]>
Commit | Line | Data |
---|---|---|
0ad19a3f | 1 | /* |
2 | * lxc: linux Container library | |
3 | * | |
4 | * (C) Copyright IBM Corp. 2007, 2008 | |
5 | * | |
6 | * Authors: | |
9afe19d6 | 7 | * Daniel Lezcano <daniel.lezcano at free.fr> |
0ad19a3f | 8 | * |
9 | * This library is free software; you can redistribute it and/or | |
10 | * modify it under the terms of the GNU Lesser General Public | |
11 | * License as published by the Free Software Foundation; either | |
12 | * version 2.1 of the License, or (at your option) any later version. | |
13 | * | |
14 | * This library is distributed in the hope that it will be useful, | |
15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
17 | * Lesser General Public License for more details. | |
18 | * | |
19 | * You should have received a copy of the GNU Lesser General Public | |
20 | * License along with this library; if not, write to the Free Software | |
250b1eec | 21 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
0ad19a3f | 22 | */ |
23 | ||
f549edcc GK |
24 | #include "config.h" |
25 | ||
0ad19a3f | 26 | #include <stdio.h> |
0ad19a3f | 27 | #include <string.h> |
28 | #include <stdlib.h> | |
29 | #include <dirent.h> | |
30 | #include <errno.h> | |
31 | #include <unistd.h> | |
32 | #include <signal.h> | |
b0a33c1e | 33 | #include <fcntl.h> |
c476bdce | 34 | #include <grp.h> |
37515ebd | 35 | #include <poll.h> |
0ad19a3f | 36 | #include <sys/param.h> |
37 | #include <sys/file.h> | |
f4d507d5 | 38 | #include <sys/mount.h> |
b4f8660e | 39 | #include <sys/stat.h> |
0ad19a3f | 40 | #include <sys/types.h> |
8173e600 | 41 | #include <sys/socket.h> |
0ad19a3f | 42 | #include <sys/prctl.h> |
ddceb1f9 | 43 | #include <sys/types.h> |
0ad19a3f | 44 | #include <sys/wait.h> |
b0a33c1e | 45 | #include <sys/un.h> |
8173e600 | 46 | #include <sys/syscall.h> |
ff218c25 | 47 | |
495d2046 SG |
48 | #if HAVE_SYS_CAPABILITY_H |
49 | #include <sys/capability.h> | |
50 | #endif | |
51 | ||
656994bb MH |
52 | #if !HAVE_DECL_PR_CAPBSET_DROP |
53 | #define PR_CAPBSET_DROP 24 | |
54 | #endif | |
55 | ||
63376d7d DL |
56 | #include "start.h" |
57 | #include "conf.h" | |
58 | #include "log.h" | |
563f2f2c | 59 | #include "cgroup.h" |
e2bcd7db | 60 | #include "error.h" |
b0a33c1e | 61 | #include "af_unix.h" |
62 | #include "mainloop.h" | |
63376d7d | 63 | #include "utils.h" |
565c2d76 | 64 | #include "lxcutmp.h" |
63376d7d | 65 | #include "monitor.h" |
96fa1ff0 | 66 | #include "commands.h" |
63376d7d | 67 | #include "console.h" |
3c22086f | 68 | #include "sync.h" |
f549edcc | 69 | #include "namespace.h" |
0d0527a9 | 70 | #include "lxcseccomp.h" |
8173e600 | 71 | #include "caps.h" |
76a26f55 | 72 | #include "bdev.h" |
fe4de9a6 | 73 | #include "lsm/lsm.h" |
36eb9bde CLG |
74 | |
75 | lxc_log_define(lxc_start, lxc); | |
76 | ||
9f30a190 MM |
77 | const struct ns_info ns_info[LXC_NS_MAX] = { |
78 | [LXC_NS_MNT] = {"mnt", CLONE_NEWNS}, | |
79 | [LXC_NS_PID] = {"pid", CLONE_NEWPID}, | |
80 | [LXC_NS_UTS] = {"uts", CLONE_NEWUTS}, | |
81 | [LXC_NS_IPC] = {"ipc", CLONE_NEWIPC}, | |
82 | [LXC_NS_USER] = {"user", CLONE_NEWUSER}, | |
83 | [LXC_NS_NET] = {"net", CLONE_NEWNET} | |
84 | }; | |
85 | ||
c8154066 SH |
86 | static void print_top_failing_dir(const char *path) |
87 | { | |
88 | size_t len = strlen(path); | |
89 | char *copy = alloca(len+1), *p, *e, saved; | |
90 | strcpy(copy, path); | |
91 | ||
92 | p = copy; | |
93 | e = copy + len; | |
94 | while (p < e) { | |
95 | while (p < e && *p == '/') p++; | |
96 | while (p < e && *p != '/') p++; | |
c8154066 SH |
97 | saved = *p; |
98 | *p = '\0'; | |
99 | if (access(copy, X_OK)) { | |
100 | SYSERROR("could not access %s. Please grant it 'x' " \ | |
101 | "access, or add an ACL for the container root.", | |
102 | copy); | |
103 | return; | |
104 | } | |
105 | *p = saved; | |
106 | } | |
107 | } | |
108 | ||
9f30a190 MM |
109 | static void close_ns(int ns_fd[LXC_NS_MAX]) { |
110 | int i; | |
111 | ||
9f30a190 MM |
112 | for (i = 0; i < LXC_NS_MAX; i++) { |
113 | if (ns_fd[i] > -1) { | |
114 | close(ns_fd[i]); | |
115 | ns_fd[i] = -1; | |
116 | } | |
117 | } | |
9f30a190 MM |
118 | } |
119 | ||
120 | static int preserve_ns(int ns_fd[LXC_NS_MAX], int clone_flags) { | |
121 | int i, saved_errno; | |
122 | char path[MAXPATHLEN]; | |
123 | ||
9f30a190 MM |
124 | for (i = 0; i < LXC_NS_MAX; i++) |
125 | ns_fd[i] = -1; | |
126 | ||
cd43d2d1 SH |
127 | if (access("/proc/self/ns", X_OK)) { |
128 | WARN("Kernel does not support attach; preserve_ns ignored"); | |
129 | return 0; | |
130 | } | |
131 | ||
9f30a190 MM |
132 | for (i = 0; i < LXC_NS_MAX; i++) { |
133 | if ((clone_flags & ns_info[i].clone_flag) == 0) | |
134 | continue; | |
135 | snprintf(path, MAXPATHLEN, "/proc/self/ns/%s", ns_info[i].proc_name); | |
9f30a190 | 136 | ns_fd[i] = open(path, O_RDONLY | O_CLOEXEC); |
9f30a190 MM |
137 | if (ns_fd[i] < 0) |
138 | goto error; | |
139 | } | |
140 | ||
141 | return 0; | |
142 | ||
143 | error: | |
144 | saved_errno = errno; | |
145 | close_ns(ns_fd); | |
146 | errno = saved_errno; | |
147 | SYSERROR("failed to open '%s'", path); | |
148 | return -1; | |
149 | } | |
150 | ||
151 | static int attach_ns(const int ns_fd[LXC_NS_MAX]) { | |
152 | int i; | |
153 | ||
154 | for (i = 0; i < LXC_NS_MAX; i++) { | |
155 | if (ns_fd[i] < 0) | |
156 | continue; | |
157 | ||
158 | if (setns(ns_fd[i], 0) != 0) | |
159 | goto error; | |
160 | } | |
161 | return 0; | |
162 | ||
163 | error: | |
164 | SYSERROR("failed to set namespace '%s'", ns_info[i].proc_name); | |
165 | return -1; | |
166 | } | |
167 | ||
80090207 CLG |
168 | static int match_fd(int fd) |
169 | { | |
170 | return (fd == 0 || fd == 1 || fd == 2); | |
171 | } | |
172 | ||
d2cf4c37 SH |
173 | /* |
174 | * Check for any fds we need to close | |
175 | * * if fd_to_ignore != -1, then if we find that fd open we will ignore it. | |
176 | * * By default we warn about open fds we find. | |
177 | * * If closeall is true, we will close open fds. | |
178 | * * If lxc-start was passed "-C", then conf->close_all_fds will be true, | |
179 | * in which case we also close all open fds. | |
180 | * * A daemonized container will always pass closeall=true. | |
181 | */ | |
182 | int lxc_check_inherited(struct lxc_conf *conf, bool closeall, int fd_to_ignore) | |
80090207 CLG |
183 | { |
184 | struct dirent dirent, *direntp; | |
185 | int fd, fddir; | |
186 | DIR *dir; | |
80090207 | 187 | |
d2cf4c37 SH |
188 | if (conf && conf->close_all_fds) |
189 | closeall = true; | |
190 | ||
b119f362 | 191 | restart: |
80090207 CLG |
192 | dir = opendir("/proc/self/fd"); |
193 | if (!dir) { | |
194 | WARN("failed to open directory: %m"); | |
195 | return -1; | |
196 | } | |
197 | ||
198 | fddir = dirfd(dir); | |
199 | ||
200 | while (!readdir_r(dir, &dirent, &direntp)) { | |
80090207 CLG |
201 | if (!direntp) |
202 | break; | |
203 | ||
204 | if (!strcmp(direntp->d_name, ".")) | |
205 | continue; | |
206 | ||
207 | if (!strcmp(direntp->d_name, "..")) | |
208 | continue; | |
209 | ||
210 | fd = atoi(direntp->d_name); | |
211 | ||
f2faa8fa | 212 | if (fd == fddir || fd == lxc_log_fd || fd == fd_to_ignore) |
80090207 CLG |
213 | continue; |
214 | ||
858377e4 SH |
215 | if (current_config && fd == current_config->logfd) |
216 | continue; | |
217 | ||
80090207 CLG |
218 | if (match_fd(fd)) |
219 | continue; | |
80090207 | 220 | |
d2cf4c37 | 221 | if (closeall) { |
b119f362 SH |
222 | close(fd); |
223 | closedir(dir); | |
224 | INFO("closed inherited fd %d", fd); | |
225 | goto restart; | |
226 | } | |
92c7f629 | 227 | WARN("inherited fd %d", fd); |
80090207 CLG |
228 | } |
229 | ||
92c7f629 GK |
230 | closedir(dir); /* cannot fail */ |
231 | return 0; | |
80090207 CLG |
232 | } |
233 | ||
83ee7875 | 234 | static int setup_signal_fd(sigset_t *oldmask) |
b0a33c1e | 235 | { |
236 | sigset_t mask; | |
237 | int fd; | |
238 | ||
f3304a29 FW |
239 | /* Block everything except serious error signals */ |
240 | if (sigfillset(&mask) || | |
241 | sigdelset(&mask, SIGILL) || | |
242 | sigdelset(&mask, SIGSEGV) || | |
243 | sigdelset(&mask, SIGBUS) || | |
b5159817 | 244 | sigdelset(&mask, SIGWINCH) || |
f3304a29 FW |
245 | sigprocmask(SIG_BLOCK, &mask, oldmask)) { |
246 | SYSERROR("failed to set signal mask"); | |
b0a33c1e | 247 | return -1; |
248 | } | |
249 | ||
250 | fd = signalfd(-1, &mask, 0); | |
251 | if (fd < 0) { | |
36eb9bde | 252 | SYSERROR("failed to create the signal fd"); |
b0a33c1e | 253 | return -1; |
254 | } | |
255 | ||
256 | if (fcntl(fd, F_SETFD, FD_CLOEXEC)) { | |
36eb9bde | 257 | SYSERROR("failed to set sigfd to close-on-exec"); |
b0a33c1e | 258 | close(fd); |
259 | return -1; | |
260 | } | |
261 | ||
1ac470c0 DL |
262 | DEBUG("sigchild handler set"); |
263 | ||
b0a33c1e | 264 | return fd; |
265 | } | |
266 | ||
84c92abd | 267 | static int signal_handler(int fd, uint32_t events, void *data, |
b0a33c1e | 268 | struct lxc_epoll_descr *descr) |
269 | { | |
15cd25fd | 270 | struct signalfd_siginfo siginfo; |
80507ee8 | 271 | siginfo_t info; |
15cd25fd | 272 | int ret; |
82d89dce | 273 | pid_t *pid = data; |
80507ee8 | 274 | bool init_died = false; |
15cd25fd DL |
275 | |
276 | ret = read(fd, &siginfo, sizeof(siginfo)); | |
277 | if (ret < 0) { | |
f3304a29 | 278 | ERROR("failed to read signal info"); |
15cd25fd DL |
279 | return -1; |
280 | } | |
281 | ||
282 | if (ret != sizeof(siginfo)) { | |
283 | ERROR("unexpected siginfo size"); | |
284 | return -1; | |
285 | } | |
286 | ||
80507ee8 SH |
287 | // check whether init is running |
288 | info.si_pid = 0; | |
289 | ret = waitid(P_PID, *pid, &info, WEXITED | WNOWAIT | WNOHANG); | |
290 | if (ret == 0 && info.si_pid == *pid) { | |
291 | init_died = true; | |
292 | } | |
293 | ||
f3304a29 FW |
294 | if (siginfo.ssi_signo != SIGCHLD) { |
295 | kill(*pid, siginfo.ssi_signo); | |
296 | INFO("forwarded signal %d to pid %d", siginfo.ssi_signo, *pid); | |
80507ee8 | 297 | return init_died ? 1 : 0; |
f3304a29 FW |
298 | } |
299 | ||
15cd25fd DL |
300 | if (siginfo.ssi_code == CLD_STOPPED || |
301 | siginfo.ssi_code == CLD_CONTINUED) { | |
302 | INFO("container init process was stopped/continued"); | |
80507ee8 | 303 | return init_died ? 1 : 0; |
15cd25fd | 304 | } |
1ac470c0 | 305 | |
82d89dce DL |
306 | /* more robustness, protect ourself from a SIGCHLD sent |
307 | * by a process different from the container init | |
308 | */ | |
309 | if (siginfo.ssi_pid != *pid) { | |
310 | WARN("invalid pid for SIGCHLD"); | |
80507ee8 | 311 | return init_died ? 1 : 0; |
82d89dce DL |
312 | } |
313 | ||
15cd25fd | 314 | DEBUG("container init process exited"); |
b0a33c1e | 315 | return 1; |
316 | } | |
317 | ||
735f2c6e | 318 | int lxc_set_state(const char *name, struct lxc_handler *handler, lxc_state_t state) |
66aeffc7 DL |
319 | { |
320 | handler->state = state; | |
9123e471 | 321 | lxc_monitor_send_state(name, state, handler->lxcpath); |
66aeffc7 DL |
322 | return 0; |
323 | } | |
324 | ||
735f2c6e | 325 | int lxc_poll(const char *name, struct lxc_handler *handler) |
b0a33c1e | 326 | { |
ca5f7926 DL |
327 | int sigfd = handler->sigfd; |
328 | int pid = handler->pid; | |
b0a33c1e | 329 | struct lxc_epoll_descr descr; |
330 | ||
a9e61274 | 331 | if (lxc_mainloop_open(&descr)) { |
36eb9bde | 332 | ERROR("failed to create mainloop"); |
50c8bf05 | 333 | goto out_sigfd; |
b0a33c1e | 334 | } |
335 | ||
83ee7875 | 336 | if (lxc_mainloop_add_handler(&descr, sigfd, signal_handler, &pid)) { |
36eb9bde | 337 | ERROR("failed to add handler for the signal"); |
b0a33c1e | 338 | goto out_mainloop_open; |
339 | } | |
340 | ||
63376d7d DL |
341 | if (lxc_console_mainloop_add(&descr, handler)) { |
342 | ERROR("failed to add console handler to mainloop"); | |
343 | goto out_mainloop_open; | |
344 | } | |
345 | ||
ef6e34ee | 346 | if (lxc_cmd_mainloop_add(name, &descr, handler)) { |
563f2f2c | 347 | ERROR("failed to add command handler to mainloop"); |
96fa1ff0 | 348 | goto out_mainloop_open; |
563f2f2c DL |
349 | } |
350 | ||
828695d9 | 351 | if (handler->conf->need_utmp_watch) { |
495d2046 | 352 | #if HAVE_SYS_CAPABILITY_H |
828695d9 SH |
353 | if (lxc_utmp_mainloop_add(&descr, handler)) { |
354 | ERROR("failed to add utmp handler to mainloop"); | |
355 | goto out_mainloop_open; | |
356 | } | |
495d2046 | 357 | #else |
959aee9c | 358 | DEBUG("not starting utmp handler as cap_sys_boot cannot be dropped without capabilities support"); |
495d2046 | 359 | #endif |
563f2f2c | 360 | } |
b0a33c1e | 361 | |
e51d4895 | 362 | return lxc_mainloop(&descr, -1); |
b0a33c1e | 363 | |
364 | out_mainloop_open: | |
365 | lxc_mainloop_close(&descr); | |
b0a33c1e | 366 | out_sigfd: |
367 | close(sigfd); | |
c3e13372 | 368 | return -1; |
b0a33c1e | 369 | } |
370 | ||
13f5be62 | 371 | struct lxc_handler *lxc_init(const char *name, struct lxc_conf *conf, const char *lxcpath) |
59eb99ba | 372 | { |
3a0f472d DL |
373 | struct lxc_handler *handler; |
374 | ||
375 | handler = malloc(sizeof(*handler)); | |
376 | if (!handler) | |
377 | return NULL; | |
59eb99ba DL |
378 | |
379 | memset(handler, 0, sizeof(*handler)); | |
380 | ||
e8bd4e43 | 381 | handler->ttysock[0] = handler->ttysock[1] = -1; |
fae349da | 382 | handler->conf = conf; |
9123e471 | 383 | handler->lxcpath = lxcpath; |
5c068da9 | 384 | handler->pinfd = -1; |
fae349da | 385 | |
fe4de9a6 DE |
386 | lsm_init(); |
387 | ||
3bdf52d7 DL |
388 | handler->name = strdup(name); |
389 | if (!handler->name) { | |
390 | ERROR("failed to allocate memory"); | |
391 | goto out_free; | |
392 | } | |
393 | ||
ef6e34ee | 394 | if (lxc_cmd_init(name, handler, lxcpath)) |
d2e30e99 DE |
395 | goto out_free_name; |
396 | ||
8f2c3a70 SH |
397 | if (lxc_read_seccomp_config(conf) != 0) { |
398 | ERROR("failed loading seccomp policy"); | |
d2e30e99 | 399 | goto out_close_maincmd_fd; |
8f2c3a70 SH |
400 | } |
401 | ||
051151de | 402 | /* Begin by setting the state to STARTING */ |
25c2aca5 | 403 | if (lxc_set_state(name, handler, STARTING)) { |
59eb99ba | 404 | ERROR("failed to set state '%s'", lxc_state2str(STARTING)); |
051151de | 405 | goto out_close_maincmd_fd; |
0ad19a3f | 406 | } |
407 | ||
f7bee6c6 MW |
408 | /* Start of environment variable setup for hooks */ |
409 | if (setenv("LXC_NAME", name, 1)) { | |
410 | SYSERROR("failed to set environment variable for container name"); | |
411 | } | |
412 | if (setenv("LXC_CONFIG_FILE", conf->rcfile, 1)) { | |
413 | SYSERROR("failed to set environment variable for config path"); | |
414 | } | |
415 | if (setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1)) { | |
416 | SYSERROR("failed to set environment variable for rootfs mount"); | |
417 | } | |
418 | if (setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1)) { | |
419 | SYSERROR("failed to set environment variable for rootfs mount"); | |
420 | } | |
421 | if (conf->console.path && setenv("LXC_CONSOLE", conf->console.path, 1)) { | |
422 | SYSERROR("failed to set environment variable for console path"); | |
423 | } | |
424 | if (conf->console.log_path && setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1)) { | |
425 | SYSERROR("failed to set environment variable for console log"); | |
426 | } | |
427 | /* End of environment variable setup for hooks */ | |
428 | ||
283678ed | 429 | if (run_lxc_hooks(name, "pre-start", conf, handler->lxcpath, NULL)) { |
773fb9ca SH |
430 | ERROR("failed to run pre-start hooks for container '%s'.", name); |
431 | goto out_aborting; | |
432 | } | |
26ddeedd | 433 | |
b0a33c1e | 434 | /* the signal fd has to be created before forking otherwise |
435 | * if the child process exits before we setup the signal fd, | |
436 | * the event will be lost and the command will be stuck */ | |
83ee7875 | 437 | handler->sigfd = setup_signal_fd(&handler->oldmask); |
59eb99ba | 438 | if (handler->sigfd < 0) { |
36eb9bde | 439 | ERROR("failed to set sigchild fd handler"); |
b5159817 DE |
440 | goto out_delete_tty; |
441 | } | |
442 | ||
443 | /* do this after setting up signals since it might unblock SIGWINCH */ | |
444 | if (lxc_console_create(conf)) { | |
445 | ERROR("failed to create console"); | |
446 | goto out_restore_sigmask; | |
b0a33c1e | 447 | } |
448 | ||
c4d10a05 SH |
449 | if (ttys_shift_ids(conf) < 0) { |
450 | ERROR("Failed to shift tty into container"); | |
451 | goto out_restore_sigmask; | |
452 | } | |
453 | ||
c3e13372 | 454 | INFO("'%s' is initialized", name); |
3a0f472d | 455 | return handler; |
59eb99ba | 456 | |
b5159817 DE |
457 | out_restore_sigmask: |
458 | sigprocmask(SIG_SETMASK, &handler->oldmask, NULL); | |
59eb99ba | 459 | out_delete_tty: |
fae349da | 460 | lxc_delete_tty(&conf->tty_info); |
59eb99ba | 461 | out_aborting: |
25c2aca5 | 462 | lxc_set_state(name, handler, ABORTING); |
d2e30e99 DE |
463 | out_close_maincmd_fd: |
464 | close(conf->maincmd_fd); | |
465 | conf->maincmd_fd = -1; | |
3bdf52d7 DL |
466 | out_free_name: |
467 | free(handler->name); | |
468 | handler->name = NULL; | |
3a0f472d DL |
469 | out_free: |
470 | free(handler); | |
c3e13372 | 471 | return NULL; |
59eb99ba DL |
472 | } |
473 | ||
3b72c4a0 | 474 | void lxc_fini(const char *name, struct lxc_handler *handler) |
59eb99ba DL |
475 | { |
476 | /* The STOPPING state is there for future cleanup code | |
477 | * which can take awhile | |
478 | */ | |
25c2aca5 MN |
479 | lxc_set_state(name, handler, STOPPING); |
480 | lxc_set_state(name, handler, STOPPED); | |
59eb99ba | 481 | |
283678ed | 482 | if (run_lxc_hooks(name, "post-stop", handler->conf, handler->lxcpath, NULL)) |
773fb9ca | 483 | ERROR("failed to run post-stop hooks for container '%s'.", name); |
26ddeedd | 484 | |
83ee7875 | 485 | /* reset mask set by setup_signal_fd */ |
8f64a3f6 MN |
486 | if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) |
487 | WARN("failed to restore sigprocmask"); | |
488 | ||
b5159817 | 489 | lxc_console_delete(&handler->conf->console); |
b2431939 | 490 | lxc_delete_tty(&handler->conf->tty_info); |
d2e30e99 DE |
491 | close(handler->conf->maincmd_fd); |
492 | handler->conf->maincmd_fd = -1; | |
3bdf52d7 | 493 | free(handler->name); |
e8bd4e43 SH |
494 | if (handler->ttysock[0] != -1) { |
495 | close(handler->ttysock[0]); | |
496 | close(handler->ttysock[1]); | |
497 | } | |
d4ef7c50 | 498 | cgroup_destroy(handler); |
b2431939 | 499 | free(handler); |
59eb99ba DL |
500 | } |
501 | ||
735f2c6e | 502 | void lxc_abort(const char *name, struct lxc_handler *handler) |
59eb99ba | 503 | { |
73e608b2 SH |
504 | int ret, status; |
505 | ||
25c2aca5 | 506 | lxc_set_state(name, handler, ABORTING); |
7d9fb3e9 DL |
507 | if (handler->pid > 0) |
508 | kill(handler->pid, SIGKILL); | |
73e608b2 | 509 | while ((ret = waitpid(-1, &status, 0)) > 0) ; |
59eb99ba DL |
510 | } |
511 | ||
828695d9 SH |
512 | #include <sys/reboot.h> |
513 | #include <linux/reboot.h> | |
514 | ||
e2fa1520 SH |
515 | /* |
516 | * reboot(LINUX_REBOOT_CMD_CAD_ON) will return -EINVAL | |
517 | * in a child pid namespace if container reboot support exists. | |
518 | * Otherwise, it will either succeed or return -EPERM. | |
519 | */ | |
520 | static int container_reboot_supported(void *arg) | |
828695d9 | 521 | { |
e2fa1520 | 522 | int *cmd = arg; |
828695d9 | 523 | int ret; |
828695d9 | 524 | |
e2fa1520 SH |
525 | ret = reboot(*cmd); |
526 | if (ret == -1 && errno == EINVAL) | |
527 | return 1; | |
528 | return 0; | |
529 | } | |
530 | ||
b60ed720 | 531 | static int must_drop_cap_sys_boot(struct lxc_conf *conf) |
e2fa1520 | 532 | { |
025ed0f3 | 533 | FILE *f; |
b60ed720 | 534 | int ret, cmd, v, flags; |
e2fa1520 | 535 | long stack_size = 4096; |
7f145a6d | 536 | void *stack = alloca(stack_size); |
e2fa1520 SH |
537 | int status; |
538 | pid_t pid; | |
539 | ||
025ed0f3 | 540 | f = fopen("/proc/sys/kernel/ctrl-alt-del", "r"); |
e2fa1520 SH |
541 | if (!f) { |
542 | DEBUG("failed to open /proc/sys/kernel/ctrl-alt-del"); | |
828695d9 | 543 | return 1; |
e2fa1520 | 544 | } |
828695d9 SH |
545 | |
546 | ret = fscanf(f, "%d", &v); | |
547 | fclose(f); | |
e2fa1520 SH |
548 | if (ret != 1) { |
549 | DEBUG("Failed to read /proc/sys/kernel/ctrl-alt-del"); | |
828695d9 | 550 | return 1; |
e2fa1520 SH |
551 | } |
552 | cmd = v ? LINUX_REBOOT_CMD_CAD_ON : LINUX_REBOOT_CMD_CAD_OFF; | |
553 | ||
b60ed720 SH |
554 | flags = CLONE_NEWPID | SIGCHLD; |
555 | if (!lxc_list_empty(&conf->id_map)) | |
556 | flags |= CLONE_NEWUSER; | |
557 | ||
7f145a6d | 558 | #ifdef __ia64__ |
959aee9c | 559 | pid = __clone2(container_reboot_supported, stack, stack_size, flags, &cmd); |
7f145a6d | 560 | #else |
959aee9c SG |
561 | stack += stack_size; |
562 | pid = clone(container_reboot_supported, stack, flags, &cmd); | |
7f145a6d | 563 | #endif |
959aee9c | 564 | if (pid < 0) { |
c08220e9 | 565 | if (flags & CLONE_NEWUSER) |
4a7e5f4f | 566 | ERROR("failed to clone (%#x): %s (includes CLONE_NEWUSER)", flags, strerror(errno)); |
c08220e9 | 567 | else |
4a7e5f4f | 568 | ERROR("failed to clone (%#x): %s", flags, strerror(errno)); |
959aee9c SG |
569 | return -1; |
570 | } | |
571 | if (wait(&status) < 0) { | |
572 | SYSERROR("unexpected wait error: %m"); | |
573 | return -1; | |
574 | } | |
e2fa1520 SH |
575 | |
576 | if (WEXITSTATUS(status) != 1) | |
828695d9 | 577 | return 1; |
e2fa1520 | 578 | |
828695d9 SH |
579 | return 0; |
580 | } | |
581 | ||
658979c5 SH |
582 | /* |
583 | * netpipe is used in the unprivileged case to transfer the ifindexes | |
584 | * from parent to child | |
585 | */ | |
586 | static int netpipe = -1; | |
587 | ||
588 | static inline int count_veths(struct lxc_list *network) | |
589 | { | |
590 | struct lxc_list *iterator; | |
591 | struct lxc_netdev *netdev; | |
592 | int count = 0; | |
593 | ||
594 | lxc_list_for_each(iterator, network) { | |
595 | netdev = iterator->elem; | |
596 | if (netdev->type != LXC_NET_VETH) | |
597 | continue; | |
598 | count++; | |
599 | } | |
600 | return count; | |
601 | } | |
602 | ||
603 | static int read_unpriv_netifindex(struct lxc_list *network) | |
604 | { | |
605 | struct lxc_list *iterator; | |
606 | struct lxc_netdev *netdev; | |
607 | ||
608 | if (netpipe == -1) | |
609 | return 0; | |
610 | lxc_list_for_each(iterator, network) { | |
611 | netdev = iterator->elem; | |
612 | if (netdev->type != LXC_NET_VETH) | |
613 | continue; | |
614 | if (!(netdev->name = malloc(IFNAMSIZ))) { | |
615 | ERROR("Out of memory"); | |
616 | close(netpipe); | |
617 | return -1; | |
618 | } | |
619 | if (read(netpipe, netdev->name, IFNAMSIZ) != IFNAMSIZ) { | |
620 | close(netpipe); | |
621 | return -1; | |
622 | } | |
623 | } | |
624 | close(netpipe); | |
625 | return 0; | |
626 | } | |
627 | ||
ffe1e01a | 628 | static int do_start(void *data) |
50e98013 | 629 | { |
7c661726 | 630 | struct lxc_list *iterator; |
23c53af9 | 631 | struct lxc_handler *handler = data; |
50e98013 DL |
632 | |
633 | if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) { | |
634 | SYSERROR("failed to set sigprocmask"); | |
9d7f9e52 | 635 | return -1; |
50e98013 DL |
636 | } |
637 | ||
743ecd2e DL |
638 | /* This prctl must be before the synchro, so if the parent |
639 | * dies before we set the parent death signal, we will detect | |
640 | * its death with the synchro right after, otherwise we have | |
641 | * a window where the parent can exit before we set the pdeath | |
642 | * signal leading to a unsupervized container. | |
643 | */ | |
644 | if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) { | |
645 | SYSERROR("failed to set pdeath signal"); | |
646 | return -1; | |
647 | } | |
648 | ||
3c22086f | 649 | lxc_sync_fini_parent(handler); |
50e98013 | 650 | |
2b0e17e4 | 651 | /* don't leak the pinfd to the container */ |
025ed0f3 | 652 | if (handler->pinfd >= 0) { |
0d03360a | 653 | close(handler->pinfd); |
025ed0f3 | 654 | } |
2b0e17e4 | 655 | |
3c22086f CLG |
656 | /* Tell the parent task it can begin to configure the |
657 | * container and wait for it to finish | |
658 | */ | |
659 | if (lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE)) | |
9d7f9e52 | 660 | return -1; |
50e98013 | 661 | |
658979c5 SH |
662 | if (read_unpriv_netifindex(&handler->conf->network) < 0) |
663 | goto out_warn_father; | |
664 | ||
f6d3e3e4 SH |
665 | /* |
666 | * if we are in a new user namespace, become root there to have | |
e06155c9 PT |
667 | * privilege over our namespace. We don't become root for lxc-execute, as |
668 | * the intent is to execute a command as the original user. | |
f6d3e3e4 | 669 | */ |
56f8ff00 PT |
670 | if (!lxc_list_empty(&handler->conf->id_map)) { |
671 | gid_t new_gid = handler->conf->is_execute ? handler->conf->parent_gid : 0; | |
672 | gid_t new_uid = handler->conf->is_execute ? handler->conf->parent_uid : 0; | |
673 | NOTICE("switching to gid/uid %d/%d in new user namespace", new_gid, new_uid); | |
674 | if (setgid(new_gid)) { | |
f6d3e3e4 SH |
675 | SYSERROR("setgid"); |
676 | goto out_warn_father; | |
677 | } | |
56f8ff00 | 678 | if (setuid(new_uid)) { |
f6d3e3e4 SH |
679 | SYSERROR("setuid"); |
680 | goto out_warn_father; | |
681 | } | |
c476bdce SH |
682 | if (setgroups(0, NULL)) { |
683 | SYSERROR("setgroups"); | |
684 | goto out_warn_father; | |
685 | } | |
f6d3e3e4 SH |
686 | } |
687 | ||
99b71824 | 688 | if (access(handler->lxcpath, X_OK)) { |
c8154066 SH |
689 | print_top_failing_dir(handler->lxcpath); |
690 | goto out_warn_father; | |
691 | } | |
692 | ||
495d2046 | 693 | #if HAVE_SYS_CAPABILITY_H |
69182a31 | 694 | if (handler->conf->need_utmp_watch) { |
828695d9 SH |
695 | if (prctl(PR_CAPBSET_DROP, CAP_SYS_BOOT, 0, 0, 0)) { |
696 | SYSERROR("failed to remove CAP_SYS_BOOT capability"); | |
c4ea60df | 697 | goto out_warn_father; |
828695d9 | 698 | } |
959aee9c | 699 | DEBUG("Dropped cap_sys_boot"); |
e2fa1520 | 700 | } |
495d2046 | 701 | #endif |
e2fa1520 SH |
702 | |
703 | /* Setup the container, ip, names, utsname, ... */ | |
d4ef7c50 | 704 | if (lxc_setup(handler)) { |
e2fa1520 SH |
705 | ERROR("failed to setup the container"); |
706 | goto out_warn_father; | |
707 | } | |
50e98013 | 708 | |
544a48a0 SH |
709 | /* ask father to setup cgroups and wait for him to finish */ |
710 | if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP)) | |
711 | return -1; | |
712 | ||
72863294 | 713 | /* Set the label to change to when we exec(2) the container's init */ |
7aff4f43 | 714 | if (lsm_process_label_set(NULL, handler->conf, 1, 1) < 0) |
e075f5d9 | 715 | goto out_warn_father; |
5112cd70 | 716 | |
0d9acb99 DE |
717 | /* Some init's such as busybox will set sane tty settings on stdin, |
718 | * stdout, stderr which it thinks is the console. We already set them | |
719 | * the way we wanted on the real terminal, and we want init to do its | |
720 | * setup on its console ie. the pty allocated in lxc_console_create() | |
721 | * so make sure that that pty is stdin,stdout,stderr. | |
722 | */ | |
723 | if (lxc_console_set_stdfds(handler) < 0) | |
724 | goto out_warn_father; | |
725 | ||
5112cd70 SH |
726 | /* If we mounted a temporary proc, then unmount it now */ |
727 | tmp_proc_unmount(handler->conf); | |
e075f5d9 | 728 | |
8f2c3a70 SH |
729 | if (lxc_seccomp_load(handler->conf) != 0) |
730 | goto out_warn_father; | |
731 | ||
283678ed | 732 | if (run_lxc_hooks(handler->name, "start", handler->conf, handler->lxcpath, NULL)) { |
773fb9ca SH |
733 | ERROR("failed to run start hooks for container '%s'.", handler->name); |
734 | goto out_warn_father; | |
735 | } | |
fc25b815 | 736 | |
f7bee6c6 | 737 | /* The clearenv() and putenv() calls have been moved here |
ec64264d | 738 | * to allow us to use environment variables passed to the various |
f7bee6c6 MW |
739 | * hooks, such as the start hook above. Not all of the |
740 | * variables like CONFIG_PATH or ROOTFS are valid in this | |
741 | * context but others are. */ | |
742 | if (clearenv()) { | |
743 | SYSERROR("failed to clear environment"); | |
744 | /* don't error out though */ | |
745 | } | |
746 | ||
7c661726 MP |
747 | lxc_list_for_each(iterator, &handler->conf->environment) { |
748 | if (putenv((char *)iterator->elem)) { | |
749 | SYSERROR("failed to set environment variable '%s'", (char *)iterator->elem); | |
750 | goto out_warn_father; | |
751 | } | |
752 | } | |
753 | ||
f7bee6c6 | 754 | if (putenv("container=lxc")) { |
7c661726 | 755 | SYSERROR("failed to set environment variable 'container=lxc'"); |
c4ea60df | 756 | goto out_warn_father; |
f7bee6c6 MW |
757 | } |
758 | ||
393903d1 SH |
759 | if (handler->conf->pty_names) { |
760 | if (putenv(handler->conf->pty_names)) { | |
761 | SYSERROR("failed to set environment variable for container ptys"); | |
762 | goto out_warn_father; | |
763 | } | |
764 | } | |
765 | ||
773fb9ca | 766 | close(handler->sigfd); |
26ddeedd | 767 | |
69aeabac TA |
768 | if (handler->backgrounded && null_stdfds() < 0) |
769 | goto out_warn_father; | |
507cee36 | 770 | |
e6126dbe MN |
771 | /* after this call, we are in error because this |
772 | * ops should not return as it execs */ | |
c4ea60df | 773 | handler->ops->start(handler, handler->data); |
50e98013 DL |
774 | |
775 | out_warn_father: | |
544a48a0 SH |
776 | /* we want the parent to know something went wrong, so any |
777 | * value other than what it expects is ok. */ | |
3c22086f | 778 | lxc_sync_wake_parent(handler, LXC_SYNC_POST_CONFIGURE); |
50e98013 DL |
779 | return -1; |
780 | } | |
781 | ||
74a3920a | 782 | static int save_phys_nics(struct lxc_conf *conf) |
7b35f3d6 SH |
783 | { |
784 | struct lxc_list *iterator; | |
785 | ||
786 | lxc_list_for_each(iterator, &conf->network) { | |
787 | struct lxc_netdev *netdev = iterator->elem; | |
788 | ||
789 | if (netdev->type != LXC_NET_PHYS) | |
790 | continue; | |
791 | conf->saved_nics = realloc(conf->saved_nics, | |
792 | (conf->num_savednics+1)*sizeof(struct saved_nic)); | |
793 | if (!conf->saved_nics) { | |
794 | SYSERROR("failed to allocate memory"); | |
795 | return -1; | |
796 | } | |
797 | conf->saved_nics[conf->num_savednics].ifindex = netdev->ifindex; | |
798 | conf->saved_nics[conf->num_savednics].orig_name = strdup(netdev->link); | |
799 | if (!conf->saved_nics[conf->num_savednics].orig_name) { | |
800 | SYSERROR("failed to allocate memory"); | |
801 | return -1; | |
802 | } | |
959aee9c | 803 | INFO("stored saved_nic #%d idx %d name %s", conf->num_savednics, |
7b35f3d6 SH |
804 | conf->saved_nics[conf->num_savednics].ifindex, |
805 | conf->saved_nics[conf->num_savednics].orig_name); | |
806 | conf->num_savednics++; | |
807 | } | |
808 | ||
809 | return 0; | |
810 | } | |
811 | ||
e8bd4e43 SH |
812 | static int recv_fd(int sock, int *fd) |
813 | { | |
814 | if (lxc_abstract_unix_recv_fd(sock, fd, NULL, 0) < 0) { | |
815 | SYSERROR("Error receiving tty fd from child"); | |
816 | return -1; | |
817 | } | |
818 | if (*fd == -1) | |
819 | return -1; | |
820 | return 0; | |
821 | } | |
822 | ||
823 | static int recv_ttys_from_child(struct lxc_handler *handler) | |
824 | { | |
825 | struct lxc_conf *conf = handler->conf; | |
826 | int i, sock = handler->ttysock[1]; | |
827 | struct lxc_tty_info *tty_info = &conf->tty_info; | |
828 | ||
829 | if (!conf->tty) | |
830 | return 0; | |
831 | ||
832 | tty_info->pty_info = malloc(sizeof(*tty_info->pty_info)*conf->tty); | |
833 | if (!tty_info->pty_info) { | |
834 | SYSERROR("failed to allocate pty_info"); | |
835 | return -1; | |
836 | } | |
837 | ||
838 | for (i = 0; i < conf->tty; i++) { | |
839 | struct lxc_pty_info *pty_info = &tty_info->pty_info[i]; | |
840 | pty_info->busy = 0; | |
841 | if (recv_fd(sock, &pty_info->slave) < 0 || | |
842 | recv_fd(sock, &pty_info->master) < 0) { | |
843 | ERROR("Error receiving tty info from child"); | |
844 | return -1; | |
845 | } | |
846 | } | |
847 | tty_info->nbtty = conf->tty; | |
848 | ||
849 | return 0; | |
850 | } | |
851 | ||
f813849c TA |
852 | void resolve_clone_flags(struct lxc_handler *handler) |
853 | { | |
854 | handler->clone_flags = CLONE_NEWPID | CLONE_NEWNS; | |
855 | ||
856 | if (!lxc_list_empty(&handler->conf->id_map)) { | |
857 | INFO("Cloning a new user namespace"); | |
858 | handler->clone_flags |= CLONE_NEWUSER; | |
859 | } | |
860 | ||
861 | if (handler->conf->inherit_ns_fd[LXC_NS_NET] == -1) { | |
862 | if (!lxc_requests_empty_network(handler)) | |
863 | handler->clone_flags |= CLONE_NEWNET; | |
864 | } else { | |
865 | INFO("Inheriting a net namespace"); | |
866 | } | |
867 | ||
868 | if (handler->conf->inherit_ns_fd[LXC_NS_IPC] == -1) { | |
869 | handler->clone_flags |= CLONE_NEWIPC; | |
870 | } else { | |
871 | INFO("Inheriting an IPC namespace"); | |
872 | } | |
873 | ||
874 | if (handler->conf->inherit_ns_fd[LXC_NS_UTS] == -1) { | |
875 | handler->clone_flags |= CLONE_NEWUTS; | |
876 | } else { | |
877 | INFO("Inheriting a UTS namespace"); | |
878 | } | |
879 | } | |
880 | ||
74a3920a | 881 | static int lxc_spawn(struct lxc_handler *handler) |
59eb99ba | 882 | { |
b98f7d6e | 883 | int failed_before_rename = 0; |
ffe1e01a | 884 | const char *name = handler->name; |
7e4dfe0b | 885 | bool cgroups_connected = false; |
9f30a190 MM |
886 | int saved_ns_fd[LXC_NS_MAX]; |
887 | int preserve_mask = 0, i; | |
658979c5 | 888 | int netpipepair[2], nveths; |
9f30a190 | 889 | |
2f2623ec SH |
890 | netpipe = -1; |
891 | ||
9f30a190 | 892 | for (i = 0; i < LXC_NS_MAX; i++) |
6c544cb3 | 893 | if (handler->conf->inherit_ns_fd[i] != -1) |
9f30a190 | 894 | preserve_mask |= ns_info[i].clone_flag; |
50e98013 | 895 | |
3c22086f | 896 | if (lxc_sync_init(handler)) |
9d7f9e52 | 897 | return -1; |
0ad19a3f | 898 | |
e8bd4e43 SH |
899 | if (socketpair(AF_UNIX, SOCK_DGRAM, 0, handler->ttysock) < 0) { |
900 | lxc_sync_fini(handler); | |
901 | return -1; | |
902 | } | |
903 | ||
f813849c | 904 | resolve_clone_flags(handler); |
9f30a190 | 905 | |
f813849c | 906 | if (handler->clone_flags & CLONE_NEWNET) { |
26b797f3 SH |
907 | if (!lxc_list_empty(&handler->conf->network)) { |
908 | ||
9f30a190 MM |
909 | /* Find gateway addresses from the link device, which is |
910 | * no longer accessible inside the container. Do this | |
911 | * before creating network interfaces, since goto | |
912 | * out_delete_net does not work before lxc_clone. */ | |
913 | if (lxc_find_gateway_addresses(handler)) { | |
914 | ERROR("failed to find gateway addresses"); | |
915 | lxc_sync_fini(handler); | |
916 | return -1; | |
917 | } | |
918 | ||
919 | /* that should be done before the clone because we will | |
920 | * fill the netdev index and use them in the child | |
921 | */ | |
922 | if (lxc_create_network(handler)) { | |
38521a3b | 923 | ERROR("failed to create the network"); |
9f30a190 MM |
924 | lxc_sync_fini(handler); |
925 | return -1; | |
926 | } | |
19a26f82 MK |
927 | } |
928 | ||
9f30a190 MM |
929 | if (save_phys_nics(handler->conf)) { |
930 | ERROR("failed to save physical nic info"); | |
931 | goto out_abort; | |
82d5ae15 DL |
932 | } |
933 | } | |
934 | ||
d4ef7c50 SH |
935 | if (!cgroup_init(handler)) { |
936 | ERROR("failed initializing cgroup support"); | |
33ad9f1a CS |
937 | goto out_delete_net; |
938 | } | |
939 | ||
7e4dfe0b SH |
940 | cgroups_connected = true; |
941 | ||
d4ef7c50 SH |
942 | if (!cgroup_create(handler)) { |
943 | ERROR("failed creating cgroups"); | |
47d8fb3b CS |
944 | goto out_delete_net; |
945 | } | |
946 | ||
0c547523 SH |
947 | /* |
948 | * if the rootfs is not a blockdev, prevent the container from | |
949 | * marking it readonly. | |
5e32a990 ÇO |
950 | * |
951 | * if the container is unprivileged then skip rootfs pinning | |
0c547523 | 952 | */ |
5e32a990 ÇO |
953 | if (lxc_list_empty(&handler->conf->id_map)) { |
954 | handler->pinfd = pin_rootfs(handler->conf->rootfs.path); | |
955 | if (handler->pinfd == -1) | |
956 | INFO("failed to pin the container's rootfs"); | |
957 | } | |
0c547523 | 958 | |
cd43d2d1 SH |
959 | if (preserve_ns(saved_ns_fd, preserve_mask) < 0) |
960 | goto out_delete_net; | |
961 | if (attach_ns(handler->conf->inherit_ns_fd) < 0) | |
962 | goto out_delete_net; | |
9f30a190 | 963 | |
658979c5 SH |
964 | if (am_unpriv() && (nveths = count_veths(&handler->conf->network))) { |
965 | if (pipe(netpipepair) < 0) { | |
966 | SYSERROR("Error creating pipe"); | |
967 | goto out_delete_net; | |
968 | } | |
969 | /* store netpipe in the global var for do_start's use */ | |
970 | netpipe = netpipepair[0]; | |
971 | } | |
972 | ||
0ad19a3f | 973 | /* Create a process in a new set of namespaces */ |
d5088cf2 | 974 | handler->pid = lxc_clone(do_start, handler, handler->clone_flags); |
59eb99ba | 975 | if (handler->pid < 0) { |
36eb9bde | 976 | SYSERROR("failed to fork into a new namespace"); |
7fef7a06 | 977 | goto out_delete_net; |
0ad19a3f | 978 | } |
979 | ||
cd43d2d1 SH |
980 | if (attach_ns(saved_ns_fd)) |
981 | WARN("failed to restore saved namespaces"); | |
9f30a190 | 982 | |
3c22086f CLG |
983 | lxc_sync_fini_child(handler); |
984 | ||
985 | if (lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE)) | |
99a6af52 | 986 | failed_before_rename = 1; |
0ad19a3f | 987 | |
d4ef7c50 SH |
988 | if (!cgroup_create_legacy(handler)) { |
989 | ERROR("failed to setup the legacy cgroups for %s", name); | |
ae5c8b8e | 990 | goto out_delete_net; |
33ad9f1a | 991 | } |
9daf6f5d SH |
992 | if (!cgroup_setup_limits(handler, false)) { |
993 | ERROR("failed to setup the cgroup limits for '%s'", name); | |
6031a6e5 DE |
994 | goto out_delete_net; |
995 | } | |
996 | ||
d4ef7c50 | 997 | if (!cgroup_enter(handler)) |
7fef7a06 | 998 | goto out_delete_net; |
218d4250 | 999 | |
0996e18a SH |
1000 | if (!cgroup_chown(handler)) |
1001 | goto out_delete_net; | |
1002 | ||
99a6af52 MN |
1003 | if (failed_before_rename) |
1004 | goto out_delete_net; | |
1005 | ||
0ad19a3f | 1006 | /* Create the network configuration */ |
d5088cf2 | 1007 | if (handler->clone_flags & CLONE_NEWNET) { |
fae349da | 1008 | if (lxc_assign_network(&handler->conf->network, handler->pid)) { |
82d5ae15 | 1009 | ERROR("failed to create the configured network"); |
7fef7a06 | 1010 | goto out_delete_net; |
82d5ae15 | 1011 | } |
0ad19a3f | 1012 | } |
1013 | ||
658979c5 SH |
1014 | if (netpipe != -1) { |
1015 | struct lxc_list *iterator; | |
1016 | struct lxc_netdev *netdev; | |
1017 | ||
1018 | close(netpipe); | |
1019 | lxc_list_for_each(iterator, &handler->conf->network) { | |
1020 | netdev = iterator->elem; | |
1021 | if (netdev->type != LXC_NET_VETH) | |
1022 | continue; | |
1023 | if (write(netpipepair[1], netdev->name, IFNAMSIZ) != IFNAMSIZ) { | |
1024 | ERROR("Error writing veth name to container"); | |
1025 | goto out_delete_net; | |
1026 | } | |
1027 | } | |
1028 | close(netpipepair[1]); | |
1029 | } | |
1030 | ||
f6d3e3e4 SH |
1031 | /* map the container uids - the container became an invalid |
1032 | * userid the moment it was cloned with CLONE_NEWUSER - this | |
1033 | * call doesn't change anything immediately, but allows the | |
1034 | * container to setuid(0) (0 being mapped to something else on | |
1035 | * the host) later to become a valid uid again */ | |
1036 | if (lxc_map_ids(&handler->conf->id_map, handler->pid)) { | |
1037 | ERROR("failed to set up id mapping"); | |
1038 | goto out_delete_net; | |
1039 | } | |
1040 | ||
544a48a0 SH |
1041 | /* Tell the child to continue its initialization. we'll get |
1042 | * LXC_SYNC_CGROUP when it is ready for us to setup cgroups | |
3c22086f CLG |
1043 | */ |
1044 | if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE)) | |
544a48a0 SH |
1045 | goto out_delete_net; |
1046 | ||
9daf6f5d | 1047 | if (!cgroup_setup_limits(handler, true)) { |
b98f7d6e SH |
1048 | ERROR("failed to setup the devices cgroup for '%s'", name); |
1049 | goto out_delete_net; | |
544a48a0 SH |
1050 | } |
1051 | ||
73d28d42 | 1052 | cgroup_disconnect(); |
7e4dfe0b | 1053 | cgroups_connected = false; |
73d28d42 | 1054 | |
e8bd4e43 SH |
1055 | /* read tty fds allocated by child */ |
1056 | if (recv_ttys_from_child(handler) < 0) { | |
1057 | ERROR("failed to receive tty info from child"); | |
1058 | goto out_delete_net; | |
1059 | } | |
1060 | ||
544a48a0 SH |
1061 | /* Tell the child to complete its initialization and wait for |
1062 | * it to exec or return an error. (the child will never | |
1063 | * return LXC_SYNC_POST_CGROUP+1. It will either close the | |
1064 | * sync pipe, causing lxc_sync_barrier_child to return | |
1065 | * success, or return a different value, causing us to error | |
1066 | * out). | |
1067 | */ | |
1068 | if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CGROUP)) | |
3c22086f | 1069 | return -1; |
0ad19a3f | 1070 | |
cc28d0b0 SH |
1071 | if (detect_shared_rootfs()) |
1072 | umount2(handler->conf->rootfs.mount, MNT_DETACH); | |
1073 | ||
23c53af9 | 1074 | if (handler->ops->post_start(handler, handler->data)) |
e6126dbe MN |
1075 | goto out_abort; |
1076 | ||
25c2aca5 | 1077 | if (lxc_set_state(name, handler, RUNNING)) { |
59eb99ba DL |
1078 | ERROR("failed to set state to %s", |
1079 | lxc_state2str(RUNNING)); | |
1080 | goto out_abort; | |
3f21c114 | 1081 | } |
22ebac19 | 1082 | |
3c22086f | 1083 | lxc_sync_fini(handler); |
0c547523 | 1084 | |
e6126dbe | 1085 | return 0; |
1ac470c0 | 1086 | |
7fef7a06 | 1087 | out_delete_net: |
7e4dfe0b SH |
1088 | if (cgroups_connected) |
1089 | cgroup_disconnect(); | |
d5088cf2 | 1090 | if (handler->clone_flags & CLONE_NEWNET) |
74a2b586 | 1091 | lxc_delete_network(handler); |
59eb99ba DL |
1092 | out_abort: |
1093 | lxc_abort(name, handler); | |
3c22086f | 1094 | lxc_sync_fini(handler); |
5c068da9 SH |
1095 | if (handler->pinfd >= 0) { |
1096 | close(handler->pinfd); | |
1097 | handler->pinfd = -1; | |
1098 | } | |
1099 | ||
b79fcd86 | 1100 | return -1; |
59eb99ba | 1101 | } |
0ad19a3f | 1102 | |
2af6bd1b SH |
1103 | int get_netns_fd(int pid) |
1104 | { | |
1105 | char path[MAXPATHLEN]; | |
1106 | int ret, fd; | |
1107 | ||
1108 | ret = snprintf(path, MAXPATHLEN, "/proc/%d/ns/net", pid); | |
1109 | if (ret < 0 || ret >= MAXPATHLEN) { | |
1110 | WARN("Failed to pin netns file for pid %d", pid); | |
1111 | return -1; | |
1112 | } | |
1113 | ||
1114 | fd = open(path, O_RDONLY); | |
1115 | if (fd < 0) { | |
1116 | WARN("Failed to pin netns file %s for pid %d: %s", | |
1117 | path, pid, strerror(errno)); | |
1118 | return -1; | |
1119 | } | |
1120 | return fd; | |
1121 | } | |
1122 | ||
ee70bf78 | 1123 | int __lxc_start(const char *name, struct lxc_conf *conf, |
507cee36 TA |
1124 | struct lxc_operations* ops, void *data, const char *lxcpath, |
1125 | bool backgrounded) | |
59eb99ba | 1126 | { |
3a0f472d | 1127 | struct lxc_handler *handler; |
e043236e | 1128 | int err = -1; |
59eb99ba | 1129 | int status; |
2af6bd1b | 1130 | int netnsfd = -1; |
80090207 | 1131 | |
13f5be62 | 1132 | handler = lxc_init(name, conf, lxcpath); |
3a0f472d | 1133 | if (!handler) { |
59eb99ba | 1134 | ERROR("failed to initialize the container"); |
66aeffc7 | 1135 | return -1; |
0ad19a3f | 1136 | } |
ee70bf78 CLG |
1137 | handler->ops = ops; |
1138 | handler->data = data; | |
507cee36 | 1139 | handler->backgrounded = backgrounded; |
e6126dbe | 1140 | |
b60ed720 | 1141 | if (must_drop_cap_sys_boot(handler->conf)) { |
495d2046 | 1142 | #if HAVE_SYS_CAPABILITY_H |
959aee9c | 1143 | DEBUG("Dropping cap_sys_boot"); |
495d2046 | 1144 | #else |
959aee9c | 1145 | DEBUG("Can't drop cap_sys_boot as capabilities aren't supported"); |
495d2046 | 1146 | #endif |
69182a31 | 1147 | } else { |
959aee9c | 1148 | DEBUG("Not dropping cap_sys_boot or watching utmp"); |
69182a31 SH |
1149 | handler->conf->need_utmp_watch = 0; |
1150 | } | |
1151 | ||
76a26f55 SH |
1152 | if (!attach_block_device(handler->conf)) { |
1153 | ERROR("Failure attaching block device"); | |
1154 | goto out_fini_nonet; | |
1155 | } | |
1156 | ||
35120d9c SH |
1157 | if (geteuid() == 0 && !lxc_list_empty(&conf->id_map)) { |
1158 | /* if the backing store is a device, mount it here and now */ | |
1159 | if (rootfs_is_blockdev(conf)) { | |
1160 | if (unshare(CLONE_NEWNS) < 0) { | |
1161 | ERROR("Error unsharing mounts"); | |
1162 | goto out_fini_nonet; | |
1163 | } | |
6a0c909a | 1164 | remount_all_slave(); |
35120d9c SH |
1165 | if (do_rootfs_setup(conf, name, lxcpath) < 0) { |
1166 | ERROR("Error setting up rootfs mount as root before spawn"); | |
1167 | goto out_fini_nonet; | |
1168 | } | |
1169 | INFO("Set up container rootfs as host root"); | |
1170 | } | |
1171 | } | |
1172 | ||
23c53af9 | 1173 | err = lxc_spawn(handler); |
59eb99ba | 1174 | if (err) { |
ee70bf78 | 1175 | ERROR("failed to spawn '%s'", name); |
76a26f55 | 1176 | goto out_detach_blockdev; |
0ad19a3f | 1177 | } |
1178 | ||
2af6bd1b SH |
1179 | netnsfd = get_netns_fd(handler->pid); |
1180 | ||
3a0f472d | 1181 | err = lxc_poll(name, handler); |
e043236e | 1182 | if (err) { |
59eb99ba | 1183 | ERROR("mainloop exited with an error"); |
4b2f98f5 SH |
1184 | if (netnsfd >= 0) |
1185 | close(netnsfd); | |
59eb99ba DL |
1186 | goto out_abort; |
1187 | } | |
0ad19a3f | 1188 | |
3a0f472d | 1189 | while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR) |
1bc5cc8c | 1190 | continue; |
e043236e | 1191 | |
8b004f07 SH |
1192 | /* |
1193 | * If the child process exited but was not signaled, | |
1194 | * it didn't call reboot. This should mean it was an | |
1195 | * lxc-execute which simply exited. In any case, treat | |
1196 | * it as a 'halt' | |
1197 | */ | |
1198 | if (WIFSIGNALED(status)) { | |
1199 | switch(WTERMSIG(status)) { | |
1200 | case SIGINT: /* halt */ | |
1201 | DEBUG("Container halting"); | |
1202 | break; | |
1203 | case SIGHUP: /* reboot */ | |
1204 | DEBUG("Container rebooting"); | |
1205 | handler->conf->reboot = 1; | |
1206 | break; | |
c2b9bd9e SH |
1207 | case SIGSYS: /* seccomp */ |
1208 | DEBUG("Container violated its seccomp policy"); | |
1209 | break; | |
8b004f07 | 1210 | default: |
959aee9c | 1211 | DEBUG("unknown exit status for init: %d", WTERMSIG(status)); |
8b004f07 SH |
1212 | break; |
1213 | } | |
828695d9 SH |
1214 | } |
1215 | ||
2af6bd1b | 1216 | lxc_rename_phys_nics_on_shutdown(netnsfd, handler->conf); |
4b2f98f5 SH |
1217 | if (netnsfd >= 0) |
1218 | close(netnsfd); | |
7b35f3d6 | 1219 | |
5c068da9 SH |
1220 | if (handler->pinfd >= 0) { |
1221 | close(handler->pinfd); | |
1222 | handler->pinfd = -1; | |
1223 | } | |
1224 | ||
1787abca | 1225 | lxc_monitor_send_exit_code(name, status, handler->lxcpath); |
3a0f472d | 1226 | err = lxc_error_set_and_log(handler->pid, status); |
9d7f9e52 | 1227 | out_fini: |
74a2b586 JK |
1228 | lxc_delete_network(handler); |
1229 | ||
76a26f55 SH |
1230 | out_detach_blockdev: |
1231 | detach_block_device(handler->conf); | |
1232 | ||
74a2b586 | 1233 | out_fini_nonet: |
3a0f472d | 1234 | lxc_fini(name, handler); |
0ad19a3f | 1235 | return err; |
1236 | ||
59eb99ba | 1237 | out_abort: |
3a0f472d | 1238 | lxc_abort(name, handler); |
9d7f9e52 | 1239 | goto out_fini; |
0ad19a3f | 1240 | } |
ee70bf78 CLG |
1241 | |
1242 | struct start_args { | |
1243 | char *const *argv; | |
1244 | }; | |
1245 | ||
1246 | static int start(struct lxc_handler *handler, void* data) | |
1247 | { | |
1248 | struct start_args *arg = data; | |
1249 | ||
1250 | NOTICE("exec'ing '%s'", arg->argv[0]); | |
1251 | ||
1252 | execvp(arg->argv[0], arg->argv); | |
1253 | SYSERROR("failed to exec %s", arg->argv[0]); | |
1254 | return 0; | |
1255 | } | |
1256 | ||
1257 | static int post_start(struct lxc_handler *handler, void* data) | |
1258 | { | |
1259 | struct start_args *arg = data; | |
1260 | ||
1261 | NOTICE("'%s' started with pid '%d'", arg->argv[0], handler->pid); | |
1262 | return 0; | |
1263 | } | |
1264 | ||
1265 | static struct lxc_operations start_ops = { | |
1266 | .start = start, | |
1267 | .post_start = post_start | |
1268 | }; | |
1269 | ||
13f5be62 | 1270 | int lxc_start(const char *name, char *const argv[], struct lxc_conf *conf, |
507cee36 | 1271 | const char *lxcpath, bool backgrounded) |
ee70bf78 CLG |
1272 | { |
1273 | struct start_args start_arg = { | |
1274 | .argv = argv, | |
1275 | }; | |
1276 | ||
828695d9 | 1277 | conf->need_utmp_watch = 1; |
507cee36 | 1278 | return __lxc_start(name, conf, &start_ops, &start_arg, lxcpath, backgrounded); |
ee70bf78 | 1279 | } |