d", info.si_code); |
55e1311e | 375 | hdlr->exit_status = 1; |
cd5177e9 TA |
376 | } |
377 | } | |
378 | ||
d4b5d7a8 | 379 | if (siginfo.ssi_signo == SIGHUP) { |
33942046 CB |
380 | if (hdlr->pidfd >= 0) |
381 | lxc_raw_pidfd_send_signal(hdlr->pidfd, SIGTERM, NULL, 0); | |
d9bb2fba CB |
382 | else |
383 | kill(hdlr->pid, SIGTERM); | |
d4b5d7a8 | 384 | INFO("Killing %d since terminal hung up", hdlr->pid); |
9b10bef5 CB |
385 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE |
386 | : LXC_MAINLOOP_CONTINUE; | |
d4b5d7a8 TA |
387 | } |
388 | ||
9cb94384 | 389 | if (siginfo.ssi_signo != SIGCHLD) { |
33942046 CB |
390 | if (hdlr->pidfd >= 0) |
391 | lxc_raw_pidfd_send_signal(hdlr->pidfd, | |
392 | siginfo.ssi_signo, NULL, 0); | |
d9bb2fba CB |
393 | else |
394 | kill(hdlr->pid, siginfo.ssi_signo); | |
9cb94384 | 395 | INFO("Forwarded signal %d to pid %d", siginfo.ssi_signo, hdlr->pid); |
9b10bef5 CB |
396 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE |
397 | : LXC_MAINLOOP_CONTINUE; | |
9cb94384 TA |
398 | } |
399 | ||
3a9e949f TA |
400 | /* More robustness, protect ourself from a SIGCHLD sent |
401 | * by a process different from the container init. | |
402 | */ | |
403 | if (siginfo.ssi_pid != hdlr->pid) { | |
6e94162a CB |
404 | NOTICE("Received %d from pid %d instead of container init %d", |
405 | siginfo.ssi_signo, siginfo.ssi_pid, hdlr->pid); | |
9b10bef5 CB |
406 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE |
407 | : LXC_MAINLOOP_CONTINUE; | |
3a9e949f TA |
408 | } |
409 | ||
408da065 | 410 | if (siginfo.ssi_code == CLD_STOPPED) { |
6e94162a | 411 | INFO("Container init process was stopped"); |
9b10bef5 CB |
412 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE |
413 | : LXC_MAINLOOP_CONTINUE; | |
414 | } | |
415 | ||
416 | if (siginfo.ssi_code == CLD_CONTINUED) { | |
6e94162a | 417 | INFO("Container init process was continued"); |
9b10bef5 CB |
418 | return hdlr->init_died ? LXC_MAINLOOP_CLOSE |
419 | : LXC_MAINLOOP_CONTINUE; | |
15cd25fd | 420 | } |
1ac470c0 | 421 | |
818a57fc | 422 | return log_debug(LXC_MAINLOOP_CLOSE, "Container init process %d exited", hdlr->pid); |
b0a33c1e | 423 | } |
424 | ||
974a8aba CB |
425 | int lxc_serve_state_clients(const char *name, struct lxc_handler *handler, |
426 | lxc_state_t state) | |
66aeffc7 | 427 | { |
9de31d5a | 428 | size_t retlen; |
dbc9832d CB |
429 | ssize_t ret; |
430 | struct lxc_list *cur, *next; | |
dbc9832d CB |
431 | struct lxc_msg msg = {.type = lxc_msg_state, .value = state}; |
432 | ||
974a8aba CB |
433 | if (state == THAWED) |
434 | handler->state = RUNNING; | |
435 | else | |
436 | handler->state = state; | |
437 | ||
9dfa4041 | 438 | TRACE("Set container state to %s", lxc_state2str(state)); |
dbc9832d | 439 | |
818a57fc CB |
440 | if (lxc_list_empty(&handler->conf->state_clients)) |
441 | return log_trace(0, "No state clients registered"); | |
dbc9832d | 442 | |
9de31d5a CB |
443 | retlen = strlcpy(msg.name, name, sizeof(msg.name)); |
444 | if (retlen >= sizeof(msg.name)) | |
445 | return -E2BIG; | |
dbc9832d | 446 | |
d39b10eb | 447 | lxc_list_for_each_safe(cur, &handler->conf->state_clients, next) { |
d85617bc | 448 | struct lxc_state_client *client = cur->elem; |
dbc9832d | 449 | |
d39b10eb | 450 | if (client->states[state] == 0) { |
9dfa4041 | 451 | TRACE("State %s not registered for state client %d", |
dbc9832d CB |
452 | lxc_state2str(state), client->clientfd); |
453 | continue; | |
454 | } | |
455 | ||
9dfa4041 | 456 | TRACE("Sending state %s to state client %d", |
dbc9832d CB |
457 | lxc_state2str(state), client->clientfd); |
458 | ||
7fbb15ec CB |
459 | ret = lxc_send_nointr(client->clientfd, &msg, sizeof(msg), MSG_NOSIGNAL); |
460 | if (ret <= 0) | |
6d1400b5 | 461 | SYSERROR("Failed to send message to client"); |
dbc9832d CB |
462 | |
463 | /* kick client from list */ | |
dbc9832d | 464 | lxc_list_del(cur); |
300d1cb4 | 465 | close(client->clientfd); |
dbc9832d CB |
466 | free(cur->elem); |
467 | free(cur); | |
468 | } | |
dbc9832d | 469 | |
5e5576a4 CB |
470 | return 0; |
471 | } | |
472 | ||
473 | static int lxc_serve_state_socket_pair(const char *name, | |
474 | struct lxc_handler *handler, | |
475 | lxc_state_t state) | |
476 | { | |
477 | ssize_t ret; | |
478 | ||
bb955810 | 479 | if (!handler->daemonize || |
5e5576a4 CB |
480 | handler->state_socket_pair[1] < 0 || |
481 | state == STARTING) | |
482 | return 0; | |
483 | ||
484 | /* Close read end of the socket pair. */ | |
a783a414 | 485 | close_prot_errno_disarm(handler->state_socket_pair[0]); |
5e5576a4 | 486 | |
ee8377bd | 487 | again: |
5e5576a4 CB |
488 | ret = lxc_abstract_unix_send_credential(handler->state_socket_pair[1], |
489 | &(int){state}, sizeof(int)); | |
9044b79e | 490 | if (ret < 0) { |
491 | SYSERROR("Failed to send state to %d", handler->state_socket_pair[1]); | |
492 | ||
ee8377bd CB |
493 | if (errno == EINTR) |
494 | goto again; | |
9044b79e | 495 | |
496 | return -1; | |
497 | } | |
498 | ||
818a57fc CB |
499 | if (ret != sizeof(int)) |
500 | return log_error(-1, "Message too long : %d", handler->state_socket_pair[1]); | |
5e5576a4 CB |
501 | |
502 | TRACE("Sent container state \"%s\" to %d", lxc_state2str(state), | |
503 | handler->state_socket_pair[1]); | |
504 | ||
505 | /* Close write end of the socket pair. */ | |
818a57fc | 506 | close_prot_errno_disarm(handler->state_socket_pair[1]); |
5e5576a4 CB |
507 | |
508 | return 0; | |
509 | } | |
510 | ||
511 | int lxc_set_state(const char *name, struct lxc_handler *handler, | |
512 | lxc_state_t state) | |
513 | { | |
514 | int ret; | |
515 | ||
516 | ret = lxc_serve_state_socket_pair(name, handler, state); | |
818a57fc CB |
517 | if (ret < 0) |
518 | return log_error(-1, "Failed to synchronize via anonymous pair of unix sockets"); | |
5e5576a4 CB |
519 | |
520 | ret = lxc_serve_state_clients(name, handler, state); | |
521 | if (ret < 0) | |
522 | return -1; | |
523 | ||
dbc9832d CB |
524 | /* This function will try to connect to the legacy lxc-monitord state |
525 | * server and only exists for backwards compatibility. | |
526 | */ | |
9123e471 | 527 | lxc_monitor_send_state(name, state, handler->lxcpath); |
dbc9832d | 528 | |
66aeffc7 DL |
529 | return 0; |
530 | } | |
531 | ||
735f2c6e | 532 | int lxc_poll(const char *name, struct lxc_handler *handler) |
b0a33c1e | 533 | { |
3c319edb | 534 | int ret; |
86530b0a | 535 | bool has_console = true; |
3c319edb | 536 | struct lxc_epoll_descr descr, descr_console; |
b0a33c1e | 537 | |
03760215 CB |
538 | if (handler->conf->console.path && |
539 | strcmp(handler->conf->console.path, "none") == 0) | |
86530b0a L |
540 | has_console = false; |
541 | ||
3c319edb CB |
542 | ret = lxc_mainloop_open(&descr); |
543 | if (ret < 0) { | |
544 | ERROR("Failed to create mainloop"); | |
50c8bf05 | 545 | goto out_sigfd; |
b0a33c1e | 546 | } |
547 | ||
30a33fbd CB |
548 | if (has_console) { |
549 | ret = lxc_mainloop_open(&descr_console); | |
550 | if (ret < 0) { | |
551 | ERROR("Failed to create console mainloop"); | |
552 | goto out_mainloop; | |
553 | } | |
3c319edb CB |
554 | } |
555 | ||
1cc8bd4b | 556 | ret = lxc_mainloop_add_handler(&descr, handler->sigfd, signal_handler, handler); |
3c319edb | 557 | if (ret < 0) { |
1cc8bd4b | 558 | ERROR("Failed to add signal handler for %d to mainloop", handler->sigfd); |
3c319edb | 559 | goto out_mainloop_console; |
b0a33c1e | 560 | } |
561 | ||
2ac0f627 CB |
562 | ret = lxc_seccomp_setup_proxy(&handler->conf->seccomp, &descr, handler); |
563 | if (ret < 0) { | |
564 | ERROR("Failed to setup seccomp proxy"); | |
c3e3c21a | 565 | goto out_mainloop_console; |
2ac0f627 | 566 | } |
cdb2a47f | 567 | |
30a33fbd | 568 | if (has_console) { |
dcad02f8 | 569 | struct lxc_terminal *console = &handler->conf->console; |
63376d7d | 570 | |
093bce5f | 571 | ret = lxc_terminal_mainloop_add(&descr, console); |
30a33fbd CB |
572 | if (ret < 0) { |
573 | ERROR("Failed to add console handlers to mainloop"); | |
574 | goto out_mainloop_console; | |
575 | } | |
576 | ||
093bce5f | 577 | ret = lxc_terminal_mainloop_add(&descr_console, console); |
30a33fbd CB |
578 | if (ret < 0) { |
579 | ERROR("Failed to add console handlers to console mainloop"); | |
580 | goto out_mainloop_console; | |
581 | } | |
a54585ad L |
582 | |
583 | handler->conf->console.descr = &descr; | |
563f2f2c DL |
584 | } |
585 | ||
3c319edb CB |
586 | ret = lxc_cmd_mainloop_add(name, &descr, handler); |
587 | if (ret < 0) { | |
588 | ERROR("Failed to add command handler to mainloop"); | |
589 | goto out_mainloop_console; | |
590 | } | |
b0a33c1e | 591 | |
3c319edb | 592 | TRACE("Mainloop is ready"); |
b0a33c1e | 593 | |
3c319edb | 594 | ret = lxc_mainloop(&descr, -1); |
a783a414 | 595 | close_prot_errno_disarm(descr.epfd); |
1cc8bd4b | 596 | if (ret < 0 || !handler->init_died) |
42b09f94 | 597 | goto out_mainloop_console; |
1cc8bd4b | 598 | |
30a33fbd CB |
599 | if (has_console) |
600 | ret = lxc_mainloop(&descr_console, 0); | |
601 | ||
3c319edb | 602 | out_mainloop_console: |
30a33fbd CB |
603 | if (has_console) { |
604 | lxc_mainloop_close(&descr_console); | |
605 | TRACE("Closed console mainloop"); | |
606 | } | |
3c319edb | 607 | |
42b09f94 CB |
608 | out_mainloop: |
609 | lxc_mainloop_close(&descr); | |
610 | TRACE("Closed mainloop"); | |
611 | ||
b0a33c1e | 612 | out_sigfd: |
1cc8bd4b | 613 | TRACE("Closed signal file descriptor %d", handler->sigfd); |
a783a414 | 614 | close_prot_errno_disarm(handler->sigfd); |
408da065 | 615 | |
1cc8bd4b | 616 | return ret; |
b0a33c1e | 617 | } |
618 | ||
a42abcce | 619 | void lxc_put_handler(struct lxc_handler *handler) |
f2e07cb6 | 620 | { |
a783a414 CB |
621 | close_prot_errno_disarm(handler->pidfd); |
622 | close_prot_errno_disarm(handler->sigfd); | |
1cc8bd4b | 623 | lxc_put_nsfds(handler); |
80308d07 | 624 | if (handler->conf && handler->conf->reboot == REBOOT_NONE) |
a783a414 CB |
625 | close_prot_errno_disarm(handler->conf->maincmd_fd); |
626 | close_prot_errno_disarm(handler->monitor_status_fd); | |
627 | close_prot_errno_disarm(handler->state_socket_pair[0]); | |
628 | close_prot_errno_disarm(handler->state_socket_pair[1]); | |
629 | cgroup_exit(handler->cgroup_ops); | |
a42abcce CB |
630 | if (handler->conf && handler->conf->reboot == REBOOT_NONE) |
631 | free_disarm(handler); | |
632 | else | |
633 | handler->conf = NULL; | |
f2e07cb6 CB |
634 | } |
635 | ||
a42abcce CB |
636 | struct lxc_handler *lxc_init_handler(struct lxc_handler *old, |
637 | const char *name, struct lxc_conf *conf, | |
5e5576a4 | 638 | const char *lxcpath, bool daemonize) |
59eb99ba | 639 | { |
85c279bb | 640 | int nr_keep_fds = 0; |
818a57fc | 641 | int ret; |
3a0f472d DL |
642 | struct lxc_handler *handler; |
643 | ||
a42abcce CB |
644 | if (!old) |
645 | handler = zalloc(sizeof(*handler)); | |
646 | else | |
647 | handler = old; | |
fdecbc9c | 648 | if (!handler) |
3a0f472d | 649 | return NULL; |
9044b79e | 650 | |
fdecbc9c CB |
651 | /* Note that am_guest_unpriv() checks the effective uid. We |
652 | * probably don't care if we are real root only if we are running | |
653 | * as root so this should be fine. | |
790255cf | 654 | */ |
e0010464 | 655 | handler->am_root = !am_guest_unpriv(); |
fae349da | 656 | handler->conf = conf; |
9123e471 | 657 | handler->lxcpath = lxcpath; |
818a57fc CB |
658 | handler->init_died = false; |
659 | handler->data_sock[0] = -EBADF; | |
660 | handler->data_sock[1] = -EBADF; | |
4d8bdfa0 | 661 | handler->monitor_status_fd = -EBADF; |
33942046 | 662 | handler->pidfd = -EBADF; |
1cc8bd4b | 663 | handler->sigfd = -EBADF; |
818a57fc CB |
664 | handler->state_socket_pair[0] = -EBADF; |
665 | handler->state_socket_pair[1] = -EBADF; | |
80308d07 | 666 | if (handler->conf->reboot == REBOOT_NONE) |
d39b10eb | 667 | lxc_list_init(&handler->conf->state_clients); |
fae349da | 668 | |
818a57fc CB |
669 | for (int i = 0; i < LXC_NS_MAX; i++) |
670 | handler->nsfd[i] = -EBADF; | |
b6b2b194 | 671 | |
f0ecc19d | 672 | handler->name = name; |
c581d2a6 CB |
673 | if (daemonize) |
674 | handler->transient_pid = lxc_raw_getpid(); | |
675 | else | |
676 | handler->transient_pid = -1; | |
3bdf52d7 | 677 | |
80308d07 | 678 | if (daemonize && handler->conf->reboot == REBOOT_NONE) { |
5e5576a4 | 679 | /* Create socketpair() to synchronize on daemonized startup. |
fdecbc9c CB |
680 | * When the container reboots we don't need to synchronize |
681 | * again currently so don't open another socketpair(). | |
5e5576a4 CB |
682 | */ |
683 | ret = socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0, | |
684 | handler->state_socket_pair); | |
685 | if (ret < 0) { | |
686 | ERROR("Failed to create anonymous pair of unix sockets"); | |
687 | goto on_error; | |
688 | } | |
9044b79e | 689 | |
5e5576a4 CB |
690 | TRACE("Created anonymous pair {%d,%d} of unix sockets", |
691 | handler->state_socket_pair[0], | |
692 | handler->state_socket_pair[1]); | |
85c279bb CB |
693 | handler->keep_fds[nr_keep_fds++] = handler->state_socket_pair[0]; |
694 | handler->keep_fds[nr_keep_fds++] = handler->state_socket_pair[1]; | |
5e5576a4 CB |
695 | } |
696 | ||
80308d07 | 697 | if (handler->conf->reboot == REBOOT_NONE) { |
d39b10eb CB |
698 | handler->conf->maincmd_fd = lxc_cmd_init(name, lxcpath, "command"); |
699 | if (handler->conf->maincmd_fd < 0) { | |
700 | ERROR("Failed to set up command socket"); | |
701 | goto on_error; | |
702 | } | |
85c279bb | 703 | handler->keep_fds[nr_keep_fds++] = handler->conf->maincmd_fd; |
aa460476 | 704 | } |
9044b79e | 705 | |
9dfa4041 | 706 | TRACE("Unix domain socket %d for command server is ready", |
aa460476 CB |
707 | handler->conf->maincmd_fd); |
708 | ||
709 | return handler; | |
710 | ||
f2e07cb6 | 711 | on_error: |
a42abcce | 712 | lxc_put_handler(handler); |
aa460476 CB |
713 | |
714 | return NULL; | |
715 | } | |
716 | ||
717 | int lxc_init(const char *name, struct lxc_handler *handler) | |
718 | { | |
f62cf1d4 | 719 | __do_close int status_fd = -EBADF; |
a2c09be0 | 720 | int ret; |
4a03ded4 | 721 | const char *loglevel; |
aa460476 CB |
722 | struct lxc_conf *conf = handler->conf; |
723 | ||
434c8e15 | 724 | handler->monitor_pid = lxc_raw_getpid(); |
4d8bdfa0 | 725 | status_fd = open("/proc/self/status", O_RDONLY | O_CLOEXEC); |
59b1b67e CB |
726 | if (status_fd < 0) |
727 | return log_error_errno(-1, errno, "Failed to open monitor status fd"); | |
434c8e15 | 728 | |
4eb19ac0 | 729 | handler->lsm_ops = lsm_init_static(); |
2170b263 | 730 | TRACE("Initialized LSM"); |
d2e30e99 | 731 | |
408da065 | 732 | /* Begin by setting the state to STARTING. */ |
2170b263 | 733 | ret = lxc_set_state(name, handler, STARTING); |
59b1b67e CB |
734 | if (ret < 0) |
735 | return log_error(-1, "Failed to set state to \"%s\"", lxc_state2str(STARTING)); | |
46800e77 | 736 | TRACE("Set container state to \"STARTING\""); |
0ad19a3f | 737 | |
408da065 | 738 | /* Start of environment variable setup for hooks. */ |
24c2a48d CB |
739 | ret = setenv("LXC_NAME", name, 1); |
740 | if (ret < 0) | |
741 | SYSERROR("Failed to set environment variable: LXC_NAME=%s", name); | |
408da065 | 742 | |
2170b263 CB |
743 | if (conf->rcfile) { |
744 | ret = setenv("LXC_CONFIG_FILE", conf->rcfile, 1); | |
745 | if (ret < 0) | |
818a57fc | 746 | SYSERROR("Failed to set environment variable: LXC_CONFIG_FILE=%s", conf->rcfile); |
2170b263 | 747 | } |
408da065 | 748 | |
2170b263 CB |
749 | if (conf->rootfs.mount) { |
750 | ret = setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1); | |
751 | if (ret < 0) | |
818a57fc | 752 | SYSERROR("Failed to set environment variable: LXC_ROOTFS_MOUNT=%s", conf->rootfs.mount); |
2170b263 | 753 | } |
408da065 | 754 | |
2170b263 CB |
755 | if (conf->rootfs.path) { |
756 | ret = setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1); | |
757 | if (ret < 0) | |
818a57fc | 758 | SYSERROR("Failed to set environment variable: LXC_ROOTFS_PATH=%s", conf->rootfs.path); |
2170b263 | 759 | } |
408da065 | 760 | |
2170b263 CB |
761 | if (conf->console.path) { |
762 | ret = setenv("LXC_CONSOLE", conf->console.path, 1); | |
763 | if (ret < 0) | |
818a57fc | 764 | SYSERROR("Failed to set environment variable: LXC_CONSOLE=%s", conf->console.path); |
2170b263 | 765 | } |
408da065 | 766 | |
2170b263 CB |
767 | if (conf->console.log_path) { |
768 | ret = setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1); | |
769 | if (ret < 0) | |
818a57fc | 770 | SYSERROR("Failed to set environment variable: LXC_CONSOLE_LOGPATH=%s", conf->console.log_path); |
2170b263 | 771 | } |
408da065 | 772 | |
2170b263 CB |
773 | if (cgns_supported()) { |
774 | ret = setenv("LXC_CGNS_AWARE", "1", 1); | |
775 | if (ret < 0) | |
818a57fc | 776 | SYSERROR("Failed to set environment variable LXC_CGNS_AWARE=1"); |
2170b263 | 777 | } |
b8f88d9b | 778 | |
4a03ded4 | 779 | loglevel = lxc_log_priority_to_string(lxc_log_get_level()); |
2170b263 CB |
780 | ret = setenv("LXC_LOG_LEVEL", loglevel, 1); |
781 | if (ret < 0) | |
818a57fc | 782 | SYSERROR("Set environment variable LXC_LOG_LEVEL=%s", loglevel); |
a2c09be0 CB |
783 | |
784 | if (conf->hooks_version == 0) | |
785 | ret = setenv("LXC_HOOK_VERSION", "0", 1); | |
786 | else | |
787 | ret = setenv("LXC_HOOK_VERSION", "1", 1); | |
788 | if (ret < 0) | |
789 | SYSERROR("Failed to set environment variable LXC_HOOK_VERSION=%u", conf->hooks_version); | |
408da065 | 790 | /* End of environment variable setup for hooks. */ |
f7bee6c6 | 791 | |
2170b263 | 792 | TRACE("Set environment variables"); |
dbc9832d | 793 | |
2170b263 | 794 | ret = run_lxc_hooks(name, "pre-start", conf, NULL); |
65a2dc8b CB |
795 | if (ret < 0) |
796 | return log_error(-1, "Failed to run lxc.hook.pre-start for container \"%s\"", name); | |
2170b263 | 797 | TRACE("Ran pre-start hooks"); |
26ddeedd | 798 | |
408da065 CB |
799 | /* The signal fd has to be created before forking otherwise if the child |
800 | * process exits before we setup the signal fd, the event will be lost | |
801 | * and the command will be stuck. | |
802 | */ | |
83ee7875 | 803 | handler->sigfd = setup_signal_fd(&handler->oldmask); |
65a2dc8b CB |
804 | if (handler->sigfd < 0) |
805 | return log_error(-1, "Failed to setup SIGCHLD fd handler."); | |
2170b263 | 806 | TRACE("Set up signal fd"); |
b5159817 | 807 | |
408da065 | 808 | /* Do this after setting up signals since it might unblock SIGWINCH. */ |
564e31c4 | 809 | ret = lxc_terminal_setup(conf); |
4d1ffb0a CB |
810 | if (ret < 0) { |
811 | ERROR("Failed to create console"); | |
b5159817 | 812 | goto out_restore_sigmask; |
b0a33c1e | 813 | } |
4d1ffb0a | 814 | TRACE("Created console"); |
b0a33c1e | 815 | |
5a087e05 | 816 | handler->cgroup_ops = cgroup_init(handler->conf); |
2202afc9 CB |
817 | if (!handler->cgroup_ops) { |
818 | ERROR("Failed to initialize cgroup driver"); | |
5b74eb3c | 819 | goto out_delete_terminal; |
2202afc9 CB |
820 | } |
821 | TRACE("Initialized cgroup driver"); | |
822 | ||
2e1361a6 CB |
823 | ret = lxc_read_seccomp_config(conf); |
824 | if (ret < 0) | |
825 | return log_error(-1, "Failed loading seccomp policy"); | |
826 | TRACE("Read seccomp policy"); | |
827 | ||
af04d847 | 828 | ret = handler->lsm_ops->prepare(handler->lsm_ops, conf, handler->lxcpath); |
1800f924 WB |
829 | if (ret < 0) { |
830 | ERROR("Failed to initialize LSM"); | |
c581d2a6 | 831 | goto out_delete_terminal; |
1800f924 WB |
832 | } |
833 | TRACE("Initialized LSM"); | |
834 | ||
2170b263 | 835 | INFO("Container \"%s\" is initialized", name); |
4d8bdfa0 | 836 | handler->monitor_status_fd = move_fd(status_fd); |
aa460476 | 837 | return 0; |
59eb99ba | 838 | |
5b74eb3c CB |
839 | out_delete_terminal: |
840 | lxc_terminal_delete(&handler->conf->console); | |
ea918412 | 841 | |
b5159817 | 842 | out_restore_sigmask: |
4c496daa | 843 | (void)pthread_sigmask(SIG_SETMASK, &handler->oldmask, NULL); |
ea918412 | 844 | |
aa460476 | 845 | return -1; |
59eb99ba DL |
846 | } |
847 | ||
fd5be714 | 848 | void lxc_end(struct lxc_handler *handler) |
59eb99ba | 849 | { |
818a57fc | 850 | int ret; |
18b3b9c1 | 851 | pid_t self; |
dbc9832d | 852 | struct lxc_list *cur, *next; |
dbc9832d | 853 | char *namespaces[LXC_NS_MAX + 1]; |
b3286b62 | 854 | size_t namespace_count = 0; |
0c5859ff | 855 | const char *name = handler->name; |
2202afc9 | 856 | struct cgroup_ops *cgroup_ops = handler->cgroup_ops; |
b6b2b194 | 857 | |
408da065 CB |
858 | /* The STOPPING state is there for future cleanup code which can take |
859 | * awhile. | |
59eb99ba | 860 | */ |
25c2aca5 | 861 | lxc_set_state(name, handler, STOPPING); |
b6b2b194 | 862 | |
0059379f | 863 | self = lxc_raw_getpid(); |
818a57fc | 864 | for (int i = 0; i < LXC_NS_MAX; i++) { |
18b3b9c1 CB |
865 | if (handler->nsfd[i] < 0) |
866 | continue; | |
867 | ||
868 | if (handler->conf->hooks_version == 0) | |
c097467f | 869 | ret = asprintf(&namespaces[namespace_count], |
18b3b9c1 CB |
870 | "%s:/proc/%d/fd/%d", ns_info[i].proc_name, |
871 | self, handler->nsfd[i]); | |
872 | else | |
c097467f | 873 | ret = asprintf(&namespaces[namespace_count], |
18b3b9c1 | 874 | "/proc/%d/fd/%d", self, handler->nsfd[i]); |
818a57fc | 875 | if (ret < 0) { |
c097467f | 876 | SYSERROR("Failed to allocate memory"); |
18b3b9c1 | 877 | break; |
b3286b62 | 878 | } |
18b3b9c1 CB |
879 | |
880 | if (handler->conf->hooks_version == 0) { | |
881 | namespace_count++; | |
882 | continue; | |
883 | } | |
884 | ||
c097467f CB |
885 | ret = setenv(ns_info[i].env_name, namespaces[namespace_count], 1); |
886 | if (ret < 0) | |
18b3b9c1 CB |
887 | SYSERROR("Failed to set environment variable %s=%s", |
888 | ns_info[i].env_name, namespaces[namespace_count]); | |
889 | else | |
890 | TRACE("Set environment variable %s=%s", | |
891 | ns_info[i].env_name, namespaces[namespace_count]); | |
892 | ||
893 | namespace_count++; | |
b3286b62 WB |
894 | } |
895 | namespaces[namespace_count] = NULL; | |
c154af98 | 896 | |
80308d07 | 897 | if (handler->conf->reboot > REBOOT_NONE) { |
c097467f CB |
898 | ret = setenv("LXC_TARGET", "reboot", 1); |
899 | if (ret < 0) | |
818a57fc | 900 | SYSERROR("Failed to set environment variable: LXC_TARGET=reboot"); |
c097467f | 901 | } |
408da065 | 902 | |
80308d07 | 903 | if (handler->conf->reboot == REBOOT_NONE) { |
c097467f CB |
904 | ret = setenv("LXC_TARGET", "stop", 1); |
905 | if (ret < 0) | |
818a57fc | 906 | SYSERROR("Failed to set environment variable: LXC_TARGET=stop"); |
c097467f | 907 | } |
c154af98 | 908 | |
18b3b9c1 | 909 | if (handler->conf->hooks_version == 0) |
c097467f | 910 | ret = run_lxc_hooks(name, "stop", handler->conf, namespaces); |
18b3b9c1 | 911 | else |
c097467f CB |
912 | ret = run_lxc_hooks(name, "stop", handler->conf, NULL); |
913 | if (ret < 0) | |
914 | ERROR("Failed to run \"lxc.hook.stop\" hook"); | |
c154af98 | 915 | |
b3286b62 WB |
916 | while (namespace_count--) |
917 | free(namespaces[namespace_count]); | |
28d9e29e | 918 | |
af04d847 | 919 | handler->lsm_ops->cleanup(handler->lsm_ops, handler->conf, handler->lxcpath); |
1800f924 | 920 | |
e2aed383 WB |
921 | if (cgroup_ops) { |
922 | cgroup_ops->payload_destroy(cgroup_ops, handler); | |
923 | cgroup_ops->monitor_destroy(cgroup_ops, handler); | |
924 | } | |
4288b79f | 925 | |
79ff643d CB |
926 | put_lxc_rootfs(&handler->conf->rootfs, true); |
927 | ||
80308d07 | 928 | if (handler->conf->reboot == REBOOT_NONE) { |
f893d898 CB |
929 | /* For all new state clients simply close the command socket. |
930 | * This will inform all state clients that the container is | |
931 | * STOPPED and also prevents a race between a open()/close() on | |
932 | * the command socket causing a new process to get ECONNREFUSED | |
933 | * because we haven't yet closed the command socket. | |
934 | */ | |
b87ee312 | 935 | close_prot_errno_disarm(handler->conf->maincmd_fd); |
c3184275 CB |
936 | TRACE("Closed command socket"); |
937 | ||
938 | /* This function will try to connect to the legacy lxc-monitord | |
939 | * state server and only exists for backwards compatibility. | |
940 | */ | |
941 | lxc_monitor_send_state(name, STOPPED, handler->lxcpath); | |
942 | ||
943 | /* The command socket is closed so no one can acces the command | |
944 | * socket anymore so there's no need to lock it. | |
945 | */ | |
946 | handler->state = STOPPED; | |
947 | TRACE("Set container state to \"STOPPED\""); | |
f893d898 CB |
948 | } else { |
949 | lxc_set_state(name, handler, STOPPED); | |
695d0e56 | 950 | TRACE("Set container state to \"STOPPED\""); |
d39b10eb | 951 | } |
4288b79f | 952 | |
048493a3 CB |
953 | /* Avoid lingering namespace references. */ |
954 | lxc_put_nsfds(handler); | |
955 | ||
46800e77 CB |
956 | ret = run_lxc_hooks(name, "post-stop", handler->conf, NULL); |
957 | if (ret < 0) { | |
c097467f | 958 | ERROR("Failed to run lxc.hook.post-stop for container \"%s\"", name); |
80308d07 | 959 | if (handler->conf->reboot > REBOOT_NONE) { |
c097467f | 960 | WARN("Container will be stopped instead of rebooted"); |
80308d07 | 961 | handler->conf->reboot = REBOOT_NONE; |
c097467f CB |
962 | |
963 | ret = setenv("LXC_TARGET", "stop", 1); | |
964 | if (ret < 0) | |
818a57fc | 965 | WARN("Failed to set environment variable: LXC_TARGET=stop"); |
f3787121 CB |
966 | } |
967 | } | |
26ddeedd | 968 | |
408da065 | 969 | /* Reset mask set by setup_signal_fd. */ |
b467714b | 970 | ret = pthread_sigmask(SIG_SETMASK, &handler->oldmask, NULL); |
c097467f | 971 | if (ret < 0) |
a24c5678 | 972 | SYSWARN("Failed to restore signal mask"); |
8f64a3f6 | 973 | |
2aac071b | 974 | lxc_terminal_delete(&handler->conf->console); |
0e4be3cf | 975 | lxc_delete_tty(&handler->conf->ttys); |
f797f05e | 976 | close_prot_errno_disarm(handler->conf->devpts_fd); |
dbc9832d | 977 | |
dbc9832d CB |
978 | /* The command socket is now closed, no more state clients can register |
979 | * themselves from now on. So free the list of state clients. | |
980 | */ | |
d39b10eb CB |
981 | lxc_list_for_each_safe(cur, &handler->conf->state_clients, next) { |
982 | struct lxc_state_client *client = cur->elem; | |
983 | ||
984 | /* Keep state clients that want to be notified about reboots. */ | |
80308d07 CB |
985 | if ((handler->conf->reboot > REBOOT_NONE) && |
986 | (client->states[RUNNING] == 2)) | |
d39b10eb CB |
987 | continue; |
988 | ||
dbc9832d | 989 | /* close state client socket */ |
dbc9832d | 990 | lxc_list_del(cur); |
c097467f | 991 | close(client->clientfd); |
dbc9832d CB |
992 | free(cur->elem); |
993 | free(cur); | |
994 | } | |
995 | ||
80308d07 | 996 | if (handler->conf->ephemeral == 1 && handler->conf->reboot != REBOOT_REQ) |
28272964 | 997 | lxc_destroy_container_on_signal(handler, name); |
2c5f2ede | 998 | |
a42abcce | 999 | lxc_put_handler(handler); |
59eb99ba DL |
1000 | } |
1001 | ||
0c5859ff | 1002 | void lxc_abort(struct lxc_handler *handler) |
59eb99ba | 1003 | { |
33942046 CB |
1004 | int ret = 0; |
1005 | int status; | |
73e608b2 | 1006 | |
0c5859ff | 1007 | lxc_set_state(handler->name, handler, ABORTING); |
0e4f9d51 | 1008 | |
c718fac1 | 1009 | if (handler->pidfd >= 0) { |
33942046 | 1010 | ret = lxc_raw_pidfd_send_signal(handler->pidfd, SIGKILL, NULL, 0); |
c718fac1 | 1011 | if (ret) |
11c7d349 CB |
1012 | SYSWARN("Failed to send SIGKILL via pidfd %d for process %d", |
1013 | handler->pidfd, handler->pid); | |
c718fac1 CB |
1014 | } |
1015 | ||
65a2dc8b | 1016 | if ((!ret || errno != ESRCH) && handler->pid > 0) |
11c7d349 CB |
1017 | if (kill(handler->pid, SIGKILL)) |
1018 | SYSWARN("Failed to send SIGKILL to %d", handler->pid); | |
0e4f9d51 | 1019 | |
33942046 CB |
1020 | do { |
1021 | ret = waitpid(-1, &status, 0); | |
1022 | } while (ret > 0); | |
59eb99ba DL |
1023 | } |
1024 | ||
ffe1e01a | 1025 | static int do_start(void *data) |
50e98013 | 1026 | { |
cdb2a47f | 1027 | struct lxc_handler *handler = data; |
f62cf1d4 | 1028 | __lxc_unused __do_close int data_sock0 = handler->data_sock[0], |
9ff57a59 | 1029 | data_sock1 = handler->data_sock[1]; |
d3103162 | 1030 | __do_close int devnull_fd = -EBADF, status_fd = -EBADF; |
8bf3abfb | 1031 | int ret; |
928b1f04 YT |
1032 | uid_t new_uid; |
1033 | gid_t new_gid; | |
e96a536c | 1034 | struct lxc_list *iterator; |
964581c2 CB |
1035 | uid_t nsuid = 0; |
1036 | gid_t nsgid = 0; | |
50e98013 | 1037 | |
18225d19 | 1038 | lxc_sync_fini_parent(handler); |
50e98013 | 1039 | |
9ff57a59 | 1040 | if (lxc_abstract_unix_recv_fds(data_sock1, &status_fd, 1, NULL, 0) < 0) { |
4d8bdfa0 CB |
1041 | ERROR("Failed to receive status file descriptor to child process"); |
1042 | goto out_warn_father; | |
1043 | } | |
1044 | ||
408da065 CB |
1045 | /* This prctl must be before the synchro, so if the parent dies before |
1046 | * we set the parent death signal, we will detect its death with the | |
1047 | * synchro right after, otherwise we have a window where the parent can | |
1048 | * exit before we set the pdeath signal leading to a unsupervized | |
1049 | * container. | |
743ecd2e | 1050 | */ |
4d8bdfa0 | 1051 | ret = lxc_set_death_signal(SIGKILL, handler->monitor_pid, status_fd); |
912314fc CB |
1052 | if (ret < 0) { |
1053 | SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL"); | |
18225d19 | 1054 | goto out_warn_father; |
743ecd2e DL |
1055 | } |
1056 | ||
611ddd34 CB |
1057 | ret = lxc_ambient_caps_up(); |
1058 | if (ret < 0) { | |
6f5e532f | 1059 | ERROR("Failed to raise ambient capabilities"); |
611ddd34 CB |
1060 | goto out_warn_father; |
1061 | } | |
1062 | ||
b467714b | 1063 | ret = pthread_sigmask(SIG_SETMASK, &handler->oldmask, NULL); |
18225d19 CB |
1064 | if (ret < 0) { |
1065 | SYSERROR("Failed to set signal mask"); | |
1066 | goto out_warn_father; | |
1067 | } | |
50e98013 | 1068 | |
6e48e7c5 | 1069 | if (!lxc_sync_wait_parent(handler, START_SYNC_STARTUP)) |
18225d19 | 1070 | goto out_warn_father; |
5b1e83cb | 1071 | |
408da065 CB |
1072 | /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See |
1073 | * https://github.com/lxc/lxd/issues/1978. | |
1074 | */ | |
f7176c3e | 1075 | if (handler->ns_unshare_flags & CLONE_NEWNET) { |
5b1e83cb SH |
1076 | ret = unshare(CLONE_NEWNET); |
1077 | if (ret < 0) { | |
e96a536c | 1078 | SYSERROR("Failed to unshare CLONE_NEWNET"); |
5b1e83cb SH |
1079 | goto out_warn_father; |
1080 | } | |
e96a536c | 1081 | INFO("Unshared CLONE_NEWNET"); |
5b1e83cb SH |
1082 | } |
1083 | ||
408da065 CB |
1084 | /* Tell the parent task it can begin to configure the container and wait |
1085 | * for it to finish. | |
3c22086f | 1086 | */ |
6e48e7c5 | 1087 | if (!lxc_sync_barrier_parent(handler, START_SYNC_CONFIGURE)) |
611ddd34 | 1088 | goto out_error; |
50e98013 | 1089 | |
e389f2af | 1090 | if (handler->ns_clone_flags & CLONE_NEWNET) { |
3c09b97c | 1091 | ret = lxc_network_recv_from_parent(handler); |
e389f2af CB |
1092 | if (ret < 0) { |
1093 | ERROR("Failed to receive veth names from parent"); | |
1094 | goto out_warn_father; | |
1095 | } | |
7ab1ba02 | 1096 | } |
658979c5 | 1097 | |
408da065 | 1098 | /* If we are in a new user namespace, become root there to have |
d08f8d2f | 1099 | * privilege over our namespace. |
f6d3e3e4 SH |
1100 | */ |
1101 | if (!lxc_list_empty(&handler->conf->id_map)) { | |
964581c2 CB |
1102 | if (!handler->conf->root_nsuid_map) |
1103 | nsuid = handler->conf->init_uid; | |
1104 | ||
1105 | if (!handler->conf->root_nsgid_map) | |
1106 | nsgid = handler->conf->init_gid; | |
4160c3a0 | 1107 | |
d08f8d2f CB |
1108 | /* Drop groups only after we switched to a valid gid in the new |
1109 | * user namespace. | |
1110 | */ | |
8917c382 | 1111 | if (!lxc_drop_groups() && |
8af07f82 | 1112 | (handler->am_root || errno != EPERM)) |
c476bdce | 1113 | goto out_warn_father; |
4b826b1f | 1114 | |
b58214ac CB |
1115 | if (!lxc_switch_uid_gid(nsuid, nsgid)) |
1116 | goto out_warn_father; | |
1117 | ||
b81689a1 CB |
1118 | ret = prctl(PR_SET_DUMPABLE, prctl_arg(1), prctl_arg(0), |
1119 | prctl_arg(0), prctl_arg(0)); | |
d7883725 CB |
1120 | if (ret < 0) |
1121 | goto out_warn_father; | |
912314fc CB |
1122 | |
1123 | /* set{g,u}id() clears deathsignal */ | |
4d8bdfa0 | 1124 | ret = lxc_set_death_signal(SIGKILL, handler->monitor_pid, status_fd); |
912314fc CB |
1125 | if (ret < 0) { |
1126 | SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL"); | |
1127 | goto out_warn_father; | |
1128 | } | |
f6d3e3e4 SH |
1129 | } |
1130 | ||
e96a536c CB |
1131 | ret = access(handler->lxcpath, X_OK); |
1132 | if (ret != 0) { | |
c8154066 SH |
1133 | print_top_failing_dir(handler->lxcpath); |
1134 | goto out_warn_father; | |
1135 | } | |
1136 | ||
7a55c157 TA |
1137 | /* In order to checkpoint restore, we need to have everything in the |
1138 | * same mount namespace. However, some containers may not have a | |
1139 | * reasonable /dev (in particular, they may not have /dev/null), so we | |
1140 | * can't set init's std fds to /dev/null by opening it from inside the | |
1141 | * container. | |
1142 | * | |
1143 | * If that's the case, fall back to using the host's /dev/null. This | |
1144 | * means that migration won't work, but at least we won't spew output | |
1145 | * where it isn't wanted. | |
1146 | */ | |
bb955810 | 1147 | if (handler->daemonize && !handler->conf->autodev) { |
2806a87d | 1148 | char path[PATH_MAX]; |
3c09b97c | 1149 | |
fa60cd7b CB |
1150 | ret = strnprintf(path, sizeof(path), "%s/dev/null", |
1151 | handler->conf->rootfs.mount); | |
1152 | if (ret < 0) | |
2806a87d | 1153 | goto out_warn_father; |
3c09b97c | 1154 | |
e96a536c CB |
1155 | ret = access(path, F_OK); |
1156 | if (ret != 0) { | |
1157 | devnull_fd = open_devnull(); | |
c44de748 | 1158 | |
e96a536c CB |
1159 | if (devnull_fd < 0) |
1160 | goto out_warn_father; | |
818a57fc | 1161 | WARN("Using /dev/null from the host for container init's standard file descriptors. Migration will not work"); |
e96a536c | 1162 | } |
c44de748 AM |
1163 | } |
1164 | ||
408da065 | 1165 | /* Ask father to setup cgroups and wait for him to finish. */ |
6e48e7c5 | 1166 | if (!lxc_sync_barrier_parent(handler, START_SYNC_CGROUP)) |
c44de748 | 1167 | goto out_error; |
544a48a0 | 1168 | |
deefdf8a CB |
1169 | /* Unshare cgroup namespace after we have setup our cgroups. If we do it |
1170 | * earlier we end up with a wrong view of /proc/self/cgroup. For | |
1171 | * example, assume we unshare(CLONE_NEWCGROUP) first, and then create | |
1172 | * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then | |
1173 | * /proc/self/cgroup would show us: | |
1174 | * | |
1175 | * 8:cpuset:/lxc/c | |
1176 | * | |
1177 | * whereas it should actually show | |
1178 | * | |
1179 | * 8:cpuset:/ | |
1180 | */ | |
f7176c3e | 1181 | if (handler->ns_unshare_flags & CLONE_NEWCGROUP) { |
e96a536c CB |
1182 | ret = unshare(CLONE_NEWCGROUP); |
1183 | if (ret < 0) { | |
bca7c59c CB |
1184 | if (errno != EINVAL) { |
1185 | SYSERROR("Failed to unshare CLONE_NEWCGROUP"); | |
1186 | goto out_warn_father; | |
1187 | } | |
1188 | ||
1189 | handler->ns_clone_flags &= ~CLONE_NEWCGROUP; | |
1190 | SYSINFO("Kernel does not support CLONE_NEWCGROUP"); | |
1191 | } else { | |
1192 | INFO("Unshared CLONE_NEWCGROUP"); | |
deefdf8a | 1193 | } |
deefdf8a CB |
1194 | } |
1195 | ||
f7176c3e | 1196 | if (handler->ns_unshare_flags & CLONE_NEWTIME) { |
70fd7fc9 CB |
1197 | ret = unshare(CLONE_NEWTIME); |
1198 | if (ret < 0) { | |
1199 | if (errno != EINVAL) { | |
1200 | SYSERROR("Failed to unshare CLONE_NEWTIME"); | |
1201 | goto out_warn_father; | |
1202 | } | |
1203 | ||
1204 | handler->ns_clone_flags &= ~CLONE_NEWTIME; | |
1205 | SYSINFO("Kernel does not support CLONE_NEWTIME"); | |
1206 | } else { | |
1207 | __do_close int timens_fd = -EBADF; | |
1208 | ||
1209 | INFO("Unshared CLONE_NEWTIME"); | |
1210 | ||
1211 | if (handler->conf->timens.s_boot) | |
1212 | ret = timens_offset_write(CLOCK_BOOTTIME, handler->conf->timens.s_boot, 0); | |
1213 | else if (handler->conf->timens.ns_boot) | |
1214 | ret = timens_offset_write(CLOCK_BOOTTIME, 0, handler->conf->timens.ns_boot); | |
1215 | if (ret) { | |
1216 | SYSERROR("Failed to write CLONE_BOOTTIME offset"); | |
1217 | goto out_warn_father; | |
1218 | } | |
1219 | TRACE("Wrote CLOCK_BOOTTIME offset"); | |
1220 | ||
1221 | if (handler->conf->timens.s_monotonic) | |
1222 | ret = timens_offset_write(CLOCK_MONOTONIC, handler->conf->timens.s_monotonic, 0); | |
1223 | else if (handler->conf->timens.ns_monotonic) | |
1224 | ret = timens_offset_write(CLOCK_MONOTONIC, 0, handler->conf->timens.ns_monotonic); | |
1225 | if (ret) { | |
1226 | SYSERROR("Failed to write CLONE_MONOTONIC offset"); | |
1227 | goto out_warn_father; | |
1228 | } | |
1229 | TRACE("Wrote CLOCK_MONOTONIC offset"); | |
1230 | ||
1231 | timens_fd = open("/proc/self/ns/time_for_children", O_RDONLY | O_CLOEXEC); | |
1232 | if (timens_fd < 0) { | |
1233 | SYSERROR("Failed to open \"/proc/self/ns/time_for_children\""); | |
1234 | goto out_warn_father; | |
1235 | } | |
1236 | ||
1237 | ret = setns(timens_fd, CLONE_NEWTIME); | |
1238 | if (ret) { | |
1239 | SYSERROR("Failed to setns(%d(\"/proc/self/ns/time_for_children\"))", timens_fd); | |
1240 | goto out_warn_father; | |
1241 | } | |
1242 | } | |
1243 | } | |
1244 | ||
149857af CB |
1245 | /* Add the requested environment variables to the current environment to |
1246 | * allow them to be used by the various hooks, such as the start hook | |
317494f1 | 1247 | * below. |
98ff08ca | 1248 | */ |
98ff08ca | 1249 | lxc_list_for_each(iterator, &handler->conf->environment) { |
e96a536c CB |
1250 | ret = putenv((char *)iterator->elem); |
1251 | if (ret < 0) { | |
1252 | SYSERROR("Failed to set environment variable: %s", | |
1253 | (char *)iterator->elem); | |
98ff08ca CB |
1254 | goto out_warn_father; |
1255 | } | |
1256 | } | |
1257 | ||
f4152036 | 1258 | /* Setup the container, ip, names, utsname, ... */ |
3b988b33 | 1259 | ret = lxc_setup(handler); |
7729f8e5 | 1260 | if (ret < 0) { |
e96a536c | 1261 | ERROR("Failed to setup container \"%s\"", handler->name); |
f4152036 CB |
1262 | goto out_warn_father; |
1263 | } | |
1264 | ||
408da065 | 1265 | /* Set the label to change to when we exec(2) the container's init. */ |
af04d847 | 1266 | ret = handler->lsm_ops->process_label_set(handler->lsm_ops, NULL, handler->conf, true); |
e96a536c | 1267 | if (ret < 0) |
e075f5d9 | 1268 | goto out_warn_father; |
5112cd70 | 1269 | |
029cdff5 | 1270 | /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it |
408da065 CB |
1271 | * before we aren't allowed anymore. |
1272 | */ | |
029cdff5 | 1273 | if (handler->conf->no_new_privs) { |
b81689a1 CB |
1274 | ret = prctl(PR_SET_NO_NEW_PRIVS, prctl_arg(1), prctl_arg(0), |
1275 | prctl_arg(0), prctl_arg(0)); | |
e96a536c | 1276 | if (ret < 0) { |
818a57fc | 1277 | SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges"); |
029cdff5 CB |
1278 | goto out_warn_father; |
1279 | } | |
818a57fc | 1280 | DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges"); |
029cdff5 CB |
1281 | } |
1282 | ||
0d9acb99 DE |
1283 | /* Some init's such as busybox will set sane tty settings on stdin, |
1284 | * stdout, stderr which it thinks is the console. We already set them | |
1285 | * the way we wanted on the real terminal, and we want init to do its | |
564e31c4 | 1286 | * setup on its console ie. the pty allocated in lxc_terminal_setup() so |
408da065 | 1287 | * make sure that that pty is stdin,stdout,stderr. |
0d9acb99 | 1288 | */ |
41808e20 | 1289 | if (handler->conf->console.pty >= 0) { |
bb955810 | 1290 | if (handler->daemonize || !handler->conf->is_execute) |
41808e20 | 1291 | ret = set_stdfds(handler->conf->console.pty); |
5d113f65 | 1292 | else |
41808e20 | 1293 | ret = lxc_terminal_set_stdfds(handler->conf->console.pty); |
5d113f65 | 1294 | if (ret < 0) { |
818a57fc | 1295 | ERROR("Failed to redirect std{in,out,err} to pty file descriptor %d", |
41808e20 | 1296 | handler->conf->console.pty); |
c5b93afb LF |
1297 | goto out_warn_father; |
1298 | } | |
5d113f65 | 1299 | } |
0d9acb99 | 1300 | |
408da065 | 1301 | /* If we mounted a temporary proc, then unmount it now. */ |
5112cd70 | 1302 | tmp_proc_unmount(handler->conf); |
e075f5d9 | 1303 | |
e96a536c CB |
1304 | ret = lxc_seccomp_load(handler->conf); |
1305 | if (ret < 0) | |
8f2c3a70 SH |
1306 | goto out_warn_father; |
1307 | ||
c3e3c21a CB |
1308 | ret = lxc_seccomp_send_notifier_fd(&handler->conf->seccomp, data_sock0); |
1309 | if (ret < 0) { | |
1310 | SYSERROR("Failed to send seccomp notify fd to parent"); | |
1311 | goto out_warn_father; | |
cdb2a47f | 1312 | } |
cdb2a47f | 1313 | |
e96a536c CB |
1314 | ret = run_lxc_hooks(handler->name, "start", handler->conf, NULL); |
1315 | if (ret < 0) { | |
1316 | ERROR("Failed to run lxc.hook.start for container \"%s\"", | |
1317 | handler->name); | |
773fb9ca SH |
1318 | goto out_warn_father; |
1319 | } | |
fc25b815 | 1320 | |
a783a414 | 1321 | close_prot_errno_disarm(handler->sigfd); |
26ddeedd | 1322 | |
41808e20 | 1323 | if (handler->conf->console.pty < 0 && handler->daemonize) { |
f4c177c3 CB |
1324 | if (devnull_fd < 0) { |
1325 | devnull_fd = open_devnull(); | |
1326 | if (devnull_fd < 0) | |
1327 | goto out_warn_father; | |
1328 | } | |
1329 | ||
e96a536c CB |
1330 | ret = set_stdfds(devnull_fd); |
1331 | if (ret < 0) { | |
1332 | ERROR("Failed to redirect std{in,out,err} to \"/dev/null\""); | |
c5b93afb LF |
1333 | goto out_warn_father; |
1334 | } | |
e96a536c | 1335 | } |
507cee36 | 1336 | |
a783a414 | 1337 | close_prot_errno_disarm(devnull_fd); |
c44de748 | 1338 | |
8c9a7665 TA |
1339 | setsid(); |
1340 | ||
e96a536c CB |
1341 | if (handler->conf->init_cwd) { |
1342 | ret = chdir(handler->conf->init_cwd); | |
1343 | if (ret < 0) { | |
1344 | SYSERROR("Could not change directory to \"%s\"", | |
1345 | handler->conf->init_cwd); | |
1346 | goto out_warn_father; | |
1347 | } | |
3c491553 L |
1348 | } |
1349 | ||
6e48e7c5 | 1350 | if (!lxc_sync_barrier_parent(handler, START_SYNC_CGROUP_LIMITS)) |
f4152036 CB |
1351 | goto out_warn_father; |
1352 | ||
149857af CB |
1353 | /* Reset the environment variables the user requested in a clear |
1354 | * environment. | |
1355 | */ | |
e96a536c CB |
1356 | ret = clearenv(); |
1357 | /* Don't error out though. */ | |
1358 | if (ret < 0) | |
149857af | 1359 | SYSERROR("Failed to clear environment."); |
149857af CB |
1360 | |
1361 | lxc_list_for_each(iterator, &handler->conf->environment) { | |
e96a536c CB |
1362 | ret = putenv((char *)iterator->elem); |
1363 | if (ret < 0) { | |
1364 | SYSERROR("Failed to set environment variable: %s", | |
1365 | (char *)iterator->elem); | |
149857af CB |
1366 | goto out_warn_father; |
1367 | } | |
1368 | } | |
1369 | ||
e96a536c CB |
1370 | ret = putenv("container=lxc"); |
1371 | if (ret < 0) { | |
1372 | SYSERROR("Failed to set environment variable: container=lxc"); | |
98ff08ca CB |
1373 | goto out_warn_father; |
1374 | } | |
1375 | ||
885766f5 CB |
1376 | if (handler->conf->ttys.tty_names) { |
1377 | ret = putenv(handler->conf->ttys.tty_names); | |
e96a536c CB |
1378 | if (ret < 0) { |
1379 | SYSERROR("Failed to set environment variable for container ptys"); | |
98ff08ca CB |
1380 | goto out_warn_father; |
1381 | } | |
1382 | } | |
1383 | ||
76bdf299 CB |
1384 | /* The container has been setup. We can now switch to an unprivileged |
1385 | * uid/gid. | |
1386 | */ | |
928b1f04 YT |
1387 | new_uid = handler->conf->init_uid; |
1388 | new_gid = handler->conf->init_gid; | |
76bdf299 | 1389 | |
964581c2 CB |
1390 | /* Avoid unnecessary syscalls. */ |
1391 | if (new_uid == nsuid) | |
1392 | new_uid = LXC_INVALID_UID; | |
1393 | ||
1394 | if (new_gid == nsgid) | |
1395 | new_gid = LXC_INVALID_GID; | |
c353b0b9 | 1396 | |
6aff5157 | 1397 | /* Make sure that the processes STDIO is correctly owned by the user that we are switching to */ |
c353b0b9 CB |
1398 | ret = fix_stdio_permissions(new_uid); |
1399 | if (ret) | |
1400 | WARN("Failed to ajust stdio permissions"); | |
964581c2 | 1401 | |
8af07f82 CB |
1402 | /* If we are in a new user namespace we already dropped all groups when |
1403 | * we switched to root in the new user namespace further above. Only | |
1404 | * drop groups if we can, so ensure that we have necessary privilege. | |
1405 | */ | |
bf31b337 | 1406 | if (lxc_list_empty(&handler->conf->id_map)) { |
8af07f82 CB |
1407 | #if HAVE_LIBCAP |
1408 | if (lxc_proc_cap_is_set(CAP_SETGID, CAP_EFFECTIVE)) | |
1409 | #endif | |
bf31b337 RJ |
1410 | { |
1411 | if (handler->conf->init_groups.size > 0) { | |
1412 | if (!lxc_setgroups(handler->conf->init_groups.list, | |
1413 | handler->conf->init_groups.size)) | |
1414 | goto out_warn_father; | |
1415 | } else { | |
1416 | if (!lxc_drop_groups()) | |
1417 | goto out_warn_father; | |
1418 | } | |
1419 | } | |
1420 | } | |
8af07f82 | 1421 | |
b58214ac CB |
1422 | if (!lxc_switch_uid_gid(new_uid, new_gid)) |
1423 | goto out_warn_father; | |
1424 | ||
611ddd34 CB |
1425 | ret = lxc_ambient_caps_down(); |
1426 | if (ret < 0) { | |
6f5e532f | 1427 | ERROR("Failed to clear ambient capabilities"); |
611ddd34 CB |
1428 | goto out_warn_father; |
1429 | } | |
1430 | ||
258f8051 | 1431 | if (handler->conf->monitor_signal_pdeath != SIGKILL) { |
4d8bdfa0 CB |
1432 | ret = lxc_set_death_signal(handler->conf->monitor_signal_pdeath, |
1433 | handler->monitor_pid, status_fd); | |
258f8051 CB |
1434 | if (ret < 0) { |
1435 | SYSERROR("Failed to set PR_SET_PDEATHSIG to %d", | |
1436 | handler->conf->monitor_signal_pdeath); | |
1437 | goto out_warn_father; | |
1438 | } | |
1439 | } | |
1440 | ||
d3103162 CB |
1441 | /* |
1442 | * After this call, we are in error because this ops should not return | |
408da065 CB |
1443 | * as it execs. |
1444 | */ | |
c4ea60df | 1445 | handler->ops->start(handler, handler->data); |
50e98013 DL |
1446 | |
1447 | out_warn_father: | |
d3103162 CB |
1448 | /* |
1449 | * We want the parent to know something went wrong, so we return a | |
408da065 CB |
1450 | * special error code. |
1451 | */ | |
946a0c6d | 1452 | lxc_sync_wake_parent(handler, SYNC_ERROR); |
c44de748 AM |
1453 | |
1454 | out_error: | |
50e98013 DL |
1455 | return -1; |
1456 | } | |
1457 | ||
ae467c54 | 1458 | static int lxc_recv_ttys_from_child(struct lxc_handler *handler) |
e8bd4e43 | 1459 | { |
ae467c54 | 1460 | int i; |
2520facd | 1461 | struct lxc_terminal_info *tty; |
ae467c54 | 1462 | int ret = -1; |
c6012571 | 1463 | int sock = handler->data_sock[1]; |
e8bd4e43 | 1464 | struct lxc_conf *conf = handler->conf; |
0e4be3cf | 1465 | struct lxc_tty_info *ttys = &conf->ttys; |
e8bd4e43 | 1466 | |
885766f5 | 1467 | if (!conf->ttys.max) |
e8bd4e43 SH |
1468 | return 0; |
1469 | ||
885766f5 | 1470 | ttys->tty = malloc(sizeof(*ttys->tty) * ttys->max); |
0e4be3cf | 1471 | if (!ttys->tty) |
e8bd4e43 | 1472 | return -1; |
e8bd4e43 | 1473 | |
885766f5 | 1474 | for (i = 0; i < conf->ttys.max; i++) { |
672c1e58 CB |
1475 | int ttyfds[2]; |
1476 | ||
1477 | ret = lxc_abstract_unix_recv_fds(sock, ttyfds, 2, NULL, 0); | |
1478 | if (ret < 0) | |
1479 | break; | |
ae467c54 | 1480 | |
0e4be3cf | 1481 | tty = &ttys->tty[i]; |
dd3de568 | 1482 | tty->busy = -1; |
36a94ce8 | 1483 | tty->ptx = ttyfds[0]; |
41808e20 CB |
1484 | tty->pty = ttyfds[1]; |
1485 | TRACE("Received pty with ptx fd %d and pty fd %d from child", tty->ptx, tty->pty); | |
e8bd4e43 | 1486 | } |
6d1400b5 | 1487 | |
ae467c54 | 1488 | if (ret < 0) |
6d1400b5 | 1489 | SYSERROR("Failed to receive %zu ttys from child", ttys->max); |
ae467c54 | 1490 | else |
885766f5 | 1491 | TRACE("Received %zu ttys from child", ttys->max); |
ae467c54 CB |
1492 | |
1493 | return ret; | |
e8bd4e43 SH |
1494 | } |
1495 | ||
5af9369b | 1496 | int resolve_clone_flags(struct lxc_handler *handler) |
f813849c | 1497 | { |
8bc8c715 CB |
1498 | int i; |
1499 | struct lxc_conf *conf = handler->conf; | |
70fd7fc9 CB |
1500 | bool wants_timens = conf->timens.s_boot || conf->timens.ns_boot || |
1501 | conf->timens.s_monotonic || conf->timens.ns_monotonic; | |
f813849c | 1502 | |
8bc8c715 | 1503 | for (i = 0; i < LXC_NS_MAX; i++) { |
f7a0c6ee CB |
1504 | if (conf->ns_keep) { |
1505 | if (!(conf->ns_keep & ns_info[i].clone_flag)) | |
becad0ec | 1506 | handler->ns_clone_flags |= ns_info[i].clone_flag; |
f7a0c6ee CB |
1507 | } else if (conf->ns_clone) { |
1508 | if ((conf->ns_clone & ns_info[i].clone_flag)) | |
becad0ec | 1509 | handler->ns_clone_flags |= ns_info[i].clone_flag; |
8bc8c715 CB |
1510 | } else { |
1511 | if (i == LXC_NS_USER && lxc_list_empty(&handler->conf->id_map)) | |
1512 | continue; | |
1513 | ||
1514 | if (i == LXC_NS_NET && lxc_requests_empty_network(handler)) | |
1515 | continue; | |
1516 | ||
7bd05339 CB |
1517 | if (i == LXC_NS_CGROUP && !cgns_supported()) |
1518 | continue; | |
1519 | ||
70fd7fc9 CB |
1520 | if (i == LXC_NS_TIME && !wants_timens) |
1521 | continue; | |
1522 | ||
becad0ec | 1523 | handler->ns_clone_flags |= ns_info[i].clone_flag; |
8bc8c715 | 1524 | } |
03df7ab5 | 1525 | |
8bc8c715 CB |
1526 | if (!conf->ns_share[i]) |
1527 | continue; | |
5af9369b | 1528 | |
becad0ec | 1529 | handler->ns_clone_flags &= ~ns_info[i].clone_flag; |
8bc8c715 | 1530 | TRACE("Sharing %s namespace", ns_info[i].proc_name); |
5af9369b CB |
1531 | } |
1532 | ||
70fd7fc9 CB |
1533 | if (wants_timens && (conf->ns_keep & ns_info[LXC_NS_TIME].clone_flag)) |
1534 | return log_trace_errno(-1, EINVAL, "Requested to keep time namespace while also specifying offsets"); | |
1535 | ||
f7176c3e CB |
1536 | /* Deal with namespaces that are unshared. */ |
1537 | if (handler->ns_clone_flags & CLONE_NEWTIME) | |
1538 | handler->ns_unshare_flags |= CLONE_NEWTIME; | |
1539 | ||
1540 | if (!pure_unified_layout(handler->cgroup_ops) && handler->ns_clone_flags & CLONE_NEWCGROUP) | |
1541 | handler->ns_unshare_flags |= CLONE_NEWCGROUP; | |
1542 | ||
1543 | if ((handler->ns_clone_flags & (CLONE_NEWNET | CLONE_NEWUSER)) == | |
1544 | (CLONE_NEWNET | CLONE_NEWUSER)) | |
1545 | handler->ns_unshare_flags |= CLONE_NEWNET; | |
1546 | ||
1547 | /* Deal with namespaces that are spawned. */ | |
1548 | handler->ns_on_clone_flags = handler->ns_clone_flags & ~handler->ns_unshare_flags; | |
1549 | ||
1550 | handler->clone_flags = handler->ns_on_clone_flags | CLONE_PIDFD; | |
1551 | ||
5af9369b | 1552 | return 0; |
f813849c TA |
1553 | } |
1554 | ||
8deca6c9 CB |
1555 | /* Note that this function is used with clone(CLONE_VM). Some glibc versions |
1556 | * used to reset the pid/tid to -1 when CLONE_VM was used without CLONE_THREAD. | |
1557 | * But since the memory between parent and child is shared on CLONE_VM this | |
1558 | * would invalidate the getpid() cache that glibc used to maintain and so | |
1559 | * getpid() in the child would return the parent's pid. This is all fixed in | |
1560 | * newer glibc versions where the getpid() cache is removed and the pid/tid is | |
1561 | * not reset anymore. | |
b35ddc5b | 1562 | * However, if for whatever reason you - dear committer - somehow need to get the |
8deca6c9 | 1563 | * pid of the dummy intermediate process for do_share_ns() you need to call |
0059379f CB |
1564 | * lxc_raw_getpid(). The next lxc_raw_clone() call does not employ CLONE_VM and |
1565 | * will be fine. | |
8deca6c9 CB |
1566 | */ |
1567 | static inline int do_share_ns(void *arg) | |
fa3a5b22 | 1568 | { |
8deca6c9 | 1569 | int i, flags, ret; |
fa3a5b22 CB |
1570 | struct lxc_handler *handler = arg; |
1571 | ||
8deca6c9 CB |
1572 | for (i = 0; i < LXC_NS_MAX; i++) { |
1573 | if (handler->nsfd[i] < 0) | |
1574 | continue; | |
fa3a5b22 | 1575 | |
8deca6c9 | 1576 | ret = setns(handler->nsfd[i], 0); |
c0b48eff CB |
1577 | if (ret < 0) { |
1578 | /* | |
1579 | * Note that joining a user and/or mount namespace | |
1580 | * requires the process is not multithreaded otherwise | |
1581 | * setns() will fail here. | |
1582 | */ | |
1583 | SYSERROR("Failed to inherit %s namespace", | |
1584 | ns_info[i].proc_name); | |
8deca6c9 | 1585 | return -1; |
c0b48eff | 1586 | } |
fa3a5b22 | 1587 | |
8deca6c9 | 1588 | DEBUG("Inherited %s namespace", ns_info[i].proc_name); |
fa3a5b22 CB |
1589 | } |
1590 | ||
becad0ec | 1591 | flags = handler->ns_on_clone_flags; |
8deca6c9 | 1592 | flags |= CLONE_PARENT; |
33942046 CB |
1593 | handler->pid = lxc_raw_clone_cb(do_start, handler, CLONE_PIDFD | flags, |
1594 | &handler->pidfd); | |
8deca6c9 | 1595 | if (handler->pid < 0) |
fa3a5b22 CB |
1596 | return -1; |
1597 | ||
8deca6c9 | 1598 | return 0; |
fa3a5b22 CB |
1599 | } |
1600 | ||
480588e6 CB |
1601 | /* lxc_spawn() performs crucial setup tasks and clone()s the new process which |
1602 | * exec()s the requested container binary. | |
1603 | * Note that lxc_spawn() runs in the parent namespaces. Any operations performed | |
1604 | * right here should be double checked if they'd pose a security risk. (For | |
1605 | * example, any {u}mount() operations performed here will be reflected on the | |
1606 | * host!) | |
1607 | */ | |
74a3920a | 1608 | static int lxc_spawn(struct lxc_handler *handler) |
59eb99ba | 1609 | { |
f62cf1d4 | 1610 | __do_close int data_sock0 = -EBADF, data_sock1 = -EBADF; |
8deca6c9 | 1611 | int i, ret; |
08dd2805 | 1612 | char pidstr[20]; |
57927bf2 CB |
1613 | bool wants_to_map_ids; |
1614 | struct lxc_list *id_map; | |
6e5fc7a5 | 1615 | const char *name = handler->name; |
28d9e29e | 1616 | const char *lxcpath = handler->lxcpath; |
2202afc9 | 1617 | bool share_ns = false; |
28d9e29e | 1618 | struct lxc_conf *conf = handler->conf; |
2202afc9 | 1619 | struct cgroup_ops *cgroup_ops = handler->cgroup_ops; |
9f30a190 | 1620 | |
28d9e29e | 1621 | id_map = &conf->id_map; |
57927bf2 | 1622 | wants_to_map_ids = !lxc_list_empty(id_map); |
2f2623ec | 1623 | |
28d9e29e | 1624 | for (i = 0; i < LXC_NS_MAX; i++) { |
b074bbf1 | 1625 | if (!conf->ns_share[i]) |
28d9e29e CB |
1626 | continue; |
1627 | ||
b074bbf1 | 1628 | handler->nsfd[i] = lxc_inherit_namespace(conf->ns_share[i], lxcpath, ns_info[i].proc_name); |
28d9e29e CB |
1629 | if (handler->nsfd[i] < 0) |
1630 | return -1; | |
fa3a5b22 | 1631 | |
8deca6c9 | 1632 | share_ns = true; |
28d9e29e | 1633 | } |
50e98013 | 1634 | |
6e48e7c5 | 1635 | if (!lxc_sync_init(handler)) |
9d7f9e52 | 1636 | return -1; |
0ad19a3f | 1637 | |
b9f522c5 | 1638 | ret = socketpair(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0, |
790255cf | 1639 | handler->data_sock); |
95ef0d7c CB |
1640 | if (ret < 0) |
1641 | goto out_sync_fini; | |
cdb2a47f CB |
1642 | data_sock0 = handler->data_sock[0]; |
1643 | data_sock1 = handler->data_sock[1]; | |
e8bd4e43 | 1644 | |
5af9369b | 1645 | ret = resolve_clone_flags(handler); |
95ef0d7c CB |
1646 | if (ret < 0) |
1647 | goto out_sync_fini; | |
9f30a190 | 1648 | |
03ca4af8 TP |
1649 | if (handler->ns_clone_flags & CLONE_NEWNET) { |
1650 | ret = lxc_find_gateway_addresses(handler); | |
1651 | if (ret) { | |
1652 | ERROR("Failed to find gateway addresses"); | |
1653 | goto out_sync_fini; | |
1654 | } | |
1655 | } | |
1656 | ||
e8b181f5 | 1657 | if (!cgroup_ops->payload_create(cgroup_ops, handler)) { |
cfc62c60 | 1658 | ERROR("Failed creating cgroups"); |
47d8fb3b CS |
1659 | goto out_delete_net; |
1660 | } | |
1661 | ||
408da065 CB |
1662 | /* If the rootfs is not a blockdev, prevent the container from marking |
1663 | * it readonly. | |
1664 | * If the container is unprivileged then skip rootfs pinning. | |
0c547523 | 1665 | */ |
79ff643d CB |
1666 | ret = lxc_rootfs_prepare(&conf->rootfs, wants_to_map_ids); |
1667 | if (ret) { | |
1668 | ERROR("Failed to handle rootfs pinning for container \"%s\"", handler->name); | |
1669 | goto out_delete_net; | |
5e32a990 | 1670 | } |
0c547523 | 1671 | |
408da065 | 1672 | /* Create a process in a new set of namespaces. */ |
4e232466 CB |
1673 | if (share_ns) { |
1674 | pid_t attacher_pid; | |
1675 | ||
cfc62c60 | 1676 | attacher_pid = lxc_clone(do_share_ns, handler, |
33258b95 | 1677 | CLONE_VFORK | CLONE_VM | CLONE_FILES, NULL); |
4e232466 CB |
1678 | if (attacher_pid < 0) { |
1679 | SYSERROR(LXC_CLONE_ERROR); | |
1680 | goto out_delete_net; | |
1681 | } | |
1682 | ||
1683 | ret = wait_for_pid(attacher_pid); | |
1684 | if (ret < 0) { | |
1685 | SYSERROR("Intermediate process failed"); | |
1686 | goto out_delete_net; | |
1687 | } | |
f7176c3e CB |
1688 | |
1689 | if (handler->pid < 0) { | |
1690 | SYSERROR(LXC_CLONE_ERROR); | |
1691 | goto out_delete_net; | |
1692 | } | |
4e232466 | 1693 | } else { |
84b66ced | 1694 | int cgroup_fd = -EBADF; |
f7176c3e CB |
1695 | |
1696 | struct lxc_clone_args clone_args = { | |
1697 | .flags = handler->clone_flags, | |
1698 | .pidfd = ptr_to_u64(&handler->pidfd), | |
1699 | .exit_signal = SIGCHLD, | |
1700 | }; | |
1701 | ||
1702 | if (handler->ns_clone_flags & CLONE_NEWCGROUP) { | |
1703 | cgroup_fd = cgroup_unified_fd(cgroup_ops); | |
1704 | if (cgroup_fd >= 0) { | |
1705 | handler->clone_flags |= CLONE_INTO_CGROUP; | |
1706 | clone_args.flags |= CLONE_INTO_CGROUP; | |
1707 | clone_args.cgroup = cgroup_fd; | |
1708 | } | |
1709 | } | |
1710 | ||
1711 | /* Try to spawn directly into target cgroup. */ | |
1712 | handler->pid = lxc_clone3(&clone_args, CLONE_ARGS_SIZE_VER2); | |
1713 | if (handler->pid < 0) { | |
1714 | SYSTRACE("Failed to spawn container directly into target cgroup"); | |
1715 | ||
1716 | /* Kernel might simply be too old for CLONE_INTO_CGROUP. */ | |
1717 | handler->clone_flags &= ~(CLONE_INTO_CGROUP | CLONE_NEWCGROUP); | |
1718 | handler->ns_on_clone_flags &= ~CLONE_NEWCGROUP; | |
1719 | handler->ns_unshare_flags |= CLONE_NEWCGROUP; | |
1720 | ||
1721 | clone_args.flags = handler->clone_flags; | |
1722 | ||
1723 | handler->pid = lxc_clone3(&clone_args, CLONE_ARGS_SIZE_VER0); | |
1724 | } else if (cgroup_fd >= 0) { | |
1725 | TRACE("Spawned container directly into target cgroup via cgroup2 fd %d", cgroup_fd); | |
1726 | } | |
1727 | ||
1728 | /* Kernel might be too old for clone3(). */ | |
1729 | if (handler->pid < 0) { | |
1730 | SYSTRACE("Failed to spawn container via clone3()"); | |
1731 | handler->pid = lxc_raw_legacy_clone(handler->clone_flags, &handler->pidfd); | |
1732 | } | |
1733 | ||
1734 | if (handler->pid < 0) { | |
1735 | SYSERROR(LXC_CLONE_ERROR); | |
1736 | goto out_delete_net; | |
1737 | } | |
1738 | ||
1739 | if (handler->pid == 0) { | |
1740 | (void)do_start(handler); | |
1741 | _exit(EXIT_FAILURE); | |
1742 | } | |
0ad19a3f | 1743 | } |
f7176c3e CB |
1744 | if (handler->pidfd < 0) |
1745 | handler->clone_flags &= ~CLONE_PIDFD; | |
fa3a5b22 | 1746 | TRACE("Cloned child process %d", handler->pid); |
1bd8d726 | 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 | |
4cb53844 | 1764 | if (!lxc_try_preserve_namespaces(handler, handler->ns_on_clone_flags, handler->pid)) { |
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 | ||
6e48e7c5 | 1794 | if (!lxc_sync_wake_child(handler, START_SYNC_STARTUP)) |
5b1e83cb | 1795 | goto out_delete_net; |
5b1e83cb | 1796 | |
6e48e7c5 | 1797 | if (!lxc_sync_wait_child(handler, START_SYNC_CONFIGURE)) |
5b1e83cb | 1798 | goto out_delete_net; |
0ad19a3f | 1799 | |
c581d2a6 | 1800 | if (!cgroup_ops->setup_limits_legacy(cgroup_ops, handler->conf, false)) { |
cfc62c60 | 1801 | ERROR("Failed to setup cgroup limits for container \"%s\"", name); |
6031a6e5 DE |
1802 | goto out_delete_net; |
1803 | } | |
1804 | ||
fe70edee | 1805 | if (!cgroup_ops->payload_enter(cgroup_ops, handler)) { |
b1ee458e | 1806 | ERROR("Failed to enter cgroups"); |
7fef7a06 | 1807 | goto out_delete_net; |
fe70edee | 1808 | } |
218d4250 | 1809 | |
c581d2a6 CB |
1810 | if (!cgroup_ops->payload_delegate_controllers(cgroup_ops)) { |
1811 | ERROR("Failed to delegate controllers to payload cgroup"); | |
1812 | goto out_delete_net; | |
1813 | } | |
1814 | ||
1815 | if (!cgroup_ops->setup_limits(cgroup_ops, handler)) { | |
1816 | ERROR("Failed to setup cgroup limits for container \"%s\"", name); | |
1817 | goto out_delete_net; | |
1818 | } | |
1819 | ||
2202afc9 | 1820 | if (!cgroup_ops->chown(cgroup_ops, handler->conf)) |
0996e18a SH |
1821 | goto out_delete_net; |
1822 | ||
aa0c0e7b RK |
1823 | /* If not done yet, we're now ready to preserve the network namespace */ |
1824 | if (handler->nsfd[LXC_NS_NET] < 0) { | |
1825 | ret = lxc_try_preserve_ns(handler->pid, "net"); | |
1826 | if (ret < 0) { | |
1827 | if (ret != -EOPNOTSUPP) { | |
1828 | SYSERROR("Failed to preserve net namespace"); | |
1829 | goto out_delete_net; | |
1830 | } | |
1831 | } else { | |
1832 | handler->nsfd[LXC_NS_NET] = ret; | |
1833 | DEBUG("Preserved net namespace via fd %d", ret); | |
4cb53844 | 1834 | } |
fa3a5b22 | 1835 | } |
aa0c0e7b RK |
1836 | ret = lxc_netns_set_nsid(handler->nsfd[LXC_NS_NET]); |
1837 | if (ret < 0) | |
1838 | SYSWARN("Failed to allocate new network namespace id"); | |
1839 | else | |
1840 | TRACE("Allocated new network namespace id"); | |
fa3a5b22 | 1841 | |
408da065 | 1842 | /* Create the network configuration. */ |
becad0ec | 1843 | if (handler->ns_clone_flags & CLONE_NEWNET) { |
e389f2af | 1844 | ret = lxc_create_network(handler); |
cfc62c60 | 1845 | if (ret < 0) { |
e389f2af | 1846 | ERROR("Failed to create the network"); |
7fef7a06 | 1847 | goto out_delete_net; |
82d5ae15 | 1848 | } |
74c6e2b0 | 1849 | |
3c09b97c | 1850 | ret = lxc_network_send_to_child(handler); |
cfc62c60 | 1851 | if (ret < 0) { |
e389f2af | 1852 | ERROR("Failed to send veth names to child"); |
74c6e2b0 CB |
1853 | goto out_delete_net; |
1854 | } | |
0ad19a3f | 1855 | } |
1856 | ||
61d7a733 YT |
1857 | if (!lxc_list_empty(&conf->procs)) { |
1858 | ret = setup_proc_filesystem(&conf->procs, handler->pid); | |
1859 | if (ret < 0) | |
1860 | goto out_delete_net; | |
1861 | } | |
1862 | ||
408da065 | 1863 | /* Tell the child to continue its initialization. We'll get |
946a0c6d | 1864 | * START_SYNC_CGROUP when it is ready for us to setup cgroups. |
3c22086f | 1865 | */ |
6e48e7c5 | 1866 | if (!lxc_sync_barrier_child(handler, START_SYNC_POST_CONFIGURE)) |
544a48a0 SH |
1867 | goto out_delete_net; |
1868 | ||
cfc62c60 CB |
1869 | if (!lxc_list_empty(&conf->limits)) { |
1870 | ret = setup_resource_limits(&conf->limits, handler->pid); | |
1871 | if (ret < 0) { | |
1872 | ERROR("Failed to setup resource limits"); | |
1873 | goto out_delete_net; | |
1874 | } | |
c6d09e15 WB |
1875 | } |
1876 | ||
6e48e7c5 | 1877 | if (!lxc_sync_barrier_child(handler, START_SYNC_CGROUP_UNSHARE)) |
f4152036 CB |
1878 | goto out_delete_net; |
1879 | ||
432faf20 WB |
1880 | /* |
1881 | * with isolation the limiting devices cgroup was already setup, so | |
1882 | * only setup devices here if we have no namespace directory | |
1883 | */ | |
1884 | if (!handler->conf->cgroup_meta.namespace_dir && | |
1885 | !cgroup_ops->setup_limits_legacy(cgroup_ops, handler->conf, true)) { | |
cfc62c60 | 1886 | ERROR("Failed to setup legacy device cgroup controller limits"); |
b98f7d6e | 1887 | goto out_delete_net; |
544a48a0 | 1888 | } |
cfc62c60 | 1889 | TRACE("Set up legacy device cgroup controller limits"); |
544a48a0 | 1890 | |
bf651989 CB |
1891 | if (!cgroup_ops->devices_activate(cgroup_ops, handler)) { |
1892 | ERROR("Failed to setup cgroup2 device controller limits"); | |
1893 | goto out_delete_net; | |
1894 | } | |
1895 | TRACE("Set up cgroup2 device controller limits"); | |
1896 | ||
f7176c3e | 1897 | if (handler->ns_unshare_flags & CLONE_NEWCGROUP) { |
8bf3abfb | 1898 | /* Now we're ready to preserve the cgroup namespace */ |
4cb53844 | 1899 | ret = lxc_try_preserve_ns(handler->pid, "cgroup"); |
8bf3abfb | 1900 | if (ret < 0) { |
4cb53844 | 1901 | if (ret != -EOPNOTSUPP) { |
6d1400b5 | 1902 | SYSERROR("Failed to preserve cgroup namespace"); |
4cb53844 CB |
1903 | goto out_delete_net; |
1904 | } | |
1905 | } else { | |
1906 | handler->nsfd[LXC_NS_CGROUP] = ret; | |
1907 | DEBUG("Preserved cgroup namespace via fd %d", ret); | |
8bf3abfb | 1908 | } |
8bf3abfb CB |
1909 | } |
1910 | ||
78eb6aa6 CB |
1911 | cgroup_ops->payload_finalize(cgroup_ops); |
1912 | TRACE("Finished setting up cgroups"); | |
1913 | ||
f7176c3e | 1914 | if (handler->ns_unshare_flags & CLONE_NEWTIME) { |
f1c43439 CB |
1915 | /* Now we're ready to preserve the cgroup namespace */ |
1916 | ret = lxc_try_preserve_ns(handler->pid, "time"); | |
1917 | if (ret < 0) { | |
1918 | if (ret != -EOPNOTSUPP) { | |
1919 | SYSERROR("Failed to preserve time namespace"); | |
1920 | goto out_delete_net; | |
1921 | } | |
1922 | } else { | |
1923 | handler->nsfd[LXC_NS_TIME] = ret; | |
1924 | DEBUG("Preserved time namespace via fd %d", ret); | |
1925 | } | |
1926 | } | |
1927 | ||
08dd2805 | 1928 | /* Run any host-side start hooks */ |
cfc62c60 CB |
1929 | ret = run_lxc_hooks(name, "start-host", conf, NULL); |
1930 | if (ret < 0) { | |
1931 | ERROR("Failed to run lxc.hook.start-host"); | |
341ed84c | 1932 | goto out_delete_net; |
08dd2805 SH |
1933 | } |
1934 | ||
408da065 CB |
1935 | /* Tell the child to complete its initialization and wait for it to exec |
1936 | * or return an error. (The child will never return | |
946a0c6d | 1937 | * START_SYNC_READY_START+1. It will either close the sync pipe, causing |
408da065 CB |
1938 | * lxc_sync_barrier_child to return success, or return a different |
1939 | * value, causing us to error out). | |
544a48a0 | 1940 | */ |
6e48e7c5 | 1941 | if (!lxc_sync_barrier_child(handler, START_SYNC_READY_START)) |
341ed84c | 1942 | goto out_delete_net; |
0ad19a3f | 1943 | |
e389f2af CB |
1944 | if (handler->ns_clone_flags & CLONE_NEWNET) { |
1945 | ret = lxc_network_recv_name_and_ifindex_from_child(handler); | |
1946 | if (ret < 0) { | |
1947 | ERROR("Failed to receive names and ifindices for network devices from child"); | |
1948 | goto out_delete_net; | |
1949 | } | |
790255cf CB |
1950 | } |
1951 | ||
68f3899e | 1952 | ret = lxc_setup_devpts_parent(handler); |
f797f05e CB |
1953 | if (ret < 0) { |
1954 | SYSERROR("Failed to receive devpts fd from child"); | |
1955 | goto out_delete_net; | |
1956 | } | |
f797f05e | 1957 | |
790255cf | 1958 | /* Now all networks are created, network devices are moved into place, |
ad2ddfcd | 1959 | * and the correct names and ifindices in the respective namespaces have |
790255cf CB |
1960 | * been recorded. The corresponding structs have now all been filled. So |
1961 | * log them for debugging purposes. | |
1962 | */ | |
28d9e29e | 1963 | lxc_log_configured_netdevs(conf); |
790255cf | 1964 | |
f4152036 | 1965 | /* Read tty fds allocated by child. */ |
cfc62c60 CB |
1966 | ret = lxc_recv_ttys_from_child(handler); |
1967 | if (ret < 0) { | |
1968 | ERROR("Failed to receive tty info from child process"); | |
f4152036 CB |
1969 | goto out_delete_net; |
1970 | } | |
1971 | ||
c3e3c21a CB |
1972 | ret = lxc_seccomp_recv_notifier_fd(&handler->conf->seccomp, data_sock1); |
1973 | if (ret < 0) { | |
1974 | SYSERROR("Failed to receive seccomp notify fd from child"); | |
1975 | goto out_delete_net; | |
cdb2a47f | 1976 | } |
cdb2a47f | 1977 | |
cfc62c60 CB |
1978 | ret = handler->ops->post_start(handler, handler->data); |
1979 | if (ret < 0) | |
e6126dbe MN |
1980 | goto out_abort; |
1981 | ||
cfc62c60 CB |
1982 | ret = lxc_set_state(name, handler, RUNNING); |
1983 | if (ret < 0) { | |
1984 | ERROR("Failed to set state to \"%s\"", lxc_state2str(RUNNING)); | |
59eb99ba | 1985 | goto out_abort; |
3f21c114 | 1986 | } |
22ebac19 | 1987 | |
3c22086f | 1988 | lxc_sync_fini(handler); |
0c547523 | 1989 | |
e6126dbe | 1990 | return 0; |
1ac470c0 | 1991 | |
7fef7a06 | 1992 | out_delete_net: |
becad0ec | 1993 | if (handler->ns_clone_flags & CLONE_NEWNET) |
bb84beda | 1994 | lxc_delete_network(handler); |
1bd8d726 | 1995 | |
59eb99ba | 1996 | out_abort: |
0c5859ff | 1997 | lxc_abort(handler); |
95ef0d7c CB |
1998 | |
1999 | out_sync_fini: | |
3c22086f | 2000 | lxc_sync_fini(handler); |
5c068da9 | 2001 | |
b79fcd86 | 2002 | return -1; |
59eb99ba | 2003 | } |
0ad19a3f | 2004 | |
7bdf97d2 CB |
2005 | int __lxc_start(struct lxc_handler *handler, struct lxc_operations *ops, |
2006 | void *data, const char *lxcpath, bool daemonize, int *error_num) | |
59eb99ba | 2007 | { |
c30e9b19 | 2008 | int ret, status; |
7bdf97d2 | 2009 | const char *name = handler->name; |
aa460476 | 2010 | struct lxc_conf *conf = handler->conf; |
72068e74 | 2011 | struct cgroup_ops *cgroup_ops; |
80090207 | 2012 | |
c30e9b19 CB |
2013 | ret = lxc_init(name, handler); |
2014 | if (ret < 0) { | |
2015 | ERROR("Failed to initialize container \"%s\"", name); | |
9ad66589 | 2016 | goto out_abort; |
0ad19a3f | 2017 | } |
ee70bf78 CLG |
2018 | handler->ops = ops; |
2019 | handler->data = data; | |
bb955810 | 2020 | handler->daemonize = daemonize; |
72068e74 | 2021 | cgroup_ops = handler->cgroup_ops; |
e6126dbe | 2022 | |
76a26f55 | 2023 | if (!attach_block_device(handler->conf)) { |
c30e9b19 | 2024 | ERROR("Failed to attach block device"); |
575ea467 | 2025 | ret = -1; |
9ad66589 | 2026 | goto out_abort; |
76a26f55 SH |
2027 | } |
2028 | ||
72068e74 CB |
2029 | if (!cgroup_ops->monitor_create(cgroup_ops, handler)) { |
2030 | ERROR("Failed to create monitor cgroup"); | |
575ea467 | 2031 | ret = -1; |
9ad66589 | 2032 | goto out_abort; |
eeef32bb CB |
2033 | } |
2034 | ||
c581d2a6 | 2035 | if (!cgroup_ops->monitor_enter(cgroup_ops, handler)) { |
eeef32bb | 2036 | ERROR("Failed to enter monitor cgroup"); |
575ea467 | 2037 | ret = -1; |
9ad66589 | 2038 | goto out_abort; |
72068e74 CB |
2039 | } |
2040 | ||
c581d2a6 CB |
2041 | if (!cgroup_ops->monitor_delegate_controllers(cgroup_ops)) { |
2042 | ERROR("Failed to delegate controllers to monitor cgroup"); | |
2043 | ret = -1; | |
9ad66589 | 2044 | goto out_abort; |
c581d2a6 CB |
2045 | } |
2046 | ||
35120d9c | 2047 | if (geteuid() == 0 && !lxc_list_empty(&conf->id_map)) { |
408da065 | 2048 | /* If the backing store is a device, mount it here and now. */ |
35120d9c | 2049 | if (rootfs_is_blockdev(conf)) { |
c30e9b19 CB |
2050 | ret = unshare(CLONE_NEWNS); |
2051 | if (ret < 0) { | |
2052 | ERROR("Failed to unshare CLONE_NEWNS"); | |
9ad66589 | 2053 | goto out_abort; |
35120d9c | 2054 | } |
c30e9b19 | 2055 | INFO("Unshared CLONE_NEWNS"); |
408da065 | 2056 | |
9e61fb1f | 2057 | turn_into_dependent_mounts(); |
8ce1abc2 | 2058 | ret = lxc_setup_rootfs_prepare_root(conf, name, lxcpath); |
c30e9b19 CB |
2059 | if (ret < 0) { |
2060 | ERROR("Error setting up rootfs mount as root before spawn"); | |
9ad66589 | 2061 | goto out_abort; |
35120d9c | 2062 | } |
c30e9b19 | 2063 | INFO("Set up container rootfs as host root"); |
35120d9c SH |
2064 | } |
2065 | } | |
2066 | ||
c30e9b19 CB |
2067 | ret = lxc_spawn(handler); |
2068 | if (ret < 0) { | |
2069 | ERROR("Failed to spawn container \"%s\"", name); | |
76a26f55 | 2070 | goto out_detach_blockdev; |
0ad19a3f | 2071 | } |
2072 | ||
80308d07 | 2073 | handler->conf->reboot = REBOOT_NONE; |
8bee8851 | 2074 | |
c30e9b19 CB |
2075 | ret = lxc_poll(name, handler); |
2076 | if (ret) { | |
2077 | ERROR("LXC mainloop exited with error: %d", ret); | |
9ad66589 | 2078 | goto out_delete_network; |
59eb99ba | 2079 | } |
0ad19a3f | 2080 | |
321db026 DJ |
2081 | if (!handler->init_died && handler->pid > 0) { |
2082 | ERROR("Child process is not killed"); | |
575ea467 | 2083 | ret = -1; |
9ad66589 | 2084 | goto out_delete_network; |
321db026 DJ |
2085 | } |
2086 | ||
c30e9b19 CB |
2087 | status = lxc_wait_for_pid_status(handler->pid); |
2088 | if (status < 0) | |
2089 | SYSERROR("Failed to retrieve status for %d", handler->pid); | |
e043236e | 2090 | |
408da065 CB |
2091 | /* If the child process exited but was not signaled, it didn't call |
2092 | * reboot. This should mean it was an lxc-execute which simply exited. | |
2093 | * In any case, treat it as a 'halt'. | |
8b004f07 | 2094 | */ |
d028235d | 2095 | if (WIFSIGNALED(status)) { |
8b004f07 SH |
2096 | switch(WTERMSIG(status)) { |
2097 | case SIGINT: /* halt */ | |
c30e9b19 | 2098 | DEBUG("Container \"%s\" is halting", name); |
8b004f07 SH |
2099 | break; |
2100 | case SIGHUP: /* reboot */ | |
c30e9b19 | 2101 | DEBUG("Container \"%s\" is rebooting", name); |
80308d07 | 2102 | handler->conf->reboot = REBOOT_REQ; |
8b004f07 | 2103 | break; |
c2b9bd9e | 2104 | case SIGSYS: /* seccomp */ |
c30e9b19 | 2105 | DEBUG("Container \"%s\" violated its seccomp policy", name); |
c2b9bd9e | 2106 | break; |
8b004f07 | 2107 | default: |
c30e9b19 | 2108 | DEBUG("Unknown exit status for container \"%s\" init %d", name, WTERMSIG(status)); |
8b004f07 SH |
2109 | break; |
2110 | } | |
d028235d | 2111 | } |
828695d9 | 2112 | |
c30e9b19 CB |
2113 | ret = lxc_restore_phys_nics_to_netns(handler); |
2114 | if (ret < 0) | |
818a57fc | 2115 | ERROR("Failed to move physical network devices back to parent network namespace"); |
ce5782df | 2116 | |
1787abca | 2117 | lxc_monitor_send_exit_code(name, status, handler->lxcpath); |
eb808539 | 2118 | lxc_error_set_and_log(handler->pid, status); |
a3b4f3d6 TA |
2119 | if (error_num) |
2120 | *error_num = handler->exit_status; | |
1bd8d726 | 2121 | |
bb84beda | 2122 | lxc_delete_network(handler); |
76a26f55 | 2123 | detach_block_device(handler->conf); |
fd5be714 | 2124 | lxc_end(handler); |
c30e9b19 | 2125 | return ret; |
0ad19a3f | 2126 | |
9ad66589 | 2127 | out_abort: |
0c5859ff | 2128 | lxc_abort(handler); |
945daa24 CB |
2129 | lxc_end(handler); |
2130 | return ret; | |
65a2dc8b CB |
2131 | |
2132 | out_detach_blockdev: | |
0c5859ff | 2133 | lxc_abort(handler); |
945daa24 CB |
2134 | detach_block_device(handler->conf); |
2135 | lxc_end(handler); | |
2136 | return ret; | |
65a2dc8b | 2137 | |
9ad66589 | 2138 | out_delete_network: |
0c5859ff | 2139 | lxc_abort(handler); |
abd953eb | 2140 | lxc_restore_phys_nics_to_netns(handler); |
945daa24 CB |
2141 | lxc_delete_network(handler); |
2142 | detach_block_device(handler->conf); | |
2143 | lxc_end(handler); | |
2144 | return ret; | |
0ad19a3f | 2145 | } |
ee70bf78 CLG |
2146 | |
2147 | struct start_args { | |
2148 | char *const *argv; | |
2149 | }; | |
2150 | ||
2151 | static int start(struct lxc_handler *handler, void* data) | |
2152 | { | |
2153 | struct start_args *arg = data; | |
2154 | ||
984984e4 | 2155 | NOTICE("Exec'ing \"%s\"", arg->argv[0]); |
ee70bf78 CLG |
2156 | |
2157 | execvp(arg->argv[0], arg->argv); | |
984984e4 | 2158 | SYSERROR("Failed to exec \"%s\"", arg->argv[0]); |
ee70bf78 CLG |
2159 | return 0; |
2160 | } | |
2161 | ||
2162 | static int post_start(struct lxc_handler *handler, void* data) | |
2163 | { | |
2164 | struct start_args *arg = data; | |
2165 | ||
4c8e880e | 2166 | NOTICE("Started \"%s\" with pid \"%d\"", arg->argv[0], handler->pid); |
ee70bf78 CLG |
2167 | return 0; |
2168 | } | |
2169 | ||
2170 | static struct lxc_operations start_ops = { | |
2171 | .start = start, | |
2172 | .post_start = post_start | |
2173 | }; | |
2174 | ||
7bdf97d2 | 2175 | int lxc_start(char *const argv[], struct lxc_handler *handler, |
bb955810 | 2176 | const char *lxcpath, bool daemonize, int *error_num) |
ee70bf78 CLG |
2177 | { |
2178 | struct start_args start_arg = { | |
2179 | .argv = argv, | |
2180 | }; | |
2181 | ||
b2efeb0b | 2182 | TRACE("Doing lxc_start"); |
7bdf97d2 | 2183 | return __lxc_start(handler, &start_ops, &start_arg, lxcpath, daemonize, error_num); |
ee70bf78 | 2184 | } |
28272964 CB |
2185 | |
2186 | static void lxc_destroy_container_on_signal(struct lxc_handler *handler, | |
2187 | const char *name) | |
2188 | { | |
245100a0 | 2189 | char destroy[PATH_MAX]; |
f01f7975 | 2190 | struct lxc_container *c; |
ae3beac9 CB |
2191 | int ret = 0; |
2192 | bool bret = true; | |
2193 | ||
df31363a | 2194 | if (handler->conf->rootfs.path && handler->conf->rootfs.mount) { |
790255cf | 2195 | bret = do_destroy_container(handler); |
28272964 | 2196 | if (!bret) { |
ae3beac9 | 2197 | ERROR("Error destroying rootfs for container \"%s\"", name); |
28272964 CB |
2198 | return; |
2199 | } | |
2200 | } | |
ae3beac9 | 2201 | INFO("Destroyed rootfs for container \"%s\"", name); |
28272964 | 2202 | |
fa60cd7b CB |
2203 | ret = strnprintf(destroy, sizeof(destroy), "%s/%s", handler->lxcpath, name); |
2204 | if (ret < 0) { | |
ae3beac9 | 2205 | ERROR("Error destroying directory for container \"%s\"", name); |
28272964 CB |
2206 | return; |
2207 | } | |
2208 | ||
f01f7975 CB |
2209 | c = lxc_container_new(name, handler->lxcpath); |
2210 | if (c) { | |
2211 | if (container_disk_lock(c)) { | |
ae3beac9 | 2212 | INFO("Could not update lxc_snapshots file"); |
f01f7975 CB |
2213 | lxc_container_put(c); |
2214 | } else { | |
2215 | mod_all_rdeps(c, false); | |
2216 | container_disk_unlock(c); | |
2217 | lxc_container_put(c); | |
2218 | } | |
2219 | } | |
2220 | ||
d0fbc7ba | 2221 | if (!handler->am_root) |
94b0a3ac CB |
2222 | ret = userns_exec_full(handler->conf, lxc_rmdir_onedev_wrapper, |
2223 | destroy, "lxc_rmdir_onedev_wrapper"); | |
28272964 CB |
2224 | else |
2225 | ret = lxc_rmdir_onedev(destroy, NULL); | |
2226 | ||
2227 | if (ret < 0) { | |
ae3beac9 | 2228 | ERROR("Error destroying directory for container \"%s\"", name); |
28272964 CB |
2229 | return; |
2230 | } | |
ae3beac9 | 2231 | INFO("Destroyed directory for container \"%s\"", name); |
28272964 CB |
2232 | } |
2233 | ||
2234 | static int lxc_rmdir_onedev_wrapper(void *data) | |
2235 | { | |
2236 | char *arg = (char *) data; | |
2237 | return lxc_rmdir_onedev(arg, NULL); | |
2238 | } | |
2239 | ||
134df645 CB |
2240 | static bool do_destroy_container(struct lxc_handler *handler) |
2241 | { | |
94b0a3ac CB |
2242 | int ret; |
2243 | ||
d0fbc7ba | 2244 | if (!handler->am_root) { |
94b0a3ac CB |
2245 | ret = userns_exec_full(handler->conf, storage_destroy_wrapper, |
2246 | handler->conf, "storage_destroy_wrapper"); | |
2247 | if (ret < 0) | |
d028235d | 2248 | return false; |
10bc1861 | 2249 | |
d028235d SG |
2250 | return true; |
2251 | } | |
10bc1861 | 2252 | |
790255cf | 2253 | return storage_destroy(handler->conf); |
28272964 | 2254 | } |