xref: /freebsd/sys/kern/kern_jail.c (revision 7a1c0d963366a31363d3705697a083dd8efee077)
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/rctl.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 #ifdef INET6
72 #include <netinet6/in6_var.h>
73 #endif /* INET6 */
74 #endif /* DDB */
75 
76 #include <security/mac/mac_framework.h>
77 
78 #define	DEFAULT_HOSTUUID	"00000000-0000-0000-0000-000000000000"
79 
80 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
81 
82 /* Keep struct prison prison0 and some code in kern_jail_set() readable. */
83 #ifdef INET
84 #ifdef INET6
85 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL|PR_IP6_SADDRSEL
86 #else
87 #define	_PR_IP_SADDRSEL	PR_IP4_SADDRSEL
88 #endif
89 #else /* !INET */
90 #ifdef INET6
91 #define	_PR_IP_SADDRSEL	PR_IP6_SADDRSEL
92 #else
93 #define	_PR_IP_SADDRSEL	0
94 #endif
95 #endif
96 
97 /* prison0 describes what is "real" about the system. */
98 struct prison prison0 = {
99 	.pr_id		= 0,
100 	.pr_name	= "0",
101 	.pr_ref		= 1,
102 	.pr_uref	= 1,
103 	.pr_path	= "/",
104 	.pr_securelevel	= -1,
105 	.pr_childmax	= JAIL_MAX,
106 	.pr_hostuuid	= DEFAULT_HOSTUUID,
107 	.pr_children	= LIST_HEAD_INITIALIZER(prison0.pr_children),
108 #ifdef VIMAGE
109 	.pr_flags	= PR_HOST|PR_VNET|_PR_IP_SADDRSEL,
110 #else
111 	.pr_flags	= PR_HOST|_PR_IP_SADDRSEL,
112 #endif
113 	.pr_allow	= PR_ALLOW_ALL,
114 };
115 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
116 
117 /* allprison and lastprid are protected by allprison_lock. */
118 struct	sx allprison_lock;
119 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
120 struct	prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
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 INET
129 static int _prison_check_ip4(struct prison *pr, struct in_addr *ia);
130 static int prison_restrict_ip4(struct prison *pr, struct in_addr *newip4);
131 #endif
132 #ifdef INET6
133 static int _prison_check_ip6(struct prison *pr, struct in6_addr *ia6);
134 static int prison_restrict_ip6(struct prison *pr, struct in6_addr *newip6);
135 #endif
136 
137 /* Flags for prison_deref */
138 #define	PD_DEREF	0x01
139 #define	PD_DEUREF	0x02
140 #define	PD_LOCKED	0x04
141 #define	PD_LIST_SLOCKED	0x08
142 #define	PD_LIST_XLOCKED	0x10
143 
144 /*
145  * Parameter names corresponding to PR_* flag values.  Size values are for kvm
146  * as we cannot figure out the size of a sparse array, or an array without a
147  * terminating entry.
148  */
149 static char *pr_flag_names[] = {
150 	[0] = "persist",
151 #ifdef INET
152 	[7] = "ip4.saddrsel",
153 #endif
154 #ifdef INET6
155 	[8] = "ip6.saddrsel",
156 #endif
157 };
158 const size_t pr_flag_names_size = sizeof(pr_flag_names);
159 
160 static char *pr_flag_nonames[] = {
161 	[0] = "nopersist",
162 #ifdef INET
163 	[7] = "ip4.nosaddrsel",
164 #endif
165 #ifdef INET6
166 	[8] = "ip6.nosaddrsel",
167 #endif
168 };
169 const size_t pr_flag_nonames_size = sizeof(pr_flag_nonames);
170 
171 struct jailsys_flags {
172 	const char	*name;
173 	unsigned	 disable;
174 	unsigned	 new;
175 } pr_flag_jailsys[] = {
176 	{ "host", 0, PR_HOST },
177 #ifdef VIMAGE
178 	{ "vnet", 0, PR_VNET },
179 #endif
180 #ifdef INET
181 	{ "ip4", PR_IP4_USER | PR_IP4_DISABLE, PR_IP4_USER },
182 #endif
183 #ifdef INET6
184 	{ "ip6", PR_IP6_USER | PR_IP6_DISABLE, PR_IP6_USER },
185 #endif
186 };
187 const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys);
188 
189 static char *pr_allow_names[] = {
190 	"allow.set_hostname",
191 	"allow.sysvipc",
192 	"allow.raw_sockets",
193 	"allow.chflags",
194 	"allow.mount",
195 	"allow.quotas",
196 	"allow.socket_af",
197 };
198 const size_t pr_allow_names_size = sizeof(pr_allow_names);
199 
200 static char *pr_allow_nonames[] = {
201 	"allow.noset_hostname",
202 	"allow.nosysvipc",
203 	"allow.noraw_sockets",
204 	"allow.nochflags",
205 	"allow.nomount",
206 	"allow.noquotas",
207 	"allow.nosocket_af",
208 };
209 const size_t pr_allow_nonames_size = sizeof(pr_allow_nonames);
210 
211 #define	JAIL_DEFAULT_ALLOW		PR_ALLOW_SET_HOSTNAME
212 #define	JAIL_DEFAULT_ENFORCE_STATFS	2
213 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
214 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
215 #if defined(INET) || defined(INET6)
216 static unsigned jail_max_af_ips = 255;
217 #endif
218 
219 #ifdef INET
220 static int
221 qcmp_v4(const void *ip1, const void *ip2)
222 {
223 	in_addr_t iaa, iab;
224 
225 	/*
226 	 * We need to compare in HBO here to get the list sorted as expected
227 	 * by the result of the code.  Sorting NBO addresses gives you
228 	 * interesting results.  If you do not understand, do not try.
229 	 */
230 	iaa = ntohl(((const struct in_addr *)ip1)->s_addr);
231 	iab = ntohl(((const struct in_addr *)ip2)->s_addr);
232 
233 	/*
234 	 * Do not simply return the difference of the two numbers, the int is
235 	 * not wide enough.
236 	 */
237 	if (iaa > iab)
238 		return (1);
239 	else if (iaa < iab)
240 		return (-1);
241 	else
242 		return (0);
243 }
244 #endif
245 
246 #ifdef INET6
247 static int
248 qcmp_v6(const void *ip1, const void *ip2)
249 {
250 	const struct in6_addr *ia6a, *ia6b;
251 	int i, rc;
252 
253 	ia6a = (const struct in6_addr *)ip1;
254 	ia6b = (const struct in6_addr *)ip2;
255 
256 	rc = 0;
257 	for (i = 0; rc == 0 && i < sizeof(struct in6_addr); i++) {
258 		if (ia6a->s6_addr[i] > ia6b->s6_addr[i])
259 			rc = 1;
260 		else if (ia6a->s6_addr[i] < ia6b->s6_addr[i])
261 			rc = -1;
262 	}
263 	return (rc);
264 }
265 #endif
266 
267 /*
268  * struct jail_args {
269  *	struct jail *jail;
270  * };
271  */
272 int
273 jail(struct thread *td, struct jail_args *uap)
274 {
275 	uint32_t version;
276 	int error;
277 	struct jail j;
278 
279 	error = copyin(uap->jail, &version, sizeof(uint32_t));
280 	if (error)
281 		return (error);
282 
283 	switch (version) {
284 	case 0:
285 	{
286 		struct jail_v0 j0;
287 
288 		/* FreeBSD single IPv4 jails. */
289 		bzero(&j, sizeof(struct jail));
290 		error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
291 		if (error)
292 			return (error);
293 		j.version = j0.version;
294 		j.path = j0.path;
295 		j.hostname = j0.hostname;
296 		j.ip4s = j0.ip_number;
297 		break;
298 	}
299 
300 	case 1:
301 		/*
302 		 * Version 1 was used by multi-IPv4 jail implementations
303 		 * that never made it into the official kernel.
304 		 */
305 		return (EINVAL);
306 
307 	case 2:	/* JAIL_API_VERSION */
308 		/* FreeBSD multi-IPv4/IPv6,noIP jails. */
309 		error = copyin(uap->jail, &j, sizeof(struct jail));
310 		if (error)
311 			return (error);
312 		break;
313 
314 	default:
315 		/* Sci-Fi jails are not supported, sorry. */
316 		return (EINVAL);
317 	}
318 	return (kern_jail(td, &j));
319 }
320 
321 int
322 kern_jail(struct thread *td, struct jail *j)
323 {
324 	struct iovec optiov[2 * (4
325 			    + sizeof(pr_allow_names) / sizeof(pr_allow_names[0])
326 #ifdef INET
327 			    + 1
328 #endif
329 #ifdef INET6
330 			    + 1
331 #endif
332 			    )];
333 	struct uio opt;
334 	char *u_path, *u_hostname, *u_name;
335 #ifdef INET
336 	uint32_t ip4s;
337 	struct in_addr *u_ip4;
338 #endif
339 #ifdef INET6
340 	struct in6_addr *u_ip6;
341 #endif
342 	size_t tmplen;
343 	int error, enforce_statfs, fi;
344 
345 	bzero(&optiov, sizeof(optiov));
346 	opt.uio_iov = optiov;
347 	opt.uio_iovcnt = 0;
348 	opt.uio_offset = -1;
349 	opt.uio_resid = -1;
350 	opt.uio_segflg = UIO_SYSSPACE;
351 	opt.uio_rw = UIO_READ;
352 	opt.uio_td = td;
353 
354 	/* Set permissions for top-level jails from sysctls. */
355 	if (!jailed(td->td_ucred)) {
356 		for (fi = 0; fi < sizeof(pr_allow_names) /
357 		     sizeof(pr_allow_names[0]); fi++) {
358 			optiov[opt.uio_iovcnt].iov_base =
359 			    (jail_default_allow & (1 << fi))
360 			    ? pr_allow_names[fi] : pr_allow_nonames[fi];
361 			optiov[opt.uio_iovcnt].iov_len =
362 			    strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
363 			opt.uio_iovcnt += 2;
364 		}
365 		optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
366 		optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
367 		opt.uio_iovcnt++;
368 		enforce_statfs = jail_default_enforce_statfs;
369 		optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
370 		optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
371 		opt.uio_iovcnt++;
372 	}
373 
374 	tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
375 #ifdef INET
376 	ip4s = (j->version == 0) ? 1 : j->ip4s;
377 	if (ip4s > jail_max_af_ips)
378 		return (EINVAL);
379 	tmplen += ip4s * sizeof(struct in_addr);
380 #else
381 	if (j->ip4s > 0)
382 		return (EINVAL);
383 #endif
384 #ifdef INET6
385 	if (j->ip6s > jail_max_af_ips)
386 		return (EINVAL);
387 	tmplen += j->ip6s * sizeof(struct in6_addr);
388 #else
389 	if (j->ip6s > 0)
390 		return (EINVAL);
391 #endif
392 	u_path = malloc(tmplen, M_TEMP, M_WAITOK);
393 	u_hostname = u_path + MAXPATHLEN;
394 	u_name = u_hostname + MAXHOSTNAMELEN;
395 #ifdef INET
396 	u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
397 #endif
398 #ifdef INET6
399 #ifdef INET
400 	u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
401 #else
402 	u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
403 #endif
404 #endif
405 	optiov[opt.uio_iovcnt].iov_base = "path";
406 	optiov[opt.uio_iovcnt].iov_len = sizeof("path");
407 	opt.uio_iovcnt++;
408 	optiov[opt.uio_iovcnt].iov_base = u_path;
409 	error = copyinstr(j->path, u_path, MAXPATHLEN,
410 	    &optiov[opt.uio_iovcnt].iov_len);
411 	if (error) {
412 		free(u_path, M_TEMP);
413 		return (error);
414 	}
415 	opt.uio_iovcnt++;
416 	optiov[opt.uio_iovcnt].iov_base = "host.hostname";
417 	optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
418 	opt.uio_iovcnt++;
419 	optiov[opt.uio_iovcnt].iov_base = u_hostname;
420 	error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
421 	    &optiov[opt.uio_iovcnt].iov_len);
422 	if (error) {
423 		free(u_path, M_TEMP);
424 		return (error);
425 	}
426 	opt.uio_iovcnt++;
427 	if (j->jailname != NULL) {
428 		optiov[opt.uio_iovcnt].iov_base = "name";
429 		optiov[opt.uio_iovcnt].iov_len = sizeof("name");
430 		opt.uio_iovcnt++;
431 		optiov[opt.uio_iovcnt].iov_base = u_name;
432 		error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
433 		    &optiov[opt.uio_iovcnt].iov_len);
434 		if (error) {
435 			free(u_path, M_TEMP);
436 			return (error);
437 		}
438 		opt.uio_iovcnt++;
439 	}
440 #ifdef INET
441 	optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
442 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
443 	opt.uio_iovcnt++;
444 	optiov[opt.uio_iovcnt].iov_base = u_ip4;
445 	optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
446 	if (j->version == 0)
447 		u_ip4->s_addr = j->ip4s;
448 	else {
449 		error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
450 		if (error) {
451 			free(u_path, M_TEMP);
452 			return (error);
453 		}
454 	}
455 	opt.uio_iovcnt++;
456 #endif
457 #ifdef INET6
458 	optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
459 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
460 	opt.uio_iovcnt++;
461 	optiov[opt.uio_iovcnt].iov_base = u_ip6;
462 	optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
463 	error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
464 	if (error) {
465 		free(u_path, M_TEMP);
466 		return (error);
467 	}
468 	opt.uio_iovcnt++;
469 #endif
470 	KASSERT(opt.uio_iovcnt <= sizeof(optiov) / sizeof(optiov[0]),
471 	    ("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
472 	error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
473 	free(u_path, M_TEMP);
474 	return (error);
475 }
476 
477 
478 /*
479  * struct jail_set_args {
480  *	struct iovec *iovp;
481  *	unsigned int iovcnt;
482  *	int flags;
483  * };
484  */
485 int
486 jail_set(struct thread *td, struct jail_set_args *uap)
487 {
488 	struct uio *auio;
489 	int error;
490 
491 	/* Check that we have an even number of iovecs. */
492 	if (uap->iovcnt & 1)
493 		return (EINVAL);
494 
495 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
496 	if (error)
497 		return (error);
498 	error = kern_jail_set(td, auio, uap->flags);
499 	free(auio, M_IOV);
500 	return (error);
501 }
502 
503 int
504 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
505 {
506 	struct nameidata nd;
507 #ifdef INET
508 	struct in_addr *ip4;
509 #endif
510 #ifdef INET6
511 	struct in6_addr *ip6;
512 #endif
513 	struct vfsopt *opt;
514 	struct vfsoptlist *opts;
515 	struct prison *pr, *deadpr, *mypr, *ppr, *tpr;
516 	struct vnode *root;
517 	char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid;
518 #if defined(INET) || defined(INET6)
519 	struct prison *tppr;
520 	void *op;
521 #endif
522 	unsigned long hid;
523 	size_t namelen, onamelen;
524 	int created, cuflags, descend, enforce, error, errmsg_len, errmsg_pos;
525 	int gotchildmax, gotenforce, gothid, gotslevel;
526 	int fi, jid, jsys, len, level;
527 	int childmax, slevel, vfslocked;
528 #if defined(INET) || defined(INET6)
529 	int ii, ij;
530 #endif
531 #ifdef INET
532 	int ip4s, redo_ip4;
533 #endif
534 #ifdef INET6
535 	int ip6s, redo_ip6;
536 #endif
537 	unsigned pr_flags, ch_flags;
538 	unsigned pr_allow, ch_allow, tallow;
539 	char numbuf[12];
540 
541 	error = priv_check(td, PRIV_JAIL_SET);
542 	if (!error && (flags & JAIL_ATTACH))
543 		error = priv_check(td, PRIV_JAIL_ATTACH);
544 	if (error)
545 		return (error);
546 	mypr = ppr = td->td_ucred->cr_prison;
547 	if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
548 		return (EPERM);
549 	if (flags & ~JAIL_SET_MASK)
550 		return (EINVAL);
551 
552 	/*
553 	 * Check all the parameters before committing to anything.  Not all
554 	 * errors can be caught early, but we may as well try.  Also, this
555 	 * takes care of some expensive stuff (path lookup) before getting
556 	 * the allprison lock.
557 	 *
558 	 * XXX Jails are not filesystems, and jail parameters are not mount
559 	 *     options.  But it makes more sense to re-use the vfsopt code
560 	 *     than duplicate it under a different name.
561 	 */
562 	error = vfs_buildopts(optuio, &opts);
563 	if (error)
564 		return (error);
565 #ifdef INET
566 	ip4 = NULL;
567 #endif
568 #ifdef INET6
569 	ip6 = NULL;
570 #endif
571 
572 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
573 	if (error == ENOENT)
574 		jid = 0;
575 	else if (error != 0)
576 		goto done_free;
577 
578 	error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
579 	if (error == ENOENT)
580 		gotslevel = 0;
581 	else if (error != 0)
582 		goto done_free;
583 	else
584 		gotslevel = 1;
585 
586 	error =
587 	    vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
588 	if (error == ENOENT)
589 		gotchildmax = 0;
590 	else if (error != 0)
591 		goto done_free;
592 	else
593 		gotchildmax = 1;
594 
595 	error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
596 	if (error == ENOENT)
597 		gotenforce = 0;
598 	else if (error != 0)
599 		goto done_free;
600 	else if (enforce < 0 || enforce > 2) {
601 		error = EINVAL;
602 		goto done_free;
603 	} else
604 		gotenforce = 1;
605 
606 	pr_flags = ch_flags = 0;
607 	for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
608 	    fi++) {
609 		if (pr_flag_names[fi] == NULL)
610 			continue;
611 		vfs_flagopt(opts, pr_flag_names[fi], &pr_flags, 1 << fi);
612 		vfs_flagopt(opts, pr_flag_nonames[fi], &ch_flags, 1 << fi);
613 	}
614 	ch_flags |= pr_flags;
615 	for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
616 	    fi++) {
617 		error = vfs_copyopt(opts, pr_flag_jailsys[fi].name, &jsys,
618 		    sizeof(jsys));
619 		if (error == ENOENT)
620 			continue;
621 		if (error != 0)
622 			goto done_free;
623 		switch (jsys) {
624 		case JAIL_SYS_DISABLE:
625 			if (!pr_flag_jailsys[fi].disable) {
626 				error = EINVAL;
627 				goto done_free;
628 			}
629 			pr_flags |= pr_flag_jailsys[fi].disable;
630 			break;
631 		case JAIL_SYS_NEW:
632 			pr_flags |= pr_flag_jailsys[fi].new;
633 			break;
634 		case JAIL_SYS_INHERIT:
635 			break;
636 		default:
637 			error = EINVAL;
638 			goto done_free;
639 		}
640 		ch_flags |=
641 		    pr_flag_jailsys[fi].new | pr_flag_jailsys[fi].disable;
642 	}
643 	if ((flags & (JAIL_CREATE | JAIL_UPDATE | JAIL_ATTACH)) == JAIL_CREATE
644 	    && !(pr_flags & PR_PERSIST)) {
645 		error = EINVAL;
646 		vfs_opterror(opts, "new jail must persist or attach");
647 		goto done_errmsg;
648 	}
649 #ifdef VIMAGE
650 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
651 		error = EINVAL;
652 		vfs_opterror(opts, "vnet cannot be changed after creation");
653 		goto done_errmsg;
654 	}
655 #endif
656 #ifdef INET
657 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
658 		error = EINVAL;
659 		vfs_opterror(opts, "ip4 cannot be changed after creation");
660 		goto done_errmsg;
661 	}
662 #endif
663 #ifdef INET6
664 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
665 		error = EINVAL;
666 		vfs_opterror(opts, "ip6 cannot be changed after creation");
667 		goto done_errmsg;
668 	}
669 #endif
670 
671 	pr_allow = ch_allow = 0;
672 	for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
673 	    fi++) {
674 		vfs_flagopt(opts, pr_allow_names[fi], &pr_allow, 1 << fi);
675 		vfs_flagopt(opts, pr_allow_nonames[fi], &ch_allow, 1 << fi);
676 	}
677 	ch_allow |= pr_allow;
678 
679 	error = vfs_getopt(opts, "name", (void **)&name, &len);
680 	if (error == ENOENT)
681 		name = NULL;
682 	else if (error != 0)
683 		goto done_free;
684 	else {
685 		if (len == 0 || name[len - 1] != '\0') {
686 			error = EINVAL;
687 			goto done_free;
688 		}
689 		if (len > MAXHOSTNAMELEN) {
690 			error = ENAMETOOLONG;
691 			goto done_free;
692 		}
693 	}
694 
695 	error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
696 	if (error == ENOENT)
697 		host = NULL;
698 	else if (error != 0)
699 		goto done_free;
700 	else {
701 		ch_flags |= PR_HOST;
702 		pr_flags |= PR_HOST;
703 		if (len == 0 || host[len - 1] != '\0') {
704 			error = EINVAL;
705 			goto done_free;
706 		}
707 		if (len > MAXHOSTNAMELEN) {
708 			error = ENAMETOOLONG;
709 			goto done_free;
710 		}
711 	}
712 
713 	error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
714 	if (error == ENOENT)
715 		domain = NULL;
716 	else if (error != 0)
717 		goto done_free;
718 	else {
719 		ch_flags |= PR_HOST;
720 		pr_flags |= PR_HOST;
721 		if (len == 0 || domain[len - 1] != '\0') {
722 			error = EINVAL;
723 			goto done_free;
724 		}
725 		if (len > MAXHOSTNAMELEN) {
726 			error = ENAMETOOLONG;
727 			goto done_free;
728 		}
729 	}
730 
731 	error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
732 	if (error == ENOENT)
733 		uuid = NULL;
734 	else if (error != 0)
735 		goto done_free;
736 	else {
737 		ch_flags |= PR_HOST;
738 		pr_flags |= PR_HOST;
739 		if (len == 0 || uuid[len - 1] != '\0') {
740 			error = EINVAL;
741 			goto done_free;
742 		}
743 		if (len > HOSTUUIDLEN) {
744 			error = ENAMETOOLONG;
745 			goto done_free;
746 		}
747 	}
748 
749 #ifdef COMPAT_FREEBSD32
750 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
751 		uint32_t hid32;
752 
753 		error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
754 		hid = hid32;
755 	} else
756 #endif
757 		error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
758 	if (error == ENOENT)
759 		gothid = 0;
760 	else if (error != 0)
761 		goto done_free;
762 	else {
763 		gothid = 1;
764 		ch_flags |= PR_HOST;
765 		pr_flags |= PR_HOST;
766 	}
767 
768 #ifdef INET
769 	error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
770 	if (error == ENOENT)
771 		ip4s = (pr_flags & PR_IP4_DISABLE) ? 0 : -1;
772 	else if (error != 0)
773 		goto done_free;
774 	else if (ip4s & (sizeof(*ip4) - 1)) {
775 		error = EINVAL;
776 		goto done_free;
777 	} else {
778 		ch_flags |= PR_IP4_USER | PR_IP4_DISABLE;
779 		if (ip4s == 0)
780 			pr_flags |= PR_IP4_USER | PR_IP4_DISABLE;
781 		else {
782 			pr_flags = (pr_flags & ~PR_IP4_DISABLE) | PR_IP4_USER;
783 			ip4s /= sizeof(*ip4);
784 			if (ip4s > jail_max_af_ips) {
785 				error = EINVAL;
786 				vfs_opterror(opts, "too many IPv4 addresses");
787 				goto done_errmsg;
788 			}
789 			ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
790 			bcopy(op, ip4, ip4s * sizeof(*ip4));
791 			/*
792 			 * IP addresses are all sorted but ip[0] to preserve
793 			 * the primary IP address as given from userland.
794 			 * This special IP is used for unbound outgoing
795 			 * connections as well for "loopback" traffic in case
796 			 * source address selection cannot find any more fitting
797 			 * address to connect from.
798 			 */
799 			if (ip4s > 1)
800 				qsort(ip4 + 1, ip4s - 1, sizeof(*ip4), qcmp_v4);
801 			/*
802 			 * Check for duplicate addresses and do some simple
803 			 * zero and broadcast checks. If users give other bogus
804 			 * addresses it is their problem.
805 			 *
806 			 * We do not have to care about byte order for these
807 			 * checks so we will do them in NBO.
808 			 */
809 			for (ii = 0; ii < ip4s; ii++) {
810 				if (ip4[ii].s_addr == INADDR_ANY ||
811 				    ip4[ii].s_addr == INADDR_BROADCAST) {
812 					error = EINVAL;
813 					goto done_free;
814 				}
815 				if ((ii+1) < ip4s &&
816 				    (ip4[0].s_addr == ip4[ii+1].s_addr ||
817 				     ip4[ii].s_addr == ip4[ii+1].s_addr)) {
818 					error = EINVAL;
819 					goto done_free;
820 				}
821 			}
822 		}
823 	}
824 #endif
825 
826 #ifdef INET6
827 	error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
828 	if (error == ENOENT)
829 		ip6s = (pr_flags & PR_IP6_DISABLE) ? 0 : -1;
830 	else if (error != 0)
831 		goto done_free;
832 	else if (ip6s & (sizeof(*ip6) - 1)) {
833 		error = EINVAL;
834 		goto done_free;
835 	} else {
836 		ch_flags |= PR_IP6_USER | PR_IP6_DISABLE;
837 		if (ip6s == 0)
838 			pr_flags |= PR_IP6_USER | PR_IP6_DISABLE;
839 		else {
840 			pr_flags = (pr_flags & ~PR_IP6_DISABLE) | PR_IP6_USER;
841 			ip6s /= sizeof(*ip6);
842 			if (ip6s > jail_max_af_ips) {
843 				error = EINVAL;
844 				vfs_opterror(opts, "too many IPv6 addresses");
845 				goto done_errmsg;
846 			}
847 			ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
848 			bcopy(op, ip6, ip6s * sizeof(*ip6));
849 			if (ip6s > 1)
850 				qsort(ip6 + 1, ip6s - 1, sizeof(*ip6), qcmp_v6);
851 			for (ii = 0; ii < ip6s; ii++) {
852 				if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
853 					error = EINVAL;
854 					goto done_free;
855 				}
856 				if ((ii+1) < ip6s &&
857 				    (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
858 				     IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
859 				{
860 					error = EINVAL;
861 					goto done_free;
862 				}
863 			}
864 		}
865 	}
866 #endif
867 
868 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
869 	if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
870 		error = EINVAL;
871 		vfs_opterror(opts,
872 		    "vnet jails cannot have IP address restrictions");
873 		goto done_errmsg;
874 	}
875 #endif
876 
877 	root = NULL;
878 	error = vfs_getopt(opts, "path", (void **)&path, &len);
879 	if (error == ENOENT)
880 		path = NULL;
881 	else if (error != 0)
882 		goto done_free;
883 	else {
884 		if (flags & JAIL_UPDATE) {
885 			error = EINVAL;
886 			vfs_opterror(opts,
887 			    "path cannot be changed after creation");
888 			goto done_errmsg;
889 		}
890 		if (len == 0 || path[len - 1] != '\0') {
891 			error = EINVAL;
892 			goto done_free;
893 		}
894 		if (len < 2 || (len == 2 && path[0] == '/'))
895 			path = NULL;
896 		else {
897 			/* Leave room for a real-root full pathname. */
898 			if (len + (path[0] == '/' && strcmp(mypr->pr_path, "/")
899 			    ? strlen(mypr->pr_path) : 0) > MAXPATHLEN) {
900 				error = ENAMETOOLONG;
901 				goto done_free;
902 			}
903 			NDINIT(&nd, LOOKUP, MPSAFE | FOLLOW, UIO_SYSSPACE,
904 			    path, td);
905 			error = namei(&nd);
906 			if (error)
907 				goto done_free;
908 			vfslocked = NDHASGIANT(&nd);
909 			root = nd.ni_vp;
910 			NDFREE(&nd, NDF_ONLY_PNBUF);
911 			if (root->v_type != VDIR) {
912 				error = ENOTDIR;
913 				vrele(root);
914 				VFS_UNLOCK_GIANT(vfslocked);
915 				goto done_free;
916 			}
917 			VFS_UNLOCK_GIANT(vfslocked);
918 		}
919 	}
920 
921 	/*
922 	 * Grab the allprison lock before letting modules check their
923 	 * parameters.  Once we have it, do not let go so we'll have a
924 	 * consistent view of the OSD list.
925 	 */
926 	sx_xlock(&allprison_lock);
927 	error = osd_jail_call(NULL, PR_METHOD_CHECK, opts);
928 	if (error)
929 		goto done_unlock_list;
930 
931 	/* By now, all parameters should have been noted. */
932 	TAILQ_FOREACH(opt, opts, link) {
933 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
934 			error = EINVAL;
935 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
936 			goto done_unlock_list;
937 		}
938 	}
939 
940 	/*
941 	 * See if we are creating a new record or updating an existing one.
942 	 * This abuses the file error codes ENOENT and EEXIST.
943 	 */
944 	cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
945 	if (!cuflags) {
946 		error = EINVAL;
947 		vfs_opterror(opts, "no valid operation (create or update)");
948 		goto done_unlock_list;
949 	}
950 	pr = NULL;
951 	namelc = NULL;
952 	if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) {
953 		namelc = strrchr(name, '.');
954 		jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10);
955 		if (*p != '\0')
956 			jid = 0;
957 	}
958 	if (jid != 0) {
959 		/*
960 		 * See if a requested jid already exists.  There is an
961 		 * information leak here if the jid exists but is not within
962 		 * the caller's jail hierarchy.  Jail creators will get EEXIST
963 		 * even though they cannot see the jail, and CREATE | UPDATE
964 		 * will return ENOENT which is not normally a valid error.
965 		 */
966 		if (jid < 0) {
967 			error = EINVAL;
968 			vfs_opterror(opts, "negative jid");
969 			goto done_unlock_list;
970 		}
971 		pr = prison_find(jid);
972 		if (pr != NULL) {
973 			ppr = pr->pr_parent;
974 			/* Create: jid must not exist. */
975 			if (cuflags == JAIL_CREATE) {
976 				mtx_unlock(&pr->pr_mtx);
977 				error = EEXIST;
978 				vfs_opterror(opts, "jail %d already exists",
979 				    jid);
980 				goto done_unlock_list;
981 			}
982 			if (!prison_ischild(mypr, pr)) {
983 				mtx_unlock(&pr->pr_mtx);
984 				pr = NULL;
985 			} else if (pr->pr_uref == 0) {
986 				if (!(flags & JAIL_DYING)) {
987 					mtx_unlock(&pr->pr_mtx);
988 					error = ENOENT;
989 					vfs_opterror(opts, "jail %d is dying",
990 					    jid);
991 					goto done_unlock_list;
992 				} else if ((flags & JAIL_ATTACH) ||
993 				    (pr_flags & PR_PERSIST)) {
994 					/*
995 					 * A dying jail might be resurrected
996 					 * (via attach or persist), but first
997 					 * it must determine if another jail
998 					 * has claimed its name.  Accomplish
999 					 * this by implicitly re-setting the
1000 					 * name.
1001 					 */
1002 					if (name == NULL)
1003 						name = prison_name(mypr, pr);
1004 				}
1005 			}
1006 		}
1007 		if (pr == NULL) {
1008 			/* Update: jid must exist. */
1009 			if (cuflags == JAIL_UPDATE) {
1010 				error = ENOENT;
1011 				vfs_opterror(opts, "jail %d not found", jid);
1012 				goto done_unlock_list;
1013 			}
1014 		}
1015 	}
1016 	/*
1017 	 * If the caller provided a name, look for a jail by that name.
1018 	 * This has different semantics for creates and updates keyed by jid
1019 	 * (where the name must not already exist in a different jail),
1020 	 * and updates keyed by the name itself (where the name must exist
1021 	 * because that is the jail being updated).
1022 	 */
1023 	if (name != NULL) {
1024 		namelc = strrchr(name, '.');
1025 		if (namelc == NULL)
1026 			namelc = name;
1027 		else {
1028 			/*
1029 			 * This is a hierarchical name.  Split it into the
1030 			 * parent and child names, and make sure the parent
1031 			 * exists or matches an already found jail.
1032 			 */
1033 			*namelc = '\0';
1034 			if (pr != NULL) {
1035 				if (strncmp(name, ppr->pr_name, namelc - name)
1036 				    || ppr->pr_name[namelc - name] != '\0') {
1037 					mtx_unlock(&pr->pr_mtx);
1038 					error = EINVAL;
1039 					vfs_opterror(opts,
1040 					    "cannot change jail's parent");
1041 					goto done_unlock_list;
1042 				}
1043 			} else {
1044 				ppr = prison_find_name(mypr, name);
1045 				if (ppr == NULL) {
1046 					error = ENOENT;
1047 					vfs_opterror(opts,
1048 					    "jail \"%s\" not found", name);
1049 					goto done_unlock_list;
1050 				}
1051 				mtx_unlock(&ppr->pr_mtx);
1052 			}
1053 			name = ++namelc;
1054 		}
1055 		if (name[0] != '\0') {
1056 			namelen =
1057 			    (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1058  name_again:
1059 			deadpr = NULL;
1060 			FOREACH_PRISON_CHILD(ppr, tpr) {
1061 				if (tpr != pr && tpr->pr_ref > 0 &&
1062 				    !strcmp(tpr->pr_name + namelen, name)) {
1063 					if (pr == NULL &&
1064 					    cuflags != JAIL_CREATE) {
1065 						mtx_lock(&tpr->pr_mtx);
1066 						if (tpr->pr_ref > 0) {
1067 							/*
1068 							 * Use this jail
1069 							 * for updates.
1070 							 */
1071 							if (tpr->pr_uref > 0) {
1072 								pr = tpr;
1073 								break;
1074 							}
1075 							deadpr = tpr;
1076 						}
1077 						mtx_unlock(&tpr->pr_mtx);
1078 					} else if (tpr->pr_uref > 0) {
1079 						/*
1080 						 * Create, or update(jid):
1081 						 * name must not exist in an
1082 						 * active sibling jail.
1083 						 */
1084 						error = EEXIST;
1085 						if (pr != NULL)
1086 							mtx_unlock(&pr->pr_mtx);
1087 						vfs_opterror(opts,
1088 						   "jail \"%s\" already exists",
1089 						   name);
1090 						goto done_unlock_list;
1091 					}
1092 				}
1093 			}
1094 			/* If no active jail is found, use a dying one. */
1095 			if (deadpr != NULL && pr == NULL) {
1096 				if (flags & JAIL_DYING) {
1097 					mtx_lock(&deadpr->pr_mtx);
1098 					if (deadpr->pr_ref == 0) {
1099 						mtx_unlock(&deadpr->pr_mtx);
1100 						goto name_again;
1101 					}
1102 					pr = deadpr;
1103 				} else if (cuflags == JAIL_UPDATE) {
1104 					error = ENOENT;
1105 					vfs_opterror(opts,
1106 					    "jail \"%s\" is dying", name);
1107 					goto done_unlock_list;
1108 				}
1109 			}
1110 			/* Update: name must exist if no jid. */
1111 			else if (cuflags == JAIL_UPDATE && pr == NULL) {
1112 				error = ENOENT;
1113 				vfs_opterror(opts, "jail \"%s\" not found",
1114 				    name);
1115 				goto done_unlock_list;
1116 			}
1117 		}
1118 	}
1119 	/* Update: must provide a jid or name. */
1120 	else if (cuflags == JAIL_UPDATE && pr == NULL) {
1121 		error = ENOENT;
1122 		vfs_opterror(opts, "update specified no jail");
1123 		goto done_unlock_list;
1124 	}
1125 
1126 	/* If there's no prison to update, create a new one and link it in. */
1127 	if (pr == NULL) {
1128 		for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1129 			if (tpr->pr_childcount >= tpr->pr_childmax) {
1130 				error = EPERM;
1131 				vfs_opterror(opts, "prison limit exceeded");
1132 				goto done_unlock_list;
1133 			}
1134 		created = 1;
1135 		mtx_lock(&ppr->pr_mtx);
1136 		if (ppr->pr_ref == 0 || (ppr->pr_flags & PR_REMOVE)) {
1137 			mtx_unlock(&ppr->pr_mtx);
1138 			error = ENOENT;
1139 			vfs_opterror(opts, "parent jail went away!");
1140 			goto done_unlock_list;
1141 		}
1142 		ppr->pr_ref++;
1143 		ppr->pr_uref++;
1144 		mtx_unlock(&ppr->pr_mtx);
1145 		pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1146 		if (jid == 0) {
1147 			/* Find the next free jid. */
1148 			jid = lastprid + 1;
1149  findnext:
1150 			if (jid == JAIL_MAX)
1151 				jid = 1;
1152 			TAILQ_FOREACH(tpr, &allprison, pr_list) {
1153 				if (tpr->pr_id < jid)
1154 					continue;
1155 				if (tpr->pr_id > jid || tpr->pr_ref == 0) {
1156 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1157 					break;
1158 				}
1159 				if (jid == lastprid) {
1160 					error = EAGAIN;
1161 					vfs_opterror(opts,
1162 					    "no available jail IDs");
1163 					free(pr, M_PRISON);
1164 					prison_deref(ppr, PD_DEREF |
1165 					    PD_DEUREF | PD_LIST_XLOCKED);
1166 					goto done_releroot;
1167 				}
1168 				jid++;
1169 				goto findnext;
1170 			}
1171 			lastprid = jid;
1172 		} else {
1173 			/*
1174 			 * The jail already has a jid (that did not yet exist),
1175 			 * so just find where to insert it.
1176 			 */
1177 			TAILQ_FOREACH(tpr, &allprison, pr_list)
1178 				if (tpr->pr_id >= jid) {
1179 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1180 					break;
1181 				}
1182 		}
1183 		if (tpr == NULL)
1184 			TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1185 		LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1186 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1187 			tpr->pr_childcount++;
1188 
1189 		pr->pr_parent = ppr;
1190 		pr->pr_id = jid;
1191 
1192 		/* Set some default values, and inherit some from the parent. */
1193 		if (name == NULL)
1194 			name = "";
1195 		if (path == NULL) {
1196 			path = "/";
1197 			root = mypr->pr_root;
1198 			vref(root);
1199 		}
1200 		racct_create(&pr->pr_racct);
1201 		strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1202 		pr->pr_flags |= PR_HOST;
1203 #if defined(INET) || defined(INET6)
1204 #ifdef VIMAGE
1205 		if (!(pr_flags & PR_VNET))
1206 #endif
1207 		{
1208 #ifdef INET
1209 			if (!(ch_flags & PR_IP4_USER))
1210 				pr->pr_flags |=
1211 				    PR_IP4 | PR_IP4_USER | PR_IP4_DISABLE;
1212 			else if (!(pr_flags & PR_IP4_USER)) {
1213 				pr->pr_flags |= ppr->pr_flags & PR_IP4;
1214 				if (ppr->pr_ip4 != NULL) {
1215 					pr->pr_ip4s = ppr->pr_ip4s;
1216 					pr->pr_ip4 = malloc(pr->pr_ip4s *
1217 					    sizeof(struct in_addr), M_PRISON,
1218 					    M_WAITOK);
1219 					bcopy(ppr->pr_ip4, pr->pr_ip4,
1220 					    pr->pr_ip4s * sizeof(*pr->pr_ip4));
1221 				}
1222 			}
1223 #endif
1224 #ifdef INET6
1225 			if (!(ch_flags & PR_IP6_USER))
1226 				pr->pr_flags |=
1227 				    PR_IP6 | PR_IP6_USER | PR_IP6_DISABLE;
1228 			else if (!(pr_flags & PR_IP6_USER)) {
1229 				pr->pr_flags |= ppr->pr_flags & PR_IP6;
1230 				if (ppr->pr_ip6 != NULL) {
1231 					pr->pr_ip6s = ppr->pr_ip6s;
1232 					pr->pr_ip6 = malloc(pr->pr_ip6s *
1233 					    sizeof(struct in6_addr), M_PRISON,
1234 					    M_WAITOK);
1235 					bcopy(ppr->pr_ip6, pr->pr_ip6,
1236 					    pr->pr_ip6s * sizeof(*pr->pr_ip6));
1237 				}
1238 			}
1239 #endif
1240 		}
1241 #endif
1242 		/* Source address selection is always on by default. */
1243 		pr->pr_flags |= _PR_IP_SADDRSEL;
1244 
1245 		pr->pr_securelevel = ppr->pr_securelevel;
1246 		pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1247 		pr->pr_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
1248 
1249 		LIST_INIT(&pr->pr_children);
1250 		mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1251 
1252 #ifdef VIMAGE
1253 		/* Allocate a new vnet if specified. */
1254 		pr->pr_vnet = (pr_flags & PR_VNET)
1255 		    ? vnet_alloc() : ppr->pr_vnet;
1256 #endif
1257 		/*
1258 		 * Allocate a dedicated cpuset for each jail.
1259 		 * Unlike other initial settings, this may return an erorr.
1260 		 */
1261 		error = cpuset_create_root(ppr, &pr->pr_cpuset);
1262 		if (error) {
1263 			prison_deref(pr, PD_LIST_XLOCKED);
1264 			goto done_releroot;
1265 		}
1266 
1267 		mtx_lock(&pr->pr_mtx);
1268 		/*
1269 		 * New prisons do not yet have a reference, because we do not
1270 		 * want other to see the incomplete prison once the
1271 		 * allprison_lock is downgraded.
1272 		 */
1273 	} else {
1274 		created = 0;
1275 		/*
1276 		 * Grab a reference for existing prisons, to ensure they
1277 		 * continue to exist for the duration of the call.
1278 		 */
1279 		pr->pr_ref++;
1280 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1281 		if ((pr->pr_flags & PR_VNET) &&
1282 		    (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1283 			error = EINVAL;
1284 			vfs_opterror(opts,
1285 			    "vnet jails cannot have IP address restrictions");
1286 			goto done_deref_locked;
1287 		}
1288 #endif
1289 #ifdef INET
1290 		if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1291 			error = EINVAL;
1292 			vfs_opterror(opts,
1293 			    "ip4 cannot be changed after creation");
1294 			goto done_deref_locked;
1295 		}
1296 #endif
1297 #ifdef INET6
1298 		if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1299 			error = EINVAL;
1300 			vfs_opterror(opts,
1301 			    "ip6 cannot be changed after creation");
1302 			goto done_deref_locked;
1303 		}
1304 #endif
1305 	}
1306 
1307 	/* Do final error checking before setting anything. */
1308 	if (gotslevel) {
1309 		if (slevel < ppr->pr_securelevel) {
1310 			error = EPERM;
1311 			goto done_deref_locked;
1312 		}
1313 	}
1314 	if (gotchildmax) {
1315 		if (childmax >= ppr->pr_childmax) {
1316 			error = EPERM;
1317 			goto done_deref_locked;
1318 		}
1319 	}
1320 	if (gotenforce) {
1321 		if (enforce < ppr->pr_enforce_statfs) {
1322 			error = EPERM;
1323 			goto done_deref_locked;
1324 		}
1325 	}
1326 #ifdef INET
1327 	if (ip4s > 0) {
1328 		if (ppr->pr_flags & PR_IP4) {
1329 			/*
1330 			 * Make sure the new set of IP addresses is a
1331 			 * subset of the parent's list.  Don't worry
1332 			 * about the parent being unlocked, as any
1333 			 * setting is done with allprison_lock held.
1334 			 */
1335 			for (ij = 0; ij < ppr->pr_ip4s; ij++)
1336 				if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
1337 					break;
1338 			if (ij == ppr->pr_ip4s) {
1339 				error = EPERM;
1340 				goto done_deref_locked;
1341 			}
1342 			if (ip4s > 1) {
1343 				for (ii = ij = 1; ii < ip4s; ii++) {
1344 					if (ip4[ii].s_addr ==
1345 					    ppr->pr_ip4[0].s_addr)
1346 						continue;
1347 					for (; ij < ppr->pr_ip4s; ij++)
1348 						if (ip4[ii].s_addr ==
1349 						    ppr->pr_ip4[ij].s_addr)
1350 							break;
1351 					if (ij == ppr->pr_ip4s)
1352 						break;
1353 				}
1354 				if (ij == ppr->pr_ip4s) {
1355 					error = EPERM;
1356 					goto done_deref_locked;
1357 				}
1358 			}
1359 		}
1360 		/*
1361 		 * Check for conflicting IP addresses.  We permit them
1362 		 * if there is no more than one IP on each jail.  If
1363 		 * there is a duplicate on a jail with more than one
1364 		 * IP stop checking and return error.
1365 		 */
1366 		tppr = ppr;
1367 #ifdef VIMAGE
1368 		for (; tppr != &prison0; tppr = tppr->pr_parent)
1369 			if (tppr->pr_flags & PR_VNET)
1370 				break;
1371 #endif
1372 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1373 			if (tpr == pr ||
1374 #ifdef VIMAGE
1375 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1376 #endif
1377 			    tpr->pr_uref == 0) {
1378 				descend = 0;
1379 				continue;
1380 			}
1381 			if (!(tpr->pr_flags & PR_IP4_USER))
1382 				continue;
1383 			descend = 0;
1384 			if (tpr->pr_ip4 == NULL ||
1385 			    (ip4s == 1 && tpr->pr_ip4s == 1))
1386 				continue;
1387 			for (ii = 0; ii < ip4s; ii++) {
1388 				if (_prison_check_ip4(tpr, &ip4[ii]) == 0) {
1389 					error = EADDRINUSE;
1390 					vfs_opterror(opts,
1391 					    "IPv4 addresses clash");
1392 					goto done_deref_locked;
1393 				}
1394 			}
1395 		}
1396 	}
1397 #endif
1398 #ifdef INET6
1399 	if (ip6s > 0) {
1400 		if (ppr->pr_flags & PR_IP6) {
1401 			/*
1402 			 * Make sure the new set of IP addresses is a
1403 			 * subset of the parent's list.
1404 			 */
1405 			for (ij = 0; ij < ppr->pr_ip6s; ij++)
1406 				if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1407 				    &ppr->pr_ip6[ij]))
1408 					break;
1409 			if (ij == ppr->pr_ip6s) {
1410 				error = EPERM;
1411 				goto done_deref_locked;
1412 			}
1413 			if (ip6s > 1) {
1414 				for (ii = ij = 1; ii < ip6s; ii++) {
1415 					if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1416 					     &ppr->pr_ip6[0]))
1417 						continue;
1418 					for (; ij < ppr->pr_ip6s; ij++)
1419 						if (IN6_ARE_ADDR_EQUAL(
1420 						    &ip6[ii], &ppr->pr_ip6[ij]))
1421 							break;
1422 					if (ij == ppr->pr_ip6s)
1423 						break;
1424 				}
1425 				if (ij == ppr->pr_ip6s) {
1426 					error = EPERM;
1427 					goto done_deref_locked;
1428 				}
1429 			}
1430 		}
1431 		/* Check for conflicting IP addresses. */
1432 		tppr = ppr;
1433 #ifdef VIMAGE
1434 		for (; tppr != &prison0; tppr = tppr->pr_parent)
1435 			if (tppr->pr_flags & PR_VNET)
1436 				break;
1437 #endif
1438 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1439 			if (tpr == pr ||
1440 #ifdef VIMAGE
1441 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1442 #endif
1443 			    tpr->pr_uref == 0) {
1444 				descend = 0;
1445 				continue;
1446 			}
1447 			if (!(tpr->pr_flags & PR_IP6_USER))
1448 				continue;
1449 			descend = 0;
1450 			if (tpr->pr_ip6 == NULL ||
1451 			    (ip6s == 1 && tpr->pr_ip6s == 1))
1452 				continue;
1453 			for (ii = 0; ii < ip6s; ii++) {
1454 				if (_prison_check_ip6(tpr, &ip6[ii]) == 0) {
1455 					error = EADDRINUSE;
1456 					vfs_opterror(opts,
1457 					    "IPv6 addresses clash");
1458 					goto done_deref_locked;
1459 				}
1460 			}
1461 		}
1462 	}
1463 #endif
1464 	onamelen = namelen = 0;
1465 	if (name != NULL) {
1466 		/* Give a default name of the jid. */
1467 		if (name[0] == '\0')
1468 			snprintf(name = numbuf, sizeof(numbuf), "%d", jid);
1469 		else if (*namelc == '0' || (strtoul(namelc, &p, 10) != jid &&
1470 		    *p == '\0')) {
1471 			error = EINVAL;
1472 			vfs_opterror(opts,
1473 			    "name cannot be numeric (unless it is the jid)");
1474 			goto done_deref_locked;
1475 		}
1476 		/*
1477 		 * Make sure the name isn't too long for the prison or its
1478 		 * children.
1479 		 */
1480 		onamelen = strlen(pr->pr_name);
1481 		namelen = strlen(name);
1482 		if (strlen(ppr->pr_name) + namelen + 2 > sizeof(pr->pr_name)) {
1483 			error = ENAMETOOLONG;
1484 			goto done_deref_locked;
1485 		}
1486 		FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1487 			if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1488 			    sizeof(pr->pr_name)) {
1489 				error = ENAMETOOLONG;
1490 				goto done_deref_locked;
1491 			}
1492 		}
1493 	}
1494 	if (pr_allow & ~ppr->pr_allow) {
1495 		error = EPERM;
1496 		goto done_deref_locked;
1497 	}
1498 
1499 	/* Set the parameters of the prison. */
1500 #ifdef INET
1501 	redo_ip4 = 0;
1502 	if (pr_flags & PR_IP4_USER) {
1503 		pr->pr_flags |= PR_IP4;
1504 		free(pr->pr_ip4, M_PRISON);
1505 		pr->pr_ip4s = ip4s;
1506 		pr->pr_ip4 = ip4;
1507 		ip4 = NULL;
1508 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1509 #ifdef VIMAGE
1510 			if (tpr->pr_flags & PR_VNET) {
1511 				descend = 0;
1512 				continue;
1513 			}
1514 #endif
1515 			if (prison_restrict_ip4(tpr, NULL)) {
1516 				redo_ip4 = 1;
1517 				descend = 0;
1518 			}
1519 		}
1520 	}
1521 #endif
1522 #ifdef INET6
1523 	redo_ip6 = 0;
1524 	if (pr_flags & PR_IP6_USER) {
1525 		pr->pr_flags |= PR_IP6;
1526 		free(pr->pr_ip6, M_PRISON);
1527 		pr->pr_ip6s = ip6s;
1528 		pr->pr_ip6 = ip6;
1529 		ip6 = NULL;
1530 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1531 #ifdef VIMAGE
1532 			if (tpr->pr_flags & PR_VNET) {
1533 				descend = 0;
1534 				continue;
1535 			}
1536 #endif
1537 			if (prison_restrict_ip6(tpr, NULL)) {
1538 				redo_ip6 = 1;
1539 				descend = 0;
1540 			}
1541 		}
1542 	}
1543 #endif
1544 	if (gotslevel) {
1545 		pr->pr_securelevel = slevel;
1546 		/* Set all child jails to be at least this level. */
1547 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1548 			if (tpr->pr_securelevel < slevel)
1549 				tpr->pr_securelevel = slevel;
1550 	}
1551 	if (gotchildmax) {
1552 		pr->pr_childmax = childmax;
1553 		/* Set all child jails to under this limit. */
1554 		FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1555 			if (tpr->pr_childmax > childmax - level)
1556 				tpr->pr_childmax = childmax > level
1557 				    ? childmax - level : 0;
1558 	}
1559 	if (gotenforce) {
1560 		pr->pr_enforce_statfs = enforce;
1561 		/* Pass this restriction on to the children. */
1562 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1563 			if (tpr->pr_enforce_statfs < enforce)
1564 				tpr->pr_enforce_statfs = enforce;
1565 	}
1566 	if (name != NULL) {
1567 		if (ppr == &prison0)
1568 			strlcpy(pr->pr_name, name, sizeof(pr->pr_name));
1569 		else
1570 			snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1571 			    ppr->pr_name, name);
1572 		/* Change this component of child names. */
1573 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1574 			bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1575 			    strlen(tpr->pr_name + onamelen) + 1);
1576 			bcopy(pr->pr_name, tpr->pr_name, namelen);
1577 		}
1578 	}
1579 	if (path != NULL) {
1580 		/* Try to keep a real-rooted full pathname. */
1581 		if (path[0] == '/' && strcmp(mypr->pr_path, "/"))
1582 			snprintf(pr->pr_path, sizeof(pr->pr_path), "%s%s",
1583 			    mypr->pr_path, path);
1584 		else
1585 			strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1586 		pr->pr_root = root;
1587 	}
1588 	if (PR_HOST & ch_flags & ~pr_flags) {
1589 		if (pr->pr_flags & PR_HOST) {
1590 			/*
1591 			 * Copy the parent's host info.  As with pr_ip4 above,
1592 			 * the lack of a lock on the parent is not a problem;
1593 			 * it is always set with allprison_lock at least
1594 			 * shared, and is held exclusively here.
1595 			 */
1596 			strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1597 			    sizeof(pr->pr_hostname));
1598 			strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1599 			    sizeof(pr->pr_domainname));
1600 			strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1601 			    sizeof(pr->pr_hostuuid));
1602 			pr->pr_hostid = pr->pr_parent->pr_hostid;
1603 		}
1604 	} else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1605 		/* Set this prison, and any descendants without PR_HOST. */
1606 		if (host != NULL)
1607 			strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1608 		if (domain != NULL)
1609 			strlcpy(pr->pr_domainname, domain,
1610 			    sizeof(pr->pr_domainname));
1611 		if (uuid != NULL)
1612 			strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1613 		if (gothid)
1614 			pr->pr_hostid = hid;
1615 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1616 			if (tpr->pr_flags & PR_HOST)
1617 				descend = 0;
1618 			else {
1619 				if (host != NULL)
1620 					strlcpy(tpr->pr_hostname,
1621 					    pr->pr_hostname,
1622 					    sizeof(tpr->pr_hostname));
1623 				if (domain != NULL)
1624 					strlcpy(tpr->pr_domainname,
1625 					    pr->pr_domainname,
1626 					    sizeof(tpr->pr_domainname));
1627 				if (uuid != NULL)
1628 					strlcpy(tpr->pr_hostuuid,
1629 					    pr->pr_hostuuid,
1630 					    sizeof(tpr->pr_hostuuid));
1631 				if (gothid)
1632 					tpr->pr_hostid = hid;
1633 			}
1634 		}
1635 	}
1636 	if ((tallow = ch_allow & ~pr_allow)) {
1637 		/* Clear allow bits in all children. */
1638 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1639 			tpr->pr_allow &= ~tallow;
1640 	}
1641 	pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1642 	/*
1643 	 * Persistent prisons get an extra reference, and prisons losing their
1644 	 * persist flag lose that reference.  Only do this for existing prisons
1645 	 * for now, so new ones will remain unseen until after the module
1646 	 * handlers have completed.
1647 	 */
1648 	if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) {
1649 		if (pr_flags & PR_PERSIST) {
1650 			pr->pr_ref++;
1651 			pr->pr_uref++;
1652 		} else {
1653 			pr->pr_ref--;
1654 			pr->pr_uref--;
1655 		}
1656 	}
1657 	pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1658 	mtx_unlock(&pr->pr_mtx);
1659 
1660 	/* Locks may have prevented a complete restriction of child IP
1661 	 * addresses.  If so, allocate some more memory and try again.
1662 	 */
1663 #ifdef INET
1664 	while (redo_ip4) {
1665 		ip4s = pr->pr_ip4s;
1666 		ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1667 		mtx_lock(&pr->pr_mtx);
1668 		redo_ip4 = 0;
1669 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1670 #ifdef VIMAGE
1671 			if (tpr->pr_flags & PR_VNET) {
1672 				descend = 0;
1673 				continue;
1674 			}
1675 #endif
1676 			if (prison_restrict_ip4(tpr, ip4)) {
1677 				if (ip4 != NULL)
1678 					ip4 = NULL;
1679 				else
1680 					redo_ip4 = 1;
1681 			}
1682 		}
1683 		mtx_unlock(&pr->pr_mtx);
1684 	}
1685 #endif
1686 #ifdef INET6
1687 	while (redo_ip6) {
1688 		ip6s = pr->pr_ip6s;
1689 		ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1690 		mtx_lock(&pr->pr_mtx);
1691 		redo_ip6 = 0;
1692 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1693 #ifdef VIMAGE
1694 			if (tpr->pr_flags & PR_VNET) {
1695 				descend = 0;
1696 				continue;
1697 			}
1698 #endif
1699 			if (prison_restrict_ip6(tpr, ip6)) {
1700 				if (ip6 != NULL)
1701 					ip6 = NULL;
1702 				else
1703 					redo_ip6 = 1;
1704 			}
1705 		}
1706 		mtx_unlock(&pr->pr_mtx);
1707 	}
1708 #endif
1709 
1710 	/* Let the modules do their work. */
1711 	sx_downgrade(&allprison_lock);
1712 	if (created) {
1713 		error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1714 		if (error) {
1715 			prison_deref(pr, PD_LIST_SLOCKED);
1716 			goto done_errmsg;
1717 		}
1718 	}
1719 	error = osd_jail_call(pr, PR_METHOD_SET, opts);
1720 	if (error) {
1721 		prison_deref(pr, created
1722 		    ? PD_LIST_SLOCKED
1723 		    : PD_DEREF | PD_LIST_SLOCKED);
1724 		goto done_errmsg;
1725 	}
1726 
1727 	/* Attach this process to the prison if requested. */
1728 	if (flags & JAIL_ATTACH) {
1729 		mtx_lock(&pr->pr_mtx);
1730 		error = do_jail_attach(td, pr);
1731 		if (error) {
1732 			vfs_opterror(opts, "attach failed");
1733 			if (!created)
1734 				prison_deref(pr, PD_DEREF);
1735 			goto done_errmsg;
1736 		}
1737 	}
1738 
1739 	/*
1740 	 * Now that it is all there, drop the temporary reference from existing
1741 	 * prisons.  Or add a reference to newly created persistent prisons
1742 	 * (which was not done earlier so that the prison would not be publicly
1743 	 * visible).
1744 	 */
1745 	if (!created) {
1746 		prison_deref(pr, (flags & JAIL_ATTACH)
1747 		    ? PD_DEREF
1748 		    : PD_DEREF | PD_LIST_SLOCKED);
1749 	} else {
1750 		if (pr_flags & PR_PERSIST) {
1751 			mtx_lock(&pr->pr_mtx);
1752 			pr->pr_ref++;
1753 			pr->pr_uref++;
1754 			mtx_unlock(&pr->pr_mtx);
1755 		}
1756 		if (!(flags & JAIL_ATTACH))
1757 			sx_sunlock(&allprison_lock);
1758 	}
1759 	td->td_retval[0] = pr->pr_id;
1760 	goto done_errmsg;
1761 
1762  done_deref_locked:
1763 	prison_deref(pr, created
1764 	    ? PD_LOCKED | PD_LIST_XLOCKED
1765 	    : PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
1766 	goto done_releroot;
1767  done_unlock_list:
1768 	sx_xunlock(&allprison_lock);
1769  done_releroot:
1770 	if (root != NULL) {
1771 		vfslocked = VFS_LOCK_GIANT(root->v_mount);
1772 		vrele(root);
1773 		VFS_UNLOCK_GIANT(vfslocked);
1774 	}
1775  done_errmsg:
1776 	if (error) {
1777 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
1778 		if (errmsg_len > 0) {
1779 			errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1780 			if (errmsg_pos > 0) {
1781 				if (optuio->uio_segflg == UIO_SYSSPACE)
1782 					bcopy(errmsg,
1783 					   optuio->uio_iov[errmsg_pos].iov_base,
1784 					   errmsg_len);
1785 				else
1786 					copyout(errmsg,
1787 					   optuio->uio_iov[errmsg_pos].iov_base,
1788 					   errmsg_len);
1789 			}
1790 		}
1791 	}
1792  done_free:
1793 #ifdef INET
1794 	free(ip4, M_PRISON);
1795 #endif
1796 #ifdef INET6
1797 	free(ip6, M_PRISON);
1798 #endif
1799 	vfs_freeopts(opts);
1800 	return (error);
1801 }
1802 
1803 
1804 /*
1805  * struct jail_get_args {
1806  *	struct iovec *iovp;
1807  *	unsigned int iovcnt;
1808  *	int flags;
1809  * };
1810  */
1811 int
1812 jail_get(struct thread *td, struct jail_get_args *uap)
1813 {
1814 	struct uio *auio;
1815 	int error;
1816 
1817 	/* Check that we have an even number of iovecs. */
1818 	if (uap->iovcnt & 1)
1819 		return (EINVAL);
1820 
1821 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
1822 	if (error)
1823 		return (error);
1824 	error = kern_jail_get(td, auio, uap->flags);
1825 	if (error == 0)
1826 		error = copyout(auio->uio_iov, uap->iovp,
1827 		    uap->iovcnt * sizeof (struct iovec));
1828 	free(auio, M_IOV);
1829 	return (error);
1830 }
1831 
1832 int
1833 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
1834 {
1835 	struct prison *pr, *mypr;
1836 	struct vfsopt *opt;
1837 	struct vfsoptlist *opts;
1838 	char *errmsg, *name;
1839 	int error, errmsg_len, errmsg_pos, fi, i, jid, len, locked, pos;
1840 
1841 	if (flags & ~JAIL_GET_MASK)
1842 		return (EINVAL);
1843 
1844 	/* Get the parameter list. */
1845 	error = vfs_buildopts(optuio, &opts);
1846 	if (error)
1847 		return (error);
1848 	errmsg_pos = vfs_getopt_pos(opts, "errmsg");
1849 	mypr = td->td_ucred->cr_prison;
1850 
1851 	/*
1852 	 * Find the prison specified by one of: lastjid, jid, name.
1853 	 */
1854 	sx_slock(&allprison_lock);
1855 	error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
1856 	if (error == 0) {
1857 		TAILQ_FOREACH(pr, &allprison, pr_list) {
1858 			if (pr->pr_id > jid && prison_ischild(mypr, pr)) {
1859 				mtx_lock(&pr->pr_mtx);
1860 				if (pr->pr_ref > 0 &&
1861 				    (pr->pr_uref > 0 || (flags & JAIL_DYING)))
1862 					break;
1863 				mtx_unlock(&pr->pr_mtx);
1864 			}
1865 		}
1866 		if (pr != NULL)
1867 			goto found_prison;
1868 		error = ENOENT;
1869 		vfs_opterror(opts, "no jail after %d", jid);
1870 		goto done_unlock_list;
1871 	} else if (error != ENOENT)
1872 		goto done_unlock_list;
1873 
1874 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1875 	if (error == 0) {
1876 		if (jid != 0) {
1877 			pr = prison_find_child(mypr, jid);
1878 			if (pr != NULL) {
1879 				if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1880 					mtx_unlock(&pr->pr_mtx);
1881 					error = ENOENT;
1882 					vfs_opterror(opts, "jail %d is dying",
1883 					    jid);
1884 					goto done_unlock_list;
1885 				}
1886 				goto found_prison;
1887 			}
1888 			error = ENOENT;
1889 			vfs_opterror(opts, "jail %d not found", jid);
1890 			goto done_unlock_list;
1891 		}
1892 	} else if (error != ENOENT)
1893 		goto done_unlock_list;
1894 
1895 	error = vfs_getopt(opts, "name", (void **)&name, &len);
1896 	if (error == 0) {
1897 		if (len == 0 || name[len - 1] != '\0') {
1898 			error = EINVAL;
1899 			goto done_unlock_list;
1900 		}
1901 		pr = prison_find_name(mypr, name);
1902 		if (pr != NULL) {
1903 			if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1904 				mtx_unlock(&pr->pr_mtx);
1905 				error = ENOENT;
1906 				vfs_opterror(opts, "jail \"%s\" is dying",
1907 				    name);
1908 				goto done_unlock_list;
1909 			}
1910 			goto found_prison;
1911 		}
1912 		error = ENOENT;
1913 		vfs_opterror(opts, "jail \"%s\" not found", name);
1914 		goto done_unlock_list;
1915 	} else if (error != ENOENT)
1916 		goto done_unlock_list;
1917 
1918 	vfs_opterror(opts, "no jail specified");
1919 	error = ENOENT;
1920 	goto done_unlock_list;
1921 
1922  found_prison:
1923 	/* Get the parameters of the prison. */
1924 	pr->pr_ref++;
1925 	locked = PD_LOCKED;
1926 	td->td_retval[0] = pr->pr_id;
1927 	error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
1928 	if (error != 0 && error != ENOENT)
1929 		goto done_deref;
1930 	i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
1931 	error = vfs_setopt(opts, "parent", &i, sizeof(i));
1932 	if (error != 0 && error != ENOENT)
1933 		goto done_deref;
1934 	error = vfs_setopts(opts, "name", prison_name(mypr, pr));
1935 	if (error != 0 && error != ENOENT)
1936 		goto done_deref;
1937 	error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
1938 	    sizeof(pr->pr_cpuset->cs_id));
1939 	if (error != 0 && error != ENOENT)
1940 		goto done_deref;
1941 	error = vfs_setopts(opts, "path", prison_path(mypr, pr));
1942 	if (error != 0 && error != ENOENT)
1943 		goto done_deref;
1944 #ifdef INET
1945 	error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
1946 	    pr->pr_ip4s * sizeof(*pr->pr_ip4));
1947 	if (error != 0 && error != ENOENT)
1948 		goto done_deref;
1949 #endif
1950 #ifdef INET6
1951 	error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
1952 	    pr->pr_ip6s * sizeof(*pr->pr_ip6));
1953 	if (error != 0 && error != ENOENT)
1954 		goto done_deref;
1955 #endif
1956 	error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
1957 	    sizeof(pr->pr_securelevel));
1958 	if (error != 0 && error != ENOENT)
1959 		goto done_deref;
1960 	error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
1961 	    sizeof(pr->pr_childcount));
1962 	if (error != 0 && error != ENOENT)
1963 		goto done_deref;
1964 	error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
1965 	    sizeof(pr->pr_childmax));
1966 	if (error != 0 && error != ENOENT)
1967 		goto done_deref;
1968 	error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
1969 	if (error != 0 && error != ENOENT)
1970 		goto done_deref;
1971 	error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
1972 	if (error != 0 && error != ENOENT)
1973 		goto done_deref;
1974 	error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
1975 	if (error != 0 && error != ENOENT)
1976 		goto done_deref;
1977 #ifdef COMPAT_FREEBSD32
1978 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
1979 		uint32_t hid32 = pr->pr_hostid;
1980 
1981 		error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
1982 	} else
1983 #endif
1984 	error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
1985 	    sizeof(pr->pr_hostid));
1986 	if (error != 0 && error != ENOENT)
1987 		goto done_deref;
1988 	error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
1989 	    sizeof(pr->pr_enforce_statfs));
1990 	if (error != 0 && error != ENOENT)
1991 		goto done_deref;
1992 	for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
1993 	    fi++) {
1994 		if (pr_flag_names[fi] == NULL)
1995 			continue;
1996 		i = (pr->pr_flags & (1 << fi)) ? 1 : 0;
1997 		error = vfs_setopt(opts, pr_flag_names[fi], &i, sizeof(i));
1998 		if (error != 0 && error != ENOENT)
1999 			goto done_deref;
2000 		i = !i;
2001 		error = vfs_setopt(opts, pr_flag_nonames[fi], &i, sizeof(i));
2002 		if (error != 0 && error != ENOENT)
2003 			goto done_deref;
2004 	}
2005 	for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
2006 	    fi++) {
2007 		i = pr->pr_flags &
2008 		    (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
2009 		i = pr_flag_jailsys[fi].disable &&
2010 		      (i == pr_flag_jailsys[fi].disable) ? JAIL_SYS_DISABLE
2011 		    : (i == pr_flag_jailsys[fi].new) ? JAIL_SYS_NEW
2012 		    : JAIL_SYS_INHERIT;
2013 		error =
2014 		    vfs_setopt(opts, pr_flag_jailsys[fi].name, &i, sizeof(i));
2015 		if (error != 0 && error != ENOENT)
2016 			goto done_deref;
2017 	}
2018 	for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
2019 	    fi++) {
2020 		if (pr_allow_names[fi] == NULL)
2021 			continue;
2022 		i = (pr->pr_allow & (1 << fi)) ? 1 : 0;
2023 		error = vfs_setopt(opts, pr_allow_names[fi], &i, sizeof(i));
2024 		if (error != 0 && error != ENOENT)
2025 			goto done_deref;
2026 		i = !i;
2027 		error = vfs_setopt(opts, pr_allow_nonames[fi], &i, sizeof(i));
2028 		if (error != 0 && error != ENOENT)
2029 			goto done_deref;
2030 	}
2031 	i = (pr->pr_uref == 0);
2032 	error = vfs_setopt(opts, "dying", &i, sizeof(i));
2033 	if (error != 0 && error != ENOENT)
2034 		goto done_deref;
2035 	i = !i;
2036 	error = vfs_setopt(opts, "nodying", &i, sizeof(i));
2037 	if (error != 0 && error != ENOENT)
2038 		goto done_deref;
2039 
2040 	/* Get the module parameters. */
2041 	mtx_unlock(&pr->pr_mtx);
2042 	locked = 0;
2043 	error = osd_jail_call(pr, PR_METHOD_GET, opts);
2044 	if (error)
2045 		goto done_deref;
2046 	prison_deref(pr, PD_DEREF | PD_LIST_SLOCKED);
2047 
2048 	/* By now, all parameters should have been noted. */
2049 	TAILQ_FOREACH(opt, opts, link) {
2050 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
2051 			error = EINVAL;
2052 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
2053 			goto done_errmsg;
2054 		}
2055 	}
2056 
2057 	/* Write the fetched parameters back to userspace. */
2058 	error = 0;
2059 	TAILQ_FOREACH(opt, opts, link) {
2060 		if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2061 			pos = 2 * opt->pos + 1;
2062 			optuio->uio_iov[pos].iov_len = opt->len;
2063 			if (opt->value != NULL) {
2064 				if (optuio->uio_segflg == UIO_SYSSPACE) {
2065 					bcopy(opt->value,
2066 					    optuio->uio_iov[pos].iov_base,
2067 					    opt->len);
2068 				} else {
2069 					error = copyout(opt->value,
2070 					    optuio->uio_iov[pos].iov_base,
2071 					    opt->len);
2072 					if (error)
2073 						break;
2074 				}
2075 			}
2076 		}
2077 	}
2078 	goto done_errmsg;
2079 
2080  done_deref:
2081 	prison_deref(pr, locked | PD_DEREF | PD_LIST_SLOCKED);
2082 	goto done_errmsg;
2083 
2084  done_unlock_list:
2085 	sx_sunlock(&allprison_lock);
2086  done_errmsg:
2087 	if (error && errmsg_pos >= 0) {
2088 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2089 		errmsg_pos = 2 * errmsg_pos + 1;
2090 		if (errmsg_len > 0) {
2091 			if (optuio->uio_segflg == UIO_SYSSPACE)
2092 				bcopy(errmsg,
2093 				    optuio->uio_iov[errmsg_pos].iov_base,
2094 				    errmsg_len);
2095 			else
2096 				copyout(errmsg,
2097 				    optuio->uio_iov[errmsg_pos].iov_base,
2098 				    errmsg_len);
2099 		}
2100 	}
2101 	vfs_freeopts(opts);
2102 	return (error);
2103 }
2104 
2105 
2106 /*
2107  * struct jail_remove_args {
2108  *	int jid;
2109  * };
2110  */
2111 int
2112 jail_remove(struct thread *td, struct jail_remove_args *uap)
2113 {
2114 	struct prison *pr, *cpr, *lpr, *tpr;
2115 	int descend, error;
2116 
2117 	error = priv_check(td, PRIV_JAIL_REMOVE);
2118 	if (error)
2119 		return (error);
2120 
2121 	sx_xlock(&allprison_lock);
2122 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2123 	if (pr == NULL) {
2124 		sx_xunlock(&allprison_lock);
2125 		return (EINVAL);
2126 	}
2127 
2128 	/* Remove all descendants of this prison, then remove this prison. */
2129 	pr->pr_ref++;
2130 	pr->pr_flags |= PR_REMOVE;
2131 	if (!LIST_EMPTY(&pr->pr_children)) {
2132 		mtx_unlock(&pr->pr_mtx);
2133 		lpr = NULL;
2134 		FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
2135 			mtx_lock(&cpr->pr_mtx);
2136 			if (cpr->pr_ref > 0) {
2137 				tpr = cpr;
2138 				cpr->pr_ref++;
2139 				cpr->pr_flags |= PR_REMOVE;
2140 			} else {
2141 				/* Already removed - do not do it again. */
2142 				tpr = NULL;
2143 			}
2144 			mtx_unlock(&cpr->pr_mtx);
2145 			if (lpr != NULL) {
2146 				mtx_lock(&lpr->pr_mtx);
2147 				prison_remove_one(lpr);
2148 				sx_xlock(&allprison_lock);
2149 			}
2150 			lpr = tpr;
2151 		}
2152 		if (lpr != NULL) {
2153 			mtx_lock(&lpr->pr_mtx);
2154 			prison_remove_one(lpr);
2155 			sx_xlock(&allprison_lock);
2156 		}
2157 		mtx_lock(&pr->pr_mtx);
2158 	}
2159 	prison_remove_one(pr);
2160 	return (0);
2161 }
2162 
2163 static void
2164 prison_remove_one(struct prison *pr)
2165 {
2166 	struct proc *p;
2167 	int deuref;
2168 
2169 	/* If the prison was persistent, it is not anymore. */
2170 	deuref = 0;
2171 	if (pr->pr_flags & PR_PERSIST) {
2172 		pr->pr_ref--;
2173 		deuref = PD_DEUREF;
2174 		pr->pr_flags &= ~PR_PERSIST;
2175 	}
2176 
2177 	/*
2178 	 * jail_remove added a reference.  If that's the only one, remove
2179 	 * the prison now.
2180 	 */
2181 	KASSERT(pr->pr_ref > 0,
2182 	    ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id));
2183 	if (pr->pr_ref == 1) {
2184 		prison_deref(pr,
2185 		    deuref | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
2186 		return;
2187 	}
2188 
2189 	mtx_unlock(&pr->pr_mtx);
2190 	sx_xunlock(&allprison_lock);
2191 	/*
2192 	 * Kill all processes unfortunate enough to be attached to this prison.
2193 	 */
2194 	sx_slock(&allproc_lock);
2195 	LIST_FOREACH(p, &allproc, p_list) {
2196 		PROC_LOCK(p);
2197 		if (p->p_state != PRS_NEW && p->p_ucred &&
2198 		    p->p_ucred->cr_prison == pr)
2199 			psignal(p, SIGKILL);
2200 		PROC_UNLOCK(p);
2201 	}
2202 	sx_sunlock(&allproc_lock);
2203 	/* Remove the temporary reference added by jail_remove. */
2204 	prison_deref(pr, deuref | PD_DEREF);
2205 }
2206 
2207 
2208 /*
2209  * struct jail_attach_args {
2210  *	int jid;
2211  * };
2212  */
2213 int
2214 jail_attach(struct thread *td, struct jail_attach_args *uap)
2215 {
2216 	struct prison *pr;
2217 	int error;
2218 
2219 	error = priv_check(td, PRIV_JAIL_ATTACH);
2220 	if (error)
2221 		return (error);
2222 
2223 	sx_slock(&allprison_lock);
2224 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2225 	if (pr == NULL) {
2226 		sx_sunlock(&allprison_lock);
2227 		return (EINVAL);
2228 	}
2229 
2230 	/*
2231 	 * Do not allow a process to attach to a prison that is not
2232 	 * considered to be "alive".
2233 	 */
2234 	if (pr->pr_uref == 0) {
2235 		mtx_unlock(&pr->pr_mtx);
2236 		sx_sunlock(&allprison_lock);
2237 		return (EINVAL);
2238 	}
2239 
2240 	return (do_jail_attach(td, pr));
2241 }
2242 
2243 static int
2244 do_jail_attach(struct thread *td, struct prison *pr)
2245 {
2246 	struct prison *ppr;
2247 	struct proc *p;
2248 	struct ucred *newcred, *oldcred;
2249 	int vfslocked, error;
2250 
2251 	/*
2252 	 * XXX: Note that there is a slight race here if two threads
2253 	 * in the same privileged process attempt to attach to two
2254 	 * different jails at the same time.  It is important for
2255 	 * user processes not to do this, or they might end up with
2256 	 * a process root from one prison, but attached to the jail
2257 	 * of another.
2258 	 */
2259 	pr->pr_ref++;
2260 	pr->pr_uref++;
2261 	mtx_unlock(&pr->pr_mtx);
2262 
2263 	/* Let modules do whatever they need to prepare for attaching. */
2264 	error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2265 	if (error) {
2266 		prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2267 		return (error);
2268 	}
2269 	sx_sunlock(&allprison_lock);
2270 
2271 	/*
2272 	 * Reparent the newly attached process to this jail.
2273 	 */
2274 	ppr = td->td_ucred->cr_prison;
2275 	p = td->td_proc;
2276 	error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2277 	if (error)
2278 		goto e_revert_osd;
2279 
2280 	vfslocked = VFS_LOCK_GIANT(pr->pr_root->v_mount);
2281 	vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2282 	if ((error = change_dir(pr->pr_root, td)) != 0)
2283 		goto e_unlock;
2284 #ifdef MAC
2285 	if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2286 		goto e_unlock;
2287 #endif
2288 	VOP_UNLOCK(pr->pr_root, 0);
2289 	if ((error = change_root(pr->pr_root, td)))
2290 		goto e_unlock_giant;
2291 	VFS_UNLOCK_GIANT(vfslocked);
2292 
2293 	newcred = crget();
2294 	PROC_LOCK(p);
2295 	oldcred = p->p_ucred;
2296 	setsugid(p);
2297 	crcopy(newcred, oldcred);
2298 	newcred->cr_prison = pr;
2299 	p->p_ucred = newcred;
2300 	PROC_UNLOCK(p);
2301 #ifdef RACCT
2302 	racct_proc_ucred_changed(p, oldcred, newcred);
2303 #endif
2304 	crfree(oldcred);
2305 	prison_deref(ppr, PD_DEREF | PD_DEUREF);
2306 	return (0);
2307  e_unlock:
2308 	VOP_UNLOCK(pr->pr_root, 0);
2309  e_unlock_giant:
2310 	VFS_UNLOCK_GIANT(vfslocked);
2311  e_revert_osd:
2312 	/* Tell modules this thread is still in its old jail after all. */
2313 	(void)osd_jail_call(ppr, PR_METHOD_ATTACH, td);
2314 	prison_deref(pr, PD_DEREF | PD_DEUREF);
2315 	return (error);
2316 }
2317 
2318 
2319 /*
2320  * Returns a locked prison instance, or NULL on failure.
2321  */
2322 struct prison *
2323 prison_find(int prid)
2324 {
2325 	struct prison *pr;
2326 
2327 	sx_assert(&allprison_lock, SX_LOCKED);
2328 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2329 		if (pr->pr_id == prid) {
2330 			mtx_lock(&pr->pr_mtx);
2331 			if (pr->pr_ref > 0)
2332 				return (pr);
2333 			mtx_unlock(&pr->pr_mtx);
2334 		}
2335 	}
2336 	return (NULL);
2337 }
2338 
2339 /*
2340  * Find a prison that is a descendant of mypr.  Returns a locked prison or NULL.
2341  */
2342 struct prison *
2343 prison_find_child(struct prison *mypr, int prid)
2344 {
2345 	struct prison *pr;
2346 	int descend;
2347 
2348 	sx_assert(&allprison_lock, SX_LOCKED);
2349 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2350 		if (pr->pr_id == prid) {
2351 			mtx_lock(&pr->pr_mtx);
2352 			if (pr->pr_ref > 0)
2353 				return (pr);
2354 			mtx_unlock(&pr->pr_mtx);
2355 		}
2356 	}
2357 	return (NULL);
2358 }
2359 
2360 /*
2361  * Look for the name relative to mypr.  Returns a locked prison or NULL.
2362  */
2363 struct prison *
2364 prison_find_name(struct prison *mypr, const char *name)
2365 {
2366 	struct prison *pr, *deadpr;
2367 	size_t mylen;
2368 	int descend;
2369 
2370 	sx_assert(&allprison_lock, SX_LOCKED);
2371 	mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2372  again:
2373 	deadpr = NULL;
2374 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2375 		if (!strcmp(pr->pr_name + mylen, name)) {
2376 			mtx_lock(&pr->pr_mtx);
2377 			if (pr->pr_ref > 0) {
2378 				if (pr->pr_uref > 0)
2379 					return (pr);
2380 				deadpr = pr;
2381 			}
2382 			mtx_unlock(&pr->pr_mtx);
2383 		}
2384 	}
2385 	/* There was no valid prison - perhaps there was a dying one. */
2386 	if (deadpr != NULL) {
2387 		mtx_lock(&deadpr->pr_mtx);
2388 		if (deadpr->pr_ref == 0) {
2389 			mtx_unlock(&deadpr->pr_mtx);
2390 			goto again;
2391 		}
2392 	}
2393 	return (deadpr);
2394 }
2395 
2396 /*
2397  * See if a prison has the specific flag set.
2398  */
2399 int
2400 prison_flag(struct ucred *cred, unsigned flag)
2401 {
2402 
2403 	/* This is an atomic read, so no locking is necessary. */
2404 	return (cred->cr_prison->pr_flags & flag);
2405 }
2406 
2407 int
2408 prison_allow(struct ucred *cred, unsigned flag)
2409 {
2410 
2411 	/* This is an atomic read, so no locking is necessary. */
2412 	return (cred->cr_prison->pr_allow & flag);
2413 }
2414 
2415 /*
2416  * Remove a prison reference.  If that was the last reference, remove the
2417  * prison itself - but not in this context in case there are locks held.
2418  */
2419 void
2420 prison_free_locked(struct prison *pr)
2421 {
2422 
2423 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2424 	pr->pr_ref--;
2425 	if (pr->pr_ref == 0) {
2426 		mtx_unlock(&pr->pr_mtx);
2427 		TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
2428 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2429 		return;
2430 	}
2431 	mtx_unlock(&pr->pr_mtx);
2432 }
2433 
2434 void
2435 prison_free(struct prison *pr)
2436 {
2437 
2438 	mtx_lock(&pr->pr_mtx);
2439 	prison_free_locked(pr);
2440 }
2441 
2442 static void
2443 prison_complete(void *context, int pending)
2444 {
2445 
2446 	prison_deref((struct prison *)context, 0);
2447 }
2448 
2449 /*
2450  * Remove a prison reference (usually).  This internal version assumes no
2451  * mutexes are held, except perhaps the prison itself.  If there are no more
2452  * references, release and delist the prison.  On completion, the prison lock
2453  * and the allprison lock are both unlocked.
2454  */
2455 static void
2456 prison_deref(struct prison *pr, int flags)
2457 {
2458 	struct prison *ppr, *tpr;
2459 	int vfslocked;
2460 
2461 	if (!(flags & PD_LOCKED))
2462 		mtx_lock(&pr->pr_mtx);
2463 	/* Decrement the user references in a separate loop. */
2464 	if (flags & PD_DEUREF) {
2465 		for (tpr = pr;; tpr = tpr->pr_parent) {
2466 			if (tpr != pr)
2467 				mtx_lock(&tpr->pr_mtx);
2468 			if (--tpr->pr_uref > 0)
2469 				break;
2470 			KASSERT(tpr != &prison0, ("prison0 pr_uref=0"));
2471 			mtx_unlock(&tpr->pr_mtx);
2472 		}
2473 		/* Done if there were only user references to remove. */
2474 		if (!(flags & PD_DEREF)) {
2475 			mtx_unlock(&tpr->pr_mtx);
2476 			if (flags & PD_LIST_SLOCKED)
2477 				sx_sunlock(&allprison_lock);
2478 			else if (flags & PD_LIST_XLOCKED)
2479 				sx_xunlock(&allprison_lock);
2480 			return;
2481 		}
2482 		if (tpr != pr) {
2483 			mtx_unlock(&tpr->pr_mtx);
2484 			mtx_lock(&pr->pr_mtx);
2485 		}
2486 	}
2487 
2488 	for (;;) {
2489 		if (flags & PD_DEREF)
2490 			pr->pr_ref--;
2491 		/* If the prison still has references, nothing else to do. */
2492 		if (pr->pr_ref > 0) {
2493 			mtx_unlock(&pr->pr_mtx);
2494 			if (flags & PD_LIST_SLOCKED)
2495 				sx_sunlock(&allprison_lock);
2496 			else if (flags & PD_LIST_XLOCKED)
2497 				sx_xunlock(&allprison_lock);
2498 			return;
2499 		}
2500 
2501 		mtx_unlock(&pr->pr_mtx);
2502 		if (flags & PD_LIST_SLOCKED) {
2503 			if (!sx_try_upgrade(&allprison_lock)) {
2504 				sx_sunlock(&allprison_lock);
2505 				sx_xlock(&allprison_lock);
2506 			}
2507 		} else if (!(flags & PD_LIST_XLOCKED))
2508 			sx_xlock(&allprison_lock);
2509 
2510 		TAILQ_REMOVE(&allprison, pr, pr_list);
2511 		LIST_REMOVE(pr, pr_sibling);
2512 		ppr = pr->pr_parent;
2513 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
2514 			tpr->pr_childcount--;
2515 		sx_xunlock(&allprison_lock);
2516 
2517 #ifdef VIMAGE
2518 		if (pr->pr_vnet != ppr->pr_vnet)
2519 			vnet_destroy(pr->pr_vnet);
2520 #endif
2521 		if (pr->pr_root != NULL) {
2522 			vfslocked = VFS_LOCK_GIANT(pr->pr_root->v_mount);
2523 			vrele(pr->pr_root);
2524 			VFS_UNLOCK_GIANT(vfslocked);
2525 		}
2526 		mtx_destroy(&pr->pr_mtx);
2527 #ifdef INET
2528 		free(pr->pr_ip4, M_PRISON);
2529 #endif
2530 #ifdef INET6
2531 		free(pr->pr_ip6, M_PRISON);
2532 #endif
2533 		if (pr->pr_cpuset != NULL)
2534 			cpuset_rel(pr->pr_cpuset);
2535 		osd_jail_exit(pr);
2536 #ifdef RCTL
2537 		rctl_racct_release(pr->pr_racct);
2538 #endif
2539 		racct_destroy(&pr->pr_racct);
2540 		free(pr, M_PRISON);
2541 
2542 		/* Removing a prison frees a reference on its parent. */
2543 		pr = ppr;
2544 		mtx_lock(&pr->pr_mtx);
2545 		flags = PD_DEREF;
2546 	}
2547 }
2548 
2549 void
2550 prison_hold_locked(struct prison *pr)
2551 {
2552 
2553 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2554 	KASSERT(pr->pr_ref > 0,
2555 	    ("Trying to hold dead prison (jid=%d).", pr->pr_id));
2556 	pr->pr_ref++;
2557 }
2558 
2559 void
2560 prison_hold(struct prison *pr)
2561 {
2562 
2563 	mtx_lock(&pr->pr_mtx);
2564 	prison_hold_locked(pr);
2565 	mtx_unlock(&pr->pr_mtx);
2566 }
2567 
2568 void
2569 prison_proc_hold(struct prison *pr)
2570 {
2571 
2572 	mtx_lock(&pr->pr_mtx);
2573 	KASSERT(pr->pr_uref > 0,
2574 	    ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2575 	pr->pr_uref++;
2576 	mtx_unlock(&pr->pr_mtx);
2577 }
2578 
2579 void
2580 prison_proc_free(struct prison *pr)
2581 {
2582 
2583 	mtx_lock(&pr->pr_mtx);
2584 	KASSERT(pr->pr_uref > 0,
2585 	    ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2586 	prison_deref(pr, PD_DEUREF | PD_LOCKED);
2587 }
2588 
2589 
2590 #ifdef INET
2591 /*
2592  * Restrict a prison's IP address list with its parent's, possibly replacing
2593  * it.  Return true if the replacement buffer was used (or would have been).
2594  */
2595 static int
2596 prison_restrict_ip4(struct prison *pr, struct in_addr *newip4)
2597 {
2598 	int ii, ij, used;
2599 	struct prison *ppr;
2600 
2601 	ppr = pr->pr_parent;
2602 	if (!(pr->pr_flags & PR_IP4_USER)) {
2603 		/* This has no user settings, so just copy the parent's list. */
2604 		if (pr->pr_ip4s < ppr->pr_ip4s) {
2605 			/*
2606 			 * There's no room for the parent's list.  Use the
2607 			 * new list buffer, which is assumed to be big enough
2608 			 * (if it was passed).  If there's no buffer, try to
2609 			 * allocate one.
2610 			 */
2611 			used = 1;
2612 			if (newip4 == NULL) {
2613 				newip4 = malloc(ppr->pr_ip4s * sizeof(*newip4),
2614 				    M_PRISON, M_NOWAIT);
2615 				if (newip4 != NULL)
2616 					used = 0;
2617 			}
2618 			if (newip4 != NULL) {
2619 				bcopy(ppr->pr_ip4, newip4,
2620 				    ppr->pr_ip4s * sizeof(*newip4));
2621 				free(pr->pr_ip4, M_PRISON);
2622 				pr->pr_ip4 = newip4;
2623 				pr->pr_ip4s = ppr->pr_ip4s;
2624 			}
2625 			return (used);
2626 		}
2627 		pr->pr_ip4s = ppr->pr_ip4s;
2628 		if (pr->pr_ip4s > 0)
2629 			bcopy(ppr->pr_ip4, pr->pr_ip4,
2630 			    pr->pr_ip4s * sizeof(*newip4));
2631 		else if (pr->pr_ip4 != NULL) {
2632 			free(pr->pr_ip4, M_PRISON);
2633 			pr->pr_ip4 = NULL;
2634 		}
2635 	} else if (pr->pr_ip4s > 0) {
2636 		/* Remove addresses that aren't in the parent. */
2637 		for (ij = 0; ij < ppr->pr_ip4s; ij++)
2638 			if (pr->pr_ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
2639 				break;
2640 		if (ij < ppr->pr_ip4s)
2641 			ii = 1;
2642 		else {
2643 			bcopy(pr->pr_ip4 + 1, pr->pr_ip4,
2644 			    --pr->pr_ip4s * sizeof(*pr->pr_ip4));
2645 			ii = 0;
2646 		}
2647 		for (ij = 1; ii < pr->pr_ip4s; ) {
2648 			if (pr->pr_ip4[ii].s_addr == ppr->pr_ip4[0].s_addr) {
2649 				ii++;
2650 				continue;
2651 			}
2652 			switch (ij >= ppr->pr_ip4s ? -1 :
2653 				qcmp_v4(&pr->pr_ip4[ii], &ppr->pr_ip4[ij])) {
2654 			case -1:
2655 				bcopy(pr->pr_ip4 + ii + 1, pr->pr_ip4 + ii,
2656 				    (--pr->pr_ip4s - ii) * sizeof(*pr->pr_ip4));
2657 				break;
2658 			case 0:
2659 				ii++;
2660 				ij++;
2661 				break;
2662 			case 1:
2663 				ij++;
2664 				break;
2665 			}
2666 		}
2667 		if (pr->pr_ip4s == 0) {
2668 			pr->pr_flags |= PR_IP4_DISABLE;
2669 			free(pr->pr_ip4, M_PRISON);
2670 			pr->pr_ip4 = NULL;
2671 		}
2672 	}
2673 	return (0);
2674 }
2675 
2676 /*
2677  * Pass back primary IPv4 address of this jail.
2678  *
2679  * If not restricted return success but do not alter the address.  Caller has
2680  * to make sure to initialize it correctly (e.g. INADDR_ANY).
2681  *
2682  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv4.
2683  * Address returned in NBO.
2684  */
2685 int
2686 prison_get_ip4(struct ucred *cred, struct in_addr *ia)
2687 {
2688 	struct prison *pr;
2689 
2690 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2691 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2692 
2693 	pr = cred->cr_prison;
2694 	if (!(pr->pr_flags & PR_IP4))
2695 		return (0);
2696 	mtx_lock(&pr->pr_mtx);
2697 	if (!(pr->pr_flags & PR_IP4)) {
2698 		mtx_unlock(&pr->pr_mtx);
2699 		return (0);
2700 	}
2701 	if (pr->pr_ip4 == NULL) {
2702 		mtx_unlock(&pr->pr_mtx);
2703 		return (EAFNOSUPPORT);
2704 	}
2705 
2706 	ia->s_addr = pr->pr_ip4[0].s_addr;
2707 	mtx_unlock(&pr->pr_mtx);
2708 	return (0);
2709 }
2710 
2711 /*
2712  * Return 1 if we should do proper source address selection or are not jailed.
2713  * We will return 0 if we should bypass source address selection in favour
2714  * of the primary jail IPv4 address. Only in this case *ia will be updated and
2715  * returned in NBO.
2716  * Return EAFNOSUPPORT, in case this jail does not allow IPv4.
2717  */
2718 int
2719 prison_saddrsel_ip4(struct ucred *cred, struct in_addr *ia)
2720 {
2721 	struct prison *pr;
2722 	struct in_addr lia;
2723 	int error;
2724 
2725 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2726 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2727 
2728 	if (!jailed(cred))
2729 		return (1);
2730 
2731 	pr = cred->cr_prison;
2732 	if (pr->pr_flags & PR_IP4_SADDRSEL)
2733 		return (1);
2734 
2735 	lia.s_addr = INADDR_ANY;
2736 	error = prison_get_ip4(cred, &lia);
2737 	if (error)
2738 		return (error);
2739 	if (lia.s_addr == INADDR_ANY)
2740 		return (1);
2741 
2742 	ia->s_addr = lia.s_addr;
2743 	return (0);
2744 }
2745 
2746 /*
2747  * Return true if pr1 and pr2 have the same IPv4 address restrictions.
2748  */
2749 int
2750 prison_equal_ip4(struct prison *pr1, struct prison *pr2)
2751 {
2752 
2753 	if (pr1 == pr2)
2754 		return (1);
2755 
2756 	/*
2757 	 * No need to lock since the PR_IP4_USER flag can't be altered for
2758 	 * existing prisons.
2759 	 */
2760 	while (pr1 != &prison0 &&
2761 #ifdef VIMAGE
2762 	       !(pr1->pr_flags & PR_VNET) &&
2763 #endif
2764 	       !(pr1->pr_flags & PR_IP4_USER))
2765 		pr1 = pr1->pr_parent;
2766 	while (pr2 != &prison0 &&
2767 #ifdef VIMAGE
2768 	       !(pr2->pr_flags & PR_VNET) &&
2769 #endif
2770 	       !(pr2->pr_flags & PR_IP4_USER))
2771 		pr2 = pr2->pr_parent;
2772 	return (pr1 == pr2);
2773 }
2774 
2775 /*
2776  * Make sure our (source) address is set to something meaningful to this
2777  * jail.
2778  *
2779  * Returns 0 if jail doesn't restrict IPv4 or if address belongs to jail,
2780  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
2781  * doesn't allow IPv4.  Address passed in in NBO and returned in NBO.
2782  */
2783 int
2784 prison_local_ip4(struct ucred *cred, struct in_addr *ia)
2785 {
2786 	struct prison *pr;
2787 	struct in_addr ia0;
2788 	int error;
2789 
2790 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2791 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2792 
2793 	pr = cred->cr_prison;
2794 	if (!(pr->pr_flags & PR_IP4))
2795 		return (0);
2796 	mtx_lock(&pr->pr_mtx);
2797 	if (!(pr->pr_flags & PR_IP4)) {
2798 		mtx_unlock(&pr->pr_mtx);
2799 		return (0);
2800 	}
2801 	if (pr->pr_ip4 == NULL) {
2802 		mtx_unlock(&pr->pr_mtx);
2803 		return (EAFNOSUPPORT);
2804 	}
2805 
2806 	ia0.s_addr = ntohl(ia->s_addr);
2807 	if (ia0.s_addr == INADDR_LOOPBACK) {
2808 		ia->s_addr = pr->pr_ip4[0].s_addr;
2809 		mtx_unlock(&pr->pr_mtx);
2810 		return (0);
2811 	}
2812 
2813 	if (ia0.s_addr == INADDR_ANY) {
2814 		/*
2815 		 * In case there is only 1 IPv4 address, bind directly.
2816 		 */
2817 		if (pr->pr_ip4s == 1)
2818 			ia->s_addr = pr->pr_ip4[0].s_addr;
2819 		mtx_unlock(&pr->pr_mtx);
2820 		return (0);
2821 	}
2822 
2823 	error = _prison_check_ip4(pr, ia);
2824 	mtx_unlock(&pr->pr_mtx);
2825 	return (error);
2826 }
2827 
2828 /*
2829  * Rewrite destination address in case we will connect to loopback address.
2830  *
2831  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv4.
2832  * Address passed in in NBO and returned in NBO.
2833  */
2834 int
2835 prison_remote_ip4(struct ucred *cred, struct in_addr *ia)
2836 {
2837 	struct prison *pr;
2838 
2839 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2840 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2841 
2842 	pr = cred->cr_prison;
2843 	if (!(pr->pr_flags & PR_IP4))
2844 		return (0);
2845 	mtx_lock(&pr->pr_mtx);
2846 	if (!(pr->pr_flags & PR_IP4)) {
2847 		mtx_unlock(&pr->pr_mtx);
2848 		return (0);
2849 	}
2850 	if (pr->pr_ip4 == NULL) {
2851 		mtx_unlock(&pr->pr_mtx);
2852 		return (EAFNOSUPPORT);
2853 	}
2854 
2855 	if (ntohl(ia->s_addr) == INADDR_LOOPBACK) {
2856 		ia->s_addr = pr->pr_ip4[0].s_addr;
2857 		mtx_unlock(&pr->pr_mtx);
2858 		return (0);
2859 	}
2860 
2861 	/*
2862 	 * Return success because nothing had to be changed.
2863 	 */
2864 	mtx_unlock(&pr->pr_mtx);
2865 	return (0);
2866 }
2867 
2868 /*
2869  * Check if given address belongs to the jail referenced by cred/prison.
2870  *
2871  * Returns 0 if jail doesn't restrict IPv4 or if address belongs to jail,
2872  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
2873  * doesn't allow IPv4.  Address passed in in NBO.
2874  */
2875 static int
2876 _prison_check_ip4(struct prison *pr, struct in_addr *ia)
2877 {
2878 	int i, a, z, d;
2879 
2880 	/*
2881 	 * Check the primary IP.
2882 	 */
2883 	if (pr->pr_ip4[0].s_addr == ia->s_addr)
2884 		return (0);
2885 
2886 	/*
2887 	 * All the other IPs are sorted so we can do a binary search.
2888 	 */
2889 	a = 0;
2890 	z = pr->pr_ip4s - 2;
2891 	while (a <= z) {
2892 		i = (a + z) / 2;
2893 		d = qcmp_v4(&pr->pr_ip4[i+1], ia);
2894 		if (d > 0)
2895 			z = i - 1;
2896 		else if (d < 0)
2897 			a = i + 1;
2898 		else
2899 			return (0);
2900 	}
2901 
2902 	return (EADDRNOTAVAIL);
2903 }
2904 
2905 int
2906 prison_check_ip4(struct ucred *cred, struct in_addr *ia)
2907 {
2908 	struct prison *pr;
2909 	int error;
2910 
2911 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2912 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2913 
2914 	pr = cred->cr_prison;
2915 	if (!(pr->pr_flags & PR_IP4))
2916 		return (0);
2917 	mtx_lock(&pr->pr_mtx);
2918 	if (!(pr->pr_flags & PR_IP4)) {
2919 		mtx_unlock(&pr->pr_mtx);
2920 		return (0);
2921 	}
2922 	if (pr->pr_ip4 == NULL) {
2923 		mtx_unlock(&pr->pr_mtx);
2924 		return (EAFNOSUPPORT);
2925 	}
2926 
2927 	error = _prison_check_ip4(pr, ia);
2928 	mtx_unlock(&pr->pr_mtx);
2929 	return (error);
2930 }
2931 #endif
2932 
2933 #ifdef INET6
2934 static int
2935 prison_restrict_ip6(struct prison *pr, struct in6_addr *newip6)
2936 {
2937 	int ii, ij, used;
2938 	struct prison *ppr;
2939 
2940 	ppr = pr->pr_parent;
2941 	if (!(pr->pr_flags & PR_IP6_USER)) {
2942 		/* This has no user settings, so just copy the parent's list. */
2943 		if (pr->pr_ip6s < ppr->pr_ip6s) {
2944 			/*
2945 			 * There's no room for the parent's list.  Use the
2946 			 * new list buffer, which is assumed to be big enough
2947 			 * (if it was passed).  If there's no buffer, try to
2948 			 * allocate one.
2949 			 */
2950 			used = 1;
2951 			if (newip6 == NULL) {
2952 				newip6 = malloc(ppr->pr_ip6s * sizeof(*newip6),
2953 				    M_PRISON, M_NOWAIT);
2954 				if (newip6 != NULL)
2955 					used = 0;
2956 			}
2957 			if (newip6 != NULL) {
2958 				bcopy(ppr->pr_ip6, newip6,
2959 				    ppr->pr_ip6s * sizeof(*newip6));
2960 				free(pr->pr_ip6, M_PRISON);
2961 				pr->pr_ip6 = newip6;
2962 				pr->pr_ip6s = ppr->pr_ip6s;
2963 			}
2964 			return (used);
2965 		}
2966 		pr->pr_ip6s = ppr->pr_ip6s;
2967 		if (pr->pr_ip6s > 0)
2968 			bcopy(ppr->pr_ip6, pr->pr_ip6,
2969 			    pr->pr_ip6s * sizeof(*newip6));
2970 		else if (pr->pr_ip6 != NULL) {
2971 			free(pr->pr_ip6, M_PRISON);
2972 			pr->pr_ip6 = NULL;
2973 		}
2974 	} else if (pr->pr_ip6s > 0) {
2975 		/* Remove addresses that aren't in the parent. */
2976 		for (ij = 0; ij < ppr->pr_ip6s; ij++)
2977 			if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[0],
2978 			    &ppr->pr_ip6[ij]))
2979 				break;
2980 		if (ij < ppr->pr_ip6s)
2981 			ii = 1;
2982 		else {
2983 			bcopy(pr->pr_ip6 + 1, pr->pr_ip6,
2984 			    --pr->pr_ip6s * sizeof(*pr->pr_ip6));
2985 			ii = 0;
2986 		}
2987 		for (ij = 1; ii < pr->pr_ip6s; ) {
2988 			if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[ii],
2989 			    &ppr->pr_ip6[0])) {
2990 				ii++;
2991 				continue;
2992 			}
2993 			switch (ij >= ppr->pr_ip4s ? -1 :
2994 				qcmp_v6(&pr->pr_ip6[ii], &ppr->pr_ip6[ij])) {
2995 			case -1:
2996 				bcopy(pr->pr_ip6 + ii + 1, pr->pr_ip6 + ii,
2997 				    (--pr->pr_ip6s - ii) * sizeof(*pr->pr_ip6));
2998 				break;
2999 			case 0:
3000 				ii++;
3001 				ij++;
3002 				break;
3003 			case 1:
3004 				ij++;
3005 				break;
3006 			}
3007 		}
3008 		if (pr->pr_ip6s == 0) {
3009 			pr->pr_flags |= PR_IP6_DISABLE;
3010 			free(pr->pr_ip6, M_PRISON);
3011 			pr->pr_ip6 = NULL;
3012 		}
3013 	}
3014 	return 0;
3015 }
3016 
3017 /*
3018  * Pass back primary IPv6 address for this jail.
3019  *
3020  * If not restricted return success but do not alter the address.  Caller has
3021  * to make sure to initialize it correctly (e.g. IN6ADDR_ANY_INIT).
3022  *
3023  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv6.
3024  */
3025 int
3026 prison_get_ip6(struct ucred *cred, struct in6_addr *ia6)
3027 {
3028 	struct prison *pr;
3029 
3030 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3031 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3032 
3033 	pr = cred->cr_prison;
3034 	if (!(pr->pr_flags & PR_IP6))
3035 		return (0);
3036 	mtx_lock(&pr->pr_mtx);
3037 	if (!(pr->pr_flags & PR_IP6)) {
3038 		mtx_unlock(&pr->pr_mtx);
3039 		return (0);
3040 	}
3041 	if (pr->pr_ip6 == NULL) {
3042 		mtx_unlock(&pr->pr_mtx);
3043 		return (EAFNOSUPPORT);
3044 	}
3045 
3046 	bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3047 	mtx_unlock(&pr->pr_mtx);
3048 	return (0);
3049 }
3050 
3051 /*
3052  * Return 1 if we should do proper source address selection or are not jailed.
3053  * We will return 0 if we should bypass source address selection in favour
3054  * of the primary jail IPv6 address. Only in this case *ia will be updated and
3055  * returned in NBO.
3056  * Return EAFNOSUPPORT, in case this jail does not allow IPv6.
3057  */
3058 int
3059 prison_saddrsel_ip6(struct ucred *cred, struct in6_addr *ia6)
3060 {
3061 	struct prison *pr;
3062 	struct in6_addr lia6;
3063 	int error;
3064 
3065 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3066 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3067 
3068 	if (!jailed(cred))
3069 		return (1);
3070 
3071 	pr = cred->cr_prison;
3072 	if (pr->pr_flags & PR_IP6_SADDRSEL)
3073 		return (1);
3074 
3075 	lia6 = in6addr_any;
3076 	error = prison_get_ip6(cred, &lia6);
3077 	if (error)
3078 		return (error);
3079 	if (IN6_IS_ADDR_UNSPECIFIED(&lia6))
3080 		return (1);
3081 
3082 	bcopy(&lia6, ia6, sizeof(struct in6_addr));
3083 	return (0);
3084 }
3085 
3086 /*
3087  * Return true if pr1 and pr2 have the same IPv6 address restrictions.
3088  */
3089 int
3090 prison_equal_ip6(struct prison *pr1, struct prison *pr2)
3091 {
3092 
3093 	if (pr1 == pr2)
3094 		return (1);
3095 
3096 	while (pr1 != &prison0 &&
3097 #ifdef VIMAGE
3098 	       !(pr1->pr_flags & PR_VNET) &&
3099 #endif
3100 	       !(pr1->pr_flags & PR_IP6_USER))
3101 		pr1 = pr1->pr_parent;
3102 	while (pr2 != &prison0 &&
3103 #ifdef VIMAGE
3104 	       !(pr2->pr_flags & PR_VNET) &&
3105 #endif
3106 	       !(pr2->pr_flags & PR_IP6_USER))
3107 		pr2 = pr2->pr_parent;
3108 	return (pr1 == pr2);
3109 }
3110 
3111 /*
3112  * Make sure our (source) address is set to something meaningful to this jail.
3113  *
3114  * v6only should be set based on (inp->inp_flags & IN6P_IPV6_V6ONLY != 0)
3115  * when needed while binding.
3116  *
3117  * Returns 0 if jail doesn't restrict IPv6 or if address belongs to jail,
3118  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3119  * doesn't allow IPv6.
3120  */
3121 int
3122 prison_local_ip6(struct ucred *cred, struct in6_addr *ia6, int v6only)
3123 {
3124 	struct prison *pr;
3125 	int error;
3126 
3127 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3128 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3129 
3130 	pr = cred->cr_prison;
3131 	if (!(pr->pr_flags & PR_IP6))
3132 		return (0);
3133 	mtx_lock(&pr->pr_mtx);
3134 	if (!(pr->pr_flags & PR_IP6)) {
3135 		mtx_unlock(&pr->pr_mtx);
3136 		return (0);
3137 	}
3138 	if (pr->pr_ip6 == NULL) {
3139 		mtx_unlock(&pr->pr_mtx);
3140 		return (EAFNOSUPPORT);
3141 	}
3142 
3143 	if (IN6_IS_ADDR_LOOPBACK(ia6)) {
3144 		bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3145 		mtx_unlock(&pr->pr_mtx);
3146 		return (0);
3147 	}
3148 
3149 	if (IN6_IS_ADDR_UNSPECIFIED(ia6)) {
3150 		/*
3151 		 * In case there is only 1 IPv6 address, and v6only is true,
3152 		 * then bind directly.
3153 		 */
3154 		if (v6only != 0 && pr->pr_ip6s == 1)
3155 			bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3156 		mtx_unlock(&pr->pr_mtx);
3157 		return (0);
3158 	}
3159 
3160 	error = _prison_check_ip6(pr, ia6);
3161 	mtx_unlock(&pr->pr_mtx);
3162 	return (error);
3163 }
3164 
3165 /*
3166  * Rewrite destination address in case we will connect to loopback address.
3167  *
3168  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv6.
3169  */
3170 int
3171 prison_remote_ip6(struct ucred *cred, struct in6_addr *ia6)
3172 {
3173 	struct prison *pr;
3174 
3175 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3176 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3177 
3178 	pr = cred->cr_prison;
3179 	if (!(pr->pr_flags & PR_IP6))
3180 		return (0);
3181 	mtx_lock(&pr->pr_mtx);
3182 	if (!(pr->pr_flags & PR_IP6)) {
3183 		mtx_unlock(&pr->pr_mtx);
3184 		return (0);
3185 	}
3186 	if (pr->pr_ip6 == NULL) {
3187 		mtx_unlock(&pr->pr_mtx);
3188 		return (EAFNOSUPPORT);
3189 	}
3190 
3191 	if (IN6_IS_ADDR_LOOPBACK(ia6)) {
3192 		bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3193 		mtx_unlock(&pr->pr_mtx);
3194 		return (0);
3195 	}
3196 
3197 	/*
3198 	 * Return success because nothing had to be changed.
3199 	 */
3200 	mtx_unlock(&pr->pr_mtx);
3201 	return (0);
3202 }
3203 
3204 /*
3205  * Check if given address belongs to the jail referenced by cred/prison.
3206  *
3207  * Returns 0 if jail doesn't restrict IPv6 or if address belongs to jail,
3208  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3209  * doesn't allow IPv6.
3210  */
3211 static int
3212 _prison_check_ip6(struct prison *pr, struct in6_addr *ia6)
3213 {
3214 	int i, a, z, d;
3215 
3216 	/*
3217 	 * Check the primary IP.
3218 	 */
3219 	if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[0], ia6))
3220 		return (0);
3221 
3222 	/*
3223 	 * All the other IPs are sorted so we can do a binary search.
3224 	 */
3225 	a = 0;
3226 	z = pr->pr_ip6s - 2;
3227 	while (a <= z) {
3228 		i = (a + z) / 2;
3229 		d = qcmp_v6(&pr->pr_ip6[i+1], ia6);
3230 		if (d > 0)
3231 			z = i - 1;
3232 		else if (d < 0)
3233 			a = i + 1;
3234 		else
3235 			return (0);
3236 	}
3237 
3238 	return (EADDRNOTAVAIL);
3239 }
3240 
3241 int
3242 prison_check_ip6(struct ucred *cred, struct in6_addr *ia6)
3243 {
3244 	struct prison *pr;
3245 	int error;
3246 
3247 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3248 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3249 
3250 	pr = cred->cr_prison;
3251 	if (!(pr->pr_flags & PR_IP6))
3252 		return (0);
3253 	mtx_lock(&pr->pr_mtx);
3254 	if (!(pr->pr_flags & PR_IP6)) {
3255 		mtx_unlock(&pr->pr_mtx);
3256 		return (0);
3257 	}
3258 	if (pr->pr_ip6 == NULL) {
3259 		mtx_unlock(&pr->pr_mtx);
3260 		return (EAFNOSUPPORT);
3261 	}
3262 
3263 	error = _prison_check_ip6(pr, ia6);
3264 	mtx_unlock(&pr->pr_mtx);
3265 	return (error);
3266 }
3267 #endif
3268 
3269 /*
3270  * Check if a jail supports the given address family.
3271  *
3272  * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
3273  * if not.
3274  */
3275 int
3276 prison_check_af(struct ucred *cred, int af)
3277 {
3278 	struct prison *pr;
3279 	int error;
3280 
3281 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3282 
3283 	pr = cred->cr_prison;
3284 #ifdef VIMAGE
3285 	/* Prisons with their own network stack are not limited. */
3286 	if (prison_owns_vnet(cred))
3287 		return (0);
3288 #endif
3289 
3290 	error = 0;
3291 	switch (af)
3292 	{
3293 #ifdef INET
3294 	case AF_INET:
3295 		if (pr->pr_flags & PR_IP4)
3296 		{
3297 			mtx_lock(&pr->pr_mtx);
3298 			if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL)
3299 				error = EAFNOSUPPORT;
3300 			mtx_unlock(&pr->pr_mtx);
3301 		}
3302 		break;
3303 #endif
3304 #ifdef INET6
3305 	case AF_INET6:
3306 		if (pr->pr_flags & PR_IP6)
3307 		{
3308 			mtx_lock(&pr->pr_mtx);
3309 			if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL)
3310 				error = EAFNOSUPPORT;
3311 			mtx_unlock(&pr->pr_mtx);
3312 		}
3313 		break;
3314 #endif
3315 	case AF_LOCAL:
3316 	case AF_ROUTE:
3317 		break;
3318 	default:
3319 		if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
3320 			error = EAFNOSUPPORT;
3321 	}
3322 	return (error);
3323 }
3324 
3325 /*
3326  * Check if given address belongs to the jail referenced by cred (wrapper to
3327  * prison_check_ip[46]).
3328  *
3329  * Returns 0 if jail doesn't restrict the address family or if address belongs
3330  * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
3331  * the jail doesn't allow the address family.  IPv4 Address passed in in NBO.
3332  */
3333 int
3334 prison_if(struct ucred *cred, struct sockaddr *sa)
3335 {
3336 #ifdef INET
3337 	struct sockaddr_in *sai;
3338 #endif
3339 #ifdef INET6
3340 	struct sockaddr_in6 *sai6;
3341 #endif
3342 	int error;
3343 
3344 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3345 	KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
3346 
3347 #ifdef VIMAGE
3348 	if (prison_owns_vnet(cred))
3349 		return (0);
3350 #endif
3351 
3352 	error = 0;
3353 	switch (sa->sa_family)
3354 	{
3355 #ifdef INET
3356 	case AF_INET:
3357 		sai = (struct sockaddr_in *)sa;
3358 		error = prison_check_ip4(cred, &sai->sin_addr);
3359 		break;
3360 #endif
3361 #ifdef INET6
3362 	case AF_INET6:
3363 		sai6 = (struct sockaddr_in6 *)sa;
3364 		error = prison_check_ip6(cred, &sai6->sin6_addr);
3365 		break;
3366 #endif
3367 	default:
3368 		if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
3369 			error = EAFNOSUPPORT;
3370 	}
3371 	return (error);
3372 }
3373 
3374 /*
3375  * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
3376  */
3377 int
3378 prison_check(struct ucred *cred1, struct ucred *cred2)
3379 {
3380 
3381 	return ((cred1->cr_prison == cred2->cr_prison ||
3382 	    prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
3383 }
3384 
3385 /*
3386  * Return 1 if p2 is a child of p1, otherwise 0.
3387  */
3388 int
3389 prison_ischild(struct prison *pr1, struct prison *pr2)
3390 {
3391 
3392 	for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
3393 		if (pr1 == pr2)
3394 			return (1);
3395 	return (0);
3396 }
3397 
3398 /*
3399  * Return 1 if the passed credential is in a jail, otherwise 0.
3400  */
3401 int
3402 jailed(struct ucred *cred)
3403 {
3404 
3405 	return (cred->cr_prison != &prison0);
3406 }
3407 
3408 /*
3409  * Return 1 if the passed credential is in a jail and that jail does not
3410  * have its own virtual network stack, otherwise 0.
3411  */
3412 int
3413 jailed_without_vnet(struct ucred *cred)
3414 {
3415 
3416 	if (!jailed(cred))
3417 		return (0);
3418 #ifdef VIMAGE
3419 	if (prison_owns_vnet(cred))
3420 		return (0);
3421 #endif
3422 
3423 	return (1);
3424 }
3425 
3426 /*
3427  * Return the correct hostname (domainname, et al) for the passed credential.
3428  */
3429 void
3430 getcredhostname(struct ucred *cred, char *buf, size_t size)
3431 {
3432 	struct prison *pr;
3433 
3434 	/*
3435 	 * A NULL credential can be used to shortcut to the physical
3436 	 * system's hostname.
3437 	 */
3438 	pr = (cred != NULL) ? cred->cr_prison : &prison0;
3439 	mtx_lock(&pr->pr_mtx);
3440 	strlcpy(buf, pr->pr_hostname, size);
3441 	mtx_unlock(&pr->pr_mtx);
3442 }
3443 
3444 void
3445 getcreddomainname(struct ucred *cred, char *buf, size_t size)
3446 {
3447 
3448 	mtx_lock(&cred->cr_prison->pr_mtx);
3449 	strlcpy(buf, cred->cr_prison->pr_domainname, size);
3450 	mtx_unlock(&cred->cr_prison->pr_mtx);
3451 }
3452 
3453 void
3454 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
3455 {
3456 
3457 	mtx_lock(&cred->cr_prison->pr_mtx);
3458 	strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
3459 	mtx_unlock(&cred->cr_prison->pr_mtx);
3460 }
3461 
3462 void
3463 getcredhostid(struct ucred *cred, unsigned long *hostid)
3464 {
3465 
3466 	mtx_lock(&cred->cr_prison->pr_mtx);
3467 	*hostid = cred->cr_prison->pr_hostid;
3468 	mtx_unlock(&cred->cr_prison->pr_mtx);
3469 }
3470 
3471 #ifdef VIMAGE
3472 /*
3473  * Determine whether the prison represented by cred owns
3474  * its vnet rather than having it inherited.
3475  *
3476  * Returns 1 in case the prison owns the vnet, 0 otherwise.
3477  */
3478 int
3479 prison_owns_vnet(struct ucred *cred)
3480 {
3481 
3482 	/*
3483 	 * vnets cannot be added/removed after jail creation,
3484 	 * so no need to lock here.
3485 	 */
3486 	return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
3487 }
3488 #endif
3489 
3490 /*
3491  * Determine whether the subject represented by cred can "see"
3492  * status of a mount point.
3493  * Returns: 0 for permitted, ENOENT otherwise.
3494  * XXX: This function should be called cr_canseemount() and should be
3495  *      placed in kern_prot.c.
3496  */
3497 int
3498 prison_canseemount(struct ucred *cred, struct mount *mp)
3499 {
3500 	struct prison *pr;
3501 	struct statfs *sp;
3502 	size_t len;
3503 
3504 	pr = cred->cr_prison;
3505 	if (pr->pr_enforce_statfs == 0)
3506 		return (0);
3507 	if (pr->pr_root->v_mount == mp)
3508 		return (0);
3509 	if (pr->pr_enforce_statfs == 2)
3510 		return (ENOENT);
3511 	/*
3512 	 * If jail's chroot directory is set to "/" we should be able to see
3513 	 * all mount-points from inside a jail.
3514 	 * This is ugly check, but this is the only situation when jail's
3515 	 * directory ends with '/'.
3516 	 */
3517 	if (strcmp(pr->pr_path, "/") == 0)
3518 		return (0);
3519 	len = strlen(pr->pr_path);
3520 	sp = &mp->mnt_stat;
3521 	if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
3522 		return (ENOENT);
3523 	/*
3524 	 * Be sure that we don't have situation where jail's root directory
3525 	 * is "/some/path" and mount point is "/some/pathpath".
3526 	 */
3527 	if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
3528 		return (ENOENT);
3529 	return (0);
3530 }
3531 
3532 void
3533 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
3534 {
3535 	char jpath[MAXPATHLEN];
3536 	struct prison *pr;
3537 	size_t len;
3538 
3539 	pr = cred->cr_prison;
3540 	if (pr->pr_enforce_statfs == 0)
3541 		return;
3542 	if (prison_canseemount(cred, mp) != 0) {
3543 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3544 		strlcpy(sp->f_mntonname, "[restricted]",
3545 		    sizeof(sp->f_mntonname));
3546 		return;
3547 	}
3548 	if (pr->pr_root->v_mount == mp) {
3549 		/*
3550 		 * Clear current buffer data, so we are sure nothing from
3551 		 * the valid path left there.
3552 		 */
3553 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3554 		*sp->f_mntonname = '/';
3555 		return;
3556 	}
3557 	/*
3558 	 * If jail's chroot directory is set to "/" we should be able to see
3559 	 * all mount-points from inside a jail.
3560 	 */
3561 	if (strcmp(pr->pr_path, "/") == 0)
3562 		return;
3563 	len = strlen(pr->pr_path);
3564 	strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3565 	/*
3566 	 * Clear current buffer data, so we are sure nothing from
3567 	 * the valid path left there.
3568 	 */
3569 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3570 	if (*jpath == '\0') {
3571 		/* Should never happen. */
3572 		*sp->f_mntonname = '/';
3573 	} else {
3574 		strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3575 	}
3576 }
3577 
3578 /*
3579  * Check with permission for a specific privilege is granted within jail.  We
3580  * have a specific list of accepted privileges; the rest are denied.
3581  */
3582 int
3583 prison_priv_check(struct ucred *cred, int priv)
3584 {
3585 
3586 	if (!jailed(cred))
3587 		return (0);
3588 
3589 #ifdef VIMAGE
3590 	/*
3591 	 * Privileges specific to prisons with a virtual network stack.
3592 	 * There might be a duplicate entry here in case the privilege
3593 	 * is only granted conditionally in the legacy jail case.
3594 	 */
3595 	switch (priv) {
3596 #ifdef notyet
3597 		/*
3598 		 * NFS-specific privileges.
3599 		 */
3600 	case PRIV_NFS_DAEMON:
3601 	case PRIV_NFS_LOCKD:
3602 #endif
3603 		/*
3604 		 * Network stack privileges.
3605 		 */
3606 	case PRIV_NET_BRIDGE:
3607 	case PRIV_NET_GRE:
3608 	case PRIV_NET_BPF:
3609 	case PRIV_NET_RAW:		/* Dup, cond. in legacy jail case. */
3610 	case PRIV_NET_ROUTE:
3611 	case PRIV_NET_TAP:
3612 	case PRIV_NET_SETIFMTU:
3613 	case PRIV_NET_SETIFFLAGS:
3614 	case PRIV_NET_SETIFCAP:
3615 	case PRIV_NET_SETIFDESCR:
3616 	case PRIV_NET_SETIFNAME	:
3617 	case PRIV_NET_SETIFMETRIC:
3618 	case PRIV_NET_SETIFPHYS:
3619 	case PRIV_NET_SETIFMAC:
3620 	case PRIV_NET_ADDMULTI:
3621 	case PRIV_NET_DELMULTI:
3622 	case PRIV_NET_HWIOCTL:
3623 	case PRIV_NET_SETLLADDR:
3624 	case PRIV_NET_ADDIFGROUP:
3625 	case PRIV_NET_DELIFGROUP:
3626 	case PRIV_NET_IFCREATE:
3627 	case PRIV_NET_IFDESTROY:
3628 	case PRIV_NET_ADDIFADDR:
3629 	case PRIV_NET_DELIFADDR:
3630 	case PRIV_NET_LAGG:
3631 	case PRIV_NET_GIF:
3632 	case PRIV_NET_SETIFVNET:
3633 
3634 		/*
3635 		 * 802.11-related privileges.
3636 		 */
3637 	case PRIV_NET80211_GETKEY:
3638 #ifdef notyet
3639 	case PRIV_NET80211_MANAGE:		/* XXX-BZ discuss with sam@ */
3640 #endif
3641 
3642 #ifdef notyet
3643 		/*
3644 		 * AppleTalk privileges.
3645 		 */
3646 	case PRIV_NETATALK_RESERVEDPORT:
3647 
3648 		/*
3649 		 * ATM privileges.
3650 		 */
3651 	case PRIV_NETATM_CFG:
3652 	case PRIV_NETATM_ADD:
3653 	case PRIV_NETATM_DEL:
3654 	case PRIV_NETATM_SET:
3655 
3656 		/*
3657 		 * Bluetooth privileges.
3658 		 */
3659 	case PRIV_NETBLUETOOTH_RAW:
3660 #endif
3661 
3662 		/*
3663 		 * Netgraph and netgraph module privileges.
3664 		 */
3665 	case PRIV_NETGRAPH_CONTROL:
3666 #ifdef notyet
3667 	case PRIV_NETGRAPH_TTY:
3668 #endif
3669 
3670 		/*
3671 		 * IPv4 and IPv6 privileges.
3672 		 */
3673 	case PRIV_NETINET_IPFW:
3674 	case PRIV_NETINET_DIVERT:
3675 	case PRIV_NETINET_PF:
3676 	case PRIV_NETINET_DUMMYNET:
3677 	case PRIV_NETINET_CARP:
3678 	case PRIV_NETINET_MROUTE:
3679 	case PRIV_NETINET_RAW:
3680 	case PRIV_NETINET_ADDRCTRL6:
3681 	case PRIV_NETINET_ND6:
3682 	case PRIV_NETINET_SCOPE6:
3683 	case PRIV_NETINET_ALIFETIME6:
3684 	case PRIV_NETINET_IPSEC:
3685 	case PRIV_NETINET_BINDANY:
3686 
3687 #ifdef notyet
3688 		/*
3689 		 * IPX/SPX privileges.
3690 		 */
3691 	case PRIV_NETIPX_RESERVEDPORT:
3692 	case PRIV_NETIPX_RAW:
3693 
3694 		/*
3695 		 * NCP privileges.
3696 		 */
3697 	case PRIV_NETNCP:
3698 
3699 		/*
3700 		 * SMB privileges.
3701 		 */
3702 	case PRIV_NETSMB:
3703 #endif
3704 
3705 	/*
3706 	 * No default: or deny here.
3707 	 * In case of no permit fall through to next switch().
3708 	 */
3709 		if (cred->cr_prison->pr_flags & PR_VNET)
3710 			return (0);
3711 	}
3712 #endif /* VIMAGE */
3713 
3714 	switch (priv) {
3715 
3716 		/*
3717 		 * Allow ktrace privileges for root in jail.
3718 		 */
3719 	case PRIV_KTRACE:
3720 
3721 #if 0
3722 		/*
3723 		 * Allow jailed processes to configure audit identity and
3724 		 * submit audit records (login, etc).  In the future we may
3725 		 * want to further refine the relationship between audit and
3726 		 * jail.
3727 		 */
3728 	case PRIV_AUDIT_GETAUDIT:
3729 	case PRIV_AUDIT_SETAUDIT:
3730 	case PRIV_AUDIT_SUBMIT:
3731 #endif
3732 
3733 		/*
3734 		 * Allow jailed processes to manipulate process UNIX
3735 		 * credentials in any way they see fit.
3736 		 */
3737 	case PRIV_CRED_SETUID:
3738 	case PRIV_CRED_SETEUID:
3739 	case PRIV_CRED_SETGID:
3740 	case PRIV_CRED_SETEGID:
3741 	case PRIV_CRED_SETGROUPS:
3742 	case PRIV_CRED_SETREUID:
3743 	case PRIV_CRED_SETREGID:
3744 	case PRIV_CRED_SETRESUID:
3745 	case PRIV_CRED_SETRESGID:
3746 
3747 		/*
3748 		 * Jail implements visibility constraints already, so allow
3749 		 * jailed root to override uid/gid-based constraints.
3750 		 */
3751 	case PRIV_SEEOTHERGIDS:
3752 	case PRIV_SEEOTHERUIDS:
3753 
3754 		/*
3755 		 * Jail implements inter-process debugging limits already, so
3756 		 * allow jailed root various debugging privileges.
3757 		 */
3758 	case PRIV_DEBUG_DIFFCRED:
3759 	case PRIV_DEBUG_SUGID:
3760 	case PRIV_DEBUG_UNPRIV:
3761 
3762 		/*
3763 		 * Allow jail to set various resource limits and login
3764 		 * properties, and for now, exceed process resource limits.
3765 		 */
3766 	case PRIV_PROC_LIMIT:
3767 	case PRIV_PROC_SETLOGIN:
3768 	case PRIV_PROC_SETRLIMIT:
3769 
3770 		/*
3771 		 * System V and POSIX IPC privileges are granted in jail.
3772 		 */
3773 	case PRIV_IPC_READ:
3774 	case PRIV_IPC_WRITE:
3775 	case PRIV_IPC_ADMIN:
3776 	case PRIV_IPC_MSGSIZE:
3777 	case PRIV_MQ_ADMIN:
3778 
3779 		/*
3780 		 * Jail operations within a jail work on child jails.
3781 		 */
3782 	case PRIV_JAIL_ATTACH:
3783 	case PRIV_JAIL_SET:
3784 	case PRIV_JAIL_REMOVE:
3785 
3786 		/*
3787 		 * Jail implements its own inter-process limits, so allow
3788 		 * root processes in jail to change scheduling on other
3789 		 * processes in the same jail.  Likewise for signalling.
3790 		 */
3791 	case PRIV_SCHED_DIFFCRED:
3792 	case PRIV_SCHED_CPUSET:
3793 	case PRIV_SIGNAL_DIFFCRED:
3794 	case PRIV_SIGNAL_SUGID:
3795 
3796 		/*
3797 		 * Allow jailed processes to write to sysctls marked as jail
3798 		 * writable.
3799 		 */
3800 	case PRIV_SYSCTL_WRITEJAIL:
3801 
3802 		/*
3803 		 * Allow root in jail to manage a variety of quota
3804 		 * properties.  These should likely be conditional on a
3805 		 * configuration option.
3806 		 */
3807 	case PRIV_VFS_GETQUOTA:
3808 	case PRIV_VFS_SETQUOTA:
3809 
3810 		/*
3811 		 * Since Jail relies on chroot() to implement file system
3812 		 * protections, grant many VFS privileges to root in jail.
3813 		 * Be careful to exclude mount-related and NFS-related
3814 		 * privileges.
3815 		 */
3816 	case PRIV_VFS_READ:
3817 	case PRIV_VFS_WRITE:
3818 	case PRIV_VFS_ADMIN:
3819 	case PRIV_VFS_EXEC:
3820 	case PRIV_VFS_LOOKUP:
3821 	case PRIV_VFS_BLOCKRESERVE:	/* XXXRW: Slightly surprising. */
3822 	case PRIV_VFS_CHFLAGS_DEV:
3823 	case PRIV_VFS_CHOWN:
3824 	case PRIV_VFS_CHROOT:
3825 	case PRIV_VFS_RETAINSUGID:
3826 	case PRIV_VFS_FCHROOT:
3827 	case PRIV_VFS_LINK:
3828 	case PRIV_VFS_SETGID:
3829 	case PRIV_VFS_STAT:
3830 	case PRIV_VFS_STICKYFILE:
3831 		return (0);
3832 
3833 		/*
3834 		 * Depending on the global setting, allow privilege of
3835 		 * setting system flags.
3836 		 */
3837 	case PRIV_VFS_SYSFLAGS:
3838 		if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
3839 			return (0);
3840 		else
3841 			return (EPERM);
3842 
3843 		/*
3844 		 * Depending on the global setting, allow privilege of
3845 		 * mounting/unmounting file systems.
3846 		 */
3847 	case PRIV_VFS_MOUNT:
3848 	case PRIV_VFS_UNMOUNT:
3849 	case PRIV_VFS_MOUNT_NONUSER:
3850 	case PRIV_VFS_MOUNT_OWNER:
3851 		if (cred->cr_prison->pr_allow & PR_ALLOW_MOUNT)
3852 			return (0);
3853 		else
3854 			return (EPERM);
3855 
3856 		/*
3857 		 * Allow jailed root to bind reserved ports and reuse in-use
3858 		 * ports.
3859 		 */
3860 	case PRIV_NETINET_RESERVEDPORT:
3861 	case PRIV_NETINET_REUSEPORT:
3862 		return (0);
3863 
3864 		/*
3865 		 * Allow jailed root to set certian IPv4/6 (option) headers.
3866 		 */
3867 	case PRIV_NETINET_SETHDROPTS:
3868 		return (0);
3869 
3870 		/*
3871 		 * Conditionally allow creating raw sockets in jail.
3872 		 */
3873 	case PRIV_NETINET_RAW:
3874 		if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
3875 			return (0);
3876 		else
3877 			return (EPERM);
3878 
3879 		/*
3880 		 * Since jail implements its own visibility limits on netstat
3881 		 * sysctls, allow getcred.  This allows identd to work in
3882 		 * jail.
3883 		 */
3884 	case PRIV_NETINET_GETCRED:
3885 		return (0);
3886 
3887 		/*
3888 		 * Allow jailed root to set loginclass.
3889 		 */
3890 	case PRIV_PROC_SETLOGINCLASS:
3891 		return (0);
3892 
3893 	default:
3894 		/*
3895 		 * In all remaining cases, deny the privilege request.  This
3896 		 * includes almost all network privileges, many system
3897 		 * configuration privileges.
3898 		 */
3899 		return (EPERM);
3900 	}
3901 }
3902 
3903 /*
3904  * Return the part of pr2's name that is relative to pr1, or the whole name
3905  * if it does not directly follow.
3906  */
3907 
3908 char *
3909 prison_name(struct prison *pr1, struct prison *pr2)
3910 {
3911 	char *name;
3912 
3913 	/* Jails see themselves as "0" (if they see themselves at all). */
3914 	if (pr1 == pr2)
3915 		return "0";
3916 	name = pr2->pr_name;
3917 	if (prison_ischild(pr1, pr2)) {
3918 		/*
3919 		 * pr1 isn't locked (and allprison_lock may not be either)
3920 		 * so its length can't be counted on.  But the number of dots
3921 		 * can be counted on - and counted.
3922 		 */
3923 		for (; pr1 != &prison0; pr1 = pr1->pr_parent)
3924 			name = strchr(name, '.') + 1;
3925 	}
3926 	return (name);
3927 }
3928 
3929 /*
3930  * Return the part of pr2's path that is relative to pr1, or the whole path
3931  * if it does not directly follow.
3932  */
3933 static char *
3934 prison_path(struct prison *pr1, struct prison *pr2)
3935 {
3936 	char *path1, *path2;
3937 	int len1;
3938 
3939 	path1 = pr1->pr_path;
3940 	path2 = pr2->pr_path;
3941 	if (!strcmp(path1, "/"))
3942 		return (path2);
3943 	len1 = strlen(path1);
3944 	if (strncmp(path1, path2, len1))
3945 		return (path2);
3946 	if (path2[len1] == '\0')
3947 		return "/";
3948 	if (path2[len1] == '/')
3949 		return (path2 + len1);
3950 	return (path2);
3951 }
3952 
3953 
3954 /*
3955  * Jail-related sysctls.
3956  */
3957 SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW, 0,
3958     "Jails");
3959 
3960 static int
3961 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
3962 {
3963 	struct xprison *xp;
3964 	struct prison *pr, *cpr;
3965 #ifdef INET
3966 	struct in_addr *ip4 = NULL;
3967 	int ip4s = 0;
3968 #endif
3969 #ifdef INET6
3970 	struct in6_addr *ip6 = NULL;
3971 	int ip6s = 0;
3972 #endif
3973 	int descend, error;
3974 
3975 	xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
3976 	pr = req->td->td_ucred->cr_prison;
3977 	error = 0;
3978 	sx_slock(&allprison_lock);
3979 	FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
3980 #if defined(INET) || defined(INET6)
3981  again:
3982 #endif
3983 		mtx_lock(&cpr->pr_mtx);
3984 #ifdef INET
3985 		if (cpr->pr_ip4s > 0) {
3986 			if (ip4s < cpr->pr_ip4s) {
3987 				ip4s = cpr->pr_ip4s;
3988 				mtx_unlock(&cpr->pr_mtx);
3989 				ip4 = realloc(ip4, ip4s *
3990 				    sizeof(struct in_addr), M_TEMP, M_WAITOK);
3991 				goto again;
3992 			}
3993 			bcopy(cpr->pr_ip4, ip4,
3994 			    cpr->pr_ip4s * sizeof(struct in_addr));
3995 		}
3996 #endif
3997 #ifdef INET6
3998 		if (cpr->pr_ip6s > 0) {
3999 			if (ip6s < cpr->pr_ip6s) {
4000 				ip6s = cpr->pr_ip6s;
4001 				mtx_unlock(&cpr->pr_mtx);
4002 				ip6 = realloc(ip6, ip6s *
4003 				    sizeof(struct in6_addr), M_TEMP, M_WAITOK);
4004 				goto again;
4005 			}
4006 			bcopy(cpr->pr_ip6, ip6,
4007 			    cpr->pr_ip6s * sizeof(struct in6_addr));
4008 		}
4009 #endif
4010 		if (cpr->pr_ref == 0) {
4011 			mtx_unlock(&cpr->pr_mtx);
4012 			continue;
4013 		}
4014 		bzero(xp, sizeof(*xp));
4015 		xp->pr_version = XPRISON_VERSION;
4016 		xp->pr_id = cpr->pr_id;
4017 		xp->pr_state = cpr->pr_uref > 0
4018 		    ? PRISON_STATE_ALIVE : PRISON_STATE_DYING;
4019 		strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
4020 		strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
4021 		strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
4022 #ifdef INET
4023 		xp->pr_ip4s = cpr->pr_ip4s;
4024 #endif
4025 #ifdef INET6
4026 		xp->pr_ip6s = cpr->pr_ip6s;
4027 #endif
4028 		mtx_unlock(&cpr->pr_mtx);
4029 		error = SYSCTL_OUT(req, xp, sizeof(*xp));
4030 		if (error)
4031 			break;
4032 #ifdef INET
4033 		if (xp->pr_ip4s > 0) {
4034 			error = SYSCTL_OUT(req, ip4,
4035 			    xp->pr_ip4s * sizeof(struct in_addr));
4036 			if (error)
4037 				break;
4038 		}
4039 #endif
4040 #ifdef INET6
4041 		if (xp->pr_ip6s > 0) {
4042 			error = SYSCTL_OUT(req, ip6,
4043 			    xp->pr_ip6s * sizeof(struct in6_addr));
4044 			if (error)
4045 				break;
4046 		}
4047 #endif
4048 	}
4049 	sx_sunlock(&allprison_lock);
4050 	free(xp, M_TEMP);
4051 #ifdef INET
4052 	free(ip4, M_TEMP);
4053 #endif
4054 #ifdef INET6
4055 	free(ip6, M_TEMP);
4056 #endif
4057 	return (error);
4058 }
4059 
4060 SYSCTL_OID(_security_jail, OID_AUTO, list,
4061     CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4062     sysctl_jail_list, "S", "List of active jails");
4063 
4064 static int
4065 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
4066 {
4067 	int error, injail;
4068 
4069 	injail = jailed(req->td->td_ucred);
4070 	error = SYSCTL_OUT(req, &injail, sizeof(injail));
4071 
4072 	return (error);
4073 }
4074 
4075 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
4076     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
4077     sysctl_jail_jailed, "I", "Process in jail?");
4078 
4079 #if defined(INET) || defined(INET6)
4080 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
4081     &jail_max_af_ips, 0,
4082     "Number of IP addresses a jail may have at most per address family");
4083 #endif
4084 
4085 /*
4086  * Default parameters for jail(2) compatability.  For historical reasons,
4087  * the sysctl names have varying similarity to the parameter names.  Prisons
4088  * just see their own parameters, and can't change them.
4089  */
4090 static int
4091 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
4092 {
4093 	struct prison *pr;
4094 	int allow, error, i;
4095 
4096 	pr = req->td->td_ucred->cr_prison;
4097 	allow = (pr == &prison0) ? jail_default_allow : pr->pr_allow;
4098 
4099 	/* Get the current flag value, and convert it to a boolean. */
4100 	i = (allow & arg2) ? 1 : 0;
4101 	if (arg1 != NULL)
4102 		i = !i;
4103 	error = sysctl_handle_int(oidp, &i, 0, req);
4104 	if (error || !req->newptr)
4105 		return (error);
4106 	i = i ? arg2 : 0;
4107 	if (arg1 != NULL)
4108 		i ^= arg2;
4109 	/*
4110 	 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
4111 	 * for writing.
4112 	 */
4113 	mtx_lock(&prison0.pr_mtx);
4114 	jail_default_allow = (jail_default_allow & ~arg2) | i;
4115 	mtx_unlock(&prison0.pr_mtx);
4116 	return (0);
4117 }
4118 
4119 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
4120     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4121     NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
4122     "Processes in jail can set their hostnames");
4123 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
4124     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4125     (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
4126     "Processes in jail are limited to creating UNIX/IP/route sockets only");
4127 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
4128     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4129     NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
4130     "Processes in jail can use System V IPC primitives");
4131 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
4132     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4133     NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
4134     "Prison root can create raw sockets");
4135 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
4136     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4137     NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
4138     "Processes in jail can alter system file flags");
4139 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
4140     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4141     NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
4142     "Processes in jail can mount/unmount jail-friendly file systems");
4143 
4144 static int
4145 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
4146 {
4147 	struct prison *pr;
4148 	int level, error;
4149 
4150 	pr = req->td->td_ucred->cr_prison;
4151 	level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
4152 	error = sysctl_handle_int(oidp, &level, 0, req);
4153 	if (error || !req->newptr)
4154 		return (error);
4155 	*(int *)arg1 = level;
4156 	return (0);
4157 }
4158 
4159 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
4160     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
4161     &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
4162     sysctl_jail_default_level, "I",
4163     "Processes in jail cannot see all mounted file systems");
4164 
4165 /*
4166  * Nodes to describe jail parameters.  Maximum length of string parameters
4167  * is returned in the string itself, and the other parameters exist merely
4168  * to make themselves and their types known.
4169  */
4170 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW, 0,
4171     "Jail parameters");
4172 
4173 int
4174 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
4175 {
4176 	int i;
4177 	long l;
4178 	size_t s;
4179 	char numbuf[12];
4180 
4181 	switch (oidp->oid_kind & CTLTYPE)
4182 	{
4183 	case CTLTYPE_LONG:
4184 	case CTLTYPE_ULONG:
4185 		l = 0;
4186 #ifdef SCTL_MASK32
4187 		if (!(req->flags & SCTL_MASK32))
4188 #endif
4189 			return (SYSCTL_OUT(req, &l, sizeof(l)));
4190 	case CTLTYPE_INT:
4191 	case CTLTYPE_UINT:
4192 		i = 0;
4193 		return (SYSCTL_OUT(req, &i, sizeof(i)));
4194 	case CTLTYPE_STRING:
4195 		snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2);
4196 		return
4197 		    (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
4198 	case CTLTYPE_STRUCT:
4199 		s = (size_t)arg2;
4200 		return (SYSCTL_OUT(req, &s, sizeof(s)));
4201 	}
4202 	return (0);
4203 }
4204 
4205 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
4206 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
4207 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
4208 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
4209 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
4210     "I", "Jail secure level");
4211 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
4212     "I", "Jail cannot see all mounted file systems");
4213 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
4214     "B", "Jail persistence");
4215 #ifdef VIMAGE
4216 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
4217     "E,jailsys", "Virtual network stack");
4218 #endif
4219 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
4220     "B", "Jail is in the process of shutting down");
4221 
4222 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
4223 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
4224     "I", "Current number of child jails");
4225 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
4226     "I", "Maximum number of child jails");
4227 
4228 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
4229 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
4230     "Jail hostname");
4231 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
4232     "Jail NIS domainname");
4233 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
4234     "Jail host UUID");
4235 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
4236     "LU", "Jail host ID");
4237 
4238 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
4239 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
4240 
4241 #ifdef INET
4242 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
4243     "Jail IPv4 address virtualization");
4244 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
4245     "S,in_addr,a", "Jail IPv4 addresses");
4246 SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4247     "B", "Do (not) use IPv4 source address selection rather than the "
4248     "primary jail IPv4 address.");
4249 #endif
4250 #ifdef INET6
4251 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
4252     "Jail IPv6 address virtualization");
4253 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
4254     "S,in6_addr,a", "Jail IPv6 addresses");
4255 SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW,
4256     "B", "Do (not) use IPv6 source address selection rather than the "
4257     "primary jail IPv6 address.");
4258 #endif
4259 
4260 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
4261 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
4262     "B", "Jail may set hostname");
4263 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
4264     "B", "Jail may use SYSV IPC");
4265 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
4266     "B", "Jail may create raw sockets");
4267 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
4268     "B", "Jail may alter system file flags");
4269 SYSCTL_JAIL_PARAM(_allow, mount, CTLTYPE_INT | CTLFLAG_RW,
4270     "B", "Jail may mount/unmount jail-friendly file systems");
4271 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
4272     "B", "Jail may set file quotas");
4273 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
4274     "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
4275 
4276 void
4277 prison_racct_foreach(void (*callback)(struct racct *racct,
4278     void *arg2, void *arg3), void *arg2, void *arg3)
4279 {
4280 	struct prison *pr;
4281 
4282 	sx_slock(&allprison_lock);
4283 	TAILQ_FOREACH(pr, &allprison, pr_list)
4284 		(callback)(pr->pr_racct, arg2, arg3);
4285 	sx_sunlock(&allprison_lock);
4286 }
4287 
4288 #ifdef DDB
4289 
4290 static void
4291 db_show_prison(struct prison *pr)
4292 {
4293 	int fi;
4294 #if defined(INET) || defined(INET6)
4295 	int ii;
4296 #endif
4297 	unsigned jsf;
4298 #ifdef INET6
4299 	char ip6buf[INET6_ADDRSTRLEN];
4300 #endif
4301 
4302 	db_printf("prison %p:\n", pr);
4303 	db_printf(" jid             = %d\n", pr->pr_id);
4304 	db_printf(" name            = %s\n", pr->pr_name);
4305 	db_printf(" parent          = %p\n", pr->pr_parent);
4306 	db_printf(" ref             = %d\n", pr->pr_ref);
4307 	db_printf(" uref            = %d\n", pr->pr_uref);
4308 	db_printf(" path            = %s\n", pr->pr_path);
4309 	db_printf(" cpuset          = %d\n", pr->pr_cpuset
4310 	    ? pr->pr_cpuset->cs_id : -1);
4311 #ifdef VIMAGE
4312 	db_printf(" vnet            = %p\n", pr->pr_vnet);
4313 #endif
4314 	db_printf(" root            = %p\n", pr->pr_root);
4315 	db_printf(" securelevel     = %d\n", pr->pr_securelevel);
4316 	db_printf(" children.max    = %d\n", pr->pr_childmax);
4317 	db_printf(" children.cur    = %d\n", pr->pr_childcount);
4318 	db_printf(" child           = %p\n", LIST_FIRST(&pr->pr_children));
4319 	db_printf(" sibling         = %p\n", LIST_NEXT(pr, pr_sibling));
4320 	db_printf(" flags           = 0x%x", pr->pr_flags);
4321 	for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
4322 	    fi++)
4323 		if (pr_flag_names[fi] != NULL && (pr->pr_flags & (1 << fi)))
4324 			db_printf(" %s", pr_flag_names[fi]);
4325 	for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
4326 	    fi++) {
4327 		jsf = pr->pr_flags &
4328 		    (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
4329 		db_printf(" %-16s= %s\n", pr_flag_jailsys[fi].name,
4330 		    pr_flag_jailsys[fi].disable &&
4331 		      (jsf == pr_flag_jailsys[fi].disable) ? "disable"
4332 		    : (jsf == pr_flag_jailsys[fi].new) ? "new"
4333 		    : "inherit");
4334 	}
4335 	db_printf(" allow           = 0x%x", pr->pr_allow);
4336 	for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
4337 	    fi++)
4338 		if (pr_allow_names[fi] != NULL && (pr->pr_allow & (1 << fi)))
4339 			db_printf(" %s", pr_allow_names[fi]);
4340 	db_printf("\n");
4341 	db_printf(" enforce_statfs  = %d\n", pr->pr_enforce_statfs);
4342 	db_printf(" host.hostname   = %s\n", pr->pr_hostname);
4343 	db_printf(" host.domainname = %s\n", pr->pr_domainname);
4344 	db_printf(" host.hostuuid   = %s\n", pr->pr_hostuuid);
4345 	db_printf(" host.hostid     = %lu\n", pr->pr_hostid);
4346 #ifdef INET
4347 	db_printf(" ip4s            = %d\n", pr->pr_ip4s);
4348 	for (ii = 0; ii < pr->pr_ip4s; ii++)
4349 		db_printf(" %s %s\n",
4350 		    ii == 0 ? "ip4.addr        =" : "                 ",
4351 		    inet_ntoa(pr->pr_ip4[ii]));
4352 #endif
4353 #ifdef INET6
4354 	db_printf(" ip6s            = %d\n", pr->pr_ip6s);
4355 	for (ii = 0; ii < pr->pr_ip6s; ii++)
4356 		db_printf(" %s %s\n",
4357 		    ii == 0 ? "ip6.addr        =" : "                 ",
4358 		    ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4359 #endif
4360 }
4361 
4362 DB_SHOW_COMMAND(prison, db_show_prison_command)
4363 {
4364 	struct prison *pr;
4365 
4366 	if (!have_addr) {
4367 		/*
4368 		 * Show all prisons in the list, and prison0 which is not
4369 		 * listed.
4370 		 */
4371 		db_show_prison(&prison0);
4372 		if (!db_pager_quit) {
4373 			TAILQ_FOREACH(pr, &allprison, pr_list) {
4374 				db_show_prison(pr);
4375 				if (db_pager_quit)
4376 					break;
4377 			}
4378 		}
4379 		return;
4380 	}
4381 
4382 	if (addr == 0)
4383 		pr = &prison0;
4384 	else {
4385 		/* Look for a prison with the ID and with references. */
4386 		TAILQ_FOREACH(pr, &allprison, pr_list)
4387 			if (pr->pr_id == addr && pr->pr_ref > 0)
4388 				break;
4389 		if (pr == NULL)
4390 			/* Look again, without requiring a reference. */
4391 			TAILQ_FOREACH(pr, &allprison, pr_list)
4392 				if (pr->pr_id == addr)
4393 					break;
4394 		if (pr == NULL)
4395 			/* Assume address points to a valid prison. */
4396 			pr = (struct prison *)addr;
4397 	}
4398 	db_show_prison(pr);
4399 }
4400 
4401 #endif /* DDB */
4402