]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * lxc: linux Container library | |
3 | * | |
4 | * (C) Copyright IBM Corp. 2007, 2008 | |
5 | * | |
6 | * Authors: | |
7 | * Daniel Lezcano <dlezcano at fr.ibm.com> | |
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 | |
21 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
22 | */ | |
23 | ||
24 | #include "../config.h" | |
25 | #include <stdio.h> | |
26 | #undef _GNU_SOURCE | |
27 | #include <string.h> | |
28 | #include <stdlib.h> | |
29 | #include <dirent.h> | |
30 | #include <errno.h> | |
31 | #include <unistd.h> | |
32 | #include <signal.h> | |
33 | #include <fcntl.h> | |
34 | #include <termios.h> | |
35 | #include <sys/param.h> | |
36 | #include <sys/file.h> | |
37 | #include <sys/mount.h> | |
38 | #include <sys/types.h> | |
39 | #include <sys/prctl.h> | |
40 | #include <sys/capability.h> | |
41 | #include <sys/wait.h> | |
42 | #include <sys/un.h> | |
43 | #include <sys/poll.h> | |
44 | ||
45 | #ifdef HAVE_SYS_SIGNALFD_H | |
46 | # include <sys/signalfd.h> | |
47 | #else | |
48 | # ifndef __NR_signalfd4 | |
49 | /* assume kernel headers are too old */ | |
50 | # if __i386__ | |
51 | # define __NR_signalfd4 327 | |
52 | # elif __x86_64__ | |
53 | # define __NR_signalfd4 289 | |
54 | # elif __powerpc__ | |
55 | # define __NR_signalfd4 313 | |
56 | # elif __s390x__ | |
57 | # define __NR_signalfd4 322 | |
58 | # endif | |
59 | #endif | |
60 | ||
61 | # ifndef __NR_signalfd | |
62 | /* assume kernel headers are too old */ | |
63 | # if __i386__ | |
64 | # define __NR_signalfd 321 | |
65 | # elif __x86_64__ | |
66 | # define __NR_signalfd 282 | |
67 | # elif __powerpc__ | |
68 | # define __NR_signalfd 305 | |
69 | # elif __s390x__ | |
70 | # define __NR_signalfd 316 | |
71 | # endif | |
72 | #endif | |
73 | ||
74 | int signalfd(int fd, const sigset_t *mask, int flags) | |
75 | { | |
76 | int retval; | |
77 | ||
78 | retval = syscall (__NR_signalfd4, fd, mask, _NSIG / 8, flags); | |
79 | if (errno == ENOSYS && flags == 0) | |
80 | retval = syscall (__NR_signalfd, fd, mask, _NSIG / 8); | |
81 | return retval; | |
82 | } | |
83 | #endif | |
84 | ||
85 | #if !HAVE_DECL_PR_CAPBSET_DROP | |
86 | #define PR_CAPBSET_DROP 24 | |
87 | #endif | |
88 | ||
89 | #include "error.h" | |
90 | #include "af_unix.h" | |
91 | #include "mainloop.h" | |
92 | ||
93 | #include <lxc/lxc.h> | |
94 | ||
95 | LXC_TTY_HANDLER(SIGINT); | |
96 | LXC_TTY_HANDLER(SIGQUIT); | |
97 | ||
98 | static int setup_sigchld_fd(sigset_t *oldmask) | |
99 | { | |
100 | sigset_t mask; | |
101 | int fd; | |
102 | ||
103 | if (sigprocmask(SIG_BLOCK, NULL, &mask)) { | |
104 | lxc_log_syserror("failed to get mask signal"); | |
105 | return -1; | |
106 | } | |
107 | ||
108 | if (sigaddset(&mask, SIGCHLD) || sigprocmask(SIG_BLOCK, &mask, oldmask)) { | |
109 | lxc_log_syserror("failed to set mask signal"); | |
110 | return -1; | |
111 | } | |
112 | ||
113 | fd = signalfd(-1, &mask, 0); | |
114 | if (fd < 0) { | |
115 | lxc_log_syserror("failed to create the signal fd"); | |
116 | return -1; | |
117 | } | |
118 | ||
119 | if (fcntl(fd, F_SETFD, FD_CLOEXEC)) { | |
120 | lxc_log_syserror("failed to set sigfd to close-on-exec"); | |
121 | close(fd); | |
122 | return -1; | |
123 | } | |
124 | ||
125 | return fd; | |
126 | } | |
127 | ||
128 | static int setup_tty_service(const char *name, int *ttyfd) | |
129 | { | |
130 | int fd; | |
131 | struct sockaddr_un addr = { 0 }; | |
132 | char *offset = &addr.sun_path[1]; | |
133 | ||
134 | strcpy(offset, name); | |
135 | addr.sun_path[0] = '\0'; | |
136 | ||
137 | fd = lxc_af_unix_open(addr.sun_path, SOCK_STREAM, 0); | |
138 | if (fd < 0) | |
139 | return -1; | |
140 | ||
141 | if (fcntl(fd, F_SETFD, FD_CLOEXEC)) { | |
142 | lxc_log_syserror("failed to close-on-exec flag"); | |
143 | close(fd); | |
144 | return -1; | |
145 | } | |
146 | ||
147 | *ttyfd = fd; | |
148 | ||
149 | return 0; | |
150 | } | |
151 | ||
152 | static int sigchld_handler(int fd, void *data, | |
153 | struct lxc_epoll_descr *descr) | |
154 | { | |
155 | pid_t *pid = data; | |
156 | ||
157 | waitpid(*pid, NULL, 0); | |
158 | ||
159 | return 1; | |
160 | } | |
161 | ||
162 | static int ttyclient_handler(int fd, void *data, | |
163 | struct lxc_epoll_descr *descr) | |
164 | { | |
165 | int i; | |
166 | struct lxc_tty_info *tty_info = data; | |
167 | ||
168 | for (i = 0; i < tty_info->nbtty; i++) { | |
169 | ||
170 | if (tty_info->pty_info[i].busy != fd) | |
171 | continue; | |
172 | ||
173 | lxc_mainloop_del_handler(descr, fd); | |
174 | tty_info->pty_info[i].busy = 0; | |
175 | close(fd); | |
176 | } | |
177 | ||
178 | return 0; | |
179 | } | |
180 | ||
181 | static int ttyservice_handler(int fd, void *data, | |
182 | struct lxc_epoll_descr *descr) | |
183 | { | |
184 | int conn, ttynum, val = 1, ret = -1; | |
185 | struct lxc_tty_info *tty_info = data; | |
186 | ||
187 | conn = accept(fd, NULL, 0); | |
188 | if (conn < 0) { | |
189 | lxc_log_syserror("failed to accept tty client"); | |
190 | return -1; | |
191 | } | |
192 | ||
193 | if (setsockopt(conn, SOL_SOCKET, SO_PASSCRED, &val, sizeof(val))) { | |
194 | lxc_log_syserror("failed to enable credential on socket"); | |
195 | goto out_close; | |
196 | } | |
197 | ||
198 | if (lxc_af_unix_rcv_credential(conn, &ttynum, sizeof(ttynum))) | |
199 | goto out_close; | |
200 | ||
201 | if (ttynum <= 0 || ttynum > tty_info->nbtty) | |
202 | goto out_close; | |
203 | ||
204 | /* fixup index array (eg. tty1 is index 0) */ | |
205 | ttynum--; | |
206 | ||
207 | if (tty_info->pty_info[ttynum].busy) | |
208 | goto out_close; | |
209 | ||
210 | if (lxc_af_unix_send_fd(conn, tty_info->pty_info[ttynum].master, | |
211 | NULL, 0) < 0) { | |
212 | lxc_log_error("failed to send tty to client"); | |
213 | goto out_close; | |
214 | } | |
215 | ||
216 | if (lxc_mainloop_add_handler(descr, conn, | |
217 | ttyclient_handler, tty_info)) { | |
218 | lxc_log_error("failed to add tty client handler"); | |
219 | goto out_close; | |
220 | } | |
221 | ||
222 | tty_info->pty_info[ttynum].busy = conn; | |
223 | ||
224 | ret = 0; | |
225 | ||
226 | out: | |
227 | return ret; | |
228 | out_close: | |
229 | close(conn); | |
230 | goto out; | |
231 | } | |
232 | ||
233 | static int mainloop(const char *name, pid_t pid, int sigfd, | |
234 | const struct lxc_tty_info *tty_info) | |
235 | { | |
236 | int nfds, ttyfd = -1, ret = -1; | |
237 | struct lxc_epoll_descr descr; | |
238 | ||
239 | if (tty_info->nbtty && setup_tty_service(name, &ttyfd)) { | |
240 | lxc_log_error("failed to create the tty service point"); | |
241 | goto out_sigfd; | |
242 | } | |
243 | ||
244 | /* sigfd + nb tty + tty service | |
245 | * if tty is enabled */ | |
246 | nfds = tty_info->nbtty + 1 + tty_info->nbtty ? 1 : 0; | |
247 | ||
248 | if (lxc_mainloop_open(nfds, &descr)) { | |
249 | lxc_log_error("failed to create mainloop"); | |
250 | goto out_ttyfd; | |
251 | } | |
252 | ||
253 | if (lxc_mainloop_add_handler(&descr, sigfd, sigchld_handler, &pid)) { | |
254 | lxc_log_error("failed to add handler for the signal"); | |
255 | goto out_mainloop_open; | |
256 | } | |
257 | ||
258 | if (tty_info->nbtty) { | |
259 | if (lxc_mainloop_add_handler(&descr, ttyfd, | |
260 | ttyservice_handler, | |
261 | (void *)tty_info)) { | |
262 | lxc_log_error("failed to add handler for the tty"); | |
263 | goto out_mainloop_open; | |
264 | } | |
265 | } | |
266 | ||
267 | ret = lxc_mainloop(&descr); | |
268 | ||
269 | out: | |
270 | return ret; | |
271 | ||
272 | out_mainloop_open: | |
273 | lxc_mainloop_close(&descr); | |
274 | out_ttyfd: | |
275 | close(ttyfd); | |
276 | out_sigfd: | |
277 | close(sigfd); | |
278 | goto out; | |
279 | } | |
280 | ||
281 | int lxc_start(const char *name, char *argv[]) | |
282 | { | |
283 | struct lxc_tty_info tty_info = { 0 }; | |
284 | sigset_t oldmask; | |
285 | char init[MAXPATHLEN]; | |
286 | char tty[MAXPATHLEN]; | |
287 | char *val = NULL; | |
288 | int fd, sigfd, lock, sv[2], sync = 0, err = -LXC_ERROR_INTERNAL; | |
289 | pid_t pid; | |
290 | int clone_flags; | |
291 | ||
292 | lock = lxc_get_lock(name); | |
293 | if (lock < 0) | |
294 | return lock; | |
295 | ||
296 | /* Begin the set the state to STARTING*/ | |
297 | if (lxc_setstate(name, STARTING)) { | |
298 | lxc_log_error("failed to set state '%s'", | |
299 | lxc_state2str(STARTING)); | |
300 | goto out; | |
301 | } | |
302 | ||
303 | /* If we are not attached to a tty, disable it */ | |
304 | if (ttyname_r(0, tty, sizeof(tty))) | |
305 | tty[0] = '\0'; | |
306 | ||
307 | if (lxc_create_tty(name, &tty_info)) { | |
308 | lxc_log_error("failed to create the ttys"); | |
309 | goto out; | |
310 | } | |
311 | ||
312 | /* the signal fd has to be created before forking otherwise | |
313 | * if the child process exits before we setup the signal fd, | |
314 | * the event will be lost and the command will be stuck */ | |
315 | sigfd = setup_sigchld_fd(&oldmask); | |
316 | if (sigfd < 0) { | |
317 | lxc_log_error("failed to set sigchild fd handler"); | |
318 | return -1; | |
319 | } | |
320 | ||
321 | /* Synchro socketpair */ | |
322 | if (socketpair(AF_LOCAL, SOCK_STREAM, 0, sv)) { | |
323 | lxc_log_syserror("failed to create communication socketpair"); | |
324 | goto out; | |
325 | } | |
326 | ||
327 | /* Avoid signals from terminal */ | |
328 | LXC_TTY_ADD_HANDLER(SIGINT); | |
329 | LXC_TTY_ADD_HANDLER(SIGQUIT); | |
330 | ||
331 | clone_flags = CLONE_NEWPID|CLONE_NEWIPC|CLONE_NEWNS; | |
332 | if (conf_has_utsname(name)) | |
333 | clone_flags |= CLONE_NEWUTS; | |
334 | if (conf_has_network(name)) | |
335 | clone_flags |= CLONE_NEWNET; | |
336 | ||
337 | /* Create a process in a new set of namespaces */ | |
338 | pid = fork_ns(clone_flags); | |
339 | if (pid < 0) { | |
340 | lxc_log_syserror("failed to fork into a new namespace"); | |
341 | goto err_fork_ns; | |
342 | } | |
343 | ||
344 | if (!pid) { | |
345 | ||
346 | if (sigprocmask(SIG_SETMASK, &oldmask, NULL)) { | |
347 | lxc_log_syserror("failed to set sigprocmask"); | |
348 | return -1; | |
349 | } | |
350 | ||
351 | close(sv[1]); | |
352 | ||
353 | /* Be sure we don't inherit this after the exec */ | |
354 | fcntl(sv[0], F_SETFD, FD_CLOEXEC); | |
355 | ||
356 | /* Tell our father he can begin to configure the container */ | |
357 | if (write(sv[0], &sync, sizeof(sync)) < 0) { | |
358 | lxc_log_syserror("failed to write socket"); | |
359 | goto out_child; | |
360 | } | |
361 | ||
362 | /* Wait for the father to finish the configuration */ | |
363 | if (read(sv[0], &sync, sizeof(sync)) < 0) { | |
364 | lxc_log_syserror("failed to read socket"); | |
365 | goto out_child; | |
366 | } | |
367 | ||
368 | /* Setup the container, ip, names, utsname, ... */ | |
369 | err = lxc_setup(name, tty, &tty_info); | |
370 | if (err) { | |
371 | lxc_log_error("failed to setup the container"); | |
372 | if (write(sv[0], &err, sizeof(err)) < 0) | |
373 | lxc_log_syserror("failed to write the socket"); | |
374 | goto out_child; | |
375 | } | |
376 | ||
377 | if (prctl(PR_CAPBSET_DROP, CAP_SYS_BOOT, 0, 0, 0)) { | |
378 | lxc_log_syserror("failed to remove CAP_SYS_BOOT capability"); | |
379 | goto out_child; | |
380 | } | |
381 | ||
382 | execvp(argv[0], argv); | |
383 | lxc_log_syserror("failed to exec %s", argv[0]); | |
384 | ||
385 | err = LXC_ERROR_WRONG_COMMAND; | |
386 | /* If the exec fails, tell that to our father */ | |
387 | if (write(sv[0], &err, sizeof(err)) < 0) | |
388 | lxc_log_syserror("failed to write the socket"); | |
389 | ||
390 | out_child: | |
391 | exit(err); | |
392 | } | |
393 | ||
394 | close(sv[0]); | |
395 | ||
396 | /* Wait for the child to be ready */ | |
397 | if (read(sv[1], &sync, sizeof(sync)) < 0) { | |
398 | lxc_log_syserror("failed to read the socket"); | |
399 | goto err_pipe_read; | |
400 | } | |
401 | ||
402 | if (lxc_link_nsgroup(name, pid)) | |
403 | lxc_log_warning("cgroupfs not found: cgroup disabled"); | |
404 | ||
405 | /* Create the network configuration */ | |
406 | if (clone_flags & CLONE_NEWNET && conf_create_network(name, pid)) { | |
407 | lxc_log_error("failed to create the configured network"); | |
408 | goto err_create_network; | |
409 | } | |
410 | ||
411 | /* Tell the child to continue its initialization */ | |
412 | if (write(sv[1], &sync, sizeof(sync)) < 0) { | |
413 | lxc_log_syserror("failed to write the socket"); | |
414 | goto err_pipe_write; | |
415 | } | |
416 | ||
417 | /* Wait for the child to exec or returning an error */ | |
418 | err = read(sv[1], &sync, sizeof(sync)); | |
419 | if (err < 0) { | |
420 | lxc_log_error("failed to read the socket"); | |
421 | goto err_pipe_read2; | |
422 | } | |
423 | ||
424 | if (err > 0) { | |
425 | err = sync; | |
426 | waitpid(pid, NULL, 0); | |
427 | goto err_child_failed; | |
428 | } | |
429 | ||
430 | asprintf(&val, "%d\n", pid); | |
431 | ||
432 | snprintf(init, MAXPATHLEN, LXCPATH "/%s/init", name); | |
433 | ||
434 | fd = open(init, O_WRONLY|O_CREAT|O_TRUNC, S_IRUSR|S_IWUSR); | |
435 | if (fd < 0) { | |
436 | lxc_log_syserror("failed to open '%s'", init); | |
437 | goto err_write; | |
438 | } | |
439 | ||
440 | if (write(fd, val, strlen(val)) < 0) { | |
441 | lxc_log_syserror("failed to write the init pid"); | |
442 | goto err_write; | |
443 | } | |
444 | ||
445 | close(fd); | |
446 | ||
447 | if (lxc_setstate(name, RUNNING)) { | |
448 | lxc_log_error("failed to set state to %s", | |
449 | lxc_state2str(RUNNING)); | |
450 | goto err_state_failed; | |
451 | } | |
452 | ||
453 | if (mainloop(name, pid, sigfd, &tty_info)) { | |
454 | lxc_log_error("mainloop exited with an error"); | |
455 | goto err_mailoop_failed; | |
456 | } | |
457 | ||
458 | if (lxc_setstate(name, STOPPING)) | |
459 | lxc_log_error("failed to set state %s", lxc_state2str(STOPPING)); | |
460 | ||
461 | if (clone_flags & CLONE_NEWNET && conf_destroy_network(name)) | |
462 | lxc_log_error("failed to destroy the network"); | |
463 | ||
464 | err = 0; | |
465 | out: | |
466 | if (lxc_setstate(name, STOPPED)) | |
467 | lxc_log_error("failed to set state %s", lxc_state2str(STOPPED)); | |
468 | ||
469 | lxc_delete_tty(&tty_info); | |
470 | lxc_unlink_nsgroup(name); | |
471 | unlink(init); | |
472 | free(val); | |
473 | lxc_put_lock(lock); | |
474 | LXC_TTY_DEL_HANDLER(SIGQUIT); | |
475 | LXC_TTY_DEL_HANDLER(SIGINT); | |
476 | ||
477 | return err; | |
478 | ||
479 | err_write: | |
480 | close(fd); | |
481 | ||
482 | err_state_failed: | |
483 | err_child_failed: | |
484 | err_pipe_read2: | |
485 | err_pipe_write: | |
486 | if (clone_flags & CLONE_NEWNET) | |
487 | conf_destroy_network(name); | |
488 | err_create_network: | |
489 | err_pipe_read: | |
490 | err_mailoop_failed: | |
491 | if (lxc_setstate(name, ABORTING)) | |
492 | lxc_log_error("failed to set state %s", lxc_state2str(STOPPED)); | |
493 | ||
494 | kill(pid, SIGKILL); | |
495 | err_fork_ns: | |
496 | close(sv[0]); | |
497 | close(sv[1]); | |
498 | goto out; | |
499 | } |