xref: /freebsd/sys/kern/kern_jail.c (revision 271c3a9060f2ee55607ebe146523f888e1db2654)
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/malloc.h>
22 #include <sys/priv.h>
23 #include <sys/proc.h>
24 #include <sys/taskqueue.h>
25 #include <sys/fcntl.h>
26 #include <sys/jail.h>
27 #include <sys/lock.h>
28 #include <sys/mutex.h>
29 #include <sys/sx.h>
30 #include <sys/namei.h>
31 #include <sys/mount.h>
32 #include <sys/queue.h>
33 #include <sys/socket.h>
34 #include <sys/syscallsubr.h>
35 #include <sys/sysctl.h>
36 #include <sys/vnode.h>
37 #include <sys/vimage.h>
38 #include <net/if.h>
39 #include <netinet/in.h>
40 
41 #include <security/mac/mac_framework.h>
42 
43 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
44 
45 SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW, 0,
46     "Jail rules");
47 
48 int	jail_set_hostname_allowed = 1;
49 SYSCTL_INT(_security_jail, OID_AUTO, set_hostname_allowed, CTLFLAG_RW,
50     &jail_set_hostname_allowed, 0,
51     "Processes in jail can set their hostnames");
52 
53 int	jail_socket_unixiproute_only = 1;
54 SYSCTL_INT(_security_jail, OID_AUTO, socket_unixiproute_only, CTLFLAG_RW,
55     &jail_socket_unixiproute_only, 0,
56     "Processes in jail are limited to creating UNIX/IPv4/route sockets only");
57 
58 int	jail_sysvipc_allowed = 0;
59 SYSCTL_INT(_security_jail, OID_AUTO, sysvipc_allowed, CTLFLAG_RW,
60     &jail_sysvipc_allowed, 0,
61     "Processes in jail can use System V IPC primitives");
62 
63 static int jail_enforce_statfs = 2;
64 SYSCTL_INT(_security_jail, OID_AUTO, enforce_statfs, CTLFLAG_RW,
65     &jail_enforce_statfs, 0,
66     "Processes in jail cannot see all mounted file systems");
67 
68 int	jail_allow_raw_sockets = 0;
69 SYSCTL_INT(_security_jail, OID_AUTO, allow_raw_sockets, CTLFLAG_RW,
70     &jail_allow_raw_sockets, 0,
71     "Prison root can create raw sockets");
72 
73 int	jail_chflags_allowed = 0;
74 SYSCTL_INT(_security_jail, OID_AUTO, chflags_allowed, CTLFLAG_RW,
75     &jail_chflags_allowed, 0,
76     "Processes in jail can alter system file flags");
77 
78 int	jail_mount_allowed = 0;
79 SYSCTL_INT(_security_jail, OID_AUTO, mount_allowed, CTLFLAG_RW,
80     &jail_mount_allowed, 0,
81     "Processes in jail can mount/unmount jail-friendly file systems");
82 
83 /* allprison, lastprid, and prisoncount are protected by allprison_lock. */
84 struct	prisonlist allprison;
85 struct	sx allprison_lock;
86 int	lastprid = 0;
87 int	prisoncount = 0;
88 
89 /*
90  * List of jail services. Protected by allprison_lock.
91  */
92 TAILQ_HEAD(prison_services_head, prison_service);
93 static struct prison_services_head prison_services =
94     TAILQ_HEAD_INITIALIZER(prison_services);
95 static int prison_service_slots = 0;
96 
97 struct prison_service {
98 	prison_create_t ps_create;
99 	prison_destroy_t ps_destroy;
100 	int		ps_slotno;
101 	TAILQ_ENTRY(prison_service) ps_next;
102 	char	ps_name[0];
103 };
104 
105 static void		 init_prison(void *);
106 static void		 prison_complete(void *context, int pending);
107 static int		 sysctl_jail_list(SYSCTL_HANDLER_ARGS);
108 
109 static void
110 init_prison(void *data __unused)
111 {
112 
113 	sx_init(&allprison_lock, "allprison");
114 	LIST_INIT(&allprison);
115 }
116 
117 SYSINIT(prison, SI_SUB_INTRINSIC, SI_ORDER_ANY, init_prison, NULL);
118 
119 /*
120  * struct jail_args {
121  *	struct jail *jail;
122  * };
123  */
124 int
125 jail(struct thread *td, struct jail_args *uap)
126 {
127 	struct nameidata nd;
128 	struct prison *pr, *tpr;
129 	struct prison_service *psrv;
130 	struct jail j;
131 	struct jail_attach_args jaa;
132 	int vfslocked, error, tryprid;
133 
134 	error = copyin(uap->jail, &j, sizeof(j));
135 	if (error)
136 		return (error);
137 	if (j.version != 0)
138 		return (EINVAL);
139 
140 	MALLOC(pr, struct prison *, sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
141 	mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF);
142 	pr->pr_ref = 1;
143 	error = copyinstr(j.path, &pr->pr_path, sizeof(pr->pr_path), 0);
144 	if (error)
145 		goto e_killmtx;
146 	NDINIT(&nd, LOOKUP, MPSAFE | FOLLOW | LOCKLEAF, UIO_SYSSPACE,
147 	    pr->pr_path, td);
148 	error = namei(&nd);
149 	if (error)
150 		goto e_killmtx;
151 	vfslocked = NDHASGIANT(&nd);
152 	pr->pr_root = nd.ni_vp;
153 	VOP_UNLOCK(nd.ni_vp, 0);
154 	NDFREE(&nd, NDF_ONLY_PNBUF);
155 	VFS_UNLOCK_GIANT(vfslocked);
156 	error = copyinstr(j.hostname, &pr->pr_host, sizeof(pr->pr_host), 0);
157 	if (error)
158 		goto e_dropvnref;
159 	pr->pr_ip = j.ip_number;
160 	pr->pr_linux = NULL;
161 	pr->pr_securelevel = securelevel;
162 	if (prison_service_slots == 0)
163 		pr->pr_slots = NULL;
164 	else {
165 		pr->pr_slots = malloc(sizeof(*pr->pr_slots) * prison_service_slots,
166 		    M_PRISON, M_ZERO | M_WAITOK);
167 	}
168 
169 	/* Determine next pr_id and add prison to allprison list. */
170 	sx_xlock(&allprison_lock);
171 	tryprid = lastprid + 1;
172 	if (tryprid == JAIL_MAX)
173 		tryprid = 1;
174 next:
175 	LIST_FOREACH(tpr, &allprison, pr_list) {
176 		if (tpr->pr_id == tryprid) {
177 			tryprid++;
178 			if (tryprid == JAIL_MAX) {
179 				sx_xunlock(&allprison_lock);
180 				error = EAGAIN;
181 				goto e_dropvnref;
182 			}
183 			goto next;
184 		}
185 	}
186 	pr->pr_id = jaa.jid = lastprid = tryprid;
187 	LIST_INSERT_HEAD(&allprison, pr, pr_list);
188 	prisoncount++;
189 	sx_downgrade(&allprison_lock);
190 	TAILQ_FOREACH(psrv, &prison_services, ps_next) {
191 		psrv->ps_create(psrv, pr);
192 	}
193 	sx_sunlock(&allprison_lock);
194 
195 	error = jail_attach(td, &jaa);
196 	if (error)
197 		goto e_dropprref;
198 	mtx_lock(&pr->pr_mtx);
199 	pr->pr_ref--;
200 	mtx_unlock(&pr->pr_mtx);
201 	td->td_retval[0] = jaa.jid;
202 	return (0);
203 e_dropprref:
204 	sx_xlock(&allprison_lock);
205 	LIST_REMOVE(pr, pr_list);
206 	prisoncount--;
207 	sx_downgrade(&allprison_lock);
208 	TAILQ_FOREACH(psrv, &prison_services, ps_next) {
209 		psrv->ps_destroy(psrv, pr);
210 	}
211 	sx_sunlock(&allprison_lock);
212 e_dropvnref:
213 	if (pr->pr_slots != NULL)
214 		FREE(pr->pr_slots, M_PRISON);
215 	vfslocked = VFS_LOCK_GIANT(pr->pr_root->v_mount);
216 	vrele(pr->pr_root);
217 	VFS_UNLOCK_GIANT(vfslocked);
218 e_killmtx:
219 	mtx_destroy(&pr->pr_mtx);
220 	FREE(pr, M_PRISON);
221 	return (error);
222 }
223 
224 /*
225  * struct jail_attach_args {
226  *	int jid;
227  * };
228  */
229 int
230 jail_attach(struct thread *td, struct jail_attach_args *uap)
231 {
232 	struct proc *p;
233 	struct ucred *newcred, *oldcred;
234 	struct prison *pr;
235 	int vfslocked, error;
236 
237 	/*
238 	 * XXX: Note that there is a slight race here if two threads
239 	 * in the same privileged process attempt to attach to two
240 	 * different jails at the same time.  It is important for
241 	 * user processes not to do this, or they might end up with
242 	 * a process root from one prison, but attached to the jail
243 	 * of another.
244 	 */
245 	error = priv_check(td, PRIV_JAIL_ATTACH);
246 	if (error)
247 		return (error);
248 
249 	p = td->td_proc;
250 	sx_slock(&allprison_lock);
251 	pr = prison_find(uap->jid);
252 	if (pr == NULL) {
253 		sx_sunlock(&allprison_lock);
254 		return (EINVAL);
255 	}
256 	pr->pr_ref++;
257 	mtx_unlock(&pr->pr_mtx);
258 	sx_sunlock(&allprison_lock);
259 
260 	vfslocked = VFS_LOCK_GIANT(pr->pr_root->v_mount);
261 	vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
262 	if ((error = change_dir(pr->pr_root, td)) != 0)
263 		goto e_unlock;
264 #ifdef MAC
265 	if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
266 		goto e_unlock;
267 #endif
268 	VOP_UNLOCK(pr->pr_root, 0);
269 	change_root(pr->pr_root, td);
270 	VFS_UNLOCK_GIANT(vfslocked);
271 
272 	newcred = crget();
273 	PROC_LOCK(p);
274 	oldcred = p->p_ucred;
275 	setsugid(p);
276 	crcopy(newcred, oldcred);
277 	newcred->cr_prison = pr;
278 	p->p_ucred = newcred;
279 	PROC_UNLOCK(p);
280 	crfree(oldcred);
281 	return (0);
282 e_unlock:
283 	VOP_UNLOCK(pr->pr_root, 0);
284 	VFS_UNLOCK_GIANT(vfslocked);
285 	mtx_lock(&pr->pr_mtx);
286 	pr->pr_ref--;
287 	mtx_unlock(&pr->pr_mtx);
288 	return (error);
289 }
290 
291 /*
292  * Returns a locked prison instance, or NULL on failure.
293  */
294 struct prison *
295 prison_find(int prid)
296 {
297 	struct prison *pr;
298 
299 	sx_assert(&allprison_lock, SX_LOCKED);
300 	LIST_FOREACH(pr, &allprison, pr_list) {
301 		if (pr->pr_id == prid) {
302 			mtx_lock(&pr->pr_mtx);
303 			if (pr->pr_ref == 0) {
304 				mtx_unlock(&pr->pr_mtx);
305 				break;
306 			}
307 			return (pr);
308 		}
309 	}
310 	return (NULL);
311 }
312 
313 void
314 prison_free(struct prison *pr)
315 {
316 
317 	mtx_lock(&pr->pr_mtx);
318 	pr->pr_ref--;
319 	if (pr->pr_ref == 0) {
320 		mtx_unlock(&pr->pr_mtx);
321 		TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
322 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
323 		return;
324 	}
325 	mtx_unlock(&pr->pr_mtx);
326 }
327 
328 static void
329 prison_complete(void *context, int pending)
330 {
331 	struct prison_service *psrv;
332 	struct prison *pr;
333 	int vfslocked;
334 
335 	pr = (struct prison *)context;
336 
337 	sx_xlock(&allprison_lock);
338 	LIST_REMOVE(pr, pr_list);
339 	prisoncount--;
340 	sx_downgrade(&allprison_lock);
341 	TAILQ_FOREACH(psrv, &prison_services, ps_next) {
342 		psrv->ps_destroy(psrv, pr);
343 	}
344 	sx_sunlock(&allprison_lock);
345 	if (pr->pr_slots != NULL)
346 		FREE(pr->pr_slots, M_PRISON);
347 
348 	vfslocked = VFS_LOCK_GIANT(pr->pr_root->v_mount);
349 	vrele(pr->pr_root);
350 	VFS_UNLOCK_GIANT(vfslocked);
351 
352 	mtx_destroy(&pr->pr_mtx);
353 	if (pr->pr_linux != NULL)
354 		FREE(pr->pr_linux, M_PRISON);
355 	FREE(pr, M_PRISON);
356 }
357 
358 void
359 prison_hold(struct prison *pr)
360 {
361 
362 	mtx_lock(&pr->pr_mtx);
363 	KASSERT(pr->pr_ref > 0,
364 	    ("Trying to hold dead prison (id=%d).", pr->pr_id));
365 	pr->pr_ref++;
366 	mtx_unlock(&pr->pr_mtx);
367 }
368 
369 u_int32_t
370 prison_getip(struct ucred *cred)
371 {
372 
373 	return (cred->cr_prison->pr_ip);
374 }
375 
376 int
377 prison_ip(struct ucred *cred, int flag, u_int32_t *ip)
378 {
379 	u_int32_t tmp;
380 
381 	if (!jailed(cred))
382 		return (0);
383 	if (flag)
384 		tmp = *ip;
385 	else
386 		tmp = ntohl(*ip);
387 	if (tmp == INADDR_ANY) {
388 		if (flag)
389 			*ip = cred->cr_prison->pr_ip;
390 		else
391 			*ip = htonl(cred->cr_prison->pr_ip);
392 		return (0);
393 	}
394 	if (tmp == INADDR_LOOPBACK) {
395 		if (flag)
396 			*ip = cred->cr_prison->pr_ip;
397 		else
398 			*ip = htonl(cred->cr_prison->pr_ip);
399 		return (0);
400 	}
401 	if (cred->cr_prison->pr_ip != tmp)
402 		return (1);
403 	return (0);
404 }
405 
406 void
407 prison_remote_ip(struct ucred *cred, int flag, u_int32_t *ip)
408 {
409 	u_int32_t tmp;
410 
411 	if (!jailed(cred))
412 		return;
413 	if (flag)
414 		tmp = *ip;
415 	else
416 		tmp = ntohl(*ip);
417 	if (tmp == INADDR_LOOPBACK) {
418 		if (flag)
419 			*ip = cred->cr_prison->pr_ip;
420 		else
421 			*ip = htonl(cred->cr_prison->pr_ip);
422 		return;
423 	}
424 	return;
425 }
426 
427 int
428 prison_if(struct ucred *cred, struct sockaddr *sa)
429 {
430 	struct sockaddr_in *sai;
431 	int ok;
432 
433 	sai = (struct sockaddr_in *)sa;
434 	if ((sai->sin_family != AF_INET) && jail_socket_unixiproute_only)
435 		ok = 1;
436 	else if (sai->sin_family != AF_INET)
437 		ok = 0;
438 	else if (cred->cr_prison->pr_ip != ntohl(sai->sin_addr.s_addr))
439 		ok = 1;
440 	else
441 		ok = 0;
442 	return (ok);
443 }
444 
445 /*
446  * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
447  */
448 int
449 prison_check(struct ucred *cred1, struct ucred *cred2)
450 {
451 
452 	if (jailed(cred1)) {
453 		if (!jailed(cred2))
454 			return (ESRCH);
455 		if (cred2->cr_prison != cred1->cr_prison)
456 			return (ESRCH);
457 	}
458 
459 	return (0);
460 }
461 
462 /*
463  * Return 1 if the passed credential is in a jail, otherwise 0.
464  */
465 int
466 jailed(struct ucred *cred)
467 {
468 
469 	return (cred->cr_prison != NULL);
470 }
471 
472 /*
473  * Return the correct hostname for the passed credential.
474  */
475 void
476 getcredhostname(struct ucred *cred, char *buf, size_t size)
477 {
478 
479 	if (jailed(cred)) {
480 		mtx_lock(&cred->cr_prison->pr_mtx);
481 		strlcpy(buf, cred->cr_prison->pr_host, size);
482 		mtx_unlock(&cred->cr_prison->pr_mtx);
483 	} else {
484 		mtx_lock(&hostname_mtx);
485 		strlcpy(buf, V_hostname, size);
486 		mtx_unlock(&hostname_mtx);
487 	}
488 }
489 
490 /*
491  * Determine whether the subject represented by cred can "see"
492  * status of a mount point.
493  * Returns: 0 for permitted, ENOENT otherwise.
494  * XXX: This function should be called cr_canseemount() and should be
495  *      placed in kern_prot.c.
496  */
497 int
498 prison_canseemount(struct ucred *cred, struct mount *mp)
499 {
500 	struct prison *pr;
501 	struct statfs *sp;
502 	size_t len;
503 
504 	if (!jailed(cred) || jail_enforce_statfs == 0)
505 		return (0);
506 	pr = cred->cr_prison;
507 	if (pr->pr_root->v_mount == mp)
508 		return (0);
509 	if (jail_enforce_statfs == 2)
510 		return (ENOENT);
511 	/*
512 	 * If jail's chroot directory is set to "/" we should be able to see
513 	 * all mount-points from inside a jail.
514 	 * This is ugly check, but this is the only situation when jail's
515 	 * directory ends with '/'.
516 	 */
517 	if (strcmp(pr->pr_path, "/") == 0)
518 		return (0);
519 	len = strlen(pr->pr_path);
520 	sp = &mp->mnt_stat;
521 	if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
522 		return (ENOENT);
523 	/*
524 	 * Be sure that we don't have situation where jail's root directory
525 	 * is "/some/path" and mount point is "/some/pathpath".
526 	 */
527 	if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
528 		return (ENOENT);
529 	return (0);
530 }
531 
532 void
533 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
534 {
535 	char jpath[MAXPATHLEN];
536 	struct prison *pr;
537 	size_t len;
538 
539 	if (!jailed(cred) || jail_enforce_statfs == 0)
540 		return;
541 	pr = cred->cr_prison;
542 	if (prison_canseemount(cred, mp) != 0) {
543 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
544 		strlcpy(sp->f_mntonname, "[restricted]",
545 		    sizeof(sp->f_mntonname));
546 		return;
547 	}
548 	if (pr->pr_root->v_mount == mp) {
549 		/*
550 		 * Clear current buffer data, so we are sure nothing from
551 		 * the valid path left there.
552 		 */
553 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
554 		*sp->f_mntonname = '/';
555 		return;
556 	}
557 	/*
558 	 * If jail's chroot directory is set to "/" we should be able to see
559 	 * all mount-points from inside a jail.
560 	 */
561 	if (strcmp(pr->pr_path, "/") == 0)
562 		return;
563 	len = strlen(pr->pr_path);
564 	strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
565 	/*
566 	 * Clear current buffer data, so we are sure nothing from
567 	 * the valid path left there.
568 	 */
569 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
570 	if (*jpath == '\0') {
571 		/* Should never happen. */
572 		*sp->f_mntonname = '/';
573 	} else {
574 		strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
575 	}
576 }
577 
578 /*
579  * Check with permission for a specific privilege is granted within jail.  We
580  * have a specific list of accepted privileges; the rest are denied.
581  */
582 int
583 prison_priv_check(struct ucred *cred, int priv)
584 {
585 
586 	if (!jailed(cred))
587 		return (0);
588 
589 	switch (priv) {
590 
591 		/*
592 		 * Allow ktrace privileges for root in jail.
593 		 */
594 	case PRIV_KTRACE:
595 
596 #if 0
597 		/*
598 		 * Allow jailed processes to configure audit identity and
599 		 * submit audit records (login, etc).  In the future we may
600 		 * want to further refine the relationship between audit and
601 		 * jail.
602 		 */
603 	case PRIV_AUDIT_GETAUDIT:
604 	case PRIV_AUDIT_SETAUDIT:
605 	case PRIV_AUDIT_SUBMIT:
606 #endif
607 
608 		/*
609 		 * Allow jailed processes to manipulate process UNIX
610 		 * credentials in any way they see fit.
611 		 */
612 	case PRIV_CRED_SETUID:
613 	case PRIV_CRED_SETEUID:
614 	case PRIV_CRED_SETGID:
615 	case PRIV_CRED_SETEGID:
616 	case PRIV_CRED_SETGROUPS:
617 	case PRIV_CRED_SETREUID:
618 	case PRIV_CRED_SETREGID:
619 	case PRIV_CRED_SETRESUID:
620 	case PRIV_CRED_SETRESGID:
621 
622 		/*
623 		 * Jail implements visibility constraints already, so allow
624 		 * jailed root to override uid/gid-based constraints.
625 		 */
626 	case PRIV_SEEOTHERGIDS:
627 	case PRIV_SEEOTHERUIDS:
628 
629 		/*
630 		 * Jail implements inter-process debugging limits already, so
631 		 * allow jailed root various debugging privileges.
632 		 */
633 	case PRIV_DEBUG_DIFFCRED:
634 	case PRIV_DEBUG_SUGID:
635 	case PRIV_DEBUG_UNPRIV:
636 
637 		/*
638 		 * Allow jail to set various resource limits and login
639 		 * properties, and for now, exceed process resource limits.
640 		 */
641 	case PRIV_PROC_LIMIT:
642 	case PRIV_PROC_SETLOGIN:
643 	case PRIV_PROC_SETRLIMIT:
644 
645 		/*
646 		 * System V and POSIX IPC privileges are granted in jail.
647 		 */
648 	case PRIV_IPC_READ:
649 	case PRIV_IPC_WRITE:
650 	case PRIV_IPC_ADMIN:
651 	case PRIV_IPC_MSGSIZE:
652 	case PRIV_MQ_ADMIN:
653 
654 		/*
655 		 * Jail implements its own inter-process limits, so allow
656 		 * root processes in jail to change scheduling on other
657 		 * processes in the same jail.  Likewise for signalling.
658 		 */
659 	case PRIV_SCHED_DIFFCRED:
660 	case PRIV_SIGNAL_DIFFCRED:
661 	case PRIV_SIGNAL_SUGID:
662 
663 		/*
664 		 * Allow jailed processes to write to sysctls marked as jail
665 		 * writable.
666 		 */
667 	case PRIV_SYSCTL_WRITEJAIL:
668 
669 		/*
670 		 * Allow root in jail to manage a variety of quota
671 		 * properties.  These should likely be conditional on a
672 		 * configuration option.
673 		 */
674 	case PRIV_VFS_GETQUOTA:
675 	case PRIV_VFS_SETQUOTA:
676 
677 		/*
678 		 * Since Jail relies on chroot() to implement file system
679 		 * protections, grant many VFS privileges to root in jail.
680 		 * Be careful to exclude mount-related and NFS-related
681 		 * privileges.
682 		 */
683 	case PRIV_VFS_READ:
684 	case PRIV_VFS_WRITE:
685 	case PRIV_VFS_ADMIN:
686 	case PRIV_VFS_EXEC:
687 	case PRIV_VFS_LOOKUP:
688 	case PRIV_VFS_BLOCKRESERVE:	/* XXXRW: Slightly surprising. */
689 	case PRIV_VFS_CHFLAGS_DEV:
690 	case PRIV_VFS_CHOWN:
691 	case PRIV_VFS_CHROOT:
692 	case PRIV_VFS_RETAINSUGID:
693 	case PRIV_VFS_FCHROOT:
694 	case PRIV_VFS_LINK:
695 	case PRIV_VFS_SETGID:
696 	case PRIV_VFS_STAT:
697 	case PRIV_VFS_STICKYFILE:
698 		return (0);
699 
700 		/*
701 		 * Depending on the global setting, allow privilege of
702 		 * setting system flags.
703 		 */
704 	case PRIV_VFS_SYSFLAGS:
705 		if (jail_chflags_allowed)
706 			return (0);
707 		else
708 			return (EPERM);
709 
710 		/*
711 		 * Depending on the global setting, allow privilege of
712 		 * mounting/unmounting file systems.
713 		 */
714 	case PRIV_VFS_MOUNT:
715 	case PRIV_VFS_UNMOUNT:
716 	case PRIV_VFS_MOUNT_NONUSER:
717 	case PRIV_VFS_MOUNT_OWNER:
718 		if (jail_mount_allowed)
719 			return (0);
720 		else
721 			return (EPERM);
722 
723 		/*
724 		 * Allow jailed root to bind reserved ports and reuse in-use
725 		 * ports.
726 		 */
727 	case PRIV_NETINET_RESERVEDPORT:
728 	case PRIV_NETINET_REUSEPORT:
729 		return (0);
730 
731 		/*
732 		 * Allow jailed root to set certian IPv4/6 (option) headers.
733 		 */
734 	case PRIV_NETINET_SETHDROPTS:
735 		return (0);
736 
737 		/*
738 		 * Conditionally allow creating raw sockets in jail.
739 		 */
740 	case PRIV_NETINET_RAW:
741 		if (jail_allow_raw_sockets)
742 			return (0);
743 		else
744 			return (EPERM);
745 
746 		/*
747 		 * Since jail implements its own visibility limits on netstat
748 		 * sysctls, allow getcred.  This allows identd to work in
749 		 * jail.
750 		 */
751 	case PRIV_NETINET_GETCRED:
752 		return (0);
753 
754 	default:
755 		/*
756 		 * In all remaining cases, deny the privilege request.  This
757 		 * includes almost all network privileges, many system
758 		 * configuration privileges.
759 		 */
760 		return (EPERM);
761 	}
762 }
763 
764 /*
765  * Register jail service. Provides 'create' and 'destroy' methods.
766  * 'create' method will be called for every existing jail and all
767  * jails in the future as they beeing created.
768  * 'destroy' method will be called for every jail going away and
769  * for all existing jails at the time of service deregistration.
770  */
771 struct prison_service *
772 prison_service_register(const char *name, prison_create_t create,
773     prison_destroy_t destroy)
774 {
775 	struct prison_service *psrv, *psrv2;
776 	struct prison *pr;
777 	int reallocate = 1, slotno = 0;
778 	void **slots, **oldslots;
779 
780 	psrv = malloc(sizeof(*psrv) + strlen(name) + 1, M_PRISON,
781 	    M_WAITOK | M_ZERO);
782 	psrv->ps_create = create;
783 	psrv->ps_destroy = destroy;
784 	strcpy(psrv->ps_name, name);
785 	/*
786 	 * Grab the allprison_lock here, so we won't miss any jail
787 	 * creation/destruction.
788 	 */
789 	sx_xlock(&allprison_lock);
790 #ifdef INVARIANTS
791 	/*
792 	 * Verify if service is not already registered.
793 	 */
794 	TAILQ_FOREACH(psrv2, &prison_services, ps_next) {
795 		KASSERT(strcmp(psrv2->ps_name, name) != 0,
796 		    ("jail service %s already registered", name));
797 	}
798 #endif
799 	/*
800 	 * Find free slot. When there is no existing free slot available,
801 	 * allocate one at the end.
802 	 */
803 	TAILQ_FOREACH(psrv2, &prison_services, ps_next) {
804 		if (psrv2->ps_slotno != slotno) {
805 			KASSERT(slotno < psrv2->ps_slotno,
806 			    ("Invalid slotno (slotno=%d >= ps_slotno=%d",
807 			    slotno, psrv2->ps_slotno));
808 			/* We found free slot. */
809 			reallocate = 0;
810 			break;
811 		}
812 		slotno++;
813 	}
814 	psrv->ps_slotno = slotno;
815 	/*
816 	 * Keep the list sorted by slot number.
817 	 */
818 	if (psrv2 != NULL) {
819 		KASSERT(reallocate == 0, ("psrv2 != NULL && reallocate != 0"));
820 		TAILQ_INSERT_BEFORE(psrv2, psrv, ps_next);
821 	} else {
822 		KASSERT(reallocate == 1, ("psrv2 == NULL && reallocate == 0"));
823 		TAILQ_INSERT_TAIL(&prison_services, psrv, ps_next);
824 	}
825 	prison_service_slots++;
826 	sx_downgrade(&allprison_lock);
827 	/*
828 	 * Allocate memory for new slot if we didn't found empty one.
829 	 * Do not use realloc(9), because pr_slots is protected with a mutex,
830 	 * so we can't sleep.
831 	 */
832 	LIST_FOREACH(pr, &allprison, pr_list) {
833 		if (reallocate) {
834 			/* First allocate memory with M_WAITOK. */
835 			slots = malloc(sizeof(*slots) * prison_service_slots,
836 			    M_PRISON, M_WAITOK);
837 			/* Now grab the mutex and replace pr_slots. */
838 			mtx_lock(&pr->pr_mtx);
839 			oldslots = pr->pr_slots;
840 			if (psrv->ps_slotno > 0) {
841 				bcopy(oldslots, slots,
842 				    sizeof(*slots) * (prison_service_slots - 1));
843 			}
844 			slots[psrv->ps_slotno] = NULL;
845 			pr->pr_slots = slots;
846 			mtx_unlock(&pr->pr_mtx);
847 			if (oldslots != NULL)
848 				free(oldslots, M_PRISON);
849 		}
850 		/*
851 		 * Call 'create' method for each existing jail.
852 		 */
853 		psrv->ps_create(psrv, pr);
854 	}
855 	sx_sunlock(&allprison_lock);
856 
857 	return (psrv);
858 }
859 
860 void
861 prison_service_deregister(struct prison_service *psrv)
862 {
863 	struct prison *pr;
864 	void **slots, **oldslots;
865 	int last = 0;
866 
867 	sx_xlock(&allprison_lock);
868 	if (TAILQ_LAST(&prison_services, prison_services_head) == psrv)
869 		last = 1;
870 	TAILQ_REMOVE(&prison_services, psrv, ps_next);
871 	prison_service_slots--;
872 	sx_downgrade(&allprison_lock);
873 	LIST_FOREACH(pr, &allprison, pr_list) {
874 		/*
875 		 * Call 'destroy' method for every currently existing jail.
876 		 */
877 		psrv->ps_destroy(psrv, pr);
878 		/*
879 		 * If this is the last slot, free the memory allocated for it.
880 		 */
881 		if (last) {
882 			if (prison_service_slots == 0)
883 				slots = NULL;
884 			else {
885 				slots = malloc(sizeof(*slots) * prison_service_slots,
886 				    M_PRISON, M_WAITOK);
887 			}
888 			mtx_lock(&pr->pr_mtx);
889 			oldslots = pr->pr_slots;
890 			/*
891 			 * We require setting slot to NULL after freeing it,
892 			 * this way we can check for memory leaks here.
893 			 */
894 			KASSERT(oldslots[psrv->ps_slotno] == NULL,
895 			    ("Slot %d (service %s, jailid=%d) still contains data?",
896 			     psrv->ps_slotno, psrv->ps_name, pr->pr_id));
897 			if (psrv->ps_slotno > 0) {
898 				bcopy(oldslots, slots,
899 				    sizeof(*slots) * prison_service_slots);
900 			}
901 			pr->pr_slots = slots;
902 			mtx_unlock(&pr->pr_mtx);
903 			KASSERT(oldslots != NULL, ("oldslots == NULL"));
904 			free(oldslots, M_PRISON);
905 		}
906 	}
907 	sx_sunlock(&allprison_lock);
908 	free(psrv, M_PRISON);
909 }
910 
911 /*
912  * Function sets data for the given jail in slot assigned for the given
913  * jail service.
914  */
915 void
916 prison_service_data_set(struct prison_service *psrv, struct prison *pr,
917     void *data)
918 {
919 
920 	mtx_assert(&pr->pr_mtx, MA_OWNED);
921 	pr->pr_slots[psrv->ps_slotno] = data;
922 }
923 
924 /*
925  * Function clears slots assigned for the given jail service in the given
926  * prison structure and returns current slot data.
927  */
928 void *
929 prison_service_data_del(struct prison_service *psrv, struct prison *pr)
930 {
931 	void *data;
932 
933 	mtx_assert(&pr->pr_mtx, MA_OWNED);
934 	data = pr->pr_slots[psrv->ps_slotno];
935 	pr->pr_slots[psrv->ps_slotno] = NULL;
936 	return (data);
937 }
938 
939 /*
940  * Function returns current data from the slot assigned to the given jail
941  * service for the given jail.
942  */
943 void *
944 prison_service_data_get(struct prison_service *psrv, struct prison *pr)
945 {
946 
947 	mtx_assert(&pr->pr_mtx, MA_OWNED);
948 	return (pr->pr_slots[psrv->ps_slotno]);
949 }
950 
951 static int
952 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
953 {
954 	struct xprison *xp, *sxp;
955 	struct prison *pr;
956 	int count, error;
957 
958 	if (jailed(req->td->td_ucred))
959 		return (0);
960 
961 	sx_slock(&allprison_lock);
962 	if ((count = prisoncount) == 0) {
963 		sx_sunlock(&allprison_lock);
964 		return (0);
965 	}
966 
967 	sxp = xp = malloc(sizeof(*xp) * count, M_TEMP, M_WAITOK | M_ZERO);
968 
969 	LIST_FOREACH(pr, &allprison, pr_list) {
970 		xp->pr_version = XPRISON_VERSION;
971 		xp->pr_id = pr->pr_id;
972 		xp->pr_ip = pr->pr_ip;
973 		strlcpy(xp->pr_path, pr->pr_path, sizeof(xp->pr_path));
974 		mtx_lock(&pr->pr_mtx);
975 		strlcpy(xp->pr_host, pr->pr_host, sizeof(xp->pr_host));
976 		mtx_unlock(&pr->pr_mtx);
977 		xp++;
978 	}
979 	sx_sunlock(&allprison_lock);
980 
981 	error = SYSCTL_OUT(req, sxp, sizeof(*sxp) * count);
982 	free(sxp, M_TEMP);
983 	return (error);
984 }
985 
986 SYSCTL_OID(_security_jail, OID_AUTO, list, CTLTYPE_STRUCT | CTLFLAG_RD,
987     NULL, 0, sysctl_jail_list, "S", "List of active jails");
988 
989 static int
990 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
991 {
992 	int error, injail;
993 
994 	injail = jailed(req->td->td_ucred);
995 	error = SYSCTL_OUT(req, &injail, sizeof(injail));
996 
997 	return (error);
998 }
999 SYSCTL_PROC(_security_jail, OID_AUTO, jailed, CTLTYPE_INT | CTLFLAG_RD,
1000     NULL, 0, sysctl_jail_jailed, "I", "Process in jail?");
1001