xref: /freebsd/sys/kern/kern_jail.c (revision 91ebf7d76582a73ac4c0a32964e4d3912010671e)
1 /*-
2  * Copyright (c) 1999 Poul-Henning Kamp.
3  * Copyright (c) 2008 Bjoern A. Zeeb.
4  * Copyright (c) 2009 James Gritton.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include "opt_compat.h"
33 #include "opt_ddb.h"
34 #include "opt_inet.h"
35 #include "opt_inet6.h"
36 
37 #include <sys/param.h>
38 #include <sys/types.h>
39 #include <sys/kernel.h>
40 #include <sys/systm.h>
41 #include <sys/errno.h>
42 #include <sys/sysproto.h>
43 #include <sys/malloc.h>
44 #include <sys/osd.h>
45 #include <sys/priv.h>
46 #include <sys/proc.h>
47 #include <sys/taskqueue.h>
48 #include <sys/fcntl.h>
49 #include <sys/jail.h>
50 #include <sys/lock.h>
51 #include <sys/mutex.h>
52 #include <sys/racct.h>
53 #include <sys/refcount.h>
54 #include <sys/sx.h>
55 #include <sys/sysent.h>
56 #include <sys/namei.h>
57 #include <sys/mount.h>
58 #include <sys/queue.h>
59 #include <sys/socket.h>
60 #include <sys/syscallsubr.h>
61 #include <sys/sysctl.h>
62 #include <sys/vnode.h>
63 
64 #include <net/if.h>
65 #include <net/vnet.h>
66 
67 #include <netinet/in.h>
68 
69 #ifdef DDB
70 #include <ddb/ddb.h>
71 #endif /* DDB */
72 
73 #include <security/mac/mac_framework.h>
74 
75 #define	DEFAULT_HOSTUUID	"00000000-0000-0000-0000-000000000000"
76 
77 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
78 static MALLOC_DEFINE(M_PRISON_RACCT, "prison_racct", "Prison racct structures");
79 
80 /* Keep struct prison prison0 and some code in kern_jail_set() readable. */
81 #ifdef INET
82 #ifdef INET6
83 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL|PR_IP6_SADDRSEL
84 #else
85 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL
86 #endif
87 #else /* !INET */
88 #ifdef INET6
89 #define	_PR_IP_SADDRSEL	PR_IP6_SADDRSEL
90 #else
91 #define	_PR_IP_SADDRSEL	0
92 #endif
93 #endif
94 
95 /* prison0 describes what is "real" about the system. */
96 struct prison prison0 = {
97 	.pr_id		= 0,
98 	.pr_name	= "0",
99 	.pr_ref		= 1,
100 	.pr_uref	= 1,
101 	.pr_path	= "/",
102 	.pr_securelevel	= -1,
103 	.pr_devfs_rsnum = 0,
104 	.pr_childmax	= JAIL_MAX,
105 	.pr_hostuuid	= DEFAULT_HOSTUUID,
106 	.pr_children	= LIST_HEAD_INITIALIZER(prison0.pr_children),
107 #ifdef VIMAGE
108 	.pr_flags	= PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
109 #else
110 	.pr_flags	= PR_HOST|_PR_IP_SADDRSEL,
111 #endif
112 	.pr_allow	= PR_ALLOW_ALL,
113 };
114 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
115 
116 /* allprison, allprison_racct and lastprid are protected by allprison_lock. */
117 struct	sx allprison_lock;
118 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
119 struct	prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
120 LIST_HEAD(, prison_racct) allprison_racct;
121 int	lastprid = 0;
122 
123 static int do_jail_attach(struct thread *td, struct prison *pr);
124 static void prison_complete(void *context, int pending);
125 static void prison_deref(struct prison *pr, int flags);
126 static char *prison_path(struct prison *pr1, struct prison *pr2);
127 static void prison_remove_one(struct prison *pr);
128 #ifdef RACCT
129 static void prison_racct_attach(struct prison *pr);
130 static void prison_racct_modify(struct prison *pr);
131 static void prison_racct_detach(struct prison *pr);
132 #endif
133 #ifdef INET
134 static int _prison_check_ip4(const struct prison *, const struct in_addr *);
135 static int prison_restrict_ip4(struct prison *pr, struct in_addr *newip4);
136 #endif
137 #ifdef INET6
138 static int _prison_check_ip6(struct prison *pr, struct in6_addr *ia6);
139 static int prison_restrict_ip6(struct prison *pr, struct in6_addr *newip6);
140 #endif
141 
142 /* Flags for prison_deref */
143 #define	PD_DEREF	0x01
144 #define	PD_DEUREF	0x02
145 #define	PD_LOCKED	0x04
146 #define	PD_LIST_SLOCKED	0x08
147 #define	PD_LIST_XLOCKED	0x10
148 
149 /*
150  * Parameter names corresponding to PR_* flag values.  Size values are for kvm
151  * as we cannot figure out the size of a sparse array, or an array without a
152  * terminating entry.
153  */
154 static char *pr_flag_names[] = {
155 	[0] = "persist",
156 #ifdef INET
157 	[7] = "ip4.saddrsel",
158 #endif
159 #ifdef INET6
160 	[8] = "ip6.saddrsel",
161 #endif
162 };
163 const size_t pr_flag_names_size = sizeof(pr_flag_names);
164 
165 static char *pr_flag_nonames[] = {
166 	[0] = "nopersist",
167 #ifdef INET
168 	[7] = "ip4.nosaddrsel",
169 #endif
170 #ifdef INET6
171 	[8] = "ip6.nosaddrsel",
172 #endif
173 };
174 const size_t pr_flag_nonames_size = sizeof(pr_flag_nonames);
175 
176 struct jailsys_flags {
177 	const char	*name;
178 	unsigned	 disable;
179 	unsigned	 new;
180 } pr_flag_jailsys[] = {
181 	{ "host", 0, PR_HOST },
182 #ifdef VIMAGE
183 	{ "vnet", 0, PR_VNET },
184 #endif
185 #ifdef INET
186 	{ "ip4", PR_IP4_USER, PR_IP4_USER },
187 #endif
188 #ifdef INET6
189 	{ "ip6", PR_IP6_USER, PR_IP6_USER },
190 #endif
191 };
192 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
193 
194 static char *pr_allow_names[] = {
195 	"allow.set_hostname",
196 	"allow.sysvipc",
197 	"allow.raw_sockets",
198 	"allow.chflags",
199 	"allow.mount",
200 	"allow.quotas",
201 	"allow.socket_af",
202 	"allow.mount.devfs",
203 	"allow.mount.nullfs",
204 	"allow.mount.zfs",
205 	"allow.mount.procfs",
206 	"allow.mount.tmpfs",
207 	"allow.mount.fdescfs",
208 	"allow.mount.linprocfs",
209 	"allow.mount.linsysfs",
210 };
211 const size_t pr_allow_names_size = sizeof(pr_allow_names);
212 
213 static char *pr_allow_nonames[] = {
214 	"allow.noset_hostname",
215 	"allow.nosysvipc",
216 	"allow.noraw_sockets",
217 	"allow.nochflags",
218 	"allow.nomount",
219 	"allow.noquotas",
220 	"allow.nosocket_af",
221 	"allow.mount.nodevfs",
222 	"allow.mount.nonullfs",
223 	"allow.mount.nozfs",
224 	"allow.mount.noprocfs",
225 	"allow.mount.notmpfs",
226 	"allow.mount.nofdescfs",
227 	"allow.mount.nolinprocfs",
228 	"allow.mount.nolinsysfs",
229 };
230 const size_t pr_allow_nonames_size = sizeof(pr_allow_nonames);
231 
232 #define	JAIL_DEFAULT_ALLOW		PR_ALLOW_SET_HOSTNAME
233 #define	JAIL_DEFAULT_ENFORCE_STATFS	2
234 #define	JAIL_DEFAULT_DEVFS_RSNUM	0
235 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
236 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
237 static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM;
238 #if defined(INET) || defined(INET6)
239 static unsigned jail_max_af_ips = 255;
240 #endif
241 
242 /*
243  * Initialize the parts of prison0 that can't be static-initialized with
244  * constants.  This is called from proc0_init() after creating thread0 cpuset.
245  */
246 void
247 prison0_init(void)
248 {
249 
250 	prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset);
251 	prison0.pr_osreldate = osreldate;
252 	strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease));
253 }
254 
255 #ifdef INET
256 static int
257 qcmp_v4(const void *ip1, const void *ip2)
258 {
259 	in_addr_t iaa, iab;
260 
261 	/*
262 	 * We need to compare in HBO here to get the list sorted as expected
263 	 * by the result of the code.  Sorting NBO addresses gives you
264 	 * interesting results.  If you do not understand, do not try.
265 	 */
266 	iaa = ntohl(((const struct in_addr *)ip1)->s_addr);
267 	iab = ntohl(((const struct in_addr *)ip2)->s_addr);
268 
269 	/*
270 	 * Do not simply return the difference of the two numbers, the int is
271 	 * not wide enough.
272 	 */
273 	if (iaa > iab)
274 		return (1);
275 	else if (iaa < iab)
276 		return (-1);
277 	else
278 		return (0);
279 }
280 #endif
281 
282 #ifdef INET6
283 static int
284 qcmp_v6(const void *ip1, const void *ip2)
285 {
286 	const struct in6_addr *ia6a, *ia6b;
287 	int i, rc;
288 
289 	ia6a = (const struct in6_addr *)ip1;
290 	ia6b = (const struct in6_addr *)ip2;
291 
292 	rc = 0;
293 	for (i = 0; rc == 0 && i < sizeof(struct in6_addr); i++) {
294 		if (ia6a->s6_addr[i] > ia6b->s6_addr[i])
295 			rc = 1;
296 		else if (ia6a->s6_addr[i] < ia6b->s6_addr[i])
297 			rc = -1;
298 	}
299 	return (rc);
300 }
301 #endif
302 
303 /*
304  * struct jail_args {
305  *	struct jail *jail;
306  * };
307  */
308 int
309 sys_jail(struct thread *td, struct jail_args *uap)
310 {
311 	uint32_t version;
312 	int error;
313 	struct jail j;
314 
315 	error = copyin(uap->jail, &version, sizeof(uint32_t));
316 	if (error)
317 		return (error);
318 
319 	switch (version) {
320 	case 0:
321 	{
322 		struct jail_v0 j0;
323 
324 		/* FreeBSD single IPv4 jails. */
325 		bzero(&j, sizeof(struct jail));
326 		error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
327 		if (error)
328 			return (error);
329 		j.version = j0.version;
330 		j.path = j0.path;
331 		j.hostname = j0.hostname;
332 		j.ip4s = htonl(j0.ip_number);	/* jail_v0 is host order */
333 		break;
334 	}
335 
336 	case 1:
337 		/*
338 		 * Version 1 was used by multi-IPv4 jail implementations
339 		 * that never made it into the official kernel.
340 		 */
341 		return (EINVAL);
342 
343 	case 2:	/* JAIL_API_VERSION */
344 		/* FreeBSD multi-IPv4/IPv6,noIP jails. */
345 		error = copyin(uap->jail, &j, sizeof(struct jail));
346 		if (error)
347 			return (error);
348 		break;
349 
350 	default:
351 		/* Sci-Fi jails are not supported, sorry. */
352 		return (EINVAL);
353 	}
354 	return (kern_jail(td, &j));
355 }
356 
357 int
358 kern_jail(struct thread *td, struct jail *j)
359 {
360 	struct iovec optiov[2 * (4 + nitems(pr_allow_names)
361 #ifdef INET
362 			    + 1
363 #endif
364 #ifdef INET6
365 			    + 1
366 #endif
367 			    )];
368 	struct uio opt;
369 	char *u_path, *u_hostname, *u_name;
370 #ifdef INET
371 	uint32_t ip4s;
372 	struct in_addr *u_ip4;
373 #endif
374 #ifdef INET6
375 	struct in6_addr *u_ip6;
376 #endif
377 	size_t tmplen;
378 	int error, enforce_statfs, fi;
379 
380 	bzero(&optiov, sizeof(optiov));
381 	opt.uio_iov = optiov;
382 	opt.uio_iovcnt = 0;
383 	opt.uio_offset = -1;
384 	opt.uio_resid = -1;
385 	opt.uio_segflg = UIO_SYSSPACE;
386 	opt.uio_rw = UIO_READ;
387 	opt.uio_td = td;
388 
389 	/* Set permissions for top-level jails from sysctls. */
390 	if (!jailed(td->td_ucred)) {
391 		for (fi = 0; fi < nitems(pr_allow_names); fi++) {
392 			optiov[opt.uio_iovcnt].iov_base =
393 			    (jail_default_allow & (1 << fi))
394 			    ? pr_allow_names[fi] : pr_allow_nonames[fi];
395 			optiov[opt.uio_iovcnt].iov_len =
396 			    strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
397 			opt.uio_iovcnt += 2;
398 		}
399 		optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
400 		optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
401 		opt.uio_iovcnt++;
402 		enforce_statfs = jail_default_enforce_statfs;
403 		optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
404 		optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
405 		opt.uio_iovcnt++;
406 	}
407 
408 	tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
409 #ifdef INET
410 	ip4s = (j->version == 0) ? 1 : j->ip4s;
411 	if (ip4s > jail_max_af_ips)
412 		return (EINVAL);
413 	tmplen += ip4s * sizeof(struct in_addr);
414 #else
415 	if (j->ip4s > 0)
416 		return (EINVAL);
417 #endif
418 #ifdef INET6
419 	if (j->ip6s > jail_max_af_ips)
420 		return (EINVAL);
421 	tmplen += j->ip6s * sizeof(struct in6_addr);
422 #else
423 	if (j->ip6s > 0)
424 		return (EINVAL);
425 #endif
426 	u_path = malloc(tmplen, M_TEMP, M_WAITOK);
427 	u_hostname = u_path + MAXPATHLEN;
428 	u_name = u_hostname + MAXHOSTNAMELEN;
429 #ifdef INET
430 	u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
431 #endif
432 #ifdef INET6
433 #ifdef INET
434 	u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
435 #else
436 	u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
437 #endif
438 #endif
439 	optiov[opt.uio_iovcnt].iov_base = "path";
440 	optiov[opt.uio_iovcnt].iov_len = sizeof("path");
441 	opt.uio_iovcnt++;
442 	optiov[opt.uio_iovcnt].iov_base = u_path;
443 	error = copyinstr(j->path, u_path, MAXPATHLEN,
444 	    &optiov[opt.uio_iovcnt].iov_len);
445 	if (error) {
446 		free(u_path, M_TEMP);
447 		return (error);
448 	}
449 	opt.uio_iovcnt++;
450 	optiov[opt.uio_iovcnt].iov_base = "host.hostname";
451 	optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
452 	opt.uio_iovcnt++;
453 	optiov[opt.uio_iovcnt].iov_base = u_hostname;
454 	error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
455 	    &optiov[opt.uio_iovcnt].iov_len);
456 	if (error) {
457 		free(u_path, M_TEMP);
458 		return (error);
459 	}
460 	opt.uio_iovcnt++;
461 	if (j->jailname != NULL) {
462 		optiov[opt.uio_iovcnt].iov_base = "name";
463 		optiov[opt.uio_iovcnt].iov_len = sizeof("name");
464 		opt.uio_iovcnt++;
465 		optiov[opt.uio_iovcnt].iov_base = u_name;
466 		error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
467 		    &optiov[opt.uio_iovcnt].iov_len);
468 		if (error) {
469 			free(u_path, M_TEMP);
470 			return (error);
471 		}
472 		opt.uio_iovcnt++;
473 	}
474 #ifdef INET
475 	optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
476 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
477 	opt.uio_iovcnt++;
478 	optiov[opt.uio_iovcnt].iov_base = u_ip4;
479 	optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
480 	if (j->version == 0)
481 		u_ip4->s_addr = j->ip4s;
482 	else {
483 		error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
484 		if (error) {
485 			free(u_path, M_TEMP);
486 			return (error);
487 		}
488 	}
489 	opt.uio_iovcnt++;
490 #endif
491 #ifdef INET6
492 	optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
493 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
494 	opt.uio_iovcnt++;
495 	optiov[opt.uio_iovcnt].iov_base = u_ip6;
496 	optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
497 	error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
498 	if (error) {
499 		free(u_path, M_TEMP);
500 		return (error);
501 	}
502 	opt.uio_iovcnt++;
503 #endif
504 	KASSERT(opt.uio_iovcnt <= nitems(optiov),
505 		("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
506 	error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
507 	free(u_path, M_TEMP);
508 	return (error);
509 }
510 
511 
512 /*
513  * struct jail_set_args {
514  *	struct iovec *iovp;
515  *	unsigned int iovcnt;
516  *	int flags;
517  * };
518  */
519 int
520 sys_jail_set(struct thread *td, struct jail_set_args *uap)
521 {
522 	struct uio *auio;
523 	int error;
524 
525 	/* Check that we have an even number of iovecs. */
526 	if (uap->iovcnt & 1)
527 		return (EINVAL);
528 
529 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
530 	if (error)
531 		return (error);
532 	error = kern_jail_set(td, auio, uap->flags);
533 	free(auio, M_IOV);
534 	return (error);
535 }
536 
537 int
538 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
539 {
540 	struct nameidata nd;
541 #ifdef INET
542 	struct in_addr *ip4;
543 #endif
544 #ifdef INET6
545 	struct in6_addr *ip6;
546 #endif
547 	struct vfsopt *opt;
548 	struct vfsoptlist *opts;
549 	struct prison *pr, *deadpr, *mypr, *ppr, *tpr;
550 	struct vnode *root;
551 	char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
552 	char *g_path, *osrelstr;
553 #if defined(INET) || defined(INET6)
554 	struct prison *tppr;
555 	void *op;
556 #endif
557 	unsigned long hid;
558 	size_t namelen, onamelen, pnamelen;
559 	int born, created, cuflags, descend, enforce;
560 	int error, errmsg_len, errmsg_pos;
561 	int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel;
562 	int fi, jid, jsys, len, level;
563 	int childmax, osreldt, rsnum, slevel;
564 	int fullpath_disabled;
565 #if defined(INET) || defined(INET6)
566 	int ii, ij;
567 #endif
568 #ifdef INET
569 	int ip4s, redo_ip4;
570 #endif
571 #ifdef INET6
572 	int ip6s, redo_ip6;
573 #endif
574 	uint64_t pr_allow, ch_allow, pr_flags, ch_flags;
575 	unsigned tallow;
576 	char numbuf[12];
577 
578 	error = priv_check(td, PRIV_JAIL_SET);
579 	if (!error && (flags & JAIL_ATTACH))
580 		error = priv_check(td, PRIV_JAIL_ATTACH);
581 	if (error)
582 		return (error);
583 	mypr = td->td_ucred->cr_prison;
584 	if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
585 		return (EPERM);
586 	if (flags & ~JAIL_SET_MASK)
587 		return (EINVAL);
588 
589 	/*
590 	 * Check all the parameters before committing to anything.  Not all
591 	 * errors can be caught early, but we may as well try.  Also, this
592 	 * takes care of some expensive stuff (path lookup) before getting
593 	 * the allprison lock.
594 	 *
595 	 * XXX Jails are not filesystems, and jail parameters are not mount
596 	 *     options.  But it makes more sense to re-use the vfsopt code
597 	 *     than duplicate it under a different name.
598 	 */
599 	error = vfs_buildopts(optuio, &opts);
600 	if (error)
601 		return (error);
602 #ifdef INET
603 	ip4 = NULL;
604 #endif
605 #ifdef INET6
606 	ip6 = NULL;
607 #endif
608 	g_path = NULL;
609 
610 	cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
611 	if (!cuflags) {
612 		error = EINVAL;
613 		vfs_opterror(opts, "no valid operation (create or update)");
614 		goto done_errmsg;
615 	}
616 
617 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
618 	if (error == ENOENT)
619 		jid = 0;
620 	else if (error != 0)
621 		goto done_free;
622 
623 	error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
624 	if (error == ENOENT)
625 		gotslevel = 0;
626 	else if (error != 0)
627 		goto done_free;
628 	else
629 		gotslevel = 1;
630 
631 	error =
632 	    vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
633 	if (error == ENOENT)
634 		gotchildmax = 0;
635 	else if (error != 0)
636 		goto done_free;
637 	else
638 		gotchildmax = 1;
639 
640 	error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
641 	if (error == ENOENT)
642 		gotenforce = 0;
643 	else if (error != 0)
644 		goto done_free;
645 	else if (enforce < 0 || enforce > 2) {
646 		error = EINVAL;
647 		goto done_free;
648 	} else
649 		gotenforce = 1;
650 
651 	error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum));
652 	if (error == ENOENT)
653 		gotrsnum = 0;
654 	else if (error != 0)
655 		goto done_free;
656 	else
657 		gotrsnum = 1;
658 
659 	pr_flags = ch_flags = 0;
660 	for (fi = 0; fi < nitems(pr_flag_names); fi++) {
661 		if (pr_flag_names[fi] == NULL)
662 			continue;
663 		vfs_flagopt(opts, pr_flag_names[fi], &pr_flags, 1 << fi);
664 		vfs_flagopt(opts, pr_flag_nonames[fi], &ch_flags, 1 << fi);
665 	}
666 	ch_flags |= pr_flags;
667 	for (fi = 0; fi < nitems(pr_flag_jailsys); fi++) {
668 		error = vfs_copyopt(opts, pr_flag_jailsys[fi].name, &jsys,
669 		    sizeof(jsys));
670 		if (error == ENOENT)
671 			continue;
672 		if (error != 0)
673 			goto done_free;
674 		switch (jsys) {
675 		case JAIL_SYS_DISABLE:
676 			if (!pr_flag_jailsys[fi].disable) {
677 				error = EINVAL;
678 				goto done_free;
679 			}
680 			pr_flags |= pr_flag_jailsys[fi].disable;
681 			break;
682 		case JAIL_SYS_NEW:
683 			pr_flags |= pr_flag_jailsys[fi].new;
684 			break;
685 		case JAIL_SYS_INHERIT:
686 			break;
687 		default:
688 			error = EINVAL;
689 			goto done_free;
690 		}
691 		ch_flags |=
692 		    pr_flag_jailsys[fi].new | pr_flag_jailsys[fi].disable;
693 	}
694 	if ((flags & (JAIL_CREATE | JAIL_UPDATE | JAIL_ATTACH)) == JAIL_CREATE
695 	    && !(pr_flags & PR_PERSIST)) {
696 		error = EINVAL;
697 		vfs_opterror(opts, "new jail must persist or attach");
698 		goto done_errmsg;
699 	}
700 #ifdef VIMAGE
701 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
702 		error = EINVAL;
703 		vfs_opterror(opts, "vnet cannot be changed after creation");
704 		goto done_errmsg;
705 	}
706 #endif
707 #ifdef INET
708 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
709 		error = EINVAL;
710 		vfs_opterror(opts, "ip4 cannot be changed after creation");
711 		goto done_errmsg;
712 	}
713 #endif
714 #ifdef INET6
715 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
716 		error = EINVAL;
717 		vfs_opterror(opts, "ip6 cannot be changed after creation");
718 		goto done_errmsg;
719 	}
720 #endif
721 
722 	pr_allow = ch_allow = 0;
723 	for (fi = 0; fi < nitems(pr_allow_names); fi++) {
724 		vfs_flagopt(opts, pr_allow_names[fi], &pr_allow, 1 << fi);
725 		vfs_flagopt(opts, pr_allow_nonames[fi], &ch_allow, 1 << fi);
726 	}
727 	ch_allow |= pr_allow;
728 
729 	error = vfs_getopt(opts, "name", (void **)&name, &len);
730 	if (error == ENOENT)
731 		name = NULL;
732 	else if (error != 0)
733 		goto done_free;
734 	else {
735 		if (len == 0 || name[len - 1] != '\0') {
736 			error = EINVAL;
737 			goto done_free;
738 		}
739 		if (len > MAXHOSTNAMELEN) {
740 			error = ENAMETOOLONG;
741 			goto done_free;
742 		}
743 	}
744 
745 	error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
746 	if (error == ENOENT)
747 		host = NULL;
748 	else if (error != 0)
749 		goto done_free;
750 	else {
751 		ch_flags |= PR_HOST;
752 		pr_flags |= PR_HOST;
753 		if (len == 0 || host[len - 1] != '\0') {
754 			error = EINVAL;
755 			goto done_free;
756 		}
757 		if (len > MAXHOSTNAMELEN) {
758 			error = ENAMETOOLONG;
759 			goto done_free;
760 		}
761 	}
762 
763 	error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
764 	if (error == ENOENT)
765 		domain = NULL;
766 	else if (error != 0)
767 		goto done_free;
768 	else {
769 		ch_flags |= PR_HOST;
770 		pr_flags |= PR_HOST;
771 		if (len == 0 || domain[len - 1] != '\0') {
772 			error = EINVAL;
773 			goto done_free;
774 		}
775 		if (len > MAXHOSTNAMELEN) {
776 			error = ENAMETOOLONG;
777 			goto done_free;
778 		}
779 	}
780 
781 	error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
782 	if (error == ENOENT)
783 		uuid = NULL;
784 	else if (error != 0)
785 		goto done_free;
786 	else {
787 		ch_flags |= PR_HOST;
788 		pr_flags |= PR_HOST;
789 		if (len == 0 || uuid[len - 1] != '\0') {
790 			error = EINVAL;
791 			goto done_free;
792 		}
793 		if (len > HOSTUUIDLEN) {
794 			error = ENAMETOOLONG;
795 			goto done_free;
796 		}
797 	}
798 
799 #ifdef COMPAT_FREEBSD32
800 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
801 		uint32_t hid32;
802 
803 		error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
804 		hid = hid32;
805 	} else
806 #endif
807 		error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
808 	if (error == ENOENT)
809 		gothid = 0;
810 	else if (error != 0)
811 		goto done_free;
812 	else {
813 		gothid = 1;
814 		ch_flags |= PR_HOST;
815 		pr_flags |= PR_HOST;
816 	}
817 
818 #ifdef INET
819 	error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
820 	if (error == ENOENT)
821 		ip4s = 0;
822 	else if (error != 0)
823 		goto done_free;
824 	else if (ip4s & (sizeof(*ip4) - 1)) {
825 		error = EINVAL;
826 		goto done_free;
827 	} else {
828 		ch_flags |= PR_IP4_USER;
829 		pr_flags |= PR_IP4_USER;
830 		if (ip4s > 0) {
831 			ip4s /= sizeof(*ip4);
832 			if (ip4s > jail_max_af_ips) {
833 				error = EINVAL;
834 				vfs_opterror(opts, "too many IPv4 addresses");
835 				goto done_errmsg;
836 			}
837 			ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
838 			bcopy(op, ip4, ip4s * sizeof(*ip4));
839 			/*
840 			 * IP addresses are all sorted but ip[0] to preserve
841 			 * the primary IP address as given from userland.
842 			 * This special IP is used for unbound outgoing
843 			 * connections as well for "loopback" traffic in case
844 			 * source address selection cannot find any more fitting
845 			 * address to connect from.
846 			 */
847 			if (ip4s > 1)
848 				qsort(ip4 + 1, ip4s - 1, sizeof(*ip4), qcmp_v4);
849 			/*
850 			 * Check for duplicate addresses and do some simple
851 			 * zero and broadcast checks. If users give other bogus
852 			 * addresses it is their problem.
853 			 *
854 			 * We do not have to care about byte order for these
855 			 * checks so we will do them in NBO.
856 			 */
857 			for (ii = 0; ii < ip4s; ii++) {
858 				if (ip4[ii].s_addr == INADDR_ANY ||
859 				    ip4[ii].s_addr == INADDR_BROADCAST) {
860 					error = EINVAL;
861 					goto done_free;
862 				}
863 				if ((ii+1) < ip4s &&
864 				    (ip4[0].s_addr == ip4[ii+1].s_addr ||
865 				     ip4[ii].s_addr == ip4[ii+1].s_addr)) {
866 					error = EINVAL;
867 					goto done_free;
868 				}
869 			}
870 		}
871 	}
872 #endif
873 
874 #ifdef INET6
875 	error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
876 	if (error == ENOENT)
877 		ip6s = 0;
878 	else if (error != 0)
879 		goto done_free;
880 	else if (ip6s & (sizeof(*ip6) - 1)) {
881 		error = EINVAL;
882 		goto done_free;
883 	} else {
884 		ch_flags |= PR_IP6_USER;
885 		pr_flags |= PR_IP6_USER;
886 		if (ip6s > 0) {
887 			ip6s /= sizeof(*ip6);
888 			if (ip6s > jail_max_af_ips) {
889 				error = EINVAL;
890 				vfs_opterror(opts, "too many IPv6 addresses");
891 				goto done_errmsg;
892 			}
893 			ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
894 			bcopy(op, ip6, ip6s * sizeof(*ip6));
895 			if (ip6s > 1)
896 				qsort(ip6 + 1, ip6s - 1, sizeof(*ip6), qcmp_v6);
897 			for (ii = 0; ii < ip6s; ii++) {
898 				if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
899 					error = EINVAL;
900 					goto done_free;
901 				}
902 				if ((ii+1) < ip6s &&
903 				    (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
904 				     IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
905 				{
906 					error = EINVAL;
907 					goto done_free;
908 				}
909 			}
910 		}
911 	}
912 #endif
913 
914 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
915 	if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
916 		error = EINVAL;
917 		vfs_opterror(opts,
918 		    "vnet jails cannot have IP address restrictions");
919 		goto done_errmsg;
920 	}
921 #endif
922 
923 	error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len);
924 	if (error == ENOENT)
925 		osrelstr = NULL;
926 	else if (error != 0)
927 		goto done_free;
928 	else {
929 		if (flags & JAIL_UPDATE) {
930 			error = EINVAL;
931 			vfs_opterror(opts,
932 			    "osrelease cannot be changed after creation");
933 			goto done_errmsg;
934 		}
935 		if (len == 0 || len >= OSRELEASELEN) {
936 			error = EINVAL;
937 			vfs_opterror(opts,
938 			    "osrelease string must be 1-%d bytes long",
939 			    OSRELEASELEN - 1);
940 			goto done_errmsg;
941 		}
942 	}
943 
944 	error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt));
945 	if (error == ENOENT)
946 		osreldt = 0;
947 	else if (error != 0)
948 		goto done_free;
949 	else {
950 		if (flags & JAIL_UPDATE) {
951 			error = EINVAL;
952 			vfs_opterror(opts,
953 			    "osreldate cannot be changed after creation");
954 			goto done_errmsg;
955 		}
956 		if (osreldt == 0) {
957 			error = EINVAL;
958 			vfs_opterror(opts, "osreldate cannot be 0");
959 			goto done_errmsg;
960 		}
961 	}
962 
963 	fullpath_disabled = 0;
964 	root = NULL;
965 	error = vfs_getopt(opts, "path", (void **)&path, &len);
966 	if (error == ENOENT)
967 		path = NULL;
968 	else if (error != 0)
969 		goto done_free;
970 	else {
971 		if (flags & JAIL_UPDATE) {
972 			error = EINVAL;
973 			vfs_opterror(opts,
974 			    "path cannot be changed after creation");
975 			goto done_errmsg;
976 		}
977 		if (len == 0 || path[len - 1] != '\0') {
978 			error = EINVAL;
979 			goto done_free;
980 		}
981 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
982 		    path, td);
983 		error = namei(&nd);
984 		if (error)
985 			goto done_free;
986 		root = nd.ni_vp;
987 		NDFREE(&nd, NDF_ONLY_PNBUF);
988 		g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
989 		strlcpy(g_path, path, MAXPATHLEN);
990 		error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN);
991 		if (error == 0)
992 			path = g_path;
993 		else if (error == ENODEV) {
994 			/* proceed if sysctl debug.disablefullpath == 1 */
995 			fullpath_disabled = 1;
996 			if (len < 2 || (len == 2 && path[0] == '/'))
997 				path = NULL;
998 		} else {
999 			/* exit on other errors */
1000 			goto done_free;
1001 		}
1002 		if (root->v_type != VDIR) {
1003 			error = ENOTDIR;
1004 			vput(root);
1005 			goto done_free;
1006 		}
1007 		VOP_UNLOCK(root, 0);
1008 		if (fullpath_disabled) {
1009 			/* Leave room for a real-root full pathname. */
1010 			if (len + (path[0] == '/' && strcmp(mypr->pr_path, "/")
1011 			    ? strlen(mypr->pr_path) : 0) > MAXPATHLEN) {
1012 				error = ENAMETOOLONG;
1013 				vrele(root);
1014 				goto done_free;
1015 			}
1016 		}
1017 	}
1018 
1019 	/*
1020 	 * Find the specified jail, or at least its parent.
1021 	 * This abuses the file error codes ENOENT and EEXIST.
1022 	 */
1023 	pr = NULL;
1024 	ppr = mypr;
1025 	if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
1026 		namelc = strrchr(name, '.');
1027 		jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
1028 		if (*p != '\0')
1029 			jid = 0;
1030 	}
1031 	sx_xlock(&allprison_lock);
1032 	if (jid != 0) {
1033 		/*
1034 		 * See if a requested jid already exists.  There is an
1035 		 * information leak here if the jid exists but is not within
1036 		 * the caller's jail hierarchy.  Jail creators will get EEXIST
1037 		 * even though they cannot see the jail, and CREATE | UPDATE
1038 		 * will return ENOENT which is not normally a valid error.
1039 		 */
1040 		if (jid < 0) {
1041 			error = EINVAL;
1042 			vfs_opterror(opts, "negative jid");
1043 			goto done_unlock_list;
1044 		}
1045 		pr = prison_find(jid);
1046 		if (pr != NULL) {
1047 			ppr = pr->pr_parent;
1048 			/* Create: jid must not exist. */
1049 			if (cuflags == JAIL_CREATE) {
1050 				mtx_unlock(&pr->pr_mtx);
1051 				error = EEXIST;
1052 				vfs_opterror(opts, "jail %d already exists",
1053 				    jid);
1054 				goto done_unlock_list;
1055 			}
1056 			if (!prison_ischild(mypr, pr)) {
1057 				mtx_unlock(&pr->pr_mtx);
1058 				pr = NULL;
1059 			} else if (pr->pr_uref == 0) {
1060 				if (!(flags & JAIL_DYING)) {
1061 					mtx_unlock(&pr->pr_mtx);
1062 					error = ENOENT;
1063 					vfs_opterror(opts, "jail %d is dying",
1064 					    jid);
1065 					goto done_unlock_list;
1066 				} else if ((flags & JAIL_ATTACH) ||
1067 				    (pr_flags & PR_PERSIST)) {
1068 					/*
1069 					 * A dying jail might be resurrected
1070 					 * (via attach or persist), but first
1071 					 * it must determine if another jail
1072 					 * has claimed its name.  Accomplish
1073 					 * this by implicitly re-setting the
1074 					 * name.
1075 					 */
1076 					if (name == NULL)
1077 						name = prison_name(mypr, pr);
1078 				}
1079 			}
1080 		}
1081 		if (pr == NULL) {
1082 			/* Update: jid must exist. */
1083 			if (cuflags == JAIL_UPDATE) {
1084 				error = ENOENT;
1085 				vfs_opterror(opts, "jail %d not found", jid);
1086 				goto done_unlock_list;
1087 			}
1088 		}
1089 	}
1090 	/*
1091 	 * If the caller provided a name, look for a jail by that name.
1092 	 * This has different semantics for creates and updates keyed by jid
1093 	 * (where the name must not already exist in a different jail),
1094 	 * and updates keyed by the name itself (where the name must exist
1095 	 * because that is the jail being updated).
1096 	 */
1097 	namelc = NULL;
1098 	if (name != NULL) {
1099 		namelc = strrchr(name, '.');
1100 		if (namelc == NULL)
1101 			namelc = name;
1102 		else {
1103 			/*
1104 			 * This is a hierarchical name.  Split it into the
1105 			 * parent and child names, and make sure the parent
1106 			 * exists or matches an already found jail.
1107 			 */
1108 			if (pr != NULL) {
1109 				if (strncmp(name, ppr->pr_name, namelc - name)
1110 				    || ppr->pr_name[namelc - name] != '\0') {
1111 					mtx_unlock(&pr->pr_mtx);
1112 					error = EINVAL;
1113 					vfs_opterror(opts,
1114 					    "cannot change jail's parent");
1115 					goto done_unlock_list;
1116 				}
1117 			} else {
1118 				*namelc = '\0';
1119 				ppr = prison_find_name(mypr, name);
1120 				if (ppr == NULL) {
1121 					error = ENOENT;
1122 					vfs_opterror(opts,
1123 					    "jail \"%s\" not found", name);
1124 					goto done_unlock_list;
1125 				}
1126 				mtx_unlock(&ppr->pr_mtx);
1127 				*namelc = '.';
1128 			}
1129 			namelc++;
1130 		}
1131 		if (namelc[0] != '\0') {
1132 			pnamelen =
1133 			    (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1134  name_again:
1135 			deadpr = NULL;
1136 			FOREACH_PRISON_CHILD(ppr, tpr) {
1137 				if (tpr != pr && tpr->pr_ref > 0 &&
1138 				    !strcmp(tpr->pr_name + pnamelen, namelc)) {
1139 					if (pr == NULL &&
1140 					    cuflags != JAIL_CREATE) {
1141 						mtx_lock(&tpr->pr_mtx);
1142 						if (tpr->pr_ref > 0) {
1143 							/*
1144 							 * Use this jail
1145 							 * for updates.
1146 							 */
1147 							if (tpr->pr_uref > 0) {
1148 								pr = tpr;
1149 								break;
1150 							}
1151 							deadpr = tpr;
1152 						}
1153 						mtx_unlock(&tpr->pr_mtx);
1154 					} else if (tpr->pr_uref > 0) {
1155 						/*
1156 						 * Create, or update(jid):
1157 						 * name must not exist in an
1158 						 * active sibling jail.
1159 						 */
1160 						error = EEXIST;
1161 						if (pr != NULL)
1162 							mtx_unlock(&pr->pr_mtx);
1163 						vfs_opterror(opts,
1164 						   "jail \"%s\" already exists",
1165 						   name);
1166 						goto done_unlock_list;
1167 					}
1168 				}
1169 			}
1170 			/* If no active jail is found, use a dying one. */
1171 			if (deadpr != NULL && pr == NULL) {
1172 				if (flags & JAIL_DYING) {
1173 					mtx_lock(&deadpr->pr_mtx);
1174 					if (deadpr->pr_ref == 0) {
1175 						mtx_unlock(&deadpr->pr_mtx);
1176 						goto name_again;
1177 					}
1178 					pr = deadpr;
1179 				} else if (cuflags == JAIL_UPDATE) {
1180 					error = ENOENT;
1181 					vfs_opterror(opts,
1182 					    "jail \"%s\" is dying", name);
1183 					goto done_unlock_list;
1184 				}
1185 			}
1186 			/* Update: name must exist if no jid. */
1187 			else if (cuflags == JAIL_UPDATE && pr == NULL) {
1188 				error = ENOENT;
1189 				vfs_opterror(opts, "jail \"%s\" not found",
1190 				    name);
1191 				goto done_unlock_list;
1192 			}
1193 		}
1194 	}
1195 	/* Update: must provide a jid or name. */
1196 	else if (cuflags == JAIL_UPDATE && pr == NULL) {
1197 		error = ENOENT;
1198 		vfs_opterror(opts, "update specified no jail");
1199 		goto done_unlock_list;
1200 	}
1201 
1202 	/* If there's no prison to update, create a new one and link it in. */
1203 	if (pr == NULL) {
1204 		for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1205 			if (tpr->pr_childcount >= tpr->pr_childmax) {
1206 				error = EPERM;
1207 				vfs_opterror(opts, "prison limit exceeded");
1208 				goto done_unlock_list;
1209 			}
1210 		created = 1;
1211 		mtx_lock(&ppr->pr_mtx);
1212 		if (ppr->pr_ref == 0) {
1213 			mtx_unlock(&ppr->pr_mtx);
1214 			error = ENOENT;
1215 			vfs_opterror(opts, "jail \"%s\" not found",
1216 			    prison_name(mypr, ppr));
1217 			goto done_unlock_list;
1218 		}
1219 		ppr->pr_ref++;
1220 		ppr->pr_uref++;
1221 		mtx_unlock(&ppr->pr_mtx);
1222 		pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1223 		if (jid == 0) {
1224 			/* Find the next free jid. */
1225 			jid = lastprid + 1;
1226  findnext:
1227 			if (jid == JAIL_MAX)
1228 				jid = 1;
1229 			TAILQ_FOREACH(tpr, &allprison, pr_list) {
1230 				if (tpr->pr_id < jid)
1231 					continue;
1232 				if (tpr->pr_id > jid || tpr->pr_ref == 0) {
1233 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1234 					break;
1235 				}
1236 				if (jid == lastprid) {
1237 					error = EAGAIN;
1238 					vfs_opterror(opts,
1239 					    "no available jail IDs");
1240 					free(pr, M_PRISON);
1241 					prison_deref(ppr, PD_DEREF |
1242 					    PD_DEUREF | PD_LIST_XLOCKED);
1243 					goto done_releroot;
1244 				}
1245 				jid++;
1246 				goto findnext;
1247 			}
1248 			lastprid = jid;
1249 		} else {
1250 			/*
1251 			 * The jail already has a jid (that did not yet exist),
1252 			 * so just find where to insert it.
1253 			 */
1254 			TAILQ_FOREACH(tpr, &allprison, pr_list)
1255 				if (tpr->pr_id >= jid) {
1256 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1257 					break;
1258 				}
1259 		}
1260 		if (tpr == NULL)
1261 			TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1262 		LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1263 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1264 			tpr->pr_childcount++;
1265 
1266 		pr->pr_parent = ppr;
1267 		pr->pr_id = jid;
1268 
1269 		/* Set some default values, and inherit some from the parent. */
1270 		if (namelc == NULL)
1271 			namelc = "";
1272 		if (path == NULL) {
1273 			path = "/";
1274 			root = mypr->pr_root;
1275 			vref(root);
1276 		}
1277 		strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1278 		pr->pr_flags |= PR_HOST;
1279 #if defined(INET) || defined(INET6)
1280 #ifdef VIMAGE
1281 		if (!(pr_flags & PR_VNET))
1282 #endif
1283 		{
1284 #ifdef INET
1285 			if (!(ch_flags & PR_IP4_USER))
1286 				pr->pr_flags |= PR_IP4 | PR_IP4_USER;
1287 			else if (!(pr_flags & PR_IP4_USER)) {
1288 				pr->pr_flags |= ppr->pr_flags & PR_IP4;
1289 				if (ppr->pr_ip4 != NULL) {
1290 					pr->pr_ip4s = ppr->pr_ip4s;
1291 					pr->pr_ip4 = malloc(pr->pr_ip4s *
1292 					    sizeof(struct in_addr), M_PRISON,
1293 					    M_WAITOK);
1294 					bcopy(ppr->pr_ip4, pr->pr_ip4,
1295 					    pr->pr_ip4s * sizeof(*pr->pr_ip4));
1296 				}
1297 			}
1298 #endif
1299 #ifdef INET6
1300 			if (!(ch_flags & PR_IP6_USER))
1301 				pr->pr_flags |= PR_IP6 | PR_IP6_USER;
1302 			else if (!(pr_flags & PR_IP6_USER)) {
1303 				pr->pr_flags |= ppr->pr_flags & PR_IP6;
1304 				if (ppr->pr_ip6 != NULL) {
1305 					pr->pr_ip6s = ppr->pr_ip6s;
1306 					pr->pr_ip6 = malloc(pr->pr_ip6s *
1307 					    sizeof(struct in6_addr), M_PRISON,
1308 					    M_WAITOK);
1309 					bcopy(ppr->pr_ip6, pr->pr_ip6,
1310 					    pr->pr_ip6s * sizeof(*pr->pr_ip6));
1311 				}
1312 			}
1313 #endif
1314 		}
1315 #endif
1316 		/* Source address selection is always on by default. */
1317 		pr->pr_flags |= _PR_IP_SADDRSEL;
1318 
1319 		pr->pr_securelevel = ppr->pr_securelevel;
1320 		pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1321 		pr->pr_enforce_statfs = jail_default_enforce_statfs;
1322 		pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum;
1323 
1324 		pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate;
1325 		if (osrelstr == NULL)
1326 		    strcpy(pr->pr_osrelease, ppr->pr_osrelease);
1327 		else
1328 		    strcpy(pr->pr_osrelease, osrelstr);
1329 
1330 		LIST_INIT(&pr->pr_children);
1331 		mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1332 		TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
1333 
1334 #ifdef VIMAGE
1335 		/* Allocate a new vnet if specified. */
1336 		pr->pr_vnet = (pr_flags & PR_VNET)
1337 		    ? vnet_alloc() : ppr->pr_vnet;
1338 #endif
1339 		/*
1340 		 * Allocate a dedicated cpuset for each jail.
1341 		 * Unlike other initial settings, this may return an erorr.
1342 		 */
1343 		error = cpuset_create_root(ppr, &pr->pr_cpuset);
1344 		if (error) {
1345 			prison_deref(pr, PD_LIST_XLOCKED);
1346 			goto done_releroot;
1347 		}
1348 
1349 		mtx_lock(&pr->pr_mtx);
1350 		/*
1351 		 * New prisons do not yet have a reference, because we do not
1352 		 * want others to see the incomplete prison once the
1353 		 * allprison_lock is downgraded.
1354 		 */
1355 	} else {
1356 		created = 0;
1357 		/*
1358 		 * Grab a reference for existing prisons, to ensure they
1359 		 * continue to exist for the duration of the call.
1360 		 */
1361 		pr->pr_ref++;
1362 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1363 		if ((pr->pr_flags & PR_VNET) &&
1364 		    (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1365 			error = EINVAL;
1366 			vfs_opterror(opts,
1367 			    "vnet jails cannot have IP address restrictions");
1368 			goto done_deref_locked;
1369 		}
1370 #endif
1371 #ifdef INET
1372 		if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1373 			error = EINVAL;
1374 			vfs_opterror(opts,
1375 			    "ip4 cannot be changed after creation");
1376 			goto done_deref_locked;
1377 		}
1378 #endif
1379 #ifdef INET6
1380 		if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1381 			error = EINVAL;
1382 			vfs_opterror(opts,
1383 			    "ip6 cannot be changed after creation");
1384 			goto done_deref_locked;
1385 		}
1386 #endif
1387 	}
1388 
1389 	/* Do final error checking before setting anything. */
1390 	if (gotslevel) {
1391 		if (slevel < ppr->pr_securelevel) {
1392 			error = EPERM;
1393 			goto done_deref_locked;
1394 		}
1395 	}
1396 	if (gotchildmax) {
1397 		if (childmax >= ppr->pr_childmax) {
1398 			error = EPERM;
1399 			goto done_deref_locked;
1400 		}
1401 	}
1402 	if (gotenforce) {
1403 		if (enforce < ppr->pr_enforce_statfs) {
1404 			error = EPERM;
1405 			goto done_deref_locked;
1406 		}
1407 	}
1408 	if (gotrsnum) {
1409 		/*
1410 		 * devfs_rsnum is a uint16_t
1411 		 */
1412 		if (rsnum < 0 || rsnum > 65535) {
1413 			error = EINVAL;
1414 			goto done_deref_locked;
1415 		}
1416 		/*
1417 		 * Nested jails always inherit parent's devfs ruleset
1418 		 */
1419 		if (jailed(td->td_ucred)) {
1420 			if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) {
1421 				error = EPERM;
1422 				goto done_deref_locked;
1423 			} else
1424 				rsnum = ppr->pr_devfs_rsnum;
1425 		}
1426 	}
1427 #ifdef INET
1428 	if (ip4s > 0) {
1429 		if (ppr->pr_flags & PR_IP4) {
1430 			/*
1431 			 * Make sure the new set of IP addresses is a
1432 			 * subset of the parent's list.  Don't worry
1433 			 * about the parent being unlocked, as any
1434 			 * setting is done with allprison_lock held.
1435 			 */
1436 			for (ij = 0; ij < ppr->pr_ip4s; ij++)
1437 				if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
1438 					break;
1439 			if (ij == ppr->pr_ip4s) {
1440 				error = EPERM;
1441 				goto done_deref_locked;
1442 			}
1443 			if (ip4s > 1) {
1444 				for (ii = ij = 1; ii < ip4s; ii++) {
1445 					if (ip4[ii].s_addr ==
1446 					    ppr->pr_ip4[0].s_addr)
1447 						continue;
1448 					for (; ij < ppr->pr_ip4s; ij++)
1449 						if (ip4[ii].s_addr ==
1450 						    ppr->pr_ip4[ij].s_addr)
1451 							break;
1452 					if (ij == ppr->pr_ip4s)
1453 						break;
1454 				}
1455 				if (ij == ppr->pr_ip4s) {
1456 					error = EPERM;
1457 					goto done_deref_locked;
1458 				}
1459 			}
1460 		}
1461 		/*
1462 		 * Check for conflicting IP addresses.  We permit them
1463 		 * if there is no more than one IP on each jail.  If
1464 		 * there is a duplicate on a jail with more than one
1465 		 * IP stop checking and return error.
1466 		 */
1467 		tppr = ppr;
1468 #ifdef VIMAGE
1469 		for (; tppr != &prison0; tppr = tppr->pr_parent)
1470 			if (tppr->pr_flags & PR_VNET)
1471 				break;
1472 #endif
1473 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1474 			if (tpr == pr ||
1475 #ifdef VIMAGE
1476 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1477 #endif
1478 			    tpr->pr_uref == 0) {
1479 				descend = 0;
1480 				continue;
1481 			}
1482 			if (!(tpr->pr_flags & PR_IP4_USER))
1483 				continue;
1484 			descend = 0;
1485 			if (tpr->pr_ip4 == NULL ||
1486 			    (ip4s == 1 && tpr->pr_ip4s == 1))
1487 				continue;
1488 			for (ii = 0; ii < ip4s; ii++) {
1489 				if (_prison_check_ip4(tpr, &ip4[ii]) == 0) {
1490 					error = EADDRINUSE;
1491 					vfs_opterror(opts,
1492 					    "IPv4 addresses clash");
1493 					goto done_deref_locked;
1494 				}
1495 			}
1496 		}
1497 	}
1498 #endif
1499 #ifdef INET6
1500 	if (ip6s > 0) {
1501 		if (ppr->pr_flags & PR_IP6) {
1502 			/*
1503 			 * Make sure the new set of IP addresses is a
1504 			 * subset of the parent's list.
1505 			 */
1506 			for (ij = 0; ij < ppr->pr_ip6s; ij++)
1507 				if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1508 				    &ppr->pr_ip6[ij]))
1509 					break;
1510 			if (ij == ppr->pr_ip6s) {
1511 				error = EPERM;
1512 				goto done_deref_locked;
1513 			}
1514 			if (ip6s > 1) {
1515 				for (ii = ij = 1; ii < ip6s; ii++) {
1516 					if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1517 					     &ppr->pr_ip6[0]))
1518 						continue;
1519 					for (; ij < ppr->pr_ip6s; ij++)
1520 						if (IN6_ARE_ADDR_EQUAL(
1521 						    &ip6[ii], &ppr->pr_ip6[ij]))
1522 							break;
1523 					if (ij == ppr->pr_ip6s)
1524 						break;
1525 				}
1526 				if (ij == ppr->pr_ip6s) {
1527 					error = EPERM;
1528 					goto done_deref_locked;
1529 				}
1530 			}
1531 		}
1532 		/* Check for conflicting IP addresses. */
1533 		tppr = ppr;
1534 #ifdef VIMAGE
1535 		for (; tppr != &prison0; tppr = tppr->pr_parent)
1536 			if (tppr->pr_flags & PR_VNET)
1537 				break;
1538 #endif
1539 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1540 			if (tpr == pr ||
1541 #ifdef VIMAGE
1542 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1543 #endif
1544 			    tpr->pr_uref == 0) {
1545 				descend = 0;
1546 				continue;
1547 			}
1548 			if (!(tpr->pr_flags & PR_IP6_USER))
1549 				continue;
1550 			descend = 0;
1551 			if (tpr->pr_ip6 == NULL ||
1552 			    (ip6s == 1 && tpr->pr_ip6s == 1))
1553 				continue;
1554 			for (ii = 0; ii < ip6s; ii++) {
1555 				if (_prison_check_ip6(tpr, &ip6[ii]) == 0) {
1556 					error = EADDRINUSE;
1557 					vfs_opterror(opts,
1558 					    "IPv6 addresses clash");
1559 					goto done_deref_locked;
1560 				}
1561 			}
1562 		}
1563 	}
1564 #endif
1565 	onamelen = namelen = 0;
1566 	if (namelc != NULL) {
1567 		/* Give a default name of the jid.  Also allow the name to be
1568 		 * explicitly the jid - but not any other number, and only in
1569 		 * normal form (no leading zero/etc).
1570 		 */
1571 		if (namelc[0] == '\0')
1572 			snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid);
1573 		else if ((strtoul(namelc, &p, 10) != jid ||
1574 			  namelc[0] < '1' || namelc[0] > '9') && *p == '\0') {
1575 			error = EINVAL;
1576 			vfs_opterror(opts,
1577 			    "name cannot be numeric (unless it is the jid)");
1578 			goto done_deref_locked;
1579 		}
1580 		/*
1581 		 * Make sure the name isn't too long for the prison or its
1582 		 * children.
1583 		 */
1584 		pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1585 		onamelen = strlen(pr->pr_name + pnamelen);
1586 		namelen = strlen(namelc);
1587 		if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) {
1588 			error = ENAMETOOLONG;
1589 			goto done_deref_locked;
1590 		}
1591 		FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1592 			if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1593 			    sizeof(pr->pr_name)) {
1594 				error = ENAMETOOLONG;
1595 				goto done_deref_locked;
1596 			}
1597 		}
1598 	}
1599 	if (pr_allow & ~ppr->pr_allow) {
1600 		error = EPERM;
1601 		goto done_deref_locked;
1602 	}
1603 
1604 	/*
1605 	 * Let modules check their parameters.  This requires unlocking and
1606 	 * then re-locking the prison, but this is still a valid state as long
1607 	 * as allprison_lock remains xlocked.
1608 	 */
1609 	mtx_unlock(&pr->pr_mtx);
1610 	error = osd_jail_call(pr, PR_METHOD_CHECK, opts);
1611 	if (error != 0) {
1612 		prison_deref(pr, created
1613 		    ? PD_LIST_XLOCKED
1614 		    : PD_DEREF | PD_LIST_XLOCKED);
1615 		goto done_releroot;
1616 	}
1617 	mtx_lock(&pr->pr_mtx);
1618 
1619 	/* At this point, all valid parameters should have been noted. */
1620 	TAILQ_FOREACH(opt, opts, link) {
1621 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
1622 			error = EINVAL;
1623 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
1624 			goto done_deref_locked;
1625 		}
1626 	}
1627 
1628 	/* Set the parameters of the prison. */
1629 #ifdef INET
1630 	redo_ip4 = 0;
1631 	if (pr_flags & PR_IP4_USER) {
1632 		pr->pr_flags |= PR_IP4;
1633 		free(pr->pr_ip4, M_PRISON);
1634 		pr->pr_ip4s = ip4s;
1635 		pr->pr_ip4 = ip4;
1636 		ip4 = NULL;
1637 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1638 #ifdef VIMAGE
1639 			if (tpr->pr_flags & PR_VNET) {
1640 				descend = 0;
1641 				continue;
1642 			}
1643 #endif
1644 			if (prison_restrict_ip4(tpr, NULL)) {
1645 				redo_ip4 = 1;
1646 				descend = 0;
1647 			}
1648 		}
1649 	}
1650 #endif
1651 #ifdef INET6
1652 	redo_ip6 = 0;
1653 	if (pr_flags & PR_IP6_USER) {
1654 		pr->pr_flags |= PR_IP6;
1655 		free(pr->pr_ip6, M_PRISON);
1656 		pr->pr_ip6s = ip6s;
1657 		pr->pr_ip6 = ip6;
1658 		ip6 = NULL;
1659 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1660 #ifdef VIMAGE
1661 			if (tpr->pr_flags & PR_VNET) {
1662 				descend = 0;
1663 				continue;
1664 			}
1665 #endif
1666 			if (prison_restrict_ip6(tpr, NULL)) {
1667 				redo_ip6 = 1;
1668 				descend = 0;
1669 			}
1670 		}
1671 	}
1672 #endif
1673 	if (gotslevel) {
1674 		pr->pr_securelevel = slevel;
1675 		/* Set all child jails to be at least this level. */
1676 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1677 			if (tpr->pr_securelevel < slevel)
1678 				tpr->pr_securelevel = slevel;
1679 	}
1680 	if (gotchildmax) {
1681 		pr->pr_childmax = childmax;
1682 		/* Set all child jails to under this limit. */
1683 		FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1684 			if (tpr->pr_childmax > childmax - level)
1685 				tpr->pr_childmax = childmax > level
1686 				    ? childmax - level : 0;
1687 	}
1688 	if (gotenforce) {
1689 		pr->pr_enforce_statfs = enforce;
1690 		/* Pass this restriction on to the children. */
1691 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1692 			if (tpr->pr_enforce_statfs < enforce)
1693 				tpr->pr_enforce_statfs = enforce;
1694 	}
1695 	if (gotrsnum) {
1696 		pr->pr_devfs_rsnum = rsnum;
1697 		/* Pass this restriction on to the children. */
1698 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1699 			tpr->pr_devfs_rsnum = rsnum;
1700 	}
1701 	if (namelc != NULL) {
1702 		if (ppr == &prison0)
1703 			strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name));
1704 		else
1705 			snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1706 			    ppr->pr_name, namelc);
1707 		/* Change this component of child names. */
1708 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1709 			bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1710 			    strlen(tpr->pr_name + onamelen) + 1);
1711 			bcopy(pr->pr_name, tpr->pr_name, namelen);
1712 		}
1713 	}
1714 	if (path != NULL) {
1715 		/* Try to keep a real-rooted full pathname. */
1716 		if (fullpath_disabled && path[0] == '/' &&
1717 		    strcmp(mypr->pr_path, "/"))
1718 			snprintf(pr->pr_path, sizeof(pr->pr_path), "%s%s",
1719 			    mypr->pr_path, path);
1720 		else
1721 			strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1722 		pr->pr_root = root;
1723 	}
1724 	if (PR_HOST & ch_flags & ~pr_flags) {
1725 		if (pr->pr_flags & PR_HOST) {
1726 			/*
1727 			 * Copy the parent's host info.  As with pr_ip4 above,
1728 			 * the lack of a lock on the parent is not a problem;
1729 			 * it is always set with allprison_lock at least
1730 			 * shared, and is held exclusively here.
1731 			 */
1732 			strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1733 			    sizeof(pr->pr_hostname));
1734 			strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1735 			    sizeof(pr->pr_domainname));
1736 			strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1737 			    sizeof(pr->pr_hostuuid));
1738 			pr->pr_hostid = pr->pr_parent->pr_hostid;
1739 		}
1740 	} else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1741 		/* Set this prison, and any descendants without PR_HOST. */
1742 		if (host != NULL)
1743 			strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1744 		if (domain != NULL)
1745 			strlcpy(pr->pr_domainname, domain,
1746 			    sizeof(pr->pr_domainname));
1747 		if (uuid != NULL)
1748 			strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1749 		if (gothid)
1750 			pr->pr_hostid = hid;
1751 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1752 			if (tpr->pr_flags & PR_HOST)
1753 				descend = 0;
1754 			else {
1755 				if (host != NULL)
1756 					strlcpy(tpr->pr_hostname,
1757 					    pr->pr_hostname,
1758 					    sizeof(tpr->pr_hostname));
1759 				if (domain != NULL)
1760 					strlcpy(tpr->pr_domainname,
1761 					    pr->pr_domainname,
1762 					    sizeof(tpr->pr_domainname));
1763 				if (uuid != NULL)
1764 					strlcpy(tpr->pr_hostuuid,
1765 					    pr->pr_hostuuid,
1766 					    sizeof(tpr->pr_hostuuid));
1767 				if (gothid)
1768 					tpr->pr_hostid = hid;
1769 			}
1770 		}
1771 	}
1772 	if ((tallow = ch_allow & ~pr_allow)) {
1773 		/* Clear allow bits in all children. */
1774 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1775 			tpr->pr_allow &= ~tallow;
1776 	}
1777 	pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1778 	/*
1779 	 * Persistent prisons get an extra reference, and prisons losing their
1780 	 * persist flag lose that reference.  Only do this for existing prisons
1781 	 * for now, so new ones will remain unseen until after the module
1782 	 * handlers have completed.
1783 	 */
1784 	born = pr->pr_uref == 0;
1785 	if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) {
1786 		if (pr_flags & PR_PERSIST) {
1787 			pr->pr_ref++;
1788 			pr->pr_uref++;
1789 		} else {
1790 			pr->pr_ref--;
1791 			pr->pr_uref--;
1792 		}
1793 	}
1794 	pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1795 	mtx_unlock(&pr->pr_mtx);
1796 
1797 #ifdef RACCT
1798 	if (racct_enable && created)
1799 		prison_racct_attach(pr);
1800 #endif
1801 
1802 	/* Locks may have prevented a complete restriction of child IP
1803 	 * addresses.  If so, allocate some more memory and try again.
1804 	 */
1805 #ifdef INET
1806 	while (redo_ip4) {
1807 		ip4s = pr->pr_ip4s;
1808 		ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1809 		mtx_lock(&pr->pr_mtx);
1810 		redo_ip4 = 0;
1811 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1812 #ifdef VIMAGE
1813 			if (tpr->pr_flags & PR_VNET) {
1814 				descend = 0;
1815 				continue;
1816 			}
1817 #endif
1818 			if (prison_restrict_ip4(tpr, ip4)) {
1819 				if (ip4 != NULL)
1820 					ip4 = NULL;
1821 				else
1822 					redo_ip4 = 1;
1823 			}
1824 		}
1825 		mtx_unlock(&pr->pr_mtx);
1826 	}
1827 #endif
1828 #ifdef INET6
1829 	while (redo_ip6) {
1830 		ip6s = pr->pr_ip6s;
1831 		ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1832 		mtx_lock(&pr->pr_mtx);
1833 		redo_ip6 = 0;
1834 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1835 #ifdef VIMAGE
1836 			if (tpr->pr_flags & PR_VNET) {
1837 				descend = 0;
1838 				continue;
1839 			}
1840 #endif
1841 			if (prison_restrict_ip6(tpr, ip6)) {
1842 				if (ip6 != NULL)
1843 					ip6 = NULL;
1844 				else
1845 					redo_ip6 = 1;
1846 			}
1847 		}
1848 		mtx_unlock(&pr->pr_mtx);
1849 	}
1850 #endif
1851 
1852 	/* Let the modules do their work. */
1853 	sx_downgrade(&allprison_lock);
1854 	if (born) {
1855 		error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1856 		if (error) {
1857 			(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1858 			prison_deref(pr, created
1859 			    ? PD_LIST_SLOCKED
1860 			    : PD_DEREF | PD_LIST_SLOCKED);
1861 			goto done_errmsg;
1862 		}
1863 	}
1864 	error = osd_jail_call(pr, PR_METHOD_SET, opts);
1865 	if (error) {
1866 		if (born)
1867 			(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
1868 		prison_deref(pr, created
1869 		    ? PD_LIST_SLOCKED
1870 		    : PD_DEREF | PD_LIST_SLOCKED);
1871 		goto done_errmsg;
1872 	}
1873 
1874 	/* Attach this process to the prison if requested. */
1875 	if (flags & JAIL_ATTACH) {
1876 		mtx_lock(&pr->pr_mtx);
1877 		error = do_jail_attach(td, pr);
1878 		if (error) {
1879 			vfs_opterror(opts, "attach failed");
1880 			if (!created)
1881 				prison_deref(pr, PD_DEREF);
1882 			goto done_errmsg;
1883 		}
1884 	}
1885 
1886 #ifdef RACCT
1887 	if (racct_enable && !created) {
1888 		if (!(flags & JAIL_ATTACH))
1889 			sx_sunlock(&allprison_lock);
1890 		prison_racct_modify(pr);
1891 		if (!(flags & JAIL_ATTACH))
1892 			sx_slock(&allprison_lock);
1893 	}
1894 #endif
1895 
1896 	td->td_retval[0] = pr->pr_id;
1897 
1898 	/*
1899 	 * Now that it is all there, drop the temporary reference from existing
1900 	 * prisons.  Or add a reference to newly created persistent prisons
1901 	 * (which was not done earlier so that the prison would not be publicly
1902 	 * visible).
1903 	 */
1904 	if (!created) {
1905 		prison_deref(pr, (flags & JAIL_ATTACH)
1906 		    ? PD_DEREF
1907 		    : PD_DEREF | PD_LIST_SLOCKED);
1908 	} else {
1909 		if (pr_flags & PR_PERSIST) {
1910 			mtx_lock(&pr->pr_mtx);
1911 			pr->pr_ref++;
1912 			pr->pr_uref++;
1913 			mtx_unlock(&pr->pr_mtx);
1914 		}
1915 		if (!(flags & JAIL_ATTACH))
1916 			sx_sunlock(&allprison_lock);
1917 	}
1918 
1919 	goto done_free;
1920 
1921  done_deref_locked:
1922 	prison_deref(pr, created
1923 	    ? PD_LOCKED | PD_LIST_XLOCKED
1924 	    : PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
1925 	goto done_releroot;
1926  done_unlock_list:
1927 	sx_xunlock(&allprison_lock);
1928  done_releroot:
1929 	if (root != NULL)
1930 		vrele(root);
1931  done_errmsg:
1932 	if (error) {
1933 		if (vfs_getopt(opts, "errmsg", (void **)&errmsg,
1934 		    &errmsg_len) == 0 && errmsg_len > 0) {
1935 			errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1936 			if (optuio->uio_segflg == UIO_SYSSPACE)
1937 				bcopy(errmsg,
1938 				    optuio->uio_iov[errmsg_pos].iov_base,
1939 				    errmsg_len);
1940 			else
1941 				copyout(errmsg,
1942 				    optuio->uio_iov[errmsg_pos].iov_base,
1943 				    errmsg_len);
1944 		}
1945 	}
1946  done_free:
1947 #ifdef INET
1948 	free(ip4, M_PRISON);
1949 #endif
1950 #ifdef INET6
1951 	free(ip6, M_PRISON);
1952 #endif
1953 	if (g_path != NULL)
1954 		free(g_path, M_TEMP);
1955 	vfs_freeopts(opts);
1956 	return (error);
1957 }
1958 
1959 
1960 /*
1961  * struct jail_get_args {
1962  *	struct iovec *iovp;
1963  *	unsigned int iovcnt;
1964  *	int flags;
1965  * };
1966  */
1967 int
1968 sys_jail_get(struct thread *td, struct jail_get_args *uap)
1969 {
1970 	struct uio *auio;
1971 	int error;
1972 
1973 	/* Check that we have an even number of iovecs. */
1974 	if (uap->iovcnt & 1)
1975 		return (EINVAL);
1976 
1977 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
1978 	if (error)
1979 		return (error);
1980 	error = kern_jail_get(td, auio, uap->flags);
1981 	if (error == 0)
1982 		error = copyout(auio->uio_iov, uap->iovp,
1983 		    uap->iovcnt * sizeof (struct iovec));
1984 	free(auio, M_IOV);
1985 	return (error);
1986 }
1987 
1988 int
1989 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
1990 {
1991 	struct prison *pr, *mypr;
1992 	struct vfsopt *opt;
1993 	struct vfsoptlist *opts;
1994 	char *errmsg, *name;
1995 	int error, errmsg_len, errmsg_pos, fi, i, jid, len, locked, pos;
1996 
1997 	if (flags & ~JAIL_GET_MASK)
1998 		return (EINVAL);
1999 
2000 	/* Get the parameter list. */
2001 	error = vfs_buildopts(optuio, &opts);
2002 	if (error)
2003 		return (error);
2004 	errmsg_pos = vfs_getopt_pos(opts, "errmsg");
2005 	mypr = td->td_ucred->cr_prison;
2006 
2007 	/*
2008 	 * Find the prison specified by one of: lastjid, jid, name.
2009 	 */
2010 	sx_slock(&allprison_lock);
2011 	error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
2012 	if (error == 0) {
2013 		TAILQ_FOREACH(pr, &allprison, pr_list) {
2014 			if (pr->pr_id > jid && prison_ischild(mypr, pr)) {
2015 				mtx_lock(&pr->pr_mtx);
2016 				if (pr->pr_ref > 0 &&
2017 				    (pr->pr_uref > 0 || (flags & JAIL_DYING)))
2018 					break;
2019 				mtx_unlock(&pr->pr_mtx);
2020 			}
2021 		}
2022 		if (pr != NULL)
2023 			goto found_prison;
2024 		error = ENOENT;
2025 		vfs_opterror(opts, "no jail after %d", jid);
2026 		goto done_unlock_list;
2027 	} else if (error != ENOENT)
2028 		goto done_unlock_list;
2029 
2030 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
2031 	if (error == 0) {
2032 		if (jid != 0) {
2033 			pr = prison_find_child(mypr, jid);
2034 			if (pr != NULL) {
2035 				if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
2036 					mtx_unlock(&pr->pr_mtx);
2037 					error = ENOENT;
2038 					vfs_opterror(opts, "jail %d is dying",
2039 					    jid);
2040 					goto done_unlock_list;
2041 				}
2042 				goto found_prison;
2043 			}
2044 			error = ENOENT;
2045 			vfs_opterror(opts, "jail %d not found", jid);
2046 			goto done_unlock_list;
2047 		}
2048 	} else if (error != ENOENT)
2049 		goto done_unlock_list;
2050 
2051 	error = vfs_getopt(opts, "name", (void **)&name, &len);
2052 	if (error == 0) {
2053 		if (len == 0 || name[len - 1] != '\0') {
2054 			error = EINVAL;
2055 			goto done_unlock_list;
2056 		}
2057 		pr = prison_find_name(mypr, name);
2058 		if (pr != NULL) {
2059 			if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
2060 				mtx_unlock(&pr->pr_mtx);
2061 				error = ENOENT;
2062 				vfs_opterror(opts, "jail \"%s\" is dying",
2063 				    name);
2064 				goto done_unlock_list;
2065 			}
2066 			goto found_prison;
2067 		}
2068 		error = ENOENT;
2069 		vfs_opterror(opts, "jail \"%s\" not found", name);
2070 		goto done_unlock_list;
2071 	} else if (error != ENOENT)
2072 		goto done_unlock_list;
2073 
2074 	vfs_opterror(opts, "no jail specified");
2075 	error = ENOENT;
2076 	goto done_unlock_list;
2077 
2078  found_prison:
2079 	/* Get the parameters of the prison. */
2080 	pr->pr_ref++;
2081 	locked = PD_LOCKED;
2082 	td->td_retval[0] = pr->pr_id;
2083 	error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
2084 	if (error != 0 && error != ENOENT)
2085 		goto done_deref;
2086 	i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
2087 	error = vfs_setopt(opts, "parent", &i, sizeof(i));
2088 	if (error != 0 && error != ENOENT)
2089 		goto done_deref;
2090 	error = vfs_setopts(opts, "name", prison_name(mypr, pr));
2091 	if (error != 0 && error != ENOENT)
2092 		goto done_deref;
2093 	error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
2094 	    sizeof(pr->pr_cpuset->cs_id));
2095 	if (error != 0 && error != ENOENT)
2096 		goto done_deref;
2097 	error = vfs_setopts(opts, "path", prison_path(mypr, pr));
2098 	if (error != 0 && error != ENOENT)
2099 		goto done_deref;
2100 #ifdef INET
2101 	error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
2102 	    pr->pr_ip4s * sizeof(*pr->pr_ip4));
2103 	if (error != 0 && error != ENOENT)
2104 		goto done_deref;
2105 #endif
2106 #ifdef INET6
2107 	error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
2108 	    pr->pr_ip6s * sizeof(*pr->pr_ip6));
2109 	if (error != 0 && error != ENOENT)
2110 		goto done_deref;
2111 #endif
2112 	error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
2113 	    sizeof(pr->pr_securelevel));
2114 	if (error != 0 && error != ENOENT)
2115 		goto done_deref;
2116 	error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
2117 	    sizeof(pr->pr_childcount));
2118 	if (error != 0 && error != ENOENT)
2119 		goto done_deref;
2120 	error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
2121 	    sizeof(pr->pr_childmax));
2122 	if (error != 0 && error != ENOENT)
2123 		goto done_deref;
2124 	error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
2125 	if (error != 0 && error != ENOENT)
2126 		goto done_deref;
2127 	error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
2128 	if (error != 0 && error != ENOENT)
2129 		goto done_deref;
2130 	error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
2131 	if (error != 0 && error != ENOENT)
2132 		goto done_deref;
2133 #ifdef COMPAT_FREEBSD32
2134 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
2135 		uint32_t hid32 = pr->pr_hostid;
2136 
2137 		error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
2138 	} else
2139 #endif
2140 	error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
2141 	    sizeof(pr->pr_hostid));
2142 	if (error != 0 && error != ENOENT)
2143 		goto done_deref;
2144 	error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
2145 	    sizeof(pr->pr_enforce_statfs));
2146 	if (error != 0 && error != ENOENT)
2147 		goto done_deref;
2148 	error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum,
2149 	    sizeof(pr->pr_devfs_rsnum));
2150 	if (error != 0 && error != ENOENT)
2151 		goto done_deref;
2152 	for (fi = 0; fi < nitems(pr_flag_names); fi++) {
2153 		if (pr_flag_names[fi] == NULL)
2154 			continue;
2155 		i = (pr->pr_flags & (1 << fi)) ? 1 : 0;
2156 		error = vfs_setopt(opts, pr_flag_names[fi], &i, sizeof(i));
2157 		if (error != 0 && error != ENOENT)
2158 			goto done_deref;
2159 		i = !i;
2160 		error = vfs_setopt(opts, pr_flag_nonames[fi], &i, sizeof(i));
2161 		if (error != 0 && error != ENOENT)
2162 			goto done_deref;
2163 	}
2164 	for (fi = 0; fi < nitems(pr_flag_jailsys); fi++) {
2165 		i = pr->pr_flags &
2166 		    (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
2167 		i = pr_flag_jailsys[fi].disable &&
2168 		      (i == pr_flag_jailsys[fi].disable) ? JAIL_SYS_DISABLE
2169 		    : (i == pr_flag_jailsys[fi].new) ? JAIL_SYS_NEW
2170 		    : JAIL_SYS_INHERIT;
2171 		error =
2172 		    vfs_setopt(opts, pr_flag_jailsys[fi].name, &i, sizeof(i));
2173 		if (error != 0 && error != ENOENT)
2174 			goto done_deref;
2175 	}
2176 	for (fi = 0; fi < nitems(pr_allow_names); fi++) {
2177 		if (pr_allow_names[fi] == NULL)
2178 			continue;
2179 		i = (pr->pr_allow & (1 << fi)) ? 1 : 0;
2180 		error = vfs_setopt(opts, pr_allow_names[fi], &i, sizeof(i));
2181 		if (error != 0 && error != ENOENT)
2182 			goto done_deref;
2183 		i = !i;
2184 		error = vfs_setopt(opts, pr_allow_nonames[fi], &i, sizeof(i));
2185 		if (error != 0 && error != ENOENT)
2186 			goto done_deref;
2187 	}
2188 	i = (pr->pr_uref == 0);
2189 	error = vfs_setopt(opts, "dying", &i, sizeof(i));
2190 	if (error != 0 && error != ENOENT)
2191 		goto done_deref;
2192 	i = !i;
2193 	error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2194 	if (error != 0 && error != ENOENT)
2195 		goto done_deref;
2196 	error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate,
2197 	    sizeof(pr->pr_osreldate));
2198 	if (error != 0 && error != ENOENT)
2199 		goto done_deref;
2200 	error = vfs_setopts(opts, "osrelease", pr->pr_osrelease);
2201 	if (error != 0 && error != ENOENT)
2202 		goto done_deref;
2203 
2204 	/* Get the module parameters. */
2205 	mtx_unlock(&pr->pr_mtx);
2206 	locked = 0;
2207 	error = osd_jail_call(pr, PR_METHOD_GET, opts);
2208 	if (error)
2209 		goto done_deref;
2210 	prison_deref(pr, PD_DEREF | PD_LIST_SLOCKED);
2211 
2212 	/* By now, all parameters should have been noted. */
2213 	TAILQ_FOREACH(opt, opts, link) {
2214 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
2215 			error = EINVAL;
2216 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
2217 			goto done_errmsg;
2218 		}
2219 	}
2220 
2221 	/* Write the fetched parameters back to userspace. */
2222 	error = 0;
2223 	TAILQ_FOREACH(opt, opts, link) {
2224 		if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2225 			pos = 2 * opt->pos + 1;
2226 			optuio->uio_iov[pos].iov_len = opt->len;
2227 			if (opt->value != NULL) {
2228 				if (optuio->uio_segflg == UIO_SYSSPACE) {
2229 					bcopy(opt->value,
2230 					    optuio->uio_iov[pos].iov_base,
2231 					    opt->len);
2232 				} else {
2233 					error = copyout(opt->value,
2234 					    optuio->uio_iov[pos].iov_base,
2235 					    opt->len);
2236 					if (error)
2237 						break;
2238 				}
2239 			}
2240 		}
2241 	}
2242 	goto done_errmsg;
2243 
2244  done_deref:
2245 	prison_deref(pr, locked | PD_DEREF | PD_LIST_SLOCKED);
2246 	goto done_errmsg;
2247 
2248  done_unlock_list:
2249 	sx_sunlock(&allprison_lock);
2250  done_errmsg:
2251 	if (error && errmsg_pos >= 0) {
2252 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2253 		errmsg_pos = 2 * errmsg_pos + 1;
2254 		if (errmsg_len > 0) {
2255 			if (optuio->uio_segflg == UIO_SYSSPACE)
2256 				bcopy(errmsg,
2257 				    optuio->uio_iov[errmsg_pos].iov_base,
2258 				    errmsg_len);
2259 			else
2260 				copyout(errmsg,
2261 				    optuio->uio_iov[errmsg_pos].iov_base,
2262 				    errmsg_len);
2263 		}
2264 	}
2265 	vfs_freeopts(opts);
2266 	return (error);
2267 }
2268 
2269 
2270 /*
2271  * struct jail_remove_args {
2272  *	int jid;
2273  * };
2274  */
2275 int
2276 sys_jail_remove(struct thread *td, struct jail_remove_args *uap)
2277 {
2278 	struct prison *pr, *cpr, *lpr, *tpr;
2279 	int descend, error;
2280 
2281 	error = priv_check(td, PRIV_JAIL_REMOVE);
2282 	if (error)
2283 		return (error);
2284 
2285 	sx_xlock(&allprison_lock);
2286 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2287 	if (pr == NULL) {
2288 		sx_xunlock(&allprison_lock);
2289 		return (EINVAL);
2290 	}
2291 
2292 	/* Remove all descendants of this prison, then remove this prison. */
2293 	pr->pr_ref++;
2294 	if (!LIST_EMPTY(&pr->pr_children)) {
2295 		mtx_unlock(&pr->pr_mtx);
2296 		lpr = NULL;
2297 		FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
2298 			mtx_lock(&cpr->pr_mtx);
2299 			if (cpr->pr_ref > 0) {
2300 				tpr = cpr;
2301 				cpr->pr_ref++;
2302 			} else {
2303 				/* Already removed - do not do it again. */
2304 				tpr = NULL;
2305 			}
2306 			mtx_unlock(&cpr->pr_mtx);
2307 			if (lpr != NULL) {
2308 				mtx_lock(&lpr->pr_mtx);
2309 				prison_remove_one(lpr);
2310 				sx_xlock(&allprison_lock);
2311 			}
2312 			lpr = tpr;
2313 		}
2314 		if (lpr != NULL) {
2315 			mtx_lock(&lpr->pr_mtx);
2316 			prison_remove_one(lpr);
2317 			sx_xlock(&allprison_lock);
2318 		}
2319 		mtx_lock(&pr->pr_mtx);
2320 	}
2321 	prison_remove_one(pr);
2322 	return (0);
2323 }
2324 
2325 static void
2326 prison_remove_one(struct prison *pr)
2327 {
2328 	struct proc *p;
2329 	int deuref;
2330 
2331 	/* If the prison was persistent, it is not anymore. */
2332 	deuref = 0;
2333 	if (pr->pr_flags & PR_PERSIST) {
2334 		pr->pr_ref--;
2335 		deuref = PD_DEUREF;
2336 		pr->pr_flags &= ~PR_PERSIST;
2337 	}
2338 
2339 	/*
2340 	 * jail_remove added a reference.  If that's the only one, remove
2341 	 * the prison now.
2342 	 */
2343 	KASSERT(pr->pr_ref > 0,
2344 	    ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id));
2345 	if (pr->pr_ref == 1) {
2346 		prison_deref(pr,
2347 		    deuref | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
2348 		return;
2349 	}
2350 
2351 	mtx_unlock(&pr->pr_mtx);
2352 	sx_xunlock(&allprison_lock);
2353 	/*
2354 	 * Kill all processes unfortunate enough to be attached to this prison.
2355 	 */
2356 	sx_slock(&allproc_lock);
2357 	LIST_FOREACH(p, &allproc, p_list) {
2358 		PROC_LOCK(p);
2359 		if (p->p_state != PRS_NEW && p->p_ucred &&
2360 		    p->p_ucred->cr_prison == pr)
2361 			kern_psignal(p, SIGKILL);
2362 		PROC_UNLOCK(p);
2363 	}
2364 	sx_sunlock(&allproc_lock);
2365 	/* Remove the temporary reference added by jail_remove. */
2366 	prison_deref(pr, deuref | PD_DEREF);
2367 }
2368 
2369 
2370 /*
2371  * struct jail_attach_args {
2372  *	int jid;
2373  * };
2374  */
2375 int
2376 sys_jail_attach(struct thread *td, struct jail_attach_args *uap)
2377 {
2378 	struct prison *pr;
2379 	int error;
2380 
2381 	error = priv_check(td, PRIV_JAIL_ATTACH);
2382 	if (error)
2383 		return (error);
2384 
2385 	/*
2386 	 * Start with exclusive hold on allprison_lock to ensure that a possible
2387 	 * PR_METHOD_REMOVE call isn't concurrent with jail_set or jail_remove.
2388 	 * But then immediately downgrade it since we don't need to stop
2389 	 * readers.
2390 	 */
2391 	sx_xlock(&allprison_lock);
2392 	sx_downgrade(&allprison_lock);
2393 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2394 	if (pr == NULL) {
2395 		sx_sunlock(&allprison_lock);
2396 		return (EINVAL);
2397 	}
2398 
2399 	/*
2400 	 * Do not allow a process to attach to a prison that is not
2401 	 * considered to be "alive".
2402 	 */
2403 	if (pr->pr_uref == 0) {
2404 		mtx_unlock(&pr->pr_mtx);
2405 		sx_sunlock(&allprison_lock);
2406 		return (EINVAL);
2407 	}
2408 
2409 	return (do_jail_attach(td, pr));
2410 }
2411 
2412 static int
2413 do_jail_attach(struct thread *td, struct prison *pr)
2414 {
2415 	struct proc *p;
2416 	struct ucred *newcred, *oldcred;
2417 	int error;
2418 
2419 	/*
2420 	 * XXX: Note that there is a slight race here if two threads
2421 	 * in the same privileged process attempt to attach to two
2422 	 * different jails at the same time.  It is important for
2423 	 * user processes not to do this, or they might end up with
2424 	 * a process root from one prison, but attached to the jail
2425 	 * of another.
2426 	 */
2427 	pr->pr_ref++;
2428 	pr->pr_uref++;
2429 	mtx_unlock(&pr->pr_mtx);
2430 
2431 	/* Let modules do whatever they need to prepare for attaching. */
2432 	error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2433 	if (error) {
2434 		prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2435 		return (error);
2436 	}
2437 	sx_sunlock(&allprison_lock);
2438 
2439 	/*
2440 	 * Reparent the newly attached process to this jail.
2441 	 */
2442 	p = td->td_proc;
2443 	error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2444 	if (error)
2445 		goto e_revert_osd;
2446 
2447 	vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2448 	if ((error = change_dir(pr->pr_root, td)) != 0)
2449 		goto e_unlock;
2450 #ifdef MAC
2451 	if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2452 		goto e_unlock;
2453 #endif
2454 	VOP_UNLOCK(pr->pr_root, 0);
2455 	if ((error = pwd_chroot(td, pr->pr_root)))
2456 		goto e_revert_osd;
2457 
2458 	newcred = crget();
2459 	PROC_LOCK(p);
2460 	oldcred = crcopysafe(p, newcred);
2461 	newcred->cr_prison = pr;
2462 	proc_set_cred(p, newcred);
2463 	setsugid(p);
2464 	PROC_UNLOCK(p);
2465 #ifdef RACCT
2466 	racct_proc_ucred_changed(p, oldcred, newcred);
2467 #endif
2468 	prison_deref(oldcred->cr_prison, PD_DEREF | PD_DEUREF);
2469 	crfree(oldcred);
2470 	return (0);
2471 
2472  e_unlock:
2473 	VOP_UNLOCK(pr->pr_root, 0);
2474  e_revert_osd:
2475 	/* Tell modules this thread is still in its old jail after all. */
2476 	(void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td);
2477 	prison_deref(pr, PD_DEREF | PD_DEUREF);
2478 	return (error);
2479 }
2480 
2481 
2482 /*
2483  * Returns a locked prison instance, or NULL on failure.
2484  */
2485 struct prison *
2486 prison_find(int prid)
2487 {
2488 	struct prison *pr;
2489 
2490 	sx_assert(&allprison_lock, SX_LOCKED);
2491 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2492 		if (pr->pr_id == prid) {
2493 			mtx_lock(&pr->pr_mtx);
2494 			if (pr->pr_ref > 0)
2495 				return (pr);
2496 			mtx_unlock(&pr->pr_mtx);
2497 		}
2498 	}
2499 	return (NULL);
2500 }
2501 
2502 /*
2503  * Find a prison that is a descendant of mypr.  Returns a locked prison or NULL.
2504  */
2505 struct prison *
2506 prison_find_child(struct prison *mypr, int prid)
2507 {
2508 	struct prison *pr;
2509 	int descend;
2510 
2511 	sx_assert(&allprison_lock, SX_LOCKED);
2512 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2513 		if (pr->pr_id == prid) {
2514 			mtx_lock(&pr->pr_mtx);
2515 			if (pr->pr_ref > 0)
2516 				return (pr);
2517 			mtx_unlock(&pr->pr_mtx);
2518 		}
2519 	}
2520 	return (NULL);
2521 }
2522 
2523 /*
2524  * Look for the name relative to mypr.  Returns a locked prison or NULL.
2525  */
2526 struct prison *
2527 prison_find_name(struct prison *mypr, const char *name)
2528 {
2529 	struct prison *pr, *deadpr;
2530 	size_t mylen;
2531 	int descend;
2532 
2533 	sx_assert(&allprison_lock, SX_LOCKED);
2534 	mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2535  again:
2536 	deadpr = NULL;
2537 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2538 		if (!strcmp(pr->pr_name + mylen, name)) {
2539 			mtx_lock(&pr->pr_mtx);
2540 			if (pr->pr_ref > 0) {
2541 				if (pr->pr_uref > 0)
2542 					return (pr);
2543 				deadpr = pr;
2544 			}
2545 			mtx_unlock(&pr->pr_mtx);
2546 		}
2547 	}
2548 	/* There was no valid prison - perhaps there was a dying one. */
2549 	if (deadpr != NULL) {
2550 		mtx_lock(&deadpr->pr_mtx);
2551 		if (deadpr->pr_ref == 0) {
2552 			mtx_unlock(&deadpr->pr_mtx);
2553 			goto again;
2554 		}
2555 	}
2556 	return (deadpr);
2557 }
2558 
2559 /*
2560  * See if a prison has the specific flag set.
2561  */
2562 int
2563 prison_flag(struct ucred *cred, unsigned flag)
2564 {
2565 
2566 	/* This is an atomic read, so no locking is necessary. */
2567 	return (cred->cr_prison->pr_flags & flag);
2568 }
2569 
2570 int
2571 prison_allow(struct ucred *cred, unsigned flag)
2572 {
2573 
2574 	/* This is an atomic read, so no locking is necessary. */
2575 	return (cred->cr_prison->pr_allow & flag);
2576 }
2577 
2578 /*
2579  * Remove a prison reference.  If that was the last reference, remove the
2580  * prison itself - but not in this context in case there are locks held.
2581  */
2582 void
2583 prison_free_locked(struct prison *pr)
2584 {
2585 	int ref;
2586 
2587 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2588 	ref = --pr->pr_ref;
2589 	mtx_unlock(&pr->pr_mtx);
2590 	if (ref == 0)
2591 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2592 }
2593 
2594 void
2595 prison_free(struct prison *pr)
2596 {
2597 
2598 	mtx_lock(&pr->pr_mtx);
2599 	prison_free_locked(pr);
2600 }
2601 
2602 /*
2603  * Complete a call to either prison_free or prison_proc_free.
2604  */
2605 static void
2606 prison_complete(void *context, int pending)
2607 {
2608 	struct prison *pr = context;
2609 
2610 	sx_xlock(&allprison_lock);
2611 	mtx_lock(&pr->pr_mtx);
2612 	prison_deref(pr, pr->pr_uref
2613 	    ? PD_DEREF | PD_DEUREF | PD_LOCKED | PD_LIST_XLOCKED
2614 	    : PD_LOCKED | PD_LIST_XLOCKED);
2615 }
2616 
2617 /*
2618  * Remove a prison reference (usually).  This internal version assumes no
2619  * mutexes are held, except perhaps the prison itself.  If there are no more
2620  * references, release and delist the prison.  On completion, the prison lock
2621  * and the allprison lock are both unlocked.
2622  */
2623 static void
2624 prison_deref(struct prison *pr, int flags)
2625 {
2626 	struct prison *ppr, *tpr;
2627 	int ref, lasturef;
2628 
2629 	if (!(flags & PD_LOCKED))
2630 		mtx_lock(&pr->pr_mtx);
2631 	for (;;) {
2632 		if (flags & PD_DEUREF) {
2633 			KASSERT(pr->pr_uref > 0,
2634 			    ("prison_deref PD_DEUREF on a dead prison (jid=%d)",
2635 			     pr->pr_id));
2636 			pr->pr_uref--;
2637 			lasturef = pr->pr_uref == 0;
2638 			if (lasturef)
2639 				pr->pr_ref++;
2640 			KASSERT(prison0.pr_uref != 0, ("prison0 pr_uref=0"));
2641 		} else
2642 			lasturef = 0;
2643 		if (flags & PD_DEREF) {
2644 			KASSERT(pr->pr_ref > 0,
2645 			    ("prison_deref PD_DEREF on a dead prison (jid=%d)",
2646 			     pr->pr_id));
2647 			pr->pr_ref--;
2648 		}
2649 		ref = pr->pr_ref;
2650 		mtx_unlock(&pr->pr_mtx);
2651 
2652 		/*
2653 		 * Tell the modules if the last user reference was removed
2654 		 * (even it sticks around in dying state).
2655 		 */
2656 		if (lasturef) {
2657 			if (!(flags & (PD_LIST_SLOCKED | PD_LIST_XLOCKED))) {
2658 				sx_xlock(&allprison_lock);
2659 				flags |= PD_LIST_XLOCKED;
2660 			}
2661 			(void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL);
2662 			mtx_lock(&pr->pr_mtx);
2663 			ref = --pr->pr_ref;
2664 			mtx_unlock(&pr->pr_mtx);
2665 		}
2666 
2667 		/* If the prison still has references, nothing else to do. */
2668 		if (ref > 0) {
2669 			if (flags & PD_LIST_SLOCKED)
2670 				sx_sunlock(&allprison_lock);
2671 			else if (flags & PD_LIST_XLOCKED)
2672 				sx_xunlock(&allprison_lock);
2673 			return;
2674 		}
2675 
2676 		if (flags & PD_LIST_SLOCKED) {
2677 			if (!sx_try_upgrade(&allprison_lock)) {
2678 				sx_sunlock(&allprison_lock);
2679 				sx_xlock(&allprison_lock);
2680 			}
2681 		} else if (!(flags & PD_LIST_XLOCKED))
2682 			sx_xlock(&allprison_lock);
2683 
2684 		TAILQ_REMOVE(&allprison, pr, pr_list);
2685 		LIST_REMOVE(pr, pr_sibling);
2686 		ppr = pr->pr_parent;
2687 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
2688 			tpr->pr_childcount--;
2689 		sx_xunlock(&allprison_lock);
2690 
2691 #ifdef VIMAGE
2692 		if (pr->pr_vnet != ppr->pr_vnet)
2693 			vnet_destroy(pr->pr_vnet);
2694 #endif
2695 		if (pr->pr_root != NULL)
2696 			vrele(pr->pr_root);
2697 		mtx_destroy(&pr->pr_mtx);
2698 #ifdef INET
2699 		free(pr->pr_ip4, M_PRISON);
2700 #endif
2701 #ifdef INET6
2702 		free(pr->pr_ip6, M_PRISON);
2703 #endif
2704 		if (pr->pr_cpuset != NULL)
2705 			cpuset_rel(pr->pr_cpuset);
2706 		osd_jail_exit(pr);
2707 #ifdef RACCT
2708 		if (racct_enable)
2709 			prison_racct_detach(pr);
2710 #endif
2711 		free(pr, M_PRISON);
2712 
2713 		/* Removing a prison frees a reference on its parent. */
2714 		pr = ppr;
2715 		mtx_lock(&pr->pr_mtx);
2716 		flags = PD_DEREF | PD_DEUREF;
2717 	}
2718 }
2719 
2720 void
2721 prison_hold_locked(struct prison *pr)
2722 {
2723 
2724 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2725 	KASSERT(pr->pr_ref > 0,
2726 	    ("Trying to hold dead prison (jid=%d).", pr->pr_id));
2727 	pr->pr_ref++;
2728 }
2729 
2730 void
2731 prison_hold(struct prison *pr)
2732 {
2733 
2734 	mtx_lock(&pr->pr_mtx);
2735 	prison_hold_locked(pr);
2736 	mtx_unlock(&pr->pr_mtx);
2737 }
2738 
2739 void
2740 prison_proc_hold(struct prison *pr)
2741 {
2742 
2743 	mtx_lock(&pr->pr_mtx);
2744 	KASSERT(pr->pr_uref > 0,
2745 	    ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2746 	pr->pr_uref++;
2747 	mtx_unlock(&pr->pr_mtx);
2748 }
2749 
2750 void
2751 prison_proc_free(struct prison *pr)
2752 {
2753 
2754 	mtx_lock(&pr->pr_mtx);
2755 	KASSERT(pr->pr_uref > 0,
2756 	    ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2757 	if (pr->pr_uref > 1)
2758 		pr->pr_uref--;
2759 	else {
2760 		/*
2761 		 * Don't remove the last user reference in this context, which
2762 		 * is expected to be a process that is not only locked, but
2763 		 * also half dead.
2764 		 */
2765 		pr->pr_ref++;
2766 		mtx_unlock(&pr->pr_mtx);
2767 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2768 		return;
2769 	}
2770 	mtx_unlock(&pr->pr_mtx);
2771 }
2772 
2773 
2774 #ifdef INET
2775 /*
2776  * Restrict a prison's IP address list with its parent's, possibly replacing
2777  * it.  Return true if the replacement buffer was used (or would have been).
2778  */
2779 static int
2780 prison_restrict_ip4(struct prison *pr, struct in_addr *newip4)
2781 {
2782 	int ii, ij, used;
2783 	struct prison *ppr;
2784 
2785 	ppr = pr->pr_parent;
2786 	if (!(pr->pr_flags & PR_IP4_USER)) {
2787 		/* This has no user settings, so just copy the parent's list. */
2788 		if (pr->pr_ip4s < ppr->pr_ip4s) {
2789 			/*
2790 			 * There's no room for the parent's list.  Use the
2791 			 * new list buffer, which is assumed to be big enough
2792 			 * (if it was passed).  If there's no buffer, try to
2793 			 * allocate one.
2794 			 */
2795 			used = 1;
2796 			if (newip4 == NULL) {
2797 				newip4 = malloc(ppr->pr_ip4s * sizeof(*newip4),
2798 				    M_PRISON, M_NOWAIT);
2799 				if (newip4 != NULL)
2800 					used = 0;
2801 			}
2802 			if (newip4 != NULL) {
2803 				bcopy(ppr->pr_ip4, newip4,
2804 				    ppr->pr_ip4s * sizeof(*newip4));
2805 				free(pr->pr_ip4, M_PRISON);
2806 				pr->pr_ip4 = newip4;
2807 				pr->pr_ip4s = ppr->pr_ip4s;
2808 			}
2809 			return (used);
2810 		}
2811 		pr->pr_ip4s = ppr->pr_ip4s;
2812 		if (pr->pr_ip4s > 0)
2813 			bcopy(ppr->pr_ip4, pr->pr_ip4,
2814 			    pr->pr_ip4s * sizeof(*newip4));
2815 		else if (pr->pr_ip4 != NULL) {
2816 			free(pr->pr_ip4, M_PRISON);
2817 			pr->pr_ip4 = NULL;
2818 		}
2819 	} else if (pr->pr_ip4s > 0) {
2820 		/* Remove addresses that aren't in the parent. */
2821 		for (ij = 0; ij < ppr->pr_ip4s; ij++)
2822 			if (pr->pr_ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
2823 				break;
2824 		if (ij < ppr->pr_ip4s)
2825 			ii = 1;
2826 		else {
2827 			bcopy(pr->pr_ip4 + 1, pr->pr_ip4,
2828 			    --pr->pr_ip4s * sizeof(*pr->pr_ip4));
2829 			ii = 0;
2830 		}
2831 		for (ij = 1; ii < pr->pr_ip4s; ) {
2832 			if (pr->pr_ip4[ii].s_addr == ppr->pr_ip4[0].s_addr) {
2833 				ii++;
2834 				continue;
2835 			}
2836 			switch (ij >= ppr->pr_ip4s ? -1 :
2837 				qcmp_v4(&pr->pr_ip4[ii], &ppr->pr_ip4[ij])) {
2838 			case -1:
2839 				bcopy(pr->pr_ip4 + ii + 1, pr->pr_ip4 + ii,
2840 				    (--pr->pr_ip4s - ii) * sizeof(*pr->pr_ip4));
2841 				break;
2842 			case 0:
2843 				ii++;
2844 				ij++;
2845 				break;
2846 			case 1:
2847 				ij++;
2848 				break;
2849 			}
2850 		}
2851 		if (pr->pr_ip4s == 0) {
2852 			free(pr->pr_ip4, M_PRISON);
2853 			pr->pr_ip4 = NULL;
2854 		}
2855 	}
2856 	return (0);
2857 }
2858 
2859 /*
2860  * Pass back primary IPv4 address of this jail.
2861  *
2862  * If not restricted return success but do not alter the address.  Caller has
2863  * to make sure to initialize it correctly (e.g. INADDR_ANY).
2864  *
2865  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv4.
2866  * Address returned in NBO.
2867  */
2868 int
2869 prison_get_ip4(struct ucred *cred, struct in_addr *ia)
2870 {
2871 	struct prison *pr;
2872 
2873 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2874 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2875 
2876 	pr = cred->cr_prison;
2877 	if (!(pr->pr_flags & PR_IP4))
2878 		return (0);
2879 	mtx_lock(&pr->pr_mtx);
2880 	if (!(pr->pr_flags & PR_IP4)) {
2881 		mtx_unlock(&pr->pr_mtx);
2882 		return (0);
2883 	}
2884 	if (pr->pr_ip4 == NULL) {
2885 		mtx_unlock(&pr->pr_mtx);
2886 		return (EAFNOSUPPORT);
2887 	}
2888 
2889 	ia->s_addr = pr->pr_ip4[0].s_addr;
2890 	mtx_unlock(&pr->pr_mtx);
2891 	return (0);
2892 }
2893 
2894 /*
2895  * Return 1 if we should do proper source address selection or are not jailed.
2896  * We will return 0 if we should bypass source address selection in favour
2897  * of the primary jail IPv4 address. Only in this case *ia will be updated and
2898  * returned in NBO.
2899  * Return EAFNOSUPPORT, in case this jail does not allow IPv4.
2900  */
2901 int
2902 prison_saddrsel_ip4(struct ucred *cred, struct in_addr *ia)
2903 {
2904 	struct prison *pr;
2905 	struct in_addr lia;
2906 	int error;
2907 
2908 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2909 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2910 
2911 	if (!jailed(cred))
2912 		return (1);
2913 
2914 	pr = cred->cr_prison;
2915 	if (pr->pr_flags & PR_IP4_SADDRSEL)
2916 		return (1);
2917 
2918 	lia.s_addr = INADDR_ANY;
2919 	error = prison_get_ip4(cred, &lia);
2920 	if (error)
2921 		return (error);
2922 	if (lia.s_addr == INADDR_ANY)
2923 		return (1);
2924 
2925 	ia->s_addr = lia.s_addr;
2926 	return (0);
2927 }
2928 
2929 /*
2930  * Return true if pr1 and pr2 have the same IPv4 address restrictions.
2931  */
2932 int
2933 prison_equal_ip4(struct prison *pr1, struct prison *pr2)
2934 {
2935 
2936 	if (pr1 == pr2)
2937 		return (1);
2938 
2939 	/*
2940 	 * No need to lock since the PR_IP4_USER flag can't be altered for
2941 	 * existing prisons.
2942 	 */
2943 	while (pr1 != &prison0 &&
2944 #ifdef VIMAGE
2945 	       !(pr1->pr_flags & PR_VNET) &&
2946 #endif
2947 	       !(pr1->pr_flags & PR_IP4_USER))
2948 		pr1 = pr1->pr_parent;
2949 	while (pr2 != &prison0 &&
2950 #ifdef VIMAGE
2951 	       !(pr2->pr_flags & PR_VNET) &&
2952 #endif
2953 	       !(pr2->pr_flags & PR_IP4_USER))
2954 		pr2 = pr2->pr_parent;
2955 	return (pr1 == pr2);
2956 }
2957 
2958 /*
2959  * Make sure our (source) address is set to something meaningful to this
2960  * jail.
2961  *
2962  * Returns 0 if jail doesn't restrict IPv4 or if address belongs to jail,
2963  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
2964  * doesn't allow IPv4.  Address passed in in NBO and returned in NBO.
2965  */
2966 int
2967 prison_local_ip4(struct ucred *cred, struct in_addr *ia)
2968 {
2969 	struct prison *pr;
2970 	struct in_addr ia0;
2971 	int error;
2972 
2973 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2974 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2975 
2976 	pr = cred->cr_prison;
2977 	if (!(pr->pr_flags & PR_IP4))
2978 		return (0);
2979 	mtx_lock(&pr->pr_mtx);
2980 	if (!(pr->pr_flags & PR_IP4)) {
2981 		mtx_unlock(&pr->pr_mtx);
2982 		return (0);
2983 	}
2984 	if (pr->pr_ip4 == NULL) {
2985 		mtx_unlock(&pr->pr_mtx);
2986 		return (EAFNOSUPPORT);
2987 	}
2988 
2989 	ia0.s_addr = ntohl(ia->s_addr);
2990 	if (ia0.s_addr == INADDR_LOOPBACK) {
2991 		ia->s_addr = pr->pr_ip4[0].s_addr;
2992 		mtx_unlock(&pr->pr_mtx);
2993 		return (0);
2994 	}
2995 
2996 	if (ia0.s_addr == INADDR_ANY) {
2997 		/*
2998 		 * In case there is only 1 IPv4 address, bind directly.
2999 		 */
3000 		if (pr->pr_ip4s == 1)
3001 			ia->s_addr = pr->pr_ip4[0].s_addr;
3002 		mtx_unlock(&pr->pr_mtx);
3003 		return (0);
3004 	}
3005 
3006 	error = _prison_check_ip4(pr, ia);
3007 	mtx_unlock(&pr->pr_mtx);
3008 	return (error);
3009 }
3010 
3011 /*
3012  * Rewrite destination address in case we will connect to loopback address.
3013  *
3014  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv4.
3015  * Address passed in in NBO and returned in NBO.
3016  */
3017 int
3018 prison_remote_ip4(struct ucred *cred, struct in_addr *ia)
3019 {
3020 	struct prison *pr;
3021 
3022 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3023 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
3024 
3025 	pr = cred->cr_prison;
3026 	if (!(pr->pr_flags & PR_IP4))
3027 		return (0);
3028 	mtx_lock(&pr->pr_mtx);
3029 	if (!(pr->pr_flags & PR_IP4)) {
3030 		mtx_unlock(&pr->pr_mtx);
3031 		return (0);
3032 	}
3033 	if (pr->pr_ip4 == NULL) {
3034 		mtx_unlock(&pr->pr_mtx);
3035 		return (EAFNOSUPPORT);
3036 	}
3037 
3038 	if (ntohl(ia->s_addr) == INADDR_LOOPBACK) {
3039 		ia->s_addr = pr->pr_ip4[0].s_addr;
3040 		mtx_unlock(&pr->pr_mtx);
3041 		return (0);
3042 	}
3043 
3044 	/*
3045 	 * Return success because nothing had to be changed.
3046 	 */
3047 	mtx_unlock(&pr->pr_mtx);
3048 	return (0);
3049 }
3050 
3051 /*
3052  * Check if given address belongs to the jail referenced by cred/prison.
3053  *
3054  * Returns 0 if jail doesn't restrict IPv4 or if address belongs to jail,
3055  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3056  * doesn't allow IPv4.  Address passed in in NBO.
3057  */
3058 static int
3059 _prison_check_ip4(const struct prison *pr, const struct in_addr *ia)
3060 {
3061 	int i, a, z, d;
3062 
3063 	/*
3064 	 * Check the primary IP.
3065 	 */
3066 	if (pr->pr_ip4[0].s_addr == ia->s_addr)
3067 		return (0);
3068 
3069 	/*
3070 	 * All the other IPs are sorted so we can do a binary search.
3071 	 */
3072 	a = 0;
3073 	z = pr->pr_ip4s - 2;
3074 	while (a <= z) {
3075 		i = (a + z) / 2;
3076 		d = qcmp_v4(&pr->pr_ip4[i+1], ia);
3077 		if (d > 0)
3078 			z = i - 1;
3079 		else if (d < 0)
3080 			a = i + 1;
3081 		else
3082 			return (0);
3083 	}
3084 
3085 	return (EADDRNOTAVAIL);
3086 }
3087 
3088 int
3089 prison_check_ip4(const struct ucred *cred, const struct in_addr *ia)
3090 {
3091 	struct prison *pr;
3092 	int error;
3093 
3094 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3095 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
3096 
3097 	pr = cred->cr_prison;
3098 	if (!(pr->pr_flags & PR_IP4))
3099 		return (0);
3100 	mtx_lock(&pr->pr_mtx);
3101 	if (!(pr->pr_flags & PR_IP4)) {
3102 		mtx_unlock(&pr->pr_mtx);
3103 		return (0);
3104 	}
3105 	if (pr->pr_ip4 == NULL) {
3106 		mtx_unlock(&pr->pr_mtx);
3107 		return (EAFNOSUPPORT);
3108 	}
3109 
3110 	error = _prison_check_ip4(pr, ia);
3111 	mtx_unlock(&pr->pr_mtx);
3112 	return (error);
3113 }
3114 #endif
3115 
3116 #ifdef INET6
3117 static int
3118 prison_restrict_ip6(struct prison *pr, struct in6_addr *newip6)
3119 {
3120 	int ii, ij, used;
3121 	struct prison *ppr;
3122 
3123 	ppr = pr->pr_parent;
3124 	if (!(pr->pr_flags & PR_IP6_USER)) {
3125 		/* This has no user settings, so just copy the parent's list. */
3126 		if (pr->pr_ip6s < ppr->pr_ip6s) {
3127 			/*
3128 			 * There's no room for the parent's list.  Use the
3129 			 * new list buffer, which is assumed to be big enough
3130 			 * (if it was passed).  If there's no buffer, try to
3131 			 * allocate one.
3132 			 */
3133 			used = 1;
3134 			if (newip6 == NULL) {
3135 				newip6 = malloc(ppr->pr_ip6s * sizeof(*newip6),
3136 				    M_PRISON, M_NOWAIT);
3137 				if (newip6 != NULL)
3138 					used = 0;
3139 			}
3140 			if (newip6 != NULL) {
3141 				bcopy(ppr->pr_ip6, newip6,
3142 				    ppr->pr_ip6s * sizeof(*newip6));
3143 				free(pr->pr_ip6, M_PRISON);
3144 				pr->pr_ip6 = newip6;
3145 				pr->pr_ip6s = ppr->pr_ip6s;
3146 			}
3147 			return (used);
3148 		}
3149 		pr->pr_ip6s = ppr->pr_ip6s;
3150 		if (pr->pr_ip6s > 0)
3151 			bcopy(ppr->pr_ip6, pr->pr_ip6,
3152 			    pr->pr_ip6s * sizeof(*newip6));
3153 		else if (pr->pr_ip6 != NULL) {
3154 			free(pr->pr_ip6, M_PRISON);
3155 			pr->pr_ip6 = NULL;
3156 		}
3157 	} else if (pr->pr_ip6s > 0) {
3158 		/* Remove addresses that aren't in the parent. */
3159 		for (ij = 0; ij < ppr->pr_ip6s; ij++)
3160 			if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[0],
3161 			    &ppr->pr_ip6[ij]))
3162 				break;
3163 		if (ij < ppr->pr_ip6s)
3164 			ii = 1;
3165 		else {
3166 			bcopy(pr->pr_ip6 + 1, pr->pr_ip6,
3167 			    --pr->pr_ip6s * sizeof(*pr->pr_ip6));
3168 			ii = 0;
3169 		}
3170 		for (ij = 1; ii < pr->pr_ip6s; ) {
3171 			if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[ii],
3172 			    &ppr->pr_ip6[0])) {
3173 				ii++;
3174 				continue;
3175 			}
3176 			switch (ij >= ppr->pr_ip6s ? -1 :
3177 				qcmp_v6(&pr->pr_ip6[ii], &ppr->pr_ip6[ij])) {
3178 			case -1:
3179 				bcopy(pr->pr_ip6 + ii + 1, pr->pr_ip6 + ii,
3180 				    (--pr->pr_ip6s - ii) * sizeof(*pr->pr_ip6));
3181 				break;
3182 			case 0:
3183 				ii++;
3184 				ij++;
3185 				break;
3186 			case 1:
3187 				ij++;
3188 				break;
3189 			}
3190 		}
3191 		if (pr->pr_ip6s == 0) {
3192 			free(pr->pr_ip6, M_PRISON);
3193 			pr->pr_ip6 = NULL;
3194 		}
3195 	}
3196 	return 0;
3197 }
3198 
3199 /*
3200  * Pass back primary IPv6 address for this jail.
3201  *
3202  * If not restricted return success but do not alter the address.  Caller has
3203  * to make sure to initialize it correctly (e.g. IN6ADDR_ANY_INIT).
3204  *
3205  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv6.
3206  */
3207 int
3208 prison_get_ip6(struct ucred *cred, struct in6_addr *ia6)
3209 {
3210 	struct prison *pr;
3211 
3212 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3213 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3214 
3215 	pr = cred->cr_prison;
3216 	if (!(pr->pr_flags & PR_IP6))
3217 		return (0);
3218 	mtx_lock(&pr->pr_mtx);
3219 	if (!(pr->pr_flags & PR_IP6)) {
3220 		mtx_unlock(&pr->pr_mtx);
3221 		return (0);
3222 	}
3223 	if (pr->pr_ip6 == NULL) {
3224 		mtx_unlock(&pr->pr_mtx);
3225 		return (EAFNOSUPPORT);
3226 	}
3227 
3228 	bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3229 	mtx_unlock(&pr->pr_mtx);
3230 	return (0);
3231 }
3232 
3233 /*
3234  * Return 1 if we should do proper source address selection or are not jailed.
3235  * We will return 0 if we should bypass source address selection in favour
3236  * of the primary jail IPv6 address. Only in this case *ia will be updated and
3237  * returned in NBO.
3238  * Return EAFNOSUPPORT, in case this jail does not allow IPv6.
3239  */
3240 int
3241 prison_saddrsel_ip6(struct ucred *cred, struct in6_addr *ia6)
3242 {
3243 	struct prison *pr;
3244 	struct in6_addr lia6;
3245 	int error;
3246 
3247 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3248 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3249 
3250 	if (!jailed(cred))
3251 		return (1);
3252 
3253 	pr = cred->cr_prison;
3254 	if (pr->pr_flags & PR_IP6_SADDRSEL)
3255 		return (1);
3256 
3257 	lia6 = in6addr_any;
3258 	error = prison_get_ip6(cred, &lia6);
3259 	if (error)
3260 		return (error);
3261 	if (IN6_IS_ADDR_UNSPECIFIED(&lia6))
3262 		return (1);
3263 
3264 	bcopy(&lia6, ia6, sizeof(struct in6_addr));
3265 	return (0);
3266 }
3267 
3268 /*
3269  * Return true if pr1 and pr2 have the same IPv6 address restrictions.
3270  */
3271 int
3272 prison_equal_ip6(struct prison *pr1, struct prison *pr2)
3273 {
3274 
3275 	if (pr1 == pr2)
3276 		return (1);
3277 
3278 	while (pr1 != &prison0 &&
3279 #ifdef VIMAGE
3280 	       !(pr1->pr_flags & PR_VNET) &&
3281 #endif
3282 	       !(pr1->pr_flags & PR_IP6_USER))
3283 		pr1 = pr1->pr_parent;
3284 	while (pr2 != &prison0 &&
3285 #ifdef VIMAGE
3286 	       !(pr2->pr_flags & PR_VNET) &&
3287 #endif
3288 	       !(pr2->pr_flags & PR_IP6_USER))
3289 		pr2 = pr2->pr_parent;
3290 	return (pr1 == pr2);
3291 }
3292 
3293 /*
3294  * Make sure our (source) address is set to something meaningful to this jail.
3295  *
3296  * v6only should be set based on (inp->inp_flags & IN6P_IPV6_V6ONLY != 0)
3297  * when needed while binding.
3298  *
3299  * Returns 0 if jail doesn't restrict IPv6 or if address belongs to jail,
3300  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3301  * doesn't allow IPv6.
3302  */
3303 int
3304 prison_local_ip6(struct ucred *cred, struct in6_addr *ia6, int v6only)
3305 {
3306 	struct prison *pr;
3307 	int error;
3308 
3309 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3310 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3311 
3312 	pr = cred->cr_prison;
3313 	if (!(pr->pr_flags & PR_IP6))
3314 		return (0);
3315 	mtx_lock(&pr->pr_mtx);
3316 	if (!(pr->pr_flags & PR_IP6)) {
3317 		mtx_unlock(&pr->pr_mtx);
3318 		return (0);
3319 	}
3320 	if (pr->pr_ip6 == NULL) {
3321 		mtx_unlock(&pr->pr_mtx);
3322 		return (EAFNOSUPPORT);
3323 	}
3324 
3325 	if (IN6_IS_ADDR_LOOPBACK(ia6)) {
3326 		bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3327 		mtx_unlock(&pr->pr_mtx);
3328 		return (0);
3329 	}
3330 
3331 	if (IN6_IS_ADDR_UNSPECIFIED(ia6)) {
3332 		/*
3333 		 * In case there is only 1 IPv6 address, and v6only is true,
3334 		 * then bind directly.
3335 		 */
3336 		if (v6only != 0 && pr->pr_ip6s == 1)
3337 			bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3338 		mtx_unlock(&pr->pr_mtx);
3339 		return (0);
3340 	}
3341 
3342 	error = _prison_check_ip6(pr, ia6);
3343 	mtx_unlock(&pr->pr_mtx);
3344 	return (error);
3345 }
3346 
3347 /*
3348  * Rewrite destination address in case we will connect to loopback address.
3349  *
3350  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv6.
3351  */
3352 int
3353 prison_remote_ip6(struct ucred *cred, struct in6_addr *ia6)
3354 {
3355 	struct prison *pr;
3356 
3357 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3358 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3359 
3360 	pr = cred->cr_prison;
3361 	if (!(pr->pr_flags & PR_IP6))
3362 		return (0);
3363 	mtx_lock(&pr->pr_mtx);
3364 	if (!(pr->pr_flags & PR_IP6)) {
3365 		mtx_unlock(&pr->pr_mtx);
3366 		return (0);
3367 	}
3368 	if (pr->pr_ip6 == NULL) {
3369 		mtx_unlock(&pr->pr_mtx);
3370 		return (EAFNOSUPPORT);
3371 	}
3372 
3373 	if (IN6_IS_ADDR_LOOPBACK(ia6)) {
3374 		bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3375 		mtx_unlock(&pr->pr_mtx);
3376 		return (0);
3377 	}
3378 
3379 	/*
3380 	 * Return success because nothing had to be changed.
3381 	 */
3382 	mtx_unlock(&pr->pr_mtx);
3383 	return (0);
3384 }
3385 
3386 /*
3387  * Check if given address belongs to the jail referenced by cred/prison.
3388  *
3389  * Returns 0 if jail doesn't restrict IPv6 or if address belongs to jail,
3390  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3391  * doesn't allow IPv6.
3392  */
3393 static int
3394 _prison_check_ip6(struct prison *pr, struct in6_addr *ia6)
3395 {
3396 	int i, a, z, d;
3397 
3398 	/*
3399 	 * Check the primary IP.
3400 	 */
3401 	if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[0], ia6))
3402 		return (0);
3403 
3404 	/*
3405 	 * All the other IPs are sorted so we can do a binary search.
3406 	 */
3407 	a = 0;
3408 	z = pr->pr_ip6s - 2;
3409 	while (a <= z) {
3410 		i = (a + z) / 2;
3411 		d = qcmp_v6(&pr->pr_ip6[i+1], ia6);
3412 		if (d > 0)
3413 			z = i - 1;
3414 		else if (d < 0)
3415 			a = i + 1;
3416 		else
3417 			return (0);
3418 	}
3419 
3420 	return (EADDRNOTAVAIL);
3421 }
3422 
3423 int
3424 prison_check_ip6(struct ucred *cred, struct in6_addr *ia6)
3425 {
3426 	struct prison *pr;
3427 	int error;
3428 
3429 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3430 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3431 
3432 	pr = cred->cr_prison;
3433 	if (!(pr->pr_flags & PR_IP6))
3434 		return (0);
3435 	mtx_lock(&pr->pr_mtx);
3436 	if (!(pr->pr_flags & PR_IP6)) {
3437 		mtx_unlock(&pr->pr_mtx);
3438 		return (0);
3439 	}
3440 	if (pr->pr_ip6 == NULL) {
3441 		mtx_unlock(&pr->pr_mtx);
3442 		return (EAFNOSUPPORT);
3443 	}
3444 
3445 	error = _prison_check_ip6(pr, ia6);
3446 	mtx_unlock(&pr->pr_mtx);
3447 	return (error);
3448 }
3449 #endif
3450 
3451 /*
3452  * Check if a jail supports the given address family.
3453  *
3454  * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
3455  * if not.
3456  */
3457 int
3458 prison_check_af(struct ucred *cred, int af)
3459 {
3460 	struct prison *pr;
3461 	int error;
3462 
3463 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3464 
3465 	pr = cred->cr_prison;
3466 #ifdef VIMAGE
3467 	/* Prisons with their own network stack are not limited. */
3468 	if (prison_owns_vnet(cred))
3469 		return (0);
3470 #endif
3471 
3472 	error = 0;
3473 	switch (af)
3474 	{
3475 #ifdef INET
3476 	case AF_INET:
3477 		if (pr->pr_flags & PR_IP4)
3478 		{
3479 			mtx_lock(&pr->pr_mtx);
3480 			if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL)
3481 				error = EAFNOSUPPORT;
3482 			mtx_unlock(&pr->pr_mtx);
3483 		}
3484 		break;
3485 #endif
3486 #ifdef INET6
3487 	case AF_INET6:
3488 		if (pr->pr_flags & PR_IP6)
3489 		{
3490 			mtx_lock(&pr->pr_mtx);
3491 			if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL)
3492 				error = EAFNOSUPPORT;
3493 			mtx_unlock(&pr->pr_mtx);
3494 		}
3495 		break;
3496 #endif
3497 	case AF_LOCAL:
3498 	case AF_ROUTE:
3499 		break;
3500 	default:
3501 		if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
3502 			error = EAFNOSUPPORT;
3503 	}
3504 	return (error);
3505 }
3506 
3507 /*
3508  * Check if given address belongs to the jail referenced by cred (wrapper to
3509  * prison_check_ip[46]).
3510  *
3511  * Returns 0 if jail doesn't restrict the address family or if address belongs
3512  * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
3513  * the jail doesn't allow the address family.  IPv4 Address passed in in NBO.
3514  */
3515 int
3516 prison_if(struct ucred *cred, struct sockaddr *sa)
3517 {
3518 #ifdef INET
3519 	struct sockaddr_in *sai;
3520 #endif
3521 #ifdef INET6
3522 	struct sockaddr_in6 *sai6;
3523 #endif
3524 	int error;
3525 
3526 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3527 	KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
3528 
3529 #ifdef VIMAGE
3530 	if (prison_owns_vnet(cred))
3531 		return (0);
3532 #endif
3533 
3534 	error = 0;
3535 	switch (sa->sa_family)
3536 	{
3537 #ifdef INET
3538 	case AF_INET:
3539 		sai = (struct sockaddr_in *)sa;
3540 		error = prison_check_ip4(cred, &sai->sin_addr);
3541 		break;
3542 #endif
3543 #ifdef INET6
3544 	case AF_INET6:
3545 		sai6 = (struct sockaddr_in6 *)sa;
3546 		error = prison_check_ip6(cred, &sai6->sin6_addr);
3547 		break;
3548 #endif
3549 	default:
3550 		if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
3551 			error = EAFNOSUPPORT;
3552 	}
3553 	return (error);
3554 }
3555 
3556 /*
3557  * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
3558  */
3559 int
3560 prison_check(struct ucred *cred1, struct ucred *cred2)
3561 {
3562 
3563 	return ((cred1->cr_prison == cred2->cr_prison ||
3564 	    prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
3565 }
3566 
3567 /*
3568  * Return 1 if p2 is a child of p1, otherwise 0.
3569  */
3570 int
3571 prison_ischild(struct prison *pr1, struct prison *pr2)
3572 {
3573 
3574 	for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
3575 		if (pr1 == pr2)
3576 			return (1);
3577 	return (0);
3578 }
3579 
3580 /*
3581  * Return 1 if the passed credential is in a jail, otherwise 0.
3582  */
3583 int
3584 jailed(struct ucred *cred)
3585 {
3586 
3587 	return (cred->cr_prison != &prison0);
3588 }
3589 
3590 /*
3591  * Return 1 if the passed credential is in a jail and that jail does not
3592  * have its own virtual network stack, otherwise 0.
3593  */
3594 int
3595 jailed_without_vnet(struct ucred *cred)
3596 {
3597 
3598 	if (!jailed(cred))
3599 		return (0);
3600 #ifdef VIMAGE
3601 	if (prison_owns_vnet(cred))
3602 		return (0);
3603 #endif
3604 
3605 	return (1);
3606 }
3607 
3608 /*
3609  * Return the correct hostname (domainname, et al) for the passed credential.
3610  */
3611 void
3612 getcredhostname(struct ucred *cred, char *buf, size_t size)
3613 {
3614 	struct prison *pr;
3615 
3616 	/*
3617 	 * A NULL credential can be used to shortcut to the physical
3618 	 * system's hostname.
3619 	 */
3620 	pr = (cred != NULL) ? cred->cr_prison : &prison0;
3621 	mtx_lock(&pr->pr_mtx);
3622 	strlcpy(buf, pr->pr_hostname, size);
3623 	mtx_unlock(&pr->pr_mtx);
3624 }
3625 
3626 void
3627 getcreddomainname(struct ucred *cred, char *buf, size_t size)
3628 {
3629 
3630 	mtx_lock(&cred->cr_prison->pr_mtx);
3631 	strlcpy(buf, cred->cr_prison->pr_domainname, size);
3632 	mtx_unlock(&cred->cr_prison->pr_mtx);
3633 }
3634 
3635 void
3636 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
3637 {
3638 
3639 	mtx_lock(&cred->cr_prison->pr_mtx);
3640 	strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
3641 	mtx_unlock(&cred->cr_prison->pr_mtx);
3642 }
3643 
3644 void
3645 getcredhostid(struct ucred *cred, unsigned long *hostid)
3646 {
3647 
3648 	mtx_lock(&cred->cr_prison->pr_mtx);
3649 	*hostid = cred->cr_prison->pr_hostid;
3650 	mtx_unlock(&cred->cr_prison->pr_mtx);
3651 }
3652 
3653 #ifdef VIMAGE
3654 /*
3655  * Determine whether the prison represented by cred owns
3656  * its vnet rather than having it inherited.
3657  *
3658  * Returns 1 in case the prison owns the vnet, 0 otherwise.
3659  */
3660 int
3661 prison_owns_vnet(struct ucred *cred)
3662 {
3663 
3664 	/*
3665 	 * vnets cannot be added/removed after jail creation,
3666 	 * so no need to lock here.
3667 	 */
3668 	return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
3669 }
3670 #endif
3671 
3672 /*
3673  * Determine whether the subject represented by cred can "see"
3674  * status of a mount point.
3675  * Returns: 0 for permitted, ENOENT otherwise.
3676  * XXX: This function should be called cr_canseemount() and should be
3677  *      placed in kern_prot.c.
3678  */
3679 int
3680 prison_canseemount(struct ucred *cred, struct mount *mp)
3681 {
3682 	struct prison *pr;
3683 	struct statfs *sp;
3684 	size_t len;
3685 
3686 	pr = cred->cr_prison;
3687 	if (pr->pr_enforce_statfs == 0)
3688 		return (0);
3689 	if (pr->pr_root->v_mount == mp)
3690 		return (0);
3691 	if (pr->pr_enforce_statfs == 2)
3692 		return (ENOENT);
3693 	/*
3694 	 * If jail's chroot directory is set to "/" we should be able to see
3695 	 * all mount-points from inside a jail.
3696 	 * This is ugly check, but this is the only situation when jail's
3697 	 * directory ends with '/'.
3698 	 */
3699 	if (strcmp(pr->pr_path, "/") == 0)
3700 		return (0);
3701 	len = strlen(pr->pr_path);
3702 	sp = &mp->mnt_stat;
3703 	if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
3704 		return (ENOENT);
3705 	/*
3706 	 * Be sure that we don't have situation where jail's root directory
3707 	 * is "/some/path" and mount point is "/some/pathpath".
3708 	 */
3709 	if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
3710 		return (ENOENT);
3711 	return (0);
3712 }
3713 
3714 void
3715 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
3716 {
3717 	char jpath[MAXPATHLEN];
3718 	struct prison *pr;
3719 	size_t len;
3720 
3721 	pr = cred->cr_prison;
3722 	if (pr->pr_enforce_statfs == 0)
3723 		return;
3724 	if (prison_canseemount(cred, mp) != 0) {
3725 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3726 		strlcpy(sp->f_mntonname, "[restricted]",
3727 		    sizeof(sp->f_mntonname));
3728 		return;
3729 	}
3730 	if (pr->pr_root->v_mount == mp) {
3731 		/*
3732 		 * Clear current buffer data, so we are sure nothing from
3733 		 * the valid path left there.
3734 		 */
3735 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3736 		*sp->f_mntonname = '/';
3737 		return;
3738 	}
3739 	/*
3740 	 * If jail's chroot directory is set to "/" we should be able to see
3741 	 * all mount-points from inside a jail.
3742 	 */
3743 	if (strcmp(pr->pr_path, "/") == 0)
3744 		return;
3745 	len = strlen(pr->pr_path);
3746 	strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3747 	/*
3748 	 * Clear current buffer data, so we are sure nothing from
3749 	 * the valid path left there.
3750 	 */
3751 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3752 	if (*jpath == '\0') {
3753 		/* Should never happen. */
3754 		*sp->f_mntonname = '/';
3755 	} else {
3756 		strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3757 	}
3758 }
3759 
3760 /*
3761  * Check with permission for a specific privilege is granted within jail.  We
3762  * have a specific list of accepted privileges; the rest are denied.
3763  */
3764 int
3765 prison_priv_check(struct ucred *cred, int priv)
3766 {
3767 
3768 	if (!jailed(cred))
3769 		return (0);
3770 
3771 #ifdef VIMAGE
3772 	/*
3773 	 * Privileges specific to prisons with a virtual network stack.
3774 	 * There might be a duplicate entry here in case the privilege
3775 	 * is only granted conditionally in the legacy jail case.
3776 	 */
3777 	switch (priv) {
3778 #ifdef notyet
3779 		/*
3780 		 * NFS-specific privileges.
3781 		 */
3782 	case PRIV_NFS_DAEMON:
3783 	case PRIV_NFS_LOCKD:
3784 #endif
3785 		/*
3786 		 * Network stack privileges.
3787 		 */
3788 	case PRIV_NET_BRIDGE:
3789 	case PRIV_NET_GRE:
3790 	case PRIV_NET_BPF:
3791 	case PRIV_NET_RAW:		/* Dup, cond. in legacy jail case. */
3792 	case PRIV_NET_ROUTE:
3793 	case PRIV_NET_TAP:
3794 	case PRIV_NET_SETIFMTU:
3795 	case PRIV_NET_SETIFFLAGS:
3796 	case PRIV_NET_SETIFCAP:
3797 	case PRIV_NET_SETIFDESCR:
3798 	case PRIV_NET_SETIFNAME	:
3799 	case PRIV_NET_SETIFMETRIC:
3800 	case PRIV_NET_SETIFPHYS:
3801 	case PRIV_NET_SETIFMAC:
3802 	case PRIV_NET_ADDMULTI:
3803 	case PRIV_NET_DELMULTI:
3804 	case PRIV_NET_HWIOCTL:
3805 	case PRIV_NET_SETLLADDR:
3806 	case PRIV_NET_ADDIFGROUP:
3807 	case PRIV_NET_DELIFGROUP:
3808 	case PRIV_NET_IFCREATE:
3809 	case PRIV_NET_IFDESTROY:
3810 	case PRIV_NET_ADDIFADDR:
3811 	case PRIV_NET_DELIFADDR:
3812 	case PRIV_NET_LAGG:
3813 	case PRIV_NET_GIF:
3814 	case PRIV_NET_SETIFVNET:
3815 	case PRIV_NET_SETIFFIB:
3816 
3817 		/*
3818 		 * 802.11-related privileges.
3819 		 */
3820 	case PRIV_NET80211_GETKEY:
3821 #ifdef notyet
3822 	case PRIV_NET80211_MANAGE:		/* XXX-BZ discuss with sam@ */
3823 #endif
3824 
3825 #ifdef notyet
3826 		/*
3827 		 * ATM privileges.
3828 		 */
3829 	case PRIV_NETATM_CFG:
3830 	case PRIV_NETATM_ADD:
3831 	case PRIV_NETATM_DEL:
3832 	case PRIV_NETATM_SET:
3833 
3834 		/*
3835 		 * Bluetooth privileges.
3836 		 */
3837 	case PRIV_NETBLUETOOTH_RAW:
3838 #endif
3839 
3840 		/*
3841 		 * Netgraph and netgraph module privileges.
3842 		 */
3843 	case PRIV_NETGRAPH_CONTROL:
3844 #ifdef notyet
3845 	case PRIV_NETGRAPH_TTY:
3846 #endif
3847 
3848 		/*
3849 		 * IPv4 and IPv6 privileges.
3850 		 */
3851 	case PRIV_NETINET_IPFW:
3852 	case PRIV_NETINET_DIVERT:
3853 	case PRIV_NETINET_PF:
3854 	case PRIV_NETINET_DUMMYNET:
3855 	case PRIV_NETINET_CARP:
3856 	case PRIV_NETINET_MROUTE:
3857 	case PRIV_NETINET_RAW:
3858 	case PRIV_NETINET_ADDRCTRL6:
3859 	case PRIV_NETINET_ND6:
3860 	case PRIV_NETINET_SCOPE6:
3861 	case PRIV_NETINET_ALIFETIME6:
3862 	case PRIV_NETINET_IPSEC:
3863 	case PRIV_NETINET_BINDANY:
3864 
3865 #ifdef notyet
3866 		/*
3867 		 * NCP privileges.
3868 		 */
3869 	case PRIV_NETNCP:
3870 
3871 		/*
3872 		 * SMB privileges.
3873 		 */
3874 	case PRIV_NETSMB:
3875 #endif
3876 
3877 	/*
3878 	 * No default: or deny here.
3879 	 * In case of no permit fall through to next switch().
3880 	 */
3881 		if (cred->cr_prison->pr_flags & PR_VNET)
3882 			return (0);
3883 	}
3884 #endif /* VIMAGE */
3885 
3886 	switch (priv) {
3887 
3888 		/*
3889 		 * Allow ktrace privileges for root in jail.
3890 		 */
3891 	case PRIV_KTRACE:
3892 
3893 #if 0
3894 		/*
3895 		 * Allow jailed processes to configure audit identity and
3896 		 * submit audit records (login, etc).  In the future we may
3897 		 * want to further refine the relationship between audit and
3898 		 * jail.
3899 		 */
3900 	case PRIV_AUDIT_GETAUDIT:
3901 	case PRIV_AUDIT_SETAUDIT:
3902 	case PRIV_AUDIT_SUBMIT:
3903 #endif
3904 
3905 		/*
3906 		 * Allow jailed processes to manipulate process UNIX
3907 		 * credentials in any way they see fit.
3908 		 */
3909 	case PRIV_CRED_SETUID:
3910 	case PRIV_CRED_SETEUID:
3911 	case PRIV_CRED_SETGID:
3912 	case PRIV_CRED_SETEGID:
3913 	case PRIV_CRED_SETGROUPS:
3914 	case PRIV_CRED_SETREUID:
3915 	case PRIV_CRED_SETREGID:
3916 	case PRIV_CRED_SETRESUID:
3917 	case PRIV_CRED_SETRESGID:
3918 
3919 		/*
3920 		 * Jail implements visibility constraints already, so allow
3921 		 * jailed root to override uid/gid-based constraints.
3922 		 */
3923 	case PRIV_SEEOTHERGIDS:
3924 	case PRIV_SEEOTHERUIDS:
3925 
3926 		/*
3927 		 * Jail implements inter-process debugging limits already, so
3928 		 * allow jailed root various debugging privileges.
3929 		 */
3930 	case PRIV_DEBUG_DIFFCRED:
3931 	case PRIV_DEBUG_SUGID:
3932 	case PRIV_DEBUG_UNPRIV:
3933 
3934 		/*
3935 		 * Allow jail to set various resource limits and login
3936 		 * properties, and for now, exceed process resource limits.
3937 		 */
3938 	case PRIV_PROC_LIMIT:
3939 	case PRIV_PROC_SETLOGIN:
3940 	case PRIV_PROC_SETRLIMIT:
3941 
3942 		/*
3943 		 * System V and POSIX IPC privileges are granted in jail.
3944 		 */
3945 	case PRIV_IPC_READ:
3946 	case PRIV_IPC_WRITE:
3947 	case PRIV_IPC_ADMIN:
3948 	case PRIV_IPC_MSGSIZE:
3949 	case PRIV_MQ_ADMIN:
3950 
3951 		/*
3952 		 * Jail operations within a jail work on child jails.
3953 		 */
3954 	case PRIV_JAIL_ATTACH:
3955 	case PRIV_JAIL_SET:
3956 	case PRIV_JAIL_REMOVE:
3957 
3958 		/*
3959 		 * Jail implements its own inter-process limits, so allow
3960 		 * root processes in jail to change scheduling on other
3961 		 * processes in the same jail.  Likewise for signalling.
3962 		 */
3963 	case PRIV_SCHED_DIFFCRED:
3964 	case PRIV_SCHED_CPUSET:
3965 	case PRIV_SIGNAL_DIFFCRED:
3966 	case PRIV_SIGNAL_SUGID:
3967 
3968 		/*
3969 		 * Allow jailed processes to write to sysctls marked as jail
3970 		 * writable.
3971 		 */
3972 	case PRIV_SYSCTL_WRITEJAIL:
3973 
3974 		/*
3975 		 * Allow root in jail to manage a variety of quota
3976 		 * properties.  These should likely be conditional on a
3977 		 * configuration option.
3978 		 */
3979 	case PRIV_VFS_GETQUOTA:
3980 	case PRIV_VFS_SETQUOTA:
3981 
3982 		/*
3983 		 * Since Jail relies on chroot() to implement file system
3984 		 * protections, grant many VFS privileges to root in jail.
3985 		 * Be careful to exclude mount-related and NFS-related
3986 		 * privileges.
3987 		 */
3988 	case PRIV_VFS_READ:
3989 	case PRIV_VFS_WRITE:
3990 	case PRIV_VFS_ADMIN:
3991 	case PRIV_VFS_EXEC:
3992 	case PRIV_VFS_LOOKUP:
3993 	case PRIV_VFS_BLOCKRESERVE:	/* XXXRW: Slightly surprising. */
3994 	case PRIV_VFS_CHFLAGS_DEV:
3995 	case PRIV_VFS_CHOWN:
3996 	case PRIV_VFS_CHROOT:
3997 	case PRIV_VFS_RETAINSUGID:
3998 	case PRIV_VFS_FCHROOT:
3999 	case PRIV_VFS_LINK:
4000 	case PRIV_VFS_SETGID:
4001 	case PRIV_VFS_STAT:
4002 	case PRIV_VFS_STICKYFILE:
4003 
4004 		/*
4005 		 * As in the non-jail case, non-root users are expected to be
4006 		 * able to read kernel/phyiscal memory (provided /dev/[k]mem
4007 		 * exists in the jail and they have permission to access it).
4008 		 */
4009 	case PRIV_KMEM_READ:
4010 		return (0);
4011 
4012 		/*
4013 		 * Depending on the global setting, allow privilege of
4014 		 * setting system flags.
4015 		 */
4016 	case PRIV_VFS_SYSFLAGS:
4017 		if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
4018 			return (0);
4019 		else
4020 			return (EPERM);
4021 
4022 		/*
4023 		 * Depending on the global setting, allow privilege of
4024 		 * mounting/unmounting file systems.
4025 		 */
4026 	case PRIV_VFS_MOUNT:
4027 	case PRIV_VFS_UNMOUNT:
4028 	case PRIV_VFS_MOUNT_NONUSER:
4029 	case PRIV_VFS_MOUNT_OWNER:
4030 		if (cred->cr_prison->pr_allow & PR_ALLOW_MOUNT &&
4031 		    cred->cr_prison->pr_enforce_statfs < 2)
4032 			return (0);
4033 		else
4034 			return (EPERM);
4035 
4036 		/*
4037 		 * Allow jailed root to bind reserved ports and reuse in-use
4038 		 * ports.
4039 		 */
4040 	case PRIV_NETINET_RESERVEDPORT:
4041 	case PRIV_NETINET_REUSEPORT:
4042 		return (0);
4043 
4044 		/*
4045 		 * Allow jailed root to set certain IPv4/6 (option) headers.
4046 		 */
4047 	case PRIV_NETINET_SETHDROPTS:
4048 		return (0);
4049 
4050 		/*
4051 		 * Conditionally allow creating raw sockets in jail.
4052 		 */
4053 	case PRIV_NETINET_RAW:
4054 		if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
4055 			return (0);
4056 		else
4057 			return (EPERM);
4058 
4059 		/*
4060 		 * Since jail implements its own visibility limits on netstat
4061 		 * sysctls, allow getcred.  This allows identd to work in
4062 		 * jail.
4063 		 */
4064 	case PRIV_NETINET_GETCRED:
4065 		return (0);
4066 
4067 		/*
4068 		 * Allow jailed root to set loginclass.
4069 		 */
4070 	case PRIV_PROC_SETLOGINCLASS:
4071 		return (0);
4072 
4073 	default:
4074 		/*
4075 		 * In all remaining cases, deny the privilege request.  This
4076 		 * includes almost all network privileges, many system
4077 		 * configuration privileges.
4078 		 */
4079 		return (EPERM);
4080 	}
4081 }
4082 
4083 /*
4084  * Return the part of pr2's name that is relative to pr1, or the whole name
4085  * if it does not directly follow.
4086  */
4087 
4088 char *
4089 prison_name(struct prison *pr1, struct prison *pr2)
4090 {
4091 	char *name;
4092 
4093 	/* Jails see themselves as "0" (if they see themselves at all). */
4094 	if (pr1 == pr2)
4095 		return "0";
4096 	name = pr2->pr_name;
4097 	if (prison_ischild(pr1, pr2)) {
4098 		/*
4099 		 * pr1 isn't locked (and allprison_lock may not be either)
4100 		 * so its length can't be counted on.  But the number of dots
4101 		 * can be counted on - and counted.
4102 		 */
4103 		for (; pr1 != &prison0; pr1 = pr1->pr_parent)
4104 			name = strchr(name, '.') + 1;
4105 	}
4106 	return (name);
4107 }
4108 
4109 /*
4110  * Return the part of pr2's path that is relative to pr1, or the whole path
4111  * if it does not directly follow.
4112  */
4113 static char *
4114 prison_path(struct prison *pr1, struct prison *pr2)
4115 {
4116 	char *path1, *path2;
4117 	int len1;
4118 
4119 	path1 = pr1->pr_path;
4120 	path2 = pr2->pr_path;
4121 	if (!strcmp(path1, "/"))
4122 		return (path2);
4123 	len1 = strlen(path1);
4124 	if (strncmp(path1, path2, len1))
4125 		return (path2);
4126 	if (path2[len1] == '\0')
4127 		return "/";
4128 	if (path2[len1] == '/')
4129 		return (path2 + len1);
4130 	return (path2);
4131 }
4132 
4133 
4134 /*
4135  * Jail-related sysctls.
4136  */
4137 static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW, 0,
4138     "Jails");
4139 
4140 static int
4141 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
4142 {
4143 	struct xprison *xp;
4144 	struct prison *pr, *cpr;
4145 #ifdef INET
4146 	struct in_addr *ip4 = NULL;
4147 	int ip4s = 0;
4148 #endif
4149 #ifdef INET6
4150 	struct in6_addr *ip6 = NULL;
4151 	int ip6s = 0;
4152 #endif
4153 	int descend, error;
4154 
4155 	xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
4156 	pr = req->td->td_ucred->cr_prison;
4157 	error = 0;
4158 	sx_slock(&allprison_lock);
4159 	FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
4160 #if defined(INET) || defined(INET6)
4161  again:
4162 #endif
4163 		mtx_lock(&cpr->pr_mtx);
4164 #ifdef INET
4165 		if (cpr->pr_ip4s > 0) {
4166 			if (ip4s < cpr->pr_ip4s) {
4167 				ip4s = cpr->pr_ip4s;
4168 				mtx_unlock(&cpr->pr_mtx);
4169 				ip4 = realloc(ip4, ip4s *
4170 				    sizeof(struct in_addr), M_TEMP, M_WAITOK);
4171 				goto again;
4172 			}
4173 			bcopy(cpr->pr_ip4, ip4,
4174 			    cpr->pr_ip4s * sizeof(struct in_addr));
4175 		}
4176 #endif
4177 #ifdef INET6
4178 		if (cpr->pr_ip6s > 0) {
4179 			if (ip6s < cpr->pr_ip6s) {
4180 				ip6s = cpr->pr_ip6s;
4181 				mtx_unlock(&cpr->pr_mtx);
4182 				ip6 = realloc(ip6, ip6s *
4183 				    sizeof(struct in6_addr), M_TEMP, M_WAITOK);
4184 				goto again;
4185 			}
4186 			bcopy(cpr->pr_ip6, ip6,
4187 			    cpr->pr_ip6s * sizeof(struct in6_addr));
4188 		}
4189 #endif
4190 		if (cpr->pr_ref == 0) {
4191 			mtx_unlock(&cpr->pr_mtx);
4192 			continue;
4193 		}
4194 		bzero(xp, sizeof(*xp));
4195 		xp->pr_version = XPRISON_VERSION;
4196 		xp->pr_id = cpr->pr_id;
4197 		xp->pr_state = cpr->pr_uref > 0
4198 		    ? PRISON_STATE_ALIVE : PRISON_STATE_DYING;
4199 		strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
4200 		strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
4201 		strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
4202 #ifdef INET
4203 		xp->pr_ip4s = cpr->pr_ip4s;
4204 #endif
4205 #ifdef INET6
4206 		xp->pr_ip6s = cpr->pr_ip6s;
4207 #endif
4208 		mtx_unlock(&cpr->pr_mtx);
4209 		error = SYSCTL_OUT(req, xp, sizeof(*xp));
4210 		if (error)
4211 			break;
4212 #ifdef INET
4213 		if (xp->pr_ip4s > 0) {
4214 			error = SYSCTL_OUT(req, ip4,
4215 			    xp->pr_ip4s * sizeof(struct in_addr));
4216 			if (error)
4217 				break;
4218 		}
4219 #endif
4220 #ifdef INET6
4221 		if (xp->pr_ip6s > 0) {
4222 			error = SYSCTL_OUT(req, ip6,
4223 			    xp->pr_ip6s * sizeof(struct in6_addr));
4224 			if (error)
4225 				break;
4226 		}
4227 #endif
4228 	}
4229 	sx_sunlock(&allprison_lock);
4230 	free(xp, M_TEMP);
4231 #ifdef INET
4232 	free(ip4, M_TEMP);
4233 #endif
4234 #ifdef INET6
4235 	free(ip6, M_TEMP);
4236 #endif
4237 	return (error);
4238 }
4239 
4240 SYSCTL_OID(_security_jail, OID_AUTO, list,
4241     CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4242     sysctl_jail_list, "S", "List of active jails");
4243 
4244 static int
4245 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
4246 {
4247 	int error, injail;
4248 
4249 	injail = jailed(req->td->td_ucred);
4250 	error = SYSCTL_OUT(req, &injail, sizeof(injail));
4251 
4252 	return (error);
4253 }
4254 
4255 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
4256     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4257     sysctl_jail_jailed, "I", "Process in jail?");
4258 
4259 static int
4260 sysctl_jail_vnet(SYSCTL_HANDLER_ARGS)
4261 {
4262 	int error, havevnet;
4263 #ifdef VIMAGE
4264 	struct ucred *cred = req->td->td_ucred;
4265 
4266 	havevnet = jailed(cred) && prison_owns_vnet(cred);
4267 #else
4268 	havevnet = 0;
4269 #endif
4270 	error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet));
4271 
4272 	return (error);
4273 }
4274 
4275 SYSCTL_PROC(_security_jail, OID_AUTO, vnet,
4276     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4277     sysctl_jail_vnet, "I", "Jail owns VNET?");
4278 
4279 #if defined(INET) || defined(INET6)
4280 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
4281     &jail_max_af_ips, 0,
4282     "Number of IP addresses a jail may have at most per address family (deprecated)");
4283 #endif
4284 
4285 /*
4286  * Default parameters for jail(2) compatibility.  For historical reasons,
4287  * the sysctl names have varying similarity to the parameter names.  Prisons
4288  * just see their own parameters, and can't change them.
4289  */
4290 static int
4291 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
4292 {
4293 	struct prison *pr;
4294 	int allow, error, i;
4295 
4296 	pr = req->td->td_ucred->cr_prison;
4297 	allow = (pr == &prison0) ? jail_default_allow : pr->pr_allow;
4298 
4299 	/* Get the current flag value, and convert it to a boolean. */
4300 	i = (allow & arg2) ? 1 : 0;
4301 	if (arg1 != NULL)
4302 		i = !i;
4303 	error = sysctl_handle_int(oidp, &i, 0, req);
4304 	if (error || !req->newptr)
4305 		return (error);
4306 	i = i ? arg2 : 0;
4307 	if (arg1 != NULL)
4308 		i ^= arg2;
4309 	/*
4310 	 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
4311 	 * for writing.
4312 	 */
4313 	mtx_lock(&prison0.pr_mtx);
4314 	jail_default_allow = (jail_default_allow & ~arg2) | i;
4315 	mtx_unlock(&prison0.pr_mtx);
4316 	return (0);
4317 }
4318 
4319 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
4320     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4321     NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
4322     "Processes in jail can set their hostnames (deprecated)");
4323 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
4324     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4325     (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
4326     "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)");
4327 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
4328     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4329     NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
4330     "Processes in jail can use System V IPC primitives (deprecated)");
4331 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
4332     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4333     NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
4334     "Prison root can create raw sockets (deprecated)");
4335 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
4336     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4337     NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
4338     "Processes in jail can alter system file flags (deprecated)");
4339 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
4340     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4341     NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
4342     "Processes in jail can mount/unmount jail-friendly file systems (deprecated)");
4343 SYSCTL_PROC(_security_jail, OID_AUTO, mount_devfs_allowed,
4344     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4345     NULL, PR_ALLOW_MOUNT_DEVFS, sysctl_jail_default_allow, "I",
4346     "Processes in jail can mount the devfs file system (deprecated)");
4347 SYSCTL_PROC(_security_jail, OID_AUTO, mount_fdescfs_allowed,
4348     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4349     NULL, PR_ALLOW_MOUNT_FDESCFS, sysctl_jail_default_allow, "I",
4350     "Processes in jail can mount the fdescfs file system (deprecated)");
4351 SYSCTL_PROC(_security_jail, OID_AUTO, mount_nullfs_allowed,
4352     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4353     NULL, PR_ALLOW_MOUNT_NULLFS, sysctl_jail_default_allow, "I",
4354     "Processes in jail can mount the nullfs file system (deprecated)");
4355 SYSCTL_PROC(_security_jail, OID_AUTO, mount_procfs_allowed,
4356     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4357     NULL, PR_ALLOW_MOUNT_PROCFS, sysctl_jail_default_allow, "I",
4358     "Processes in jail can mount the procfs file system (deprecated)");
4359 SYSCTL_PROC(_security_jail, OID_AUTO, mount_linprocfs_allowed,
4360     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4361     NULL, PR_ALLOW_MOUNT_LINPROCFS, sysctl_jail_default_allow, "I",
4362     "Processes in jail can mount the linprocfs file system (deprecated)");
4363 SYSCTL_PROC(_security_jail, OID_AUTO, mount_linsysfs_allowed,
4364     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4365     NULL, PR_ALLOW_MOUNT_LINSYSFS, sysctl_jail_default_allow, "I",
4366     "Processes in jail can mount the linsysfs file system (deprecated)");
4367 SYSCTL_PROC(_security_jail, OID_AUTO, mount_tmpfs_allowed,
4368     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4369     NULL, PR_ALLOW_MOUNT_TMPFS, sysctl_jail_default_allow, "I",
4370     "Processes in jail can mount the tmpfs file system (deprecated)");
4371 SYSCTL_PROC(_security_jail, OID_AUTO, mount_zfs_allowed,
4372     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4373     NULL, PR_ALLOW_MOUNT_ZFS, sysctl_jail_default_allow, "I",
4374     "Processes in jail can mount the zfs file system (deprecated)");
4375 
4376 static int
4377 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
4378 {
4379 	struct prison *pr;
4380 	int level, error;
4381 
4382 	pr = req->td->td_ucred->cr_prison;
4383 	level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
4384 	error = sysctl_handle_int(oidp, &level, 0, req);
4385 	if (error || !req->newptr)
4386 		return (error);
4387 	*(int *)arg1 = level;
4388 	return (0);
4389 }
4390 
4391 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
4392     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4393     &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
4394     sysctl_jail_default_level, "I",
4395     "Processes in jail cannot see all mounted file systems (deprecated)");
4396 
4397 SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset,
4398     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
4399     &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum),
4400     sysctl_jail_default_level, "I",
4401     "Ruleset for the devfs filesystem in jail (deprecated)");
4402 
4403 /*
4404  * Nodes to describe jail parameters.  Maximum length of string parameters
4405  * is returned in the string itself, and the other parameters exist merely
4406  * to make themselves and their types known.
4407  */
4408 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW, 0,
4409     "Jail parameters");
4410 
4411 int
4412 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
4413 {
4414 	int i;
4415 	long l;
4416 	size_t s;
4417 	char numbuf[12];
4418 
4419 	switch (oidp->oid_kind & CTLTYPE)
4420 	{
4421 	case CTLTYPE_LONG:
4422 	case CTLTYPE_ULONG:
4423 		l = 0;
4424 #ifdef SCTL_MASK32
4425 		if (!(req->flags & SCTL_MASK32))
4426 #endif
4427 			return (SYSCTL_OUT(req, &l, sizeof(l)));
4428 	case CTLTYPE_INT:
4429 	case CTLTYPE_UINT:
4430 		i = 0;
4431 		return (SYSCTL_OUT(req, &i, sizeof(i)));
4432 	case CTLTYPE_STRING:
4433 		snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
4434 		return
4435 		    (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
4436 	case CTLTYPE_STRUCT:
4437 		s = (size_t)arg2;
4438 		return (SYSCTL_OUT(req, &s, sizeof(s)));
4439 	}
4440 	return (0);
4441 }
4442 
4443 /*
4444  * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at
4445  * jail creation time but cannot be changed in an existing jail.
4446  */
4447 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
4448 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
4449 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
4450 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
4451 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
4452     "I", "Jail secure level");
4453 SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I",
4454     "Jail value for kern.osreldate and uname -K");
4455 SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN,
4456     "Jail value for kern.osrelease and uname -r");
4457 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
4458     "I", "Jail cannot see all mounted file systems");
4459 SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW,
4460     "I", "Ruleset for in-jail devfs mounts");
4461 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
4462     "B", "Jail persistence");
4463 #ifdef VIMAGE
4464 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
4465     "E,jailsys", "Virtual network stack");
4466 #endif
4467 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
4468     "B", "Jail is in the process of shutting down");
4469 
4470 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
4471 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
4472     "I", "Current number of child jails");
4473 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
4474     "I", "Maximum number of child jails");
4475 
4476 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
4477 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
4478     "Jail hostname");
4479 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
4480     "Jail NIS domainname");
4481 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
4482     "Jail host UUID");
4483 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
4484     "LU", "Jail host ID");
4485 
4486 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
4487 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
4488 
4489 #ifdef INET
4490 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
4491     "Jail IPv4 address virtualization");
4492 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
4493     "S,in_addr,a", "Jail IPv4 addresses");
4494 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4495     "B", "Do (not) use IPv4 source address selection rather than the "
4496     "primary jail IPv4 address.");
4497 #endif
4498 #ifdef INET6
4499 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
4500     "Jail IPv6 address virtualization");
4501 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
4502     "S,in6_addr,a", "Jail IPv6 addresses");
4503 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4504     "B", "Do (not) use IPv6 source address selection rather than the "
4505     "primary jail IPv6 address.");
4506 #endif
4507 
4508 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
4509 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
4510     "B", "Jail may set hostname");
4511 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
4512     "B", "Jail may use SYSV IPC");
4513 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
4514     "B", "Jail may create raw sockets");
4515 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
4516     "B", "Jail may alter system file flags");
4517 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
4518     "B", "Jail may set file quotas");
4519 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
4520     "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
4521 
4522 SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags");
4523 SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW,
4524     "B", "Jail may mount/unmount jail-friendly file systems in general");
4525 SYSCTL_JAIL_PARAM(_allow_mount, devfs, CTLTYPE_INT | CTLFLAG_RW,
4526     "B", "Jail may mount the devfs file system");
4527 SYSCTL_JAIL_PARAM(_allow_mount, fdescfs, CTLTYPE_INT | CTLFLAG_RW,
4528     "B", "Jail may mount the fdescfs file system");
4529 SYSCTL_JAIL_PARAM(_allow_mount, nullfs, CTLTYPE_INT | CTLFLAG_RW,
4530     "B", "Jail may mount the nullfs file system");
4531 SYSCTL_JAIL_PARAM(_allow_mount, procfs, CTLTYPE_INT | CTLFLAG_RW,
4532     "B", "Jail may mount the procfs file system");
4533 SYSCTL_JAIL_PARAM(_allow_mount, linprocfs, CTLTYPE_INT | CTLFLAG_RW,
4534     "B", "Jail may mount the linprocfs file system");
4535 SYSCTL_JAIL_PARAM(_allow_mount, linsysfs, CTLTYPE_INT | CTLFLAG_RW,
4536     "B", "Jail may mount the linsysfs file system");
4537 SYSCTL_JAIL_PARAM(_allow_mount, tmpfs, CTLTYPE_INT | CTLFLAG_RW,
4538     "B", "Jail may mount the tmpfs file system");
4539 SYSCTL_JAIL_PARAM(_allow_mount, zfs, CTLTYPE_INT | CTLFLAG_RW,
4540     "B", "Jail may mount the zfs file system");
4541 
4542 #ifdef RACCT
4543 void
4544 prison_racct_foreach(void (*callback)(struct racct *racct,
4545     void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
4546     void *arg2, void *arg3)
4547 {
4548 	struct prison_racct *prr;
4549 
4550 	ASSERT_RACCT_ENABLED();
4551 
4552 	sx_slock(&allprison_lock);
4553 	if (pre != NULL)
4554 		(pre)();
4555 	LIST_FOREACH(prr, &allprison_racct, prr_next)
4556 		(callback)(prr->prr_racct, arg2, arg3);
4557 	if (post != NULL)
4558 		(post)();
4559 	sx_sunlock(&allprison_lock);
4560 }
4561 
4562 static struct prison_racct *
4563 prison_racct_find_locked(const char *name)
4564 {
4565 	struct prison_racct *prr;
4566 
4567 	ASSERT_RACCT_ENABLED();
4568 	sx_assert(&allprison_lock, SA_XLOCKED);
4569 
4570 	if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN)
4571 		return (NULL);
4572 
4573 	LIST_FOREACH(prr, &allprison_racct, prr_next) {
4574 		if (strcmp(name, prr->prr_name) != 0)
4575 			continue;
4576 
4577 		/* Found prison_racct with a matching name? */
4578 		prison_racct_hold(prr);
4579 		return (prr);
4580 	}
4581 
4582 	/* Add new prison_racct. */
4583 	prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK);
4584 	racct_create(&prr->prr_racct);
4585 
4586 	strcpy(prr->prr_name, name);
4587 	refcount_init(&prr->prr_refcount, 1);
4588 	LIST_INSERT_HEAD(&allprison_racct, prr, prr_next);
4589 
4590 	return (prr);
4591 }
4592 
4593 struct prison_racct *
4594 prison_racct_find(const char *name)
4595 {
4596 	struct prison_racct *prr;
4597 
4598 	ASSERT_RACCT_ENABLED();
4599 
4600 	sx_xlock(&allprison_lock);
4601 	prr = prison_racct_find_locked(name);
4602 	sx_xunlock(&allprison_lock);
4603 	return (prr);
4604 }
4605 
4606 void
4607 prison_racct_hold(struct prison_racct *prr)
4608 {
4609 
4610 	ASSERT_RACCT_ENABLED();
4611 
4612 	refcount_acquire(&prr->prr_refcount);
4613 }
4614 
4615 static void
4616 prison_racct_free_locked(struct prison_racct *prr)
4617 {
4618 
4619 	ASSERT_RACCT_ENABLED();
4620 	sx_assert(&allprison_lock, SA_XLOCKED);
4621 
4622 	if (refcount_release(&prr->prr_refcount)) {
4623 		racct_destroy(&prr->prr_racct);
4624 		LIST_REMOVE(prr, prr_next);
4625 		free(prr, M_PRISON_RACCT);
4626 	}
4627 }
4628 
4629 void
4630 prison_racct_free(struct prison_racct *prr)
4631 {
4632 	int old;
4633 
4634 	ASSERT_RACCT_ENABLED();
4635 	sx_assert(&allprison_lock, SA_UNLOCKED);
4636 
4637 	old = prr->prr_refcount;
4638 	if (old > 1 && atomic_cmpset_int(&prr->prr_refcount, old, old - 1))
4639 		return;
4640 
4641 	sx_xlock(&allprison_lock);
4642 	prison_racct_free_locked(prr);
4643 	sx_xunlock(&allprison_lock);
4644 }
4645 
4646 static void
4647 prison_racct_attach(struct prison *pr)
4648 {
4649 	struct prison_racct *prr;
4650 
4651 	ASSERT_RACCT_ENABLED();
4652 	sx_assert(&allprison_lock, SA_XLOCKED);
4653 
4654 	prr = prison_racct_find_locked(pr->pr_name);
4655 	KASSERT(prr != NULL, ("cannot find prison_racct"));
4656 
4657 	pr->pr_prison_racct = prr;
4658 }
4659 
4660 /*
4661  * Handle jail renaming.  From the racct point of view, renaming means
4662  * moving from one prison_racct to another.
4663  */
4664 static void
4665 prison_racct_modify(struct prison *pr)
4666 {
4667 	struct proc *p;
4668 	struct ucred *cred;
4669 	struct prison_racct *oldprr;
4670 
4671 	ASSERT_RACCT_ENABLED();
4672 
4673 	sx_slock(&allproc_lock);
4674 	sx_xlock(&allprison_lock);
4675 
4676 	if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) {
4677 		sx_xunlock(&allprison_lock);
4678 		sx_sunlock(&allproc_lock);
4679 		return;
4680 	}
4681 
4682 	oldprr = pr->pr_prison_racct;
4683 	pr->pr_prison_racct = NULL;
4684 
4685 	prison_racct_attach(pr);
4686 
4687 	/*
4688 	 * Move resource utilisation records.
4689 	 */
4690 	racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct);
4691 
4692 	/*
4693 	 * Force rctl to reattach rules to processes.
4694 	 */
4695 	FOREACH_PROC_IN_SYSTEM(p) {
4696 		PROC_LOCK(p);
4697 		cred = crhold(p->p_ucred);
4698 		PROC_UNLOCK(p);
4699 		racct_proc_ucred_changed(p, cred, cred);
4700 		crfree(cred);
4701 	}
4702 
4703 	sx_sunlock(&allproc_lock);
4704 	prison_racct_free_locked(oldprr);
4705 	sx_xunlock(&allprison_lock);
4706 }
4707 
4708 static void
4709 prison_racct_detach(struct prison *pr)
4710 {
4711 
4712 	ASSERT_RACCT_ENABLED();
4713 	sx_assert(&allprison_lock, SA_UNLOCKED);
4714 
4715 	if (pr->pr_prison_racct == NULL)
4716 		return;
4717 	prison_racct_free(pr->pr_prison_racct);
4718 	pr->pr_prison_racct = NULL;
4719 }
4720 #endif /* RACCT */
4721 
4722 #ifdef DDB
4723 
4724 static void
4725 db_show_prison(struct prison *pr)
4726 {
4727 	int fi;
4728 #if defined(INET) || defined(INET6)
4729 	int ii;
4730 #endif
4731 	unsigned jsf;
4732 #ifdef INET6
4733 	char ip6buf[INET6_ADDRSTRLEN];
4734 #endif
4735 
4736 	db_printf("prison %p:\n", pr);
4737 	db_printf(" jid             = %d\n", pr->pr_id);
4738 	db_printf(" name            = %s\n", pr->pr_name);
4739 	db_printf(" parent          = %p\n", pr->pr_parent);
4740 	db_printf(" ref             = %d\n", pr->pr_ref);
4741 	db_printf(" uref            = %d\n", pr->pr_uref);
4742 	db_printf(" path            = %s\n", pr->pr_path);
4743 	db_printf(" cpuset          = %d\n", pr->pr_cpuset
4744 	    ? pr->pr_cpuset->cs_id : -1);
4745 #ifdef VIMAGE
4746 	db_printf(" vnet            = %p\n", pr->pr_vnet);
4747 #endif
4748 	db_printf(" root            = %p\n", pr->pr_root);
4749 	db_printf(" securelevel     = %d\n", pr->pr_securelevel);
4750 	db_printf(" devfs_rsnum     = %d\n", pr->pr_devfs_rsnum);
4751 	db_printf(" children.max    = %d\n", pr->pr_childmax);
4752 	db_printf(" children.cur    = %d\n", pr->pr_childcount);
4753 	db_printf(" child           = %p\n", LIST_FIRST(&pr->pr_children));
4754 	db_printf(" sibling         = %p\n", LIST_NEXT(pr, pr_sibling));
4755 	db_printf(" flags           = 0x%x", pr->pr_flags);
4756 	for (fi = 0; fi < nitems(pr_flag_names); fi++)
4757 		if (pr_flag_names[fi] != NULL && (pr->pr_flags & (1 << fi)))
4758 			db_printf(" %s", pr_flag_names[fi]);
4759 	for (fi = 0; fi < nitems(pr_flag_jailsys); fi++) {
4760 		jsf = pr->pr_flags &
4761 		    (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
4762 		db_printf(" %-16s= %s\n", pr_flag_jailsys[fi].name,
4763 		    pr_flag_jailsys[fi].disable &&
4764 		      (jsf == pr_flag_jailsys[fi].disable) ? "disable"
4765 		    : (jsf == pr_flag_jailsys[fi].new) ? "new"
4766 		    : "inherit");
4767 	}
4768 	db_printf(" allow           = 0x%x", pr->pr_allow);
4769 	for (fi = 0; fi < nitems(pr_allow_names); fi++)
4770 		if (pr_allow_names[fi] != NULL && (pr->pr_allow & (1 << fi)))
4771 			db_printf(" %s", pr_allow_names[fi]);
4772 	db_printf("\n");
4773 	db_printf(" enforce_statfs  = %d\n", pr->pr_enforce_statfs);
4774 	db_printf(" host.hostname   = %s\n", pr->pr_hostname);
4775 	db_printf(" host.domainname = %s\n", pr->pr_domainname);
4776 	db_printf(" host.hostuuid   = %s\n", pr->pr_hostuuid);
4777 	db_printf(" host.hostid     = %lu\n", pr->pr_hostid);
4778 #ifdef INET
4779 	db_printf(" ip4s            = %d\n", pr->pr_ip4s);
4780 	for (ii = 0; ii < pr->pr_ip4s; ii++)
4781 		db_printf(" %s %s\n",
4782 		    ii == 0 ? "ip4.addr        =" : "                 ",
4783 		    inet_ntoa(pr->pr_ip4[ii]));
4784 #endif
4785 #ifdef INET6
4786 	db_printf(" ip6s            = %d\n", pr->pr_ip6s);
4787 	for (ii = 0; ii < pr->pr_ip6s; ii++)
4788 		db_printf(" %s %s\n",
4789 		    ii == 0 ? "ip6.addr        =" : "                 ",
4790 		    ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4791 #endif
4792 }
4793 
4794 DB_SHOW_COMMAND(prison, db_show_prison_command)
4795 {
4796 	struct prison *pr;
4797 
4798 	if (!have_addr) {
4799 		/*
4800 		 * Show all prisons in the list, and prison0 which is not
4801 		 * listed.
4802 		 */
4803 		db_show_prison(&prison0);
4804 		if (!db_pager_quit) {
4805 			TAILQ_FOREACH(pr, &allprison, pr_list) {
4806 				db_show_prison(pr);
4807 				if (db_pager_quit)
4808 					break;
4809 			}
4810 		}
4811 		return;
4812 	}
4813 
4814 	if (addr == 0)
4815 		pr = &prison0;
4816 	else {
4817 		/* Look for a prison with the ID and with references. */
4818 		TAILQ_FOREACH(pr, &allprison, pr_list)
4819 			if (pr->pr_id == addr && pr->pr_ref > 0)
4820 				break;
4821 		if (pr == NULL)
4822 			/* Look again, without requiring a reference. */
4823 			TAILQ_FOREACH(pr, &allprison, pr_list)
4824 				if (pr->pr_id == addr)
4825 					break;
4826 		if (pr == NULL)
4827 			/* Assume address points to a valid prison. */
4828 			pr = (struct prison *)addr;
4829 	}
4830 	db_show_prison(pr);
4831 }
4832 
4833 #endif /* DDB */
4834