xref: /freebsd/sys/kern/kern_jail.c (revision 037479ff5ee18977b1c48e1e59770aad2f200a5a)
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 /*
441  * struct jail_set_args {
442  *	struct iovec *iovp;
443  *	unsigned int iovcnt;
444  *	int flags;
445  * };
446  */
447 int
448 sys_jail_set(struct thread *td, struct jail_set_args *uap)
449 {
450 	struct uio *auio;
451 	int error;
452 
453 	/* Check that we have an even number of iovecs. */
454 	if (uap->iovcnt & 1)
455 		return (EINVAL);
456 
457 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
458 	if (error)
459 		return (error);
460 	error = kern_jail_set(td, auio, uap->flags);
461 	free(auio, M_IOV);
462 	return (error);
463 }
464 
465 int
466 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
467 {
468 	struct nameidata nd;
469 #ifdef INET
470 	struct in_addr *ip4;
471 #endif
472 #ifdef INET6
473 	struct in6_addr *ip6;
474 #endif
475 	struct vfsopt *opt;
476 	struct vfsoptlist *opts;
477 	struct prison *pr, *deadpr, *mypr, *ppr, *tpr;
478 	struct vnode *root;
479 	char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
480 	char *g_path, *osrelstr;
481 	struct bool_flags *bf;
482 	struct jailsys_flags *jsf;
483 #if defined(INET) || defined(INET6)
484 	struct prison *tppr;
485 	void *op;
486 #endif
487 	unsigned long hid;
488 	size_t namelen, onamelen, pnamelen;
489 	int born, created, cuflags, descend, enforce;
490 	int error, errmsg_len, errmsg_pos;
491 	int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
492 	int jid, jsys, len, level;
493 	int childmax, osreldt, rsnum, slevel;
494 	int fullpath_disabled;
495 #if defined(INET) || defined(INET6)
496 	int ii, ij;
497 #endif
498 #ifdef INET
499 	int ip4s, redo_ip4;
500 #endif
501 #ifdef INET6
502 	int ip6s, redo_ip6;
503 #endif
504 	uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
505 	uint64_t pr_allow_diff;
506 	unsigned tallow;
507 	char numbuf[12];
508 
509 	error = priv_check(td, PRIV_JAIL_SET);
510 	if (!error && (flags & JAIL_ATTACH))
511 		error = priv_check(td, PRIV_JAIL_ATTACH);
512 	if (error)
513 		return (error);
514 	mypr = td->td_ucred->cr_prison;
515 	if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
516 		return (EPERM);
517 	if (flags & ~JAIL_SET_MASK)
518 		return (EINVAL);
519 
520 	/*
521 	 * Check all the parameters before committing to anything.  Not all
522 	 * errors can be caught early, but we may as well try.  Also, this
523 	 * takes care of some expensive stuff (path lookup) before getting
524 	 * the allprison lock.
525 	 *
526 	 * XXX Jails are not filesystems, and jail parameters are not mount
527 	 *     options.  But it makes more sense to re-use the vfsopt code
528 	 *     than duplicate it under a different name.
529 	 */
530 	error = vfs_buildopts(optuio, &opts);
531 	if (error)
532 		return (error);
533 #ifdef INET
534 	ip4 = NULL;
535 #endif
536 #ifdef INET6
537 	ip6 = NULL;
538 #endif
539 	g_path = NULL;
540 
541 	cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
542 	if (!cuflags) {
543 		error = EINVAL;
544 		vfs_opterror(opts, "no valid operation (create or update)");
545 		goto done_errmsg;
546 	}
547 
548 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
549 	if (error == ENOENT)
550 		jid = 0;
551 	else if (error != 0)
552 		goto done_free;
553 
554 	error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
555 	if (error == ENOENT)
556 		gotslevel = 0;
557 	else if (error != 0)
558 		goto done_free;
559 	else
560 		gotslevel = 1;
561 
562 	error =
563 	    vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
564 	if (error == ENOENT)
565 		gotchildmax = 0;
566 	else if (error != 0)
567 		goto done_free;
568 	else
569 		gotchildmax = 1;
570 
571 	error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
572 	if (error == ENOENT)
573 		gotenforce = 0;
574 	else if (error != 0)
575 		goto done_free;
576 	else if (enforce < 0 || enforce > 2) {
577 		error = EINVAL;
578 		goto done_free;
579 	} else
580 		gotenforce = 1;
581 
582 	error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
583 	if (error == ENOENT)
584 		gotrsnum = 0;
585 	else if (error != 0)
586 		goto done_free;
587 	else
588 		gotrsnum = 1;
589 
590 	pr_flags = ch_flags = 0;
591 	for (bf = pr_flag_bool;
592 	     bf < pr_flag_bool + nitems(pr_flag_bool);
593 	     bf++) {
594 		vfs_flagopt(opts, bf->name, &pr_flags, bf->flag);
595 		vfs_flagopt(opts, bf->noname, &ch_flags, bf->flag);
596 	}
597 	ch_flags |= pr_flags;
598 	for (jsf = pr_flag_jailsys;
599 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
600 	     jsf++) {
601 		error = vfs_copyopt(opts, jsf->name, &jsys, sizeof(jsys));
602 		if (error == ENOENT)
603 			continue;
604 		if (error != 0)
605 			goto done_free;
606 		switch (jsys) {
607 		case JAIL_SYS_DISABLE:
608 			if (!jsf->disable) {
609 				error = EINVAL;
610 				goto done_free;
611 			}
612 			pr_flags |= jsf->disable;
613 			break;
614 		case JAIL_SYS_NEW:
615 			pr_flags |= jsf->new;
616 			break;
617 		case JAIL_SYS_INHERIT:
618 			break;
619 		default:
620 			error = EINVAL;
621 			goto done_free;
622 		}
623 		ch_flags |= jsf->new | jsf->disable;
624 	}
625 	if ((flags & (JAIL_CREATE | JAIL_UPDATE | JAIL_ATTACH)) == JAIL_CREATE
626 	    && !(pr_flags & PR_PERSIST)) {
627 		error = EINVAL;
628 		vfs_opterror(opts, "new jail must persist or attach");
629 		goto done_errmsg;
630 	}
631 #ifdef VIMAGE
632 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
633 		error = EINVAL;
634 		vfs_opterror(opts, "vnet cannot be changed after creation");
635 		goto done_errmsg;
636 	}
637 #endif
638 #ifdef INET
639 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
640 		error = EINVAL;
641 		vfs_opterror(opts, "ip4 cannot be changed after creation");
642 		goto done_errmsg;
643 	}
644 #endif
645 #ifdef INET6
646 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
647 		error = EINVAL;
648 		vfs_opterror(opts, "ip6 cannot be changed after creation");
649 		goto done_errmsg;
650 	}
651 #endif
652 
653 	pr_allow = ch_allow = 0;
654 	for (bf = pr_flag_allow;
655 	     bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
656 	     bf++) {
657 		vfs_flagopt(opts, bf->name, &pr_allow, bf->flag);
658 		vfs_flagopt(opts, bf->noname, &ch_allow, bf->flag);
659 	}
660 	ch_allow |= pr_allow;
661 
662 	error = vfs_getopt(opts, "name", (void **)&name, &len);
663 	if (error == ENOENT)
664 		name = NULL;
665 	else if (error != 0)
666 		goto done_free;
667 	else {
668 		if (len == 0 || name[len - 1] != '\0') {
669 			error = EINVAL;
670 			goto done_free;
671 		}
672 		if (len > MAXHOSTNAMELEN) {
673 			error = ENAMETOOLONG;
674 			goto done_free;
675 		}
676 	}
677 
678 	error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
679 	if (error == ENOENT)
680 		host = NULL;
681 	else if (error != 0)
682 		goto done_free;
683 	else {
684 		ch_flags |= PR_HOST;
685 		pr_flags |= PR_HOST;
686 		if (len == 0 || host[len - 1] != '\0') {
687 			error = EINVAL;
688 			goto done_free;
689 		}
690 		if (len > MAXHOSTNAMELEN) {
691 			error = ENAMETOOLONG;
692 			goto done_free;
693 		}
694 	}
695 
696 	error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
697 	if (error == ENOENT)
698 		domain = NULL;
699 	else if (error != 0)
700 		goto done_free;
701 	else {
702 		ch_flags |= PR_HOST;
703 		pr_flags |= PR_HOST;
704 		if (len == 0 || domain[len - 1] != '\0') {
705 			error = EINVAL;
706 			goto done_free;
707 		}
708 		if (len > MAXHOSTNAMELEN) {
709 			error = ENAMETOOLONG;
710 			goto done_free;
711 		}
712 	}
713 
714 	error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
715 	if (error == ENOENT)
716 		uuid = NULL;
717 	else if (error != 0)
718 		goto done_free;
719 	else {
720 		ch_flags |= PR_HOST;
721 		pr_flags |= PR_HOST;
722 		if (len == 0 || uuid[len - 1] != '\0') {
723 			error = EINVAL;
724 			goto done_free;
725 		}
726 		if (len > HOSTUUIDLEN) {
727 			error = ENAMETOOLONG;
728 			goto done_free;
729 		}
730 	}
731 
732 #ifdef COMPAT_FREEBSD32
733 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
734 		uint32_t hid32;
735 
736 		error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
737 		hid = hid32;
738 	} else
739 #endif
740 		error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
741 	if (error == ENOENT)
742 		gothid = 0;
743 	else if (error != 0)
744 		goto done_free;
745 	else {
746 		gothid = 1;
747 		ch_flags |= PR_HOST;
748 		pr_flags |= PR_HOST;
749 	}
750 
751 #ifdef INET
752 	error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
753 	if (error == ENOENT)
754 		ip4s = 0;
755 	else if (error != 0)
756 		goto done_free;
757 	else if (ip4s & (sizeof(*ip4) - 1)) {
758 		error = EINVAL;
759 		goto done_free;
760 	} else {
761 		ch_flags |= PR_IP4_USER;
762 		pr_flags |= PR_IP4_USER;
763 		if (ip4s > 0) {
764 			ip4s /= sizeof(*ip4);
765 			if (ip4s > jail_max_af_ips) {
766 				error = EINVAL;
767 				vfs_opterror(opts, "too many IPv4 addresses");
768 				goto done_errmsg;
769 			}
770 			ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
771 			bcopy(op, ip4, ip4s * sizeof(*ip4));
772 			/*
773 			 * IP addresses are all sorted but ip[0] to preserve
774 			 * the primary IP address as given from userland.
775 			 * This special IP is used for unbound outgoing
776 			 * connections as well for "loopback" traffic in case
777 			 * source address selection cannot find any more fitting
778 			 * address to connect from.
779 			 */
780 			if (ip4s > 1)
781 				qsort(ip4 + 1, ip4s - 1, sizeof(*ip4),
782 				    prison_qcmp_v4);
783 			/*
784 			 * Check for duplicate addresses and do some simple
785 			 * zero and broadcast checks. If users give other bogus
786 			 * addresses it is their problem.
787 			 *
788 			 * We do not have to care about byte order for these
789 			 * checks so we will do them in NBO.
790 			 */
791 			for (ii = 0; ii < ip4s; ii++) {
792 				if (ip4[ii].s_addr == INADDR_ANY ||
793 				    ip4[ii].s_addr == INADDR_BROADCAST) {
794 					error = EINVAL;
795 					goto done_free;
796 				}
797 				if ((ii+1) < ip4s &&
798 				    (ip4[0].s_addr == ip4[ii+1].s_addr ||
799 				     ip4[ii].s_addr == ip4[ii+1].s_addr)) {
800 					error = EINVAL;
801 					goto done_free;
802 				}
803 			}
804 		}
805 	}
806 #endif
807 
808 #ifdef INET6
809 	error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
810 	if (error == ENOENT)
811 		ip6s = 0;
812 	else if (error != 0)
813 		goto done_free;
814 	else if (ip6s & (sizeof(*ip6) - 1)) {
815 		error = EINVAL;
816 		goto done_free;
817 	} else {
818 		ch_flags |= PR_IP6_USER;
819 		pr_flags |= PR_IP6_USER;
820 		if (ip6s > 0) {
821 			ip6s /= sizeof(*ip6);
822 			if (ip6s > jail_max_af_ips) {
823 				error = EINVAL;
824 				vfs_opterror(opts, "too many IPv6 addresses");
825 				goto done_errmsg;
826 			}
827 			ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
828 			bcopy(op, ip6, ip6s * sizeof(*ip6));
829 			if (ip6s > 1)
830 				qsort(ip6 + 1, ip6s - 1, sizeof(*ip6),
831 				    prison_qcmp_v6);
832 			for (ii = 0; ii < ip6s; ii++) {
833 				if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
834 					error = EINVAL;
835 					goto done_free;
836 				}
837 				if ((ii+1) < ip6s &&
838 				    (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
839 				     IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
840 				{
841 					error = EINVAL;
842 					goto done_free;
843 				}
844 			}
845 		}
846 	}
847 #endif
848 
849 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
850 	if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
851 		error = EINVAL;
852 		vfs_opterror(opts,
853 		    "vnet jails cannot have IP address restrictions");
854 		goto done_errmsg;
855 	}
856 #endif
857 
858 	error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
859 	if (error == ENOENT)
860 		osrelstr = NULL;
861 	else if (error != 0)
862 		goto done_free;
863 	else {
864 		if (flags & JAIL_UPDATE) {
865 			error = EINVAL;
866 			vfs_opterror(opts,
867 			    "osrelease cannot be changed after creation");
868 			goto done_errmsg;
869 		}
870 		if (len == 0 || len >= OSRELEASELEN) {
871 			error = EINVAL;
872 			vfs_opterror(opts,
873 			    "osrelease string must be 1-%d bytes long",
874 			    OSRELEASELEN - 1);
875 			goto done_errmsg;
876 		}
877 	}
878 
879 	error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
880 	if (error == ENOENT)
881 		osreldt = 0;
882 	else if (error != 0)
883 		goto done_free;
884 	else {
885 		if (flags & JAIL_UPDATE) {
886 			error = EINVAL;
887 			vfs_opterror(opts,
888 			    "osreldate cannot be changed after creation");
889 			goto done_errmsg;
890 		}
891 		if (osreldt == 0) {
892 			error = EINVAL;
893 			vfs_opterror(opts, "osreldate cannot be 0");
894 			goto done_errmsg;
895 		}
896 	}
897 
898 	fullpath_disabled = 0;
899 	root = NULL;
900 	error = vfs_getopt(opts, "path", (void **)&path, &len);
901 	if (error == ENOENT)
902 		path = NULL;
903 	else if (error != 0)
904 		goto done_free;
905 	else {
906 		if (flags & JAIL_UPDATE) {
907 			error = EINVAL;
908 			vfs_opterror(opts,
909 			    "path cannot be changed after creation");
910 			goto done_errmsg;
911 		}
912 		if (len == 0 || path[len - 1] != '\0') {
913 			error = EINVAL;
914 			goto done_free;
915 		}
916 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
917 		    path, td);
918 		error = namei(&nd);
919 		if (error)
920 			goto done_free;
921 		root = nd.ni_vp;
922 		NDFREE(&nd, NDF_ONLY_PNBUF);
923 		g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
924 		strlcpy(g_path, path, MAXPATHLEN);
925 		error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
926 		if (error == 0)
927 			path = g_path;
928 		else if (error == ENODEV) {
929 			/* proceed if sysctl debug.disablefullpath == 1 */
930 			fullpath_disabled = 1;
931 			if (len < 2 || (len == 2 && path[0] == '/'))
932 				path = NULL;
933 		} else {
934 			/* exit on other errors */
935 			goto done_free;
936 		}
937 		if (root->v_type != VDIR) {
938 			error = ENOTDIR;
939 			vput(root);
940 			goto done_free;
941 		}
942 		VOP_UNLOCK(root, 0);
943 		if (fullpath_disabled) {
944 			/* Leave room for a real-root full pathname. */
945 			if (len + (path[0] == '/' && strcmp(mypr->pr_path, "/")
946 			    ? strlen(mypr->pr_path) : 0) > MAXPATHLEN) {
947 				error = ENAMETOOLONG;
948 				vrele(root);
949 				goto done_free;
950 			}
951 		}
952 	}
953 
954 	/*
955 	 * Find the specified jail, or at least its parent.
956 	 * This abuses the file error codes ENOENT and EEXIST.
957 	 */
958 	pr = NULL;
959 	ppr = mypr;
960 	if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
961 		namelc = strrchr(name, '.');
962 		jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
963 		if (*p != '\0')
964 			jid = 0;
965 	}
966 	sx_xlock(&allprison_lock);
967 	if (jid != 0) {
968 		/*
969 		 * See if a requested jid already exists.  There is an
970 		 * information leak here if the jid exists but is not within
971 		 * the caller's jail hierarchy.  Jail creators will get EEXIST
972 		 * even though they cannot see the jail, and CREATE | UPDATE
973 		 * will return ENOENT which is not normally a valid error.
974 		 */
975 		if (jid < 0) {
976 			error = EINVAL;
977 			vfs_opterror(opts, "negative jid");
978 			goto done_unlock_list;
979 		}
980 		pr = prison_find(jid);
981 		if (pr != NULL) {
982 			ppr = pr->pr_parent;
983 			/* Create: jid must not exist. */
984 			if (cuflags == JAIL_CREATE) {
985 				mtx_unlock(&pr->pr_mtx);
986 				error = EEXIST;
987 				vfs_opterror(opts, "jail %d already exists",
988 				    jid);
989 				goto done_unlock_list;
990 			}
991 			if (!prison_ischild(mypr, pr)) {
992 				mtx_unlock(&pr->pr_mtx);
993 				pr = NULL;
994 			} else if (pr->pr_uref == 0) {
995 				if (!(flags & JAIL_DYING)) {
996 					mtx_unlock(&pr->pr_mtx);
997 					error = ENOENT;
998 					vfs_opterror(opts, "jail %d is dying",
999 					    jid);
1000 					goto done_unlock_list;
1001 				} else if ((flags & JAIL_ATTACH) ||
1002 				    (pr_flags & PR_PERSIST)) {
1003 					/*
1004 					 * A dying jail might be resurrected
1005 					 * (via attach or persist), but first
1006 					 * it must determine if another jail
1007 					 * has claimed its name.  Accomplish
1008 					 * this by implicitly re-setting the
1009 					 * name.
1010 					 */
1011 					if (name == NULL)
1012 						name = prison_name(mypr, pr);
1013 				}
1014 			}
1015 		}
1016 		if (pr == NULL) {
1017 			/* Update: jid must exist. */
1018 			if (cuflags == JAIL_UPDATE) {
1019 				error = ENOENT;
1020 				vfs_opterror(opts, "jail %d not found", jid);
1021 				goto done_unlock_list;
1022 			}
1023 		}
1024 	}
1025 	/*
1026 	 * If the caller provided a name, look for a jail by that name.
1027 	 * This has different semantics for creates and updates keyed by jid
1028 	 * (where the name must not already exist in a different jail),
1029 	 * and updates keyed by the name itself (where the name must exist
1030 	 * because that is the jail being updated).
1031 	 */
1032 	namelc = NULL;
1033 	if (name != NULL) {
1034 		namelc = strrchr(name, '.');
1035 		if (namelc == NULL)
1036 			namelc = name;
1037 		else {
1038 			/*
1039 			 * This is a hierarchical name.  Split it into the
1040 			 * parent and child names, and make sure the parent
1041 			 * exists or matches an already found jail.
1042 			 */
1043 			if (pr != NULL) {
1044 				if (strncmp(name, ppr->pr_name, namelc - name)
1045 				    || ppr->pr_name[namelc - name] != '\0') {
1046 					mtx_unlock(&pr->pr_mtx);
1047 					error = EINVAL;
1048 					vfs_opterror(opts,
1049 					    "cannot change jail's parent");
1050 					goto done_unlock_list;
1051 				}
1052 			} else {
1053 				*namelc = '\0';
1054 				ppr = prison_find_name(mypr, name);
1055 				if (ppr == NULL) {
1056 					error = ENOENT;
1057 					vfs_opterror(opts,
1058 					    "jail \"%s\" not found", name);
1059 					goto done_unlock_list;
1060 				}
1061 				mtx_unlock(&ppr->pr_mtx);
1062 				*namelc = '.';
1063 			}
1064 			namelc++;
1065 		}
1066 		if (namelc[0] != '\0') {
1067 			pnamelen =
1068 			    (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1069  name_again:
1070 			deadpr = NULL;
1071 			FOREACH_PRISON_CHILD(ppr, tpr) {
1072 				if (tpr != pr && tpr->pr_ref > 0 &&
1073 				    !strcmp(tpr->pr_name + pnamelen, namelc)) {
1074 					if (pr == NULL &&
1075 					    cuflags != JAIL_CREATE) {
1076 						mtx_lock(&tpr->pr_mtx);
1077 						if (tpr->pr_ref > 0) {
1078 							/*
1079 							 * Use this jail
1080 							 * for updates.
1081 							 */
1082 							if (tpr->pr_uref > 0) {
1083 								pr = tpr;
1084 								break;
1085 							}
1086 							deadpr = tpr;
1087 						}
1088 						mtx_unlock(&tpr->pr_mtx);
1089 					} else if (tpr->pr_uref > 0) {
1090 						/*
1091 						 * Create, or update(jid):
1092 						 * name must not exist in an
1093 						 * active sibling jail.
1094 						 */
1095 						error = EEXIST;
1096 						if (pr != NULL)
1097 							mtx_unlock(&pr->pr_mtx);
1098 						vfs_opterror(opts,
1099 						   "jail \"%s\" already exists",
1100 						   name);
1101 						goto done_unlock_list;
1102 					}
1103 				}
1104 			}
1105 			/* If no active jail is found, use a dying one. */
1106 			if (deadpr != NULL && pr == NULL) {
1107 				if (flags & JAIL_DYING) {
1108 					mtx_lock(&deadpr->pr_mtx);
1109 					if (deadpr->pr_ref == 0) {
1110 						mtx_unlock(&deadpr->pr_mtx);
1111 						goto name_again;
1112 					}
1113 					pr = deadpr;
1114 				} else if (cuflags == JAIL_UPDATE) {
1115 					error = ENOENT;
1116 					vfs_opterror(opts,
1117 					    "jail \"%s\" is dying", name);
1118 					goto done_unlock_list;
1119 				}
1120 			}
1121 			/* Update: name must exist if no jid. */
1122 			else if (cuflags == JAIL_UPDATE && pr == NULL) {
1123 				error = ENOENT;
1124 				vfs_opterror(opts, "jail \"%s\" not found",
1125 				    name);
1126 				goto done_unlock_list;
1127 			}
1128 		}
1129 	}
1130 	/* Update: must provide a jid or name. */
1131 	else if (cuflags == JAIL_UPDATE && pr == NULL) {
1132 		error = ENOENT;
1133 		vfs_opterror(opts, "update specified no jail");
1134 		goto done_unlock_list;
1135 	}
1136 
1137 	/* If there's no prison to update, create a new one and link it in. */
1138 	if (pr == NULL) {
1139 		for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1140 			if (tpr->pr_childcount >= tpr->pr_childmax) {
1141 				error = EPERM;
1142 				vfs_opterror(opts, "prison limit exceeded");
1143 				goto done_unlock_list;
1144 			}
1145 		created = 1;
1146 		mtx_lock(&ppr->pr_mtx);
1147 		if (ppr->pr_ref == 0) {
1148 			mtx_unlock(&ppr->pr_mtx);
1149 			error = ENOENT;
1150 			vfs_opterror(opts, "jail \"%s\" not found",
1151 			    prison_name(mypr, ppr));
1152 			goto done_unlock_list;
1153 		}
1154 		ppr->pr_ref++;
1155 		ppr->pr_uref++;
1156 		mtx_unlock(&ppr->pr_mtx);
1157 		pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1158 		if (jid == 0) {
1159 			/* Find the next free jid. */
1160 			jid = lastprid + 1;
1161  findnext:
1162 			if (jid == JAIL_MAX)
1163 				jid = 1;
1164 			TAILQ_FOREACH(tpr, &allprison, pr_list) {
1165 				if (tpr->pr_id < jid)
1166 					continue;
1167 				if (tpr->pr_id > jid || tpr->pr_ref == 0) {
1168 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1169 					break;
1170 				}
1171 				if (jid == lastprid) {
1172 					error = EAGAIN;
1173 					vfs_opterror(opts,
1174 					    "no available jail IDs");
1175 					free(pr, M_PRISON);
1176 					prison_deref(ppr, PD_DEREF |
1177 					    PD_DEUREF | PD_LIST_XLOCKED);
1178 					goto done_releroot;
1179 				}
1180 				jid++;
1181 				goto findnext;
1182 			}
1183 			lastprid = jid;
1184 		} else {
1185 			/*
1186 			 * The jail already has a jid (that did not yet exist),
1187 			 * so just find where to insert it.
1188 			 */
1189 			TAILQ_FOREACH(tpr, &allprison, pr_list)
1190 				if (tpr->pr_id >= jid) {
1191 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1192 					break;
1193 				}
1194 		}
1195 		if (tpr == NULL)
1196 			TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1197 		LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1198 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1199 			tpr->pr_childcount++;
1200 
1201 		pr->pr_parent = ppr;
1202 		pr->pr_id = jid;
1203 
1204 		/* Set some default values, and inherit some from the parent. */
1205 		if (namelc == NULL)
1206 			namelc = "";
1207 		if (path == NULL) {
1208 			path = "/";
1209 			root = mypr->pr_root;
1210 			vref(root);
1211 		}
1212 		strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1213 		pr->pr_flags |= PR_HOST;
1214 #if defined(INET) || defined(INET6)
1215 #ifdef VIMAGE
1216 		if (!(pr_flags & PR_VNET))
1217 #endif
1218 		{
1219 #ifdef INET
1220 			if (!(ch_flags & PR_IP4_USER))
1221 				pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1222 			else if (!(pr_flags & PR_IP4_USER)) {
1223 				pr->pr_flags |= ppr->pr_flags & PR_IP4;
1224 				if (ppr->pr_ip4 != NULL) {
1225 					pr->pr_ip4s = ppr->pr_ip4s;
1226 					pr->pr_ip4 = malloc(pr->pr_ip4s *
1227 					    sizeof(struct in_addr), M_PRISON,
1228 					    M_WAITOK);
1229 					bcopy(ppr->pr_ip4, pr->pr_ip4,
1230 					    pr->pr_ip4s * sizeof(*pr->pr_ip4));
1231 				}
1232 			}
1233 #endif
1234 #ifdef INET6
1235 			if (!(ch_flags & PR_IP6_USER))
1236 				pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1237 			else if (!(pr_flags & PR_IP6_USER)) {
1238 				pr->pr_flags |= ppr->pr_flags & PR_IP6;
1239 				if (ppr->pr_ip6 != NULL) {
1240 					pr->pr_ip6s = ppr->pr_ip6s;
1241 					pr->pr_ip6 = malloc(pr->pr_ip6s *
1242 					    sizeof(struct in6_addr), M_PRISON,
1243 					    M_WAITOK);
1244 					bcopy(ppr->pr_ip6, pr->pr_ip6,
1245 					    pr->pr_ip6s * sizeof(*pr->pr_ip6));
1246 				}
1247 			}
1248 #endif
1249 		}
1250 #endif
1251 		/* Source address selection is always on by default. */
1252 		pr->pr_flags |= _PR_IP_SADDRSEL;
1253 
1254 		pr->pr_securelevel = ppr->pr_securelevel;
1255 		pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1256 		pr->pr_enforce_statfs = jail_default_enforce_statfs;
1257 		pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1258 
1259 		pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1260 		if (osrelstr == NULL)
1261 		    strcpy(pr->pr_osrelease, ppr->pr_osrelease);
1262 		else
1263 		    strcpy(pr->pr_osrelease, osrelstr);
1264 
1265 		LIST_INIT(&pr->pr_children);
1266 		mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1267 		TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1268 
1269 #ifdef VIMAGE
1270 		/* Allocate a new vnet if specified. */
1271 		pr->pr_vnet = (pr_flags & PR_VNET)
1272 		    ? vnet_alloc() : ppr->pr_vnet;
1273 #endif
1274 		/*
1275 		 * Allocate a dedicated cpuset for each jail.
1276 		 * Unlike other initial settings, this may return an erorr.
1277 		 */
1278 		error = cpuset_create_root(ppr, &pr->pr_cpuset);
1279 		if (error) {
1280 			prison_deref(pr, PD_LIST_XLOCKED);
1281 			goto done_releroot;
1282 		}
1283 
1284 		mtx_lock(&pr->pr_mtx);
1285 		/*
1286 		 * New prisons do not yet have a reference, because we do not
1287 		 * want others to see the incomplete prison once the
1288 		 * allprison_lock is downgraded.
1289 		 */
1290 	} else {
1291 		created = 0;
1292 		/*
1293 		 * Grab a reference for existing prisons, to ensure they
1294 		 * continue to exist for the duration of the call.
1295 		 */
1296 		pr->pr_ref++;
1297 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1298 		if ((pr->pr_flags & PR_VNET) &&
1299 		    (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1300 			error = EINVAL;
1301 			vfs_opterror(opts,
1302 			    "vnet jails cannot have IP address restrictions");
1303 			goto done_deref_locked;
1304 		}
1305 #endif
1306 #ifdef INET
1307 		if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1308 			error = EINVAL;
1309 			vfs_opterror(opts,
1310 			    "ip4 cannot be changed after creation");
1311 			goto done_deref_locked;
1312 		}
1313 #endif
1314 #ifdef INET6
1315 		if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1316 			error = EINVAL;
1317 			vfs_opterror(opts,
1318 			    "ip6 cannot be changed after creation");
1319 			goto done_deref_locked;
1320 		}
1321 #endif
1322 	}
1323 
1324 	/* Do final error checking before setting anything. */
1325 	if (gotslevel) {
1326 		if (slevel < ppr->pr_securelevel) {
1327 			error = EPERM;
1328 			goto done_deref_locked;
1329 		}
1330 	}
1331 	if (gotchildmax) {
1332 		if (childmax >= ppr->pr_childmax) {
1333 			error = EPERM;
1334 			goto done_deref_locked;
1335 		}
1336 	}
1337 	if (gotenforce) {
1338 		if (enforce < ppr->pr_enforce_statfs) {
1339 			error = EPERM;
1340 			goto done_deref_locked;
1341 		}
1342 	}
1343 	if (gotrsnum) {
1344 		/*
1345 		 * devfs_rsnum is a uint16_t
1346 		 */
1347 		if (rsnum < 0 || rsnum > 65535) {
1348 			error = EINVAL;
1349 			goto done_deref_locked;
1350 		}
1351 		/*
1352 		 * Nested jails always inherit parent's devfs ruleset
1353 		 */
1354 		if (jailed(td->td_ucred)) {
1355 			if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1356 				error = EPERM;
1357 				goto done_deref_locked;
1358 			} else
1359 				rsnum = ppr->pr_devfs_rsnum;
1360 		}
1361 	}
1362 #ifdef INET
1363 	if (ip4s > 0) {
1364 		if (ppr->pr_flags & PR_IP4) {
1365 			/*
1366 			 * Make sure the new set of IP addresses is a
1367 			 * subset of the parent's list.  Don't worry
1368 			 * about the parent being unlocked, as any
1369 			 * setting is done with allprison_lock held.
1370 			 */
1371 			for (ij = 0; ij < ppr->pr_ip4s; ij++)
1372 				if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
1373 					break;
1374 			if (ij == ppr->pr_ip4s) {
1375 				error = EPERM;
1376 				goto done_deref_locked;
1377 			}
1378 			if (ip4s > 1) {
1379 				for (ii = ij = 1; ii < ip4s; ii++) {
1380 					if (ip4[ii].s_addr ==
1381 					    ppr->pr_ip4[0].s_addr)
1382 						continue;
1383 					for (; ij < ppr->pr_ip4s; ij++)
1384 						if (ip4[ii].s_addr ==
1385 						    ppr->pr_ip4[ij].s_addr)
1386 							break;
1387 					if (ij == ppr->pr_ip4s)
1388 						break;
1389 				}
1390 				if (ij == ppr->pr_ip4s) {
1391 					error = EPERM;
1392 					goto done_deref_locked;
1393 				}
1394 			}
1395 		}
1396 		/*
1397 		 * Check for conflicting IP addresses.  We permit them
1398 		 * if there is no more than one IP on each jail.  If
1399 		 * there is a duplicate on a jail with more than one
1400 		 * IP stop checking and return error.
1401 		 */
1402 #ifdef VIMAGE
1403 		for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
1404 			if (tppr->pr_flags & PR_VNET)
1405 				break;
1406 #else
1407 		tppr = &prison0;
1408 #endif
1409 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1410 			if (tpr == pr ||
1411 #ifdef VIMAGE
1412 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1413 #endif
1414 			    tpr->pr_uref == 0) {
1415 				descend = 0;
1416 				continue;
1417 			}
1418 			if (!(tpr->pr_flags & PR_IP4_USER))
1419 				continue;
1420 			descend = 0;
1421 			if (tpr->pr_ip4 == NULL ||
1422 			    (ip4s == 1 && tpr->pr_ip4s == 1))
1423 				continue;
1424 			for (ii = 0; ii < ip4s; ii++) {
1425 				if (prison_check_ip4_locked(tpr, &ip4[ii]) ==
1426 				    0) {
1427 					error = EADDRINUSE;
1428 					vfs_opterror(opts,
1429 					    "IPv4 addresses clash");
1430 					goto done_deref_locked;
1431 				}
1432 			}
1433 		}
1434 	}
1435 #endif
1436 #ifdef INET6
1437 	if (ip6s > 0) {
1438 		if (ppr->pr_flags & PR_IP6) {
1439 			/*
1440 			 * Make sure the new set of IP addresses is a
1441 			 * subset of the parent's list.
1442 			 */
1443 			for (ij = 0; ij < ppr->pr_ip6s; ij++)
1444 				if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1445 				    &ppr->pr_ip6[ij]))
1446 					break;
1447 			if (ij == ppr->pr_ip6s) {
1448 				error = EPERM;
1449 				goto done_deref_locked;
1450 			}
1451 			if (ip6s > 1) {
1452 				for (ii = ij = 1; ii < ip6s; ii++) {
1453 					if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1454 					     &ppr->pr_ip6[0]))
1455 						continue;
1456 					for (; ij < ppr->pr_ip6s; ij++)
1457 						if (IN6_ARE_ADDR_EQUAL(
1458 						    &ip6[ii], &ppr->pr_ip6[ij]))
1459 							break;
1460 					if (ij == ppr->pr_ip6s)
1461 						break;
1462 				}
1463 				if (ij == ppr->pr_ip6s) {
1464 					error = EPERM;
1465 					goto done_deref_locked;
1466 				}
1467 			}
1468 		}
1469 		/* Check for conflicting IP addresses. */
1470 #ifdef VIMAGE
1471 		for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent)
1472 			if (tppr->pr_flags & PR_VNET)
1473 				break;
1474 #else
1475 		tppr = &prison0;
1476 #endif
1477 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1478 			if (tpr == pr ||
1479 #ifdef VIMAGE
1480 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1481 #endif
1482 			    tpr->pr_uref == 0) {
1483 				descend = 0;
1484 				continue;
1485 			}
1486 			if (!(tpr->pr_flags & PR_IP6_USER))
1487 				continue;
1488 			descend = 0;
1489 			if (tpr->pr_ip6 == NULL ||
1490 			    (ip6s == 1 && tpr->pr_ip6s == 1))
1491 				continue;
1492 			for (ii = 0; ii < ip6s; ii++) {
1493 				if (prison_check_ip6_locked(tpr, &ip6[ii]) ==
1494 				    0) {
1495 					error = EADDRINUSE;
1496 					vfs_opterror(opts,
1497 					    "IPv6 addresses clash");
1498 					goto done_deref_locked;
1499 				}
1500 			}
1501 		}
1502 	}
1503 #endif
1504 	onamelen = namelen = 0;
1505 	if (namelc != NULL) {
1506 		/* Give a default name of the jid.  Also allow the name to be
1507 		 * explicitly the jid - but not any other number, and only in
1508 		 * normal form (no leading zero/etc).
1509 		 */
1510 		if (namelc[0] == '\0')
1511 			snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1512 		else if ((strtoul(namelc, &p, 10) != jid ||
1513 			  namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1514 			error = EINVAL;
1515 			vfs_opterror(opts,
1516 			    "name cannot be numeric (unless it is the jid)");
1517 			goto done_deref_locked;
1518 		}
1519 		/*
1520 		 * Make sure the name isn't too long for the prison or its
1521 		 * children.
1522 		 */
1523 		pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1524 		onamelen = strlen(pr->pr_name + pnamelen);
1525 		namelen = strlen(namelc);
1526 		if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1527 			error = ENAMETOOLONG;
1528 			goto done_deref_locked;
1529 		}
1530 		FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1531 			if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1532 			    sizeof(pr->pr_name)) {
1533 				error = ENAMETOOLONG;
1534 				goto done_deref_locked;
1535 			}
1536 		}
1537 	}
1538 	pr_allow_diff = pr_allow & ~ppr->pr_allow;
1539 	if (pr_allow_diff & ~PR_ALLOW_DIFFERENCES) {
1540 		error = EPERM;
1541 		goto done_deref_locked;
1542 	}
1543 
1544 	/*
1545 	 * Let modules check their parameters.  This requires unlocking and
1546 	 * then re-locking the prison, but this is still a valid state as long
1547 	 * as allprison_lock remains xlocked.
1548 	 */
1549 	mtx_unlock(&pr->pr_mtx);
1550 	error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1551 	if (error != 0) {
1552 		prison_deref(pr, created
1553 		    ? PD_LIST_XLOCKED
1554 		    : PD_DEREF | PD_LIST_XLOCKED);
1555 		goto done_releroot;
1556 	}
1557 	mtx_lock(&pr->pr_mtx);
1558 
1559 	/* At this point, all valid parameters should have been noted. */
1560 	TAILQ_FOREACH(opt, opts, link) {
1561 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
1562 			error = EINVAL;
1563 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
1564 			goto done_deref_locked;
1565 		}
1566 	}
1567 
1568 	/* Set the parameters of the prison. */
1569 #ifdef INET
1570 	redo_ip4 = 0;
1571 	if (pr_flags & PR_IP4_USER) {
1572 		pr->pr_flags |= PR_IP4;
1573 		free(pr->pr_ip4, M_PRISON);
1574 		pr->pr_ip4s = ip4s;
1575 		pr->pr_ip4 = ip4;
1576 		ip4 = NULL;
1577 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1578 #ifdef VIMAGE
1579 			if (tpr->pr_flags & PR_VNET) {
1580 				descend = 0;
1581 				continue;
1582 			}
1583 #endif
1584 			if (prison_restrict_ip4(tpr, NULL)) {
1585 				redo_ip4 = 1;
1586 				descend = 0;
1587 			}
1588 		}
1589 	}
1590 #endif
1591 #ifdef INET6
1592 	redo_ip6 = 0;
1593 	if (pr_flags & PR_IP6_USER) {
1594 		pr->pr_flags |= PR_IP6;
1595 		free(pr->pr_ip6, M_PRISON);
1596 		pr->pr_ip6s = ip6s;
1597 		pr->pr_ip6 = ip6;
1598 		ip6 = NULL;
1599 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1600 #ifdef VIMAGE
1601 			if (tpr->pr_flags & PR_VNET) {
1602 				descend = 0;
1603 				continue;
1604 			}
1605 #endif
1606 			if (prison_restrict_ip6(tpr, NULL)) {
1607 				redo_ip6 = 1;
1608 				descend = 0;
1609 			}
1610 		}
1611 	}
1612 #endif
1613 	if (gotslevel) {
1614 		pr->pr_securelevel = slevel;
1615 		/* Set all child jails to be at least this level. */
1616 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1617 			if (tpr->pr_securelevel < slevel)
1618 				tpr->pr_securelevel = slevel;
1619 	}
1620 	if (gotchildmax) {
1621 		pr->pr_childmax = childmax;
1622 		/* Set all child jails to under this limit. */
1623 		FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1624 			if (tpr->pr_childmax > childmax - level)
1625 				tpr->pr_childmax = childmax > level
1626 				    ? childmax - level : 0;
1627 	}
1628 	if (gotenforce) {
1629 		pr->pr_enforce_statfs = enforce;
1630 		/* Pass this restriction on to the children. */
1631 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1632 			if (tpr->pr_enforce_statfs < enforce)
1633 				tpr->pr_enforce_statfs = enforce;
1634 	}
1635 	if (gotrsnum) {
1636 		pr->pr_devfs_rsnum = rsnum;
1637 		/* Pass this restriction on to the children. */
1638 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1639 			tpr->pr_devfs_rsnum = rsnum;
1640 	}
1641 	if (namelc != NULL) {
1642 		if (ppr == &prison0)
1643 			strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
1644 		else
1645 			snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1646 			    ppr->pr_name, namelc);
1647 		/* Change this component of child names. */
1648 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1649 			bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1650 			    strlen(tpr->pr_name + onamelen) + 1);
1651 			bcopy(pr->pr_name, tpr->pr_name, namelen);
1652 		}
1653 	}
1654 	if (path != NULL) {
1655 		/* Try to keep a real-rooted full pathname. */
1656 		if (fullpath_disabled && path[0] == '/' &&
1657 		    strcmp(mypr->pr_path, "/"))
1658 			snprintf(pr->pr_path, sizeof(pr->pr_path), "%s%s",
1659 			    mypr->pr_path, path);
1660 		else
1661 			strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1662 		pr->pr_root = root;
1663 	}
1664 	if (PR_HOST & ch_flags & ~pr_flags) {
1665 		if (pr->pr_flags & PR_HOST) {
1666 			/*
1667 			 * Copy the parent's host info.  As with pr_ip4 above,
1668 			 * the lack of a lock on the parent is not a problem;
1669 			 * it is always set with allprison_lock at least
1670 			 * shared, and is held exclusively here.
1671 			 */
1672 			strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1673 			    sizeof(pr->pr_hostname));
1674 			strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1675 			    sizeof(pr->pr_domainname));
1676 			strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1677 			    sizeof(pr->pr_hostuuid));
1678 			pr->pr_hostid = pr->pr_parent->pr_hostid;
1679 		}
1680 	} else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1681 		/* Set this prison, and any descendants without PR_HOST. */
1682 		if (host != NULL)
1683 			strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1684 		if (domain != NULL)
1685 			strlcpy(pr->pr_domainname, domain,
1686 			    sizeof(pr->pr_domainname));
1687 		if (uuid != NULL)
1688 			strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1689 		if (gothid)
1690 			pr->pr_hostid = hid;
1691 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1692 			if (tpr->pr_flags & PR_HOST)
1693 				descend = 0;
1694 			else {
1695 				if (host != NULL)
1696 					strlcpy(tpr->pr_hostname,
1697 					    pr->pr_hostname,
1698 					    sizeof(tpr->pr_hostname));
1699 				if (domain != NULL)
1700 					strlcpy(tpr->pr_domainname,
1701 					    pr->pr_domainname,
1702 					    sizeof(tpr->pr_domainname));
1703 				if (uuid != NULL)
1704 					strlcpy(tpr->pr_hostuuid,
1705 					    pr->pr_hostuuid,
1706 					    sizeof(tpr->pr_hostuuid));
1707 				if (gothid)
1708 					tpr->pr_hostid = hid;
1709 			}
1710 		}
1711 	}
1712 	if ((tallow = ch_allow & ~pr_allow)) {
1713 		/* Clear allow bits in all children. */
1714 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1715 			tpr->pr_allow &= ~tallow;
1716 	}
1717 	pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1718 	/*
1719 	 * Persistent prisons get an extra reference, and prisons losing their
1720 	 * persist flag lose that reference.  Only do this for existing prisons
1721 	 * for now, so new ones will remain unseen until after the module
1722 	 * handlers have completed.
1723 	 */
1724 	born = pr->pr_uref == 0;
1725 	if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) {
1726 		if (pr_flags & PR_PERSIST) {
1727 			pr->pr_ref++;
1728 			pr->pr_uref++;
1729 		} else {
1730 			pr->pr_ref--;
1731 			pr->pr_uref--;
1732 		}
1733 	}
1734 	pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1735 	mtx_unlock(&pr->pr_mtx);
1736 
1737 #ifdef RACCT
1738 	if (racct_enable && created)
1739 		prison_racct_attach(pr);
1740 #endif
1741 
1742 	/* Locks may have prevented a complete restriction of child IP
1743 	 * addresses.  If so, allocate some more memory and try again.
1744 	 */
1745 #ifdef INET
1746 	while (redo_ip4) {
1747 		ip4s = pr->pr_ip4s;
1748 		ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1749 		mtx_lock(&pr->pr_mtx);
1750 		redo_ip4 = 0;
1751 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1752 #ifdef VIMAGE
1753 			if (tpr->pr_flags & PR_VNET) {
1754 				descend = 0;
1755 				continue;
1756 			}
1757 #endif
1758 			if (prison_restrict_ip4(tpr, ip4)) {
1759 				if (ip4 != NULL)
1760 					ip4 = NULL;
1761 				else
1762 					redo_ip4 = 1;
1763 			}
1764 		}
1765 		mtx_unlock(&pr->pr_mtx);
1766 	}
1767 #endif
1768 #ifdef INET6
1769 	while (redo_ip6) {
1770 		ip6s = pr->pr_ip6s;
1771 		ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1772 		mtx_lock(&pr->pr_mtx);
1773 		redo_ip6 = 0;
1774 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1775 #ifdef VIMAGE
1776 			if (tpr->pr_flags & PR_VNET) {
1777 				descend = 0;
1778 				continue;
1779 			}
1780 #endif
1781 			if (prison_restrict_ip6(tpr, ip6)) {
1782 				if (ip6 != NULL)
1783 					ip6 = NULL;
1784 				else
1785 					redo_ip6 = 1;
1786 			}
1787 		}
1788 		mtx_unlock(&pr->pr_mtx);
1789 	}
1790 #endif
1791 
1792 	/* Let the modules do their work. */
1793 	sx_downgrade(&allprison_lock);
1794 	if (born) {
1795 		error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1796 		if (error) {
1797 			(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1798 			prison_deref(pr, created
1799 			    ? PD_LIST_SLOCKED
1800 			    : PD_DEREF | PD_LIST_SLOCKED);
1801 			goto done_errmsg;
1802 		}
1803 	}
1804 	error = osd_jail_call(pr, PR_METHOD_SET, opts);
1805 	if (error) {
1806 		if (born)
1807 			(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1808 		prison_deref(pr, created
1809 		    ? PD_LIST_SLOCKED
1810 		    : PD_DEREF | PD_LIST_SLOCKED);
1811 		goto done_errmsg;
1812 	}
1813 
1814 	/* Attach this process to the prison if requested. */
1815 	if (flags & JAIL_ATTACH) {
1816 		mtx_lock(&pr->pr_mtx);
1817 		error = do_jail_attach(td, pr);
1818 		if (error) {
1819 			vfs_opterror(opts, "attach failed");
1820 			if (!created)
1821 				prison_deref(pr, PD_DEREF);
1822 			goto done_errmsg;
1823 		}
1824 	}
1825 
1826 #ifdef RACCT
1827 	if (racct_enable && !created) {
1828 		if (!(flags & JAIL_ATTACH))
1829 			sx_sunlock(&allprison_lock);
1830 		prison_racct_modify(pr);
1831 		if (!(flags & JAIL_ATTACH))
1832 			sx_slock(&allprison_lock);
1833 	}
1834 #endif
1835 
1836 	td->td_retval[0] = pr->pr_id;
1837 
1838 	/*
1839 	 * Now that it is all there, drop the temporary reference from existing
1840 	 * prisons.  Or add a reference to newly created persistent prisons
1841 	 * (which was not done earlier so that the prison would not be publicly
1842 	 * visible).
1843 	 */
1844 	if (!created) {
1845 		prison_deref(pr, (flags & JAIL_ATTACH)
1846 		    ? PD_DEREF
1847 		    : PD_DEREF | PD_LIST_SLOCKED);
1848 	} else {
1849 		if (pr_flags & PR_PERSIST) {
1850 			mtx_lock(&pr->pr_mtx);
1851 			pr->pr_ref++;
1852 			pr->pr_uref++;
1853 			mtx_unlock(&pr->pr_mtx);
1854 		}
1855 		if (!(flags & JAIL_ATTACH))
1856 			sx_sunlock(&allprison_lock);
1857 	}
1858 
1859 	goto done_free;
1860 
1861  done_deref_locked:
1862 	prison_deref(pr, created
1863 	    ? PD_LOCKED | PD_LIST_XLOCKED
1864 	    : PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
1865 	goto done_releroot;
1866  done_unlock_list:
1867 	sx_xunlock(&allprison_lock);
1868  done_releroot:
1869 	if (root != NULL)
1870 		vrele(root);
1871  done_errmsg:
1872 	if (error) {
1873 		if (vfs_getopt(opts, "errmsg", (void **)&errmsg,
1874 		    &errmsg_len) == 0 && errmsg_len > 0) {
1875 			errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1876 			if (optuio->uio_segflg == UIO_SYSSPACE)
1877 				bcopy(errmsg,
1878 				    optuio->uio_iov[errmsg_pos].iov_base,
1879 				    errmsg_len);
1880 			else
1881 				copyout(errmsg,
1882 				    optuio->uio_iov[errmsg_pos].iov_base,
1883 				    errmsg_len);
1884 		}
1885 	}
1886  done_free:
1887 #ifdef INET
1888 	free(ip4, M_PRISON);
1889 #endif
1890 #ifdef INET6
1891 	free(ip6, M_PRISON);
1892 #endif
1893 	if (g_path != NULL)
1894 		free(g_path, M_TEMP);
1895 	vfs_freeopts(opts);
1896 	return (error);
1897 }
1898 
1899 
1900 /*
1901  * struct jail_get_args {
1902  *	struct iovec *iovp;
1903  *	unsigned int iovcnt;
1904  *	int flags;
1905  * };
1906  */
1907 int
1908 sys_jail_get(struct thread *td, struct jail_get_args *uap)
1909 {
1910 	struct uio *auio;
1911 	int error;
1912 
1913 	/* Check that we have an even number of iovecs. */
1914 	if (uap->iovcnt & 1)
1915 		return (EINVAL);
1916 
1917 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
1918 	if (error)
1919 		return (error);
1920 	error = kern_jail_get(td, auio, uap->flags);
1921 	if (error == 0)
1922 		error = copyout(auio->uio_iov, uap->iovp,
1923 		    uap->iovcnt * sizeof (struct iovec));
1924 	free(auio, M_IOV);
1925 	return (error);
1926 }
1927 
1928 int
1929 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
1930 {
1931 	struct bool_flags *bf;
1932 	struct jailsys_flags *jsf;
1933 	struct prison *pr, *mypr;
1934 	struct vfsopt *opt;
1935 	struct vfsoptlist *opts;
1936 	char *errmsg, *name;
1937 	int error, errmsg_len, errmsg_pos, i, jid, len, locked, pos;
1938 	unsigned f;
1939 
1940 	if (flags & ~JAIL_GET_MASK)
1941 		return (EINVAL);
1942 
1943 	/* Get the parameter list. */
1944 	error = vfs_buildopts(optuio, &opts);
1945 	if (error)
1946 		return (error);
1947 	errmsg_pos = vfs_getopt_pos(opts, "errmsg");
1948 	mypr = td->td_ucred->cr_prison;
1949 
1950 	/*
1951 	 * Find the prison specified by one of: lastjid, jid, name.
1952 	 */
1953 	sx_slock(&allprison_lock);
1954 	error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
1955 	if (error == 0) {
1956 		TAILQ_FOREACH(pr, &allprison, pr_list) {
1957 			if (pr->pr_id > jid && prison_ischild(mypr, pr)) {
1958 				mtx_lock(&pr->pr_mtx);
1959 				if (pr->pr_ref > 0 &&
1960 				    (pr->pr_uref > 0 || (flags & JAIL_DYING)))
1961 					break;
1962 				mtx_unlock(&pr->pr_mtx);
1963 			}
1964 		}
1965 		if (pr != NULL)
1966 			goto found_prison;
1967 		error = ENOENT;
1968 		vfs_opterror(opts, "no jail after %d", jid);
1969 		goto done_unlock_list;
1970 	} else if (error != ENOENT)
1971 		goto done_unlock_list;
1972 
1973 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1974 	if (error == 0) {
1975 		if (jid != 0) {
1976 			pr = prison_find_child(mypr, jid);
1977 			if (pr != NULL) {
1978 				if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1979 					mtx_unlock(&pr->pr_mtx);
1980 					error = ENOENT;
1981 					vfs_opterror(opts, "jail %d is dying",
1982 					    jid);
1983 					goto done_unlock_list;
1984 				}
1985 				goto found_prison;
1986 			}
1987 			error = ENOENT;
1988 			vfs_opterror(opts, "jail %d not found", jid);
1989 			goto done_unlock_list;
1990 		}
1991 	} else if (error != ENOENT)
1992 		goto done_unlock_list;
1993 
1994 	error = vfs_getopt(opts, "name", (void **)&name, &len);
1995 	if (error == 0) {
1996 		if (len == 0 || name[len - 1] != '\0') {
1997 			error = EINVAL;
1998 			goto done_unlock_list;
1999 		}
2000 		pr = prison_find_name(mypr, name);
2001 		if (pr != NULL) {
2002 			if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
2003 				mtx_unlock(&pr->pr_mtx);
2004 				error = ENOENT;
2005 				vfs_opterror(opts, "jail \"%s\" is dying",
2006 				    name);
2007 				goto done_unlock_list;
2008 			}
2009 			goto found_prison;
2010 		}
2011 		error = ENOENT;
2012 		vfs_opterror(opts, "jail \"%s\" not found", name);
2013 		goto done_unlock_list;
2014 	} else if (error != ENOENT)
2015 		goto done_unlock_list;
2016 
2017 	vfs_opterror(opts, "no jail specified");
2018 	error = ENOENT;
2019 	goto done_unlock_list;
2020 
2021  found_prison:
2022 	/* Get the parameters of the prison. */
2023 	pr->pr_ref++;
2024 	locked = PD_LOCKED;
2025 	td->td_retval[0] = pr->pr_id;
2026 	error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2027 	if (error != 0 && error != ENOENT)
2028 		goto done_deref;
2029 	i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2030 	error = vfs_setopt(opts, "parent", &i, sizeof(i));
2031 	if (error != 0 && error != ENOENT)
2032 		goto done_deref;
2033 	error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2034 	if (error != 0 && error != ENOENT)
2035 		goto done_deref;
2036 	error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2037 	    sizeof(pr->pr_cpuset->cs_id));
2038 	if (error != 0 && error != ENOENT)
2039 		goto done_deref;
2040 	error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2041 	if (error != 0 && error != ENOENT)
2042 		goto done_deref;
2043 #ifdef INET
2044 	error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
2045 	    pr->pr_ip4s * sizeof(*pr->pr_ip4));
2046 	if (error != 0 && error != ENOENT)
2047 		goto done_deref;
2048 #endif
2049 #ifdef INET6
2050 	error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
2051 	    pr->pr_ip6s * sizeof(*pr->pr_ip6));
2052 	if (error != 0 && error != ENOENT)
2053 		goto done_deref;
2054 #endif
2055 	error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2056 	    sizeof(pr->pr_securelevel));
2057 	if (error != 0 && error != ENOENT)
2058 		goto done_deref;
2059 	error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2060 	    sizeof(pr->pr_childcount));
2061 	if (error != 0 && error != ENOENT)
2062 		goto done_deref;
2063 	error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2064 	    sizeof(pr->pr_childmax));
2065 	if (error != 0 && error != ENOENT)
2066 		goto done_deref;
2067 	error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2068 	if (error != 0 && error != ENOENT)
2069 		goto done_deref;
2070 	error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2071 	if (error != 0 && error != ENOENT)
2072 		goto done_deref;
2073 	error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2074 	if (error != 0 && error != ENOENT)
2075 		goto done_deref;
2076 #ifdef COMPAT_FREEBSD32
2077 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2078 		uint32_t hid32 = pr->pr_hostid;
2079 
2080 		error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2081 	} else
2082 #endif
2083 	error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2084 	    sizeof(pr->pr_hostid));
2085 	if (error != 0 && error != ENOENT)
2086 		goto done_deref;
2087 	error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2088 	    sizeof(pr->pr_enforce_statfs));
2089 	if (error != 0 && error != ENOENT)
2090 		goto done_deref;
2091 	error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2092 	    sizeof(pr->pr_devfs_rsnum));
2093 	if (error != 0 && error != ENOENT)
2094 		goto done_deref;
2095 	for (bf = pr_flag_bool;
2096 	     bf < pr_flag_bool + nitems(pr_flag_bool);
2097 	     bf++) {
2098 		i = (pr->pr_flags & bf->flag) ? 1 : 0;
2099 		error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2100 		if (error != 0 && error != ENOENT)
2101 			goto done_deref;
2102 		i = !i;
2103 		error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2104 		if (error != 0 && error != ENOENT)
2105 			goto done_deref;
2106 	}
2107 	for (jsf = pr_flag_jailsys;
2108 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
2109 	     jsf++) {
2110 		f = pr->pr_flags & (jsf->disable | jsf->new);
2111 		i = (f != 0 && f == jsf->disable) ? JAIL_SYS_DISABLE
2112 		    : (f == jsf->new) ? JAIL_SYS_NEW
2113 		    : JAIL_SYS_INHERIT;
2114 		error = vfs_setopt(opts, jsf->name, &i, sizeof(i));
2115 		if (error != 0 && error != ENOENT)
2116 			goto done_deref;
2117 	}
2118 	for (bf = pr_flag_allow;
2119 	     bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
2120 	     bf++) {
2121 		i = (pr->pr_allow & bf->flag) ? 1 : 0;
2122 		error = vfs_setopt(opts, bf->name, &i, sizeof(i));
2123 		if (error != 0 && error != ENOENT)
2124 			goto done_deref;
2125 		i = !i;
2126 		error = vfs_setopt(opts, bf->noname, &i, sizeof(i));
2127 		if (error != 0 && error != ENOENT)
2128 			goto done_deref;
2129 	}
2130 	i = (pr->pr_uref == 0);
2131 	error = vfs_setopt(opts, "dying", &i, sizeof(i));
2132 	if (error != 0 && error != ENOENT)
2133 		goto done_deref;
2134 	i = !i;
2135 	error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2136 	if (error != 0 && error != ENOENT)
2137 		goto done_deref;
2138 	error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2139 	    sizeof(pr->pr_osreldate));
2140 	if (error != 0 && error != ENOENT)
2141 		goto done_deref;
2142 	error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2143 	if (error != 0 && error != ENOENT)
2144 		goto done_deref;
2145 
2146 	/* Get the module parameters. */
2147 	mtx_unlock(&pr->pr_mtx);
2148 	locked = 0;
2149 	error = osd_jail_call(pr, PR_METHOD_GET, opts);
2150 	if (error)
2151 		goto done_deref;
2152 	prison_deref(pr, PD_DEREF | PD_LIST_SLOCKED);
2153 
2154 	/* By now, all parameters should have been noted. */
2155 	TAILQ_FOREACH(opt, opts, link) {
2156 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
2157 			error = EINVAL;
2158 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
2159 			goto done_errmsg;
2160 		}
2161 	}
2162 
2163 	/* Write the fetched parameters back to userspace. */
2164 	error = 0;
2165 	TAILQ_FOREACH(opt, opts, link) {
2166 		if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2167 			pos = 2 * opt->pos + 1;
2168 			optuio->uio_iov[pos].iov_len = opt->len;
2169 			if (opt->value != NULL) {
2170 				if (optuio->uio_segflg == UIO_SYSSPACE) {
2171 					bcopy(opt->value,
2172 					    optuio->uio_iov[pos].iov_base,
2173 					    opt->len);
2174 				} else {
2175 					error = copyout(opt->value,
2176 					    optuio->uio_iov[pos].iov_base,
2177 					    opt->len);
2178 					if (error)
2179 						break;
2180 				}
2181 			}
2182 		}
2183 	}
2184 	goto done_errmsg;
2185 
2186  done_deref:
2187 	prison_deref(pr, locked | PD_DEREF | PD_LIST_SLOCKED);
2188 	goto done_errmsg;
2189 
2190  done_unlock_list:
2191 	sx_sunlock(&allprison_lock);
2192  done_errmsg:
2193 	if (error && errmsg_pos >= 0) {
2194 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2195 		errmsg_pos = 2 * errmsg_pos + 1;
2196 		if (errmsg_len > 0) {
2197 			if (optuio->uio_segflg == UIO_SYSSPACE)
2198 				bcopy(errmsg,
2199 				    optuio->uio_iov[errmsg_pos].iov_base,
2200 				    errmsg_len);
2201 			else
2202 				copyout(errmsg,
2203 				    optuio->uio_iov[errmsg_pos].iov_base,
2204 				    errmsg_len);
2205 		}
2206 	}
2207 	vfs_freeopts(opts);
2208 	return (error);
2209 }
2210 
2211 
2212 /*
2213  * struct jail_remove_args {
2214  *	int jid;
2215  * };
2216  */
2217 int
2218 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2219 {
2220 	struct prison *pr, *cpr, *lpr, *tpr;
2221 	int descend, error;
2222 
2223 	error = priv_check(td, PRIV_JAIL_REMOVE);
2224 	if (error)
2225 		return (error);
2226 
2227 	sx_xlock(&allprison_lock);
2228 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2229 	if (pr == NULL) {
2230 		sx_xunlock(&allprison_lock);
2231 		return (EINVAL);
2232 	}
2233 
2234 	/* Remove all descendants of this prison, then remove this prison. */
2235 	pr->pr_ref++;
2236 	if (!LIST_EMPTY(&pr->pr_children)) {
2237 		mtx_unlock(&pr->pr_mtx);
2238 		lpr = NULL;
2239 		FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
2240 			mtx_lock(&cpr->pr_mtx);
2241 			if (cpr->pr_ref > 0) {
2242 				tpr = cpr;
2243 				cpr->pr_ref++;
2244 			} else {
2245 				/* Already removed - do not do it again. */
2246 				tpr = NULL;
2247 			}
2248 			mtx_unlock(&cpr->pr_mtx);
2249 			if (lpr != NULL) {
2250 				mtx_lock(&lpr->pr_mtx);
2251 				prison_remove_one(lpr);
2252 				sx_xlock(&allprison_lock);
2253 			}
2254 			lpr = tpr;
2255 		}
2256 		if (lpr != NULL) {
2257 			mtx_lock(&lpr->pr_mtx);
2258 			prison_remove_one(lpr);
2259 			sx_xlock(&allprison_lock);
2260 		}
2261 		mtx_lock(&pr->pr_mtx);
2262 	}
2263 	prison_remove_one(pr);
2264 	return (0);
2265 }
2266 
2267 static void
2268 prison_remove_one(struct prison *pr)
2269 {
2270 	struct proc *p;
2271 	int deuref;
2272 
2273 	/* If the prison was persistent, it is not anymore. */
2274 	deuref = 0;
2275 	if (pr->pr_flags & PR_PERSIST) {
2276 		pr->pr_ref--;
2277 		deuref = PD_DEUREF;
2278 		pr->pr_flags &= ~PR_PERSIST;
2279 	}
2280 
2281 	/*
2282 	 * jail_remove added a reference.  If that's the only one, remove
2283 	 * the prison now.
2284 	 */
2285 	KASSERT(pr->pr_ref > 0,
2286 	    ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id));
2287 	if (pr->pr_ref == 1) {
2288 		prison_deref(pr,
2289 		    deuref | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
2290 		return;
2291 	}
2292 
2293 	mtx_unlock(&pr->pr_mtx);
2294 	sx_xunlock(&allprison_lock);
2295 	/*
2296 	 * Kill all processes unfortunate enough to be attached to this prison.
2297 	 */
2298 	sx_slock(&allproc_lock);
2299 	FOREACH_PROC_IN_SYSTEM(p) {
2300 		PROC_LOCK(p);
2301 		if (p->p_state != PRS_NEW && p->p_ucred &&
2302 		    p->p_ucred->cr_prison == pr)
2303 			kern_psignal(p, SIGKILL);
2304 		PROC_UNLOCK(p);
2305 	}
2306 	sx_sunlock(&allproc_lock);
2307 	/* Remove the temporary reference added by jail_remove. */
2308 	prison_deref(pr, deuref | PD_DEREF);
2309 }
2310 
2311 
2312 /*
2313  * struct jail_attach_args {
2314  *	int jid;
2315  * };
2316  */
2317 int
2318 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2319 {
2320 	struct prison *pr;
2321 	int error;
2322 
2323 	error = priv_check(td, PRIV_JAIL_ATTACH);
2324 	if (error)
2325 		return (error);
2326 
2327 	/*
2328 	 * Start with exclusive hold on allprison_lock to ensure that a possible
2329 	 * PR_METHOD_REMOVE call isn't concurrent with jail_set or jail_remove.
2330 	 * But then immediately downgrade it since we don't need to stop
2331 	 * readers.
2332 	 */
2333 	sx_xlock(&allprison_lock);
2334 	sx_downgrade(&allprison_lock);
2335 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2336 	if (pr == NULL) {
2337 		sx_sunlock(&allprison_lock);
2338 		return (EINVAL);
2339 	}
2340 
2341 	/*
2342 	 * Do not allow a process to attach to a prison that is not
2343 	 * considered to be "alive".
2344 	 */
2345 	if (pr->pr_uref == 0) {
2346 		mtx_unlock(&pr->pr_mtx);
2347 		sx_sunlock(&allprison_lock);
2348 		return (EINVAL);
2349 	}
2350 
2351 	return (do_jail_attach(td, pr));
2352 }
2353 
2354 static int
2355 do_jail_attach(struct thread *td, struct prison *pr)
2356 {
2357 	struct proc *p;
2358 	struct ucred *newcred, *oldcred;
2359 	int error;
2360 
2361 	/*
2362 	 * XXX: Note that there is a slight race here if two threads
2363 	 * in the same privileged process attempt to attach to two
2364 	 * different jails at the same time.  It is important for
2365 	 * user processes not to do this, or they might end up with
2366 	 * a process root from one prison, but attached to the jail
2367 	 * of another.
2368 	 */
2369 	pr->pr_ref++;
2370 	pr->pr_uref++;
2371 	mtx_unlock(&pr->pr_mtx);
2372 
2373 	/* Let modules do whatever they need to prepare for attaching. */
2374 	error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2375 	if (error) {
2376 		prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2377 		return (error);
2378 	}
2379 	sx_sunlock(&allprison_lock);
2380 
2381 	/*
2382 	 * Reparent the newly attached process to this jail.
2383 	 */
2384 	p = td->td_proc;
2385 	error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2386 	if (error)
2387 		goto e_revert_osd;
2388 
2389 	vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2390 	if ((error = change_dir(pr->pr_root, td)) != 0)
2391 		goto e_unlock;
2392 #ifdef MAC
2393 	if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2394 		goto e_unlock;
2395 #endif
2396 	VOP_UNLOCK(pr->pr_root, 0);
2397 	if ((error = pwd_chroot(td, pr->pr_root)))
2398 		goto e_revert_osd;
2399 
2400 	newcred = crget();
2401 	PROC_LOCK(p);
2402 	oldcred = crcopysafe(p, newcred);
2403 	newcred->cr_prison = pr;
2404 	proc_set_cred(p, newcred);
2405 	setsugid(p);
2406 #ifdef RACCT
2407 	racct_proc_ucred_changed(p, oldcred, newcred);
2408 	crhold(newcred);
2409 #endif
2410 	PROC_UNLOCK(p);
2411 #ifdef RCTL
2412 	rctl_proc_ucred_changed(p, newcred);
2413 	crfree(newcred);
2414 #endif
2415 	prison_deref(oldcred->cr_prison, PD_DEREF | PD_DEUREF);
2416 	crfree(oldcred);
2417 	return (0);
2418 
2419  e_unlock:
2420 	VOP_UNLOCK(pr->pr_root, 0);
2421  e_revert_osd:
2422 	/* Tell modules this thread is still in its old jail after all. */
2423 	(void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
2424 	prison_deref(pr, PD_DEREF | PD_DEUREF);
2425 	return (error);
2426 }
2427 
2428 
2429 /*
2430  * Returns a locked prison instance, or NULL on failure.
2431  */
2432 struct prison *
2433 prison_find(int prid)
2434 {
2435 	struct prison *pr;
2436 
2437 	sx_assert(&allprison_lock, SX_LOCKED);
2438 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2439 		if (pr->pr_id == prid) {
2440 			mtx_lock(&pr->pr_mtx);
2441 			if (pr->pr_ref > 0)
2442 				return (pr);
2443 			mtx_unlock(&pr->pr_mtx);
2444 		}
2445 	}
2446 	return (NULL);
2447 }
2448 
2449 /*
2450  * Find a prison that is a descendant of mypr.  Returns a locked prison or NULL.
2451  */
2452 struct prison *
2453 prison_find_child(struct prison *mypr, int prid)
2454 {
2455 	struct prison *pr;
2456 	int descend;
2457 
2458 	sx_assert(&allprison_lock, SX_LOCKED);
2459 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2460 		if (pr->pr_id == prid) {
2461 			mtx_lock(&pr->pr_mtx);
2462 			if (pr->pr_ref > 0)
2463 				return (pr);
2464 			mtx_unlock(&pr->pr_mtx);
2465 		}
2466 	}
2467 	return (NULL);
2468 }
2469 
2470 /*
2471  * Look for the name relative to mypr.  Returns a locked prison or NULL.
2472  */
2473 struct prison *
2474 prison_find_name(struct prison *mypr, const char *name)
2475 {
2476 	struct prison *pr, *deadpr;
2477 	size_t mylen;
2478 	int descend;
2479 
2480 	sx_assert(&allprison_lock, SX_LOCKED);
2481 	mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2482  again:
2483 	deadpr = NULL;
2484 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2485 		if (!strcmp(pr->pr_name + mylen, name)) {
2486 			mtx_lock(&pr->pr_mtx);
2487 			if (pr->pr_ref > 0) {
2488 				if (pr->pr_uref > 0)
2489 					return (pr);
2490 				deadpr = pr;
2491 			}
2492 			mtx_unlock(&pr->pr_mtx);
2493 		}
2494 	}
2495 	/* There was no valid prison - perhaps there was a dying one. */
2496 	if (deadpr != NULL) {
2497 		mtx_lock(&deadpr->pr_mtx);
2498 		if (deadpr->pr_ref == 0) {
2499 			mtx_unlock(&deadpr->pr_mtx);
2500 			goto again;
2501 		}
2502 	}
2503 	return (deadpr);
2504 }
2505 
2506 /*
2507  * See if a prison has the specific flag set.
2508  */
2509 int
2510 prison_flag(struct ucred *cred, unsigned flag)
2511 {
2512 
2513 	/* This is an atomic read, so no locking is necessary. */
2514 	return (cred->cr_prison->pr_flags & flag);
2515 }
2516 
2517 int
2518 prison_allow(struct ucred *cred, unsigned flag)
2519 {
2520 
2521 	/* This is an atomic read, so no locking is necessary. */
2522 	return (cred->cr_prison->pr_allow & flag);
2523 }
2524 
2525 /*
2526  * Remove a prison reference.  If that was the last reference, remove the
2527  * prison itself - but not in this context in case there are locks held.
2528  */
2529 void
2530 prison_free_locked(struct prison *pr)
2531 {
2532 	int ref;
2533 
2534 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2535 	ref = --pr->pr_ref;
2536 	mtx_unlock(&pr->pr_mtx);
2537 	if (ref == 0)
2538 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2539 }
2540 
2541 void
2542 prison_free(struct prison *pr)
2543 {
2544 
2545 	mtx_lock(&pr->pr_mtx);
2546 	prison_free_locked(pr);
2547 }
2548 
2549 /*
2550  * Complete a call to either prison_free or prison_proc_free.
2551  */
2552 static void
2553 prison_complete(void *context, int pending)
2554 {
2555 	struct prison *pr = context;
2556 
2557 	sx_xlock(&allprison_lock);
2558 	mtx_lock(&pr->pr_mtx);
2559 	prison_deref(pr, pr->pr_uref
2560 	    ? PD_DEREF | PD_DEUREF | PD_LOCKED | PD_LIST_XLOCKED
2561 	    : PD_LOCKED | PD_LIST_XLOCKED);
2562 }
2563 
2564 /*
2565  * Remove a prison reference (usually).  This internal version assumes no
2566  * mutexes are held, except perhaps the prison itself.  If there are no more
2567  * references, release and delist the prison.  On completion, the prison lock
2568  * and the allprison lock are both unlocked.
2569  */
2570 static void
2571 prison_deref(struct prison *pr, int flags)
2572 {
2573 	struct prison *ppr, *tpr;
2574 	int ref, lasturef;
2575 
2576 	if (!(flags & PD_LOCKED))
2577 		mtx_lock(&pr->pr_mtx);
2578 	for (;;) {
2579 		if (flags & PD_DEUREF) {
2580 			KASSERT(pr->pr_uref > 0,
2581 			    ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
2582 			     pr->pr_id));
2583 			pr->pr_uref--;
2584 			lasturef = pr->pr_uref == 0;
2585 			if (lasturef)
2586 				pr->pr_ref++;
2587 			KASSERT(prison0.pr_uref != 0, ("prison0 pr_uref=0"));
2588 		} else
2589 			lasturef = 0;
2590 		if (flags & PD_DEREF) {
2591 			KASSERT(pr->pr_ref > 0,
2592 			    ("prison_deref PD_DEREF on a dead prison (jid=%d)",
2593 			     pr->pr_id));
2594 			pr->pr_ref--;
2595 		}
2596 		ref = pr->pr_ref;
2597 		mtx_unlock(&pr->pr_mtx);
2598 
2599 		/*
2600 		 * Tell the modules if the last user reference was removed
2601 		 * (even it sticks around in dying state).
2602 		 */
2603 		if (lasturef) {
2604 			if (!(flags & (PD_LIST_SLOCKED | PD_LIST_XLOCKED))) {
2605 				sx_xlock(&allprison_lock);
2606 				flags |= PD_LIST_XLOCKED;
2607 			}
2608 			(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
2609 			mtx_lock(&pr->pr_mtx);
2610 			ref = --pr->pr_ref;
2611 			mtx_unlock(&pr->pr_mtx);
2612 		}
2613 
2614 		/* If the prison still has references, nothing else to do. */
2615 		if (ref > 0) {
2616 			if (flags & PD_LIST_SLOCKED)
2617 				sx_sunlock(&allprison_lock);
2618 			else if (flags & PD_LIST_XLOCKED)
2619 				sx_xunlock(&allprison_lock);
2620 			return;
2621 		}
2622 
2623 		if (flags & PD_LIST_SLOCKED) {
2624 			if (!sx_try_upgrade(&allprison_lock)) {
2625 				sx_sunlock(&allprison_lock);
2626 				sx_xlock(&allprison_lock);
2627 			}
2628 		} else if (!(flags & PD_LIST_XLOCKED))
2629 			sx_xlock(&allprison_lock);
2630 
2631 		TAILQ_REMOVE(&allprison, pr, pr_list);
2632 		LIST_REMOVE(pr, pr_sibling);
2633 		ppr = pr->pr_parent;
2634 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
2635 			tpr->pr_childcount--;
2636 		sx_xunlock(&allprison_lock);
2637 
2638 #ifdef VIMAGE
2639 		if (pr->pr_vnet != ppr->pr_vnet)
2640 			vnet_destroy(pr->pr_vnet);
2641 #endif
2642 		if (pr->pr_root != NULL)
2643 			vrele(pr->pr_root);
2644 		mtx_destroy(&pr->pr_mtx);
2645 #ifdef INET
2646 		free(pr->pr_ip4, M_PRISON);
2647 #endif
2648 #ifdef INET6
2649 		free(pr->pr_ip6, M_PRISON);
2650 #endif
2651 		if (pr->pr_cpuset != NULL)
2652 			cpuset_rel(pr->pr_cpuset);
2653 		osd_jail_exit(pr);
2654 #ifdef RACCT
2655 		if (racct_enable)
2656 			prison_racct_detach(pr);
2657 #endif
2658 		free(pr, M_PRISON);
2659 
2660 		/* Removing a prison frees a reference on its parent. */
2661 		pr = ppr;
2662 		mtx_lock(&pr->pr_mtx);
2663 		flags = PD_DEREF | PD_DEUREF;
2664 	}
2665 }
2666 
2667 void
2668 prison_hold_locked(struct prison *pr)
2669 {
2670 
2671 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2672 	KASSERT(pr->pr_ref > 0,
2673 	    ("Trying to hold dead prison %p (jid=%d).", pr, pr->pr_id));
2674 	pr->pr_ref++;
2675 }
2676 
2677 void
2678 prison_hold(struct prison *pr)
2679 {
2680 
2681 	mtx_lock(&pr->pr_mtx);
2682 	prison_hold_locked(pr);
2683 	mtx_unlock(&pr->pr_mtx);
2684 }
2685 
2686 void
2687 prison_proc_hold(struct prison *pr)
2688 {
2689 
2690 	mtx_lock(&pr->pr_mtx);
2691 	KASSERT(pr->pr_uref > 0,
2692 	    ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2693 	pr->pr_uref++;
2694 	mtx_unlock(&pr->pr_mtx);
2695 }
2696 
2697 void
2698 prison_proc_free(struct prison *pr)
2699 {
2700 
2701 	mtx_lock(&pr->pr_mtx);
2702 	KASSERT(pr->pr_uref > 0,
2703 	    ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2704 	if (pr->pr_uref > 1)
2705 		pr->pr_uref--;
2706 	else {
2707 		/*
2708 		 * Don't remove the last user reference in this context, which
2709 		 * is expected to be a process that is not only locked, but
2710 		 * also half dead.
2711 		 */
2712 		pr->pr_ref++;
2713 		mtx_unlock(&pr->pr_mtx);
2714 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2715 		return;
2716 	}
2717 	mtx_unlock(&pr->pr_mtx);
2718 }
2719 
2720 /*
2721  * Check if a jail supports the given address family.
2722  *
2723  * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
2724  * if not.
2725  */
2726 int
2727 prison_check_af(struct ucred *cred, int af)
2728 {
2729 	struct prison *pr;
2730 	int error;
2731 
2732 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2733 
2734 	pr = cred->cr_prison;
2735 #ifdef VIMAGE
2736 	/* Prisons with their own network stack are not limited. */
2737 	if (prison_owns_vnet(cred))
2738 		return (0);
2739 #endif
2740 
2741 	error = 0;
2742 	switch (af)
2743 	{
2744 #ifdef INET
2745 	case AF_INET:
2746 		if (pr->pr_flags & PR_IP4)
2747 		{
2748 			mtx_lock(&pr->pr_mtx);
2749 			if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL)
2750 				error = EAFNOSUPPORT;
2751 			mtx_unlock(&pr->pr_mtx);
2752 		}
2753 		break;
2754 #endif
2755 #ifdef INET6
2756 	case AF_INET6:
2757 		if (pr->pr_flags & PR_IP6)
2758 		{
2759 			mtx_lock(&pr->pr_mtx);
2760 			if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL)
2761 				error = EAFNOSUPPORT;
2762 			mtx_unlock(&pr->pr_mtx);
2763 		}
2764 		break;
2765 #endif
2766 	case AF_LOCAL:
2767 	case AF_ROUTE:
2768 		break;
2769 	default:
2770 		if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
2771 			error = EAFNOSUPPORT;
2772 	}
2773 	return (error);
2774 }
2775 
2776 /*
2777  * Check if given address belongs to the jail referenced by cred (wrapper to
2778  * prison_check_ip[46]).
2779  *
2780  * Returns 0 if jail doesn't restrict the address family or if address belongs
2781  * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
2782  * the jail doesn't allow the address family.  IPv4 Address passed in in NBO.
2783  */
2784 int
2785 prison_if(struct ucred *cred, struct sockaddr *sa)
2786 {
2787 #ifdef INET
2788 	struct sockaddr_in *sai;
2789 #endif
2790 #ifdef INET6
2791 	struct sockaddr_in6 *sai6;
2792 #endif
2793 	int error;
2794 
2795 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2796 	KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
2797 
2798 #ifdef VIMAGE
2799 	if (prison_owns_vnet(cred))
2800 		return (0);
2801 #endif
2802 
2803 	error = 0;
2804 	switch (sa->sa_family)
2805 	{
2806 #ifdef INET
2807 	case AF_INET:
2808 		sai = (struct sockaddr_in *)sa;
2809 		error = prison_check_ip4(cred, &sai->sin_addr);
2810 		break;
2811 #endif
2812 #ifdef INET6
2813 	case AF_INET6:
2814 		sai6 = (struct sockaddr_in6 *)sa;
2815 		error = prison_check_ip6(cred, &sai6->sin6_addr);
2816 		break;
2817 #endif
2818 	default:
2819 		if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
2820 			error = EAFNOSUPPORT;
2821 	}
2822 	return (error);
2823 }
2824 
2825 /*
2826  * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
2827  */
2828 int
2829 prison_check(struct ucred *cred1, struct ucred *cred2)
2830 {
2831 
2832 	return ((cred1->cr_prison == cred2->cr_prison ||
2833 	    prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
2834 }
2835 
2836 /*
2837  * Return 1 if p2 is a child of p1, otherwise 0.
2838  */
2839 int
2840 prison_ischild(struct prison *pr1, struct prison *pr2)
2841 {
2842 
2843 	for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
2844 		if (pr1 == pr2)
2845 			return (1);
2846 	return (0);
2847 }
2848 
2849 /*
2850  * Return 1 if the passed credential is in a jail, otherwise 0.
2851  */
2852 int
2853 jailed(struct ucred *cred)
2854 {
2855 
2856 	return (cred->cr_prison != &prison0);
2857 }
2858 
2859 /*
2860  * Return 1 if the passed credential is in a jail and that jail does not
2861  * have its own virtual network stack, otherwise 0.
2862  */
2863 int
2864 jailed_without_vnet(struct ucred *cred)
2865 {
2866 
2867 	if (!jailed(cred))
2868 		return (0);
2869 #ifdef VIMAGE
2870 	if (prison_owns_vnet(cred))
2871 		return (0);
2872 #endif
2873 
2874 	return (1);
2875 }
2876 
2877 /*
2878  * Return the correct hostname (domainname, et al) for the passed credential.
2879  */
2880 void
2881 getcredhostname(struct ucred *cred, char *buf, size_t size)
2882 {
2883 	struct prison *pr;
2884 
2885 	/*
2886 	 * A NULL credential can be used to shortcut to the physical
2887 	 * system's hostname.
2888 	 */
2889 	pr = (cred != NULL) ? cred->cr_prison : &prison0;
2890 	mtx_lock(&pr->pr_mtx);
2891 	strlcpy(buf, pr->pr_hostname, size);
2892 	mtx_unlock(&pr->pr_mtx);
2893 }
2894 
2895 void
2896 getcreddomainname(struct ucred *cred, char *buf, size_t size)
2897 {
2898 
2899 	mtx_lock(&cred->cr_prison->pr_mtx);
2900 	strlcpy(buf, cred->cr_prison->pr_domainname, size);
2901 	mtx_unlock(&cred->cr_prison->pr_mtx);
2902 }
2903 
2904 void
2905 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
2906 {
2907 
2908 	mtx_lock(&cred->cr_prison->pr_mtx);
2909 	strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
2910 	mtx_unlock(&cred->cr_prison->pr_mtx);
2911 }
2912 
2913 void
2914 getcredhostid(struct ucred *cred, unsigned long *hostid)
2915 {
2916 
2917 	mtx_lock(&cred->cr_prison->pr_mtx);
2918 	*hostid = cred->cr_prison->pr_hostid;
2919 	mtx_unlock(&cred->cr_prison->pr_mtx);
2920 }
2921 
2922 #ifdef VIMAGE
2923 /*
2924  * Determine whether the prison represented by cred owns
2925  * its vnet rather than having it inherited.
2926  *
2927  * Returns 1 in case the prison owns the vnet, 0 otherwise.
2928  */
2929 int
2930 prison_owns_vnet(struct ucred *cred)
2931 {
2932 
2933 	/*
2934 	 * vnets cannot be added/removed after jail creation,
2935 	 * so no need to lock here.
2936 	 */
2937 	return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
2938 }
2939 #endif
2940 
2941 /*
2942  * Determine whether the subject represented by cred can "see"
2943  * status of a mount point.
2944  * Returns: 0 for permitted, ENOENT otherwise.
2945  * XXX: This function should be called cr_canseemount() and should be
2946  *      placed in kern_prot.c.
2947  */
2948 int
2949 prison_canseemount(struct ucred *cred, struct mount *mp)
2950 {
2951 	struct prison *pr;
2952 	struct statfs *sp;
2953 	size_t len;
2954 
2955 	pr = cred->cr_prison;
2956 	if (pr->pr_enforce_statfs == 0)
2957 		return (0);
2958 	if (pr->pr_root->v_mount == mp)
2959 		return (0);
2960 	if (pr->pr_enforce_statfs == 2)
2961 		return (ENOENT);
2962 	/*
2963 	 * If jail's chroot directory is set to "/" we should be able to see
2964 	 * all mount-points from inside a jail.
2965 	 * This is ugly check, but this is the only situation when jail's
2966 	 * directory ends with '/'.
2967 	 */
2968 	if (strcmp(pr->pr_path, "/") == 0)
2969 		return (0);
2970 	len = strlen(pr->pr_path);
2971 	sp = &mp->mnt_stat;
2972 	if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
2973 		return (ENOENT);
2974 	/*
2975 	 * Be sure that we don't have situation where jail's root directory
2976 	 * is "/some/path" and mount point is "/some/pathpath".
2977 	 */
2978 	if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
2979 		return (ENOENT);
2980 	return (0);
2981 }
2982 
2983 void
2984 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
2985 {
2986 	char jpath[MAXPATHLEN];
2987 	struct prison *pr;
2988 	size_t len;
2989 
2990 	pr = cred->cr_prison;
2991 	if (pr->pr_enforce_statfs == 0)
2992 		return;
2993 	if (prison_canseemount(cred, mp) != 0) {
2994 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
2995 		strlcpy(sp->f_mntonname, "[restricted]",
2996 		    sizeof(sp->f_mntonname));
2997 		return;
2998 	}
2999 	if (pr->pr_root->v_mount == mp) {
3000 		/*
3001 		 * Clear current buffer data, so we are sure nothing from
3002 		 * the valid path left there.
3003 		 */
3004 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3005 		*sp->f_mntonname = '/';
3006 		return;
3007 	}
3008 	/*
3009 	 * If jail's chroot directory is set to "/" we should be able to see
3010 	 * all mount-points from inside a jail.
3011 	 */
3012 	if (strcmp(pr->pr_path, "/") == 0)
3013 		return;
3014 	len = strlen(pr->pr_path);
3015 	strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3016 	/*
3017 	 * Clear current buffer data, so we are sure nothing from
3018 	 * the valid path left there.
3019 	 */
3020 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3021 	if (*jpath == '\0') {
3022 		/* Should never happen. */
3023 		*sp->f_mntonname = '/';
3024 	} else {
3025 		strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3026 	}
3027 }
3028 
3029 /*
3030  * Check with permission for a specific privilege is granted within jail.  We
3031  * have a specific list of accepted privileges; the rest are denied.
3032  */
3033 int
3034 prison_priv_check(struct ucred *cred, int priv)
3035 {
3036 
3037 	if (!jailed(cred))
3038 		return (0);
3039 
3040 #ifdef VIMAGE
3041 	/*
3042 	 * Privileges specific to prisons with a virtual network stack.
3043 	 * There might be a duplicate entry here in case the privilege
3044 	 * is only granted conditionally in the legacy jail case.
3045 	 */
3046 	switch (priv) {
3047 #ifdef notyet
3048 		/*
3049 		 * NFS-specific privileges.
3050 		 */
3051 	case PRIV_NFS_DAEMON:
3052 	case PRIV_NFS_LOCKD:
3053 #endif
3054 		/*
3055 		 * Network stack privileges.
3056 		 */
3057 	case PRIV_NET_BRIDGE:
3058 	case PRIV_NET_GRE:
3059 	case PRIV_NET_BPF:
3060 	case PRIV_NET_RAW:		/* Dup, cond. in legacy jail case. */
3061 	case PRIV_NET_ROUTE:
3062 	case PRIV_NET_TAP:
3063 	case PRIV_NET_SETIFMTU:
3064 	case PRIV_NET_SETIFFLAGS:
3065 	case PRIV_NET_SETIFCAP:
3066 	case PRIV_NET_SETIFDESCR:
3067 	case PRIV_NET_SETIFNAME	:
3068 	case PRIV_NET_SETIFMETRIC:
3069 	case PRIV_NET_SETIFPHYS:
3070 	case PRIV_NET_SETIFMAC:
3071 	case PRIV_NET_SETLANPCP:
3072 	case PRIV_NET_ADDMULTI:
3073 	case PRIV_NET_DELMULTI:
3074 	case PRIV_NET_HWIOCTL:
3075 	case PRIV_NET_SETLLADDR:
3076 	case PRIV_NET_ADDIFGROUP:
3077 	case PRIV_NET_DELIFGROUP:
3078 	case PRIV_NET_IFCREATE:
3079 	case PRIV_NET_IFDESTROY:
3080 	case PRIV_NET_ADDIFADDR:
3081 	case PRIV_NET_DELIFADDR:
3082 	case PRIV_NET_LAGG:
3083 	case PRIV_NET_GIF:
3084 	case PRIV_NET_SETIFVNET:
3085 	case PRIV_NET_SETIFFIB:
3086 
3087 		/*
3088 		 * 802.11-related privileges.
3089 		 */
3090 	case PRIV_NET80211_GETKEY:
3091 #ifdef notyet
3092 	case PRIV_NET80211_MANAGE:		/* XXX-BZ discuss with sam@ */
3093 #endif
3094 
3095 #ifdef notyet
3096 		/*
3097 		 * ATM privileges.
3098 		 */
3099 	case PRIV_NETATM_CFG:
3100 	case PRIV_NETATM_ADD:
3101 	case PRIV_NETATM_DEL:
3102 	case PRIV_NETATM_SET:
3103 
3104 		/*
3105 		 * Bluetooth privileges.
3106 		 */
3107 	case PRIV_NETBLUETOOTH_RAW:
3108 #endif
3109 
3110 		/*
3111 		 * Netgraph and netgraph module privileges.
3112 		 */
3113 	case PRIV_NETGRAPH_CONTROL:
3114 #ifdef notyet
3115 	case PRIV_NETGRAPH_TTY:
3116 #endif
3117 
3118 		/*
3119 		 * IPv4 and IPv6 privileges.
3120 		 */
3121 	case PRIV_NETINET_IPFW:
3122 	case PRIV_NETINET_DIVERT:
3123 	case PRIV_NETINET_PF:
3124 	case PRIV_NETINET_DUMMYNET:
3125 	case PRIV_NETINET_CARP:
3126 	case PRIV_NETINET_MROUTE:
3127 	case PRIV_NETINET_RAW:
3128 	case PRIV_NETINET_ADDRCTRL6:
3129 	case PRIV_NETINET_ND6:
3130 	case PRIV_NETINET_SCOPE6:
3131 	case PRIV_NETINET_ALIFETIME6:
3132 	case PRIV_NETINET_IPSEC:
3133 	case PRIV_NETINET_BINDANY:
3134 
3135 #ifdef notyet
3136 		/*
3137 		 * NCP privileges.
3138 		 */
3139 	case PRIV_NETNCP:
3140 
3141 		/*
3142 		 * SMB privileges.
3143 		 */
3144 	case PRIV_NETSMB:
3145 #endif
3146 
3147 	/*
3148 	 * No default: or deny here.
3149 	 * In case of no permit fall through to next switch().
3150 	 */
3151 		if (cred->cr_prison->pr_flags & PR_VNET)
3152 			return (0);
3153 	}
3154 #endif /* VIMAGE */
3155 
3156 	switch (priv) {
3157 
3158 		/*
3159 		 * Allow ktrace privileges for root in jail.
3160 		 */
3161 	case PRIV_KTRACE:
3162 
3163 #if 0
3164 		/*
3165 		 * Allow jailed processes to configure audit identity and
3166 		 * submit audit records (login, etc).  In the future we may
3167 		 * want to further refine the relationship between audit and
3168 		 * jail.
3169 		 */
3170 	case PRIV_AUDIT_GETAUDIT:
3171 	case PRIV_AUDIT_SETAUDIT:
3172 	case PRIV_AUDIT_SUBMIT:
3173 #endif
3174 
3175 		/*
3176 		 * Allow jailed processes to manipulate process UNIX
3177 		 * credentials in any way they see fit.
3178 		 */
3179 	case PRIV_CRED_SETUID:
3180 	case PRIV_CRED_SETEUID:
3181 	case PRIV_CRED_SETGID:
3182 	case PRIV_CRED_SETEGID:
3183 	case PRIV_CRED_SETGROUPS:
3184 	case PRIV_CRED_SETREUID:
3185 	case PRIV_CRED_SETREGID:
3186 	case PRIV_CRED_SETRESUID:
3187 	case PRIV_CRED_SETRESGID:
3188 
3189 		/*
3190 		 * Jail implements visibility constraints already, so allow
3191 		 * jailed root to override uid/gid-based constraints.
3192 		 */
3193 	case PRIV_SEEOTHERGIDS:
3194 	case PRIV_SEEOTHERUIDS:
3195 
3196 		/*
3197 		 * Jail implements inter-process debugging limits already, so
3198 		 * allow jailed root various debugging privileges.
3199 		 */
3200 	case PRIV_DEBUG_DIFFCRED:
3201 	case PRIV_DEBUG_SUGID:
3202 	case PRIV_DEBUG_UNPRIV:
3203 
3204 		/*
3205 		 * Allow jail to set various resource limits and login
3206 		 * properties, and for now, exceed process resource limits.
3207 		 */
3208 	case PRIV_PROC_LIMIT:
3209 	case PRIV_PROC_SETLOGIN:
3210 	case PRIV_PROC_SETRLIMIT:
3211 
3212 		/*
3213 		 * System V and POSIX IPC privileges are granted in jail.
3214 		 */
3215 	case PRIV_IPC_READ:
3216 	case PRIV_IPC_WRITE:
3217 	case PRIV_IPC_ADMIN:
3218 	case PRIV_IPC_MSGSIZE:
3219 	case PRIV_MQ_ADMIN:
3220 
3221 		/*
3222 		 * Jail operations within a jail work on child jails.
3223 		 */
3224 	case PRIV_JAIL_ATTACH:
3225 	case PRIV_JAIL_SET:
3226 	case PRIV_JAIL_REMOVE:
3227 
3228 		/*
3229 		 * Jail implements its own inter-process limits, so allow
3230 		 * root processes in jail to change scheduling on other
3231 		 * processes in the same jail.  Likewise for signalling.
3232 		 */
3233 	case PRIV_SCHED_DIFFCRED:
3234 	case PRIV_SCHED_CPUSET:
3235 	case PRIV_SIGNAL_DIFFCRED:
3236 	case PRIV_SIGNAL_SUGID:
3237 
3238 		/*
3239 		 * Allow jailed processes to write to sysctls marked as jail
3240 		 * writable.
3241 		 */
3242 	case PRIV_SYSCTL_WRITEJAIL:
3243 
3244 		/*
3245 		 * Allow root in jail to manage a variety of quota
3246 		 * properties.  These should likely be conditional on a
3247 		 * configuration option.
3248 		 */
3249 	case PRIV_VFS_GETQUOTA:
3250 	case PRIV_VFS_SETQUOTA:
3251 
3252 		/*
3253 		 * Since Jail relies on chroot() to implement file system
3254 		 * protections, grant many VFS privileges to root in jail.
3255 		 * Be careful to exclude mount-related and NFS-related
3256 		 * privileges.
3257 		 */
3258 	case PRIV_VFS_READ:
3259 	case PRIV_VFS_WRITE:
3260 	case PRIV_VFS_ADMIN:
3261 	case PRIV_VFS_EXEC:
3262 	case PRIV_VFS_LOOKUP:
3263 	case PRIV_VFS_BLOCKRESERVE:	/* XXXRW: Slightly surprising. */
3264 	case PRIV_VFS_CHFLAGS_DEV:
3265 	case PRIV_VFS_CHOWN:
3266 	case PRIV_VFS_CHROOT:
3267 	case PRIV_VFS_RETAINSUGID:
3268 	case PRIV_VFS_FCHROOT:
3269 	case PRIV_VFS_LINK:
3270 	case PRIV_VFS_SETGID:
3271 	case PRIV_VFS_STAT:
3272 	case PRIV_VFS_STICKYFILE:
3273 
3274 		/*
3275 		 * As in the non-jail case, non-root users are expected to be
3276 		 * able to read kernel/phyiscal memory (provided /dev/[k]mem
3277 		 * exists in the jail and they have permission to access it).
3278 		 */
3279 	case PRIV_KMEM_READ:
3280 		return (0);
3281 
3282 		/*
3283 		 * Depending on the global setting, allow privilege of
3284 		 * setting system flags.
3285 		 */
3286 	case PRIV_VFS_SYSFLAGS:
3287 		if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
3288 			return (0);
3289 		else
3290 			return (EPERM);
3291 
3292 		/*
3293 		 * Depending on the global setting, allow privilege of
3294 		 * mounting/unmounting file systems.
3295 		 */
3296 	case PRIV_VFS_MOUNT:
3297 	case PRIV_VFS_UNMOUNT:
3298 	case PRIV_VFS_MOUNT_NONUSER:
3299 	case PRIV_VFS_MOUNT_OWNER:
3300 		if (cred->cr_prison->pr_allow & PR_ALLOW_MOUNT &&
3301 		    cred->cr_prison->pr_enforce_statfs < 2)
3302 			return (0);
3303 		else
3304 			return (EPERM);
3305 
3306 		/*
3307 		 * Conditionnaly allow locking (unlocking) physical pages
3308 		 * in memory.
3309 		 */
3310 	case PRIV_VM_MLOCK:
3311 	case PRIV_VM_MUNLOCK:
3312 		if (cred->cr_prison->pr_allow & PR_ALLOW_MLOCK)
3313 			return (0);
3314 		else
3315 			return (EPERM);
3316 
3317 		/*
3318 		 * Conditionally allow jailed root to bind reserved ports.
3319 		 */
3320 	case PRIV_NETINET_RESERVEDPORT:
3321 		if (cred->cr_prison->pr_allow & PR_ALLOW_RESERVED_PORTS)
3322 			return (0);
3323 		else
3324 			return (EPERM);
3325 
3326 		/*
3327 		 * Allow jailed root to reuse in-use ports.
3328 		 */
3329 	case PRIV_NETINET_REUSEPORT:
3330 		return (0);
3331 
3332 		/*
3333 		 * Allow jailed root to set certain IPv4/6 (option) headers.
3334 		 */
3335 	case PRIV_NETINET_SETHDROPTS:
3336 		return (0);
3337 
3338 		/*
3339 		 * Conditionally allow creating raw sockets in jail.
3340 		 */
3341 	case PRIV_NETINET_RAW:
3342 		if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
3343 			return (0);
3344 		else
3345 			return (EPERM);
3346 
3347 		/*
3348 		 * Since jail implements its own visibility limits on netstat
3349 		 * sysctls, allow getcred.  This allows identd to work in
3350 		 * jail.
3351 		 */
3352 	case PRIV_NETINET_GETCRED:
3353 		return (0);
3354 
3355 		/*
3356 		 * Allow jailed root to set loginclass.
3357 		 */
3358 	case PRIV_PROC_SETLOGINCLASS:
3359 		return (0);
3360 
3361 		/*
3362 		 * Do not allow a process inside a jail to read the kernel
3363 		 * message buffer unless explicitly permitted.
3364 		 */
3365 	case PRIV_MSGBUF:
3366 		if (cred->cr_prison->pr_allow & PR_ALLOW_READ_MSGBUF)
3367 			return (0);
3368 		return (EPERM);
3369 
3370 	default:
3371 		/*
3372 		 * In all remaining cases, deny the privilege request.  This
3373 		 * includes almost all network privileges, many system
3374 		 * configuration privileges.
3375 		 */
3376 		return (EPERM);
3377 	}
3378 }
3379 
3380 /*
3381  * Return the part of pr2's name that is relative to pr1, or the whole name
3382  * if it does not directly follow.
3383  */
3384 
3385 char *
3386 prison_name(struct prison *pr1, struct prison *pr2)
3387 {
3388 	char *name;
3389 
3390 	/* Jails see themselves as "0" (if they see themselves at all). */
3391 	if (pr1 == pr2)
3392 		return "0";
3393 	name = pr2->pr_name;
3394 	if (prison_ischild(pr1, pr2)) {
3395 		/*
3396 		 * pr1 isn't locked (and allprison_lock may not be either)
3397 		 * so its length can't be counted on.  But the number of dots
3398 		 * can be counted on - and counted.
3399 		 */
3400 		for (; pr1 != &prison0; pr1 = pr1->pr_parent)
3401 			name = strchr(name, '.') + 1;
3402 	}
3403 	return (name);
3404 }
3405 
3406 /*
3407  * Return the part of pr2's path that is relative to pr1, or the whole path
3408  * if it does not directly follow.
3409  */
3410 static char *
3411 prison_path(struct prison *pr1, struct prison *pr2)
3412 {
3413 	char *path1, *path2;
3414 	int len1;
3415 
3416 	path1 = pr1->pr_path;
3417 	path2 = pr2->pr_path;
3418 	if (!strcmp(path1, "/"))
3419 		return (path2);
3420 	len1 = strlen(path1);
3421 	if (strncmp(path1, path2, len1))
3422 		return (path2);
3423 	if (path2[len1] == '\0')
3424 		return "/";
3425 	if (path2[len1] == '/')
3426 		return (path2 + len1);
3427 	return (path2);
3428 }
3429 
3430 
3431 /*
3432  * Jail-related sysctls.
3433  */
3434 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW, 0,
3435     "Jails");
3436 
3437 static int
3438 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
3439 {
3440 	struct xprison *xp;
3441 	struct prison *pr, *cpr;
3442 #ifdef INET
3443 	struct in_addr *ip4 = NULL;
3444 	int ip4s = 0;
3445 #endif
3446 #ifdef INET6
3447 	struct in6_addr *ip6 = NULL;
3448 	int ip6s = 0;
3449 #endif
3450 	int descend, error;
3451 
3452 	xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
3453 	pr = req->td->td_ucred->cr_prison;
3454 	error = 0;
3455 	sx_slock(&allprison_lock);
3456 	FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
3457 #if defined(INET) || defined(INET6)
3458  again:
3459 #endif
3460 		mtx_lock(&cpr->pr_mtx);
3461 #ifdef INET
3462 		if (cpr->pr_ip4s > 0) {
3463 			if (ip4s < cpr->pr_ip4s) {
3464 				ip4s = cpr->pr_ip4s;
3465 				mtx_unlock(&cpr->pr_mtx);
3466 				ip4 = realloc(ip4, ip4s *
3467 				    sizeof(struct in_addr), M_TEMP, M_WAITOK);
3468 				goto again;
3469 			}
3470 			bcopy(cpr->pr_ip4, ip4,
3471 			    cpr->pr_ip4s * sizeof(struct in_addr));
3472 		}
3473 #endif
3474 #ifdef INET6
3475 		if (cpr->pr_ip6s > 0) {
3476 			if (ip6s < cpr->pr_ip6s) {
3477 				ip6s = cpr->pr_ip6s;
3478 				mtx_unlock(&cpr->pr_mtx);
3479 				ip6 = realloc(ip6, ip6s *
3480 				    sizeof(struct in6_addr), M_TEMP, M_WAITOK);
3481 				goto again;
3482 			}
3483 			bcopy(cpr->pr_ip6, ip6,
3484 			    cpr->pr_ip6s * sizeof(struct in6_addr));
3485 		}
3486 #endif
3487 		if (cpr->pr_ref == 0) {
3488 			mtx_unlock(&cpr->pr_mtx);
3489 			continue;
3490 		}
3491 		bzero(xp, sizeof(*xp));
3492 		xp->pr_version = XPRISON_VERSION;
3493 		xp->pr_id = cpr->pr_id;
3494 		xp->pr_state = cpr->pr_uref > 0
3495 		    ? PRISON_STATE_ALIVE : PRISON_STATE_DYING;
3496 		strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
3497 		strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
3498 		strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
3499 #ifdef INET
3500 		xp->pr_ip4s = cpr->pr_ip4s;
3501 #endif
3502 #ifdef INET6
3503 		xp->pr_ip6s = cpr->pr_ip6s;
3504 #endif
3505 		mtx_unlock(&cpr->pr_mtx);
3506 		error = SYSCTL_OUT(req, xp, sizeof(*xp));
3507 		if (error)
3508 			break;
3509 #ifdef INET
3510 		if (xp->pr_ip4s > 0) {
3511 			error = SYSCTL_OUT(req, ip4,
3512 			    xp->pr_ip4s * sizeof(struct in_addr));
3513 			if (error)
3514 				break;
3515 		}
3516 #endif
3517 #ifdef INET6
3518 		if (xp->pr_ip6s > 0) {
3519 			error = SYSCTL_OUT(req, ip6,
3520 			    xp->pr_ip6s * sizeof(struct in6_addr));
3521 			if (error)
3522 				break;
3523 		}
3524 #endif
3525 	}
3526 	sx_sunlock(&allprison_lock);
3527 	free(xp, M_TEMP);
3528 #ifdef INET
3529 	free(ip4, M_TEMP);
3530 #endif
3531 #ifdef INET6
3532 	free(ip6, M_TEMP);
3533 #endif
3534 	return (error);
3535 }
3536 
3537 SYSCTL_OID(_security_jail, OID_AUTO, list,
3538     CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3539     sysctl_jail_list, "S", "List of active jails");
3540 
3541 static int
3542 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
3543 {
3544 	int error, injail;
3545 
3546 	injail = jailed(req->td->td_ucred);
3547 	error = SYSCTL_OUT(req, &injail, sizeof(injail));
3548 
3549 	return (error);
3550 }
3551 
3552 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
3553     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3554     sysctl_jail_jailed, "I", "Process in jail?");
3555 
3556 static int
3557 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
3558 {
3559 	int error, havevnet;
3560 #ifdef VIMAGE
3561 	struct ucred *cred = req->td->td_ucred;
3562 
3563 	havevnet = jailed(cred) && prison_owns_vnet(cred);
3564 #else
3565 	havevnet = 0;
3566 #endif
3567 	error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
3568 
3569 	return (error);
3570 }
3571 
3572 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
3573     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3574     sysctl_jail_vnet, "I", "Jail owns vnet?");
3575 
3576 #if defined(INET) || defined(INET6)
3577 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
3578     &jail_max_af_ips, 0,
3579     "Number of IP addresses a jail may have at most per address family (deprecated)");
3580 #endif
3581 
3582 /*
3583  * Default parameters for jail(2) compatibility.  For historical reasons,
3584  * the sysctl names have varying similarity to the parameter names.  Prisons
3585  * just see their own parameters, and can't change them.
3586  */
3587 static int
3588 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
3589 {
3590 	struct prison *pr;
3591 	int allow, error, i;
3592 
3593 	pr = req->td->td_ucred->cr_prison;
3594 	allow = (pr == &prison0) ? jail_default_allow : pr->pr_allow;
3595 
3596 	/* Get the current flag value, and convert it to a boolean. */
3597 	i = (allow & arg2) ? 1 : 0;
3598 	if (arg1 != NULL)
3599 		i = !i;
3600 	error = sysctl_handle_int(oidp, &i, 0, req);
3601 	if (error || !req->newptr)
3602 		return (error);
3603 	i = i ? arg2 : 0;
3604 	if (arg1 != NULL)
3605 		i ^= arg2;
3606 	/*
3607 	 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
3608 	 * for writing.
3609 	 */
3610 	mtx_lock(&prison0.pr_mtx);
3611 	jail_default_allow = (jail_default_allow & ~arg2) | i;
3612 	mtx_unlock(&prison0.pr_mtx);
3613 	return (0);
3614 }
3615 
3616 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
3617     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3618     NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
3619     "Processes in jail can set their hostnames (deprecated)");
3620 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
3621     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3622     (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
3623     "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
3624 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
3625     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3626     NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
3627     "Processes in jail can use System V IPC primitives (deprecated)");
3628 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
3629     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3630     NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
3631     "Prison root can create raw sockets (deprecated)");
3632 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
3633     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3634     NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
3635     "Processes in jail can alter system file flags (deprecated)");
3636 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
3637     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3638     NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
3639     "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
3640 
3641 static int
3642 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
3643 {
3644 	struct prison *pr;
3645 	int level, error;
3646 
3647 	pr = req->td->td_ucred->cr_prison;
3648 	level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
3649 	error = sysctl_handle_int(oidp, &level, 0, req);
3650 	if (error || !req->newptr)
3651 		return (error);
3652 	*(int *)arg1 = level;
3653 	return (0);
3654 }
3655 
3656 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
3657     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3658     &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
3659     sysctl_jail_default_level, "I",
3660     "Processes in jail cannot see all mounted file systems (deprecated)");
3661 
3662 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
3663     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
3664     &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
3665     sysctl_jail_default_level, "I",
3666     "Ruleset for the devfs filesystem in jail (deprecated)");
3667 
3668 /*
3669  * Nodes to describe jail parameters.  Maximum length of string parameters
3670  * is returned in the string itself, and the other parameters exist merely
3671  * to make themselves and their types known.
3672  */
3673 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW, 0,
3674     "Jail parameters");
3675 
3676 int
3677 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
3678 {
3679 	int i;
3680 	long l;
3681 	size_t s;
3682 	char numbuf[12];
3683 
3684 	switch (oidp->oid_kind & CTLTYPE)
3685 	{
3686 	case CTLTYPE_LONG:
3687 	case CTLTYPE_ULONG:
3688 		l = 0;
3689 #ifdef SCTL_MASK32
3690 		if (!(req->flags & SCTL_MASK32))
3691 #endif
3692 			return (SYSCTL_OUT(req, &l, sizeof(l)));
3693 	case CTLTYPE_INT:
3694 	case CTLTYPE_UINT:
3695 		i = 0;
3696 		return (SYSCTL_OUT(req, &i, sizeof(i)));
3697 	case CTLTYPE_STRING:
3698 		snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
3699 		return
3700 		    (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
3701 	case CTLTYPE_STRUCT:
3702 		s = (size_t)arg2;
3703 		return (SYSCTL_OUT(req, &s, sizeof(s)));
3704 	}
3705 	return (0);
3706 }
3707 
3708 /*
3709  * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
3710  * jail creation time but cannot be changed in an existing jail.
3711  */
3712 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
3713 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
3714 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
3715 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
3716 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
3717     "I", "Jail secure level");
3718 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
3719     "Jail value for kern.osreldate and uname -K");
3720 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
3721     "Jail value for kern.osrelease and uname -r");
3722 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
3723     "I", "Jail cannot see all mounted file systems");
3724 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
3725     "I", "Ruleset for in-jail devfs mounts");
3726 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
3727     "B", "Jail persistence");
3728 #ifdef VIMAGE
3729 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
3730     "E,jailsys", "Virtual network stack");
3731 #endif
3732 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
3733     "B", "Jail is in the process of shutting down");
3734 
3735 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
3736 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
3737     "I", "Current number of child jails");
3738 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
3739     "I", "Maximum number of child jails");
3740 
3741 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
3742 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
3743     "Jail hostname");
3744 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
3745     "Jail NIS domainname");
3746 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
3747     "Jail host UUID");
3748 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
3749     "LU", "Jail host ID");
3750 
3751 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
3752 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
3753 
3754 #ifdef INET
3755 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
3756     "Jail IPv4 address virtualization");
3757 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
3758     "S,in_addr,a", "Jail IPv4 addresses");
3759 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
3760     "B", "Do (not) use IPv4 source address selection rather than the "
3761     "primary jail IPv4 address.");
3762 #endif
3763 #ifdef INET6
3764 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
3765     "Jail IPv6 address virtualization");
3766 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
3767     "S,in6_addr,a", "Jail IPv6 addresses");
3768 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
3769     "B", "Do (not) use IPv6 source address selection rather than the "
3770     "primary jail IPv6 address.");
3771 #endif
3772 
3773 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
3774 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
3775     "B", "Jail may set hostname");
3776 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
3777     "B", "Jail may use SYSV IPC");
3778 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
3779     "B", "Jail may create raw sockets");
3780 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
3781     "B", "Jail may alter system file flags");
3782 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
3783     "B", "Jail may set file quotas");
3784 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
3785     "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
3786 SYSCTL_JAIL_PARAM(_allow, mlock, CTLTYPE_INT | CTLFLAG_RW,
3787     "B", "Jail may lock (unlock) physical pages in memory");
3788 SYSCTL_JAIL_PARAM(_allow, reserved_ports, CTLTYPE_INT | CTLFLAG_RW,
3789     "B", "Jail may bind sockets to reserved ports");
3790 SYSCTL_JAIL_PARAM(_allow, read_msgbuf, CTLTYPE_INT | CTLFLAG_RW,
3791     "B", "Jail may read the kernel message buffer");
3792 SYSCTL_JAIL_PARAM(_allow, unprivileged_proc_debug, CTLTYPE_INT | CTLFLAG_RW,
3793     "B", "Unprivileged processes may use process debugging facilities");
3794 
3795 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
3796 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
3797     "B", "Jail may mount/unmount jail-friendly file systems in general");
3798 
3799 /*
3800  * Add a dynamic parameter allow.<name>, or allow.<prefix>.<name>.  Return
3801  * its associated bit in the pr_allow bitmask, or zero if the parameter was
3802  * not created.
3803  */
3804 unsigned
3805 prison_add_allow(const char *prefix, const char *name, const char *prefix_descr,
3806     const char *descr)
3807 {
3808 	struct bool_flags *bf;
3809 	struct sysctl_oid *parent;
3810 	char *allow_name, *allow_noname, *allowed;
3811 #ifndef NO_SYSCTL_DESCR
3812 	char *descr_deprecated;
3813 #endif
3814 	unsigned allow_flag;
3815 
3816 	if (prefix
3817 	    ? asprintf(&allow_name, M_PRISON, "allow.%s.%s", prefix, name)
3818 		< 0 ||
3819 	      asprintf(&allow_noname, M_PRISON, "allow.%s.no%s", prefix, name)
3820 		< 0
3821 	    : asprintf(&allow_name, M_PRISON, "allow.%s", name) < 0 ||
3822 	      asprintf(&allow_noname, M_PRISON, "allow.no%s", name) < 0) {
3823 		free(allow_name, M_PRISON);
3824 		return 0;
3825 	}
3826 
3827 	/*
3828 	 * See if this parameter has already beed added, i.e. a module was
3829 	 * previously loaded/unloaded.
3830 	 */
3831 	mtx_lock(&prison0.pr_mtx);
3832 	for (bf = pr_flag_allow;
3833 	     bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
3834 	     bf++) {
3835 		if (strcmp(bf->name, allow_name) == 0) {
3836 			allow_flag = bf->flag;
3837 			goto no_add;
3838 		}
3839 	}
3840 
3841 	/*
3842 	 * Find a free bit in prison0's pr_allow, failing if there are none
3843 	 * (which shouldn't happen as long as we keep track of how many
3844 	 * potential dynamic flags exist).
3845 	 *
3846 	 * Due to per-jail unprivileged process debugging support
3847 	 * using pr_allow, also verify against PR_ALLOW_ALL_STATIC.
3848 	 * prison0 may have unprivileged process debugging unset.
3849 	 */
3850 	for (allow_flag = 1;; allow_flag <<= 1) {
3851 		if (allow_flag == 0)
3852 			goto no_add;
3853 		if (allow_flag & PR_ALLOW_ALL_STATIC)
3854 			continue;
3855 		if ((prison0.pr_allow & allow_flag) == 0)
3856 			break;
3857 	}
3858 
3859 	/*
3860 	 * Note the parameter in the next open slot in pr_flag_allow.
3861 	 * Set the flag last so code that checks pr_flag_allow can do so
3862 	 * without locking.
3863 	 */
3864 	for (bf = pr_flag_allow; bf->flag != 0; bf++)
3865 		if (bf == pr_flag_allow + nitems(pr_flag_allow)) {
3866 			/* This should never happen, but is not fatal. */
3867 			allow_flag = 0;
3868 			goto no_add;
3869 		}
3870 	prison0.pr_allow |= allow_flag;
3871 	bf->name = allow_name;
3872 	bf->noname = allow_noname;
3873 	bf->flag = allow_flag;
3874 	mtx_unlock(&prison0.pr_mtx);
3875 
3876 	/*
3877 	 * Create sysctls for the paramter, and the back-compat global
3878 	 * permission.
3879 	 */
3880 	parent = prefix
3881 	    ? SYSCTL_ADD_NODE(NULL,
3882 		  SYSCTL_CHILDREN(&sysctl___security_jail_param_allow),
3883 		  OID_AUTO, prefix, 0, 0, prefix_descr)
3884 	    : &sysctl___security_jail_param_allow;
3885 	(void)SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(parent), OID_AUTO,
3886 	    name, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3887 	    NULL, 0, sysctl_jail_param, "B", descr);
3888 	if ((prefix
3889 	     ? asprintf(&allowed, M_TEMP, "%s_%s_allowed", prefix, name)
3890 	     : asprintf(&allowed, M_TEMP, "%s_allowed", name)) >= 0) {
3891 #ifndef NO_SYSCTL_DESCR
3892 		(void)asprintf(&descr_deprecated, M_TEMP, "%s (deprecated)",
3893 		    descr);
3894 #endif
3895 		(void)SYSCTL_ADD_PROC(NULL,
3896 		    SYSCTL_CHILDREN(&sysctl___security_jail), OID_AUTO, allowed,
3897 		    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, allow_flag,
3898 		    sysctl_jail_default_allow, "I", descr_deprecated);
3899 #ifndef NO_SYSCTL_DESCR
3900 		free(descr_deprecated, M_TEMP);
3901 #endif
3902 		free(allowed, M_TEMP);
3903 	}
3904 	return allow_flag;
3905 
3906  no_add:
3907 	mtx_unlock(&prison0.pr_mtx);
3908 	free(allow_name, M_PRISON);
3909 	free(allow_noname, M_PRISON);
3910 	return allow_flag;
3911 }
3912 
3913 /*
3914  * The VFS system will register jail-aware filesystems here.  They each get
3915  * a parameter allow.mount.xxxfs and a flag to check when a jailed user
3916  * attempts to mount.
3917  */
3918 void
3919 prison_add_vfs(struct vfsconf *vfsp)
3920 {
3921 #ifdef NO_SYSCTL_DESCR
3922 
3923 	vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
3924 	    NULL, NULL);
3925 #else
3926 	char *descr;
3927 
3928 	(void)asprintf(&descr, M_TEMP, "Jail may mount the %s file system",
3929 	    vfsp->vfc_name);
3930 	vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name,
3931 	    NULL, descr);
3932 	free(descr, M_TEMP);
3933 #endif
3934 }
3935 
3936 #ifdef RACCT
3937 void
3938 prison_racct_foreach(void (*callback)(struct racct *racct,
3939     void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
3940     void *arg2, void *arg3)
3941 {
3942 	struct prison_racct *prr;
3943 
3944 	ASSERT_RACCT_ENABLED();
3945 
3946 	sx_slock(&allprison_lock);
3947 	if (pre != NULL)
3948 		(pre)();
3949 	LIST_FOREACH(prr, &allprison_racct, prr_next)
3950 		(callback)(prr->prr_racct, arg2, arg3);
3951 	if (post != NULL)
3952 		(post)();
3953 	sx_sunlock(&allprison_lock);
3954 }
3955 
3956 static struct prison_racct *
3957 prison_racct_find_locked(const char *name)
3958 {
3959 	struct prison_racct *prr;
3960 
3961 	ASSERT_RACCT_ENABLED();
3962 	sx_assert(&allprison_lock, SA_XLOCKED);
3963 
3964 	if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
3965 		return (NULL);
3966 
3967 	LIST_FOREACH(prr, &allprison_racct, prr_next) {
3968 		if (strcmp(name, prr->prr_name) != 0)
3969 			continue;
3970 
3971 		/* Found prison_racct with a matching name? */
3972 		prison_racct_hold(prr);
3973 		return (prr);
3974 	}
3975 
3976 	/* Add new prison_racct. */
3977 	prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
3978 	racct_create(&prr->prr_racct);
3979 
3980 	strcpy(prr->prr_name, name);
3981 	refcount_init(&prr->prr_refcount, 1);
3982 	LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
3983 
3984 	return (prr);
3985 }
3986 
3987 struct prison_racct *
3988 prison_racct_find(const char *name)
3989 {
3990 	struct prison_racct *prr;
3991 
3992 	ASSERT_RACCT_ENABLED();
3993 
3994 	sx_xlock(&allprison_lock);
3995 	prr = prison_racct_find_locked(name);
3996 	sx_xunlock(&allprison_lock);
3997 	return (prr);
3998 }
3999 
4000 void
4001 prison_racct_hold(struct prison_racct *prr)
4002 {
4003 
4004 	ASSERT_RACCT_ENABLED();
4005 
4006 	refcount_acquire(&prr->prr_refcount);
4007 }
4008 
4009 static void
4010 prison_racct_free_locked(struct prison_racct *prr)
4011 {
4012 
4013 	ASSERT_RACCT_ENABLED();
4014 	sx_assert(&allprison_lock, SA_XLOCKED);
4015 
4016 	if (refcount_release(&prr->prr_refcount)) {
4017 		racct_destroy(&prr->prr_racct);
4018 		LIST_REMOVE(prr, prr_next);
4019 		free(prr, M_PRISON_RACCT);
4020 	}
4021 }
4022 
4023 void
4024 prison_racct_free(struct prison_racct *prr)
4025 {
4026 	int old;
4027 
4028 	ASSERT_RACCT_ENABLED();
4029 	sx_assert(&allprison_lock, SA_UNLOCKED);
4030 
4031 	old = prr->prr_refcount;
4032 	if (old > 1 && atomic_cmpset_int(&prr->prr_refcount, old, old - 1))
4033 		return;
4034 
4035 	sx_xlock(&allprison_lock);
4036 	prison_racct_free_locked(prr);
4037 	sx_xunlock(&allprison_lock);
4038 }
4039 
4040 static void
4041 prison_racct_attach(struct prison *pr)
4042 {
4043 	struct prison_racct *prr;
4044 
4045 	ASSERT_RACCT_ENABLED();
4046 	sx_assert(&allprison_lock, SA_XLOCKED);
4047 
4048 	prr = prison_racct_find_locked(pr->pr_name);
4049 	KASSERT(prr != NULL, ("cannot find prison_racct"));
4050 
4051 	pr->pr_prison_racct = prr;
4052 }
4053 
4054 /*
4055  * Handle jail renaming.  From the racct point of view, renaming means
4056  * moving from one prison_racct to another.
4057  */
4058 static void
4059 prison_racct_modify(struct prison *pr)
4060 {
4061 #ifdef RCTL
4062 	struct proc *p;
4063 	struct ucred *cred;
4064 #endif
4065 	struct prison_racct *oldprr;
4066 
4067 	ASSERT_RACCT_ENABLED();
4068 
4069 	sx_slock(&allproc_lock);
4070 	sx_xlock(&allprison_lock);
4071 
4072 	if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4073 		sx_xunlock(&allprison_lock);
4074 		sx_sunlock(&allproc_lock);
4075 		return;
4076 	}
4077 
4078 	oldprr = pr->pr_prison_racct;
4079 	pr->pr_prison_racct = NULL;
4080 
4081 	prison_racct_attach(pr);
4082 
4083 	/*
4084 	 * Move resource utilisation records.
4085 	 */
4086 	racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
4087 
4088 #ifdef RCTL
4089 	/*
4090 	 * Force rctl to reattach rules to processes.
4091 	 */
4092 	FOREACH_PROC_IN_SYSTEM(p) {
4093 		PROC_LOCK(p);
4094 		cred = crhold(p->p_ucred);
4095 		PROC_UNLOCK(p);
4096 		rctl_proc_ucred_changed(p, cred);
4097 		crfree(cred);
4098 	}
4099 #endif
4100 
4101 	sx_sunlock(&allproc_lock);
4102 	prison_racct_free_locked(oldprr);
4103 	sx_xunlock(&allprison_lock);
4104 }
4105 
4106 static void
4107 prison_racct_detach(struct prison *pr)
4108 {
4109 
4110 	ASSERT_RACCT_ENABLED();
4111 	sx_assert(&allprison_lock, SA_UNLOCKED);
4112 
4113 	if (pr->pr_prison_racct == NULL)
4114 		return;
4115 	prison_racct_free(pr->pr_prison_racct);
4116 	pr->pr_prison_racct = NULL;
4117 }
4118 #endif /* RACCT */
4119 
4120 #ifdef DDB
4121 
4122 static void
4123 db_show_prison(struct prison *pr)
4124 {
4125 	struct bool_flags *bf;
4126 	struct jailsys_flags *jsf;
4127 #if defined(INET) || defined(INET6)
4128 	int ii;
4129 #endif
4130 	unsigned f;
4131 #ifdef INET
4132 	char ip4buf[INET_ADDRSTRLEN];
4133 #endif
4134 #ifdef INET6
4135 	char ip6buf[INET6_ADDRSTRLEN];
4136 #endif
4137 
4138 	db_printf("prison %p:\n", pr);
4139 	db_printf(" jid             = %d\n", pr->pr_id);
4140 	db_printf(" name            = %s\n", pr->pr_name);
4141 	db_printf(" parent          = %p\n", pr->pr_parent);
4142 	db_printf(" ref             = %d\n", pr->pr_ref);
4143 	db_printf(" uref            = %d\n", pr->pr_uref);
4144 	db_printf(" path            = %s\n", pr->pr_path);
4145 	db_printf(" cpuset          = %d\n", pr->pr_cpuset
4146 	    ? pr->pr_cpuset->cs_id : -1);
4147 #ifdef VIMAGE
4148 	db_printf(" vnet            = %p\n", pr->pr_vnet);
4149 #endif
4150 	db_printf(" root            = %p\n", pr->pr_root);
4151 	db_printf(" securelevel     = %d\n", pr->pr_securelevel);
4152 	db_printf(" devfs_rsnum     = %d\n", pr->pr_devfs_rsnum);
4153 	db_printf(" children.max    = %d\n", pr->pr_childmax);
4154 	db_printf(" children.cur    = %d\n", pr->pr_childcount);
4155 	db_printf(" child           = %p\n", LIST_FIRST(&pr->pr_children));
4156 	db_printf(" sibling         = %p\n", LIST_NEXT(pr, pr_sibling));
4157 	db_printf(" flags           = 0x%x", pr->pr_flags);
4158 	for (bf = pr_flag_bool; bf < pr_flag_bool + nitems(pr_flag_bool); bf++)
4159 		if (pr->pr_flags & bf->flag)
4160 			db_printf(" %s", bf->name);
4161 	for (jsf = pr_flag_jailsys;
4162 	     jsf < pr_flag_jailsys + nitems(pr_flag_jailsys);
4163 	     jsf++) {
4164 		f = pr->pr_flags & (jsf->disable | jsf->new);
4165 		db_printf(" %-16s= %s\n", jsf->name,
4166 		    (f != 0 && f == jsf->disable) ? "disable"
4167 		    : (f == jsf->new) ? "new"
4168 		    : "inherit");
4169 	}
4170 	db_printf(" allow           = 0x%x", pr->pr_allow);
4171 	for (bf = pr_flag_allow;
4172 	     bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0;
4173 	     bf++)
4174 		if (pr->pr_allow & bf->flag)
4175 			db_printf(" %s", bf->name);
4176 	db_printf("\n");
4177 	db_printf(" enforce_statfs  = %d\n", pr->pr_enforce_statfs);
4178 	db_printf(" host.hostname   = %s\n", pr->pr_hostname);
4179 	db_printf(" host.domainname = %s\n", pr->pr_domainname);
4180 	db_printf(" host.hostuuid   = %s\n", pr->pr_hostuuid);
4181 	db_printf(" host.hostid     = %lu\n", pr->pr_hostid);
4182 #ifdef INET
4183 	db_printf(" ip4s            = %d\n", pr->pr_ip4s);
4184 	for (ii = 0; ii < pr->pr_ip4s; ii++)
4185 		db_printf(" %s %s\n",
4186 		    ii == 0 ? "ip4.addr        =" : "                 ",
4187 		    inet_ntoa_r(pr->pr_ip4[ii], ip4buf));
4188 #endif
4189 #ifdef INET6
4190 	db_printf(" ip6s            = %d\n", pr->pr_ip6s);
4191 	for (ii = 0; ii < pr->pr_ip6s; ii++)
4192 		db_printf(" %s %s\n",
4193 		    ii == 0 ? "ip6.addr        =" : "                 ",
4194 		    ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4195 #endif
4196 }
4197 
4198 DB_SHOW_COMMAND(prison, db_show_prison_command)
4199 {
4200 	struct prison *pr;
4201 
4202 	if (!have_addr) {
4203 		/*
4204 		 * Show all prisons in the list, and prison0 which is not
4205 		 * listed.
4206 		 */
4207 		db_show_prison(&prison0);
4208 		if (!db_pager_quit) {
4209 			TAILQ_FOREACH(pr, &allprison, pr_list) {
4210 				db_show_prison(pr);
4211 				if (db_pager_quit)
4212 					break;
4213 			}
4214 		}
4215 		return;
4216 	}
4217 
4218 	if (addr == 0)
4219 		pr = &prison0;
4220 	else {
4221 		/* Look for a prison with the ID and with references. */
4222 		TAILQ_FOREACH(pr, &allprison, pr_list)
4223 			if (pr->pr_id == addr && pr->pr_ref > 0)
4224 				break;
4225 		if (pr == NULL)
4226 			/* Look again, without requiring a reference. */
4227 			TAILQ_FOREACH(pr, &allprison, pr_list)
4228 				if (pr->pr_id == addr)
4229 					break;
4230 		if (pr == NULL)
4231 			/* Assume address points to a valid prison. */
4232 			pr = (struct prison *)addr;
4233 	}
4234 	db_show_prison(pr);
4235 }
4236 
4237 #endif /* DDB */
4238