xref: /freebsd/sys/kern/kern_jail.c (revision ce6a89e27cd190313be39bb479880aeda4778436)
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 #if defined(INET) || defined(INET6)
494 	int ii, ij;
495 #endif
496 #ifdef INET
497 	int ip4s, redo_ip4;
498 #endif
499 #ifdef INET6
500 	int ip6s, redo_ip6;
501 #endif
502 	uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
503 	uint64_t pr_allow_diff;
504 	unsigned tallow;
505 	char numbuf[12];
506 
507 	error = priv_check(td, PRIV_JAIL_SET);
508 	if (!error && (flags & JAIL_ATTACH))
509 		error = priv_check(td, PRIV_JAIL_ATTACH);
510 	if (error)
511 		return (error);
512 	mypr = td->td_ucred->cr_prison;
513 	if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
514 		return (EPERM);
515 	if (flags & ~JAIL_SET_MASK)
516 		return (EINVAL);
517 
518 	/*
519 	 * Check all the parameters before committing to anything.  Not all
520 	 * errors can be caught early, but we may as well try.  Also, this
521 	 * takes care of some expensive stuff (path lookup) before getting
522 	 * the allprison lock.
523 	 *
524 	 * XXX Jails are not filesystems, and jail parameters are not mount
525 	 *     options.  But it makes more sense to re-use the vfsopt code
526 	 *     than duplicate it under a different name.
527 	 */
528 	error = vfs_buildopts(optuio, &opts);
529 	if (error)
530 		return (error);
531 #ifdef INET
532 	ip4 = NULL;
533 #endif
534 #ifdef INET6
535 	ip6 = NULL;
536 #endif
537 	g_path = NULL;
538 
539 	cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
540 	if (!cuflags) {
541 		error = EINVAL;
542 		vfs_opterror(opts, "no valid operation (create or update)");
543 		goto done_errmsg;
544 	}
545 
546 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
547 	if (error == ENOENT)
548 		jid = 0;
549 	else if (error != 0)
550 		goto done_free;
551 
552 	error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
553 	if (error == ENOENT)
554 		gotslevel = 0;
555 	else if (error != 0)
556 		goto done_free;
557 	else
558 		gotslevel = 1;
559 
560 	error =
561 	    vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
562 	if (error == ENOENT)
563 		gotchildmax = 0;
564 	else if (error != 0)
565 		goto done_free;
566 	else
567 		gotchildmax = 1;
568 
569 	error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
570 	if (error == ENOENT)
571 		gotenforce = 0;
572 	else if (error != 0)
573 		goto done_free;
574 	else if (enforce < 0 || enforce > 2) {
575 		error = EINVAL;
576 		goto done_free;
577 	} else
578 		gotenforce = 1;
579 
580 	error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
581 	if (error == ENOENT)
582 		gotrsnum = 0;
583 	else if (error != 0)
584 		goto done_free;
585 	else
586 		gotrsnum = 1;
587 
588 	pr_flags = ch_flags = 0;
589 	for (bf = pr_flag_bool;
590 	     bf < pr_flag_bool + nitems(pr_flag_bool);
591 	     bf++) {
592 		vfs_flagopt(opts, bf->name, &pr_flags, bf->flag);
593 		vfs_flagopt(opts, bf->noname, &ch_flags, bf->flag);
594 	}
595 	ch_flags |= pr_flags;
596 	for (jsf = pr_flag_jailsys;
597 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
598 	     jsf++) {
599 		error = vfs_copyopt(opts, jsf->name, &jsys, sizeof(jsys));
600 		if (error == ENOENT)
601 			continue;
602 		if (error != 0)
603 			goto done_free;
604 		switch (jsys) {
605 		case JAIL_SYS_DISABLE:
606 			if (!jsf->disable) {
607 				error = EINVAL;
608 				goto done_free;
609 			}
610 			pr_flags |= jsf->disable;
611 			break;
612 		case JAIL_SYS_NEW:
613 			pr_flags |= jsf->new;
614 			break;
615 		case JAIL_SYS_INHERIT:
616 			break;
617 		default:
618 			error = EINVAL;
619 			goto done_free;
620 		}
621 		ch_flags |= jsf->new | jsf->disable;
622 	}
623 	if ((flags & (JAIL_CREATE | JAIL_UPDATE | JAIL_ATTACH)) == JAIL_CREATE
624 	    && !(pr_flags & PR_PERSIST)) {
625 		error = EINVAL;
626 		vfs_opterror(opts, "new jail must persist or attach");
627 		goto done_errmsg;
628 	}
629 #ifdef VIMAGE
630 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
631 		error = EINVAL;
632 		vfs_opterror(opts, "vnet cannot be changed after creation");
633 		goto done_errmsg;
634 	}
635 #endif
636 #ifdef INET
637 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
638 		error = EINVAL;
639 		vfs_opterror(opts, "ip4 cannot be changed after creation");
640 		goto done_errmsg;
641 	}
642 #endif
643 #ifdef INET6
644 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
645 		error = EINVAL;
646 		vfs_opterror(opts, "ip6 cannot be changed after creation");
647 		goto done_errmsg;
648 	}
649 #endif
650 
651 	pr_allow = ch_allow = 0;
652 	for (bf = pr_flag_allow;
653 	     bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
654 	     bf++) {
655 		vfs_flagopt(opts, bf->name, &pr_allow, bf->flag);
656 		vfs_flagopt(opts, bf->noname, &ch_allow, bf->flag);
657 	}
658 	ch_allow |= pr_allow;
659 
660 	error = vfs_getopt(opts, "name", (void **)&name, &len);
661 	if (error == ENOENT)
662 		name = NULL;
663 	else if (error != 0)
664 		goto done_free;
665 	else {
666 		if (len == 0 || name[len - 1] != '\0') {
667 			error = EINVAL;
668 			goto done_free;
669 		}
670 		if (len > MAXHOSTNAMELEN) {
671 			error = ENAMETOOLONG;
672 			goto done_free;
673 		}
674 	}
675 
676 	error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
677 	if (error == ENOENT)
678 		host = NULL;
679 	else if (error != 0)
680 		goto done_free;
681 	else {
682 		ch_flags |= PR_HOST;
683 		pr_flags |= PR_HOST;
684 		if (len == 0 || host[len - 1] != '\0') {
685 			error = EINVAL;
686 			goto done_free;
687 		}
688 		if (len > MAXHOSTNAMELEN) {
689 			error = ENAMETOOLONG;
690 			goto done_free;
691 		}
692 	}
693 
694 	error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
695 	if (error == ENOENT)
696 		domain = NULL;
697 	else if (error != 0)
698 		goto done_free;
699 	else {
700 		ch_flags |= PR_HOST;
701 		pr_flags |= PR_HOST;
702 		if (len == 0 || domain[len - 1] != '\0') {
703 			error = EINVAL;
704 			goto done_free;
705 		}
706 		if (len > MAXHOSTNAMELEN) {
707 			error = ENAMETOOLONG;
708 			goto done_free;
709 		}
710 	}
711 
712 	error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
713 	if (error == ENOENT)
714 		uuid = NULL;
715 	else if (error != 0)
716 		goto done_free;
717 	else {
718 		ch_flags |= PR_HOST;
719 		pr_flags |= PR_HOST;
720 		if (len == 0 || uuid[len - 1] != '\0') {
721 			error = EINVAL;
722 			goto done_free;
723 		}
724 		if (len > HOSTUUIDLEN) {
725 			error = ENAMETOOLONG;
726 			goto done_free;
727 		}
728 	}
729 
730 #ifdef COMPAT_FREEBSD32
731 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
732 		uint32_t hid32;
733 
734 		error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
735 		hid = hid32;
736 	} else
737 #endif
738 		error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
739 	if (error == ENOENT)
740 		gothid = 0;
741 	else if (error != 0)
742 		goto done_free;
743 	else {
744 		gothid = 1;
745 		ch_flags |= PR_HOST;
746 		pr_flags |= PR_HOST;
747 	}
748 
749 #ifdef INET
750 	error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
751 	if (error == ENOENT)
752 		ip4s = 0;
753 	else if (error != 0)
754 		goto done_free;
755 	else if (ip4s & (sizeof(*ip4) - 1)) {
756 		error = EINVAL;
757 		goto done_free;
758 	} else {
759 		ch_flags |= PR_IP4_USER;
760 		pr_flags |= PR_IP4_USER;
761 		if (ip4s > 0) {
762 			ip4s /= sizeof(*ip4);
763 			if (ip4s > jail_max_af_ips) {
764 				error = EINVAL;
765 				vfs_opterror(opts, "too many IPv4 addresses");
766 				goto done_errmsg;
767 			}
768 			ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
769 			bcopy(op, ip4, ip4s * sizeof(*ip4));
770 			/*
771 			 * IP addresses are all sorted but ip[0] to preserve
772 			 * the primary IP address as given from userland.
773 			 * This special IP is used for unbound outgoing
774 			 * connections as well for "loopback" traffic in case
775 			 * source address selection cannot find any more fitting
776 			 * address to connect from.
777 			 */
778 			if (ip4s > 1)
779 				qsort(ip4 + 1, ip4s - 1, sizeof(*ip4),
780 				    prison_qcmp_v4);
781 			/*
782 			 * Check for duplicate addresses and do some simple
783 			 * zero and broadcast checks. If users give other bogus
784 			 * addresses it is their problem.
785 			 *
786 			 * We do not have to care about byte order for these
787 			 * checks so we will do them in NBO.
788 			 */
789 			for (ii = 0; ii < ip4s; ii++) {
790 				if (ip4[ii].s_addr == INADDR_ANY ||
791 				    ip4[ii].s_addr == INADDR_BROADCAST) {
792 					error = EINVAL;
793 					goto done_free;
794 				}
795 				if ((ii+1) < ip4s &&
796 				    (ip4[0].s_addr == ip4[ii+1].s_addr ||
797 				     ip4[ii].s_addr == ip4[ii+1].s_addr)) {
798 					error = EINVAL;
799 					goto done_free;
800 				}
801 			}
802 		}
803 	}
804 #endif
805 
806 #ifdef INET6
807 	error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
808 	if (error == ENOENT)
809 		ip6s = 0;
810 	else if (error != 0)
811 		goto done_free;
812 	else if (ip6s & (sizeof(*ip6) - 1)) {
813 		error = EINVAL;
814 		goto done_free;
815 	} else {
816 		ch_flags |= PR_IP6_USER;
817 		pr_flags |= PR_IP6_USER;
818 		if (ip6s > 0) {
819 			ip6s /= sizeof(*ip6);
820 			if (ip6s > jail_max_af_ips) {
821 				error = EINVAL;
822 				vfs_opterror(opts, "too many IPv6 addresses");
823 				goto done_errmsg;
824 			}
825 			ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
826 			bcopy(op, ip6, ip6s * sizeof(*ip6));
827 			if (ip6s > 1)
828 				qsort(ip6 + 1, ip6s - 1, sizeof(*ip6),
829 				    prison_qcmp_v6);
830 			for (ii = 0; ii < ip6s; ii++) {
831 				if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
832 					error = EINVAL;
833 					goto done_free;
834 				}
835 				if ((ii+1) < ip6s &&
836 				    (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
837 				     IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
838 				{
839 					error = EINVAL;
840 					goto done_free;
841 				}
842 			}
843 		}
844 	}
845 #endif
846 
847 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
848 	if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
849 		error = EINVAL;
850 		vfs_opterror(opts,
851 		    "vnet jails cannot have IP address restrictions");
852 		goto done_errmsg;
853 	}
854 #endif
855 
856 	error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
857 	if (error == ENOENT)
858 		osrelstr = NULL;
859 	else if (error != 0)
860 		goto done_free;
861 	else {
862 		if (flags & JAIL_UPDATE) {
863 			error = EINVAL;
864 			vfs_opterror(opts,
865 			    "osrelease cannot be changed after creation");
866 			goto done_errmsg;
867 		}
868 		if (len == 0 || osrelstr[len - 1] != '\0') {
869 			error = EINVAL;
870 			goto done_free;
871 		}
872 		if (len >= OSRELEASELEN) {
873 			error = ENAMETOOLONG;
874 			vfs_opterror(opts,
875 			    "osrelease string must be 1-%d bytes long",
876 			    OSRELEASELEN - 1);
877 			goto done_errmsg;
878 		}
879 	}
880 
881 	error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
882 	if (error == ENOENT)
883 		osreldt = 0;
884 	else if (error != 0)
885 		goto done_free;
886 	else {
887 		if (flags & JAIL_UPDATE) {
888 			error = EINVAL;
889 			vfs_opterror(opts,
890 			    "osreldate cannot be changed after creation");
891 			goto done_errmsg;
892 		}
893 		if (osreldt == 0) {
894 			error = EINVAL;
895 			vfs_opterror(opts, "osreldate cannot be 0");
896 			goto done_errmsg;
897 		}
898 	}
899 
900 	root = NULL;
901 	error = vfs_getopt(opts, "path", (void **)&path, &len);
902 	if (error == ENOENT)
903 		path = NULL;
904 	else if (error != 0)
905 		goto done_free;
906 	else {
907 		if (flags & JAIL_UPDATE) {
908 			error = EINVAL;
909 			vfs_opterror(opts,
910 			    "path cannot be changed after creation");
911 			goto done_errmsg;
912 		}
913 		if (len == 0 || path[len - 1] != '\0') {
914 			error = EINVAL;
915 			goto done_free;
916 		}
917 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
918 		    path, td);
919 		error = namei(&nd);
920 		if (error)
921 			goto done_free;
922 		root = nd.ni_vp;
923 		NDFREE(&nd, NDF_ONLY_PNBUF);
924 		g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
925 		strlcpy(g_path, path, MAXPATHLEN);
926 		error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
927 		if (error == 0) {
928 			path = g_path;
929 		} else {
930 			/* exit on other errors */
931 			goto done_free;
932 		}
933 		if (root->v_type != VDIR) {
934 			error = ENOTDIR;
935 			vput(root);
936 			goto done_free;
937 		}
938 		VOP_UNLOCK(root);
939 	}
940 
941 	/*
942 	 * Find the specified jail, or at least its parent.
943 	 * This abuses the file error codes ENOENT and EEXIST.
944 	 */
945 	pr = NULL;
946 	ppr = mypr;
947 	if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
948 		namelc = strrchr(name, '.');
949 		jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
950 		if (*p != '\0')
951 			jid = 0;
952 	}
953 	sx_xlock(&allprison_lock);
954 	if (jid != 0) {
955 		/*
956 		 * See if a requested jid already exists.  There is an
957 		 * information leak here if the jid exists but is not within
958 		 * the caller's jail hierarchy.  Jail creators will get EEXIST
959 		 * even though they cannot see the jail, and CREATE | UPDATE
960 		 * will return ENOENT which is not normally a valid error.
961 		 */
962 		if (jid < 0) {
963 			error = EINVAL;
964 			vfs_opterror(opts, "negative jid");
965 			goto done_unlock_list;
966 		}
967 		pr = prison_find(jid);
968 		if (pr != NULL) {
969 			ppr = pr->pr_parent;
970 			/* Create: jid must not exist. */
971 			if (cuflags == JAIL_CREATE) {
972 				mtx_unlock(&pr->pr_mtx);
973 				error = EEXIST;
974 				vfs_opterror(opts, "jail %d already exists",
975 				    jid);
976 				goto done_unlock_list;
977 			}
978 			if (!prison_ischild(mypr, pr)) {
979 				mtx_unlock(&pr->pr_mtx);
980 				pr = NULL;
981 			} else if (pr->pr_uref == 0) {
982 				if (!(flags & JAIL_DYING)) {
983 					mtx_unlock(&pr->pr_mtx);
984 					error = ENOENT;
985 					vfs_opterror(opts, "jail %d is dying",
986 					    jid);
987 					goto done_unlock_list;
988 				} else if ((flags & JAIL_ATTACH) ||
989 				    (pr_flags & PR_PERSIST)) {
990 					/*
991 					 * A dying jail might be resurrected
992 					 * (via attach or persist), but first
993 					 * it must determine if another jail
994 					 * has claimed its name.  Accomplish
995 					 * this by implicitly re-setting the
996 					 * name.
997 					 */
998 					if (name == NULL)
999 						name = prison_name(mypr, pr);
1000 				}
1001 			}
1002 		}
1003 		if (pr == NULL) {
1004 			/* Update: jid must exist. */
1005 			if (cuflags == JAIL_UPDATE) {
1006 				error = ENOENT;
1007 				vfs_opterror(opts, "jail %d not found", jid);
1008 				goto done_unlock_list;
1009 			}
1010 		}
1011 	}
1012 	/*
1013 	 * If the caller provided a name, look for a jail by that name.
1014 	 * This has different semantics for creates and updates keyed by jid
1015 	 * (where the name must not already exist in a different jail),
1016 	 * and updates keyed by the name itself (where the name must exist
1017 	 * because that is the jail being updated).
1018 	 */
1019 	namelc = NULL;
1020 	if (name != NULL) {
1021 		namelc = strrchr(name, '.');
1022 		if (namelc == NULL)
1023 			namelc = name;
1024 		else {
1025 			/*
1026 			 * This is a hierarchical name.  Split it into the
1027 			 * parent and child names, and make sure the parent
1028 			 * exists or matches an already found jail.
1029 			 */
1030 			if (pr != NULL) {
1031 				if (strncmp(name, ppr->pr_name, namelc - name)
1032 				    || ppr->pr_name[namelc - name] != '\0') {
1033 					mtx_unlock(&pr->pr_mtx);
1034 					error = EINVAL;
1035 					vfs_opterror(opts,
1036 					    "cannot change jail's parent");
1037 					goto done_unlock_list;
1038 				}
1039 			} else {
1040 				*namelc = '\0';
1041 				ppr = prison_find_name(mypr, name);
1042 				if (ppr == NULL) {
1043 					error = ENOENT;
1044 					vfs_opterror(opts,
1045 					    "jail \"%s\" not found", name);
1046 					goto done_unlock_list;
1047 				}
1048 				mtx_unlock(&ppr->pr_mtx);
1049 				*namelc = '.';
1050 			}
1051 			namelc++;
1052 		}
1053 		if (namelc[0] != '\0') {
1054 			pnamelen =
1055 			    (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1056  name_again:
1057 			deadpr = NULL;
1058 			FOREACH_PRISON_CHILD(ppr, tpr) {
1059 				if (tpr != pr && tpr->pr_ref > 0 &&
1060 				    !strcmp(tpr->pr_name + pnamelen, namelc)) {
1061 					if (pr == NULL &&
1062 					    cuflags != JAIL_CREATE) {
1063 						mtx_lock(&tpr->pr_mtx);
1064 						if (tpr->pr_ref > 0) {
1065 							/*
1066 							 * Use this jail
1067 							 * for updates.
1068 							 */
1069 							if (tpr->pr_uref > 0) {
1070 								pr = tpr;
1071 								break;
1072 							}
1073 							deadpr = tpr;
1074 						}
1075 						mtx_unlock(&tpr->pr_mtx);
1076 					} else if (tpr->pr_uref > 0) {
1077 						/*
1078 						 * Create, or update(jid):
1079 						 * name must not exist in an
1080 						 * active sibling jail.
1081 						 */
1082 						error = EEXIST;
1083 						if (pr != NULL)
1084 							mtx_unlock(&pr->pr_mtx);
1085 						vfs_opterror(opts,
1086 						   "jail \"%s\" already exists",
1087 						   name);
1088 						goto done_unlock_list;
1089 					}
1090 				}
1091 			}
1092 			/* If no active jail is found, use a dying one. */
1093 			if (deadpr != NULL && pr == NULL) {
1094 				if (flags & JAIL_DYING) {
1095 					mtx_lock(&deadpr->pr_mtx);
1096 					if (deadpr->pr_ref == 0) {
1097 						mtx_unlock(&deadpr->pr_mtx);
1098 						goto name_again;
1099 					}
1100 					pr = deadpr;
1101 				} else if (cuflags == JAIL_UPDATE) {
1102 					error = ENOENT;
1103 					vfs_opterror(opts,
1104 					    "jail \"%s\" is dying", name);
1105 					goto done_unlock_list;
1106 				}
1107 			}
1108 			/* Update: name must exist if no jid. */
1109 			else if (cuflags == JAIL_UPDATE && pr == NULL) {
1110 				error = ENOENT;
1111 				vfs_opterror(opts, "jail \"%s\" not found",
1112 				    name);
1113 				goto done_unlock_list;
1114 			}
1115 		}
1116 	}
1117 	/* Update: must provide a jid or name. */
1118 	else if (cuflags == JAIL_UPDATE && pr == NULL) {
1119 		error = ENOENT;
1120 		vfs_opterror(opts, "update specified no jail");
1121 		goto done_unlock_list;
1122 	}
1123 
1124 	/* If there's no prison to update, create a new one and link it in. */
1125 	if (pr == NULL) {
1126 		for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1127 			if (tpr->pr_childcount >= tpr->pr_childmax) {
1128 				error = EPERM;
1129 				vfs_opterror(opts, "prison limit exceeded");
1130 				goto done_unlock_list;
1131 			}
1132 		created = 1;
1133 		mtx_lock(&ppr->pr_mtx);
1134 		if (ppr->pr_ref == 0) {
1135 			mtx_unlock(&ppr->pr_mtx);
1136 			error = ENOENT;
1137 			vfs_opterror(opts, "jail \"%s\" not found",
1138 			    prison_name(mypr, ppr));
1139 			goto done_unlock_list;
1140 		}
1141 		ppr->pr_ref++;
1142 		ppr->pr_uref++;
1143 		mtx_unlock(&ppr->pr_mtx);
1144 		pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1145 		if (jid == 0) {
1146 			/* Find the next free jid. */
1147 			jid = lastprid + 1;
1148  findnext:
1149 			if (jid == JAIL_MAX)
1150 				jid = 1;
1151 			TAILQ_FOREACH(tpr, &allprison, pr_list) {
1152 				if (tpr->pr_id < jid)
1153 					continue;
1154 				if (tpr->pr_id > jid || tpr->pr_ref == 0) {
1155 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1156 					break;
1157 				}
1158 				if (jid == lastprid) {
1159 					error = EAGAIN;
1160 					vfs_opterror(opts,
1161 					    "no available jail IDs");
1162 					free(pr, M_PRISON);
1163 					prison_deref(ppr, PD_DEREF |
1164 					    PD_DEUREF | PD_LIST_XLOCKED);
1165 					goto done_releroot;
1166 				}
1167 				jid++;
1168 				goto findnext;
1169 			}
1170 			lastprid = jid;
1171 		} else {
1172 			/*
1173 			 * The jail already has a jid (that did not yet exist),
1174 			 * so just find where to insert it.
1175 			 */
1176 			TAILQ_FOREACH(tpr, &allprison, pr_list)
1177 				if (tpr->pr_id >= jid) {
1178 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1179 					break;
1180 				}
1181 		}
1182 		if (tpr == NULL)
1183 			TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1184 		LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1185 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1186 			tpr->pr_childcount++;
1187 
1188 		pr->pr_parent = ppr;
1189 		pr->pr_id = jid;
1190 
1191 		/* Set some default values, and inherit some from the parent. */
1192 		if (namelc == NULL)
1193 			namelc = "";
1194 		if (path == NULL) {
1195 			path = "/";
1196 			root = mypr->pr_root;
1197 			vref(root);
1198 		}
1199 		strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1200 		pr->pr_flags |= PR_HOST;
1201 #if defined(INET) || defined(INET6)
1202 #ifdef VIMAGE
1203 		if (!(pr_flags & PR_VNET))
1204 #endif
1205 		{
1206 #ifdef INET
1207 			if (!(ch_flags & PR_IP4_USER))
1208 				pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1209 			else if (!(pr_flags & PR_IP4_USER)) {
1210 				pr->pr_flags |= ppr->pr_flags & PR_IP4;
1211 				if (ppr->pr_ip4 != NULL) {
1212 					pr->pr_ip4s = ppr->pr_ip4s;
1213 					pr->pr_ip4 = malloc(pr->pr_ip4s *
1214 					    sizeof(struct in_addr), M_PRISON,
1215 					    M_WAITOK);
1216 					bcopy(ppr->pr_ip4, pr->pr_ip4,
1217 					    pr->pr_ip4s * sizeof(*pr->pr_ip4));
1218 				}
1219 			}
1220 #endif
1221 #ifdef INET6
1222 			if (!(ch_flags & PR_IP6_USER))
1223 				pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1224 			else if (!(pr_flags & PR_IP6_USER)) {
1225 				pr->pr_flags |= ppr->pr_flags & PR_IP6;
1226 				if (ppr->pr_ip6 != NULL) {
1227 					pr->pr_ip6s = ppr->pr_ip6s;
1228 					pr->pr_ip6 = malloc(pr->pr_ip6s *
1229 					    sizeof(struct in6_addr), M_PRISON,
1230 					    M_WAITOK);
1231 					bcopy(ppr->pr_ip6, pr->pr_ip6,
1232 					    pr->pr_ip6s * sizeof(*pr->pr_ip6));
1233 				}
1234 			}
1235 #endif
1236 		}
1237 #endif
1238 		/* Source address selection is always on by default. */
1239 		pr->pr_flags |= _PR_IP_SADDRSEL;
1240 
1241 		pr->pr_securelevel = ppr->pr_securelevel;
1242 		pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1243 		pr->pr_enforce_statfs = jail_default_enforce_statfs;
1244 		pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1245 
1246 		pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1247 		if (osrelstr == NULL)
1248 			strlcpy(pr->pr_osrelease, ppr->pr_osrelease,
1249 			    sizeof(pr->pr_osrelease));
1250 		else
1251 			strlcpy(pr->pr_osrelease, osrelstr,
1252 			    sizeof(pr->pr_osrelease));
1253 
1254 		LIST_INIT(&pr->pr_children);
1255 		mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1256 		TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1257 
1258 #ifdef VIMAGE
1259 		/* Allocate a new vnet if specified. */
1260 		pr->pr_vnet = (pr_flags & PR_VNET)
1261 		    ? vnet_alloc() : ppr->pr_vnet;
1262 #endif
1263 		/*
1264 		 * Allocate a dedicated cpuset for each jail.
1265 		 * Unlike other initial settings, this may return an erorr.
1266 		 */
1267 		error = cpuset_create_root(ppr, &pr->pr_cpuset);
1268 		if (error) {
1269 			prison_deref(pr, PD_LIST_XLOCKED);
1270 			goto done_releroot;
1271 		}
1272 
1273 		mtx_lock(&pr->pr_mtx);
1274 		/*
1275 		 * New prisons do not yet have a reference, because we do not
1276 		 * want others to see the incomplete prison once the
1277 		 * allprison_lock is downgraded.
1278 		 */
1279 	} else {
1280 		created = 0;
1281 		/*
1282 		 * Grab a reference for existing prisons, to ensure they
1283 		 * continue to exist for the duration of the call.
1284 		 */
1285 		pr->pr_ref++;
1286 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1287 		if ((pr->pr_flags & PR_VNET) &&
1288 		    (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1289 			error = EINVAL;
1290 			vfs_opterror(opts,
1291 			    "vnet jails cannot have IP address restrictions");
1292 			goto done_deref_locked;
1293 		}
1294 #endif
1295 #ifdef INET
1296 		if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1297 			error = EINVAL;
1298 			vfs_opterror(opts,
1299 			    "ip4 cannot be changed after creation");
1300 			goto done_deref_locked;
1301 		}
1302 #endif
1303 #ifdef INET6
1304 		if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1305 			error = EINVAL;
1306 			vfs_opterror(opts,
1307 			    "ip6 cannot be changed after creation");
1308 			goto done_deref_locked;
1309 		}
1310 #endif
1311 	}
1312 
1313 	/* Do final error checking before setting anything. */
1314 	if (gotslevel) {
1315 		if (slevel < ppr->pr_securelevel) {
1316 			error = EPERM;
1317 			goto done_deref_locked;
1318 		}
1319 	}
1320 	if (gotchildmax) {
1321 		if (childmax >= ppr->pr_childmax) {
1322 			error = EPERM;
1323 			goto done_deref_locked;
1324 		}
1325 	}
1326 	if (gotenforce) {
1327 		if (enforce < ppr->pr_enforce_statfs) {
1328 			error = EPERM;
1329 			goto done_deref_locked;
1330 		}
1331 	}
1332 	if (gotrsnum) {
1333 		/*
1334 		 * devfs_rsnum is a uint16_t
1335 		 */
1336 		if (rsnum < 0 || rsnum > 65535) {
1337 			error = EINVAL;
1338 			goto done_deref_locked;
1339 		}
1340 		/*
1341 		 * Nested jails always inherit parent's devfs ruleset
1342 		 */
1343 		if (jailed(td->td_ucred)) {
1344 			if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1345 				error = EPERM;
1346 				goto done_deref_locked;
1347 			} else
1348 				rsnum = ppr->pr_devfs_rsnum;
1349 		}
1350 	}
1351 #ifdef INET
1352 	if (ip4s > 0) {
1353 		if (ppr->pr_flags & PR_IP4) {
1354 			/*
1355 			 * Make sure the new set of IP addresses is a
1356 			 * subset of the parent's list.  Don't worry
1357 			 * about the parent being unlocked, as any
1358 			 * setting is done with allprison_lock held.
1359 			 */
1360 			for (ij = 0; ij < ppr->pr_ip4s; ij++)
1361 				if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
1362 					break;
1363 			if (ij == ppr->pr_ip4s) {
1364 				error = EPERM;
1365 				goto done_deref_locked;
1366 			}
1367 			if (ip4s > 1) {
1368 				for (ii = ij = 1; ii < ip4s; ii++) {
1369 					if (ip4[ii].s_addr ==
1370 					    ppr->pr_ip4[0].s_addr)
1371 						continue;
1372 					for (; ij < ppr->pr_ip4s; ij++)
1373 						if (ip4[ii].s_addr ==
1374 						    ppr->pr_ip4[ij].s_addr)
1375 							break;
1376 					if (ij == ppr->pr_ip4s)
1377 						break;
1378 				}
1379 				if (ij == ppr->pr_ip4s) {
1380 					error = EPERM;
1381 					goto done_deref_locked;
1382 				}
1383 			}
1384 		}
1385 		/*
1386 		 * Check for conflicting IP addresses.  We permit them
1387 		 * if there is no more than one IP on each jail.  If
1388 		 * there is a duplicate on a jail with more than one
1389 		 * IP stop checking and return error.
1390 		 */
1391 #ifdef VIMAGE
1392 		for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
1393 			if (tppr->pr_flags & PR_VNET)
1394 				break;
1395 #else
1396 		tppr = &prison0;
1397 #endif
1398 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1399 			if (tpr == pr ||
1400 #ifdef VIMAGE
1401 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1402 #endif
1403 			    tpr->pr_uref == 0) {
1404 				descend = 0;
1405 				continue;
1406 			}
1407 			if (!(tpr->pr_flags & PR_IP4_USER))
1408 				continue;
1409 			descend = 0;
1410 			if (tpr->pr_ip4 == NULL ||
1411 			    (ip4s == 1 && tpr->pr_ip4s == 1))
1412 				continue;
1413 			for (ii = 0; ii < ip4s; ii++) {
1414 				if (prison_check_ip4_locked(tpr, &ip4[ii]) ==
1415 				    0) {
1416 					error = EADDRINUSE;
1417 					vfs_opterror(opts,
1418 					    "IPv4 addresses clash");
1419 					goto done_deref_locked;
1420 				}
1421 			}
1422 		}
1423 	}
1424 #endif
1425 #ifdef INET6
1426 	if (ip6s > 0) {
1427 		if (ppr->pr_flags & PR_IP6) {
1428 			/*
1429 			 * Make sure the new set of IP addresses is a
1430 			 * subset of the parent's list.
1431 			 */
1432 			for (ij = 0; ij < ppr->pr_ip6s; ij++)
1433 				if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1434 				    &ppr->pr_ip6[ij]))
1435 					break;
1436 			if (ij == ppr->pr_ip6s) {
1437 				error = EPERM;
1438 				goto done_deref_locked;
1439 			}
1440 			if (ip6s > 1) {
1441 				for (ii = ij = 1; ii < ip6s; ii++) {
1442 					if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1443 					     &ppr->pr_ip6[0]))
1444 						continue;
1445 					for (; ij < ppr->pr_ip6s; ij++)
1446 						if (IN6_ARE_ADDR_EQUAL(
1447 						    &ip6[ii], &ppr->pr_ip6[ij]))
1448 							break;
1449 					if (ij == ppr->pr_ip6s)
1450 						break;
1451 				}
1452 				if (ij == ppr->pr_ip6s) {
1453 					error = EPERM;
1454 					goto done_deref_locked;
1455 				}
1456 			}
1457 		}
1458 		/* Check for conflicting IP addresses. */
1459 #ifdef VIMAGE
1460 		for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
1461 			if (tppr->pr_flags & PR_VNET)
1462 				break;
1463 #else
1464 		tppr = &prison0;
1465 #endif
1466 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1467 			if (tpr == pr ||
1468 #ifdef VIMAGE
1469 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1470 #endif
1471 			    tpr->pr_uref == 0) {
1472 				descend = 0;
1473 				continue;
1474 			}
1475 			if (!(tpr->pr_flags & PR_IP6_USER))
1476 				continue;
1477 			descend = 0;
1478 			if (tpr->pr_ip6 == NULL ||
1479 			    (ip6s == 1 && tpr->pr_ip6s == 1))
1480 				continue;
1481 			for (ii = 0; ii < ip6s; ii++) {
1482 				if (prison_check_ip6_locked(tpr, &ip6[ii]) ==
1483 				    0) {
1484 					error = EADDRINUSE;
1485 					vfs_opterror(opts,
1486 					    "IPv6 addresses clash");
1487 					goto done_deref_locked;
1488 				}
1489 			}
1490 		}
1491 	}
1492 #endif
1493 	onamelen = namelen = 0;
1494 	if (namelc != NULL) {
1495 		/* Give a default name of the jid.  Also allow the name to be
1496 		 * explicitly the jid - but not any other number, and only in
1497 		 * normal form (no leading zero/etc).
1498 		 */
1499 		if (namelc[0] == '\0')
1500 			snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1501 		else if ((strtoul(namelc, &p, 10) != jid ||
1502 			  namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1503 			error = EINVAL;
1504 			vfs_opterror(opts,
1505 			    "name cannot be numeric (unless it is the jid)");
1506 			goto done_deref_locked;
1507 		}
1508 		/*
1509 		 * Make sure the name isn't too long for the prison or its
1510 		 * children.
1511 		 */
1512 		pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1513 		onamelen = strlen(pr->pr_name + pnamelen);
1514 		namelen = strlen(namelc);
1515 		if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1516 			error = ENAMETOOLONG;
1517 			goto done_deref_locked;
1518 		}
1519 		FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1520 			if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1521 			    sizeof(pr->pr_name)) {
1522 				error = ENAMETOOLONG;
1523 				goto done_deref_locked;
1524 			}
1525 		}
1526 	}
1527 	pr_allow_diff = pr_allow & ~ppr->pr_allow;
1528 	if (pr_allow_diff & ~PR_ALLOW_DIFFERENCES) {
1529 		error = EPERM;
1530 		goto done_deref_locked;
1531 	}
1532 
1533 	/*
1534 	 * Let modules check their parameters.  This requires unlocking and
1535 	 * then re-locking the prison, but this is still a valid state as long
1536 	 * as allprison_lock remains xlocked.
1537 	 */
1538 	mtx_unlock(&pr->pr_mtx);
1539 	error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1540 	if (error != 0) {
1541 		prison_deref(pr, created
1542 		    ? PD_LIST_XLOCKED
1543 		    : PD_DEREF | PD_LIST_XLOCKED);
1544 		goto done_releroot;
1545 	}
1546 	mtx_lock(&pr->pr_mtx);
1547 
1548 	/* At this point, all valid parameters should have been noted. */
1549 	TAILQ_FOREACH(opt, opts, link) {
1550 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
1551 			error = EINVAL;
1552 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
1553 			goto done_deref_locked;
1554 		}
1555 	}
1556 
1557 	/* Set the parameters of the prison. */
1558 #ifdef INET
1559 	redo_ip4 = 0;
1560 	if (pr_flags & PR_IP4_USER) {
1561 		pr->pr_flags |= PR_IP4;
1562 		free(pr->pr_ip4, M_PRISON);
1563 		pr->pr_ip4s = ip4s;
1564 		pr->pr_ip4 = ip4;
1565 		ip4 = NULL;
1566 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1567 #ifdef VIMAGE
1568 			if (tpr->pr_flags & PR_VNET) {
1569 				descend = 0;
1570 				continue;
1571 			}
1572 #endif
1573 			if (prison_restrict_ip4(tpr, NULL)) {
1574 				redo_ip4 = 1;
1575 				descend = 0;
1576 			}
1577 		}
1578 	}
1579 #endif
1580 #ifdef INET6
1581 	redo_ip6 = 0;
1582 	if (pr_flags & PR_IP6_USER) {
1583 		pr->pr_flags |= PR_IP6;
1584 		free(pr->pr_ip6, M_PRISON);
1585 		pr->pr_ip6s = ip6s;
1586 		pr->pr_ip6 = ip6;
1587 		ip6 = NULL;
1588 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1589 #ifdef VIMAGE
1590 			if (tpr->pr_flags & PR_VNET) {
1591 				descend = 0;
1592 				continue;
1593 			}
1594 #endif
1595 			if (prison_restrict_ip6(tpr, NULL)) {
1596 				redo_ip6 = 1;
1597 				descend = 0;
1598 			}
1599 		}
1600 	}
1601 #endif
1602 	if (gotslevel) {
1603 		pr->pr_securelevel = slevel;
1604 		/* Set all child jails to be at least this level. */
1605 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1606 			if (tpr->pr_securelevel < slevel)
1607 				tpr->pr_securelevel = slevel;
1608 	}
1609 	if (gotchildmax) {
1610 		pr->pr_childmax = childmax;
1611 		/* Set all child jails to under this limit. */
1612 		FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1613 			if (tpr->pr_childmax > childmax - level)
1614 				tpr->pr_childmax = childmax > level
1615 				    ? childmax - level : 0;
1616 	}
1617 	if (gotenforce) {
1618 		pr->pr_enforce_statfs = enforce;
1619 		/* Pass this restriction on to the children. */
1620 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1621 			if (tpr->pr_enforce_statfs < enforce)
1622 				tpr->pr_enforce_statfs = enforce;
1623 	}
1624 	if (gotrsnum) {
1625 		pr->pr_devfs_rsnum = rsnum;
1626 		/* Pass this restriction on to the children. */
1627 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1628 			tpr->pr_devfs_rsnum = rsnum;
1629 	}
1630 	if (namelc != NULL) {
1631 		if (ppr == &prison0)
1632 			strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
1633 		else
1634 			snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1635 			    ppr->pr_name, namelc);
1636 		/* Change this component of child names. */
1637 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1638 			bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1639 			    strlen(tpr->pr_name + onamelen) + 1);
1640 			bcopy(pr->pr_name, tpr->pr_name, namelen);
1641 		}
1642 	}
1643 	if (path != NULL) {
1644 		/* Try to keep a real-rooted full pathname. */
1645 		strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1646 		pr->pr_root = root;
1647 	}
1648 	if (PR_HOST & ch_flags & ~pr_flags) {
1649 		if (pr->pr_flags & PR_HOST) {
1650 			/*
1651 			 * Copy the parent's host info.  As with pr_ip4 above,
1652 			 * the lack of a lock on the parent is not a problem;
1653 			 * it is always set with allprison_lock at least
1654 			 * shared, and is held exclusively here.
1655 			 */
1656 			strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1657 			    sizeof(pr->pr_hostname));
1658 			strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1659 			    sizeof(pr->pr_domainname));
1660 			strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1661 			    sizeof(pr->pr_hostuuid));
1662 			pr->pr_hostid = pr->pr_parent->pr_hostid;
1663 		}
1664 	} else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1665 		/* Set this prison, and any descendants without PR_HOST. */
1666 		if (host != NULL)
1667 			strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1668 		if (domain != NULL)
1669 			strlcpy(pr->pr_domainname, domain,
1670 			    sizeof(pr->pr_domainname));
1671 		if (uuid != NULL)
1672 			strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1673 		if (gothid)
1674 			pr->pr_hostid = hid;
1675 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1676 			if (tpr->pr_flags & PR_HOST)
1677 				descend = 0;
1678 			else {
1679 				if (host != NULL)
1680 					strlcpy(tpr->pr_hostname,
1681 					    pr->pr_hostname,
1682 					    sizeof(tpr->pr_hostname));
1683 				if (domain != NULL)
1684 					strlcpy(tpr->pr_domainname,
1685 					    pr->pr_domainname,
1686 					    sizeof(tpr->pr_domainname));
1687 				if (uuid != NULL)
1688 					strlcpy(tpr->pr_hostuuid,
1689 					    pr->pr_hostuuid,
1690 					    sizeof(tpr->pr_hostuuid));
1691 				if (gothid)
1692 					tpr->pr_hostid = hid;
1693 			}
1694 		}
1695 	}
1696 	if ((tallow = ch_allow & ~pr_allow)) {
1697 		/* Clear allow bits in all children. */
1698 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1699 			tpr->pr_allow &= ~tallow;
1700 	}
1701 	pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1702 	/*
1703 	 * Persistent prisons get an extra reference, and prisons losing their
1704 	 * persist flag lose that reference.  Only do this for existing prisons
1705 	 * for now, so new ones will remain unseen until after the module
1706 	 * handlers have completed.
1707 	 */
1708 	born = pr->pr_uref == 0;
1709 	if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) {
1710 		if (pr_flags & PR_PERSIST) {
1711 			pr->pr_ref++;
1712 			pr->pr_uref++;
1713 		} else {
1714 			pr->pr_ref--;
1715 			pr->pr_uref--;
1716 		}
1717 	}
1718 	pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1719 	mtx_unlock(&pr->pr_mtx);
1720 
1721 #ifdef RACCT
1722 	if (racct_enable && created)
1723 		prison_racct_attach(pr);
1724 #endif
1725 
1726 	/* Locks may have prevented a complete restriction of child IP
1727 	 * addresses.  If so, allocate some more memory and try again.
1728 	 */
1729 #ifdef INET
1730 	while (redo_ip4) {
1731 		ip4s = pr->pr_ip4s;
1732 		ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1733 		mtx_lock(&pr->pr_mtx);
1734 		redo_ip4 = 0;
1735 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1736 #ifdef VIMAGE
1737 			if (tpr->pr_flags & PR_VNET) {
1738 				descend = 0;
1739 				continue;
1740 			}
1741 #endif
1742 			if (prison_restrict_ip4(tpr, ip4)) {
1743 				if (ip4 != NULL)
1744 					ip4 = NULL;
1745 				else
1746 					redo_ip4 = 1;
1747 			}
1748 		}
1749 		mtx_unlock(&pr->pr_mtx);
1750 	}
1751 #endif
1752 #ifdef INET6
1753 	while (redo_ip6) {
1754 		ip6s = pr->pr_ip6s;
1755 		ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1756 		mtx_lock(&pr->pr_mtx);
1757 		redo_ip6 = 0;
1758 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1759 #ifdef VIMAGE
1760 			if (tpr->pr_flags & PR_VNET) {
1761 				descend = 0;
1762 				continue;
1763 			}
1764 #endif
1765 			if (prison_restrict_ip6(tpr, ip6)) {
1766 				if (ip6 != NULL)
1767 					ip6 = NULL;
1768 				else
1769 					redo_ip6 = 1;
1770 			}
1771 		}
1772 		mtx_unlock(&pr->pr_mtx);
1773 	}
1774 #endif
1775 
1776 	/* Let the modules do their work. */
1777 	sx_downgrade(&allprison_lock);
1778 	if (born) {
1779 		error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1780 		if (error) {
1781 			(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1782 			prison_deref(pr, created
1783 			    ? PD_LIST_SLOCKED
1784 			    : PD_DEREF | PD_LIST_SLOCKED);
1785 			goto done_errmsg;
1786 		}
1787 	}
1788 	error = osd_jail_call(pr, PR_METHOD_SET, opts);
1789 	if (error) {
1790 		if (born)
1791 			(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1792 		prison_deref(pr, created
1793 		    ? PD_LIST_SLOCKED
1794 		    : PD_DEREF | PD_LIST_SLOCKED);
1795 		goto done_errmsg;
1796 	}
1797 
1798 	/* Attach this process to the prison if requested. */
1799 	if (flags & JAIL_ATTACH) {
1800 		mtx_lock(&pr->pr_mtx);
1801 		error = do_jail_attach(td, pr);
1802 		if (error) {
1803 			vfs_opterror(opts, "attach failed");
1804 			if (!created)
1805 				prison_deref(pr, PD_DEREF);
1806 			goto done_errmsg;
1807 		}
1808 	}
1809 
1810 #ifdef RACCT
1811 	if (racct_enable && !created) {
1812 		if (!(flags & JAIL_ATTACH))
1813 			sx_sunlock(&allprison_lock);
1814 		prison_racct_modify(pr);
1815 		if (!(flags & JAIL_ATTACH))
1816 			sx_slock(&allprison_lock);
1817 	}
1818 #endif
1819 
1820 	td->td_retval[0] = pr->pr_id;
1821 
1822 	/*
1823 	 * Now that it is all there, drop the temporary reference from existing
1824 	 * prisons.  Or add a reference to newly created persistent prisons
1825 	 * (which was not done earlier so that the prison would not be publicly
1826 	 * visible).
1827 	 */
1828 	if (!created) {
1829 		prison_deref(pr, (flags & JAIL_ATTACH)
1830 		    ? PD_DEREF
1831 		    : PD_DEREF | PD_LIST_SLOCKED);
1832 	} else {
1833 		if (pr_flags & PR_PERSIST) {
1834 			mtx_lock(&pr->pr_mtx);
1835 			pr->pr_ref++;
1836 			pr->pr_uref++;
1837 			mtx_unlock(&pr->pr_mtx);
1838 		}
1839 		if (!(flags & JAIL_ATTACH))
1840 			sx_sunlock(&allprison_lock);
1841 	}
1842 
1843 	goto done_free;
1844 
1845  done_deref_locked:
1846 	prison_deref(pr, created
1847 	    ? PD_LOCKED | PD_LIST_XLOCKED
1848 	    : PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
1849 	goto done_releroot;
1850  done_unlock_list:
1851 	sx_xunlock(&allprison_lock);
1852  done_releroot:
1853 	if (root != NULL)
1854 		vrele(root);
1855  done_errmsg:
1856 	if (error) {
1857 		if (vfs_getopt(opts, "errmsg", (void **)&errmsg,
1858 		    &errmsg_len) == 0 && errmsg_len > 0) {
1859 			errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1860 			if (optuio->uio_segflg == UIO_SYSSPACE)
1861 				bcopy(errmsg,
1862 				    optuio->uio_iov[errmsg_pos].iov_base,
1863 				    errmsg_len);
1864 			else
1865 				copyout(errmsg,
1866 				    optuio->uio_iov[errmsg_pos].iov_base,
1867 				    errmsg_len);
1868 		}
1869 	}
1870  done_free:
1871 #ifdef INET
1872 	free(ip4, M_PRISON);
1873 #endif
1874 #ifdef INET6
1875 	free(ip6, M_PRISON);
1876 #endif
1877 	if (g_path != NULL)
1878 		free(g_path, M_TEMP);
1879 	vfs_freeopts(opts);
1880 	return (error);
1881 }
1882 
1883 /*
1884  * struct jail_get_args {
1885  *	struct iovec *iovp;
1886  *	unsigned int iovcnt;
1887  *	int flags;
1888  * };
1889  */
1890 int
1891 sys_jail_get(struct thread *td, struct jail_get_args *uap)
1892 {
1893 	struct uio *auio;
1894 	int error;
1895 
1896 	/* Check that we have an even number of iovecs. */
1897 	if (uap->iovcnt & 1)
1898 		return (EINVAL);
1899 
1900 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
1901 	if (error)
1902 		return (error);
1903 	error = kern_jail_get(td, auio, uap->flags);
1904 	if (error == 0)
1905 		error = copyout(auio->uio_iov, uap->iovp,
1906 		    uap->iovcnt * sizeof (struct iovec));
1907 	free(auio, M_IOV);
1908 	return (error);
1909 }
1910 
1911 int
1912 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
1913 {
1914 	struct bool_flags *bf;
1915 	struct jailsys_flags *jsf;
1916 	struct prison *pr, *mypr;
1917 	struct vfsopt *opt;
1918 	struct vfsoptlist *opts;
1919 	char *errmsg, *name;
1920 	int error, errmsg_len, errmsg_pos, i, jid, len, locked, pos;
1921 	unsigned f;
1922 
1923 	if (flags & ~JAIL_GET_MASK)
1924 		return (EINVAL);
1925 
1926 	/* Get the parameter list. */
1927 	error = vfs_buildopts(optuio, &opts);
1928 	if (error)
1929 		return (error);
1930 	errmsg_pos = vfs_getopt_pos(opts, "errmsg");
1931 	mypr = td->td_ucred->cr_prison;
1932 
1933 	/*
1934 	 * Find the prison specified by one of: lastjid, jid, name.
1935 	 */
1936 	sx_slock(&allprison_lock);
1937 	error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
1938 	if (error == 0) {
1939 		TAILQ_FOREACH(pr, &allprison, pr_list) {
1940 			if (pr->pr_id > jid && prison_ischild(mypr, pr)) {
1941 				mtx_lock(&pr->pr_mtx);
1942 				if (pr->pr_ref > 0 &&
1943 				    (pr->pr_uref > 0 || (flags & JAIL_DYING)))
1944 					break;
1945 				mtx_unlock(&pr->pr_mtx);
1946 			}
1947 		}
1948 		if (pr != NULL)
1949 			goto found_prison;
1950 		error = ENOENT;
1951 		vfs_opterror(opts, "no jail after %d", jid);
1952 		goto done_unlock_list;
1953 	} else if (error != ENOENT)
1954 		goto done_unlock_list;
1955 
1956 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1957 	if (error == 0) {
1958 		if (jid != 0) {
1959 			pr = prison_find_child(mypr, jid);
1960 			if (pr != NULL) {
1961 				if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1962 					mtx_unlock(&pr->pr_mtx);
1963 					error = ENOENT;
1964 					vfs_opterror(opts, "jail %d is dying",
1965 					    jid);
1966 					goto done_unlock_list;
1967 				}
1968 				goto found_prison;
1969 			}
1970 			error = ENOENT;
1971 			vfs_opterror(opts, "jail %d not found", jid);
1972 			goto done_unlock_list;
1973 		}
1974 	} else if (error != ENOENT)
1975 		goto done_unlock_list;
1976 
1977 	error = vfs_getopt(opts, "name", (void **)&name, &len);
1978 	if (error == 0) {
1979 		if (len == 0 || name[len - 1] != '\0') {
1980 			error = EINVAL;
1981 			goto done_unlock_list;
1982 		}
1983 		pr = prison_find_name(mypr, name);
1984 		if (pr != NULL) {
1985 			if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1986 				mtx_unlock(&pr->pr_mtx);
1987 				error = ENOENT;
1988 				vfs_opterror(opts, "jail \"%s\" is dying",
1989 				    name);
1990 				goto done_unlock_list;
1991 			}
1992 			goto found_prison;
1993 		}
1994 		error = ENOENT;
1995 		vfs_opterror(opts, "jail \"%s\" not found", name);
1996 		goto done_unlock_list;
1997 	} else if (error != ENOENT)
1998 		goto done_unlock_list;
1999 
2000 	vfs_opterror(opts, "no jail specified");
2001 	error = ENOENT;
2002 	goto done_unlock_list;
2003 
2004  found_prison:
2005 	/* Get the parameters of the prison. */
2006 	pr->pr_ref++;
2007 	locked = PD_LOCKED;
2008 	td->td_retval[0] = pr->pr_id;
2009 	error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2010 	if (error != 0 && error != ENOENT)
2011 		goto done_deref;
2012 	i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2013 	error = vfs_setopt(opts, "parent", &i, sizeof(i));
2014 	if (error != 0 && error != ENOENT)
2015 		goto done_deref;
2016 	error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2017 	if (error != 0 && error != ENOENT)
2018 		goto done_deref;
2019 	error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2020 	    sizeof(pr->pr_cpuset->cs_id));
2021 	if (error != 0 && error != ENOENT)
2022 		goto done_deref;
2023 	error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2024 	if (error != 0 && error != ENOENT)
2025 		goto done_deref;
2026 #ifdef INET
2027 	error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
2028 	    pr->pr_ip4s * sizeof(*pr->pr_ip4));
2029 	if (error != 0 && error != ENOENT)
2030 		goto done_deref;
2031 #endif
2032 #ifdef INET6
2033 	error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
2034 	    pr->pr_ip6s * sizeof(*pr->pr_ip6));
2035 	if (error != 0 && error != ENOENT)
2036 		goto done_deref;
2037 #endif
2038 	error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2039 	    sizeof(pr->pr_securelevel));
2040 	if (error != 0 && error != ENOENT)
2041 		goto done_deref;
2042 	error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2043 	    sizeof(pr->pr_childcount));
2044 	if (error != 0 && error != ENOENT)
2045 		goto done_deref;
2046 	error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2047 	    sizeof(pr->pr_childmax));
2048 	if (error != 0 && error != ENOENT)
2049 		goto done_deref;
2050 	error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2051 	if (error != 0 && error != ENOENT)
2052 		goto done_deref;
2053 	error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2054 	if (error != 0 && error != ENOENT)
2055 		goto done_deref;
2056 	error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2057 	if (error != 0 && error != ENOENT)
2058 		goto done_deref;
2059 #ifdef COMPAT_FREEBSD32
2060 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2061 		uint32_t hid32 = pr->pr_hostid;
2062 
2063 		error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2064 	} else
2065 #endif
2066 	error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2067 	    sizeof(pr->pr_hostid));
2068 	if (error != 0 && error != ENOENT)
2069 		goto done_deref;
2070 	error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2071 	    sizeof(pr->pr_enforce_statfs));
2072 	if (error != 0 && error != ENOENT)
2073 		goto done_deref;
2074 	error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2075 	    sizeof(pr->pr_devfs_rsnum));
2076 	if (error != 0 && error != ENOENT)
2077 		goto done_deref;
2078 	for (bf = pr_flag_bool;
2079 	     bf < pr_flag_bool + nitems(pr_flag_bool);
2080 	     bf++) {
2081 		i = (pr->pr_flags & bf->flag) ? 1 : 0;
2082 		error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2083 		if (error != 0 && error != ENOENT)
2084 			goto done_deref;
2085 		i = !i;
2086 		error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2087 		if (error != 0 && error != ENOENT)
2088 			goto done_deref;
2089 	}
2090 	for (jsf = pr_flag_jailsys;
2091 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
2092 	     jsf++) {
2093 		f = pr->pr_flags & (jsf->disable | jsf->new);
2094 		i = (f != 0 && f == jsf->disable) ? JAIL_SYS_DISABLE
2095 		    : (f == jsf->new) ? JAIL_SYS_NEW
2096 		    : JAIL_SYS_INHERIT;
2097 		error = vfs_setopt(opts, jsf->name, &i, sizeof(i));
2098 		if (error != 0 && error != ENOENT)
2099 			goto done_deref;
2100 	}
2101 	for (bf = pr_flag_allow;
2102 	     bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
2103 	     bf++) {
2104 		i = (pr->pr_allow & bf->flag) ? 1 : 0;
2105 		error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2106 		if (error != 0 && error != ENOENT)
2107 			goto done_deref;
2108 		i = !i;
2109 		error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2110 		if (error != 0 && error != ENOENT)
2111 			goto done_deref;
2112 	}
2113 	i = (pr->pr_uref == 0);
2114 	error = vfs_setopt(opts, "dying", &i, sizeof(i));
2115 	if (error != 0 && error != ENOENT)
2116 		goto done_deref;
2117 	i = !i;
2118 	error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2119 	if (error != 0 && error != ENOENT)
2120 		goto done_deref;
2121 	error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2122 	    sizeof(pr->pr_osreldate));
2123 	if (error != 0 && error != ENOENT)
2124 		goto done_deref;
2125 	error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2126 	if (error != 0 && error != ENOENT)
2127 		goto done_deref;
2128 
2129 	/* Get the module parameters. */
2130 	mtx_unlock(&pr->pr_mtx);
2131 	locked = 0;
2132 	error = osd_jail_call(pr, PR_METHOD_GET, opts);
2133 	if (error)
2134 		goto done_deref;
2135 	prison_deref(pr, PD_DEREF | PD_LIST_SLOCKED);
2136 
2137 	/* By now, all parameters should have been noted. */
2138 	TAILQ_FOREACH(opt, opts, link) {
2139 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
2140 			error = EINVAL;
2141 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
2142 			goto done_errmsg;
2143 		}
2144 	}
2145 
2146 	/* Write the fetched parameters back to userspace. */
2147 	error = 0;
2148 	TAILQ_FOREACH(opt, opts, link) {
2149 		if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2150 			pos = 2 * opt->pos + 1;
2151 			optuio->uio_iov[pos].iov_len = opt->len;
2152 			if (opt->value != NULL) {
2153 				if (optuio->uio_segflg == UIO_SYSSPACE) {
2154 					bcopy(opt->value,
2155 					    optuio->uio_iov[pos].iov_base,
2156 					    opt->len);
2157 				} else {
2158 					error = copyout(opt->value,
2159 					    optuio->uio_iov[pos].iov_base,
2160 					    opt->len);
2161 					if (error)
2162 						break;
2163 				}
2164 			}
2165 		}
2166 	}
2167 	goto done_errmsg;
2168 
2169  done_deref:
2170 	prison_deref(pr, locked | PD_DEREF | PD_LIST_SLOCKED);
2171 	goto done_errmsg;
2172 
2173  done_unlock_list:
2174 	sx_sunlock(&allprison_lock);
2175  done_errmsg:
2176 	if (error && errmsg_pos >= 0) {
2177 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2178 		errmsg_pos = 2 * errmsg_pos + 1;
2179 		if (errmsg_len > 0) {
2180 			if (optuio->uio_segflg == UIO_SYSSPACE)
2181 				bcopy(errmsg,
2182 				    optuio->uio_iov[errmsg_pos].iov_base,
2183 				    errmsg_len);
2184 			else
2185 				copyout(errmsg,
2186 				    optuio->uio_iov[errmsg_pos].iov_base,
2187 				    errmsg_len);
2188 		}
2189 	}
2190 	vfs_freeopts(opts);
2191 	return (error);
2192 }
2193 
2194 /*
2195  * struct jail_remove_args {
2196  *	int jid;
2197  * };
2198  */
2199 int
2200 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2201 {
2202 	struct prison *pr, *cpr, *lpr, *tpr;
2203 	int descend, error;
2204 
2205 	error = priv_check(td, PRIV_JAIL_REMOVE);
2206 	if (error)
2207 		return (error);
2208 
2209 	sx_xlock(&allprison_lock);
2210 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2211 	if (pr == NULL) {
2212 		sx_xunlock(&allprison_lock);
2213 		return (EINVAL);
2214 	}
2215 
2216 	/* Remove all descendants of this prison, then remove this prison. */
2217 	pr->pr_ref++;
2218 	if (!LIST_EMPTY(&pr->pr_children)) {
2219 		mtx_unlock(&pr->pr_mtx);
2220 		lpr = NULL;
2221 		FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
2222 			mtx_lock(&cpr->pr_mtx);
2223 			if (cpr->pr_ref > 0) {
2224 				tpr = cpr;
2225 				cpr->pr_ref++;
2226 			} else {
2227 				/* Already removed - do not do it again. */
2228 				tpr = NULL;
2229 			}
2230 			mtx_unlock(&cpr->pr_mtx);
2231 			if (lpr != NULL) {
2232 				mtx_lock(&lpr->pr_mtx);
2233 				prison_remove_one(lpr);
2234 				sx_xlock(&allprison_lock);
2235 			}
2236 			lpr = tpr;
2237 		}
2238 		if (lpr != NULL) {
2239 			mtx_lock(&lpr->pr_mtx);
2240 			prison_remove_one(lpr);
2241 			sx_xlock(&allprison_lock);
2242 		}
2243 		mtx_lock(&pr->pr_mtx);
2244 	}
2245 	prison_remove_one(pr);
2246 	return (0);
2247 }
2248 
2249 static void
2250 prison_remove_one(struct prison *pr)
2251 {
2252 	struct proc *p;
2253 	int deuref;
2254 
2255 	/* If the prison was persistent, it is not anymore. */
2256 	deuref = 0;
2257 	if (pr->pr_flags & PR_PERSIST) {
2258 		pr->pr_ref--;
2259 		deuref = PD_DEUREF;
2260 		pr->pr_flags &= ~PR_PERSIST;
2261 	}
2262 
2263 	/*
2264 	 * jail_remove added a reference.  If that's the only one, remove
2265 	 * the prison now.
2266 	 */
2267 	KASSERT(pr->pr_ref > 0,
2268 	    ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id));
2269 	if (pr->pr_ref == 1) {
2270 		prison_deref(pr,
2271 		    deuref | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
2272 		return;
2273 	}
2274 
2275 	mtx_unlock(&pr->pr_mtx);
2276 	sx_xunlock(&allprison_lock);
2277 	/*
2278 	 * Kill all processes unfortunate enough to be attached to this prison.
2279 	 */
2280 	sx_slock(&allproc_lock);
2281 	FOREACH_PROC_IN_SYSTEM(p) {
2282 		PROC_LOCK(p);
2283 		if (p->p_state != PRS_NEW && p->p_ucred &&
2284 		    p->p_ucred->cr_prison == pr)
2285 			kern_psignal(p, SIGKILL);
2286 		PROC_UNLOCK(p);
2287 	}
2288 	sx_sunlock(&allproc_lock);
2289 	/* Remove the temporary reference added by jail_remove. */
2290 	prison_deref(pr, deuref | PD_DEREF);
2291 }
2292 
2293 /*
2294  * struct jail_attach_args {
2295  *	int jid;
2296  * };
2297  */
2298 int
2299 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2300 {
2301 	struct prison *pr;
2302 	int error;
2303 
2304 	error = priv_check(td, PRIV_JAIL_ATTACH);
2305 	if (error)
2306 		return (error);
2307 
2308 	/*
2309 	 * Start with exclusive hold on allprison_lock to ensure that a possible
2310 	 * PR_METHOD_REMOVE call isn't concurrent with jail_set or jail_remove.
2311 	 * But then immediately downgrade it since we don't need to stop
2312 	 * readers.
2313 	 */
2314 	sx_xlock(&allprison_lock);
2315 	sx_downgrade(&allprison_lock);
2316 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2317 	if (pr == NULL) {
2318 		sx_sunlock(&allprison_lock);
2319 		return (EINVAL);
2320 	}
2321 
2322 	/*
2323 	 * Do not allow a process to attach to a prison that is not
2324 	 * considered to be "alive".
2325 	 */
2326 	if (pr->pr_uref == 0) {
2327 		mtx_unlock(&pr->pr_mtx);
2328 		sx_sunlock(&allprison_lock);
2329 		return (EINVAL);
2330 	}
2331 
2332 	return (do_jail_attach(td, pr));
2333 }
2334 
2335 static int
2336 do_jail_attach(struct thread *td, struct prison *pr)
2337 {
2338 	struct proc *p;
2339 	struct ucred *newcred, *oldcred;
2340 	int error;
2341 
2342 	/*
2343 	 * XXX: Note that there is a slight race here if two threads
2344 	 * in the same privileged process attempt to attach to two
2345 	 * different jails at the same time.  It is important for
2346 	 * user processes not to do this, or they might end up with
2347 	 * a process root from one prison, but attached to the jail
2348 	 * of another.
2349 	 */
2350 	pr->pr_ref++;
2351 	pr->pr_uref++;
2352 	mtx_unlock(&pr->pr_mtx);
2353 
2354 	/* Let modules do whatever they need to prepare for attaching. */
2355 	error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2356 	if (error) {
2357 		prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2358 		return (error);
2359 	}
2360 	sx_sunlock(&allprison_lock);
2361 
2362 	/*
2363 	 * Reparent the newly attached process to this jail.
2364 	 */
2365 	p = td->td_proc;
2366 	error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2367 	if (error)
2368 		goto e_revert_osd;
2369 
2370 	vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2371 	if ((error = change_dir(pr->pr_root, td)) != 0)
2372 		goto e_unlock;
2373 #ifdef MAC
2374 	if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2375 		goto e_unlock;
2376 #endif
2377 	VOP_UNLOCK(pr->pr_root);
2378 	if ((error = pwd_chroot(td, pr->pr_root)))
2379 		goto e_revert_osd;
2380 
2381 	newcred = crget();
2382 	PROC_LOCK(p);
2383 	oldcred = crcopysafe(p, newcred);
2384 	newcred->cr_prison = pr;
2385 	proc_set_cred(p, newcred);
2386 	setsugid(p);
2387 #ifdef RACCT
2388 	racct_proc_ucred_changed(p, oldcred, newcred);
2389 	crhold(newcred);
2390 #endif
2391 	PROC_UNLOCK(p);
2392 #ifdef RCTL
2393 	rctl_proc_ucred_changed(p, newcred);
2394 	crfree(newcred);
2395 #endif
2396 	prison_deref(oldcred->cr_prison, PD_DEREF | PD_DEUREF);
2397 	crfree(oldcred);
2398 	return (0);
2399 
2400  e_unlock:
2401 	VOP_UNLOCK(pr->pr_root);
2402  e_revert_osd:
2403 	/* Tell modules this thread is still in its old jail after all. */
2404 	(void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
2405 	prison_deref(pr, PD_DEREF | PD_DEUREF);
2406 	return (error);
2407 }
2408 
2409 /*
2410  * Returns a locked prison instance, or NULL on failure.
2411  */
2412 struct prison *
2413 prison_find(int prid)
2414 {
2415 	struct prison *pr;
2416 
2417 	sx_assert(&allprison_lock, SX_LOCKED);
2418 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2419 		if (pr->pr_id == prid) {
2420 			mtx_lock(&pr->pr_mtx);
2421 			if (pr->pr_ref > 0)
2422 				return (pr);
2423 			mtx_unlock(&pr->pr_mtx);
2424 		}
2425 	}
2426 	return (NULL);
2427 }
2428 
2429 /*
2430  * Find a prison that is a descendant of mypr.  Returns a locked prison or NULL.
2431  */
2432 struct prison *
2433 prison_find_child(struct prison *mypr, int prid)
2434 {
2435 	struct prison *pr;
2436 	int descend;
2437 
2438 	sx_assert(&allprison_lock, SX_LOCKED);
2439 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2440 		if (pr->pr_id == prid) {
2441 			mtx_lock(&pr->pr_mtx);
2442 			if (pr->pr_ref > 0)
2443 				return (pr);
2444 			mtx_unlock(&pr->pr_mtx);
2445 		}
2446 	}
2447 	return (NULL);
2448 }
2449 
2450 /*
2451  * Look for the name relative to mypr.  Returns a locked prison or NULL.
2452  */
2453 struct prison *
2454 prison_find_name(struct prison *mypr, const char *name)
2455 {
2456 	struct prison *pr, *deadpr;
2457 	size_t mylen;
2458 	int descend;
2459 
2460 	sx_assert(&allprison_lock, SX_LOCKED);
2461 	mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2462  again:
2463 	deadpr = NULL;
2464 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2465 		if (!strcmp(pr->pr_name + mylen, name)) {
2466 			mtx_lock(&pr->pr_mtx);
2467 			if (pr->pr_ref > 0) {
2468 				if (pr->pr_uref > 0)
2469 					return (pr);
2470 				deadpr = pr;
2471 			}
2472 			mtx_unlock(&pr->pr_mtx);
2473 		}
2474 	}
2475 	/* There was no valid prison - perhaps there was a dying one. */
2476 	if (deadpr != NULL) {
2477 		mtx_lock(&deadpr->pr_mtx);
2478 		if (deadpr->pr_ref == 0) {
2479 			mtx_unlock(&deadpr->pr_mtx);
2480 			goto again;
2481 		}
2482 	}
2483 	return (deadpr);
2484 }
2485 
2486 /*
2487  * See if a prison has the specific flag set.
2488  */
2489 int
2490 prison_flag(struct ucred *cred, unsigned flag)
2491 {
2492 
2493 	/* This is an atomic read, so no locking is necessary. */
2494 	return (cred->cr_prison->pr_flags & flag);
2495 }
2496 
2497 int
2498 prison_allow(struct ucred *cred, unsigned flag)
2499 {
2500 
2501 	/* This is an atomic read, so no locking is necessary. */
2502 	return (cred->cr_prison->pr_allow & flag);
2503 }
2504 
2505 /*
2506  * Remove a prison reference.  If that was the last reference, remove the
2507  * prison itself - but not in this context in case there are locks held.
2508  */
2509 void
2510 prison_free_locked(struct prison *pr)
2511 {
2512 	int ref;
2513 
2514 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2515 	ref = --pr->pr_ref;
2516 	mtx_unlock(&pr->pr_mtx);
2517 	if (ref == 0)
2518 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2519 }
2520 
2521 void
2522 prison_free(struct prison *pr)
2523 {
2524 
2525 	mtx_lock(&pr->pr_mtx);
2526 	prison_free_locked(pr);
2527 }
2528 
2529 /*
2530  * Complete a call to either prison_free or prison_proc_free.
2531  */
2532 static void
2533 prison_complete(void *context, int pending)
2534 {
2535 	struct prison *pr = context;
2536 
2537 	sx_xlock(&allprison_lock);
2538 	mtx_lock(&pr->pr_mtx);
2539 	prison_deref(pr, pr->pr_uref
2540 	    ? PD_DEREF | PD_DEUREF | PD_LOCKED | PD_LIST_XLOCKED
2541 	    : PD_LOCKED | PD_LIST_XLOCKED);
2542 }
2543 
2544 /*
2545  * Remove a prison reference (usually).  This internal version assumes no
2546  * mutexes are held, except perhaps the prison itself.  If there are no more
2547  * references, release and delist the prison.  On completion, the prison lock
2548  * and the allprison lock are both unlocked.
2549  */
2550 static void
2551 prison_deref(struct prison *pr, int flags)
2552 {
2553 	struct prison *ppr, *tpr;
2554 	int ref, lasturef;
2555 
2556 	if (!(flags & PD_LOCKED))
2557 		mtx_lock(&pr->pr_mtx);
2558 	for (;;) {
2559 		if (flags & PD_DEUREF) {
2560 			KASSERT(pr->pr_uref > 0,
2561 			    ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
2562 			     pr->pr_id));
2563 			pr->pr_uref--;
2564 			lasturef = pr->pr_uref == 0;
2565 			if (lasturef)
2566 				pr->pr_ref++;
2567 			KASSERT(prison0.pr_uref != 0, ("prison0 pr_uref=0"));
2568 		} else
2569 			lasturef = 0;
2570 		if (flags & PD_DEREF) {
2571 			KASSERT(pr->pr_ref > 0,
2572 			    ("prison_deref PD_DEREF on a dead prison (jid=%d)",
2573 			     pr->pr_id));
2574 			pr->pr_ref--;
2575 		}
2576 		ref = pr->pr_ref;
2577 		mtx_unlock(&pr->pr_mtx);
2578 
2579 		/*
2580 		 * Tell the modules if the last user reference was removed
2581 		 * (even it sticks around in dying state).
2582 		 */
2583 		if (lasturef) {
2584 			if (!(flags & (PD_LIST_SLOCKED | PD_LIST_XLOCKED))) {
2585 				sx_xlock(&allprison_lock);
2586 				flags |= PD_LIST_XLOCKED;
2587 			}
2588 			(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
2589 			mtx_lock(&pr->pr_mtx);
2590 			ref = --pr->pr_ref;
2591 			mtx_unlock(&pr->pr_mtx);
2592 		}
2593 
2594 		/* If the prison still has references, nothing else to do. */
2595 		if (ref > 0) {
2596 			if (flags & PD_LIST_SLOCKED)
2597 				sx_sunlock(&allprison_lock);
2598 			else if (flags & PD_LIST_XLOCKED)
2599 				sx_xunlock(&allprison_lock);
2600 			return;
2601 		}
2602 
2603 		if (flags & PD_LIST_SLOCKED) {
2604 			if (!sx_try_upgrade(&allprison_lock)) {
2605 				sx_sunlock(&allprison_lock);
2606 				sx_xlock(&allprison_lock);
2607 			}
2608 		} else if (!(flags & PD_LIST_XLOCKED))
2609 			sx_xlock(&allprison_lock);
2610 
2611 		TAILQ_REMOVE(&allprison, pr, pr_list);
2612 		LIST_REMOVE(pr, pr_sibling);
2613 		ppr = pr->pr_parent;
2614 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
2615 			tpr->pr_childcount--;
2616 		sx_xunlock(&allprison_lock);
2617 
2618 #ifdef VIMAGE
2619 		if (pr->pr_vnet != ppr->pr_vnet)
2620 			vnet_destroy(pr->pr_vnet);
2621 #endif
2622 		if (pr->pr_root != NULL)
2623 			vrele(pr->pr_root);
2624 		mtx_destroy(&pr->pr_mtx);
2625 #ifdef INET
2626 		free(pr->pr_ip4, M_PRISON);
2627 #endif
2628 #ifdef INET6
2629 		free(pr->pr_ip6, M_PRISON);
2630 #endif
2631 		if (pr->pr_cpuset != NULL)
2632 			cpuset_rel(pr->pr_cpuset);
2633 		osd_jail_exit(pr);
2634 #ifdef RACCT
2635 		if (racct_enable)
2636 			prison_racct_detach(pr);
2637 #endif
2638 		free(pr, M_PRISON);
2639 
2640 		/* Removing a prison frees a reference on its parent. */
2641 		pr = ppr;
2642 		mtx_lock(&pr->pr_mtx);
2643 		flags = PD_DEREF | PD_DEUREF;
2644 	}
2645 }
2646 
2647 void
2648 prison_hold_locked(struct prison *pr)
2649 {
2650 
2651 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2652 	KASSERT(pr->pr_ref > 0,
2653 	    ("Trying to hold dead prison %p (jid=%d).", pr, pr->pr_id));
2654 	pr->pr_ref++;
2655 }
2656 
2657 void
2658 prison_hold(struct prison *pr)
2659 {
2660 
2661 	mtx_lock(&pr->pr_mtx);
2662 	prison_hold_locked(pr);
2663 	mtx_unlock(&pr->pr_mtx);
2664 }
2665 
2666 void
2667 prison_proc_hold(struct prison *pr)
2668 {
2669 
2670 	mtx_lock(&pr->pr_mtx);
2671 	KASSERT(pr->pr_uref > 0,
2672 	    ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2673 	pr->pr_uref++;
2674 	mtx_unlock(&pr->pr_mtx);
2675 }
2676 
2677 void
2678 prison_proc_free(struct prison *pr)
2679 {
2680 
2681 	mtx_lock(&pr->pr_mtx);
2682 	KASSERT(pr->pr_uref > 0,
2683 	    ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2684 	if (pr->pr_uref > 1)
2685 		pr->pr_uref--;
2686 	else {
2687 		/*
2688 		 * Don't remove the last user reference in this context, which
2689 		 * is expected to be a process that is not only locked, but
2690 		 * also half dead.
2691 		 */
2692 		pr->pr_ref++;
2693 		mtx_unlock(&pr->pr_mtx);
2694 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2695 		return;
2696 	}
2697 	mtx_unlock(&pr->pr_mtx);
2698 }
2699 
2700 /*
2701  * Check if a jail supports the given address family.
2702  *
2703  * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
2704  * if not.
2705  */
2706 int
2707 prison_check_af(struct ucred *cred, int af)
2708 {
2709 	struct prison *pr;
2710 	int error;
2711 
2712 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2713 
2714 	pr = cred->cr_prison;
2715 #ifdef VIMAGE
2716 	/* Prisons with their own network stack are not limited. */
2717 	if (prison_owns_vnet(cred))
2718 		return (0);
2719 #endif
2720 
2721 	error = 0;
2722 	switch (af)
2723 	{
2724 #ifdef INET
2725 	case AF_INET:
2726 		if (pr->pr_flags & PR_IP4)
2727 		{
2728 			mtx_lock(&pr->pr_mtx);
2729 			if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL)
2730 				error = EAFNOSUPPORT;
2731 			mtx_unlock(&pr->pr_mtx);
2732 		}
2733 		break;
2734 #endif
2735 #ifdef INET6
2736 	case AF_INET6:
2737 		if (pr->pr_flags & PR_IP6)
2738 		{
2739 			mtx_lock(&pr->pr_mtx);
2740 			if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL)
2741 				error = EAFNOSUPPORT;
2742 			mtx_unlock(&pr->pr_mtx);
2743 		}
2744 		break;
2745 #endif
2746 	case AF_LOCAL:
2747 	case AF_ROUTE:
2748 		break;
2749 	default:
2750 		if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
2751 			error = EAFNOSUPPORT;
2752 	}
2753 	return (error);
2754 }
2755 
2756 /*
2757  * Check if given address belongs to the jail referenced by cred (wrapper to
2758  * prison_check_ip[46]).
2759  *
2760  * Returns 0 if jail doesn't restrict the address family or if address belongs
2761  * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
2762  * the jail doesn't allow the address family.  IPv4 Address passed in in NBO.
2763  */
2764 int
2765 prison_if(struct ucred *cred, const struct sockaddr *sa)
2766 {
2767 #ifdef INET
2768 	const struct sockaddr_in *sai;
2769 #endif
2770 #ifdef INET6
2771 	const struct sockaddr_in6 *sai6;
2772 #endif
2773 	int error;
2774 
2775 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2776 	KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
2777 
2778 #ifdef VIMAGE
2779 	if (prison_owns_vnet(cred))
2780 		return (0);
2781 #endif
2782 
2783 	error = 0;
2784 	switch (sa->sa_family)
2785 	{
2786 #ifdef INET
2787 	case AF_INET:
2788 		sai = (const struct sockaddr_in *)sa;
2789 		error = prison_check_ip4(cred, &sai->sin_addr);
2790 		break;
2791 #endif
2792 #ifdef INET6
2793 	case AF_INET6:
2794 		sai6 = (const struct sockaddr_in6 *)sa;
2795 		error = prison_check_ip6(cred, &sai6->sin6_addr);
2796 		break;
2797 #endif
2798 	default:
2799 		if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
2800 			error = EAFNOSUPPORT;
2801 	}
2802 	return (error);
2803 }
2804 
2805 /*
2806  * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
2807  */
2808 int
2809 prison_check(struct ucred *cred1, struct ucred *cred2)
2810 {
2811 
2812 	return ((cred1->cr_prison == cred2->cr_prison ||
2813 	    prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
2814 }
2815 
2816 /*
2817  * Return 1 if p2 is a child of p1, otherwise 0.
2818  */
2819 int
2820 prison_ischild(struct prison *pr1, struct prison *pr2)
2821 {
2822 
2823 	for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
2824 		if (pr1 == pr2)
2825 			return (1);
2826 	return (0);
2827 }
2828 
2829 /*
2830  * Return 1 if the passed credential is in a jail and that jail does not
2831  * have its own virtual network stack, otherwise 0.
2832  */
2833 int
2834 jailed_without_vnet(struct ucred *cred)
2835 {
2836 
2837 	if (!jailed(cred))
2838 		return (0);
2839 #ifdef VIMAGE
2840 	if (prison_owns_vnet(cred))
2841 		return (0);
2842 #endif
2843 
2844 	return (1);
2845 }
2846 
2847 /*
2848  * Return the correct hostname (domainname, et al) for the passed credential.
2849  */
2850 void
2851 getcredhostname(struct ucred *cred, char *buf, size_t size)
2852 {
2853 	struct prison *pr;
2854 
2855 	/*
2856 	 * A NULL credential can be used to shortcut to the physical
2857 	 * system's hostname.
2858 	 */
2859 	pr = (cred != NULL) ? cred->cr_prison : &prison0;
2860 	mtx_lock(&pr->pr_mtx);
2861 	strlcpy(buf, pr->pr_hostname, size);
2862 	mtx_unlock(&pr->pr_mtx);
2863 }
2864 
2865 void
2866 getcreddomainname(struct ucred *cred, char *buf, size_t size)
2867 {
2868 
2869 	mtx_lock(&cred->cr_prison->pr_mtx);
2870 	strlcpy(buf, cred->cr_prison->pr_domainname, size);
2871 	mtx_unlock(&cred->cr_prison->pr_mtx);
2872 }
2873 
2874 void
2875 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
2876 {
2877 
2878 	mtx_lock(&cred->cr_prison->pr_mtx);
2879 	strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
2880 	mtx_unlock(&cred->cr_prison->pr_mtx);
2881 }
2882 
2883 void
2884 getcredhostid(struct ucred *cred, unsigned long *hostid)
2885 {
2886 
2887 	mtx_lock(&cred->cr_prison->pr_mtx);
2888 	*hostid = cred->cr_prison->pr_hostid;
2889 	mtx_unlock(&cred->cr_prison->pr_mtx);
2890 }
2891 
2892 #ifdef VIMAGE
2893 /*
2894  * Determine whether the prison represented by cred owns
2895  * its vnet rather than having it inherited.
2896  *
2897  * Returns 1 in case the prison owns the vnet, 0 otherwise.
2898  */
2899 int
2900 prison_owns_vnet(struct ucred *cred)
2901 {
2902 
2903 	/*
2904 	 * vnets cannot be added/removed after jail creation,
2905 	 * so no need to lock here.
2906 	 */
2907 	return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
2908 }
2909 #endif
2910 
2911 /*
2912  * Determine whether the subject represented by cred can "see"
2913  * status of a mount point.
2914  * Returns: 0 for permitted, ENOENT otherwise.
2915  * XXX: This function should be called cr_canseemount() and should be
2916  *      placed in kern_prot.c.
2917  */
2918 int
2919 prison_canseemount(struct ucred *cred, struct mount *mp)
2920 {
2921 	struct prison *pr;
2922 	struct statfs *sp;
2923 	size_t len;
2924 
2925 	pr = cred->cr_prison;
2926 	if (pr->pr_enforce_statfs == 0)
2927 		return (0);
2928 	if (pr->pr_root->v_mount == mp)
2929 		return (0);
2930 	if (pr->pr_enforce_statfs == 2)
2931 		return (ENOENT);
2932 	/*
2933 	 * If jail's chroot directory is set to "/" we should be able to see
2934 	 * all mount-points from inside a jail.
2935 	 * This is ugly check, but this is the only situation when jail's
2936 	 * directory ends with '/'.
2937 	 */
2938 	if (strcmp(pr->pr_path, "/") == 0)
2939 		return (0);
2940 	len = strlen(pr->pr_path);
2941 	sp = &mp->mnt_stat;
2942 	if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
2943 		return (ENOENT);
2944 	/*
2945 	 * Be sure that we don't have situation where jail's root directory
2946 	 * is "/some/path" and mount point is "/some/pathpath".
2947 	 */
2948 	if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
2949 		return (ENOENT);
2950 	return (0);
2951 }
2952 
2953 void
2954 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
2955 {
2956 	char jpath[MAXPATHLEN];
2957 	struct prison *pr;
2958 	size_t len;
2959 
2960 	pr = cred->cr_prison;
2961 	if (pr->pr_enforce_statfs == 0)
2962 		return;
2963 	if (prison_canseemount(cred, mp) != 0) {
2964 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
2965 		strlcpy(sp->f_mntonname, "[restricted]",
2966 		    sizeof(sp->f_mntonname));
2967 		return;
2968 	}
2969 	if (pr->pr_root->v_mount == mp) {
2970 		/*
2971 		 * Clear current buffer data, so we are sure nothing from
2972 		 * the valid path left there.
2973 		 */
2974 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
2975 		*sp->f_mntonname = '/';
2976 		return;
2977 	}
2978 	/*
2979 	 * If jail's chroot directory is set to "/" we should be able to see
2980 	 * all mount-points from inside a jail.
2981 	 */
2982 	if (strcmp(pr->pr_path, "/") == 0)
2983 		return;
2984 	len = strlen(pr->pr_path);
2985 	strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
2986 	/*
2987 	 * Clear current buffer data, so we are sure nothing from
2988 	 * the valid path left there.
2989 	 */
2990 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
2991 	if (*jpath == '\0') {
2992 		/* Should never happen. */
2993 		*sp->f_mntonname = '/';
2994 	} else {
2995 		strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
2996 	}
2997 }
2998 
2999 /*
3000  * Check with permission for a specific privilege is granted within jail.  We
3001  * have a specific list of accepted privileges; the rest are denied.
3002  */
3003 int
3004 prison_priv_check(struct ucred *cred, int priv)
3005 {
3006 
3007 	/*
3008 	 * Some policies have custom handlers. This routine should not be
3009 	 * called for them. See priv_check_cred().
3010 	 */
3011 	switch (priv) {
3012 	case PRIV_VFS_GENERATION:
3013 		KASSERT(0, ("prison_priv_check instead of a custom handler "
3014 		    "called for %d\n", priv));
3015 	}
3016 
3017 	if (!jailed(cred))
3018 		return (0);
3019 
3020 #ifdef VIMAGE
3021 	/*
3022 	 * Privileges specific to prisons with a virtual network stack.
3023 	 * There might be a duplicate entry here in case the privilege
3024 	 * is only granted conditionally in the legacy jail case.
3025 	 */
3026 	switch (priv) {
3027 #ifdef notyet
3028 		/*
3029 		 * NFS-specific privileges.
3030 		 */
3031 	case PRIV_NFS_DAEMON:
3032 	case PRIV_NFS_LOCKD:
3033 #endif
3034 		/*
3035 		 * Network stack privileges.
3036 		 */
3037 	case PRIV_NET_BRIDGE:
3038 	case PRIV_NET_GRE:
3039 	case PRIV_NET_BPF:
3040 	case PRIV_NET_RAW:		/* Dup, cond. in legacy jail case. */
3041 	case PRIV_NET_ROUTE:
3042 	case PRIV_NET_TAP:
3043 	case PRIV_NET_SETIFMTU:
3044 	case PRIV_NET_SETIFFLAGS:
3045 	case PRIV_NET_SETIFCAP:
3046 	case PRIV_NET_SETIFDESCR:
3047 	case PRIV_NET_SETIFNAME	:
3048 	case PRIV_NET_SETIFMETRIC:
3049 	case PRIV_NET_SETIFPHYS:
3050 	case PRIV_NET_SETIFMAC:
3051 	case PRIV_NET_SETLANPCP:
3052 	case PRIV_NET_ADDMULTI:
3053 	case PRIV_NET_DELMULTI:
3054 	case PRIV_NET_HWIOCTL:
3055 	case PRIV_NET_SETLLADDR:
3056 	case PRIV_NET_ADDIFGROUP:
3057 	case PRIV_NET_DELIFGROUP:
3058 	case PRIV_NET_IFCREATE:
3059 	case PRIV_NET_IFDESTROY:
3060 	case PRIV_NET_ADDIFADDR:
3061 	case PRIV_NET_DELIFADDR:
3062 	case PRIV_NET_LAGG:
3063 	case PRIV_NET_GIF:
3064 	case PRIV_NET_SETIFVNET:
3065 	case PRIV_NET_SETIFFIB:
3066 
3067 		/*
3068 		 * 802.11-related privileges.
3069 		 */
3070 	case PRIV_NET80211_GETKEY:
3071 #ifdef notyet
3072 	case PRIV_NET80211_MANAGE:		/* XXX-BZ discuss with sam@ */
3073 #endif
3074 
3075 #ifdef notyet
3076 		/*
3077 		 * ATM privileges.
3078 		 */
3079 	case PRIV_NETATM_CFG:
3080 	case PRIV_NETATM_ADD:
3081 	case PRIV_NETATM_DEL:
3082 	case PRIV_NETATM_SET:
3083 
3084 		/*
3085 		 * Bluetooth privileges.
3086 		 */
3087 	case PRIV_NETBLUETOOTH_RAW:
3088 #endif
3089 
3090 		/*
3091 		 * Netgraph and netgraph module privileges.
3092 		 */
3093 	case PRIV_NETGRAPH_CONTROL:
3094 #ifdef notyet
3095 	case PRIV_NETGRAPH_TTY:
3096 #endif
3097 
3098 		/*
3099 		 * IPv4 and IPv6 privileges.
3100 		 */
3101 	case PRIV_NETINET_IPFW:
3102 	case PRIV_NETINET_DIVERT:
3103 	case PRIV_NETINET_PF:
3104 	case PRIV_NETINET_DUMMYNET:
3105 	case PRIV_NETINET_CARP:
3106 	case PRIV_NETINET_MROUTE:
3107 	case PRIV_NETINET_RAW:
3108 	case PRIV_NETINET_ADDRCTRL6:
3109 	case PRIV_NETINET_ND6:
3110 	case PRIV_NETINET_SCOPE6:
3111 	case PRIV_NETINET_ALIFETIME6:
3112 	case PRIV_NETINET_IPSEC:
3113 	case PRIV_NETINET_BINDANY:
3114 
3115 #ifdef notyet
3116 		/*
3117 		 * NCP privileges.
3118 		 */
3119 	case PRIV_NETNCP:
3120 
3121 		/*
3122 		 * SMB privileges.
3123 		 */
3124 	case PRIV_NETSMB:
3125 #endif
3126 
3127 	/*
3128 	 * No default: or deny here.
3129 	 * In case of no permit fall through to next switch().
3130 	 */
3131 		if (cred->cr_prison->pr_flags & PR_VNET)
3132 			return (0);
3133 	}
3134 #endif /* VIMAGE */
3135 
3136 	switch (priv) {
3137 
3138 		/*
3139 		 * Allow ktrace privileges for root in jail.
3140 		 */
3141 	case PRIV_KTRACE:
3142 
3143 #if 0
3144 		/*
3145 		 * Allow jailed processes to configure audit identity and
3146 		 * submit audit records (login, etc).  In the future we may
3147 		 * want to further refine the relationship between audit and
3148 		 * jail.
3149 		 */
3150 	case PRIV_AUDIT_GETAUDIT:
3151 	case PRIV_AUDIT_SETAUDIT:
3152 	case PRIV_AUDIT_SUBMIT:
3153 #endif
3154 
3155 		/*
3156 		 * Allow jailed processes to manipulate process UNIX
3157 		 * credentials in any way they see fit.
3158 		 */
3159 	case PRIV_CRED_SETUID:
3160 	case PRIV_CRED_SETEUID:
3161 	case PRIV_CRED_SETGID:
3162 	case PRIV_CRED_SETEGID:
3163 	case PRIV_CRED_SETGROUPS:
3164 	case PRIV_CRED_SETREUID:
3165 	case PRIV_CRED_SETREGID:
3166 	case PRIV_CRED_SETRESUID:
3167 	case PRIV_CRED_SETRESGID:
3168 
3169 		/*
3170 		 * Jail implements visibility constraints already, so allow
3171 		 * jailed root to override uid/gid-based constraints.
3172 		 */
3173 	case PRIV_SEEOTHERGIDS:
3174 	case PRIV_SEEOTHERUIDS:
3175 
3176 		/*
3177 		 * Jail implements inter-process debugging limits already, so
3178 		 * allow jailed root various debugging privileges.
3179 		 */
3180 	case PRIV_DEBUG_DIFFCRED:
3181 	case PRIV_DEBUG_SUGID:
3182 	case PRIV_DEBUG_UNPRIV:
3183 
3184 		/*
3185 		 * Allow jail to set various resource limits and login
3186 		 * properties, and for now, exceed process resource limits.
3187 		 */
3188 	case PRIV_PROC_LIMIT:
3189 	case PRIV_PROC_SETLOGIN:
3190 	case PRIV_PROC_SETRLIMIT:
3191 
3192 		/*
3193 		 * System V and POSIX IPC privileges are granted in jail.
3194 		 */
3195 	case PRIV_IPC_READ:
3196 	case PRIV_IPC_WRITE:
3197 	case PRIV_IPC_ADMIN:
3198 	case PRIV_IPC_MSGSIZE:
3199 	case PRIV_MQ_ADMIN:
3200 
3201 		/*
3202 		 * Jail operations within a jail work on child jails.
3203 		 */
3204 	case PRIV_JAIL_ATTACH:
3205 	case PRIV_JAIL_SET:
3206 	case PRIV_JAIL_REMOVE:
3207 
3208 		/*
3209 		 * Jail implements its own inter-process limits, so allow
3210 		 * root processes in jail to change scheduling on other
3211 		 * processes in the same jail.  Likewise for signalling.
3212 		 */
3213 	case PRIV_SCHED_DIFFCRED:
3214 	case PRIV_SCHED_CPUSET:
3215 	case PRIV_SIGNAL_DIFFCRED:
3216 	case PRIV_SIGNAL_SUGID:
3217 
3218 		/*
3219 		 * Allow jailed processes to write to sysctls marked as jail
3220 		 * writable.
3221 		 */
3222 	case PRIV_SYSCTL_WRITEJAIL:
3223 
3224 		/*
3225 		 * Allow root in jail to manage a variety of quota
3226 		 * properties.  These should likely be conditional on a
3227 		 * configuration option.
3228 		 */
3229 	case PRIV_VFS_GETQUOTA:
3230 	case PRIV_VFS_SETQUOTA:
3231 
3232 		/*
3233 		 * Since Jail relies on chroot() to implement file system
3234 		 * protections, grant many VFS privileges to root in jail.
3235 		 * Be careful to exclude mount-related and NFS-related
3236 		 * privileges.
3237 		 */
3238 	case PRIV_VFS_READ:
3239 	case PRIV_VFS_WRITE:
3240 	case PRIV_VFS_ADMIN:
3241 	case PRIV_VFS_EXEC:
3242 	case PRIV_VFS_LOOKUP:
3243 	case PRIV_VFS_BLOCKRESERVE:	/* XXXRW: Slightly surprising. */
3244 	case PRIV_VFS_CHFLAGS_DEV:
3245 	case PRIV_VFS_CHOWN:
3246 	case PRIV_VFS_CHROOT:
3247 	case PRIV_VFS_RETAINSUGID:
3248 	case PRIV_VFS_FCHROOT:
3249 	case PRIV_VFS_LINK:
3250 	case PRIV_VFS_SETGID:
3251 	case PRIV_VFS_STAT:
3252 	case PRIV_VFS_STICKYFILE:
3253 
3254 		/*
3255 		 * As in the non-jail case, non-root users are expected to be
3256 		 * able to read kernel/phyiscal memory (provided /dev/[k]mem
3257 		 * exists in the jail and they have permission to access it).
3258 		 */
3259 	case PRIV_KMEM_READ:
3260 		return (0);
3261 
3262 		/*
3263 		 * Depending on the global setting, allow privilege of
3264 		 * setting system flags.
3265 		 */
3266 	case PRIV_VFS_SYSFLAGS:
3267 		if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
3268 			return (0);
3269 		else
3270 			return (EPERM);
3271 
3272 		/*
3273 		 * Depending on the global setting, allow privilege of
3274 		 * mounting/unmounting file systems.
3275 		 */
3276 	case PRIV_VFS_MOUNT:
3277 	case PRIV_VFS_UNMOUNT:
3278 	case PRIV_VFS_MOUNT_NONUSER:
3279 	case PRIV_VFS_MOUNT_OWNER:
3280 		if (cred->cr_prison->pr_allow & PR_ALLOW_MOUNT &&
3281 		    cred->cr_prison->pr_enforce_statfs < 2)
3282 			return (0);
3283 		else
3284 			return (EPERM);
3285 
3286 		/*
3287 		 * Conditionnaly allow locking (unlocking) physical pages
3288 		 * in memory.
3289 		 */
3290 	case PRIV_VM_MLOCK:
3291 	case PRIV_VM_MUNLOCK:
3292 		if (cred->cr_prison->pr_allow & PR_ALLOW_MLOCK)
3293 			return (0);
3294 		else
3295 			return (EPERM);
3296 
3297 		/*
3298 		 * Conditionally allow jailed root to bind reserved ports.
3299 		 */
3300 	case PRIV_NETINET_RESERVEDPORT:
3301 		if (cred->cr_prison->pr_allow & PR_ALLOW_RESERVED_PORTS)
3302 			return (0);
3303 		else
3304 			return (EPERM);
3305 
3306 		/*
3307 		 * Allow jailed root to reuse in-use ports.
3308 		 */
3309 	case PRIV_NETINET_REUSEPORT:
3310 		return (0);
3311 
3312 		/*
3313 		 * Allow jailed root to set certain IPv4/6 (option) headers.
3314 		 */
3315 	case PRIV_NETINET_SETHDROPTS:
3316 		return (0);
3317 
3318 		/*
3319 		 * Conditionally allow creating raw sockets in jail.
3320 		 */
3321 	case PRIV_NETINET_RAW:
3322 		if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
3323 			return (0);
3324 		else
3325 			return (EPERM);
3326 
3327 		/*
3328 		 * Since jail implements its own visibility limits on netstat
3329 		 * sysctls, allow getcred.  This allows identd to work in
3330 		 * jail.
3331 		 */
3332 	case PRIV_NETINET_GETCRED:
3333 		return (0);
3334 
3335 		/*
3336 		 * Allow jailed root to set loginclass.
3337 		 */
3338 	case PRIV_PROC_SETLOGINCLASS:
3339 		return (0);
3340 
3341 		/*
3342 		 * Do not allow a process inside a jail to read the kernel
3343 		 * message buffer unless explicitly permitted.
3344 		 */
3345 	case PRIV_MSGBUF:
3346 		if (cred->cr_prison->pr_allow & PR_ALLOW_READ_MSGBUF)
3347 			return (0);
3348 		return (EPERM);
3349 
3350 	default:
3351 		/*
3352 		 * In all remaining cases, deny the privilege request.  This
3353 		 * includes almost all network privileges, many system
3354 		 * configuration privileges.
3355 		 */
3356 		return (EPERM);
3357 	}
3358 }
3359 
3360 /*
3361  * Return the part of pr2's name that is relative to pr1, or the whole name
3362  * if it does not directly follow.
3363  */
3364 
3365 char *
3366 prison_name(struct prison *pr1, struct prison *pr2)
3367 {
3368 	char *name;
3369 
3370 	/* Jails see themselves as "0" (if they see themselves at all). */
3371 	if (pr1 == pr2)
3372 		return "0";
3373 	name = pr2->pr_name;
3374 	if (prison_ischild(pr1, pr2)) {
3375 		/*
3376 		 * pr1 isn't locked (and allprison_lock may not be either)
3377 		 * so its length can't be counted on.  But the number of dots
3378 		 * can be counted on - and counted.
3379 		 */
3380 		for (; pr1 != &prison0; pr1 = pr1->pr_parent)
3381 			name = strchr(name, '.') + 1;
3382 	}
3383 	return (name);
3384 }
3385 
3386 /*
3387  * Return the part of pr2's path that is relative to pr1, or the whole path
3388  * if it does not directly follow.
3389  */
3390 static char *
3391 prison_path(struct prison *pr1, struct prison *pr2)
3392 {
3393 	char *path1, *path2;
3394 	int len1;
3395 
3396 	path1 = pr1->pr_path;
3397 	path2 = pr2->pr_path;
3398 	if (!strcmp(path1, "/"))
3399 		return (path2);
3400 	len1 = strlen(path1);
3401 	if (strncmp(path1, path2, len1))
3402 		return (path2);
3403 	if (path2[len1] == '\0')
3404 		return "/";
3405 	if (path2[len1] == '/')
3406 		return (path2 + len1);
3407 	return (path2);
3408 }
3409 
3410 /*
3411  * Jail-related sysctls.
3412  */
3413 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
3414     "Jails");
3415 
3416 static int
3417 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
3418 {
3419 	struct xprison *xp;
3420 	struct prison *pr, *cpr;
3421 #ifdef INET
3422 	struct in_addr *ip4 = NULL;
3423 	int ip4s = 0;
3424 #endif
3425 #ifdef INET6
3426 	struct in6_addr *ip6 = NULL;
3427 	int ip6s = 0;
3428 #endif
3429 	int descend, error;
3430 
3431 	xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
3432 	pr = req->td->td_ucred->cr_prison;
3433 	error = 0;
3434 	sx_slock(&allprison_lock);
3435 	FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
3436 #if defined(INET) || defined(INET6)
3437  again:
3438 #endif
3439 		mtx_lock(&cpr->pr_mtx);
3440 #ifdef INET
3441 		if (cpr->pr_ip4s > 0) {
3442 			if (ip4s < cpr->pr_ip4s) {
3443 				ip4s = cpr->pr_ip4s;
3444 				mtx_unlock(&cpr->pr_mtx);
3445 				ip4 = realloc(ip4, ip4s *
3446 				    sizeof(struct in_addr), M_TEMP, M_WAITOK);
3447 				goto again;
3448 			}
3449 			bcopy(cpr->pr_ip4, ip4,
3450 			    cpr->pr_ip4s * sizeof(struct in_addr));
3451 		}
3452 #endif
3453 #ifdef INET6
3454 		if (cpr->pr_ip6s > 0) {
3455 			if (ip6s < cpr->pr_ip6s) {
3456 				ip6s = cpr->pr_ip6s;
3457 				mtx_unlock(&cpr->pr_mtx);
3458 				ip6 = realloc(ip6, ip6s *
3459 				    sizeof(struct in6_addr), M_TEMP, M_WAITOK);
3460 				goto again;
3461 			}
3462 			bcopy(cpr->pr_ip6, ip6,
3463 			    cpr->pr_ip6s * sizeof(struct in6_addr));
3464 		}
3465 #endif
3466 		if (cpr->pr_ref == 0) {
3467 			mtx_unlock(&cpr->pr_mtx);
3468 			continue;
3469 		}
3470 		bzero(xp, sizeof(*xp));
3471 		xp->pr_version = XPRISON_VERSION;
3472 		xp->pr_id = cpr->pr_id;
3473 		xp->pr_state = cpr->pr_uref > 0
3474 		    ? PRISON_STATE_ALIVE : PRISON_STATE_DYING;
3475 		strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
3476 		strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
3477 		strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
3478 #ifdef INET
3479 		xp->pr_ip4s = cpr->pr_ip4s;
3480 #endif
3481 #ifdef INET6
3482 		xp->pr_ip6s = cpr->pr_ip6s;
3483 #endif
3484 		mtx_unlock(&cpr->pr_mtx);
3485 		error = SYSCTL_OUT(req, xp, sizeof(*xp));
3486 		if (error)
3487 			break;
3488 #ifdef INET
3489 		if (xp->pr_ip4s > 0) {
3490 			error = SYSCTL_OUT(req, ip4,
3491 			    xp->pr_ip4s * sizeof(struct in_addr));
3492 			if (error)
3493 				break;
3494 		}
3495 #endif
3496 #ifdef INET6
3497 		if (xp->pr_ip6s > 0) {
3498 			error = SYSCTL_OUT(req, ip6,
3499 			    xp->pr_ip6s * sizeof(struct in6_addr));
3500 			if (error)
3501 				break;
3502 		}
3503 #endif
3504 	}
3505 	sx_sunlock(&allprison_lock);
3506 	free(xp, M_TEMP);
3507 #ifdef INET
3508 	free(ip4, M_TEMP);
3509 #endif
3510 #ifdef INET6
3511 	free(ip6, M_TEMP);
3512 #endif
3513 	return (error);
3514 }
3515 
3516 SYSCTL_OID(_security_jail, OID_AUTO, list,
3517     CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3518     sysctl_jail_list, "S", "List of active jails");
3519 
3520 static int
3521 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
3522 {
3523 	int error, injail;
3524 
3525 	injail = jailed(req->td->td_ucred);
3526 	error = SYSCTL_OUT(req, &injail, sizeof(injail));
3527 
3528 	return (error);
3529 }
3530 
3531 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
3532     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3533     sysctl_jail_jailed, "I", "Process in jail?");
3534 
3535 static int
3536 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
3537 {
3538 	int error, havevnet;
3539 #ifdef VIMAGE
3540 	struct ucred *cred = req->td->td_ucred;
3541 
3542 	havevnet = jailed(cred) && prison_owns_vnet(cred);
3543 #else
3544 	havevnet = 0;
3545 #endif
3546 	error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
3547 
3548 	return (error);
3549 }
3550 
3551 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
3552     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3553     sysctl_jail_vnet, "I", "Jail owns vnet?");
3554 
3555 #if defined(INET) || defined(INET6)
3556 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
3557     &jail_max_af_ips, 0,
3558     "Number of IP addresses a jail may have at most per address family (deprecated)");
3559 #endif
3560 
3561 /*
3562  * Default parameters for jail(2) compatibility.  For historical reasons,
3563  * the sysctl names have varying similarity to the parameter names.  Prisons
3564  * just see their own parameters, and can't change them.
3565  */
3566 static int
3567 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
3568 {
3569 	struct prison *pr;
3570 	int allow, error, i;
3571 
3572 	pr = req->td->td_ucred->cr_prison;
3573 	allow = (pr == &prison0) ? jail_default_allow : pr->pr_allow;
3574 
3575 	/* Get the current flag value, and convert it to a boolean. */
3576 	i = (allow & arg2) ? 1 : 0;
3577 	if (arg1 != NULL)
3578 		i = !i;
3579 	error = sysctl_handle_int(oidp, &i, 0, req);
3580 	if (error || !req->newptr)
3581 		return (error);
3582 	i = i ? arg2 : 0;
3583 	if (arg1 != NULL)
3584 		i ^= arg2;
3585 	/*
3586 	 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
3587 	 * for writing.
3588 	 */
3589 	mtx_lock(&prison0.pr_mtx);
3590 	jail_default_allow = (jail_default_allow & ~arg2) | i;
3591 	mtx_unlock(&prison0.pr_mtx);
3592 	return (0);
3593 }
3594 
3595 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
3596     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3597     NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
3598     "Processes in jail can set their hostnames (deprecated)");
3599 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
3600     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3601     (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
3602     "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
3603 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
3604     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3605     NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
3606     "Processes in jail can use System V IPC primitives (deprecated)");
3607 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
3608     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3609     NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
3610     "Prison root can create raw sockets (deprecated)");
3611 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
3612     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3613     NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
3614     "Processes in jail can alter system file flags (deprecated)");
3615 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
3616     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3617     NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
3618     "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
3619 
3620 static int
3621 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
3622 {
3623 	struct prison *pr;
3624 	int level, error;
3625 
3626 	pr = req->td->td_ucred->cr_prison;
3627 	level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
3628 	error = sysctl_handle_int(oidp, &level, 0, req);
3629 	if (error || !req->newptr)
3630 		return (error);
3631 	*(int *)arg1 = level;
3632 	return (0);
3633 }
3634 
3635 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
3636     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3637     &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
3638     sysctl_jail_default_level, "I",
3639     "Processes in jail cannot see all mounted file systems (deprecated)");
3640 
3641 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
3642     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
3643     &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
3644     sysctl_jail_default_level, "I",
3645     "Ruleset for the devfs filesystem in jail (deprecated)");
3646 
3647 /*
3648  * Nodes to describe jail parameters.  Maximum length of string parameters
3649  * is returned in the string itself, and the other parameters exist merely
3650  * to make themselves and their types known.
3651  */
3652 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
3653     "Jail parameters");
3654 
3655 int
3656 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
3657 {
3658 	int i;
3659 	long l;
3660 	size_t s;
3661 	char numbuf[12];
3662 
3663 	switch (oidp->oid_kind & CTLTYPE)
3664 	{
3665 	case CTLTYPE_LONG:
3666 	case CTLTYPE_ULONG:
3667 		l = 0;
3668 #ifdef SCTL_MASK32
3669 		if (!(req->flags & SCTL_MASK32))
3670 #endif
3671 			return (SYSCTL_OUT(req, &l, sizeof(l)));
3672 	case CTLTYPE_INT:
3673 	case CTLTYPE_UINT:
3674 		i = 0;
3675 		return (SYSCTL_OUT(req, &i, sizeof(i)));
3676 	case CTLTYPE_STRING:
3677 		snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
3678 		return
3679 		    (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
3680 	case CTLTYPE_STRUCT:
3681 		s = (size_t)arg2;
3682 		return (SYSCTL_OUT(req, &s, sizeof(s)));
3683 	}
3684 	return (0);
3685 }
3686 
3687 /*
3688  * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
3689  * jail creation time but cannot be changed in an existing jail.
3690  */
3691 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
3692 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
3693 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
3694 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
3695 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
3696     "I", "Jail secure level");
3697 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
3698     "Jail value for kern.osreldate and uname -K");
3699 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
3700     "Jail value for kern.osrelease and uname -r");
3701 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
3702     "I", "Jail cannot see all mounted file systems");
3703 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
3704     "I", "Ruleset for in-jail devfs mounts");
3705 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
3706     "B", "Jail persistence");
3707 #ifdef VIMAGE
3708 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
3709     "E,jailsys", "Virtual network stack");
3710 #endif
3711 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
3712     "B", "Jail is in the process of shutting down");
3713 
3714 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
3715 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
3716     "I", "Current number of child jails");
3717 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
3718     "I", "Maximum number of child jails");
3719 
3720 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
3721 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
3722     "Jail hostname");
3723 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
3724     "Jail NIS domainname");
3725 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
3726     "Jail host UUID");
3727 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
3728     "LU", "Jail host ID");
3729 
3730 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
3731 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
3732 
3733 #ifdef INET
3734 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
3735     "Jail IPv4 address virtualization");
3736 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
3737     "S,in_addr,a", "Jail IPv4 addresses");
3738 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
3739     "B", "Do (not) use IPv4 source address selection rather than the "
3740     "primary jail IPv4 address.");
3741 #endif
3742 #ifdef INET6
3743 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
3744     "Jail IPv6 address virtualization");
3745 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
3746     "S,in6_addr,a", "Jail IPv6 addresses");
3747 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
3748     "B", "Do (not) use IPv6 source address selection rather than the "
3749     "primary jail IPv6 address.");
3750 #endif
3751 
3752 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
3753 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
3754     "B", "Jail may set hostname");
3755 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
3756     "B", "Jail may use SYSV IPC");
3757 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
3758     "B", "Jail may create raw sockets");
3759 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
3760     "B", "Jail may alter system file flags");
3761 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
3762     "B", "Jail may set file quotas");
3763 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
3764     "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
3765 SYSCTL_JAIL_PARAM(_allow, mlock, CTLTYPE_INT | CTLFLAG_RW,
3766     "B", "Jail may lock (unlock) physical pages in memory");
3767 SYSCTL_JAIL_PARAM(_allow, reserved_ports, CTLTYPE_INT | CTLFLAG_RW,
3768     "B", "Jail may bind sockets to reserved ports");
3769 SYSCTL_JAIL_PARAM(_allow, read_msgbuf, CTLTYPE_INT | CTLFLAG_RW,
3770     "B", "Jail may read the kernel message buffer");
3771 SYSCTL_JAIL_PARAM(_allow, unprivileged_proc_debug, CTLTYPE_INT | CTLFLAG_RW,
3772     "B", "Unprivileged processes may use process debugging facilities");
3773 
3774 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
3775 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
3776     "B", "Jail may mount/unmount jail-friendly file systems in general");
3777 
3778 /*
3779  * Add a dynamic parameter allow.<name>, or allow.<prefix>.<name>.  Return
3780  * its associated bit in the pr_allow bitmask, or zero if the parameter was
3781  * not created.
3782  */
3783 unsigned
3784 prison_add_allow(const char *prefix, const char *name, const char *prefix_descr,
3785     const char *descr)
3786 {
3787 	struct bool_flags *bf;
3788 	struct sysctl_oid *parent;
3789 	char *allow_name, *allow_noname, *allowed;
3790 #ifndef NO_SYSCTL_DESCR
3791 	char *descr_deprecated;
3792 #endif
3793 	unsigned allow_flag;
3794 
3795 	if (prefix
3796 	    ? asprintf(&allow_name, M_PRISON, "allow.%s.%s", prefix, name)
3797 		< 0 ||
3798 	      asprintf(&allow_noname, M_PRISON, "allow.%s.no%s", prefix, name)
3799 		< 0
3800 	    : asprintf(&allow_name, M_PRISON, "allow.%s", name) < 0 ||
3801 	      asprintf(&allow_noname, M_PRISON, "allow.no%s", name) < 0) {
3802 		free(allow_name, M_PRISON);
3803 		return 0;
3804 	}
3805 
3806 	/*
3807 	 * See if this parameter has already beed added, i.e. a module was
3808 	 * previously loaded/unloaded.
3809 	 */
3810 	mtx_lock(&prison0.pr_mtx);
3811 	for (bf = pr_flag_allow;
3812 	     bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
3813 	     bf++) {
3814 		if (strcmp(bf->name, allow_name) == 0) {
3815 			allow_flag = bf->flag;
3816 			goto no_add;
3817 		}
3818 	}
3819 
3820 	/*
3821 	 * Find a free bit in prison0's pr_allow, failing if there are none
3822 	 * (which shouldn't happen as long as we keep track of how many
3823 	 * potential dynamic flags exist).
3824 	 *
3825 	 * Due to per-jail unprivileged process debugging support
3826 	 * using pr_allow, also verify against PR_ALLOW_ALL_STATIC.
3827 	 * prison0 may have unprivileged process debugging unset.
3828 	 */
3829 	for (allow_flag = 1;; allow_flag <<= 1) {
3830 		if (allow_flag == 0)
3831 			goto no_add;
3832 		if (allow_flag & PR_ALLOW_ALL_STATIC)
3833 			continue;
3834 		if ((prison0.pr_allow & allow_flag) == 0)
3835 			break;
3836 	}
3837 
3838 	/*
3839 	 * Note the parameter in the next open slot in pr_flag_allow.
3840 	 * Set the flag last so code that checks pr_flag_allow can do so
3841 	 * without locking.
3842 	 */
3843 	for (bf = pr_flag_allow; bf->flag != 0; bf++)
3844 		if (bf == pr_flag_allow + nitems(pr_flag_allow)) {
3845 			/* This should never happen, but is not fatal. */
3846 			allow_flag = 0;
3847 			goto no_add;
3848 		}
3849 	prison0.pr_allow |= allow_flag;
3850 	bf->name = allow_name;
3851 	bf->noname = allow_noname;
3852 	bf->flag = allow_flag;
3853 	mtx_unlock(&prison0.pr_mtx);
3854 
3855 	/*
3856 	 * Create sysctls for the paramter, and the back-compat global
3857 	 * permission.
3858 	 */
3859 	parent = prefix
3860 	    ? SYSCTL_ADD_NODE(NULL,
3861 		  SYSCTL_CHILDREN(&sysctl___security_jail_param_allow),
3862 		  OID_AUTO, prefix, CTLFLAG_MPSAFE, 0, prefix_descr)
3863 	    : &sysctl___security_jail_param_allow;
3864 	(void)SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(parent), OID_AUTO,
3865 	    name, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3866 	    NULL, 0, sysctl_jail_param, "B", descr);
3867 	if ((prefix
3868 	     ? asprintf(&allowed, M_TEMP, "%s_%s_allowed", prefix, name)
3869 	     : asprintf(&allowed, M_TEMP, "%s_allowed", name)) >= 0) {
3870 #ifndef NO_SYSCTL_DESCR
3871 		(void)asprintf(&descr_deprecated, M_TEMP, "%s (deprecated)",
3872 		    descr);
3873 #endif
3874 		(void)SYSCTL_ADD_PROC(NULL,
3875 		    SYSCTL_CHILDREN(&sysctl___security_jail), OID_AUTO, allowed,
3876 		    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, allow_flag,
3877 		    sysctl_jail_default_allow, "I", descr_deprecated);
3878 #ifndef NO_SYSCTL_DESCR
3879 		free(descr_deprecated, M_TEMP);
3880 #endif
3881 		free(allowed, M_TEMP);
3882 	}
3883 	return allow_flag;
3884 
3885  no_add:
3886 	mtx_unlock(&prison0.pr_mtx);
3887 	free(allow_name, M_PRISON);
3888 	free(allow_noname, M_PRISON);
3889 	return allow_flag;
3890 }
3891 
3892 /*
3893  * The VFS system will register jail-aware filesystems here.  They each get
3894  * a parameter allow.mount.xxxfs and a flag to check when a jailed user
3895  * attempts to mount.
3896  */
3897 void
3898 prison_add_vfs(struct vfsconf *vfsp)
3899 {
3900 #ifdef NO_SYSCTL_DESCR
3901 
3902 	vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
3903 	    NULL, NULL);
3904 #else
3905 	char *descr;
3906 
3907 	(void)asprintf(&descr, M_TEMP, "Jail may mount the %s file system",
3908 	    vfsp->vfc_name);
3909 	vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
3910 	    NULL, descr);
3911 	free(descr, M_TEMP);
3912 #endif
3913 }
3914 
3915 #ifdef RACCT
3916 void
3917 prison_racct_foreach(void (*callback)(struct racct *racct,
3918     void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
3919     void *arg2, void *arg3)
3920 {
3921 	struct prison_racct *prr;
3922 
3923 	ASSERT_RACCT_ENABLED();
3924 
3925 	sx_slock(&allprison_lock);
3926 	if (pre != NULL)
3927 		(pre)();
3928 	LIST_FOREACH(prr, &allprison_racct, prr_next)
3929 		(callback)(prr->prr_racct, arg2, arg3);
3930 	if (post != NULL)
3931 		(post)();
3932 	sx_sunlock(&allprison_lock);
3933 }
3934 
3935 static struct prison_racct *
3936 prison_racct_find_locked(const char *name)
3937 {
3938 	struct prison_racct *prr;
3939 
3940 	ASSERT_RACCT_ENABLED();
3941 	sx_assert(&allprison_lock, SA_XLOCKED);
3942 
3943 	if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
3944 		return (NULL);
3945 
3946 	LIST_FOREACH(prr, &allprison_racct, prr_next) {
3947 		if (strcmp(name, prr->prr_name) != 0)
3948 			continue;
3949 
3950 		/* Found prison_racct with a matching name? */
3951 		prison_racct_hold(prr);
3952 		return (prr);
3953 	}
3954 
3955 	/* Add new prison_racct. */
3956 	prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
3957 	racct_create(&prr->prr_racct);
3958 
3959 	strcpy(prr->prr_name, name);
3960 	refcount_init(&prr->prr_refcount, 1);
3961 	LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
3962 
3963 	return (prr);
3964 }
3965 
3966 struct prison_racct *
3967 prison_racct_find(const char *name)
3968 {
3969 	struct prison_racct *prr;
3970 
3971 	ASSERT_RACCT_ENABLED();
3972 
3973 	sx_xlock(&allprison_lock);
3974 	prr = prison_racct_find_locked(name);
3975 	sx_xunlock(&allprison_lock);
3976 	return (prr);
3977 }
3978 
3979 void
3980 prison_racct_hold(struct prison_racct *prr)
3981 {
3982 
3983 	ASSERT_RACCT_ENABLED();
3984 
3985 	refcount_acquire(&prr->prr_refcount);
3986 }
3987 
3988 static void
3989 prison_racct_free_locked(struct prison_racct *prr)
3990 {
3991 
3992 	ASSERT_RACCT_ENABLED();
3993 	sx_assert(&allprison_lock, SA_XLOCKED);
3994 
3995 	if (refcount_release(&prr->prr_refcount)) {
3996 		racct_destroy(&prr->prr_racct);
3997 		LIST_REMOVE(prr, prr_next);
3998 		free(prr, M_PRISON_RACCT);
3999 	}
4000 }
4001 
4002 void
4003 prison_racct_free(struct prison_racct *prr)
4004 {
4005 
4006 	ASSERT_RACCT_ENABLED();
4007 	sx_assert(&allprison_lock, SA_UNLOCKED);
4008 
4009 	if (refcount_release_if_not_last(&prr->prr_refcount))
4010 		return;
4011 
4012 	sx_xlock(&allprison_lock);
4013 	prison_racct_free_locked(prr);
4014 	sx_xunlock(&allprison_lock);
4015 }
4016 
4017 static void
4018 prison_racct_attach(struct prison *pr)
4019 {
4020 	struct prison_racct *prr;
4021 
4022 	ASSERT_RACCT_ENABLED();
4023 	sx_assert(&allprison_lock, SA_XLOCKED);
4024 
4025 	prr = prison_racct_find_locked(pr->pr_name);
4026 	KASSERT(prr != NULL, ("cannot find prison_racct"));
4027 
4028 	pr->pr_prison_racct = prr;
4029 }
4030 
4031 /*
4032  * Handle jail renaming.  From the racct point of view, renaming means
4033  * moving from one prison_racct to another.
4034  */
4035 static void
4036 prison_racct_modify(struct prison *pr)
4037 {
4038 #ifdef RCTL
4039 	struct proc *p;
4040 	struct ucred *cred;
4041 #endif
4042 	struct prison_racct *oldprr;
4043 
4044 	ASSERT_RACCT_ENABLED();
4045 
4046 	sx_slock(&allproc_lock);
4047 	sx_xlock(&allprison_lock);
4048 
4049 	if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4050 		sx_xunlock(&allprison_lock);
4051 		sx_sunlock(&allproc_lock);
4052 		return;
4053 	}
4054 
4055 	oldprr = pr->pr_prison_racct;
4056 	pr->pr_prison_racct = NULL;
4057 
4058 	prison_racct_attach(pr);
4059 
4060 	/*
4061 	 * Move resource utilisation records.
4062 	 */
4063 	racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
4064 
4065 #ifdef RCTL
4066 	/*
4067 	 * Force rctl to reattach rules to processes.
4068 	 */
4069 	FOREACH_PROC_IN_SYSTEM(p) {
4070 		PROC_LOCK(p);
4071 		cred = crhold(p->p_ucred);
4072 		PROC_UNLOCK(p);
4073 		rctl_proc_ucred_changed(p, cred);
4074 		crfree(cred);
4075 	}
4076 #endif
4077 
4078 	sx_sunlock(&allproc_lock);
4079 	prison_racct_free_locked(oldprr);
4080 	sx_xunlock(&allprison_lock);
4081 }
4082 
4083 static void
4084 prison_racct_detach(struct prison *pr)
4085 {
4086 
4087 	ASSERT_RACCT_ENABLED();
4088 	sx_assert(&allprison_lock, SA_UNLOCKED);
4089 
4090 	if (pr->pr_prison_racct == NULL)
4091 		return;
4092 	prison_racct_free(pr->pr_prison_racct);
4093 	pr->pr_prison_racct = NULL;
4094 }
4095 #endif /* RACCT */
4096 
4097 #ifdef DDB
4098 
4099 static void
4100 db_show_prison(struct prison *pr)
4101 {
4102 	struct bool_flags *bf;
4103 	struct jailsys_flags *jsf;
4104 #if defined(INET) || defined(INET6)
4105 	int ii;
4106 #endif
4107 	unsigned f;
4108 #ifdef INET
4109 	char ip4buf[INET_ADDRSTRLEN];
4110 #endif
4111 #ifdef INET6
4112 	char ip6buf[INET6_ADDRSTRLEN];
4113 #endif
4114 
4115 	db_printf("prison %p:\n", pr);
4116 	db_printf(" jid             = %d\n", pr->pr_id);
4117 	db_printf(" name            = %s\n", pr->pr_name);
4118 	db_printf(" parent          = %p\n", pr->pr_parent);
4119 	db_printf(" ref             = %d\n", pr->pr_ref);
4120 	db_printf(" uref            = %d\n", pr->pr_uref);
4121 	db_printf(" path            = %s\n", pr->pr_path);
4122 	db_printf(" cpuset          = %d\n", pr->pr_cpuset
4123 	    ? pr->pr_cpuset->cs_id : -1);
4124 #ifdef VIMAGE
4125 	db_printf(" vnet            = %p\n", pr->pr_vnet);
4126 #endif
4127 	db_printf(" root            = %p\n", pr->pr_root);
4128 	db_printf(" securelevel     = %d\n", pr->pr_securelevel);
4129 	db_printf(" devfs_rsnum     = %d\n", pr->pr_devfs_rsnum);
4130 	db_printf(" children.max    = %d\n", pr->pr_childmax);
4131 	db_printf(" children.cur    = %d\n", pr->pr_childcount);
4132 	db_printf(" child           = %p\n", LIST_FIRST(&pr->pr_children));
4133 	db_printf(" sibling         = %p\n", LIST_NEXT(pr, pr_sibling));
4134 	db_printf(" flags           = 0x%x", pr->pr_flags);
4135 	for (bf = pr_flag_bool; bf < pr_flag_bool + nitems(pr_flag_bool); bf++)
4136 		if (pr->pr_flags & bf->flag)
4137 			db_printf(" %s", bf->name);
4138 	for (jsf = pr_flag_jailsys;
4139 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
4140 	     jsf++) {
4141 		f = pr->pr_flags & (jsf->disable | jsf->new);
4142 		db_printf(" %-16s= %s\n", jsf->name,
4143 		    (f != 0 && f == jsf->disable) ? "disable"
4144 		    : (f == jsf->new) ? "new"
4145 		    : "inherit");
4146 	}
4147 	db_printf(" allow           = 0x%x", pr->pr_allow);
4148 	for (bf = pr_flag_allow;
4149 	     bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
4150 	     bf++)
4151 		if (pr->pr_allow & bf->flag)
4152 			db_printf(" %s", bf->name);
4153 	db_printf("\n");
4154 	db_printf(" enforce_statfs  = %d\n", pr->pr_enforce_statfs);
4155 	db_printf(" host.hostname   = %s\n", pr->pr_hostname);
4156 	db_printf(" host.domainname = %s\n", pr->pr_domainname);
4157 	db_printf(" host.hostuuid   = %s\n", pr->pr_hostuuid);
4158 	db_printf(" host.hostid     = %lu\n", pr->pr_hostid);
4159 #ifdef INET
4160 	db_printf(" ip4s            = %d\n", pr->pr_ip4s);
4161 	for (ii = 0; ii < pr->pr_ip4s; ii++)
4162 		db_printf(" %s %s\n",
4163 		    ii == 0 ? "ip4.addr        =" : "                 ",
4164 		    inet_ntoa_r(pr->pr_ip4[ii], ip4buf));
4165 #endif
4166 #ifdef INET6
4167 	db_printf(" ip6s            = %d\n", pr->pr_ip6s);
4168 	for (ii = 0; ii < pr->pr_ip6s; ii++)
4169 		db_printf(" %s %s\n",
4170 		    ii == 0 ? "ip6.addr        =" : "                 ",
4171 		    ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4172 #endif
4173 }
4174 
4175 DB_SHOW_COMMAND(prison, db_show_prison_command)
4176 {
4177 	struct prison *pr;
4178 
4179 	if (!have_addr) {
4180 		/*
4181 		 * Show all prisons in the list, and prison0 which is not
4182 		 * listed.
4183 		 */
4184 		db_show_prison(&prison0);
4185 		if (!db_pager_quit) {
4186 			TAILQ_FOREACH(pr, &allprison, pr_list) {
4187 				db_show_prison(pr);
4188 				if (db_pager_quit)
4189 					break;
4190 			}
4191 		}
4192 		return;
4193 	}
4194 
4195 	if (addr == 0)
4196 		pr = &prison0;
4197 	else {
4198 		/* Look for a prison with the ID and with references. */
4199 		TAILQ_FOREACH(pr, &allprison, pr_list)
4200 			if (pr->pr_id == addr && pr->pr_ref > 0)
4201 				break;
4202 		if (pr == NULL)
4203 			/* Look again, without requiring a reference. */
4204 			TAILQ_FOREACH(pr, &allprison, pr_list)
4205 				if (pr->pr_id == addr)
4206 					break;
4207 		if (pr == NULL)
4208 			/* Assume address points to a valid prison. */
4209 			pr = (struct prison *)addr;
4210 	}
4211 	db_show_prison(pr);
4212 }
4213 
4214 #endif /* DDB */
4215