xref: /freebsd/sys/kern/kern_jail.c (revision a9148abd9da5db2f1c682fb17bed791845fc41c9)
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 	pr = malloc(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 	INIT_VPROCG(cred->cr_vimage->v_procg);
479 
480 	if (jailed(cred)) {
481 		mtx_lock(&cred->cr_prison->pr_mtx);
482 		strlcpy(buf, cred->cr_prison->pr_host, size);
483 		mtx_unlock(&cred->cr_prison->pr_mtx);
484 	} else {
485 		mtx_lock(&hostname_mtx);
486 		strlcpy(buf, V_hostname, size);
487 		mtx_unlock(&hostname_mtx);
488 	}
489 }
490 
491 /*
492  * Determine whether the subject represented by cred can "see"
493  * status of a mount point.
494  * Returns: 0 for permitted, ENOENT otherwise.
495  * XXX: This function should be called cr_canseemount() and should be
496  *      placed in kern_prot.c.
497  */
498 int
499 prison_canseemount(struct ucred *cred, struct mount *mp)
500 {
501 	struct prison *pr;
502 	struct statfs *sp;
503 	size_t len;
504 
505 	if (!jailed(cred) || jail_enforce_statfs == 0)
506 		return (0);
507 	pr = cred->cr_prison;
508 	if (pr->pr_root->v_mount == mp)
509 		return (0);
510 	if (jail_enforce_statfs == 2)
511 		return (ENOENT);
512 	/*
513 	 * If jail's chroot directory is set to "/" we should be able to see
514 	 * all mount-points from inside a jail.
515 	 * This is ugly check, but this is the only situation when jail's
516 	 * directory ends with '/'.
517 	 */
518 	if (strcmp(pr->pr_path, "/") == 0)
519 		return (0);
520 	len = strlen(pr->pr_path);
521 	sp = &mp->mnt_stat;
522 	if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
523 		return (ENOENT);
524 	/*
525 	 * Be sure that we don't have situation where jail's root directory
526 	 * is "/some/path" and mount point is "/some/pathpath".
527 	 */
528 	if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
529 		return (ENOENT);
530 	return (0);
531 }
532 
533 void
534 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
535 {
536 	char jpath[MAXPATHLEN];
537 	struct prison *pr;
538 	size_t len;
539 
540 	if (!jailed(cred) || jail_enforce_statfs == 0)
541 		return;
542 	pr = cred->cr_prison;
543 	if (prison_canseemount(cred, mp) != 0) {
544 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
545 		strlcpy(sp->f_mntonname, "[restricted]",
546 		    sizeof(sp->f_mntonname));
547 		return;
548 	}
549 	if (pr->pr_root->v_mount == mp) {
550 		/*
551 		 * Clear current buffer data, so we are sure nothing from
552 		 * the valid path left there.
553 		 */
554 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
555 		*sp->f_mntonname = '/';
556 		return;
557 	}
558 	/*
559 	 * If jail's chroot directory is set to "/" we should be able to see
560 	 * all mount-points from inside a jail.
561 	 */
562 	if (strcmp(pr->pr_path, "/") == 0)
563 		return;
564 	len = strlen(pr->pr_path);
565 	strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
566 	/*
567 	 * Clear current buffer data, so we are sure nothing from
568 	 * the valid path left there.
569 	 */
570 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
571 	if (*jpath == '\0') {
572 		/* Should never happen. */
573 		*sp->f_mntonname = '/';
574 	} else {
575 		strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
576 	}
577 }
578 
579 /*
580  * Check with permission for a specific privilege is granted within jail.  We
581  * have a specific list of accepted privileges; the rest are denied.
582  */
583 int
584 prison_priv_check(struct ucred *cred, int priv)
585 {
586 
587 	if (!jailed(cred))
588 		return (0);
589 
590 	switch (priv) {
591 
592 		/*
593 		 * Allow ktrace privileges for root in jail.
594 		 */
595 	case PRIV_KTRACE:
596 
597 #if 0
598 		/*
599 		 * Allow jailed processes to configure audit identity and
600 		 * submit audit records (login, etc).  In the future we may
601 		 * want to further refine the relationship between audit and
602 		 * jail.
603 		 */
604 	case PRIV_AUDIT_GETAUDIT:
605 	case PRIV_AUDIT_SETAUDIT:
606 	case PRIV_AUDIT_SUBMIT:
607 #endif
608 
609 		/*
610 		 * Allow jailed processes to manipulate process UNIX
611 		 * credentials in any way they see fit.
612 		 */
613 	case PRIV_CRED_SETUID:
614 	case PRIV_CRED_SETEUID:
615 	case PRIV_CRED_SETGID:
616 	case PRIV_CRED_SETEGID:
617 	case PRIV_CRED_SETGROUPS:
618 	case PRIV_CRED_SETREUID:
619 	case PRIV_CRED_SETREGID:
620 	case PRIV_CRED_SETRESUID:
621 	case PRIV_CRED_SETRESGID:
622 
623 		/*
624 		 * Jail implements visibility constraints already, so allow
625 		 * jailed root to override uid/gid-based constraints.
626 		 */
627 	case PRIV_SEEOTHERGIDS:
628 	case PRIV_SEEOTHERUIDS:
629 
630 		/*
631 		 * Jail implements inter-process debugging limits already, so
632 		 * allow jailed root various debugging privileges.
633 		 */
634 	case PRIV_DEBUG_DIFFCRED:
635 	case PRIV_DEBUG_SUGID:
636 	case PRIV_DEBUG_UNPRIV:
637 
638 		/*
639 		 * Allow jail to set various resource limits and login
640 		 * properties, and for now, exceed process resource limits.
641 		 */
642 	case PRIV_PROC_LIMIT:
643 	case PRIV_PROC_SETLOGIN:
644 	case PRIV_PROC_SETRLIMIT:
645 
646 		/*
647 		 * System V and POSIX IPC privileges are granted in jail.
648 		 */
649 	case PRIV_IPC_READ:
650 	case PRIV_IPC_WRITE:
651 	case PRIV_IPC_ADMIN:
652 	case PRIV_IPC_MSGSIZE:
653 	case PRIV_MQ_ADMIN:
654 
655 		/*
656 		 * Jail implements its own inter-process limits, so allow
657 		 * root processes in jail to change scheduling on other
658 		 * processes in the same jail.  Likewise for signalling.
659 		 */
660 	case PRIV_SCHED_DIFFCRED:
661 	case PRIV_SIGNAL_DIFFCRED:
662 	case PRIV_SIGNAL_SUGID:
663 
664 		/*
665 		 * Allow jailed processes to write to sysctls marked as jail
666 		 * writable.
667 		 */
668 	case PRIV_SYSCTL_WRITEJAIL:
669 
670 		/*
671 		 * Allow root in jail to manage a variety of quota
672 		 * properties.  These should likely be conditional on a
673 		 * configuration option.
674 		 */
675 	case PRIV_VFS_GETQUOTA:
676 	case PRIV_VFS_SETQUOTA:
677 
678 		/*
679 		 * Since Jail relies on chroot() to implement file system
680 		 * protections, grant many VFS privileges to root in jail.
681 		 * Be careful to exclude mount-related and NFS-related
682 		 * privileges.
683 		 */
684 	case PRIV_VFS_READ:
685 	case PRIV_VFS_WRITE:
686 	case PRIV_VFS_ADMIN:
687 	case PRIV_VFS_EXEC:
688 	case PRIV_VFS_LOOKUP:
689 	case PRIV_VFS_BLOCKRESERVE:	/* XXXRW: Slightly surprising. */
690 	case PRIV_VFS_CHFLAGS_DEV:
691 	case PRIV_VFS_CHOWN:
692 	case PRIV_VFS_CHROOT:
693 	case PRIV_VFS_RETAINSUGID:
694 	case PRIV_VFS_FCHROOT:
695 	case PRIV_VFS_LINK:
696 	case PRIV_VFS_SETGID:
697 	case PRIV_VFS_STAT:
698 	case PRIV_VFS_STICKYFILE:
699 		return (0);
700 
701 		/*
702 		 * Depending on the global setting, allow privilege of
703 		 * setting system flags.
704 		 */
705 	case PRIV_VFS_SYSFLAGS:
706 		if (jail_chflags_allowed)
707 			return (0);
708 		else
709 			return (EPERM);
710 
711 		/*
712 		 * Depending on the global setting, allow privilege of
713 		 * mounting/unmounting file systems.
714 		 */
715 	case PRIV_VFS_MOUNT:
716 	case PRIV_VFS_UNMOUNT:
717 	case PRIV_VFS_MOUNT_NONUSER:
718 	case PRIV_VFS_MOUNT_OWNER:
719 		if (jail_mount_allowed)
720 			return (0);
721 		else
722 			return (EPERM);
723 
724 		/*
725 		 * Allow jailed root to bind reserved ports and reuse in-use
726 		 * ports.
727 		 */
728 	case PRIV_NETINET_RESERVEDPORT:
729 	case PRIV_NETINET_REUSEPORT:
730 		return (0);
731 
732 		/*
733 		 * Allow jailed root to set certian IPv4/6 (option) headers.
734 		 */
735 	case PRIV_NETINET_SETHDROPTS:
736 		return (0);
737 
738 		/*
739 		 * Conditionally allow creating raw sockets in jail.
740 		 */
741 	case PRIV_NETINET_RAW:
742 		if (jail_allow_raw_sockets)
743 			return (0);
744 		else
745 			return (EPERM);
746 
747 		/*
748 		 * Since jail implements its own visibility limits on netstat
749 		 * sysctls, allow getcred.  This allows identd to work in
750 		 * jail.
751 		 */
752 	case PRIV_NETINET_GETCRED:
753 		return (0);
754 
755 	default:
756 		/*
757 		 * In all remaining cases, deny the privilege request.  This
758 		 * includes almost all network privileges, many system
759 		 * configuration privileges.
760 		 */
761 		return (EPERM);
762 	}
763 }
764 
765 /*
766  * Register jail service. Provides 'create' and 'destroy' methods.
767  * 'create' method will be called for every existing jail and all
768  * jails in the future as they beeing created.
769  * 'destroy' method will be called for every jail going away and
770  * for all existing jails at the time of service deregistration.
771  */
772 struct prison_service *
773 prison_service_register(const char *name, prison_create_t create,
774     prison_destroy_t destroy)
775 {
776 	struct prison_service *psrv, *psrv2;
777 	struct prison *pr;
778 	int reallocate = 1, slotno = 0;
779 	void **slots, **oldslots;
780 
781 	psrv = malloc(sizeof(*psrv) + strlen(name) + 1, M_PRISON,
782 	    M_WAITOK | M_ZERO);
783 	psrv->ps_create = create;
784 	psrv->ps_destroy = destroy;
785 	strcpy(psrv->ps_name, name);
786 	/*
787 	 * Grab the allprison_lock here, so we won't miss any jail
788 	 * creation/destruction.
789 	 */
790 	sx_xlock(&allprison_lock);
791 #ifdef INVARIANTS
792 	/*
793 	 * Verify if service is not already registered.
794 	 */
795 	TAILQ_FOREACH(psrv2, &prison_services, ps_next) {
796 		KASSERT(strcmp(psrv2->ps_name, name) != 0,
797 		    ("jail service %s already registered", name));
798 	}
799 #endif
800 	/*
801 	 * Find free slot. When there is no existing free slot available,
802 	 * allocate one at the end.
803 	 */
804 	TAILQ_FOREACH(psrv2, &prison_services, ps_next) {
805 		if (psrv2->ps_slotno != slotno) {
806 			KASSERT(slotno < psrv2->ps_slotno,
807 			    ("Invalid slotno (slotno=%d >= ps_slotno=%d",
808 			    slotno, psrv2->ps_slotno));
809 			/* We found free slot. */
810 			reallocate = 0;
811 			break;
812 		}
813 		slotno++;
814 	}
815 	psrv->ps_slotno = slotno;
816 	/*
817 	 * Keep the list sorted by slot number.
818 	 */
819 	if (psrv2 != NULL) {
820 		KASSERT(reallocate == 0, ("psrv2 != NULL && reallocate != 0"));
821 		TAILQ_INSERT_BEFORE(psrv2, psrv, ps_next);
822 	} else {
823 		KASSERT(reallocate == 1, ("psrv2 == NULL && reallocate == 0"));
824 		TAILQ_INSERT_TAIL(&prison_services, psrv, ps_next);
825 	}
826 	prison_service_slots++;
827 	sx_downgrade(&allprison_lock);
828 	/*
829 	 * Allocate memory for new slot if we didn't found empty one.
830 	 * Do not use realloc(9), because pr_slots is protected with a mutex,
831 	 * so we can't sleep.
832 	 */
833 	LIST_FOREACH(pr, &allprison, pr_list) {
834 		if (reallocate) {
835 			/* First allocate memory with M_WAITOK. */
836 			slots = malloc(sizeof(*slots) * prison_service_slots,
837 			    M_PRISON, M_WAITOK);
838 			/* Now grab the mutex and replace pr_slots. */
839 			mtx_lock(&pr->pr_mtx);
840 			oldslots = pr->pr_slots;
841 			if (psrv->ps_slotno > 0) {
842 				bcopy(oldslots, slots,
843 				    sizeof(*slots) * (prison_service_slots - 1));
844 			}
845 			slots[psrv->ps_slotno] = NULL;
846 			pr->pr_slots = slots;
847 			mtx_unlock(&pr->pr_mtx);
848 			if (oldslots != NULL)
849 				free(oldslots, M_PRISON);
850 		}
851 		/*
852 		 * Call 'create' method for each existing jail.
853 		 */
854 		psrv->ps_create(psrv, pr);
855 	}
856 	sx_sunlock(&allprison_lock);
857 
858 	return (psrv);
859 }
860 
861 void
862 prison_service_deregister(struct prison_service *psrv)
863 {
864 	struct prison *pr;
865 	void **slots, **oldslots;
866 	int last = 0;
867 
868 	sx_xlock(&allprison_lock);
869 	if (TAILQ_LAST(&prison_services, prison_services_head) == psrv)
870 		last = 1;
871 	TAILQ_REMOVE(&prison_services, psrv, ps_next);
872 	prison_service_slots--;
873 	sx_downgrade(&allprison_lock);
874 	LIST_FOREACH(pr, &allprison, pr_list) {
875 		/*
876 		 * Call 'destroy' method for every currently existing jail.
877 		 */
878 		psrv->ps_destroy(psrv, pr);
879 		/*
880 		 * If this is the last slot, free the memory allocated for it.
881 		 */
882 		if (last) {
883 			if (prison_service_slots == 0)
884 				slots = NULL;
885 			else {
886 				slots = malloc(sizeof(*slots) * prison_service_slots,
887 				    M_PRISON, M_WAITOK);
888 			}
889 			mtx_lock(&pr->pr_mtx);
890 			oldslots = pr->pr_slots;
891 			/*
892 			 * We require setting slot to NULL after freeing it,
893 			 * this way we can check for memory leaks here.
894 			 */
895 			KASSERT(oldslots[psrv->ps_slotno] == NULL,
896 			    ("Slot %d (service %s, jailid=%d) still contains data?",
897 			     psrv->ps_slotno, psrv->ps_name, pr->pr_id));
898 			if (psrv->ps_slotno > 0) {
899 				bcopy(oldslots, slots,
900 				    sizeof(*slots) * prison_service_slots);
901 			}
902 			pr->pr_slots = slots;
903 			mtx_unlock(&pr->pr_mtx);
904 			KASSERT(oldslots != NULL, ("oldslots == NULL"));
905 			free(oldslots, M_PRISON);
906 		}
907 	}
908 	sx_sunlock(&allprison_lock);
909 	free(psrv, M_PRISON);
910 }
911 
912 /*
913  * Function sets data for the given jail in slot assigned for the given
914  * jail service.
915  */
916 void
917 prison_service_data_set(struct prison_service *psrv, struct prison *pr,
918     void *data)
919 {
920 
921 	mtx_assert(&pr->pr_mtx, MA_OWNED);
922 	pr->pr_slots[psrv->ps_slotno] = data;
923 }
924 
925 /*
926  * Function clears slots assigned for the given jail service in the given
927  * prison structure and returns current slot data.
928  */
929 void *
930 prison_service_data_del(struct prison_service *psrv, struct prison *pr)
931 {
932 	void *data;
933 
934 	mtx_assert(&pr->pr_mtx, MA_OWNED);
935 	data = pr->pr_slots[psrv->ps_slotno];
936 	pr->pr_slots[psrv->ps_slotno] = NULL;
937 	return (data);
938 }
939 
940 /*
941  * Function returns current data from the slot assigned to the given jail
942  * service for the given jail.
943  */
944 void *
945 prison_service_data_get(struct prison_service *psrv, struct prison *pr)
946 {
947 
948 	mtx_assert(&pr->pr_mtx, MA_OWNED);
949 	return (pr->pr_slots[psrv->ps_slotno]);
950 }
951 
952 static int
953 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
954 {
955 	struct xprison *xp, *sxp;
956 	struct prison *pr;
957 	int count, error;
958 
959 	if (jailed(req->td->td_ucred))
960 		return (0);
961 
962 	sx_slock(&allprison_lock);
963 	if ((count = prisoncount) == 0) {
964 		sx_sunlock(&allprison_lock);
965 		return (0);
966 	}
967 
968 	sxp = xp = malloc(sizeof(*xp) * count, M_TEMP, M_WAITOK | M_ZERO);
969 
970 	LIST_FOREACH(pr, &allprison, pr_list) {
971 		xp->pr_version = XPRISON_VERSION;
972 		xp->pr_id = pr->pr_id;
973 		xp->pr_ip = pr->pr_ip;
974 		strlcpy(xp->pr_path, pr->pr_path, sizeof(xp->pr_path));
975 		mtx_lock(&pr->pr_mtx);
976 		strlcpy(xp->pr_host, pr->pr_host, sizeof(xp->pr_host));
977 		mtx_unlock(&pr->pr_mtx);
978 		xp++;
979 	}
980 	sx_sunlock(&allprison_lock);
981 
982 	error = SYSCTL_OUT(req, sxp, sizeof(*sxp) * count);
983 	free(sxp, M_TEMP);
984 	return (error);
985 }
986 
987 SYSCTL_OID(_security_jail, OID_AUTO, list, CTLTYPE_STRUCT | CTLFLAG_RD,
988     NULL, 0, sysctl_jail_list, "S", "List of active jails");
989 
990 static int
991 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
992 {
993 	int error, injail;
994 
995 	injail = jailed(req->td->td_ucred);
996 	error = SYSCTL_OUT(req, &injail, sizeof(injail));
997 
998 	return (error);
999 }
1000 SYSCTL_PROC(_security_jail, OID_AUTO, jailed, CTLTYPE_INT | CTLFLAG_RD,
1001     NULL, 0, sysctl_jail_jailed, "I", "Process in jail?");
1002