xref: /freebsd/sys/kern/kern_jail.c (revision 53071ed1c96db7f89defc99c95b0ad1031d48f45)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 1999 Poul-Henning Kamp.
5  * Copyright (c) 2008 Bjoern A. Zeeb.
6  * Copyright (c) 2009 James Gritton.
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include "opt_ddb.h"
35 #include "opt_inet.h"
36 #include "opt_inet6.h"
37 
38 #include <sys/param.h>
39 #include <sys/types.h>
40 #include <sys/kernel.h>
41 #include <sys/systm.h>
42 #include <sys/errno.h>
43 #include <sys/sysproto.h>
44 #include <sys/malloc.h>
45 #include <sys/osd.h>
46 #include <sys/priv.h>
47 #include <sys/proc.h>
48 #include <sys/taskqueue.h>
49 #include <sys/fcntl.h>
50 #include <sys/jail.h>
51 #include <sys/lock.h>
52 #include <sys/mutex.h>
53 #include <sys/racct.h>
54 #include <sys/rctl.h>
55 #include <sys/refcount.h>
56 #include <sys/sx.h>
57 #include <sys/sysent.h>
58 #include <sys/namei.h>
59 #include <sys/mount.h>
60 #include <sys/queue.h>
61 #include <sys/socket.h>
62 #include <sys/syscallsubr.h>
63 #include <sys/sysctl.h>
64 #include <sys/vnode.h>
65 
66 #include <net/if.h>
67 #include <net/vnet.h>
68 
69 #include <netinet/in.h>
70 
71 #ifdef DDB
72 #include <ddb/ddb.h>
73 #endif /* DDB */
74 
75 #include <security/mac/mac_framework.h>
76 
77 #define	DEFAULT_HOSTUUID	"00000000-0000-0000-0000-000000000000"
78 
79 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
80 static MALLOC_DEFINE(M_PRISON_RACCT, "prison_racct", "Prison racct structures");
81 
82 /* Keep struct prison prison0 and some code in kern_jail_set() readable. */
83 #ifdef INET
84 #ifdef INET6
85 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL|PR_IP6_SADDRSEL
86 #else
87 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL
88 #endif
89 #else /* !INET */
90 #ifdef INET6
91 #define	_PR_IP_SADDRSEL	PR_IP6_SADDRSEL
92 #else
93 #define	_PR_IP_SADDRSEL	0
94 #endif
95 #endif
96 
97 /* prison0 describes what is "real" about the system. */
98 struct prison prison0 = {
99 	.pr_id		= 0,
100 	.pr_name	= "0",
101 	.pr_ref		= 1,
102 	.pr_uref	= 1,
103 	.pr_path	= "/",
104 	.pr_securelevel	= -1,
105 	.pr_devfs_rsnum = 0,
106 	.pr_childmax	= JAIL_MAX,
107 	.pr_hostuuid	= DEFAULT_HOSTUUID,
108 	.pr_children	= LIST_HEAD_INITIALIZER(prison0.pr_children),
109 #ifdef VIMAGE
110 	.pr_flags	= PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
111 #else
112 	.pr_flags	= PR_HOST|_PR_IP_SADDRSEL,
113 #endif
114 	.pr_allow	= PR_ALLOW_ALL_STATIC,
115 };
116 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
117 
118 struct bool_flags {
119 	const char	*name;
120 	const char	*noname;
121 	unsigned	 flag;
122 };
123 struct jailsys_flags {
124 	const char	*name;
125 	unsigned	 disable;
126 	unsigned	 new;
127 };
128 
129 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
130 struct	sx allprison_lock;
131 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
132 struct	prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
133 LIST_HEAD(, prison_racct) allprison_racct;
134 int	lastprid = 0;
135 
136 static int do_jail_attach(struct thread *td, struct prison *pr);
137 static void prison_complete(void *context, int pending);
138 static void prison_deref(struct prison *pr, int flags);
139 static char *prison_path(struct prison *pr1, struct prison *pr2);
140 static void prison_remove_one(struct prison *pr);
141 #ifdef RACCT
142 static void prison_racct_attach(struct prison *pr);
143 static void prison_racct_modify(struct prison *pr);
144 static void prison_racct_detach(struct prison *pr);
145 #endif
146 
147 /* Flags for prison_deref */
148 #define	PD_DEREF	0x01
149 #define	PD_DEUREF	0x02
150 #define	PD_LOCKED	0x04
151 #define	PD_LIST_SLOCKED	0x08
152 #define	PD_LIST_XLOCKED	0x10
153 
154 /*
155  * Parameter names corresponding to PR_* flag values.  Size values are for kvm
156  * as we cannot figure out the size of a sparse array, or an array without a
157  * terminating entry.
158  */
159 static struct bool_flags pr_flag_bool[] = {
160 	{"persist", "nopersist", PR_PERSIST},
161 #ifdef INET
162 	{"ip4.saddrsel", "ip4.nosaddrsel", PR_IP4_SADDRSEL},
163 #endif
164 #ifdef INET6
165 	{"ip6.saddrsel", "ip6.nosaddrsel", PR_IP6_SADDRSEL},
166 #endif
167 };
168 const size_t pr_flag_bool_size = sizeof(pr_flag_bool);
169 
170 static struct jailsys_flags pr_flag_jailsys[] = {
171 	{"host", 0, PR_HOST},
172 #ifdef VIMAGE
173 	{"vnet", 0, PR_VNET},
174 #endif
175 #ifdef INET
176 	{"ip4", PR_IP4_USER, PR_IP4_USER},
177 #endif
178 #ifdef INET6
179 	{"ip6", PR_IP6_USER, PR_IP6_USER},
180 #endif
181 };
182 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
183 
184 /* Make this array full-size so dynamic parameters can be added. */
185 static struct bool_flags pr_flag_allow[NBBY * NBPW] = {
186 	{"allow.set_hostname", "allow.noset_hostname", PR_ALLOW_SET_HOSTNAME},
187 	{"allow.sysvipc", "allow.nosysvipc", PR_ALLOW_SYSVIPC},
188 	{"allow.raw_sockets", "allow.noraw_sockets", PR_ALLOW_RAW_SOCKETS},
189 	{"allow.chflags", "allow.nochflags", PR_ALLOW_CHFLAGS},
190 	{"allow.mount", "allow.nomount", PR_ALLOW_MOUNT},
191 	{"allow.quotas", "allow.noquotas", PR_ALLOW_QUOTAS},
192 	{"allow.socket_af", "allow.nosocket_af", PR_ALLOW_SOCKET_AF},
193 	{"allow.mlock", "allow.nomlock", PR_ALLOW_MLOCK},
194 	{"allow.reserved_ports", "allow.noreserved_ports",
195 	 PR_ALLOW_RESERVED_PORTS},
196 	{"allow.read_msgbuf", "allow.noread_msgbuf", PR_ALLOW_READ_MSGBUF},
197 	{"allow.unprivileged_proc_debug", "allow.nounprivileged_proc_debug",
198 	 PR_ALLOW_UNPRIV_DEBUG},
199 };
200 const size_t pr_flag_allow_size = sizeof(pr_flag_allow);
201 
202 #define	JAIL_DEFAULT_ALLOW		(PR_ALLOW_SET_HOSTNAME | \
203 					 PR_ALLOW_RESERVED_PORTS | \
204 					 PR_ALLOW_UNPRIV_DEBUG)
205 #define	JAIL_DEFAULT_ENFORCE_STATFS	2
206 #define	JAIL_DEFAULT_DEVFS_RSNUM	0
207 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
208 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
209 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
210 #if defined(INET) || defined(INET6)
211 static unsigned jail_max_af_ips = 255;
212 #endif
213 
214 /*
215  * Initialize the parts of prison0 that can't be static-initialized with
216  * constants.  This is called from proc0_init() after creating thread0 cpuset.
217  */
218 void
219 prison0_init(void)
220 {
221 
222 	prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset);
223 	prison0.pr_osreldate = osreldate;
224 	strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease));
225 }
226 
227 /*
228  * struct jail_args {
229  *	struct jail *jail;
230  * };
231  */
232 int
233 sys_jail(struct thread *td, struct jail_args *uap)
234 {
235 	uint32_t version;
236 	int error;
237 	struct jail j;
238 
239 	error = copyin(uap->jail, &version, sizeof(uint32_t));
240 	if (error)
241 		return (error);
242 
243 	switch (version) {
244 	case 0:
245 	{
246 		struct jail_v0 j0;
247 
248 		/* FreeBSD single IPv4 jails. */
249 		bzero(&j, sizeof(struct jail));
250 		error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
251 		if (error)
252 			return (error);
253 		j.version = j0.version;
254 		j.path = j0.path;
255 		j.hostname = j0.hostname;
256 		j.ip4s = htonl(j0.ip_number);	/* jail_v0 is host order */
257 		break;
258 	}
259 
260 	case 1:
261 		/*
262 		 * Version 1 was used by multi-IPv4 jail implementations
263 		 * that never made it into the official kernel.
264 		 */
265 		return (EINVAL);
266 
267 	case 2:	/* JAIL_API_VERSION */
268 		/* FreeBSD multi-IPv4/IPv6,noIP jails. */
269 		error = copyin(uap->jail, &j, sizeof(struct jail));
270 		if (error)
271 			return (error);
272 		break;
273 
274 	default:
275 		/* Sci-Fi jails are not supported, sorry. */
276 		return (EINVAL);
277 	}
278 	return (kern_jail(td, &j));
279 }
280 
281 int
282 kern_jail(struct thread *td, struct jail *j)
283 {
284 	struct iovec optiov[2 * (4 + nitems(pr_flag_allow)
285 #ifdef INET
286 			    + 1
287 #endif
288 #ifdef INET6
289 			    + 1
290 #endif
291 			    )];
292 	struct uio opt;
293 	char *u_path, *u_hostname, *u_name;
294 	struct bool_flags *bf;
295 #ifdef INET
296 	uint32_t ip4s;
297 	struct in_addr *u_ip4;
298 #endif
299 #ifdef INET6
300 	struct in6_addr *u_ip6;
301 #endif
302 	size_t tmplen;
303 	int error, enforce_statfs;
304 
305 	bzero(&optiov, sizeof(optiov));
306 	opt.uio_iov = optiov;
307 	opt.uio_iovcnt = 0;
308 	opt.uio_offset = -1;
309 	opt.uio_resid = -1;
310 	opt.uio_segflg = UIO_SYSSPACE;
311 	opt.uio_rw = UIO_READ;
312 	opt.uio_td = td;
313 
314 	/* Set permissions for top-level jails from sysctls. */
315 	if (!jailed(td->td_ucred)) {
316 		for (bf = pr_flag_allow;
317 		     bf < pr_flag_allow + nitems(pr_flag_allow) &&
318 			bf->flag != 0;
319 		     bf++) {
320 			optiov[opt.uio_iovcnt].iov_base = __DECONST(char *,
321 			    (jail_default_allow & bf->flag)
322 			    ? bf->name : bf->noname);
323 			optiov[opt.uio_iovcnt].iov_len =
324 			    strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
325 			opt.uio_iovcnt += 2;
326 		}
327 		optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
328 		optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
329 		opt.uio_iovcnt++;
330 		enforce_statfs = jail_default_enforce_statfs;
331 		optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
332 		optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
333 		opt.uio_iovcnt++;
334 	}
335 
336 	tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
337 #ifdef INET
338 	ip4s = (j->version == 0) ? 1 : j->ip4s;
339 	if (ip4s > jail_max_af_ips)
340 		return (EINVAL);
341 	tmplen += ip4s * sizeof(struct in_addr);
342 #else
343 	if (j->ip4s > 0)
344 		return (EINVAL);
345 #endif
346 #ifdef INET6
347 	if (j->ip6s > jail_max_af_ips)
348 		return (EINVAL);
349 	tmplen += j->ip6s * sizeof(struct in6_addr);
350 #else
351 	if (j->ip6s > 0)
352 		return (EINVAL);
353 #endif
354 	u_path = malloc(tmplen, M_TEMP, M_WAITOK);
355 	u_hostname = u_path + MAXPATHLEN;
356 	u_name = u_hostname + MAXHOSTNAMELEN;
357 #ifdef INET
358 	u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
359 #endif
360 #ifdef INET6
361 #ifdef INET
362 	u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
363 #else
364 	u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
365 #endif
366 #endif
367 	optiov[opt.uio_iovcnt].iov_base = "path";
368 	optiov[opt.uio_iovcnt].iov_len = sizeof("path");
369 	opt.uio_iovcnt++;
370 	optiov[opt.uio_iovcnt].iov_base = u_path;
371 	error = copyinstr(j->path, u_path, MAXPATHLEN,
372 	    &optiov[opt.uio_iovcnt].iov_len);
373 	if (error) {
374 		free(u_path, M_TEMP);
375 		return (error);
376 	}
377 	opt.uio_iovcnt++;
378 	optiov[opt.uio_iovcnt].iov_base = "host.hostname";
379 	optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
380 	opt.uio_iovcnt++;
381 	optiov[opt.uio_iovcnt].iov_base = u_hostname;
382 	error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
383 	    &optiov[opt.uio_iovcnt].iov_len);
384 	if (error) {
385 		free(u_path, M_TEMP);
386 		return (error);
387 	}
388 	opt.uio_iovcnt++;
389 	if (j->jailname != NULL) {
390 		optiov[opt.uio_iovcnt].iov_base = "name";
391 		optiov[opt.uio_iovcnt].iov_len = sizeof("name");
392 		opt.uio_iovcnt++;
393 		optiov[opt.uio_iovcnt].iov_base = u_name;
394 		error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
395 		    &optiov[opt.uio_iovcnt].iov_len);
396 		if (error) {
397 			free(u_path, M_TEMP);
398 			return (error);
399 		}
400 		opt.uio_iovcnt++;
401 	}
402 #ifdef INET
403 	optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
404 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
405 	opt.uio_iovcnt++;
406 	optiov[opt.uio_iovcnt].iov_base = u_ip4;
407 	optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
408 	if (j->version == 0)
409 		u_ip4->s_addr = j->ip4s;
410 	else {
411 		error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
412 		if (error) {
413 			free(u_path, M_TEMP);
414 			return (error);
415 		}
416 	}
417 	opt.uio_iovcnt++;
418 #endif
419 #ifdef INET6
420 	optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
421 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
422 	opt.uio_iovcnt++;
423 	optiov[opt.uio_iovcnt].iov_base = u_ip6;
424 	optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
425 	error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
426 	if (error) {
427 		free(u_path, M_TEMP);
428 		return (error);
429 	}
430 	opt.uio_iovcnt++;
431 #endif
432 	KASSERT(opt.uio_iovcnt <= nitems(optiov),
433 		("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
434 	error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
435 	free(u_path, M_TEMP);
436 	return (error);
437 }
438 
439 /*
440  * struct jail_set_args {
441  *	struct iovec *iovp;
442  *	unsigned int iovcnt;
443  *	int flags;
444  * };
445  */
446 int
447 sys_jail_set(struct thread *td, struct jail_set_args *uap)
448 {
449 	struct uio *auio;
450 	int error;
451 
452 	/* Check that we have an even number of iovecs. */
453 	if (uap->iovcnt & 1)
454 		return (EINVAL);
455 
456 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
457 	if (error)
458 		return (error);
459 	error = kern_jail_set(td, auio, uap->flags);
460 	free(auio, M_IOV);
461 	return (error);
462 }
463 
464 int
465 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
466 {
467 	struct nameidata nd;
468 #ifdef INET
469 	struct in_addr *ip4;
470 #endif
471 #ifdef INET6
472 	struct in6_addr *ip6;
473 #endif
474 	struct vfsopt *opt;
475 	struct vfsoptlist *opts;
476 	struct prison *pr, *deadpr, *mypr, *ppr, *tpr;
477 	struct vnode *root;
478 	char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
479 	char *g_path, *osrelstr;
480 	struct bool_flags *bf;
481 	struct jailsys_flags *jsf;
482 #if defined(INET) || defined(INET6)
483 	struct prison *tppr;
484 	void *op;
485 #endif
486 	unsigned long hid;
487 	size_t namelen, onamelen, pnamelen;
488 	int born, created, cuflags, descend, enforce;
489 	int error, errmsg_len, errmsg_pos;
490 	int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
491 	int jid, jsys, len, level;
492 	int childmax, osreldt, rsnum, slevel;
493 	int fullpath_disabled;
494 #if defined(INET) || defined(INET6)
495 	int ii, ij;
496 #endif
497 #ifdef INET
498 	int ip4s, redo_ip4;
499 #endif
500 #ifdef INET6
501 	int ip6s, redo_ip6;
502 #endif
503 	uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
504 	uint64_t pr_allow_diff;
505 	unsigned tallow;
506 	char numbuf[12];
507 
508 	error = priv_check(td, PRIV_JAIL_SET);
509 	if (!error && (flags & JAIL_ATTACH))
510 		error = priv_check(td, PRIV_JAIL_ATTACH);
511 	if (error)
512 		return (error);
513 	mypr = td->td_ucred->cr_prison;
514 	if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
515 		return (EPERM);
516 	if (flags & ~JAIL_SET_MASK)
517 		return (EINVAL);
518 
519 	/*
520 	 * Check all the parameters before committing to anything.  Not all
521 	 * errors can be caught early, but we may as well try.  Also, this
522 	 * takes care of some expensive stuff (path lookup) before getting
523 	 * the allprison lock.
524 	 *
525 	 * XXX Jails are not filesystems, and jail parameters are not mount
526 	 *     options.  But it makes more sense to re-use the vfsopt code
527 	 *     than duplicate it under a different name.
528 	 */
529 	error = vfs_buildopts(optuio, &opts);
530 	if (error)
531 		return (error);
532 #ifdef INET
533 	ip4 = NULL;
534 #endif
535 #ifdef INET6
536 	ip6 = NULL;
537 #endif
538 	g_path = NULL;
539 
540 	cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
541 	if (!cuflags) {
542 		error = EINVAL;
543 		vfs_opterror(opts, "no valid operation (create or update)");
544 		goto done_errmsg;
545 	}
546 
547 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
548 	if (error == ENOENT)
549 		jid = 0;
550 	else if (error != 0)
551 		goto done_free;
552 
553 	error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
554 	if (error == ENOENT)
555 		gotslevel = 0;
556 	else if (error != 0)
557 		goto done_free;
558 	else
559 		gotslevel = 1;
560 
561 	error =
562 	    vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
563 	if (error == ENOENT)
564 		gotchildmax = 0;
565 	else if (error != 0)
566 		goto done_free;
567 	else
568 		gotchildmax = 1;
569 
570 	error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
571 	if (error == ENOENT)
572 		gotenforce = 0;
573 	else if (error != 0)
574 		goto done_free;
575 	else if (enforce < 0 || enforce > 2) {
576 		error = EINVAL;
577 		goto done_free;
578 	} else
579 		gotenforce = 1;
580 
581 	error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
582 	if (error == ENOENT)
583 		gotrsnum = 0;
584 	else if (error != 0)
585 		goto done_free;
586 	else
587 		gotrsnum = 1;
588 
589 	pr_flags = ch_flags = 0;
590 	for (bf = pr_flag_bool;
591 	     bf < pr_flag_bool + nitems(pr_flag_bool);
592 	     bf++) {
593 		vfs_flagopt(opts, bf->name, &pr_flags, bf->flag);
594 		vfs_flagopt(opts, bf->noname, &ch_flags, bf->flag);
595 	}
596 	ch_flags |= pr_flags;
597 	for (jsf = pr_flag_jailsys;
598 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
599 	     jsf++) {
600 		error = vfs_copyopt(opts, jsf->name, &jsys, sizeof(jsys));
601 		if (error == ENOENT)
602 			continue;
603 		if (error != 0)
604 			goto done_free;
605 		switch (jsys) {
606 		case JAIL_SYS_DISABLE:
607 			if (!jsf->disable) {
608 				error = EINVAL;
609 				goto done_free;
610 			}
611 			pr_flags |= jsf->disable;
612 			break;
613 		case JAIL_SYS_NEW:
614 			pr_flags |= jsf->new;
615 			break;
616 		case JAIL_SYS_INHERIT:
617 			break;
618 		default:
619 			error = EINVAL;
620 			goto done_free;
621 		}
622 		ch_flags |= jsf->new | jsf->disable;
623 	}
624 	if ((flags & (JAIL_CREATE | JAIL_UPDATE | JAIL_ATTACH)) == JAIL_CREATE
625 	    && !(pr_flags & PR_PERSIST)) {
626 		error = EINVAL;
627 		vfs_opterror(opts, "new jail must persist or attach");
628 		goto done_errmsg;
629 	}
630 #ifdef VIMAGE
631 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
632 		error = EINVAL;
633 		vfs_opterror(opts, "vnet cannot be changed after creation");
634 		goto done_errmsg;
635 	}
636 #endif
637 #ifdef INET
638 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
639 		error = EINVAL;
640 		vfs_opterror(opts, "ip4 cannot be changed after creation");
641 		goto done_errmsg;
642 	}
643 #endif
644 #ifdef INET6
645 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
646 		error = EINVAL;
647 		vfs_opterror(opts, "ip6 cannot be changed after creation");
648 		goto done_errmsg;
649 	}
650 #endif
651 
652 	pr_allow = ch_allow = 0;
653 	for (bf = pr_flag_allow;
654 	     bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
655 	     bf++) {
656 		vfs_flagopt(opts, bf->name, &pr_allow, bf->flag);
657 		vfs_flagopt(opts, bf->noname, &ch_allow, bf->flag);
658 	}
659 	ch_allow |= pr_allow;
660 
661 	error = vfs_getopt(opts, "name", (void **)&name, &len);
662 	if (error == ENOENT)
663 		name = NULL;
664 	else if (error != 0)
665 		goto done_free;
666 	else {
667 		if (len == 0 || name[len - 1] != '\0') {
668 			error = EINVAL;
669 			goto done_free;
670 		}
671 		if (len > MAXHOSTNAMELEN) {
672 			error = ENAMETOOLONG;
673 			goto done_free;
674 		}
675 	}
676 
677 	error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
678 	if (error == ENOENT)
679 		host = NULL;
680 	else if (error != 0)
681 		goto done_free;
682 	else {
683 		ch_flags |= PR_HOST;
684 		pr_flags |= PR_HOST;
685 		if (len == 0 || host[len - 1] != '\0') {
686 			error = EINVAL;
687 			goto done_free;
688 		}
689 		if (len > MAXHOSTNAMELEN) {
690 			error = ENAMETOOLONG;
691 			goto done_free;
692 		}
693 	}
694 
695 	error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
696 	if (error == ENOENT)
697 		domain = NULL;
698 	else if (error != 0)
699 		goto done_free;
700 	else {
701 		ch_flags |= PR_HOST;
702 		pr_flags |= PR_HOST;
703 		if (len == 0 || domain[len - 1] != '\0') {
704 			error = EINVAL;
705 			goto done_free;
706 		}
707 		if (len > MAXHOSTNAMELEN) {
708 			error = ENAMETOOLONG;
709 			goto done_free;
710 		}
711 	}
712 
713 	error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
714 	if (error == ENOENT)
715 		uuid = NULL;
716 	else if (error != 0)
717 		goto done_free;
718 	else {
719 		ch_flags |= PR_HOST;
720 		pr_flags |= PR_HOST;
721 		if (len == 0 || uuid[len - 1] != '\0') {
722 			error = EINVAL;
723 			goto done_free;
724 		}
725 		if (len > HOSTUUIDLEN) {
726 			error = ENAMETOOLONG;
727 			goto done_free;
728 		}
729 	}
730 
731 #ifdef COMPAT_FREEBSD32
732 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
733 		uint32_t hid32;
734 
735 		error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
736 		hid = hid32;
737 	} else
738 #endif
739 		error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
740 	if (error == ENOENT)
741 		gothid = 0;
742 	else if (error != 0)
743 		goto done_free;
744 	else {
745 		gothid = 1;
746 		ch_flags |= PR_HOST;
747 		pr_flags |= PR_HOST;
748 	}
749 
750 #ifdef INET
751 	error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
752 	if (error == ENOENT)
753 		ip4s = 0;
754 	else if (error != 0)
755 		goto done_free;
756 	else if (ip4s & (sizeof(*ip4) - 1)) {
757 		error = EINVAL;
758 		goto done_free;
759 	} else {
760 		ch_flags |= PR_IP4_USER;
761 		pr_flags |= PR_IP4_USER;
762 		if (ip4s > 0) {
763 			ip4s /= sizeof(*ip4);
764 			if (ip4s > jail_max_af_ips) {
765 				error = EINVAL;
766 				vfs_opterror(opts, "too many IPv4 addresses");
767 				goto done_errmsg;
768 			}
769 			ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
770 			bcopy(op, ip4, ip4s * sizeof(*ip4));
771 			/*
772 			 * IP addresses are all sorted but ip[0] to preserve
773 			 * the primary IP address as given from userland.
774 			 * This special IP is used for unbound outgoing
775 			 * connections as well for "loopback" traffic in case
776 			 * source address selection cannot find any more fitting
777 			 * address to connect from.
778 			 */
779 			if (ip4s > 1)
780 				qsort(ip4 + 1, ip4s - 1, sizeof(*ip4),
781 				    prison_qcmp_v4);
782 			/*
783 			 * Check for duplicate addresses and do some simple
784 			 * zero and broadcast checks. If users give other bogus
785 			 * addresses it is their problem.
786 			 *
787 			 * We do not have to care about byte order for these
788 			 * checks so we will do them in NBO.
789 			 */
790 			for (ii = 0; ii < ip4s; ii++) {
791 				if (ip4[ii].s_addr == INADDR_ANY ||
792 				    ip4[ii].s_addr == INADDR_BROADCAST) {
793 					error = EINVAL;
794 					goto done_free;
795 				}
796 				if ((ii+1) < ip4s &&
797 				    (ip4[0].s_addr == ip4[ii+1].s_addr ||
798 				     ip4[ii].s_addr == ip4[ii+1].s_addr)) {
799 					error = EINVAL;
800 					goto done_free;
801 				}
802 			}
803 		}
804 	}
805 #endif
806 
807 #ifdef INET6
808 	error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
809 	if (error == ENOENT)
810 		ip6s = 0;
811 	else if (error != 0)
812 		goto done_free;
813 	else if (ip6s & (sizeof(*ip6) - 1)) {
814 		error = EINVAL;
815 		goto done_free;
816 	} else {
817 		ch_flags |= PR_IP6_USER;
818 		pr_flags |= PR_IP6_USER;
819 		if (ip6s > 0) {
820 			ip6s /= sizeof(*ip6);
821 			if (ip6s > jail_max_af_ips) {
822 				error = EINVAL;
823 				vfs_opterror(opts, "too many IPv6 addresses");
824 				goto done_errmsg;
825 			}
826 			ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
827 			bcopy(op, ip6, ip6s * sizeof(*ip6));
828 			if (ip6s > 1)
829 				qsort(ip6 + 1, ip6s - 1, sizeof(*ip6),
830 				    prison_qcmp_v6);
831 			for (ii = 0; ii < ip6s; ii++) {
832 				if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
833 					error = EINVAL;
834 					goto done_free;
835 				}
836 				if ((ii+1) < ip6s &&
837 				    (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
838 				     IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
839 				{
840 					error = EINVAL;
841 					goto done_free;
842 				}
843 			}
844 		}
845 	}
846 #endif
847 
848 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
849 	if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
850 		error = EINVAL;
851 		vfs_opterror(opts,
852 		    "vnet jails cannot have IP address restrictions");
853 		goto done_errmsg;
854 	}
855 #endif
856 
857 	error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
858 	if (error == ENOENT)
859 		osrelstr = NULL;
860 	else if (error != 0)
861 		goto done_free;
862 	else {
863 		if (flags & JAIL_UPDATE) {
864 			error = EINVAL;
865 			vfs_opterror(opts,
866 			    "osrelease cannot be changed after creation");
867 			goto done_errmsg;
868 		}
869 		if (len == 0 || len >= OSRELEASELEN) {
870 			error = EINVAL;
871 			vfs_opterror(opts,
872 			    "osrelease string must be 1-%d bytes long",
873 			    OSRELEASELEN - 1);
874 			goto done_errmsg;
875 		}
876 	}
877 
878 	error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
879 	if (error == ENOENT)
880 		osreldt = 0;
881 	else if (error != 0)
882 		goto done_free;
883 	else {
884 		if (flags & JAIL_UPDATE) {
885 			error = EINVAL;
886 			vfs_opterror(opts,
887 			    "osreldate cannot be changed after creation");
888 			goto done_errmsg;
889 		}
890 		if (osreldt == 0) {
891 			error = EINVAL;
892 			vfs_opterror(opts, "osreldate cannot be 0");
893 			goto done_errmsg;
894 		}
895 	}
896 
897 	fullpath_disabled = 0;
898 	root = NULL;
899 	error = vfs_getopt(opts, "path", (void **)&path, &len);
900 	if (error == ENOENT)
901 		path = NULL;
902 	else if (error != 0)
903 		goto done_free;
904 	else {
905 		if (flags & JAIL_UPDATE) {
906 			error = EINVAL;
907 			vfs_opterror(opts,
908 			    "path cannot be changed after creation");
909 			goto done_errmsg;
910 		}
911 		if (len == 0 || path[len - 1] != '\0') {
912 			error = EINVAL;
913 			goto done_free;
914 		}
915 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
916 		    path, td);
917 		error = namei(&nd);
918 		if (error)
919 			goto done_free;
920 		root = nd.ni_vp;
921 		NDFREE(&nd, NDF_ONLY_PNBUF);
922 		g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
923 		strlcpy(g_path, path, MAXPATHLEN);
924 		error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
925 		if (error == 0)
926 			path = g_path;
927 		else if (error == ENODEV) {
928 			/* proceed if sysctl debug.disablefullpath == 1 */
929 			fullpath_disabled = 1;
930 			if (len < 2 || (len == 2 && path[0] == '/'))
931 				path = NULL;
932 		} else {
933 			/* exit on other errors */
934 			goto done_free;
935 		}
936 		if (root->v_type != VDIR) {
937 			error = ENOTDIR;
938 			vput(root);
939 			goto done_free;
940 		}
941 		VOP_UNLOCK(root);
942 		if (fullpath_disabled) {
943 			/* Leave room for a real-root full pathname. */
944 			if (len + (path[0] == '/' && strcmp(mypr->pr_path, "/")
945 			    ? strlen(mypr->pr_path) : 0) > MAXPATHLEN) {
946 				error = ENAMETOOLONG;
947 				vrele(root);
948 				goto done_free;
949 			}
950 		}
951 	}
952 
953 	/*
954 	 * Find the specified jail, or at least its parent.
955 	 * This abuses the file error codes ENOENT and EEXIST.
956 	 */
957 	pr = NULL;
958 	ppr = mypr;
959 	if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
960 		namelc = strrchr(name, '.');
961 		jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
962 		if (*p != '\0')
963 			jid = 0;
964 	}
965 	sx_xlock(&allprison_lock);
966 	if (jid != 0) {
967 		/*
968 		 * See if a requested jid already exists.  There is an
969 		 * information leak here if the jid exists but is not within
970 		 * the caller's jail hierarchy.  Jail creators will get EEXIST
971 		 * even though they cannot see the jail, and CREATE | UPDATE
972 		 * will return ENOENT which is not normally a valid error.
973 		 */
974 		if (jid < 0) {
975 			error = EINVAL;
976 			vfs_opterror(opts, "negative jid");
977 			goto done_unlock_list;
978 		}
979 		pr = prison_find(jid);
980 		if (pr != NULL) {
981 			ppr = pr->pr_parent;
982 			/* Create: jid must not exist. */
983 			if (cuflags == JAIL_CREATE) {
984 				mtx_unlock(&pr->pr_mtx);
985 				error = EEXIST;
986 				vfs_opterror(opts, "jail %d already exists",
987 				    jid);
988 				goto done_unlock_list;
989 			}
990 			if (!prison_ischild(mypr, pr)) {
991 				mtx_unlock(&pr->pr_mtx);
992 				pr = NULL;
993 			} else if (pr->pr_uref == 0) {
994 				if (!(flags & JAIL_DYING)) {
995 					mtx_unlock(&pr->pr_mtx);
996 					error = ENOENT;
997 					vfs_opterror(opts, "jail %d is dying",
998 					    jid);
999 					goto done_unlock_list;
1000 				} else if ((flags & JAIL_ATTACH) ||
1001 				    (pr_flags & PR_PERSIST)) {
1002 					/*
1003 					 * A dying jail might be resurrected
1004 					 * (via attach or persist), but first
1005 					 * it must determine if another jail
1006 					 * has claimed its name.  Accomplish
1007 					 * this by implicitly re-setting the
1008 					 * name.
1009 					 */
1010 					if (name == NULL)
1011 						name = prison_name(mypr, pr);
1012 				}
1013 			}
1014 		}
1015 		if (pr == NULL) {
1016 			/* Update: jid must exist. */
1017 			if (cuflags == JAIL_UPDATE) {
1018 				error = ENOENT;
1019 				vfs_opterror(opts, "jail %d not found", jid);
1020 				goto done_unlock_list;
1021 			}
1022 		}
1023 	}
1024 	/*
1025 	 * If the caller provided a name, look for a jail by that name.
1026 	 * This has different semantics for creates and updates keyed by jid
1027 	 * (where the name must not already exist in a different jail),
1028 	 * and updates keyed by the name itself (where the name must exist
1029 	 * because that is the jail being updated).
1030 	 */
1031 	namelc = NULL;
1032 	if (name != NULL) {
1033 		namelc = strrchr(name, '.');
1034 		if (namelc == NULL)
1035 			namelc = name;
1036 		else {
1037 			/*
1038 			 * This is a hierarchical name.  Split it into the
1039 			 * parent and child names, and make sure the parent
1040 			 * exists or matches an already found jail.
1041 			 */
1042 			if (pr != NULL) {
1043 				if (strncmp(name, ppr->pr_name, namelc - name)
1044 				    || ppr->pr_name[namelc - name] != '\0') {
1045 					mtx_unlock(&pr->pr_mtx);
1046 					error = EINVAL;
1047 					vfs_opterror(opts,
1048 					    "cannot change jail's parent");
1049 					goto done_unlock_list;
1050 				}
1051 			} else {
1052 				*namelc = '\0';
1053 				ppr = prison_find_name(mypr, name);
1054 				if (ppr == NULL) {
1055 					error = ENOENT;
1056 					vfs_opterror(opts,
1057 					    "jail \"%s\" not found", name);
1058 					goto done_unlock_list;
1059 				}
1060 				mtx_unlock(&ppr->pr_mtx);
1061 				*namelc = '.';
1062 			}
1063 			namelc++;
1064 		}
1065 		if (namelc[0] != '\0') {
1066 			pnamelen =
1067 			    (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1068  name_again:
1069 			deadpr = NULL;
1070 			FOREACH_PRISON_CHILD(ppr, tpr) {
1071 				if (tpr != pr && tpr->pr_ref > 0 &&
1072 				    !strcmp(tpr->pr_name + pnamelen, namelc)) {
1073 					if (pr == NULL &&
1074 					    cuflags != JAIL_CREATE) {
1075 						mtx_lock(&tpr->pr_mtx);
1076 						if (tpr->pr_ref > 0) {
1077 							/*
1078 							 * Use this jail
1079 							 * for updates.
1080 							 */
1081 							if (tpr->pr_uref > 0) {
1082 								pr = tpr;
1083 								break;
1084 							}
1085 							deadpr = tpr;
1086 						}
1087 						mtx_unlock(&tpr->pr_mtx);
1088 					} else if (tpr->pr_uref > 0) {
1089 						/*
1090 						 * Create, or update(jid):
1091 						 * name must not exist in an
1092 						 * active sibling jail.
1093 						 */
1094 						error = EEXIST;
1095 						if (pr != NULL)
1096 							mtx_unlock(&pr->pr_mtx);
1097 						vfs_opterror(opts,
1098 						   "jail \"%s\" already exists",
1099 						   name);
1100 						goto done_unlock_list;
1101 					}
1102 				}
1103 			}
1104 			/* If no active jail is found, use a dying one. */
1105 			if (deadpr != NULL && pr == NULL) {
1106 				if (flags & JAIL_DYING) {
1107 					mtx_lock(&deadpr->pr_mtx);
1108 					if (deadpr->pr_ref == 0) {
1109 						mtx_unlock(&deadpr->pr_mtx);
1110 						goto name_again;
1111 					}
1112 					pr = deadpr;
1113 				} else if (cuflags == JAIL_UPDATE) {
1114 					error = ENOENT;
1115 					vfs_opterror(opts,
1116 					    "jail \"%s\" is dying", name);
1117 					goto done_unlock_list;
1118 				}
1119 			}
1120 			/* Update: name must exist if no jid. */
1121 			else if (cuflags == JAIL_UPDATE && pr == NULL) {
1122 				error = ENOENT;
1123 				vfs_opterror(opts, "jail \"%s\" not found",
1124 				    name);
1125 				goto done_unlock_list;
1126 			}
1127 		}
1128 	}
1129 	/* Update: must provide a jid or name. */
1130 	else if (cuflags == JAIL_UPDATE && pr == NULL) {
1131 		error = ENOENT;
1132 		vfs_opterror(opts, "update specified no jail");
1133 		goto done_unlock_list;
1134 	}
1135 
1136 	/* If there's no prison to update, create a new one and link it in. */
1137 	if (pr == NULL) {
1138 		for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1139 			if (tpr->pr_childcount >= tpr->pr_childmax) {
1140 				error = EPERM;
1141 				vfs_opterror(opts, "prison limit exceeded");
1142 				goto done_unlock_list;
1143 			}
1144 		created = 1;
1145 		mtx_lock(&ppr->pr_mtx);
1146 		if (ppr->pr_ref == 0) {
1147 			mtx_unlock(&ppr->pr_mtx);
1148 			error = ENOENT;
1149 			vfs_opterror(opts, "jail \"%s\" not found",
1150 			    prison_name(mypr, ppr));
1151 			goto done_unlock_list;
1152 		}
1153 		ppr->pr_ref++;
1154 		ppr->pr_uref++;
1155 		mtx_unlock(&ppr->pr_mtx);
1156 		pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1157 		if (jid == 0) {
1158 			/* Find the next free jid. */
1159 			jid = lastprid + 1;
1160  findnext:
1161 			if (jid == JAIL_MAX)
1162 				jid = 1;
1163 			TAILQ_FOREACH(tpr, &allprison, pr_list) {
1164 				if (tpr->pr_id < jid)
1165 					continue;
1166 				if (tpr->pr_id > jid || tpr->pr_ref == 0) {
1167 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1168 					break;
1169 				}
1170 				if (jid == lastprid) {
1171 					error = EAGAIN;
1172 					vfs_opterror(opts,
1173 					    "no available jail IDs");
1174 					free(pr, M_PRISON);
1175 					prison_deref(ppr, PD_DEREF |
1176 					    PD_DEUREF | PD_LIST_XLOCKED);
1177 					goto done_releroot;
1178 				}
1179 				jid++;
1180 				goto findnext;
1181 			}
1182 			lastprid = jid;
1183 		} else {
1184 			/*
1185 			 * The jail already has a jid (that did not yet exist),
1186 			 * so just find where to insert it.
1187 			 */
1188 			TAILQ_FOREACH(tpr, &allprison, pr_list)
1189 				if (tpr->pr_id >= jid) {
1190 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1191 					break;
1192 				}
1193 		}
1194 		if (tpr == NULL)
1195 			TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1196 		LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1197 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1198 			tpr->pr_childcount++;
1199 
1200 		pr->pr_parent = ppr;
1201 		pr->pr_id = jid;
1202 
1203 		/* Set some default values, and inherit some from the parent. */
1204 		if (namelc == NULL)
1205 			namelc = "";
1206 		if (path == NULL) {
1207 			path = "/";
1208 			root = mypr->pr_root;
1209 			vref(root);
1210 		}
1211 		strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1212 		pr->pr_flags |= PR_HOST;
1213 #if defined(INET) || defined(INET6)
1214 #ifdef VIMAGE
1215 		if (!(pr_flags & PR_VNET))
1216 #endif
1217 		{
1218 #ifdef INET
1219 			if (!(ch_flags & PR_IP4_USER))
1220 				pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1221 			else if (!(pr_flags & PR_IP4_USER)) {
1222 				pr->pr_flags |= ppr->pr_flags & PR_IP4;
1223 				if (ppr->pr_ip4 != NULL) {
1224 					pr->pr_ip4s = ppr->pr_ip4s;
1225 					pr->pr_ip4 = malloc(pr->pr_ip4s *
1226 					    sizeof(struct in_addr), M_PRISON,
1227 					    M_WAITOK);
1228 					bcopy(ppr->pr_ip4, pr->pr_ip4,
1229 					    pr->pr_ip4s * sizeof(*pr->pr_ip4));
1230 				}
1231 			}
1232 #endif
1233 #ifdef INET6
1234 			if (!(ch_flags & PR_IP6_USER))
1235 				pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1236 			else if (!(pr_flags & PR_IP6_USER)) {
1237 				pr->pr_flags |= ppr->pr_flags & PR_IP6;
1238 				if (ppr->pr_ip6 != NULL) {
1239 					pr->pr_ip6s = ppr->pr_ip6s;
1240 					pr->pr_ip6 = malloc(pr->pr_ip6s *
1241 					    sizeof(struct in6_addr), M_PRISON,
1242 					    M_WAITOK);
1243 					bcopy(ppr->pr_ip6, pr->pr_ip6,
1244 					    pr->pr_ip6s * sizeof(*pr->pr_ip6));
1245 				}
1246 			}
1247 #endif
1248 		}
1249 #endif
1250 		/* Source address selection is always on by default. */
1251 		pr->pr_flags |= _PR_IP_SADDRSEL;
1252 
1253 		pr->pr_securelevel = ppr->pr_securelevel;
1254 		pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1255 		pr->pr_enforce_statfs = jail_default_enforce_statfs;
1256 		pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1257 
1258 		pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1259 		if (osrelstr == NULL)
1260 		    strcpy(pr->pr_osrelease, ppr->pr_osrelease);
1261 		else
1262 		    strcpy(pr->pr_osrelease, osrelstr);
1263 
1264 		LIST_INIT(&pr->pr_children);
1265 		mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1266 		TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1267 
1268 #ifdef VIMAGE
1269 		/* Allocate a new vnet if specified. */
1270 		pr->pr_vnet = (pr_flags & PR_VNET)
1271 		    ? vnet_alloc() : ppr->pr_vnet;
1272 #endif
1273 		/*
1274 		 * Allocate a dedicated cpuset for each jail.
1275 		 * Unlike other initial settings, this may return an erorr.
1276 		 */
1277 		error = cpuset_create_root(ppr, &pr->pr_cpuset);
1278 		if (error) {
1279 			prison_deref(pr, PD_LIST_XLOCKED);
1280 			goto done_releroot;
1281 		}
1282 
1283 		mtx_lock(&pr->pr_mtx);
1284 		/*
1285 		 * New prisons do not yet have a reference, because we do not
1286 		 * want others to see the incomplete prison once the
1287 		 * allprison_lock is downgraded.
1288 		 */
1289 	} else {
1290 		created = 0;
1291 		/*
1292 		 * Grab a reference for existing prisons, to ensure they
1293 		 * continue to exist for the duration of the call.
1294 		 */
1295 		pr->pr_ref++;
1296 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1297 		if ((pr->pr_flags & PR_VNET) &&
1298 		    (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1299 			error = EINVAL;
1300 			vfs_opterror(opts,
1301 			    "vnet jails cannot have IP address restrictions");
1302 			goto done_deref_locked;
1303 		}
1304 #endif
1305 #ifdef INET
1306 		if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1307 			error = EINVAL;
1308 			vfs_opterror(opts,
1309 			    "ip4 cannot be changed after creation");
1310 			goto done_deref_locked;
1311 		}
1312 #endif
1313 #ifdef INET6
1314 		if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1315 			error = EINVAL;
1316 			vfs_opterror(opts,
1317 			    "ip6 cannot be changed after creation");
1318 			goto done_deref_locked;
1319 		}
1320 #endif
1321 	}
1322 
1323 	/* Do final error checking before setting anything. */
1324 	if (gotslevel) {
1325 		if (slevel < ppr->pr_securelevel) {
1326 			error = EPERM;
1327 			goto done_deref_locked;
1328 		}
1329 	}
1330 	if (gotchildmax) {
1331 		if (childmax >= ppr->pr_childmax) {
1332 			error = EPERM;
1333 			goto done_deref_locked;
1334 		}
1335 	}
1336 	if (gotenforce) {
1337 		if (enforce < ppr->pr_enforce_statfs) {
1338 			error = EPERM;
1339 			goto done_deref_locked;
1340 		}
1341 	}
1342 	if (gotrsnum) {
1343 		/*
1344 		 * devfs_rsnum is a uint16_t
1345 		 */
1346 		if (rsnum < 0 || rsnum > 65535) {
1347 			error = EINVAL;
1348 			goto done_deref_locked;
1349 		}
1350 		/*
1351 		 * Nested jails always inherit parent's devfs ruleset
1352 		 */
1353 		if (jailed(td->td_ucred)) {
1354 			if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1355 				error = EPERM;
1356 				goto done_deref_locked;
1357 			} else
1358 				rsnum = ppr->pr_devfs_rsnum;
1359 		}
1360 	}
1361 #ifdef INET
1362 	if (ip4s > 0) {
1363 		if (ppr->pr_flags & PR_IP4) {
1364 			/*
1365 			 * Make sure the new set of IP addresses is a
1366 			 * subset of the parent's list.  Don't worry
1367 			 * about the parent being unlocked, as any
1368 			 * setting is done with allprison_lock held.
1369 			 */
1370 			for (ij = 0; ij < ppr->pr_ip4s; ij++)
1371 				if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
1372 					break;
1373 			if (ij == ppr->pr_ip4s) {
1374 				error = EPERM;
1375 				goto done_deref_locked;
1376 			}
1377 			if (ip4s > 1) {
1378 				for (ii = ij = 1; ii < ip4s; ii++) {
1379 					if (ip4[ii].s_addr ==
1380 					    ppr->pr_ip4[0].s_addr)
1381 						continue;
1382 					for (; ij < ppr->pr_ip4s; ij++)
1383 						if (ip4[ii].s_addr ==
1384 						    ppr->pr_ip4[ij].s_addr)
1385 							break;
1386 					if (ij == ppr->pr_ip4s)
1387 						break;
1388 				}
1389 				if (ij == ppr->pr_ip4s) {
1390 					error = EPERM;
1391 					goto done_deref_locked;
1392 				}
1393 			}
1394 		}
1395 		/*
1396 		 * Check for conflicting IP addresses.  We permit them
1397 		 * if there is no more than one IP on each jail.  If
1398 		 * there is a duplicate on a jail with more than one
1399 		 * IP stop checking and return error.
1400 		 */
1401 #ifdef VIMAGE
1402 		for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
1403 			if (tppr->pr_flags & PR_VNET)
1404 				break;
1405 #else
1406 		tppr = &prison0;
1407 #endif
1408 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1409 			if (tpr == pr ||
1410 #ifdef VIMAGE
1411 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1412 #endif
1413 			    tpr->pr_uref == 0) {
1414 				descend = 0;
1415 				continue;
1416 			}
1417 			if (!(tpr->pr_flags & PR_IP4_USER))
1418 				continue;
1419 			descend = 0;
1420 			if (tpr->pr_ip4 == NULL ||
1421 			    (ip4s == 1 && tpr->pr_ip4s == 1))
1422 				continue;
1423 			for (ii = 0; ii < ip4s; ii++) {
1424 				if (prison_check_ip4_locked(tpr, &ip4[ii]) ==
1425 				    0) {
1426 					error = EADDRINUSE;
1427 					vfs_opterror(opts,
1428 					    "IPv4 addresses clash");
1429 					goto done_deref_locked;
1430 				}
1431 			}
1432 		}
1433 	}
1434 #endif
1435 #ifdef INET6
1436 	if (ip6s > 0) {
1437 		if (ppr->pr_flags & PR_IP6) {
1438 			/*
1439 			 * Make sure the new set of IP addresses is a
1440 			 * subset of the parent's list.
1441 			 */
1442 			for (ij = 0; ij < ppr->pr_ip6s; ij++)
1443 				if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1444 				    &ppr->pr_ip6[ij]))
1445 					break;
1446 			if (ij == ppr->pr_ip6s) {
1447 				error = EPERM;
1448 				goto done_deref_locked;
1449 			}
1450 			if (ip6s > 1) {
1451 				for (ii = ij = 1; ii < ip6s; ii++) {
1452 					if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1453 					     &ppr->pr_ip6[0]))
1454 						continue;
1455 					for (; ij < ppr->pr_ip6s; ij++)
1456 						if (IN6_ARE_ADDR_EQUAL(
1457 						    &ip6[ii], &ppr->pr_ip6[ij]))
1458 							break;
1459 					if (ij == ppr->pr_ip6s)
1460 						break;
1461 				}
1462 				if (ij == ppr->pr_ip6s) {
1463 					error = EPERM;
1464 					goto done_deref_locked;
1465 				}
1466 			}
1467 		}
1468 		/* Check for conflicting IP addresses. */
1469 #ifdef VIMAGE
1470 		for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
1471 			if (tppr->pr_flags & PR_VNET)
1472 				break;
1473 #else
1474 		tppr = &prison0;
1475 #endif
1476 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1477 			if (tpr == pr ||
1478 #ifdef VIMAGE
1479 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1480 #endif
1481 			    tpr->pr_uref == 0) {
1482 				descend = 0;
1483 				continue;
1484 			}
1485 			if (!(tpr->pr_flags & PR_IP6_USER))
1486 				continue;
1487 			descend = 0;
1488 			if (tpr->pr_ip6 == NULL ||
1489 			    (ip6s == 1 && tpr->pr_ip6s == 1))
1490 				continue;
1491 			for (ii = 0; ii < ip6s; ii++) {
1492 				if (prison_check_ip6_locked(tpr, &ip6[ii]) ==
1493 				    0) {
1494 					error = EADDRINUSE;
1495 					vfs_opterror(opts,
1496 					    "IPv6 addresses clash");
1497 					goto done_deref_locked;
1498 				}
1499 			}
1500 		}
1501 	}
1502 #endif
1503 	onamelen = namelen = 0;
1504 	if (namelc != NULL) {
1505 		/* Give a default name of the jid.  Also allow the name to be
1506 		 * explicitly the jid - but not any other number, and only in
1507 		 * normal form (no leading zero/etc).
1508 		 */
1509 		if (namelc[0] == '\0')
1510 			snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1511 		else if ((strtoul(namelc, &p, 10) != jid ||
1512 			  namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1513 			error = EINVAL;
1514 			vfs_opterror(opts,
1515 			    "name cannot be numeric (unless it is the jid)");
1516 			goto done_deref_locked;
1517 		}
1518 		/*
1519 		 * Make sure the name isn't too long for the prison or its
1520 		 * children.
1521 		 */
1522 		pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1523 		onamelen = strlen(pr->pr_name + pnamelen);
1524 		namelen = strlen(namelc);
1525 		if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1526 			error = ENAMETOOLONG;
1527 			goto done_deref_locked;
1528 		}
1529 		FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1530 			if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1531 			    sizeof(pr->pr_name)) {
1532 				error = ENAMETOOLONG;
1533 				goto done_deref_locked;
1534 			}
1535 		}
1536 	}
1537 	pr_allow_diff = pr_allow & ~ppr->pr_allow;
1538 	if (pr_allow_diff & ~PR_ALLOW_DIFFERENCES) {
1539 		error = EPERM;
1540 		goto done_deref_locked;
1541 	}
1542 
1543 	/*
1544 	 * Let modules check their parameters.  This requires unlocking and
1545 	 * then re-locking the prison, but this is still a valid state as long
1546 	 * as allprison_lock remains xlocked.
1547 	 */
1548 	mtx_unlock(&pr->pr_mtx);
1549 	error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1550 	if (error != 0) {
1551 		prison_deref(pr, created
1552 		    ? PD_LIST_XLOCKED
1553 		    : PD_DEREF | PD_LIST_XLOCKED);
1554 		goto done_releroot;
1555 	}
1556 	mtx_lock(&pr->pr_mtx);
1557 
1558 	/* At this point, all valid parameters should have been noted. */
1559 	TAILQ_FOREACH(opt, opts, link) {
1560 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
1561 			error = EINVAL;
1562 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
1563 			goto done_deref_locked;
1564 		}
1565 	}
1566 
1567 	/* Set the parameters of the prison. */
1568 #ifdef INET
1569 	redo_ip4 = 0;
1570 	if (pr_flags & PR_IP4_USER) {
1571 		pr->pr_flags |= PR_IP4;
1572 		free(pr->pr_ip4, M_PRISON);
1573 		pr->pr_ip4s = ip4s;
1574 		pr->pr_ip4 = ip4;
1575 		ip4 = NULL;
1576 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1577 #ifdef VIMAGE
1578 			if (tpr->pr_flags & PR_VNET) {
1579 				descend = 0;
1580 				continue;
1581 			}
1582 #endif
1583 			if (prison_restrict_ip4(tpr, NULL)) {
1584 				redo_ip4 = 1;
1585 				descend = 0;
1586 			}
1587 		}
1588 	}
1589 #endif
1590 #ifdef INET6
1591 	redo_ip6 = 0;
1592 	if (pr_flags & PR_IP6_USER) {
1593 		pr->pr_flags |= PR_IP6;
1594 		free(pr->pr_ip6, M_PRISON);
1595 		pr->pr_ip6s = ip6s;
1596 		pr->pr_ip6 = ip6;
1597 		ip6 = NULL;
1598 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1599 #ifdef VIMAGE
1600 			if (tpr->pr_flags & PR_VNET) {
1601 				descend = 0;
1602 				continue;
1603 			}
1604 #endif
1605 			if (prison_restrict_ip6(tpr, NULL)) {
1606 				redo_ip6 = 1;
1607 				descend = 0;
1608 			}
1609 		}
1610 	}
1611 #endif
1612 	if (gotslevel) {
1613 		pr->pr_securelevel = slevel;
1614 		/* Set all child jails to be at least this level. */
1615 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1616 			if (tpr->pr_securelevel < slevel)
1617 				tpr->pr_securelevel = slevel;
1618 	}
1619 	if (gotchildmax) {
1620 		pr->pr_childmax = childmax;
1621 		/* Set all child jails to under this limit. */
1622 		FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1623 			if (tpr->pr_childmax > childmax - level)
1624 				tpr->pr_childmax = childmax > level
1625 				    ? childmax - level : 0;
1626 	}
1627 	if (gotenforce) {
1628 		pr->pr_enforce_statfs = enforce;
1629 		/* Pass this restriction on to the children. */
1630 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1631 			if (tpr->pr_enforce_statfs < enforce)
1632 				tpr->pr_enforce_statfs = enforce;
1633 	}
1634 	if (gotrsnum) {
1635 		pr->pr_devfs_rsnum = rsnum;
1636 		/* Pass this restriction on to the children. */
1637 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1638 			tpr->pr_devfs_rsnum = rsnum;
1639 	}
1640 	if (namelc != NULL) {
1641 		if (ppr == &prison0)
1642 			strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
1643 		else
1644 			snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1645 			    ppr->pr_name, namelc);
1646 		/* Change this component of child names. */
1647 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1648 			bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1649 			    strlen(tpr->pr_name + onamelen) + 1);
1650 			bcopy(pr->pr_name, tpr->pr_name, namelen);
1651 		}
1652 	}
1653 	if (path != NULL) {
1654 		/* Try to keep a real-rooted full pathname. */
1655 		if (fullpath_disabled && path[0] == '/' &&
1656 		    strcmp(mypr->pr_path, "/"))
1657 			snprintf(pr->pr_path, sizeof(pr->pr_path), "%s%s",
1658 			    mypr->pr_path, path);
1659 		else
1660 			strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1661 		pr->pr_root = root;
1662 	}
1663 	if (PR_HOST & ch_flags & ~pr_flags) {
1664 		if (pr->pr_flags & PR_HOST) {
1665 			/*
1666 			 * Copy the parent's host info.  As with pr_ip4 above,
1667 			 * the lack of a lock on the parent is not a problem;
1668 			 * it is always set with allprison_lock at least
1669 			 * shared, and is held exclusively here.
1670 			 */
1671 			strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1672 			    sizeof(pr->pr_hostname));
1673 			strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1674 			    sizeof(pr->pr_domainname));
1675 			strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1676 			    sizeof(pr->pr_hostuuid));
1677 			pr->pr_hostid = pr->pr_parent->pr_hostid;
1678 		}
1679 	} else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1680 		/* Set this prison, and any descendants without PR_HOST. */
1681 		if (host != NULL)
1682 			strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1683 		if (domain != NULL)
1684 			strlcpy(pr->pr_domainname, domain,
1685 			    sizeof(pr->pr_domainname));
1686 		if (uuid != NULL)
1687 			strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1688 		if (gothid)
1689 			pr->pr_hostid = hid;
1690 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1691 			if (tpr->pr_flags & PR_HOST)
1692 				descend = 0;
1693 			else {
1694 				if (host != NULL)
1695 					strlcpy(tpr->pr_hostname,
1696 					    pr->pr_hostname,
1697 					    sizeof(tpr->pr_hostname));
1698 				if (domain != NULL)
1699 					strlcpy(tpr->pr_domainname,
1700 					    pr->pr_domainname,
1701 					    sizeof(tpr->pr_domainname));
1702 				if (uuid != NULL)
1703 					strlcpy(tpr->pr_hostuuid,
1704 					    pr->pr_hostuuid,
1705 					    sizeof(tpr->pr_hostuuid));
1706 				if (gothid)
1707 					tpr->pr_hostid = hid;
1708 			}
1709 		}
1710 	}
1711 	if ((tallow = ch_allow & ~pr_allow)) {
1712 		/* Clear allow bits in all children. */
1713 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1714 			tpr->pr_allow &= ~tallow;
1715 	}
1716 	pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1717 	/*
1718 	 * Persistent prisons get an extra reference, and prisons losing their
1719 	 * persist flag lose that reference.  Only do this for existing prisons
1720 	 * for now, so new ones will remain unseen until after the module
1721 	 * handlers have completed.
1722 	 */
1723 	born = pr->pr_uref == 0;
1724 	if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) {
1725 		if (pr_flags & PR_PERSIST) {
1726 			pr->pr_ref++;
1727 			pr->pr_uref++;
1728 		} else {
1729 			pr->pr_ref--;
1730 			pr->pr_uref--;
1731 		}
1732 	}
1733 	pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1734 	mtx_unlock(&pr->pr_mtx);
1735 
1736 #ifdef RACCT
1737 	if (racct_enable && created)
1738 		prison_racct_attach(pr);
1739 #endif
1740 
1741 	/* Locks may have prevented a complete restriction of child IP
1742 	 * addresses.  If so, allocate some more memory and try again.
1743 	 */
1744 #ifdef INET
1745 	while (redo_ip4) {
1746 		ip4s = pr->pr_ip4s;
1747 		ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1748 		mtx_lock(&pr->pr_mtx);
1749 		redo_ip4 = 0;
1750 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1751 #ifdef VIMAGE
1752 			if (tpr->pr_flags & PR_VNET) {
1753 				descend = 0;
1754 				continue;
1755 			}
1756 #endif
1757 			if (prison_restrict_ip4(tpr, ip4)) {
1758 				if (ip4 != NULL)
1759 					ip4 = NULL;
1760 				else
1761 					redo_ip4 = 1;
1762 			}
1763 		}
1764 		mtx_unlock(&pr->pr_mtx);
1765 	}
1766 #endif
1767 #ifdef INET6
1768 	while (redo_ip6) {
1769 		ip6s = pr->pr_ip6s;
1770 		ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1771 		mtx_lock(&pr->pr_mtx);
1772 		redo_ip6 = 0;
1773 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1774 #ifdef VIMAGE
1775 			if (tpr->pr_flags & PR_VNET) {
1776 				descend = 0;
1777 				continue;
1778 			}
1779 #endif
1780 			if (prison_restrict_ip6(tpr, ip6)) {
1781 				if (ip6 != NULL)
1782 					ip6 = NULL;
1783 				else
1784 					redo_ip6 = 1;
1785 			}
1786 		}
1787 		mtx_unlock(&pr->pr_mtx);
1788 	}
1789 #endif
1790 
1791 	/* Let the modules do their work. */
1792 	sx_downgrade(&allprison_lock);
1793 	if (born) {
1794 		error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1795 		if (error) {
1796 			(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1797 			prison_deref(pr, created
1798 			    ? PD_LIST_SLOCKED
1799 			    : PD_DEREF | PD_LIST_SLOCKED);
1800 			goto done_errmsg;
1801 		}
1802 	}
1803 	error = osd_jail_call(pr, PR_METHOD_SET, opts);
1804 	if (error) {
1805 		if (born)
1806 			(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1807 		prison_deref(pr, created
1808 		    ? PD_LIST_SLOCKED
1809 		    : PD_DEREF | PD_LIST_SLOCKED);
1810 		goto done_errmsg;
1811 	}
1812 
1813 	/* Attach this process to the prison if requested. */
1814 	if (flags & JAIL_ATTACH) {
1815 		mtx_lock(&pr->pr_mtx);
1816 		error = do_jail_attach(td, pr);
1817 		if (error) {
1818 			vfs_opterror(opts, "attach failed");
1819 			if (!created)
1820 				prison_deref(pr, PD_DEREF);
1821 			goto done_errmsg;
1822 		}
1823 	}
1824 
1825 #ifdef RACCT
1826 	if (racct_enable && !created) {
1827 		if (!(flags & JAIL_ATTACH))
1828 			sx_sunlock(&allprison_lock);
1829 		prison_racct_modify(pr);
1830 		if (!(flags & JAIL_ATTACH))
1831 			sx_slock(&allprison_lock);
1832 	}
1833 #endif
1834 
1835 	td->td_retval[0] = pr->pr_id;
1836 
1837 	/*
1838 	 * Now that it is all there, drop the temporary reference from existing
1839 	 * prisons.  Or add a reference to newly created persistent prisons
1840 	 * (which was not done earlier so that the prison would not be publicly
1841 	 * visible).
1842 	 */
1843 	if (!created) {
1844 		prison_deref(pr, (flags & JAIL_ATTACH)
1845 		    ? PD_DEREF
1846 		    : PD_DEREF | PD_LIST_SLOCKED);
1847 	} else {
1848 		if (pr_flags & PR_PERSIST) {
1849 			mtx_lock(&pr->pr_mtx);
1850 			pr->pr_ref++;
1851 			pr->pr_uref++;
1852 			mtx_unlock(&pr->pr_mtx);
1853 		}
1854 		if (!(flags & JAIL_ATTACH))
1855 			sx_sunlock(&allprison_lock);
1856 	}
1857 
1858 	goto done_free;
1859 
1860  done_deref_locked:
1861 	prison_deref(pr, created
1862 	    ? PD_LOCKED | PD_LIST_XLOCKED
1863 	    : PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
1864 	goto done_releroot;
1865  done_unlock_list:
1866 	sx_xunlock(&allprison_lock);
1867  done_releroot:
1868 	if (root != NULL)
1869 		vrele(root);
1870  done_errmsg:
1871 	if (error) {
1872 		if (vfs_getopt(opts, "errmsg", (void **)&errmsg,
1873 		    &errmsg_len) == 0 && errmsg_len > 0) {
1874 			errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1875 			if (optuio->uio_segflg == UIO_SYSSPACE)
1876 				bcopy(errmsg,
1877 				    optuio->uio_iov[errmsg_pos].iov_base,
1878 				    errmsg_len);
1879 			else
1880 				copyout(errmsg,
1881 				    optuio->uio_iov[errmsg_pos].iov_base,
1882 				    errmsg_len);
1883 		}
1884 	}
1885  done_free:
1886 #ifdef INET
1887 	free(ip4, M_PRISON);
1888 #endif
1889 #ifdef INET6
1890 	free(ip6, M_PRISON);
1891 #endif
1892 	if (g_path != NULL)
1893 		free(g_path, M_TEMP);
1894 	vfs_freeopts(opts);
1895 	return (error);
1896 }
1897 
1898 /*
1899  * struct jail_get_args {
1900  *	struct iovec *iovp;
1901  *	unsigned int iovcnt;
1902  *	int flags;
1903  * };
1904  */
1905 int
1906 sys_jail_get(struct thread *td, struct jail_get_args *uap)
1907 {
1908 	struct uio *auio;
1909 	int error;
1910 
1911 	/* Check that we have an even number of iovecs. */
1912 	if (uap->iovcnt & 1)
1913 		return (EINVAL);
1914 
1915 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
1916 	if (error)
1917 		return (error);
1918 	error = kern_jail_get(td, auio, uap->flags);
1919 	if (error == 0)
1920 		error = copyout(auio->uio_iov, uap->iovp,
1921 		    uap->iovcnt * sizeof (struct iovec));
1922 	free(auio, M_IOV);
1923 	return (error);
1924 }
1925 
1926 int
1927 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
1928 {
1929 	struct bool_flags *bf;
1930 	struct jailsys_flags *jsf;
1931 	struct prison *pr, *mypr;
1932 	struct vfsopt *opt;
1933 	struct vfsoptlist *opts;
1934 	char *errmsg, *name;
1935 	int error, errmsg_len, errmsg_pos, i, jid, len, locked, pos;
1936 	unsigned f;
1937 
1938 	if (flags & ~JAIL_GET_MASK)
1939 		return (EINVAL);
1940 
1941 	/* Get the parameter list. */
1942 	error = vfs_buildopts(optuio, &opts);
1943 	if (error)
1944 		return (error);
1945 	errmsg_pos = vfs_getopt_pos(opts, "errmsg");
1946 	mypr = td->td_ucred->cr_prison;
1947 
1948 	/*
1949 	 * Find the prison specified by one of: lastjid, jid, name.
1950 	 */
1951 	sx_slock(&allprison_lock);
1952 	error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
1953 	if (error == 0) {
1954 		TAILQ_FOREACH(pr, &allprison, pr_list) {
1955 			if (pr->pr_id > jid && prison_ischild(mypr, pr)) {
1956 				mtx_lock(&pr->pr_mtx);
1957 				if (pr->pr_ref > 0 &&
1958 				    (pr->pr_uref > 0 || (flags & JAIL_DYING)))
1959 					break;
1960 				mtx_unlock(&pr->pr_mtx);
1961 			}
1962 		}
1963 		if (pr != NULL)
1964 			goto found_prison;
1965 		error = ENOENT;
1966 		vfs_opterror(opts, "no jail after %d", jid);
1967 		goto done_unlock_list;
1968 	} else if (error != ENOENT)
1969 		goto done_unlock_list;
1970 
1971 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1972 	if (error == 0) {
1973 		if (jid != 0) {
1974 			pr = prison_find_child(mypr, jid);
1975 			if (pr != NULL) {
1976 				if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1977 					mtx_unlock(&pr->pr_mtx);
1978 					error = ENOENT;
1979 					vfs_opterror(opts, "jail %d is dying",
1980 					    jid);
1981 					goto done_unlock_list;
1982 				}
1983 				goto found_prison;
1984 			}
1985 			error = ENOENT;
1986 			vfs_opterror(opts, "jail %d not found", jid);
1987 			goto done_unlock_list;
1988 		}
1989 	} else if (error != ENOENT)
1990 		goto done_unlock_list;
1991 
1992 	error = vfs_getopt(opts, "name", (void **)&name, &len);
1993 	if (error == 0) {
1994 		if (len == 0 || name[len - 1] != '\0') {
1995 			error = EINVAL;
1996 			goto done_unlock_list;
1997 		}
1998 		pr = prison_find_name(mypr, name);
1999 		if (pr != NULL) {
2000 			if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
2001 				mtx_unlock(&pr->pr_mtx);
2002 				error = ENOENT;
2003 				vfs_opterror(opts, "jail \"%s\" is dying",
2004 				    name);
2005 				goto done_unlock_list;
2006 			}
2007 			goto found_prison;
2008 		}
2009 		error = ENOENT;
2010 		vfs_opterror(opts, "jail \"%s\" not found", name);
2011 		goto done_unlock_list;
2012 	} else if (error != ENOENT)
2013 		goto done_unlock_list;
2014 
2015 	vfs_opterror(opts, "no jail specified");
2016 	error = ENOENT;
2017 	goto done_unlock_list;
2018 
2019  found_prison:
2020 	/* Get the parameters of the prison. */
2021 	pr->pr_ref++;
2022 	locked = PD_LOCKED;
2023 	td->td_retval[0] = pr->pr_id;
2024 	error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2025 	if (error != 0 && error != ENOENT)
2026 		goto done_deref;
2027 	i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2028 	error = vfs_setopt(opts, "parent", &i, sizeof(i));
2029 	if (error != 0 && error != ENOENT)
2030 		goto done_deref;
2031 	error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2032 	if (error != 0 && error != ENOENT)
2033 		goto done_deref;
2034 	error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2035 	    sizeof(pr->pr_cpuset->cs_id));
2036 	if (error != 0 && error != ENOENT)
2037 		goto done_deref;
2038 	error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2039 	if (error != 0 && error != ENOENT)
2040 		goto done_deref;
2041 #ifdef INET
2042 	error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
2043 	    pr->pr_ip4s * sizeof(*pr->pr_ip4));
2044 	if (error != 0 && error != ENOENT)
2045 		goto done_deref;
2046 #endif
2047 #ifdef INET6
2048 	error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
2049 	    pr->pr_ip6s * sizeof(*pr->pr_ip6));
2050 	if (error != 0 && error != ENOENT)
2051 		goto done_deref;
2052 #endif
2053 	error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2054 	    sizeof(pr->pr_securelevel));
2055 	if (error != 0 && error != ENOENT)
2056 		goto done_deref;
2057 	error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2058 	    sizeof(pr->pr_childcount));
2059 	if (error != 0 && error != ENOENT)
2060 		goto done_deref;
2061 	error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2062 	    sizeof(pr->pr_childmax));
2063 	if (error != 0 && error != ENOENT)
2064 		goto done_deref;
2065 	error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2066 	if (error != 0 && error != ENOENT)
2067 		goto done_deref;
2068 	error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2069 	if (error != 0 && error != ENOENT)
2070 		goto done_deref;
2071 	error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2072 	if (error != 0 && error != ENOENT)
2073 		goto done_deref;
2074 #ifdef COMPAT_FREEBSD32
2075 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2076 		uint32_t hid32 = pr->pr_hostid;
2077 
2078 		error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2079 	} else
2080 #endif
2081 	error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2082 	    sizeof(pr->pr_hostid));
2083 	if (error != 0 && error != ENOENT)
2084 		goto done_deref;
2085 	error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2086 	    sizeof(pr->pr_enforce_statfs));
2087 	if (error != 0 && error != ENOENT)
2088 		goto done_deref;
2089 	error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2090 	    sizeof(pr->pr_devfs_rsnum));
2091 	if (error != 0 && error != ENOENT)
2092 		goto done_deref;
2093 	for (bf = pr_flag_bool;
2094 	     bf < pr_flag_bool + nitems(pr_flag_bool);
2095 	     bf++) {
2096 		i = (pr->pr_flags & bf->flag) ? 1 : 0;
2097 		error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2098 		if (error != 0 && error != ENOENT)
2099 			goto done_deref;
2100 		i = !i;
2101 		error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2102 		if (error != 0 && error != ENOENT)
2103 			goto done_deref;
2104 	}
2105 	for (jsf = pr_flag_jailsys;
2106 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
2107 	     jsf++) {
2108 		f = pr->pr_flags & (jsf->disable | jsf->new);
2109 		i = (f != 0 && f == jsf->disable) ? JAIL_SYS_DISABLE
2110 		    : (f == jsf->new) ? JAIL_SYS_NEW
2111 		    : JAIL_SYS_INHERIT;
2112 		error = vfs_setopt(opts, jsf->name, &i, sizeof(i));
2113 		if (error != 0 && error != ENOENT)
2114 			goto done_deref;
2115 	}
2116 	for (bf = pr_flag_allow;
2117 	     bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
2118 	     bf++) {
2119 		i = (pr->pr_allow & bf->flag) ? 1 : 0;
2120 		error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2121 		if (error != 0 && error != ENOENT)
2122 			goto done_deref;
2123 		i = !i;
2124 		error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2125 		if (error != 0 && error != ENOENT)
2126 			goto done_deref;
2127 	}
2128 	i = (pr->pr_uref == 0);
2129 	error = vfs_setopt(opts, "dying", &i, sizeof(i));
2130 	if (error != 0 && error != ENOENT)
2131 		goto done_deref;
2132 	i = !i;
2133 	error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2134 	if (error != 0 && error != ENOENT)
2135 		goto done_deref;
2136 	error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2137 	    sizeof(pr->pr_osreldate));
2138 	if (error != 0 && error != ENOENT)
2139 		goto done_deref;
2140 	error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2141 	if (error != 0 && error != ENOENT)
2142 		goto done_deref;
2143 
2144 	/* Get the module parameters. */
2145 	mtx_unlock(&pr->pr_mtx);
2146 	locked = 0;
2147 	error = osd_jail_call(pr, PR_METHOD_GET, opts);
2148 	if (error)
2149 		goto done_deref;
2150 	prison_deref(pr, PD_DEREF | PD_LIST_SLOCKED);
2151 
2152 	/* By now, all parameters should have been noted. */
2153 	TAILQ_FOREACH(opt, opts, link) {
2154 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
2155 			error = EINVAL;
2156 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
2157 			goto done_errmsg;
2158 		}
2159 	}
2160 
2161 	/* Write the fetched parameters back to userspace. */
2162 	error = 0;
2163 	TAILQ_FOREACH(opt, opts, link) {
2164 		if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2165 			pos = 2 * opt->pos + 1;
2166 			optuio->uio_iov[pos].iov_len = opt->len;
2167 			if (opt->value != NULL) {
2168 				if (optuio->uio_segflg == UIO_SYSSPACE) {
2169 					bcopy(opt->value,
2170 					    optuio->uio_iov[pos].iov_base,
2171 					    opt->len);
2172 				} else {
2173 					error = copyout(opt->value,
2174 					    optuio->uio_iov[pos].iov_base,
2175 					    opt->len);
2176 					if (error)
2177 						break;
2178 				}
2179 			}
2180 		}
2181 	}
2182 	goto done_errmsg;
2183 
2184  done_deref:
2185 	prison_deref(pr, locked | PD_DEREF | PD_LIST_SLOCKED);
2186 	goto done_errmsg;
2187 
2188  done_unlock_list:
2189 	sx_sunlock(&allprison_lock);
2190  done_errmsg:
2191 	if (error && errmsg_pos >= 0) {
2192 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2193 		errmsg_pos = 2 * errmsg_pos + 1;
2194 		if (errmsg_len > 0) {
2195 			if (optuio->uio_segflg == UIO_SYSSPACE)
2196 				bcopy(errmsg,
2197 				    optuio->uio_iov[errmsg_pos].iov_base,
2198 				    errmsg_len);
2199 			else
2200 				copyout(errmsg,
2201 				    optuio->uio_iov[errmsg_pos].iov_base,
2202 				    errmsg_len);
2203 		}
2204 	}
2205 	vfs_freeopts(opts);
2206 	return (error);
2207 }
2208 
2209 /*
2210  * struct jail_remove_args {
2211  *	int jid;
2212  * };
2213  */
2214 int
2215 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2216 {
2217 	struct prison *pr, *cpr, *lpr, *tpr;
2218 	int descend, error;
2219 
2220 	error = priv_check(td, PRIV_JAIL_REMOVE);
2221 	if (error)
2222 		return (error);
2223 
2224 	sx_xlock(&allprison_lock);
2225 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2226 	if (pr == NULL) {
2227 		sx_xunlock(&allprison_lock);
2228 		return (EINVAL);
2229 	}
2230 
2231 	/* Remove all descendants of this prison, then remove this prison. */
2232 	pr->pr_ref++;
2233 	if (!LIST_EMPTY(&pr->pr_children)) {
2234 		mtx_unlock(&pr->pr_mtx);
2235 		lpr = NULL;
2236 		FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
2237 			mtx_lock(&cpr->pr_mtx);
2238 			if (cpr->pr_ref > 0) {
2239 				tpr = cpr;
2240 				cpr->pr_ref++;
2241 			} else {
2242 				/* Already removed - do not do it again. */
2243 				tpr = NULL;
2244 			}
2245 			mtx_unlock(&cpr->pr_mtx);
2246 			if (lpr != NULL) {
2247 				mtx_lock(&lpr->pr_mtx);
2248 				prison_remove_one(lpr);
2249 				sx_xlock(&allprison_lock);
2250 			}
2251 			lpr = tpr;
2252 		}
2253 		if (lpr != NULL) {
2254 			mtx_lock(&lpr->pr_mtx);
2255 			prison_remove_one(lpr);
2256 			sx_xlock(&allprison_lock);
2257 		}
2258 		mtx_lock(&pr->pr_mtx);
2259 	}
2260 	prison_remove_one(pr);
2261 	return (0);
2262 }
2263 
2264 static void
2265 prison_remove_one(struct prison *pr)
2266 {
2267 	struct proc *p;
2268 	int deuref;
2269 
2270 	/* If the prison was persistent, it is not anymore. */
2271 	deuref = 0;
2272 	if (pr->pr_flags & PR_PERSIST) {
2273 		pr->pr_ref--;
2274 		deuref = PD_DEUREF;
2275 		pr->pr_flags &= ~PR_PERSIST;
2276 	}
2277 
2278 	/*
2279 	 * jail_remove added a reference.  If that's the only one, remove
2280 	 * the prison now.
2281 	 */
2282 	KASSERT(pr->pr_ref > 0,
2283 	    ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id));
2284 	if (pr->pr_ref == 1) {
2285 		prison_deref(pr,
2286 		    deuref | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
2287 		return;
2288 	}
2289 
2290 	mtx_unlock(&pr->pr_mtx);
2291 	sx_xunlock(&allprison_lock);
2292 	/*
2293 	 * Kill all processes unfortunate enough to be attached to this prison.
2294 	 */
2295 	sx_slock(&allproc_lock);
2296 	FOREACH_PROC_IN_SYSTEM(p) {
2297 		PROC_LOCK(p);
2298 		if (p->p_state != PRS_NEW && p->p_ucred &&
2299 		    p->p_ucred->cr_prison == pr)
2300 			kern_psignal(p, SIGKILL);
2301 		PROC_UNLOCK(p);
2302 	}
2303 	sx_sunlock(&allproc_lock);
2304 	/* Remove the temporary reference added by jail_remove. */
2305 	prison_deref(pr, deuref | PD_DEREF);
2306 }
2307 
2308 /*
2309  * struct jail_attach_args {
2310  *	int jid;
2311  * };
2312  */
2313 int
2314 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2315 {
2316 	struct prison *pr;
2317 	int error;
2318 
2319 	error = priv_check(td, PRIV_JAIL_ATTACH);
2320 	if (error)
2321 		return (error);
2322 
2323 	/*
2324 	 * Start with exclusive hold on allprison_lock to ensure that a possible
2325 	 * PR_METHOD_REMOVE call isn't concurrent with jail_set or jail_remove.
2326 	 * But then immediately downgrade it since we don't need to stop
2327 	 * readers.
2328 	 */
2329 	sx_xlock(&allprison_lock);
2330 	sx_downgrade(&allprison_lock);
2331 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2332 	if (pr == NULL) {
2333 		sx_sunlock(&allprison_lock);
2334 		return (EINVAL);
2335 	}
2336 
2337 	/*
2338 	 * Do not allow a process to attach to a prison that is not
2339 	 * considered to be "alive".
2340 	 */
2341 	if (pr->pr_uref == 0) {
2342 		mtx_unlock(&pr->pr_mtx);
2343 		sx_sunlock(&allprison_lock);
2344 		return (EINVAL);
2345 	}
2346 
2347 	return (do_jail_attach(td, pr));
2348 }
2349 
2350 static int
2351 do_jail_attach(struct thread *td, struct prison *pr)
2352 {
2353 	struct proc *p;
2354 	struct ucred *newcred, *oldcred;
2355 	int error;
2356 
2357 	/*
2358 	 * XXX: Note that there is a slight race here if two threads
2359 	 * in the same privileged process attempt to attach to two
2360 	 * different jails at the same time.  It is important for
2361 	 * user processes not to do this, or they might end up with
2362 	 * a process root from one prison, but attached to the jail
2363 	 * of another.
2364 	 */
2365 	pr->pr_ref++;
2366 	pr->pr_uref++;
2367 	mtx_unlock(&pr->pr_mtx);
2368 
2369 	/* Let modules do whatever they need to prepare for attaching. */
2370 	error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2371 	if (error) {
2372 		prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2373 		return (error);
2374 	}
2375 	sx_sunlock(&allprison_lock);
2376 
2377 	/*
2378 	 * Reparent the newly attached process to this jail.
2379 	 */
2380 	p = td->td_proc;
2381 	error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2382 	if (error)
2383 		goto e_revert_osd;
2384 
2385 	vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2386 	if ((error = change_dir(pr->pr_root, td)) != 0)
2387 		goto e_unlock;
2388 #ifdef MAC
2389 	if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2390 		goto e_unlock;
2391 #endif
2392 	VOP_UNLOCK(pr->pr_root);
2393 	if ((error = pwd_chroot(td, pr->pr_root)))
2394 		goto e_revert_osd;
2395 
2396 	newcred = crget();
2397 	PROC_LOCK(p);
2398 	oldcred = crcopysafe(p, newcred);
2399 	newcred->cr_prison = pr;
2400 	proc_set_cred(p, newcred);
2401 	setsugid(p);
2402 #ifdef RACCT
2403 	racct_proc_ucred_changed(p, oldcred, newcred);
2404 	crhold(newcred);
2405 #endif
2406 	PROC_UNLOCK(p);
2407 #ifdef RCTL
2408 	rctl_proc_ucred_changed(p, newcred);
2409 	crfree(newcred);
2410 #endif
2411 	prison_deref(oldcred->cr_prison, PD_DEREF | PD_DEUREF);
2412 	crfree(oldcred);
2413 	return (0);
2414 
2415  e_unlock:
2416 	VOP_UNLOCK(pr->pr_root);
2417  e_revert_osd:
2418 	/* Tell modules this thread is still in its old jail after all. */
2419 	(void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
2420 	prison_deref(pr, PD_DEREF | PD_DEUREF);
2421 	return (error);
2422 }
2423 
2424 /*
2425  * Returns a locked prison instance, or NULL on failure.
2426  */
2427 struct prison *
2428 prison_find(int prid)
2429 {
2430 	struct prison *pr;
2431 
2432 	sx_assert(&allprison_lock, SX_LOCKED);
2433 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2434 		if (pr->pr_id == prid) {
2435 			mtx_lock(&pr->pr_mtx);
2436 			if (pr->pr_ref > 0)
2437 				return (pr);
2438 			mtx_unlock(&pr->pr_mtx);
2439 		}
2440 	}
2441 	return (NULL);
2442 }
2443 
2444 /*
2445  * Find a prison that is a descendant of mypr.  Returns a locked prison or NULL.
2446  */
2447 struct prison *
2448 prison_find_child(struct prison *mypr, int prid)
2449 {
2450 	struct prison *pr;
2451 	int descend;
2452 
2453 	sx_assert(&allprison_lock, SX_LOCKED);
2454 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2455 		if (pr->pr_id == prid) {
2456 			mtx_lock(&pr->pr_mtx);
2457 			if (pr->pr_ref > 0)
2458 				return (pr);
2459 			mtx_unlock(&pr->pr_mtx);
2460 		}
2461 	}
2462 	return (NULL);
2463 }
2464 
2465 /*
2466  * Look for the name relative to mypr.  Returns a locked prison or NULL.
2467  */
2468 struct prison *
2469 prison_find_name(struct prison *mypr, const char *name)
2470 {
2471 	struct prison *pr, *deadpr;
2472 	size_t mylen;
2473 	int descend;
2474 
2475 	sx_assert(&allprison_lock, SX_LOCKED);
2476 	mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2477  again:
2478 	deadpr = NULL;
2479 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2480 		if (!strcmp(pr->pr_name + mylen, name)) {
2481 			mtx_lock(&pr->pr_mtx);
2482 			if (pr->pr_ref > 0) {
2483 				if (pr->pr_uref > 0)
2484 					return (pr);
2485 				deadpr = pr;
2486 			}
2487 			mtx_unlock(&pr->pr_mtx);
2488 		}
2489 	}
2490 	/* There was no valid prison - perhaps there was a dying one. */
2491 	if (deadpr != NULL) {
2492 		mtx_lock(&deadpr->pr_mtx);
2493 		if (deadpr->pr_ref == 0) {
2494 			mtx_unlock(&deadpr->pr_mtx);
2495 			goto again;
2496 		}
2497 	}
2498 	return (deadpr);
2499 }
2500 
2501 /*
2502  * See if a prison has the specific flag set.
2503  */
2504 int
2505 prison_flag(struct ucred *cred, unsigned flag)
2506 {
2507 
2508 	/* This is an atomic read, so no locking is necessary. */
2509 	return (cred->cr_prison->pr_flags & flag);
2510 }
2511 
2512 int
2513 prison_allow(struct ucred *cred, unsigned flag)
2514 {
2515 
2516 	/* This is an atomic read, so no locking is necessary. */
2517 	return (cred->cr_prison->pr_allow & flag);
2518 }
2519 
2520 /*
2521  * Remove a prison reference.  If that was the last reference, remove the
2522  * prison itself - but not in this context in case there are locks held.
2523  */
2524 void
2525 prison_free_locked(struct prison *pr)
2526 {
2527 	int ref;
2528 
2529 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2530 	ref = --pr->pr_ref;
2531 	mtx_unlock(&pr->pr_mtx);
2532 	if (ref == 0)
2533 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2534 }
2535 
2536 void
2537 prison_free(struct prison *pr)
2538 {
2539 
2540 	mtx_lock(&pr->pr_mtx);
2541 	prison_free_locked(pr);
2542 }
2543 
2544 /*
2545  * Complete a call to either prison_free or prison_proc_free.
2546  */
2547 static void
2548 prison_complete(void *context, int pending)
2549 {
2550 	struct prison *pr = context;
2551 
2552 	sx_xlock(&allprison_lock);
2553 	mtx_lock(&pr->pr_mtx);
2554 	prison_deref(pr, pr->pr_uref
2555 	    ? PD_DEREF | PD_DEUREF | PD_LOCKED | PD_LIST_XLOCKED
2556 	    : PD_LOCKED | PD_LIST_XLOCKED);
2557 }
2558 
2559 /*
2560  * Remove a prison reference (usually).  This internal version assumes no
2561  * mutexes are held, except perhaps the prison itself.  If there are no more
2562  * references, release and delist the prison.  On completion, the prison lock
2563  * and the allprison lock are both unlocked.
2564  */
2565 static void
2566 prison_deref(struct prison *pr, int flags)
2567 {
2568 	struct prison *ppr, *tpr;
2569 	int ref, lasturef;
2570 
2571 	if (!(flags & PD_LOCKED))
2572 		mtx_lock(&pr->pr_mtx);
2573 	for (;;) {
2574 		if (flags & PD_DEUREF) {
2575 			KASSERT(pr->pr_uref > 0,
2576 			    ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
2577 			     pr->pr_id));
2578 			pr->pr_uref--;
2579 			lasturef = pr->pr_uref == 0;
2580 			if (lasturef)
2581 				pr->pr_ref++;
2582 			KASSERT(prison0.pr_uref != 0, ("prison0 pr_uref=0"));
2583 		} else
2584 			lasturef = 0;
2585 		if (flags & PD_DEREF) {
2586 			KASSERT(pr->pr_ref > 0,
2587 			    ("prison_deref PD_DEREF on a dead prison (jid=%d)",
2588 			     pr->pr_id));
2589 			pr->pr_ref--;
2590 		}
2591 		ref = pr->pr_ref;
2592 		mtx_unlock(&pr->pr_mtx);
2593 
2594 		/*
2595 		 * Tell the modules if the last user reference was removed
2596 		 * (even it sticks around in dying state).
2597 		 */
2598 		if (lasturef) {
2599 			if (!(flags & (PD_LIST_SLOCKED | PD_LIST_XLOCKED))) {
2600 				sx_xlock(&allprison_lock);
2601 				flags |= PD_LIST_XLOCKED;
2602 			}
2603 			(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
2604 			mtx_lock(&pr->pr_mtx);
2605 			ref = --pr->pr_ref;
2606 			mtx_unlock(&pr->pr_mtx);
2607 		}
2608 
2609 		/* If the prison still has references, nothing else to do. */
2610 		if (ref > 0) {
2611 			if (flags & PD_LIST_SLOCKED)
2612 				sx_sunlock(&allprison_lock);
2613 			else if (flags & PD_LIST_XLOCKED)
2614 				sx_xunlock(&allprison_lock);
2615 			return;
2616 		}
2617 
2618 		if (flags & PD_LIST_SLOCKED) {
2619 			if (!sx_try_upgrade(&allprison_lock)) {
2620 				sx_sunlock(&allprison_lock);
2621 				sx_xlock(&allprison_lock);
2622 			}
2623 		} else if (!(flags & PD_LIST_XLOCKED))
2624 			sx_xlock(&allprison_lock);
2625 
2626 		TAILQ_REMOVE(&allprison, pr, pr_list);
2627 		LIST_REMOVE(pr, pr_sibling);
2628 		ppr = pr->pr_parent;
2629 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
2630 			tpr->pr_childcount--;
2631 		sx_xunlock(&allprison_lock);
2632 
2633 #ifdef VIMAGE
2634 		if (pr->pr_vnet != ppr->pr_vnet)
2635 			vnet_destroy(pr->pr_vnet);
2636 #endif
2637 		if (pr->pr_root != NULL)
2638 			vrele(pr->pr_root);
2639 		mtx_destroy(&pr->pr_mtx);
2640 #ifdef INET
2641 		free(pr->pr_ip4, M_PRISON);
2642 #endif
2643 #ifdef INET6
2644 		free(pr->pr_ip6, M_PRISON);
2645 #endif
2646 		if (pr->pr_cpuset != NULL)
2647 			cpuset_rel(pr->pr_cpuset);
2648 		osd_jail_exit(pr);
2649 #ifdef RACCT
2650 		if (racct_enable)
2651 			prison_racct_detach(pr);
2652 #endif
2653 		free(pr, M_PRISON);
2654 
2655 		/* Removing a prison frees a reference on its parent. */
2656 		pr = ppr;
2657 		mtx_lock(&pr->pr_mtx);
2658 		flags = PD_DEREF | PD_DEUREF;
2659 	}
2660 }
2661 
2662 void
2663 prison_hold_locked(struct prison *pr)
2664 {
2665 
2666 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2667 	KASSERT(pr->pr_ref > 0,
2668 	    ("Trying to hold dead prison %p (jid=%d).", pr, pr->pr_id));
2669 	pr->pr_ref++;
2670 }
2671 
2672 void
2673 prison_hold(struct prison *pr)
2674 {
2675 
2676 	mtx_lock(&pr->pr_mtx);
2677 	prison_hold_locked(pr);
2678 	mtx_unlock(&pr->pr_mtx);
2679 }
2680 
2681 void
2682 prison_proc_hold(struct prison *pr)
2683 {
2684 
2685 	mtx_lock(&pr->pr_mtx);
2686 	KASSERT(pr->pr_uref > 0,
2687 	    ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2688 	pr->pr_uref++;
2689 	mtx_unlock(&pr->pr_mtx);
2690 }
2691 
2692 void
2693 prison_proc_free(struct prison *pr)
2694 {
2695 
2696 	mtx_lock(&pr->pr_mtx);
2697 	KASSERT(pr->pr_uref > 0,
2698 	    ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2699 	if (pr->pr_uref > 1)
2700 		pr->pr_uref--;
2701 	else {
2702 		/*
2703 		 * Don't remove the last user reference in this context, which
2704 		 * is expected to be a process that is not only locked, but
2705 		 * also half dead.
2706 		 */
2707 		pr->pr_ref++;
2708 		mtx_unlock(&pr->pr_mtx);
2709 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2710 		return;
2711 	}
2712 	mtx_unlock(&pr->pr_mtx);
2713 }
2714 
2715 /*
2716  * Check if a jail supports the given address family.
2717  *
2718  * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
2719  * if not.
2720  */
2721 int
2722 prison_check_af(struct ucred *cred, int af)
2723 {
2724 	struct prison *pr;
2725 	int error;
2726 
2727 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2728 
2729 	pr = cred->cr_prison;
2730 #ifdef VIMAGE
2731 	/* Prisons with their own network stack are not limited. */
2732 	if (prison_owns_vnet(cred))
2733 		return (0);
2734 #endif
2735 
2736 	error = 0;
2737 	switch (af)
2738 	{
2739 #ifdef INET
2740 	case AF_INET:
2741 		if (pr->pr_flags & PR_IP4)
2742 		{
2743 			mtx_lock(&pr->pr_mtx);
2744 			if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL)
2745 				error = EAFNOSUPPORT;
2746 			mtx_unlock(&pr->pr_mtx);
2747 		}
2748 		break;
2749 #endif
2750 #ifdef INET6
2751 	case AF_INET6:
2752 		if (pr->pr_flags & PR_IP6)
2753 		{
2754 			mtx_lock(&pr->pr_mtx);
2755 			if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL)
2756 				error = EAFNOSUPPORT;
2757 			mtx_unlock(&pr->pr_mtx);
2758 		}
2759 		break;
2760 #endif
2761 	case AF_LOCAL:
2762 	case AF_ROUTE:
2763 		break;
2764 	default:
2765 		if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
2766 			error = EAFNOSUPPORT;
2767 	}
2768 	return (error);
2769 }
2770 
2771 /*
2772  * Check if given address belongs to the jail referenced by cred (wrapper to
2773  * prison_check_ip[46]).
2774  *
2775  * Returns 0 if jail doesn't restrict the address family or if address belongs
2776  * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
2777  * the jail doesn't allow the address family.  IPv4 Address passed in in NBO.
2778  */
2779 int
2780 prison_if(struct ucred *cred, const struct sockaddr *sa)
2781 {
2782 #ifdef INET
2783 	const struct sockaddr_in *sai;
2784 #endif
2785 #ifdef INET6
2786 	const struct sockaddr_in6 *sai6;
2787 #endif
2788 	int error;
2789 
2790 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2791 	KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
2792 
2793 #ifdef VIMAGE
2794 	if (prison_owns_vnet(cred))
2795 		return (0);
2796 #endif
2797 
2798 	error = 0;
2799 	switch (sa->sa_family)
2800 	{
2801 #ifdef INET
2802 	case AF_INET:
2803 		sai = (const struct sockaddr_in *)sa;
2804 		error = prison_check_ip4(cred, &sai->sin_addr);
2805 		break;
2806 #endif
2807 #ifdef INET6
2808 	case AF_INET6:
2809 		sai6 = (const struct sockaddr_in6 *)sa;
2810 		error = prison_check_ip6(cred, &sai6->sin6_addr);
2811 		break;
2812 #endif
2813 	default:
2814 		if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
2815 			error = EAFNOSUPPORT;
2816 	}
2817 	return (error);
2818 }
2819 
2820 /*
2821  * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
2822  */
2823 int
2824 prison_check(struct ucred *cred1, struct ucred *cred2)
2825 {
2826 
2827 	return ((cred1->cr_prison == cred2->cr_prison ||
2828 	    prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
2829 }
2830 
2831 /*
2832  * Return 1 if p2 is a child of p1, otherwise 0.
2833  */
2834 int
2835 prison_ischild(struct prison *pr1, struct prison *pr2)
2836 {
2837 
2838 	for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
2839 		if (pr1 == pr2)
2840 			return (1);
2841 	return (0);
2842 }
2843 
2844 /*
2845  * Return 1 if the passed credential is in a jail, otherwise 0.
2846  */
2847 int
2848 jailed(struct ucred *cred)
2849 {
2850 
2851 	return (cred->cr_prison != &prison0);
2852 }
2853 
2854 /*
2855  * Return 1 if the passed credential is in a jail and that jail does not
2856  * have its own virtual network stack, otherwise 0.
2857  */
2858 int
2859 jailed_without_vnet(struct ucred *cred)
2860 {
2861 
2862 	if (!jailed(cred))
2863 		return (0);
2864 #ifdef VIMAGE
2865 	if (prison_owns_vnet(cred))
2866 		return (0);
2867 #endif
2868 
2869 	return (1);
2870 }
2871 
2872 /*
2873  * Return the correct hostname (domainname, et al) for the passed credential.
2874  */
2875 void
2876 getcredhostname(struct ucred *cred, char *buf, size_t size)
2877 {
2878 	struct prison *pr;
2879 
2880 	/*
2881 	 * A NULL credential can be used to shortcut to the physical
2882 	 * system's hostname.
2883 	 */
2884 	pr = (cred != NULL) ? cred->cr_prison : &prison0;
2885 	mtx_lock(&pr->pr_mtx);
2886 	strlcpy(buf, pr->pr_hostname, size);
2887 	mtx_unlock(&pr->pr_mtx);
2888 }
2889 
2890 void
2891 getcreddomainname(struct ucred *cred, char *buf, size_t size)
2892 {
2893 
2894 	mtx_lock(&cred->cr_prison->pr_mtx);
2895 	strlcpy(buf, cred->cr_prison->pr_domainname, size);
2896 	mtx_unlock(&cred->cr_prison->pr_mtx);
2897 }
2898 
2899 void
2900 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
2901 {
2902 
2903 	mtx_lock(&cred->cr_prison->pr_mtx);
2904 	strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
2905 	mtx_unlock(&cred->cr_prison->pr_mtx);
2906 }
2907 
2908 void
2909 getcredhostid(struct ucred *cred, unsigned long *hostid)
2910 {
2911 
2912 	mtx_lock(&cred->cr_prison->pr_mtx);
2913 	*hostid = cred->cr_prison->pr_hostid;
2914 	mtx_unlock(&cred->cr_prison->pr_mtx);
2915 }
2916 
2917 #ifdef VIMAGE
2918 /*
2919  * Determine whether the prison represented by cred owns
2920  * its vnet rather than having it inherited.
2921  *
2922  * Returns 1 in case the prison owns the vnet, 0 otherwise.
2923  */
2924 int
2925 prison_owns_vnet(struct ucred *cred)
2926 {
2927 
2928 	/*
2929 	 * vnets cannot be added/removed after jail creation,
2930 	 * so no need to lock here.
2931 	 */
2932 	return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
2933 }
2934 #endif
2935 
2936 /*
2937  * Determine whether the subject represented by cred can "see"
2938  * status of a mount point.
2939  * Returns: 0 for permitted, ENOENT otherwise.
2940  * XXX: This function should be called cr_canseemount() and should be
2941  *      placed in kern_prot.c.
2942  */
2943 int
2944 prison_canseemount(struct ucred *cred, struct mount *mp)
2945 {
2946 	struct prison *pr;
2947 	struct statfs *sp;
2948 	size_t len;
2949 
2950 	pr = cred->cr_prison;
2951 	if (pr->pr_enforce_statfs == 0)
2952 		return (0);
2953 	if (pr->pr_root->v_mount == mp)
2954 		return (0);
2955 	if (pr->pr_enforce_statfs == 2)
2956 		return (ENOENT);
2957 	/*
2958 	 * If jail's chroot directory is set to "/" we should be able to see
2959 	 * all mount-points from inside a jail.
2960 	 * This is ugly check, but this is the only situation when jail's
2961 	 * directory ends with '/'.
2962 	 */
2963 	if (strcmp(pr->pr_path, "/") == 0)
2964 		return (0);
2965 	len = strlen(pr->pr_path);
2966 	sp = &mp->mnt_stat;
2967 	if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
2968 		return (ENOENT);
2969 	/*
2970 	 * Be sure that we don't have situation where jail's root directory
2971 	 * is "/some/path" and mount point is "/some/pathpath".
2972 	 */
2973 	if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
2974 		return (ENOENT);
2975 	return (0);
2976 }
2977 
2978 void
2979 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
2980 {
2981 	char jpath[MAXPATHLEN];
2982 	struct prison *pr;
2983 	size_t len;
2984 
2985 	pr = cred->cr_prison;
2986 	if (pr->pr_enforce_statfs == 0)
2987 		return;
2988 	if (prison_canseemount(cred, mp) != 0) {
2989 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
2990 		strlcpy(sp->f_mntonname, "[restricted]",
2991 		    sizeof(sp->f_mntonname));
2992 		return;
2993 	}
2994 	if (pr->pr_root->v_mount == mp) {
2995 		/*
2996 		 * Clear current buffer data, so we are sure nothing from
2997 		 * the valid path left there.
2998 		 */
2999 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3000 		*sp->f_mntonname = '/';
3001 		return;
3002 	}
3003 	/*
3004 	 * If jail's chroot directory is set to "/" we should be able to see
3005 	 * all mount-points from inside a jail.
3006 	 */
3007 	if (strcmp(pr->pr_path, "/") == 0)
3008 		return;
3009 	len = strlen(pr->pr_path);
3010 	strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3011 	/*
3012 	 * Clear current buffer data, so we are sure nothing from
3013 	 * the valid path left there.
3014 	 */
3015 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3016 	if (*jpath == '\0') {
3017 		/* Should never happen. */
3018 		*sp->f_mntonname = '/';
3019 	} else {
3020 		strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3021 	}
3022 }
3023 
3024 /*
3025  * Check with permission for a specific privilege is granted within jail.  We
3026  * have a specific list of accepted privileges; the rest are denied.
3027  */
3028 int
3029 prison_priv_check(struct ucred *cred, int priv)
3030 {
3031 
3032 	if (!jailed(cred))
3033 		return (0);
3034 
3035 #ifdef VIMAGE
3036 	/*
3037 	 * Privileges specific to prisons with a virtual network stack.
3038 	 * There might be a duplicate entry here in case the privilege
3039 	 * is only granted conditionally in the legacy jail case.
3040 	 */
3041 	switch (priv) {
3042 #ifdef notyet
3043 		/*
3044 		 * NFS-specific privileges.
3045 		 */
3046 	case PRIV_NFS_DAEMON:
3047 	case PRIV_NFS_LOCKD:
3048 #endif
3049 		/*
3050 		 * Network stack privileges.
3051 		 */
3052 	case PRIV_NET_BRIDGE:
3053 	case PRIV_NET_GRE:
3054 	case PRIV_NET_BPF:
3055 	case PRIV_NET_RAW:		/* Dup, cond. in legacy jail case. */
3056 	case PRIV_NET_ROUTE:
3057 	case PRIV_NET_TAP:
3058 	case PRIV_NET_SETIFMTU:
3059 	case PRIV_NET_SETIFFLAGS:
3060 	case PRIV_NET_SETIFCAP:
3061 	case PRIV_NET_SETIFDESCR:
3062 	case PRIV_NET_SETIFNAME	:
3063 	case PRIV_NET_SETIFMETRIC:
3064 	case PRIV_NET_SETIFPHYS:
3065 	case PRIV_NET_SETIFMAC:
3066 	case PRIV_NET_SETLANPCP:
3067 	case PRIV_NET_ADDMULTI:
3068 	case PRIV_NET_DELMULTI:
3069 	case PRIV_NET_HWIOCTL:
3070 	case PRIV_NET_SETLLADDR:
3071 	case PRIV_NET_ADDIFGROUP:
3072 	case PRIV_NET_DELIFGROUP:
3073 	case PRIV_NET_IFCREATE:
3074 	case PRIV_NET_IFDESTROY:
3075 	case PRIV_NET_ADDIFADDR:
3076 	case PRIV_NET_DELIFADDR:
3077 	case PRIV_NET_LAGG:
3078 	case PRIV_NET_GIF:
3079 	case PRIV_NET_SETIFVNET:
3080 	case PRIV_NET_SETIFFIB:
3081 
3082 		/*
3083 		 * 802.11-related privileges.
3084 		 */
3085 	case PRIV_NET80211_GETKEY:
3086 #ifdef notyet
3087 	case PRIV_NET80211_MANAGE:		/* XXX-BZ discuss with sam@ */
3088 #endif
3089 
3090 #ifdef notyet
3091 		/*
3092 		 * ATM privileges.
3093 		 */
3094 	case PRIV_NETATM_CFG:
3095 	case PRIV_NETATM_ADD:
3096 	case PRIV_NETATM_DEL:
3097 	case PRIV_NETATM_SET:
3098 
3099 		/*
3100 		 * Bluetooth privileges.
3101 		 */
3102 	case PRIV_NETBLUETOOTH_RAW:
3103 #endif
3104 
3105 		/*
3106 		 * Netgraph and netgraph module privileges.
3107 		 */
3108 	case PRIV_NETGRAPH_CONTROL:
3109 #ifdef notyet
3110 	case PRIV_NETGRAPH_TTY:
3111 #endif
3112 
3113 		/*
3114 		 * IPv4 and IPv6 privileges.
3115 		 */
3116 	case PRIV_NETINET_IPFW:
3117 	case PRIV_NETINET_DIVERT:
3118 	case PRIV_NETINET_PF:
3119 	case PRIV_NETINET_DUMMYNET:
3120 	case PRIV_NETINET_CARP:
3121 	case PRIV_NETINET_MROUTE:
3122 	case PRIV_NETINET_RAW:
3123 	case PRIV_NETINET_ADDRCTRL6:
3124 	case PRIV_NETINET_ND6:
3125 	case PRIV_NETINET_SCOPE6:
3126 	case PRIV_NETINET_ALIFETIME6:
3127 	case PRIV_NETINET_IPSEC:
3128 	case PRIV_NETINET_BINDANY:
3129 
3130 #ifdef notyet
3131 		/*
3132 		 * NCP privileges.
3133 		 */
3134 	case PRIV_NETNCP:
3135 
3136 		/*
3137 		 * SMB privileges.
3138 		 */
3139 	case PRIV_NETSMB:
3140 #endif
3141 
3142 	/*
3143 	 * No default: or deny here.
3144 	 * In case of no permit fall through to next switch().
3145 	 */
3146 		if (cred->cr_prison->pr_flags & PR_VNET)
3147 			return (0);
3148 	}
3149 #endif /* VIMAGE */
3150 
3151 	switch (priv) {
3152 
3153 		/*
3154 		 * Allow ktrace privileges for root in jail.
3155 		 */
3156 	case PRIV_KTRACE:
3157 
3158 #if 0
3159 		/*
3160 		 * Allow jailed processes to configure audit identity and
3161 		 * submit audit records (login, etc).  In the future we may
3162 		 * want to further refine the relationship between audit and
3163 		 * jail.
3164 		 */
3165 	case PRIV_AUDIT_GETAUDIT:
3166 	case PRIV_AUDIT_SETAUDIT:
3167 	case PRIV_AUDIT_SUBMIT:
3168 #endif
3169 
3170 		/*
3171 		 * Allow jailed processes to manipulate process UNIX
3172 		 * credentials in any way they see fit.
3173 		 */
3174 	case PRIV_CRED_SETUID:
3175 	case PRIV_CRED_SETEUID:
3176 	case PRIV_CRED_SETGID:
3177 	case PRIV_CRED_SETEGID:
3178 	case PRIV_CRED_SETGROUPS:
3179 	case PRIV_CRED_SETREUID:
3180 	case PRIV_CRED_SETREGID:
3181 	case PRIV_CRED_SETRESUID:
3182 	case PRIV_CRED_SETRESGID:
3183 
3184 		/*
3185 		 * Jail implements visibility constraints already, so allow
3186 		 * jailed root to override uid/gid-based constraints.
3187 		 */
3188 	case PRIV_SEEOTHERGIDS:
3189 	case PRIV_SEEOTHERUIDS:
3190 
3191 		/*
3192 		 * Jail implements inter-process debugging limits already, so
3193 		 * allow jailed root various debugging privileges.
3194 		 */
3195 	case PRIV_DEBUG_DIFFCRED:
3196 	case PRIV_DEBUG_SUGID:
3197 	case PRIV_DEBUG_UNPRIV:
3198 
3199 		/*
3200 		 * Allow jail to set various resource limits and login
3201 		 * properties, and for now, exceed process resource limits.
3202 		 */
3203 	case PRIV_PROC_LIMIT:
3204 	case PRIV_PROC_SETLOGIN:
3205 	case PRIV_PROC_SETRLIMIT:
3206 
3207 		/*
3208 		 * System V and POSIX IPC privileges are granted in jail.
3209 		 */
3210 	case PRIV_IPC_READ:
3211 	case PRIV_IPC_WRITE:
3212 	case PRIV_IPC_ADMIN:
3213 	case PRIV_IPC_MSGSIZE:
3214 	case PRIV_MQ_ADMIN:
3215 
3216 		/*
3217 		 * Jail operations within a jail work on child jails.
3218 		 */
3219 	case PRIV_JAIL_ATTACH:
3220 	case PRIV_JAIL_SET:
3221 	case PRIV_JAIL_REMOVE:
3222 
3223 		/*
3224 		 * Jail implements its own inter-process limits, so allow
3225 		 * root processes in jail to change scheduling on other
3226 		 * processes in the same jail.  Likewise for signalling.
3227 		 */
3228 	case PRIV_SCHED_DIFFCRED:
3229 	case PRIV_SCHED_CPUSET:
3230 	case PRIV_SIGNAL_DIFFCRED:
3231 	case PRIV_SIGNAL_SUGID:
3232 
3233 		/*
3234 		 * Allow jailed processes to write to sysctls marked as jail
3235 		 * writable.
3236 		 */
3237 	case PRIV_SYSCTL_WRITEJAIL:
3238 
3239 		/*
3240 		 * Allow root in jail to manage a variety of quota
3241 		 * properties.  These should likely be conditional on a
3242 		 * configuration option.
3243 		 */
3244 	case PRIV_VFS_GETQUOTA:
3245 	case PRIV_VFS_SETQUOTA:
3246 
3247 		/*
3248 		 * Since Jail relies on chroot() to implement file system
3249 		 * protections, grant many VFS privileges to root in jail.
3250 		 * Be careful to exclude mount-related and NFS-related
3251 		 * privileges.
3252 		 */
3253 	case PRIV_VFS_READ:
3254 	case PRIV_VFS_WRITE:
3255 	case PRIV_VFS_ADMIN:
3256 	case PRIV_VFS_EXEC:
3257 	case PRIV_VFS_LOOKUP:
3258 	case PRIV_VFS_BLOCKRESERVE:	/* XXXRW: Slightly surprising. */
3259 	case PRIV_VFS_CHFLAGS_DEV:
3260 	case PRIV_VFS_CHOWN:
3261 	case PRIV_VFS_CHROOT:
3262 	case PRIV_VFS_RETAINSUGID:
3263 	case PRIV_VFS_FCHROOT:
3264 	case PRIV_VFS_LINK:
3265 	case PRIV_VFS_SETGID:
3266 	case PRIV_VFS_STAT:
3267 	case PRIV_VFS_STICKYFILE:
3268 
3269 		/*
3270 		 * As in the non-jail case, non-root users are expected to be
3271 		 * able to read kernel/phyiscal memory (provided /dev/[k]mem
3272 		 * exists in the jail and they have permission to access it).
3273 		 */
3274 	case PRIV_KMEM_READ:
3275 		return (0);
3276 
3277 		/*
3278 		 * Depending on the global setting, allow privilege of
3279 		 * setting system flags.
3280 		 */
3281 	case PRIV_VFS_SYSFLAGS:
3282 		if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
3283 			return (0);
3284 		else
3285 			return (EPERM);
3286 
3287 		/*
3288 		 * Depending on the global setting, allow privilege of
3289 		 * mounting/unmounting file systems.
3290 		 */
3291 	case PRIV_VFS_MOUNT:
3292 	case PRIV_VFS_UNMOUNT:
3293 	case PRIV_VFS_MOUNT_NONUSER:
3294 	case PRIV_VFS_MOUNT_OWNER:
3295 		if (cred->cr_prison->pr_allow & PR_ALLOW_MOUNT &&
3296 		    cred->cr_prison->pr_enforce_statfs < 2)
3297 			return (0);
3298 		else
3299 			return (EPERM);
3300 
3301 		/*
3302 		 * Conditionnaly allow locking (unlocking) physical pages
3303 		 * in memory.
3304 		 */
3305 	case PRIV_VM_MLOCK:
3306 	case PRIV_VM_MUNLOCK:
3307 		if (cred->cr_prison->pr_allow & PR_ALLOW_MLOCK)
3308 			return (0);
3309 		else
3310 			return (EPERM);
3311 
3312 		/*
3313 		 * Conditionally allow jailed root to bind reserved ports.
3314 		 */
3315 	case PRIV_NETINET_RESERVEDPORT:
3316 		if (cred->cr_prison->pr_allow & PR_ALLOW_RESERVED_PORTS)
3317 			return (0);
3318 		else
3319 			return (EPERM);
3320 
3321 		/*
3322 		 * Allow jailed root to reuse in-use ports.
3323 		 */
3324 	case PRIV_NETINET_REUSEPORT:
3325 		return (0);
3326 
3327 		/*
3328 		 * Allow jailed root to set certain IPv4/6 (option) headers.
3329 		 */
3330 	case PRIV_NETINET_SETHDROPTS:
3331 		return (0);
3332 
3333 		/*
3334 		 * Conditionally allow creating raw sockets in jail.
3335 		 */
3336 	case PRIV_NETINET_RAW:
3337 		if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
3338 			return (0);
3339 		else
3340 			return (EPERM);
3341 
3342 		/*
3343 		 * Since jail implements its own visibility limits on netstat
3344 		 * sysctls, allow getcred.  This allows identd to work in
3345 		 * jail.
3346 		 */
3347 	case PRIV_NETINET_GETCRED:
3348 		return (0);
3349 
3350 		/*
3351 		 * Allow jailed root to set loginclass.
3352 		 */
3353 	case PRIV_PROC_SETLOGINCLASS:
3354 		return (0);
3355 
3356 		/*
3357 		 * Do not allow a process inside a jail to read the kernel
3358 		 * message buffer unless explicitly permitted.
3359 		 */
3360 	case PRIV_MSGBUF:
3361 		if (cred->cr_prison->pr_allow & PR_ALLOW_READ_MSGBUF)
3362 			return (0);
3363 		return (EPERM);
3364 
3365 	default:
3366 		/*
3367 		 * In all remaining cases, deny the privilege request.  This
3368 		 * includes almost all network privileges, many system
3369 		 * configuration privileges.
3370 		 */
3371 		return (EPERM);
3372 	}
3373 }
3374 
3375 /*
3376  * Return the part of pr2's name that is relative to pr1, or the whole name
3377  * if it does not directly follow.
3378  */
3379 
3380 char *
3381 prison_name(struct prison *pr1, struct prison *pr2)
3382 {
3383 	char *name;
3384 
3385 	/* Jails see themselves as "0" (if they see themselves at all). */
3386 	if (pr1 == pr2)
3387 		return "0";
3388 	name = pr2->pr_name;
3389 	if (prison_ischild(pr1, pr2)) {
3390 		/*
3391 		 * pr1 isn't locked (and allprison_lock may not be either)
3392 		 * so its length can't be counted on.  But the number of dots
3393 		 * can be counted on - and counted.
3394 		 */
3395 		for (; pr1 != &prison0; pr1 = pr1->pr_parent)
3396 			name = strchr(name, '.') + 1;
3397 	}
3398 	return (name);
3399 }
3400 
3401 /*
3402  * Return the part of pr2's path that is relative to pr1, or the whole path
3403  * if it does not directly follow.
3404  */
3405 static char *
3406 prison_path(struct prison *pr1, struct prison *pr2)
3407 {
3408 	char *path1, *path2;
3409 	int len1;
3410 
3411 	path1 = pr1->pr_path;
3412 	path2 = pr2->pr_path;
3413 	if (!strcmp(path1, "/"))
3414 		return (path2);
3415 	len1 = strlen(path1);
3416 	if (strncmp(path1, path2, len1))
3417 		return (path2);
3418 	if (path2[len1] == '\0')
3419 		return "/";
3420 	if (path2[len1] == '/')
3421 		return (path2 + len1);
3422 	return (path2);
3423 }
3424 
3425 /*
3426  * Jail-related sysctls.
3427  */
3428 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW, 0,
3429     "Jails");
3430 
3431 static int
3432 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
3433 {
3434 	struct xprison *xp;
3435 	struct prison *pr, *cpr;
3436 #ifdef INET
3437 	struct in_addr *ip4 = NULL;
3438 	int ip4s = 0;
3439 #endif
3440 #ifdef INET6
3441 	struct in6_addr *ip6 = NULL;
3442 	int ip6s = 0;
3443 #endif
3444 	int descend, error;
3445 
3446 	xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
3447 	pr = req->td->td_ucred->cr_prison;
3448 	error = 0;
3449 	sx_slock(&allprison_lock);
3450 	FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
3451 #if defined(INET) || defined(INET6)
3452  again:
3453 #endif
3454 		mtx_lock(&cpr->pr_mtx);
3455 #ifdef INET
3456 		if (cpr->pr_ip4s > 0) {
3457 			if (ip4s < cpr->pr_ip4s) {
3458 				ip4s = cpr->pr_ip4s;
3459 				mtx_unlock(&cpr->pr_mtx);
3460 				ip4 = realloc(ip4, ip4s *
3461 				    sizeof(struct in_addr), M_TEMP, M_WAITOK);
3462 				goto again;
3463 			}
3464 			bcopy(cpr->pr_ip4, ip4,
3465 			    cpr->pr_ip4s * sizeof(struct in_addr));
3466 		}
3467 #endif
3468 #ifdef INET6
3469 		if (cpr->pr_ip6s > 0) {
3470 			if (ip6s < cpr->pr_ip6s) {
3471 				ip6s = cpr->pr_ip6s;
3472 				mtx_unlock(&cpr->pr_mtx);
3473 				ip6 = realloc(ip6, ip6s *
3474 				    sizeof(struct in6_addr), M_TEMP, M_WAITOK);
3475 				goto again;
3476 			}
3477 			bcopy(cpr->pr_ip6, ip6,
3478 			    cpr->pr_ip6s * sizeof(struct in6_addr));
3479 		}
3480 #endif
3481 		if (cpr->pr_ref == 0) {
3482 			mtx_unlock(&cpr->pr_mtx);
3483 			continue;
3484 		}
3485 		bzero(xp, sizeof(*xp));
3486 		xp->pr_version = XPRISON_VERSION;
3487 		xp->pr_id = cpr->pr_id;
3488 		xp->pr_state = cpr->pr_uref > 0
3489 		    ? PRISON_STATE_ALIVE : PRISON_STATE_DYING;
3490 		strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
3491 		strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
3492 		strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
3493 #ifdef INET
3494 		xp->pr_ip4s = cpr->pr_ip4s;
3495 #endif
3496 #ifdef INET6
3497 		xp->pr_ip6s = cpr->pr_ip6s;
3498 #endif
3499 		mtx_unlock(&cpr->pr_mtx);
3500 		error = SYSCTL_OUT(req, xp, sizeof(*xp));
3501 		if (error)
3502 			break;
3503 #ifdef INET
3504 		if (xp->pr_ip4s > 0) {
3505 			error = SYSCTL_OUT(req, ip4,
3506 			    xp->pr_ip4s * sizeof(struct in_addr));
3507 			if (error)
3508 				break;
3509 		}
3510 #endif
3511 #ifdef INET6
3512 		if (xp->pr_ip6s > 0) {
3513 			error = SYSCTL_OUT(req, ip6,
3514 			    xp->pr_ip6s * sizeof(struct in6_addr));
3515 			if (error)
3516 				break;
3517 		}
3518 #endif
3519 	}
3520 	sx_sunlock(&allprison_lock);
3521 	free(xp, M_TEMP);
3522 #ifdef INET
3523 	free(ip4, M_TEMP);
3524 #endif
3525 #ifdef INET6
3526 	free(ip6, M_TEMP);
3527 #endif
3528 	return (error);
3529 }
3530 
3531 SYSCTL_OID(_security_jail, OID_AUTO, list,
3532     CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3533     sysctl_jail_list, "S", "List of active jails");
3534 
3535 static int
3536 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
3537 {
3538 	int error, injail;
3539 
3540 	injail = jailed(req->td->td_ucred);
3541 	error = SYSCTL_OUT(req, &injail, sizeof(injail));
3542 
3543 	return (error);
3544 }
3545 
3546 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
3547     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3548     sysctl_jail_jailed, "I", "Process in jail?");
3549 
3550 static int
3551 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
3552 {
3553 	int error, havevnet;
3554 #ifdef VIMAGE
3555 	struct ucred *cred = req->td->td_ucred;
3556 
3557 	havevnet = jailed(cred) && prison_owns_vnet(cred);
3558 #else
3559 	havevnet = 0;
3560 #endif
3561 	error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
3562 
3563 	return (error);
3564 }
3565 
3566 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
3567     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3568     sysctl_jail_vnet, "I", "Jail owns vnet?");
3569 
3570 #if defined(INET) || defined(INET6)
3571 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
3572     &jail_max_af_ips, 0,
3573     "Number of IP addresses a jail may have at most per address family (deprecated)");
3574 #endif
3575 
3576 /*
3577  * Default parameters for jail(2) compatibility.  For historical reasons,
3578  * the sysctl names have varying similarity to the parameter names.  Prisons
3579  * just see their own parameters, and can't change them.
3580  */
3581 static int
3582 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
3583 {
3584 	struct prison *pr;
3585 	int allow, error, i;
3586 
3587 	pr = req->td->td_ucred->cr_prison;
3588 	allow = (pr == &prison0) ? jail_default_allow : pr->pr_allow;
3589 
3590 	/* Get the current flag value, and convert it to a boolean. */
3591 	i = (allow & arg2) ? 1 : 0;
3592 	if (arg1 != NULL)
3593 		i = !i;
3594 	error = sysctl_handle_int(oidp, &i, 0, req);
3595 	if (error || !req->newptr)
3596 		return (error);
3597 	i = i ? arg2 : 0;
3598 	if (arg1 != NULL)
3599 		i ^= arg2;
3600 	/*
3601 	 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
3602 	 * for writing.
3603 	 */
3604 	mtx_lock(&prison0.pr_mtx);
3605 	jail_default_allow = (jail_default_allow & ~arg2) | i;
3606 	mtx_unlock(&prison0.pr_mtx);
3607 	return (0);
3608 }
3609 
3610 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
3611     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3612     NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
3613     "Processes in jail can set their hostnames (deprecated)");
3614 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
3615     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3616     (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
3617     "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
3618 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
3619     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3620     NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
3621     "Processes in jail can use System V IPC primitives (deprecated)");
3622 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
3623     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3624     NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
3625     "Prison root can create raw sockets (deprecated)");
3626 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
3627     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3628     NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
3629     "Processes in jail can alter system file flags (deprecated)");
3630 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
3631     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3632     NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
3633     "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
3634 
3635 static int
3636 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
3637 {
3638 	struct prison *pr;
3639 	int level, error;
3640 
3641 	pr = req->td->td_ucred->cr_prison;
3642 	level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
3643 	error = sysctl_handle_int(oidp, &level, 0, req);
3644 	if (error || !req->newptr)
3645 		return (error);
3646 	*(int *)arg1 = level;
3647 	return (0);
3648 }
3649 
3650 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
3651     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3652     &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
3653     sysctl_jail_default_level, "I",
3654     "Processes in jail cannot see all mounted file systems (deprecated)");
3655 
3656 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
3657     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
3658     &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
3659     sysctl_jail_default_level, "I",
3660     "Ruleset for the devfs filesystem in jail (deprecated)");
3661 
3662 /*
3663  * Nodes to describe jail parameters.  Maximum length of string parameters
3664  * is returned in the string itself, and the other parameters exist merely
3665  * to make themselves and their types known.
3666  */
3667 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW, 0,
3668     "Jail parameters");
3669 
3670 int
3671 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
3672 {
3673 	int i;
3674 	long l;
3675 	size_t s;
3676 	char numbuf[12];
3677 
3678 	switch (oidp->oid_kind & CTLTYPE)
3679 	{
3680 	case CTLTYPE_LONG:
3681 	case CTLTYPE_ULONG:
3682 		l = 0;
3683 #ifdef SCTL_MASK32
3684 		if (!(req->flags & SCTL_MASK32))
3685 #endif
3686 			return (SYSCTL_OUT(req, &l, sizeof(l)));
3687 	case CTLTYPE_INT:
3688 	case CTLTYPE_UINT:
3689 		i = 0;
3690 		return (SYSCTL_OUT(req, &i, sizeof(i)));
3691 	case CTLTYPE_STRING:
3692 		snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
3693 		return
3694 		    (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
3695 	case CTLTYPE_STRUCT:
3696 		s = (size_t)arg2;
3697 		return (SYSCTL_OUT(req, &s, sizeof(s)));
3698 	}
3699 	return (0);
3700 }
3701 
3702 /*
3703  * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
3704  * jail creation time but cannot be changed in an existing jail.
3705  */
3706 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
3707 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
3708 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
3709 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
3710 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
3711     "I", "Jail secure level");
3712 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
3713     "Jail value for kern.osreldate and uname -K");
3714 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
3715     "Jail value for kern.osrelease and uname -r");
3716 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
3717     "I", "Jail cannot see all mounted file systems");
3718 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
3719     "I", "Ruleset for in-jail devfs mounts");
3720 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
3721     "B", "Jail persistence");
3722 #ifdef VIMAGE
3723 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
3724     "E,jailsys", "Virtual network stack");
3725 #endif
3726 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
3727     "B", "Jail is in the process of shutting down");
3728 
3729 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
3730 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
3731     "I", "Current number of child jails");
3732 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
3733     "I", "Maximum number of child jails");
3734 
3735 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
3736 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
3737     "Jail hostname");
3738 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
3739     "Jail NIS domainname");
3740 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
3741     "Jail host UUID");
3742 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
3743     "LU", "Jail host ID");
3744 
3745 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
3746 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
3747 
3748 #ifdef INET
3749 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
3750     "Jail IPv4 address virtualization");
3751 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
3752     "S,in_addr,a", "Jail IPv4 addresses");
3753 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
3754     "B", "Do (not) use IPv4 source address selection rather than the "
3755     "primary jail IPv4 address.");
3756 #endif
3757 #ifdef INET6
3758 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
3759     "Jail IPv6 address virtualization");
3760 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
3761     "S,in6_addr,a", "Jail IPv6 addresses");
3762 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
3763     "B", "Do (not) use IPv6 source address selection rather than the "
3764     "primary jail IPv6 address.");
3765 #endif
3766 
3767 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
3768 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
3769     "B", "Jail may set hostname");
3770 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
3771     "B", "Jail may use SYSV IPC");
3772 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
3773     "B", "Jail may create raw sockets");
3774 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
3775     "B", "Jail may alter system file flags");
3776 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
3777     "B", "Jail may set file quotas");
3778 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
3779     "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
3780 SYSCTL_JAIL_PARAM(_allow, mlock, CTLTYPE_INT | CTLFLAG_RW,
3781     "B", "Jail may lock (unlock) physical pages in memory");
3782 SYSCTL_JAIL_PARAM(_allow, reserved_ports, CTLTYPE_INT | CTLFLAG_RW,
3783     "B", "Jail may bind sockets to reserved ports");
3784 SYSCTL_JAIL_PARAM(_allow, read_msgbuf, CTLTYPE_INT | CTLFLAG_RW,
3785     "B", "Jail may read the kernel message buffer");
3786 SYSCTL_JAIL_PARAM(_allow, unprivileged_proc_debug, CTLTYPE_INT | CTLFLAG_RW,
3787     "B", "Unprivileged processes may use process debugging facilities");
3788 
3789 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
3790 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
3791     "B", "Jail may mount/unmount jail-friendly file systems in general");
3792 
3793 /*
3794  * Add a dynamic parameter allow.<name>, or allow.<prefix>.<name>.  Return
3795  * its associated bit in the pr_allow bitmask, or zero if the parameter was
3796  * not created.
3797  */
3798 unsigned
3799 prison_add_allow(const char *prefix, const char *name, const char *prefix_descr,
3800     const char *descr)
3801 {
3802 	struct bool_flags *bf;
3803 	struct sysctl_oid *parent;
3804 	char *allow_name, *allow_noname, *allowed;
3805 #ifndef NO_SYSCTL_DESCR
3806 	char *descr_deprecated;
3807 #endif
3808 	unsigned allow_flag;
3809 
3810 	if (prefix
3811 	    ? asprintf(&allow_name, M_PRISON, "allow.%s.%s", prefix, name)
3812 		< 0 ||
3813 	      asprintf(&allow_noname, M_PRISON, "allow.%s.no%s", prefix, name)
3814 		< 0
3815 	    : asprintf(&allow_name, M_PRISON, "allow.%s", name) < 0 ||
3816 	      asprintf(&allow_noname, M_PRISON, "allow.no%s", name) < 0) {
3817 		free(allow_name, M_PRISON);
3818 		return 0;
3819 	}
3820 
3821 	/*
3822 	 * See if this parameter has already beed added, i.e. a module was
3823 	 * previously loaded/unloaded.
3824 	 */
3825 	mtx_lock(&prison0.pr_mtx);
3826 	for (bf = pr_flag_allow;
3827 	     bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
3828 	     bf++) {
3829 		if (strcmp(bf->name, allow_name) == 0) {
3830 			allow_flag = bf->flag;
3831 			goto no_add;
3832 		}
3833 	}
3834 
3835 	/*
3836 	 * Find a free bit in prison0's pr_allow, failing if there are none
3837 	 * (which shouldn't happen as long as we keep track of how many
3838 	 * potential dynamic flags exist).
3839 	 *
3840 	 * Due to per-jail unprivileged process debugging support
3841 	 * using pr_allow, also verify against PR_ALLOW_ALL_STATIC.
3842 	 * prison0 may have unprivileged process debugging unset.
3843 	 */
3844 	for (allow_flag = 1;; allow_flag <<= 1) {
3845 		if (allow_flag == 0)
3846 			goto no_add;
3847 		if (allow_flag & PR_ALLOW_ALL_STATIC)
3848 			continue;
3849 		if ((prison0.pr_allow & allow_flag) == 0)
3850 			break;
3851 	}
3852 
3853 	/*
3854 	 * Note the parameter in the next open slot in pr_flag_allow.
3855 	 * Set the flag last so code that checks pr_flag_allow can do so
3856 	 * without locking.
3857 	 */
3858 	for (bf = pr_flag_allow; bf->flag != 0; bf++)
3859 		if (bf == pr_flag_allow + nitems(pr_flag_allow)) {
3860 			/* This should never happen, but is not fatal. */
3861 			allow_flag = 0;
3862 			goto no_add;
3863 		}
3864 	prison0.pr_allow |= allow_flag;
3865 	bf->name = allow_name;
3866 	bf->noname = allow_noname;
3867 	bf->flag = allow_flag;
3868 	mtx_unlock(&prison0.pr_mtx);
3869 
3870 	/*
3871 	 * Create sysctls for the paramter, and the back-compat global
3872 	 * permission.
3873 	 */
3874 	parent = prefix
3875 	    ? SYSCTL_ADD_NODE(NULL,
3876 		  SYSCTL_CHILDREN(&sysctl___security_jail_param_allow),
3877 		  OID_AUTO, prefix, 0, 0, prefix_descr)
3878 	    : &sysctl___security_jail_param_allow;
3879 	(void)SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(parent), OID_AUTO,
3880 	    name, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3881 	    NULL, 0, sysctl_jail_param, "B", descr);
3882 	if ((prefix
3883 	     ? asprintf(&allowed, M_TEMP, "%s_%s_allowed", prefix, name)
3884 	     : asprintf(&allowed, M_TEMP, "%s_allowed", name)) >= 0) {
3885 #ifndef NO_SYSCTL_DESCR
3886 		(void)asprintf(&descr_deprecated, M_TEMP, "%s (deprecated)",
3887 		    descr);
3888 #endif
3889 		(void)SYSCTL_ADD_PROC(NULL,
3890 		    SYSCTL_CHILDREN(&sysctl___security_jail), OID_AUTO, allowed,
3891 		    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, allow_flag,
3892 		    sysctl_jail_default_allow, "I", descr_deprecated);
3893 #ifndef NO_SYSCTL_DESCR
3894 		free(descr_deprecated, M_TEMP);
3895 #endif
3896 		free(allowed, M_TEMP);
3897 	}
3898 	return allow_flag;
3899 
3900  no_add:
3901 	mtx_unlock(&prison0.pr_mtx);
3902 	free(allow_name, M_PRISON);
3903 	free(allow_noname, M_PRISON);
3904 	return allow_flag;
3905 }
3906 
3907 /*
3908  * The VFS system will register jail-aware filesystems here.  They each get
3909  * a parameter allow.mount.xxxfs and a flag to check when a jailed user
3910  * attempts to mount.
3911  */
3912 void
3913 prison_add_vfs(struct vfsconf *vfsp)
3914 {
3915 #ifdef NO_SYSCTL_DESCR
3916 
3917 	vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
3918 	    NULL, NULL);
3919 #else
3920 	char *descr;
3921 
3922 	(void)asprintf(&descr, M_TEMP, "Jail may mount the %s file system",
3923 	    vfsp->vfc_name);
3924 	vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
3925 	    NULL, descr);
3926 	free(descr, M_TEMP);
3927 #endif
3928 }
3929 
3930 #ifdef RACCT
3931 void
3932 prison_racct_foreach(void (*callback)(struct racct *racct,
3933     void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
3934     void *arg2, void *arg3)
3935 {
3936 	struct prison_racct *prr;
3937 
3938 	ASSERT_RACCT_ENABLED();
3939 
3940 	sx_slock(&allprison_lock);
3941 	if (pre != NULL)
3942 		(pre)();
3943 	LIST_FOREACH(prr, &allprison_racct, prr_next)
3944 		(callback)(prr->prr_racct, arg2, arg3);
3945 	if (post != NULL)
3946 		(post)();
3947 	sx_sunlock(&allprison_lock);
3948 }
3949 
3950 static struct prison_racct *
3951 prison_racct_find_locked(const char *name)
3952 {
3953 	struct prison_racct *prr;
3954 
3955 	ASSERT_RACCT_ENABLED();
3956 	sx_assert(&allprison_lock, SA_XLOCKED);
3957 
3958 	if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
3959 		return (NULL);
3960 
3961 	LIST_FOREACH(prr, &allprison_racct, prr_next) {
3962 		if (strcmp(name, prr->prr_name) != 0)
3963 			continue;
3964 
3965 		/* Found prison_racct with a matching name? */
3966 		prison_racct_hold(prr);
3967 		return (prr);
3968 	}
3969 
3970 	/* Add new prison_racct. */
3971 	prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
3972 	racct_create(&prr->prr_racct);
3973 
3974 	strcpy(prr->prr_name, name);
3975 	refcount_init(&prr->prr_refcount, 1);
3976 	LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
3977 
3978 	return (prr);
3979 }
3980 
3981 struct prison_racct *
3982 prison_racct_find(const char *name)
3983 {
3984 	struct prison_racct *prr;
3985 
3986 	ASSERT_RACCT_ENABLED();
3987 
3988 	sx_xlock(&allprison_lock);
3989 	prr = prison_racct_find_locked(name);
3990 	sx_xunlock(&allprison_lock);
3991 	return (prr);
3992 }
3993 
3994 void
3995 prison_racct_hold(struct prison_racct *prr)
3996 {
3997 
3998 	ASSERT_RACCT_ENABLED();
3999 
4000 	refcount_acquire(&prr->prr_refcount);
4001 }
4002 
4003 static void
4004 prison_racct_free_locked(struct prison_racct *prr)
4005 {
4006 
4007 	ASSERT_RACCT_ENABLED();
4008 	sx_assert(&allprison_lock, SA_XLOCKED);
4009 
4010 	if (refcount_release(&prr->prr_refcount)) {
4011 		racct_destroy(&prr->prr_racct);
4012 		LIST_REMOVE(prr, prr_next);
4013 		free(prr, M_PRISON_RACCT);
4014 	}
4015 }
4016 
4017 void
4018 prison_racct_free(struct prison_racct *prr)
4019 {
4020 
4021 	ASSERT_RACCT_ENABLED();
4022 	sx_assert(&allprison_lock, SA_UNLOCKED);
4023 
4024 	if (refcount_release_if_not_last(&prr->prr_refcount))
4025 		return;
4026 
4027 	sx_xlock(&allprison_lock);
4028 	prison_racct_free_locked(prr);
4029 	sx_xunlock(&allprison_lock);
4030 }
4031 
4032 static void
4033 prison_racct_attach(struct prison *pr)
4034 {
4035 	struct prison_racct *prr;
4036 
4037 	ASSERT_RACCT_ENABLED();
4038 	sx_assert(&allprison_lock, SA_XLOCKED);
4039 
4040 	prr = prison_racct_find_locked(pr->pr_name);
4041 	KASSERT(prr != NULL, ("cannot find prison_racct"));
4042 
4043 	pr->pr_prison_racct = prr;
4044 }
4045 
4046 /*
4047  * Handle jail renaming.  From the racct point of view, renaming means
4048  * moving from one prison_racct to another.
4049  */
4050 static void
4051 prison_racct_modify(struct prison *pr)
4052 {
4053 #ifdef RCTL
4054 	struct proc *p;
4055 	struct ucred *cred;
4056 #endif
4057 	struct prison_racct *oldprr;
4058 
4059 	ASSERT_RACCT_ENABLED();
4060 
4061 	sx_slock(&allproc_lock);
4062 	sx_xlock(&allprison_lock);
4063 
4064 	if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4065 		sx_xunlock(&allprison_lock);
4066 		sx_sunlock(&allproc_lock);
4067 		return;
4068 	}
4069 
4070 	oldprr = pr->pr_prison_racct;
4071 	pr->pr_prison_racct = NULL;
4072 
4073 	prison_racct_attach(pr);
4074 
4075 	/*
4076 	 * Move resource utilisation records.
4077 	 */
4078 	racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
4079 
4080 #ifdef RCTL
4081 	/*
4082 	 * Force rctl to reattach rules to processes.
4083 	 */
4084 	FOREACH_PROC_IN_SYSTEM(p) {
4085 		PROC_LOCK(p);
4086 		cred = crhold(p->p_ucred);
4087 		PROC_UNLOCK(p);
4088 		rctl_proc_ucred_changed(p, cred);
4089 		crfree(cred);
4090 	}
4091 #endif
4092 
4093 	sx_sunlock(&allproc_lock);
4094 	prison_racct_free_locked(oldprr);
4095 	sx_xunlock(&allprison_lock);
4096 }
4097 
4098 static void
4099 prison_racct_detach(struct prison *pr)
4100 {
4101 
4102 	ASSERT_RACCT_ENABLED();
4103 	sx_assert(&allprison_lock, SA_UNLOCKED);
4104 
4105 	if (pr->pr_prison_racct == NULL)
4106 		return;
4107 	prison_racct_free(pr->pr_prison_racct);
4108 	pr->pr_prison_racct = NULL;
4109 }
4110 #endif /* RACCT */
4111 
4112 #ifdef DDB
4113 
4114 static void
4115 db_show_prison(struct prison *pr)
4116 {
4117 	struct bool_flags *bf;
4118 	struct jailsys_flags *jsf;
4119 #if defined(INET) || defined(INET6)
4120 	int ii;
4121 #endif
4122 	unsigned f;
4123 #ifdef INET
4124 	char ip4buf[INET_ADDRSTRLEN];
4125 #endif
4126 #ifdef INET6
4127 	char ip6buf[INET6_ADDRSTRLEN];
4128 #endif
4129 
4130 	db_printf("prison %p:\n", pr);
4131 	db_printf(" jid             = %d\n", pr->pr_id);
4132 	db_printf(" name            = %s\n", pr->pr_name);
4133 	db_printf(" parent          = %p\n", pr->pr_parent);
4134 	db_printf(" ref             = %d\n", pr->pr_ref);
4135 	db_printf(" uref            = %d\n", pr->pr_uref);
4136 	db_printf(" path            = %s\n", pr->pr_path);
4137 	db_printf(" cpuset          = %d\n", pr->pr_cpuset
4138 	    ? pr->pr_cpuset->cs_id : -1);
4139 #ifdef VIMAGE
4140 	db_printf(" vnet            = %p\n", pr->pr_vnet);
4141 #endif
4142 	db_printf(" root            = %p\n", pr->pr_root);
4143 	db_printf(" securelevel     = %d\n", pr->pr_securelevel);
4144 	db_printf(" devfs_rsnum     = %d\n", pr->pr_devfs_rsnum);
4145 	db_printf(" children.max    = %d\n", pr->pr_childmax);
4146 	db_printf(" children.cur    = %d\n", pr->pr_childcount);
4147 	db_printf(" child           = %p\n", LIST_FIRST(&pr->pr_children));
4148 	db_printf(" sibling         = %p\n", LIST_NEXT(pr, pr_sibling));
4149 	db_printf(" flags           = 0x%x", pr->pr_flags);
4150 	for (bf = pr_flag_bool; bf < pr_flag_bool + nitems(pr_flag_bool); bf++)
4151 		if (pr->pr_flags & bf->flag)
4152 			db_printf(" %s", bf->name);
4153 	for (jsf = pr_flag_jailsys;
4154 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
4155 	     jsf++) {
4156 		f = pr->pr_flags & (jsf->disable | jsf->new);
4157 		db_printf(" %-16s= %s\n", jsf->name,
4158 		    (f != 0 && f == jsf->disable) ? "disable"
4159 		    : (f == jsf->new) ? "new"
4160 		    : "inherit");
4161 	}
4162 	db_printf(" allow           = 0x%x", pr->pr_allow);
4163 	for (bf = pr_flag_allow;
4164 	     bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
4165 	     bf++)
4166 		if (pr->pr_allow & bf->flag)
4167 			db_printf(" %s", bf->name);
4168 	db_printf("\n");
4169 	db_printf(" enforce_statfs  = %d\n", pr->pr_enforce_statfs);
4170 	db_printf(" host.hostname   = %s\n", pr->pr_hostname);
4171 	db_printf(" host.domainname = %s\n", pr->pr_domainname);
4172 	db_printf(" host.hostuuid   = %s\n", pr->pr_hostuuid);
4173 	db_printf(" host.hostid     = %lu\n", pr->pr_hostid);
4174 #ifdef INET
4175 	db_printf(" ip4s            = %d\n", pr->pr_ip4s);
4176 	for (ii = 0; ii < pr->pr_ip4s; ii++)
4177 		db_printf(" %s %s\n",
4178 		    ii == 0 ? "ip4.addr        =" : "                 ",
4179 		    inet_ntoa_r(pr->pr_ip4[ii], ip4buf));
4180 #endif
4181 #ifdef INET6
4182 	db_printf(" ip6s            = %d\n", pr->pr_ip6s);
4183 	for (ii = 0; ii < pr->pr_ip6s; ii++)
4184 		db_printf(" %s %s\n",
4185 		    ii == 0 ? "ip6.addr        =" : "                 ",
4186 		    ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4187 #endif
4188 }
4189 
4190 DB_SHOW_COMMAND(prison, db_show_prison_command)
4191 {
4192 	struct prison *pr;
4193 
4194 	if (!have_addr) {
4195 		/*
4196 		 * Show all prisons in the list, and prison0 which is not
4197 		 * listed.
4198 		 */
4199 		db_show_prison(&prison0);
4200 		if (!db_pager_quit) {
4201 			TAILQ_FOREACH(pr, &allprison, pr_list) {
4202 				db_show_prison(pr);
4203 				if (db_pager_quit)
4204 					break;
4205 			}
4206 		}
4207 		return;
4208 	}
4209 
4210 	if (addr == 0)
4211 		pr = &prison0;
4212 	else {
4213 		/* Look for a prison with the ID and with references. */
4214 		TAILQ_FOREACH(pr, &allprison, pr_list)
4215 			if (pr->pr_id == addr && pr->pr_ref > 0)
4216 				break;
4217 		if (pr == NULL)
4218 			/* Look again, without requiring a reference. */
4219 			TAILQ_FOREACH(pr, &allprison, pr_list)
4220 				if (pr->pr_id == addr)
4221 					break;
4222 		if (pr == NULL)
4223 			/* Assume address points to a valid prison. */
4224 			pr = (struct prison *)addr;
4225 	}
4226 	db_show_prison(pr);
4227 }
4228 
4229 #endif /* DDB */
4230