xref: /freebsd/sys/kern/kern_jail.c (revision 87569f75a91f298c52a71823c04d41cf53c88889)
1 /*-
2  * ----------------------------------------------------------------------------
3  * "THE BEER-WARE LICENSE" (Revision 42):
4  * <phk@FreeBSD.ORG> wrote this file.  As long as you retain this notice you
5  * can do whatever you want with this stuff. If we meet some day, and you think
6  * this stuff is worth it, you can buy me a beer in return.   Poul-Henning Kamp
7  * ----------------------------------------------------------------------------
8  */
9 
10 #include <sys/cdefs.h>
11 __FBSDID("$FreeBSD$");
12 
13 #include "opt_mac.h"
14 
15 #include <sys/param.h>
16 #include <sys/types.h>
17 #include <sys/kernel.h>
18 #include <sys/systm.h>
19 #include <sys/errno.h>
20 #include <sys/sysproto.h>
21 #include <sys/mac.h>
22 #include <sys/malloc.h>
23 #include <sys/proc.h>
24 #include <sys/taskqueue.h>
25 #include <sys/jail.h>
26 #include <sys/lock.h>
27 #include <sys/mutex.h>
28 #include <sys/namei.h>
29 #include <sys/mount.h>
30 #include <sys/queue.h>
31 #include <sys/socket.h>
32 #include <sys/syscallsubr.h>
33 #include <sys/sysctl.h>
34 #include <sys/vnode.h>
35 #include <net/if.h>
36 #include <netinet/in.h>
37 
38 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
39 
40 SYSCTL_DECL(_security);
41 SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW, 0,
42     "Jail rules");
43 
44 int	jail_set_hostname_allowed = 1;
45 SYSCTL_INT(_security_jail, OID_AUTO, set_hostname_allowed, CTLFLAG_RW,
46     &jail_set_hostname_allowed, 0,
47     "Processes in jail can set their hostnames");
48 
49 int	jail_socket_unixiproute_only = 1;
50 SYSCTL_INT(_security_jail, OID_AUTO, socket_unixiproute_only, CTLFLAG_RW,
51     &jail_socket_unixiproute_only, 0,
52     "Processes in jail are limited to creating UNIX/IPv4/route sockets only");
53 
54 int	jail_sysvipc_allowed = 0;
55 SYSCTL_INT(_security_jail, OID_AUTO, sysvipc_allowed, CTLFLAG_RW,
56     &jail_sysvipc_allowed, 0,
57     "Processes in jail can use System V IPC primitives");
58 
59 static int jail_enforce_statfs = 2;
60 SYSCTL_INT(_security_jail, OID_AUTO, enforce_statfs, CTLFLAG_RW,
61     &jail_enforce_statfs, 0,
62     "Processes in jail cannot see all mounted file systems");
63 
64 int	jail_allow_raw_sockets = 0;
65 SYSCTL_INT(_security_jail, OID_AUTO, allow_raw_sockets, CTLFLAG_RW,
66     &jail_allow_raw_sockets, 0,
67     "Prison root can create raw sockets");
68 
69 int	jail_chflags_allowed = 0;
70 SYSCTL_INT(_security_jail, OID_AUTO, chflags_allowed, CTLFLAG_RW,
71     &jail_chflags_allowed, 0,
72     "Processes in jail can alter system file flags");
73 
74 /* allprison, lastprid, and prisoncount are protected by allprison_mtx. */
75 struct	prisonlist allprison;
76 struct	mtx allprison_mtx;
77 int	lastprid = 0;
78 int	prisoncount = 0;
79 
80 static void		 init_prison(void *);
81 static void		 prison_complete(void *context, int pending);
82 static struct prison	*prison_find(int);
83 static int		 sysctl_jail_list(SYSCTL_HANDLER_ARGS);
84 
85 static void
86 init_prison(void *data __unused)
87 {
88 
89 	mtx_init(&allprison_mtx, "allprison", NULL, MTX_DEF);
90 	LIST_INIT(&allprison);
91 }
92 
93 SYSINIT(prison, SI_SUB_INTRINSIC, SI_ORDER_ANY, init_prison, NULL);
94 
95 /*
96  * MPSAFE
97  *
98  * struct jail_args {
99  *	struct jail *jail;
100  * };
101  */
102 int
103 jail(struct thread *td, struct jail_args *uap)
104 {
105 	struct nameidata nd;
106 	struct prison *pr, *tpr;
107 	struct jail j;
108 	struct jail_attach_args jaa;
109 	int vfslocked, error, tryprid;
110 
111 	error = copyin(uap->jail, &j, sizeof(j));
112 	if (error)
113 		return (error);
114 	if (j.version != 0)
115 		return (EINVAL);
116 
117 	MALLOC(pr, struct prison *, sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
118 	mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF);
119 	pr->pr_ref = 1;
120 	error = copyinstr(j.path, &pr->pr_path, sizeof(pr->pr_path), 0);
121 	if (error)
122 		goto e_killmtx;
123 	NDINIT(&nd, LOOKUP, MPSAFE | FOLLOW | LOCKLEAF, UIO_SYSSPACE,
124 	    pr->pr_path, td);
125 	error = namei(&nd);
126 	if (error)
127 		goto e_killmtx;
128 	vfslocked = NDHASGIANT(&nd);
129 	pr->pr_root = nd.ni_vp;
130 	VOP_UNLOCK(nd.ni_vp, 0, td);
131 	NDFREE(&nd, NDF_ONLY_PNBUF);
132 	VFS_UNLOCK_GIANT(vfslocked);
133 	error = copyinstr(j.hostname, &pr->pr_host, sizeof(pr->pr_host), 0);
134 	if (error)
135 		goto e_dropvnref;
136 	pr->pr_ip = j.ip_number;
137 	pr->pr_linux = NULL;
138 	pr->pr_securelevel = securelevel;
139 
140 	/* Determine next pr_id and add prison to allprison list. */
141 	mtx_lock(&allprison_mtx);
142 	tryprid = lastprid + 1;
143 	if (tryprid == JAIL_MAX)
144 		tryprid = 1;
145 next:
146 	LIST_FOREACH(tpr, &allprison, pr_list) {
147 		if (tpr->pr_id == tryprid) {
148 			tryprid++;
149 			if (tryprid == JAIL_MAX) {
150 				mtx_unlock(&allprison_mtx);
151 				error = EAGAIN;
152 				goto e_dropvnref;
153 			}
154 			goto next;
155 		}
156 	}
157 	pr->pr_id = jaa.jid = lastprid = tryprid;
158 	LIST_INSERT_HEAD(&allprison, pr, pr_list);
159 	prisoncount++;
160 	mtx_unlock(&allprison_mtx);
161 
162 	error = jail_attach(td, &jaa);
163 	if (error)
164 		goto e_dropprref;
165 	mtx_lock(&pr->pr_mtx);
166 	pr->pr_ref--;
167 	mtx_unlock(&pr->pr_mtx);
168 	td->td_retval[0] = jaa.jid;
169 	return (0);
170 e_dropprref:
171 	mtx_lock(&allprison_mtx);
172 	LIST_REMOVE(pr, pr_list);
173 	prisoncount--;
174 	mtx_unlock(&allprison_mtx);
175 e_dropvnref:
176 	vfslocked = VFS_LOCK_GIANT(pr->pr_root->v_mount);
177 	vrele(pr->pr_root);
178 	VFS_UNLOCK_GIANT(vfslocked);
179 e_killmtx:
180 	mtx_destroy(&pr->pr_mtx);
181 	FREE(pr, M_PRISON);
182 	return (error);
183 }
184 
185 /*
186  * MPSAFE
187  *
188  * struct jail_attach_args {
189  *	int jid;
190  * };
191  */
192 int
193 jail_attach(struct thread *td, struct jail_attach_args *uap)
194 {
195 	struct proc *p;
196 	struct ucred *newcred, *oldcred;
197 	struct prison *pr;
198 	int vfslocked, error;
199 
200 	/*
201 	 * XXX: Note that there is a slight race here if two threads
202 	 * in the same privileged process attempt to attach to two
203 	 * different jails at the same time.  It is important for
204 	 * user processes not to do this, or they might end up with
205 	 * a process root from one prison, but attached to the jail
206 	 * of another.
207 	 */
208 	error = suser(td);
209 	if (error)
210 		return (error);
211 
212 	p = td->td_proc;
213 	mtx_lock(&allprison_mtx);
214 	pr = prison_find(uap->jid);
215 	if (pr == NULL) {
216 		mtx_unlock(&allprison_mtx);
217 		return (EINVAL);
218 	}
219 	pr->pr_ref++;
220 	mtx_unlock(&pr->pr_mtx);
221 	mtx_unlock(&allprison_mtx);
222 
223 	vfslocked = VFS_LOCK_GIANT(pr->pr_root->v_mount);
224 	vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY, td);
225 	if ((error = change_dir(pr->pr_root, td)) != 0)
226 		goto e_unlock;
227 #ifdef MAC
228 	if ((error = mac_check_vnode_chroot(td->td_ucred, pr->pr_root)))
229 		goto e_unlock;
230 #endif
231 	VOP_UNLOCK(pr->pr_root, 0, td);
232 	change_root(pr->pr_root, td);
233 	VFS_UNLOCK_GIANT(vfslocked);
234 
235 	newcred = crget();
236 	PROC_LOCK(p);
237 	oldcred = p->p_ucred;
238 	setsugid(p);
239 	crcopy(newcred, oldcred);
240 	newcred->cr_prison = pr;
241 	p->p_ucred = newcred;
242 	PROC_UNLOCK(p);
243 	crfree(oldcred);
244 	return (0);
245 e_unlock:
246 	VOP_UNLOCK(pr->pr_root, 0, td);
247 	VFS_UNLOCK_GIANT(vfslocked);
248 	mtx_lock(&pr->pr_mtx);
249 	pr->pr_ref--;
250 	mtx_unlock(&pr->pr_mtx);
251 	return (error);
252 }
253 
254 /*
255  * Returns a locked prison instance, or NULL on failure.
256  */
257 static struct prison *
258 prison_find(int prid)
259 {
260 	struct prison *pr;
261 
262 	mtx_assert(&allprison_mtx, MA_OWNED);
263 	LIST_FOREACH(pr, &allprison, pr_list) {
264 		if (pr->pr_id == prid) {
265 			mtx_lock(&pr->pr_mtx);
266 			return (pr);
267 		}
268 	}
269 	return (NULL);
270 }
271 
272 void
273 prison_free(struct prison *pr)
274 {
275 
276 	mtx_lock(&allprison_mtx);
277 	mtx_lock(&pr->pr_mtx);
278 	pr->pr_ref--;
279 	if (pr->pr_ref == 0) {
280 		LIST_REMOVE(pr, pr_list);
281 		mtx_unlock(&pr->pr_mtx);
282 		prisoncount--;
283 		mtx_unlock(&allprison_mtx);
284 
285 		TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
286 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
287 		return;
288 	}
289 	mtx_unlock(&pr->pr_mtx);
290 	mtx_unlock(&allprison_mtx);
291 }
292 
293 static void
294 prison_complete(void *context, int pending)
295 {
296 	struct prison *pr;
297 	int vfslocked;
298 
299 	pr = (struct prison *)context;
300 
301 	vfslocked = VFS_LOCK_GIANT(pr->pr_root->v_mount);
302 	vrele(pr->pr_root);
303 	VFS_UNLOCK_GIANT(vfslocked);
304 
305 	mtx_destroy(&pr->pr_mtx);
306 	if (pr->pr_linux != NULL)
307 		FREE(pr->pr_linux, M_PRISON);
308 	FREE(pr, M_PRISON);
309 }
310 
311 void
312 prison_hold(struct prison *pr)
313 {
314 
315 	mtx_lock(&pr->pr_mtx);
316 	pr->pr_ref++;
317 	mtx_unlock(&pr->pr_mtx);
318 }
319 
320 u_int32_t
321 prison_getip(struct ucred *cred)
322 {
323 
324 	return (cred->cr_prison->pr_ip);
325 }
326 
327 int
328 prison_ip(struct ucred *cred, int flag, u_int32_t *ip)
329 {
330 	u_int32_t tmp;
331 
332 	if (!jailed(cred))
333 		return (0);
334 	if (flag)
335 		tmp = *ip;
336 	else
337 		tmp = ntohl(*ip);
338 	if (tmp == INADDR_ANY) {
339 		if (flag)
340 			*ip = cred->cr_prison->pr_ip;
341 		else
342 			*ip = htonl(cred->cr_prison->pr_ip);
343 		return (0);
344 	}
345 	if (tmp == INADDR_LOOPBACK) {
346 		if (flag)
347 			*ip = cred->cr_prison->pr_ip;
348 		else
349 			*ip = htonl(cred->cr_prison->pr_ip);
350 		return (0);
351 	}
352 	if (cred->cr_prison->pr_ip != tmp)
353 		return (1);
354 	return (0);
355 }
356 
357 void
358 prison_remote_ip(struct ucred *cred, int flag, u_int32_t *ip)
359 {
360 	u_int32_t tmp;
361 
362 	if (!jailed(cred))
363 		return;
364 	if (flag)
365 		tmp = *ip;
366 	else
367 		tmp = ntohl(*ip);
368 	if (tmp == INADDR_LOOPBACK) {
369 		if (flag)
370 			*ip = cred->cr_prison->pr_ip;
371 		else
372 			*ip = htonl(cred->cr_prison->pr_ip);
373 		return;
374 	}
375 	return;
376 }
377 
378 int
379 prison_if(struct ucred *cred, struct sockaddr *sa)
380 {
381 	struct sockaddr_in *sai;
382 	int ok;
383 
384 	sai = (struct sockaddr_in *)sa;
385 	if ((sai->sin_family != AF_INET) && jail_socket_unixiproute_only)
386 		ok = 1;
387 	else if (sai->sin_family != AF_INET)
388 		ok = 0;
389 	else if (cred->cr_prison->pr_ip != ntohl(sai->sin_addr.s_addr))
390 		ok = 1;
391 	else
392 		ok = 0;
393 	return (ok);
394 }
395 
396 /*
397  * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
398  */
399 int
400 prison_check(struct ucred *cred1, struct ucred *cred2)
401 {
402 
403 	if (jailed(cred1)) {
404 		if (!jailed(cred2))
405 			return (ESRCH);
406 		if (cred2->cr_prison != cred1->cr_prison)
407 			return (ESRCH);
408 	}
409 
410 	return (0);
411 }
412 
413 /*
414  * Return 1 if the passed credential is in a jail, otherwise 0.
415  */
416 int
417 jailed(struct ucred *cred)
418 {
419 
420 	return (cred->cr_prison != NULL);
421 }
422 
423 /*
424  * Return the correct hostname for the passed credential.
425  */
426 void
427 getcredhostname(struct ucred *cred, char *buf, size_t size)
428 {
429 
430 	if (jailed(cred)) {
431 		mtx_lock(&cred->cr_prison->pr_mtx);
432 		strlcpy(buf, cred->cr_prison->pr_host, size);
433 		mtx_unlock(&cred->cr_prison->pr_mtx);
434 	} else
435 		strlcpy(buf, hostname, size);
436 }
437 
438 /*
439  * Determine whether the subject represented by cred can "see"
440  * status of a mount point.
441  * Returns: 0 for permitted, ENOENT otherwise.
442  * XXX: This function should be called cr_canseemount() and should be
443  *      placed in kern_prot.c.
444  */
445 int
446 prison_canseemount(struct ucred *cred, struct mount *mp)
447 {
448 	struct prison *pr;
449 	struct statfs *sp;
450 	size_t len;
451 
452 	if (!jailed(cred) || jail_enforce_statfs == 0)
453 		return (0);
454 	pr = cred->cr_prison;
455 	if (pr->pr_root->v_mount == mp)
456 		return (0);
457 	if (jail_enforce_statfs == 2)
458 		return (ENOENT);
459 	/*
460 	 * If jail's chroot directory is set to "/" we should be able to see
461 	 * all mount-points from inside a jail.
462 	 * This is ugly check, but this is the only situation when jail's
463 	 * directory ends with '/'.
464 	 */
465 	if (strcmp(pr->pr_path, "/") == 0)
466 		return (0);
467 	len = strlen(pr->pr_path);
468 	sp = &mp->mnt_stat;
469 	if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
470 		return (ENOENT);
471 	/*
472 	 * Be sure that we don't have situation where jail's root directory
473 	 * is "/some/path" and mount point is "/some/pathpath".
474 	 */
475 	if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
476 		return (ENOENT);
477 	return (0);
478 }
479 
480 void
481 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
482 {
483 	char jpath[MAXPATHLEN];
484 	struct prison *pr;
485 	size_t len;
486 
487 	if (!jailed(cred) || jail_enforce_statfs == 0)
488 		return;
489 	pr = cred->cr_prison;
490 	if (prison_canseemount(cred, mp) != 0) {
491 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
492 		strlcpy(sp->f_mntonname, "[restricted]",
493 		    sizeof(sp->f_mntonname));
494 		return;
495 	}
496 	if (pr->pr_root->v_mount == mp) {
497 		/*
498 		 * Clear current buffer data, so we are sure nothing from
499 		 * the valid path left there.
500 		 */
501 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
502 		*sp->f_mntonname = '/';
503 		return;
504 	}
505 	/*
506 	 * If jail's chroot directory is set to "/" we should be able to see
507 	 * all mount-points from inside a jail.
508 	 */
509 	if (strcmp(pr->pr_path, "/") == 0)
510 		return;
511 	len = strlen(pr->pr_path);
512 	strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
513 	/*
514 	 * Clear current buffer data, so we are sure nothing from
515 	 * the valid path left there.
516 	 */
517 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
518 	if (*jpath == '\0') {
519 		/* Should never happen. */
520 		*sp->f_mntonname = '/';
521 	} else {
522 		strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
523 	}
524 }
525 
526 static int
527 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
528 {
529 	struct xprison *xp, *sxp;
530 	struct prison *pr;
531 	int count, error;
532 
533 	if (jailed(req->td->td_ucred))
534 		return (0);
535 retry:
536 	mtx_lock(&allprison_mtx);
537 	count = prisoncount;
538 	mtx_unlock(&allprison_mtx);
539 
540 	if (count == 0)
541 		return (0);
542 
543 	sxp = xp = malloc(sizeof(*xp) * count, M_TEMP, M_WAITOK | M_ZERO);
544 	mtx_lock(&allprison_mtx);
545 	if (count != prisoncount) {
546 		mtx_unlock(&allprison_mtx);
547 		free(sxp, M_TEMP);
548 		goto retry;
549 	}
550 
551 	LIST_FOREACH(pr, &allprison, pr_list) {
552 		mtx_lock(&pr->pr_mtx);
553 		xp->pr_version = XPRISON_VERSION;
554 		xp->pr_id = pr->pr_id;
555 		strlcpy(xp->pr_path, pr->pr_path, sizeof(xp->pr_path));
556 		strlcpy(xp->pr_host, pr->pr_host, sizeof(xp->pr_host));
557 		xp->pr_ip = pr->pr_ip;
558 		mtx_unlock(&pr->pr_mtx);
559 		xp++;
560 	}
561 	mtx_unlock(&allprison_mtx);
562 
563 	error = SYSCTL_OUT(req, sxp, sizeof(*sxp) * count);
564 	free(sxp, M_TEMP);
565 	if (error)
566 		return (error);
567 	return (0);
568 }
569 
570 SYSCTL_OID(_security_jail, OID_AUTO, list, CTLTYPE_STRUCT | CTLFLAG_RD,
571     NULL, 0, sysctl_jail_list, "S", "List of active jails");
572 
573 static int
574 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
575 {
576 	int error, injail;
577 
578 	injail = jailed(req->td->td_ucred);
579 	error = SYSCTL_OUT(req, &injail, sizeof(injail));
580 
581 	return (error);
582 }
583 SYSCTL_PROC(_security_jail, OID_AUTO, jailed, CTLTYPE_INT | CTLFLAG_RD,
584     NULL, 0, sysctl_jail_jailed, "I", "Process in jail?");
585