xref: /freebsd/sys/kern/kern_jail.c (revision 830940567b49bb0c08dfaed40418999e76616909)
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/sx.h>
53 #include <sys/sysent.h>
54 #include <sys/namei.h>
55 #include <sys/mount.h>
56 #include <sys/queue.h>
57 #include <sys/socket.h>
58 #include <sys/syscallsubr.h>
59 #include <sys/sysctl.h>
60 #include <sys/vnode.h>
61 
62 #include <net/if.h>
63 #include <net/vnet.h>
64 
65 #include <netinet/in.h>
66 
67 #ifdef DDB
68 #include <ddb/ddb.h>
69 #ifdef INET6
70 #include <netinet6/in6_var.h>
71 #endif /* INET6 */
72 #endif /* DDB */
73 
74 #include <security/mac/mac_framework.h>
75 
76 #define	DEFAULT_HOSTUUID	"00000000-0000-0000-0000-000000000000"
77 
78 MALLOC_DEFINE(M_PRISON, "prison", "Prison structures");
79 
80 /* prison0 describes what is "real" about the system. */
81 struct prison prison0 = {
82 	.pr_id		= 0,
83 	.pr_name	= "0",
84 	.pr_ref		= 1,
85 	.pr_uref	= 1,
86 	.pr_path	= "/",
87 	.pr_securelevel	= -1,
88 	.pr_childmax	= JAIL_MAX,
89 	.pr_hostuuid	= DEFAULT_HOSTUUID,
90 	.pr_children	= LIST_HEAD_INITIALIZER(&prison0.pr_children),
91 #ifdef VIMAGE
92 	.pr_flags	= PR_HOST|PR_VNET,
93 #else
94 	.pr_flags	= PR_HOST,
95 #endif
96 	.pr_allow	= PR_ALLOW_ALL,
97 };
98 MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF);
99 
100 /* allprison and lastprid are protected by allprison_lock. */
101 struct	sx allprison_lock;
102 SX_SYSINIT(allprison_lock, &allprison_lock, "allprison");
103 struct	prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison);
104 int	lastprid = 0;
105 
106 static int do_jail_attach(struct thread *td, struct prison *pr);
107 static void prison_complete(void *context, int pending);
108 static void prison_deref(struct prison *pr, int flags);
109 static char *prison_path(struct prison *pr1, struct prison *pr2);
110 static void prison_remove_one(struct prison *pr);
111 #ifdef INET
112 static int _prison_check_ip4(struct prison *pr, struct in_addr *ia);
113 static int prison_restrict_ip4(struct prison *pr, struct in_addr *newip4);
114 #endif
115 #ifdef INET6
116 static int _prison_check_ip6(struct prison *pr, struct in6_addr *ia6);
117 static int prison_restrict_ip6(struct prison *pr, struct in6_addr *newip6);
118 #endif
119 
120 /* Flags for prison_deref */
121 #define	PD_DEREF	0x01
122 #define	PD_DEUREF	0x02
123 #define	PD_LOCKED	0x04
124 #define	PD_LIST_SLOCKED	0x08
125 #define	PD_LIST_XLOCKED	0x10
126 
127 /*
128  * Parameter names corresponding to PR_* flag values
129  */
130 static char *pr_flag_names[] = {
131 	[0] = "persist",
132 };
133 
134 static char *pr_flag_nonames[] = {
135 	[0] = "nopersist",
136 };
137 
138 struct jailsys_flags {
139 	const char	*name;
140 	unsigned	 disable;
141 	unsigned	 new;
142 } pr_flag_jailsys[] = {
143 	{ "host", 0, PR_HOST },
144 #ifdef VIMAGE
145 	{ "vnet", 0, PR_VNET },
146 #endif
147 #ifdef INET
148 	{ "ip4", PR_IP4_USER | PR_IP4_DISABLE, PR_IP4_USER },
149 #endif
150 #ifdef INET6
151 	{ "ip6", PR_IP6_USER | PR_IP6_DISABLE, PR_IP6_USER },
152 #endif
153 };
154 
155 static char *pr_allow_names[] = {
156 	"allow.set_hostname",
157 	"allow.sysvipc",
158 	"allow.raw_sockets",
159 	"allow.chflags",
160 	"allow.mount",
161 	"allow.quotas",
162 	"allow.socket_af",
163 };
164 
165 static char *pr_allow_nonames[] = {
166 	"allow.noset_hostname",
167 	"allow.nosysvipc",
168 	"allow.noraw_sockets",
169 	"allow.nochflags",
170 	"allow.nomount",
171 	"allow.noquotas",
172 	"allow.nosocket_af",
173 };
174 
175 #define	JAIL_DEFAULT_ALLOW		PR_ALLOW_SET_HOSTNAME
176 #define	JAIL_DEFAULT_ENFORCE_STATFS	2
177 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW;
178 static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
179 #if defined(INET) || defined(INET6)
180 static unsigned jail_max_af_ips = 255;
181 #endif
182 
183 #ifdef INET
184 static int
185 qcmp_v4(const void *ip1, const void *ip2)
186 {
187 	in_addr_t iaa, iab;
188 
189 	/*
190 	 * We need to compare in HBO here to get the list sorted as expected
191 	 * by the result of the code.  Sorting NBO addresses gives you
192 	 * interesting results.  If you do not understand, do not try.
193 	 */
194 	iaa = ntohl(((const struct in_addr *)ip1)->s_addr);
195 	iab = ntohl(((const struct in_addr *)ip2)->s_addr);
196 
197 	/*
198 	 * Do not simply return the difference of the two numbers, the int is
199 	 * not wide enough.
200 	 */
201 	if (iaa > iab)
202 		return (1);
203 	else if (iaa < iab)
204 		return (-1);
205 	else
206 		return (0);
207 }
208 #endif
209 
210 #ifdef INET6
211 static int
212 qcmp_v6(const void *ip1, const void *ip2)
213 {
214 	const struct in6_addr *ia6a, *ia6b;
215 	int i, rc;
216 
217 	ia6a = (const struct in6_addr *)ip1;
218 	ia6b = (const struct in6_addr *)ip2;
219 
220 	rc = 0;
221 	for (i = 0; rc == 0 && i < sizeof(struct in6_addr); i++) {
222 		if (ia6a->s6_addr[i] > ia6b->s6_addr[i])
223 			rc = 1;
224 		else if (ia6a->s6_addr[i] < ia6b->s6_addr[i])
225 			rc = -1;
226 	}
227 	return (rc);
228 }
229 #endif
230 
231 /*
232  * struct jail_args {
233  *	struct jail *jail;
234  * };
235  */
236 int
237 jail(struct thread *td, struct jail_args *uap)
238 {
239 	uint32_t version;
240 	int error;
241 	struct jail j;
242 
243 	error = copyin(uap->jail, &version, sizeof(uint32_t));
244 	if (error)
245 		return (error);
246 
247 	switch (version) {
248 	case 0:
249 	{
250 		struct jail_v0 j0;
251 
252 		/* FreeBSD single IPv4 jails. */
253 		bzero(&j, sizeof(struct jail));
254 		error = copyin(uap->jail, &j0, sizeof(struct jail_v0));
255 		if (error)
256 			return (error);
257 		j.version = j0.version;
258 		j.path = j0.path;
259 		j.hostname = j0.hostname;
260 		j.ip4s = j0.ip_number;
261 		break;
262 	}
263 
264 	case 1:
265 		/*
266 		 * Version 1 was used by multi-IPv4 jail implementations
267 		 * that never made it into the official kernel.
268 		 */
269 		return (EINVAL);
270 
271 	case 2:	/* JAIL_API_VERSION */
272 		/* FreeBSD multi-IPv4/IPv6,noIP jails. */
273 		error = copyin(uap->jail, &j, sizeof(struct jail));
274 		if (error)
275 			return (error);
276 		break;
277 
278 	default:
279 		/* Sci-Fi jails are not supported, sorry. */
280 		return (EINVAL);
281 	}
282 	return (kern_jail(td, &j));
283 }
284 
285 int
286 kern_jail(struct thread *td, struct jail *j)
287 {
288 	struct iovec optiov[2 * (4
289 			    + sizeof(pr_allow_names) / sizeof(pr_allow_names[0])
290 #ifdef INET
291 			    + 1
292 #endif
293 #ifdef INET6
294 			    + 1
295 #endif
296 			    )];
297 	struct uio opt;
298 	char *u_path, *u_hostname, *u_name;
299 #ifdef INET
300 	uint32_t ip4s;
301 	struct in_addr *u_ip4;
302 #endif
303 #ifdef INET6
304 	struct in6_addr *u_ip6;
305 #endif
306 	size_t tmplen;
307 	int error, enforce_statfs, fi;
308 
309 	bzero(&optiov, sizeof(optiov));
310 	opt.uio_iov = optiov;
311 	opt.uio_iovcnt = 0;
312 	opt.uio_offset = -1;
313 	opt.uio_resid = -1;
314 	opt.uio_segflg = UIO_SYSSPACE;
315 	opt.uio_rw = UIO_READ;
316 	opt.uio_td = td;
317 
318 	/* Set permissions for top-level jails from sysctls. */
319 	if (!jailed(td->td_ucred)) {
320 		for (fi = 0; fi < sizeof(pr_allow_names) /
321 		     sizeof(pr_allow_names[0]); fi++) {
322 			optiov[opt.uio_iovcnt].iov_base =
323 			    (jail_default_allow & (1 << fi))
324 			    ? pr_allow_names[fi] : pr_allow_nonames[fi];
325 			optiov[opt.uio_iovcnt].iov_len =
326 			    strlen(optiov[opt.uio_iovcnt].iov_base) + 1;
327 			opt.uio_iovcnt += 2;
328 		}
329 		optiov[opt.uio_iovcnt].iov_base = "enforce_statfs";
330 		optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs");
331 		opt.uio_iovcnt++;
332 		enforce_statfs = jail_default_enforce_statfs;
333 		optiov[opt.uio_iovcnt].iov_base = &enforce_statfs;
334 		optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs);
335 		opt.uio_iovcnt++;
336 	}
337 
338 	tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN;
339 #ifdef INET
340 	ip4s = (j->version == 0) ? 1 : j->ip4s;
341 	if (ip4s > jail_max_af_ips)
342 		return (EINVAL);
343 	tmplen += ip4s * sizeof(struct in_addr);
344 #else
345 	if (j->ip4s > 0)
346 		return (EINVAL);
347 #endif
348 #ifdef INET6
349 	if (j->ip6s > jail_max_af_ips)
350 		return (EINVAL);
351 	tmplen += j->ip6s * sizeof(struct in6_addr);
352 #else
353 	if (j->ip6s > 0)
354 		return (EINVAL);
355 #endif
356 	u_path = malloc(tmplen, M_TEMP, M_WAITOK);
357 	u_hostname = u_path + MAXPATHLEN;
358 	u_name = u_hostname + MAXHOSTNAMELEN;
359 #ifdef INET
360 	u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN);
361 #endif
362 #ifdef INET6
363 #ifdef INET
364 	u_ip6 = (struct in6_addr *)(u_ip4 + ip4s);
365 #else
366 	u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN);
367 #endif
368 #endif
369 	optiov[opt.uio_iovcnt].iov_base = "path";
370 	optiov[opt.uio_iovcnt].iov_len = sizeof("path");
371 	opt.uio_iovcnt++;
372 	optiov[opt.uio_iovcnt].iov_base = u_path;
373 	error = copyinstr(j->path, u_path, MAXPATHLEN,
374 	    &optiov[opt.uio_iovcnt].iov_len);
375 	if (error) {
376 		free(u_path, M_TEMP);
377 		return (error);
378 	}
379 	opt.uio_iovcnt++;
380 	optiov[opt.uio_iovcnt].iov_base = "host.hostname";
381 	optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname");
382 	opt.uio_iovcnt++;
383 	optiov[opt.uio_iovcnt].iov_base = u_hostname;
384 	error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN,
385 	    &optiov[opt.uio_iovcnt].iov_len);
386 	if (error) {
387 		free(u_path, M_TEMP);
388 		return (error);
389 	}
390 	opt.uio_iovcnt++;
391 	if (j->jailname != NULL) {
392 		optiov[opt.uio_iovcnt].iov_base = "name";
393 		optiov[opt.uio_iovcnt].iov_len = sizeof("name");
394 		opt.uio_iovcnt++;
395 		optiov[opt.uio_iovcnt].iov_base = u_name;
396 		error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN,
397 		    &optiov[opt.uio_iovcnt].iov_len);
398 		if (error) {
399 			free(u_path, M_TEMP);
400 			return (error);
401 		}
402 		opt.uio_iovcnt++;
403 	}
404 #ifdef INET
405 	optiov[opt.uio_iovcnt].iov_base = "ip4.addr";
406 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr");
407 	opt.uio_iovcnt++;
408 	optiov[opt.uio_iovcnt].iov_base = u_ip4;
409 	optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr);
410 	if (j->version == 0)
411 		u_ip4->s_addr = j->ip4s;
412 	else {
413 		error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len);
414 		if (error) {
415 			free(u_path, M_TEMP);
416 			return (error);
417 		}
418 	}
419 	opt.uio_iovcnt++;
420 #endif
421 #ifdef INET6
422 	optiov[opt.uio_iovcnt].iov_base = "ip6.addr";
423 	optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr");
424 	opt.uio_iovcnt++;
425 	optiov[opt.uio_iovcnt].iov_base = u_ip6;
426 	optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr);
427 	error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len);
428 	if (error) {
429 		free(u_path, M_TEMP);
430 		return (error);
431 	}
432 	opt.uio_iovcnt++;
433 #endif
434 	KASSERT(opt.uio_iovcnt <= sizeof(optiov) / sizeof(optiov[0]),
435 	    ("kern_jail: too many iovecs (%d)", opt.uio_iovcnt));
436 	error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH);
437 	free(u_path, M_TEMP);
438 	return (error);
439 }
440 
441 
442 /*
443  * struct jail_set_args {
444  *	struct iovec *iovp;
445  *	unsigned int iovcnt;
446  *	int flags;
447  * };
448  */
449 int
450 jail_set(struct thread *td, struct jail_set_args *uap)
451 {
452 	struct uio *auio;
453 	int error;
454 
455 	/* Check that we have an even number of iovecs. */
456 	if (uap->iovcnt & 1)
457 		return (EINVAL);
458 
459 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
460 	if (error)
461 		return (error);
462 	error = kern_jail_set(td, auio, uap->flags);
463 	free(auio, M_IOV);
464 	return (error);
465 }
466 
467 int
468 kern_jail_set(struct thread *td, struct uio *optuio, int flags)
469 {
470 	struct nameidata nd;
471 #ifdef INET
472 	struct in_addr *ip4;
473 #endif
474 #ifdef INET6
475 	struct in6_addr *ip6;
476 #endif
477 	struct vfsopt *opt;
478 	struct vfsoptlist *opts;
479 	struct prison *pr, *deadpr, *mypr, *ppr, *tpr;
480 	struct vnode *root;
481 	char *domain, *errmsg, *host, *name, *p, *path, *uuid;
482 #if defined(INET) || defined(INET6)
483 	struct prison *tppr;
484 	void *op;
485 #endif
486 	unsigned long hid;
487 	size_t namelen, onamelen;
488 	int created, cuflags, descend, enforce, error, errmsg_len, errmsg_pos;
489 	int gotchildmax, gotenforce, gothid, gotslevel;
490 	int fi, jid, jsys, len, level;
491 	int childmax, slevel, vfslocked;
492 #if defined(INET) || defined(INET6)
493 	int ii, ij;
494 #endif
495 #ifdef INET
496 	int ip4s, redo_ip4;
497 #endif
498 #ifdef INET6
499 	int ip6s, redo_ip6;
500 #endif
501 	unsigned pr_flags, ch_flags;
502 	unsigned pr_allow, ch_allow, tallow;
503 	char numbuf[12];
504 
505 	error = priv_check(td, PRIV_JAIL_SET);
506 	if (!error && (flags & JAIL_ATTACH))
507 		error = priv_check(td, PRIV_JAIL_ATTACH);
508 	if (error)
509 		return (error);
510 	mypr = ppr = td->td_ucred->cr_prison;
511 	if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0)
512 		return (EPERM);
513 	if (flags & ~JAIL_SET_MASK)
514 		return (EINVAL);
515 
516 	/*
517 	 * Check all the parameters before committing to anything.  Not all
518 	 * errors can be caught early, but we may as well try.  Also, this
519 	 * takes care of some expensive stuff (path lookup) before getting
520 	 * the allprison lock.
521 	 *
522 	 * XXX Jails are not filesystems, and jail parameters are not mount
523 	 *     options.  But it makes more sense to re-use the vfsopt code
524 	 *     than duplicate it under a different name.
525 	 */
526 	error = vfs_buildopts(optuio, &opts);
527 	if (error)
528 		return (error);
529 #ifdef INET
530 	ip4 = NULL;
531 #endif
532 #ifdef INET6
533 	ip6 = NULL;
534 #endif
535 
536 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
537 	if (error == ENOENT)
538 		jid = 0;
539 	else if (error != 0)
540 		goto done_free;
541 
542 	error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel));
543 	if (error == ENOENT)
544 		gotslevel = 0;
545 	else if (error != 0)
546 		goto done_free;
547 	else
548 		gotslevel = 1;
549 
550 	error =
551 	    vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax));
552 	if (error == ENOENT)
553 		gotchildmax = 0;
554 	else if (error != 0)
555 		goto done_free;
556 	else
557 		gotchildmax = 1;
558 
559 	error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce));
560 	gotenforce = (error == 0);
561 	if (gotenforce) {
562 		if (enforce < 0 || enforce > 2)
563 			return (EINVAL);
564 	} else if (error != ENOENT)
565 		goto done_free;
566 
567 	pr_flags = ch_flags = 0;
568 	for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
569 	    fi++) {
570 		if (pr_flag_names[fi] == NULL)
571 			continue;
572 		vfs_flagopt(opts, pr_flag_names[fi], &pr_flags, 1 << fi);
573 		vfs_flagopt(opts, pr_flag_nonames[fi], &ch_flags, 1 << fi);
574 	}
575 	ch_flags |= pr_flags;
576 	for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
577 	    fi++) {
578 		error = vfs_copyopt(opts, pr_flag_jailsys[fi].name, &jsys,
579 		    sizeof(jsys));
580 		if (error == ENOENT)
581 			continue;
582 		if (error != 0)
583 			goto done_free;
584 		switch (jsys) {
585 		case JAIL_SYS_DISABLE:
586 			if (!pr_flag_jailsys[fi].disable) {
587 				error = EINVAL;
588 				goto done_free;
589 			}
590 			pr_flags |= pr_flag_jailsys[fi].disable;
591 			break;
592 		case JAIL_SYS_NEW:
593 			pr_flags |= pr_flag_jailsys[fi].new;
594 			break;
595 		case JAIL_SYS_INHERIT:
596 			break;
597 		default:
598 			error = EINVAL;
599 			goto done_free;
600 		}
601 		ch_flags |=
602 		    pr_flag_jailsys[fi].new | pr_flag_jailsys[fi].disable;
603 	}
604 	if ((flags & (JAIL_CREATE | JAIL_UPDATE | JAIL_ATTACH)) == JAIL_CREATE
605 	    && !(pr_flags & PR_PERSIST)) {
606 		error = EINVAL;
607 		vfs_opterror(opts, "new jail must persist or attach");
608 		goto done_errmsg;
609 	}
610 #ifdef VIMAGE
611 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) {
612 		error = EINVAL;
613 		vfs_opterror(opts, "vnet cannot be changed after creation");
614 		goto done_errmsg;
615 	}
616 #endif
617 #ifdef INET
618 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) {
619 		error = EINVAL;
620 		vfs_opterror(opts, "ip4 cannot be changed after creation");
621 		goto done_errmsg;
622 	}
623 #endif
624 #ifdef INET6
625 	if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) {
626 		error = EINVAL;
627 		vfs_opterror(opts, "ip6 cannot be changed after creation");
628 		goto done_errmsg;
629 	}
630 #endif
631 
632 	pr_allow = ch_allow = 0;
633 	for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
634 	    fi++) {
635 		vfs_flagopt(opts, pr_allow_names[fi], &pr_allow, 1 << fi);
636 		vfs_flagopt(opts, pr_allow_nonames[fi], &ch_allow, 1 << fi);
637 	}
638 	ch_allow |= pr_allow;
639 
640 	error = vfs_getopt(opts, "name", (void **)&name, &len);
641 	if (error == ENOENT)
642 		name = NULL;
643 	else if (error != 0)
644 		goto done_free;
645 	else {
646 		if (len == 0 || name[len - 1] != '\0') {
647 			error = EINVAL;
648 			goto done_free;
649 		}
650 		if (len > MAXHOSTNAMELEN) {
651 			error = ENAMETOOLONG;
652 			goto done_free;
653 		}
654 	}
655 
656 	error = vfs_getopt(opts, "host.hostname", (void **)&host, &len);
657 	if (error == ENOENT)
658 		host = NULL;
659 	else if (error != 0)
660 		goto done_free;
661 	else {
662 		ch_flags |= PR_HOST;
663 		pr_flags |= PR_HOST;
664 		if (len == 0 || host[len - 1] != '\0') {
665 			error = EINVAL;
666 			goto done_free;
667 		}
668 		if (len > MAXHOSTNAMELEN) {
669 			error = ENAMETOOLONG;
670 			goto done_free;
671 		}
672 	}
673 
674 	error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len);
675 	if (error == ENOENT)
676 		domain = NULL;
677 	else if (error != 0)
678 		goto done_free;
679 	else {
680 		ch_flags |= PR_HOST;
681 		pr_flags |= PR_HOST;
682 		if (len == 0 || domain[len - 1] != '\0') {
683 			error = EINVAL;
684 			goto done_free;
685 		}
686 		if (len > MAXHOSTNAMELEN) {
687 			error = ENAMETOOLONG;
688 			goto done_free;
689 		}
690 	}
691 
692 	error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len);
693 	if (error == ENOENT)
694 		uuid = NULL;
695 	else if (error != 0)
696 		goto done_free;
697 	else {
698 		ch_flags |= PR_HOST;
699 		pr_flags |= PR_HOST;
700 		if (len == 0 || uuid[len - 1] != '\0') {
701 			error = EINVAL;
702 			goto done_free;
703 		}
704 		if (len > HOSTUUIDLEN) {
705 			error = ENAMETOOLONG;
706 			goto done_free;
707 		}
708 	}
709 
710 #ifdef COMPAT_IA32
711 	if (td->td_proc->p_sysent->sv_flags & SV_IA32) {
712 		uint32_t hid32;
713 
714 		error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32));
715 		hid = hid32;
716 	} else
717 #endif
718 		error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid));
719 	if (error == ENOENT)
720 		gothid = 0;
721 	else if (error != 0)
722 		goto done_free;
723 	else {
724 		gothid = 1;
725 		ch_flags |= PR_HOST;
726 		pr_flags |= PR_HOST;
727 	}
728 
729 #ifdef INET
730 	error = vfs_getopt(opts, "ip4.addr", &op, &ip4s);
731 	if (error == ENOENT)
732 		ip4s = (pr_flags & PR_IP4_DISABLE) ? 0 : -1;
733 	else if (error != 0)
734 		goto done_free;
735 	else if (ip4s & (sizeof(*ip4) - 1)) {
736 		error = EINVAL;
737 		goto done_free;
738 	} else {
739 		ch_flags |= PR_IP4_USER | PR_IP4_DISABLE;
740 		if (ip4s == 0)
741 			pr_flags |= PR_IP4_USER | PR_IP4_DISABLE;
742 		else {
743 			pr_flags = (pr_flags & ~PR_IP4_DISABLE) | PR_IP4_USER;
744 			ip4s /= sizeof(*ip4);
745 			if (ip4s > jail_max_af_ips) {
746 				error = EINVAL;
747 				vfs_opterror(opts, "too many IPv4 addresses");
748 				goto done_errmsg;
749 			}
750 			ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
751 			bcopy(op, ip4, ip4s * sizeof(*ip4));
752 			/*
753 			 * IP addresses are all sorted but ip[0] to preserve
754 			 * the primary IP address as given from userland.
755 			 * This special IP is used for unbound outgoing
756 			 * connections as well for "loopback" traffic.
757 			 */
758 			if (ip4s > 1)
759 				qsort(ip4 + 1, ip4s - 1, sizeof(*ip4), qcmp_v4);
760 			/*
761 			 * Check for duplicate addresses and do some simple
762 			 * zero and broadcast checks. If users give other bogus
763 			 * addresses it is their problem.
764 			 *
765 			 * We do not have to care about byte order for these
766 			 * checks so we will do them in NBO.
767 			 */
768 			for (ii = 0; ii < ip4s; ii++) {
769 				if (ip4[ii].s_addr == INADDR_ANY ||
770 				    ip4[ii].s_addr == INADDR_BROADCAST) {
771 					error = EINVAL;
772 					goto done_free;
773 				}
774 				if ((ii+1) < ip4s &&
775 				    (ip4[0].s_addr == ip4[ii+1].s_addr ||
776 				     ip4[ii].s_addr == ip4[ii+1].s_addr)) {
777 					error = EINVAL;
778 					goto done_free;
779 				}
780 			}
781 		}
782 	}
783 #endif
784 
785 #ifdef INET6
786 	error = vfs_getopt(opts, "ip6.addr", &op, &ip6s);
787 	if (error == ENOENT)
788 		ip6s = (pr_flags & PR_IP6_DISABLE) ? 0 : -1;
789 	else if (error != 0)
790 		goto done_free;
791 	else if (ip6s & (sizeof(*ip6) - 1)) {
792 		error = EINVAL;
793 		goto done_free;
794 	} else {
795 		ch_flags |= PR_IP6_USER | PR_IP6_DISABLE;
796 		if (ip6s == 0)
797 			pr_flags |= PR_IP6_USER | PR_IP6_DISABLE;
798 		else {
799 			pr_flags = (pr_flags & ~PR_IP6_DISABLE) | PR_IP6_USER;
800 			ip6s /= sizeof(*ip6);
801 			if (ip6s > jail_max_af_ips) {
802 				error = EINVAL;
803 				vfs_opterror(opts, "too many IPv6 addresses");
804 				goto done_errmsg;
805 			}
806 			ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
807 			bcopy(op, ip6, ip6s * sizeof(*ip6));
808 			if (ip6s > 1)
809 				qsort(ip6 + 1, ip6s - 1, sizeof(*ip6), qcmp_v6);
810 			for (ii = 0; ii < ip6s; ii++) {
811 				if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) {
812 					error = EINVAL;
813 					goto done_free;
814 				}
815 				if ((ii+1) < ip6s &&
816 				    (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) ||
817 				     IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1])))
818 				{
819 					error = EINVAL;
820 					goto done_free;
821 				}
822 			}
823 		}
824 	}
825 #endif
826 
827 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
828 	if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
829 		error = EINVAL;
830 		vfs_opterror(opts,
831 		    "vnet jails cannot have IP address restrictions");
832 		goto done_errmsg;
833 	}
834 #endif
835 
836 	root = NULL;
837 	error = vfs_getopt(opts, "path", (void **)&path, &len);
838 	if (error == ENOENT)
839 		path = NULL;
840 	else if (error != 0)
841 		goto done_free;
842 	else {
843 		if (flags & JAIL_UPDATE) {
844 			error = EINVAL;
845 			vfs_opterror(opts,
846 			    "path cannot be changed after creation");
847 			goto done_errmsg;
848 		}
849 		if (len == 0 || path[len - 1] != '\0') {
850 			error = EINVAL;
851 			goto done_free;
852 		}
853 		if (len < 2 || (len == 2 && path[0] == '/'))
854 			path = NULL;
855 		else {
856 			/* Leave room for a real-root full pathname. */
857 			if (len + (path[0] == '/' && strcmp(mypr->pr_path, "/")
858 			    ? strlen(mypr->pr_path) : 0) > MAXPATHLEN) {
859 				error = ENAMETOOLONG;
860 				goto done_free;
861 			}
862 			NDINIT(&nd, LOOKUP, MPSAFE | FOLLOW, UIO_SYSSPACE,
863 			    path, td);
864 			error = namei(&nd);
865 			if (error)
866 				goto done_free;
867 			vfslocked = NDHASGIANT(&nd);
868 			root = nd.ni_vp;
869 			NDFREE(&nd, NDF_ONLY_PNBUF);
870 			if (root->v_type != VDIR) {
871 				error = ENOTDIR;
872 				vrele(root);
873 				VFS_UNLOCK_GIANT(vfslocked);
874 				goto done_free;
875 			}
876 			VFS_UNLOCK_GIANT(vfslocked);
877 		}
878 	}
879 
880 	/*
881 	 * Grab the allprison lock before letting modules check their
882 	 * parameters.  Once we have it, do not let go so we'll have a
883 	 * consistent view of the OSD list.
884 	 */
885 	sx_xlock(&allprison_lock);
886 	error = osd_jail_call(NULL, PR_METHOD_CHECK, opts);
887 	if (error)
888 		goto done_unlock_list;
889 
890 	/* By now, all parameters should have been noted. */
891 	TAILQ_FOREACH(opt, opts, link) {
892 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
893 			error = EINVAL;
894 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
895 			goto done_unlock_list;
896 		}
897 	}
898 
899 	/*
900 	 * See if we are creating a new record or updating an existing one.
901 	 * This abuses the file error codes ENOENT and EEXIST.
902 	 */
903 	cuflags = flags & (JAIL_CREATE | JAIL_UPDATE);
904 	if (!cuflags) {
905 		error = EINVAL;
906 		vfs_opterror(opts, "no valid operation (create or update)");
907 		goto done_unlock_list;
908 	}
909 	pr = NULL;
910 	if (jid != 0) {
911 		/*
912 		 * See if a requested jid already exists.  There is an
913 		 * information leak here if the jid exists but is not within
914 		 * the caller's jail hierarchy.  Jail creators will get EEXIST
915 		 * even though they cannot see the jail, and CREATE | UPDATE
916 		 * will return ENOENT which is not normally a valid error.
917 		 */
918 		if (jid < 0) {
919 			error = EINVAL;
920 			vfs_opterror(opts, "negative jid");
921 			goto done_unlock_list;
922 		}
923 		pr = prison_find(jid);
924 		if (pr != NULL) {
925 			ppr = pr->pr_parent;
926 			/* Create: jid must not exist. */
927 			if (cuflags == JAIL_CREATE) {
928 				mtx_unlock(&pr->pr_mtx);
929 				error = EEXIST;
930 				vfs_opterror(opts, "jail %d already exists",
931 				    jid);
932 				goto done_unlock_list;
933 			}
934 			if (!prison_ischild(mypr, pr)) {
935 				mtx_unlock(&pr->pr_mtx);
936 				pr = NULL;
937 			} else if (pr->pr_uref == 0) {
938 				if (!(flags & JAIL_DYING)) {
939 					mtx_unlock(&pr->pr_mtx);
940 					error = ENOENT;
941 					vfs_opterror(opts, "jail %d is dying",
942 					    jid);
943 					goto done_unlock_list;
944 				} else if ((flags & JAIL_ATTACH) ||
945 				    (pr_flags & PR_PERSIST)) {
946 					/*
947 					 * A dying jail might be resurrected
948 					 * (via attach or persist), but first
949 					 * it must determine if another jail
950 					 * has claimed its name.  Accomplish
951 					 * this by implicitly re-setting the
952 					 * name.
953 					 */
954 					if (name == NULL)
955 						name = prison_name(mypr, pr);
956 				}
957 			}
958 		}
959 		if (pr == NULL) {
960 			/* Update: jid must exist. */
961 			if (cuflags == JAIL_UPDATE) {
962 				error = ENOENT;
963 				vfs_opterror(opts, "jail %d not found", jid);
964 				goto done_unlock_list;
965 			}
966 		}
967 	}
968 	/*
969 	 * If the caller provided a name, look for a jail by that name.
970 	 * This has different semantics for creates and updates keyed by jid
971 	 * (where the name must not already exist in a different jail),
972 	 * and updates keyed by the name itself (where the name must exist
973 	 * because that is the jail being updated).
974 	 */
975 	if (name != NULL) {
976 		p = strrchr(name, '.');
977 		if (p != NULL) {
978 			/*
979 			 * This is a hierarchical name.  Split it into the
980 			 * parent and child names, and make sure the parent
981 			 * exists or matches an already found jail.
982 			 */
983 			*p = '\0';
984 			if (pr != NULL) {
985 				if (strncmp(name, ppr->pr_name, p - name) ||
986 				    ppr->pr_name[p - name] != '\0') {
987 					mtx_unlock(&pr->pr_mtx);
988 					error = EINVAL;
989 					vfs_opterror(opts,
990 					    "cannot change jail's parent");
991 					goto done_unlock_list;
992 				}
993 			} else {
994 				ppr = prison_find_name(mypr, name);
995 				if (ppr == NULL) {
996 					error = ENOENT;
997 					vfs_opterror(opts,
998 					    "jail \"%s\" not found", name);
999 					goto done_unlock_list;
1000 				}
1001 				mtx_unlock(&ppr->pr_mtx);
1002 			}
1003 			name = p + 1;
1004 		}
1005 		if (name[0] != '\0') {
1006 			namelen =
1007 			    (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1;
1008  name_again:
1009 			deadpr = NULL;
1010 			FOREACH_PRISON_CHILD(ppr, tpr) {
1011 				if (tpr != pr && tpr->pr_ref > 0 &&
1012 				    !strcmp(tpr->pr_name + namelen, name)) {
1013 					if (pr == NULL &&
1014 					    cuflags != JAIL_CREATE) {
1015 						mtx_lock(&tpr->pr_mtx);
1016 						if (tpr->pr_ref > 0) {
1017 							/*
1018 							 * Use this jail
1019 							 * for updates.
1020 							 */
1021 							if (tpr->pr_uref > 0) {
1022 								pr = tpr;
1023 								break;
1024 							}
1025 							deadpr = tpr;
1026 						}
1027 						mtx_unlock(&tpr->pr_mtx);
1028 					} else if (tpr->pr_uref > 0) {
1029 						/*
1030 						 * Create, or update(jid):
1031 						 * name must not exist in an
1032 						 * active sibling jail.
1033 						 */
1034 						error = EEXIST;
1035 						if (pr != NULL)
1036 							mtx_unlock(&pr->pr_mtx);
1037 						vfs_opterror(opts,
1038 						   "jail \"%s\" already exists",
1039 						   name);
1040 						goto done_unlock_list;
1041 					}
1042 				}
1043 			}
1044 			/* If no active jail is found, use a dying one. */
1045 			if (deadpr != NULL && pr == NULL) {
1046 				if (flags & JAIL_DYING) {
1047 					mtx_lock(&deadpr->pr_mtx);
1048 					if (deadpr->pr_ref == 0) {
1049 						mtx_unlock(&deadpr->pr_mtx);
1050 						goto name_again;
1051 					}
1052 					pr = deadpr;
1053 				} else if (cuflags == JAIL_UPDATE) {
1054 					error = ENOENT;
1055 					vfs_opterror(opts,
1056 					    "jail \"%s\" is dying", name);
1057 					goto done_unlock_list;
1058 				}
1059 			}
1060 			/* Update: name must exist if no jid. */
1061 			else if (cuflags == JAIL_UPDATE && pr == NULL) {
1062 				error = ENOENT;
1063 				vfs_opterror(opts, "jail \"%s\" not found",
1064 				    name);
1065 				goto done_unlock_list;
1066 			}
1067 		}
1068 	}
1069 	/* Update: must provide a jid or name. */
1070 	else if (cuflags == JAIL_UPDATE && pr == NULL) {
1071 		error = ENOENT;
1072 		vfs_opterror(opts, "update specified no jail");
1073 		goto done_unlock_list;
1074 	}
1075 
1076 	/* If there's no prison to update, create a new one and link it in. */
1077 	if (pr == NULL) {
1078 		for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent)
1079 			if (tpr->pr_childcount >= tpr->pr_childmax) {
1080 				error = EPERM;
1081 				vfs_opterror(opts, "prison limit exceeded");
1082 				goto done_unlock_list;
1083 			}
1084 		created = 1;
1085 		mtx_lock(&ppr->pr_mtx);
1086 		if (ppr->pr_ref == 0 || (ppr->pr_flags & PR_REMOVE)) {
1087 			mtx_unlock(&ppr->pr_mtx);
1088 			error = ENOENT;
1089 			vfs_opterror(opts, "parent jail went away!");
1090 			goto done_unlock_list;
1091 		}
1092 		ppr->pr_ref++;
1093 		ppr->pr_uref++;
1094 		mtx_unlock(&ppr->pr_mtx);
1095 		pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO);
1096 		if (jid == 0) {
1097 			/* Find the next free jid. */
1098 			jid = lastprid + 1;
1099  findnext:
1100 			if (jid == JAIL_MAX)
1101 				jid = 1;
1102 			TAILQ_FOREACH(tpr, &allprison, pr_list) {
1103 				if (tpr->pr_id < jid)
1104 					continue;
1105 				if (tpr->pr_id > jid || tpr->pr_ref == 0) {
1106 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1107 					break;
1108 				}
1109 				if (jid == lastprid) {
1110 					error = EAGAIN;
1111 					vfs_opterror(opts,
1112 					    "no available jail IDs");
1113 					free(pr, M_PRISON);
1114 					prison_deref(ppr, PD_DEREF |
1115 					    PD_DEUREF | PD_LIST_XLOCKED);
1116 					goto done_releroot;
1117 				}
1118 				jid++;
1119 				goto findnext;
1120 			}
1121 			lastprid = jid;
1122 		} else {
1123 			/*
1124 			 * The jail already has a jid (that did not yet exist),
1125 			 * so just find where to insert it.
1126 			 */
1127 			TAILQ_FOREACH(tpr, &allprison, pr_list)
1128 				if (tpr->pr_id >= jid) {
1129 					TAILQ_INSERT_BEFORE(tpr, pr, pr_list);
1130 					break;
1131 				}
1132 		}
1133 		if (tpr == NULL)
1134 			TAILQ_INSERT_TAIL(&allprison, pr, pr_list);
1135 		LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling);
1136 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
1137 			tpr->pr_childcount++;
1138 
1139 		pr->pr_parent = ppr;
1140 		pr->pr_id = jid;
1141 
1142 		/* Set some default values, and inherit some from the parent. */
1143 		if (name == NULL)
1144 			name = "";
1145 		if (path == NULL) {
1146 			path = "/";
1147 			root = mypr->pr_root;
1148 			vref(root);
1149 		}
1150 		strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN);
1151 		pr->pr_flags |= PR_HOST;
1152 #if defined(INET) || defined(INET6)
1153 #ifdef VIMAGE
1154 		if (!(pr_flags & PR_VNET))
1155 #endif
1156 		{
1157 #ifdef INET
1158 			if (!(ch_flags & PR_IP4_USER))
1159 				pr->pr_flags |=
1160 				    PR_IP4 | PR_IP4_USER | PR_IP4_DISABLE;
1161 			else if (!(pr_flags & PR_IP4_USER)) {
1162 				pr->pr_flags |= ppr->pr_flags & PR_IP4;
1163 				if (ppr->pr_ip4 != NULL) {
1164 					pr->pr_ip4s = ppr->pr_ip4s;
1165 					pr->pr_ip4 = malloc(pr->pr_ip4s *
1166 					    sizeof(struct in_addr), M_PRISON,
1167 					    M_WAITOK);
1168 					bcopy(ppr->pr_ip4, pr->pr_ip4,
1169 					    pr->pr_ip4s * sizeof(*pr->pr_ip4));
1170 				}
1171 			}
1172 #endif
1173 #ifdef INET6
1174 			if (!(ch_flags & PR_IP6_USER))
1175 				pr->pr_flags |=
1176 				    PR_IP6 | PR_IP6_USER | PR_IP6_DISABLE;
1177 			else if (!(pr_flags & PR_IP6_USER)) {
1178 				pr->pr_flags |= ppr->pr_flags & PR_IP6;
1179 				if (ppr->pr_ip6 != NULL) {
1180 					pr->pr_ip6s = ppr->pr_ip6s;
1181 					pr->pr_ip6 = malloc(pr->pr_ip6s *
1182 					    sizeof(struct in6_addr), M_PRISON,
1183 					    M_WAITOK);
1184 					bcopy(ppr->pr_ip6, pr->pr_ip6,
1185 					    pr->pr_ip6s * sizeof(*pr->pr_ip6));
1186 				}
1187 			}
1188 #endif
1189 		}
1190 #endif
1191 		pr->pr_securelevel = ppr->pr_securelevel;
1192 		pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow;
1193 		pr->pr_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS;
1194 
1195 		LIST_INIT(&pr->pr_children);
1196 		mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK);
1197 
1198 #ifdef VIMAGE
1199 		/* Allocate a new vnet if specified. */
1200 		pr->pr_vnet = (pr_flags & PR_VNET)
1201 		    ? vnet_alloc() : ppr->pr_vnet;
1202 #endif
1203 		/*
1204 		 * Allocate a dedicated cpuset for each jail.
1205 		 * Unlike other initial settings, this may return an erorr.
1206 		 */
1207 		error = cpuset_create_root(ppr, &pr->pr_cpuset);
1208 		if (error) {
1209 			prison_deref(pr, PD_LIST_XLOCKED);
1210 			goto done_releroot;
1211 		}
1212 
1213 		mtx_lock(&pr->pr_mtx);
1214 		/*
1215 		 * New prisons do not yet have a reference, because we do not
1216 		 * want other to see the incomplete prison once the
1217 		 * allprison_lock is downgraded.
1218 		 */
1219 	} else {
1220 		created = 0;
1221 		/*
1222 		 * Grab a reference for existing prisons, to ensure they
1223 		 * continue to exist for the duration of the call.
1224 		 */
1225 		pr->pr_ref++;
1226 #if defined(VIMAGE) && (defined(INET) || defined(INET6))
1227 		if ((pr->pr_flags & PR_VNET) &&
1228 		    (ch_flags & (PR_IP4_USER | PR_IP6_USER))) {
1229 			error = EINVAL;
1230 			vfs_opterror(opts,
1231 			    "vnet jails cannot have IP address restrictions");
1232 			goto done_deref_locked;
1233 		}
1234 #endif
1235 #ifdef INET
1236 		if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1237 			error = EINVAL;
1238 			vfs_opterror(opts,
1239 			    "ip4 cannot be changed after creation");
1240 			goto done_deref_locked;
1241 		}
1242 #endif
1243 #ifdef INET6
1244 		if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) {
1245 			error = EINVAL;
1246 			vfs_opterror(opts,
1247 			    "ip6 cannot be changed after creation");
1248 			goto done_deref_locked;
1249 		}
1250 #endif
1251 	}
1252 
1253 	/* Do final error checking before setting anything. */
1254 	if (gotslevel) {
1255 		if (slevel < ppr->pr_securelevel) {
1256 			error = EPERM;
1257 			goto done_deref_locked;
1258 		}
1259 	}
1260 	if (gotchildmax) {
1261 		if (childmax >= ppr->pr_childmax) {
1262 			error = EPERM;
1263 			goto done_deref_locked;
1264 		}
1265 	}
1266 	if (gotenforce) {
1267 		if (enforce < ppr->pr_enforce_statfs) {
1268 			error = EPERM;
1269 			goto done_deref_locked;
1270 		}
1271 	}
1272 #ifdef INET
1273 	if (ip4s > 0) {
1274 		if (ppr->pr_flags & PR_IP4) {
1275 			/*
1276 			 * Make sure the new set of IP addresses is a
1277 			 * subset of the parent's list.  Don't worry
1278 			 * about the parent being unlocked, as any
1279 			 * setting is done with allprison_lock held.
1280 			 */
1281 			for (ij = 0; ij < ppr->pr_ip4s; ij++)
1282 				if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
1283 					break;
1284 			if (ij == ppr->pr_ip4s) {
1285 				error = EPERM;
1286 				goto done_deref_locked;
1287 			}
1288 			if (ip4s > 1) {
1289 				for (ii = ij = 1; ii < ip4s; ii++) {
1290 					if (ip4[ii].s_addr ==
1291 					    ppr->pr_ip4[0].s_addr)
1292 						continue;
1293 					for (; ij < ppr->pr_ip4s; ij++)
1294 						if (ip4[ii].s_addr ==
1295 						    ppr->pr_ip4[ij].s_addr)
1296 							break;
1297 					if (ij == ppr->pr_ip4s)
1298 						break;
1299 				}
1300 				if (ij == ppr->pr_ip4s) {
1301 					error = EPERM;
1302 					goto done_deref_locked;
1303 				}
1304 			}
1305 		}
1306 		/*
1307 		 * Check for conflicting IP addresses.  We permit them
1308 		 * if there is no more than one IP on each jail.  If
1309 		 * there is a duplicate on a jail with more than one
1310 		 * IP stop checking and return error.
1311 		 */
1312 		tppr = ppr;
1313 #ifdef VIMAGE
1314 		for (; tppr != &prison0; tppr = tppr->pr_parent)
1315 			if (tppr->pr_flags & PR_VNET)
1316 				break;
1317 #endif
1318 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1319 			if (tpr == pr ||
1320 #ifdef VIMAGE
1321 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1322 #endif
1323 			    tpr->pr_uref == 0) {
1324 				descend = 0;
1325 				continue;
1326 			}
1327 			if (!(tpr->pr_flags & PR_IP4_USER))
1328 				continue;
1329 			descend = 0;
1330 			if (tpr->pr_ip4 == NULL ||
1331 			    (ip4s == 1 && tpr->pr_ip4s == 1))
1332 				continue;
1333 			for (ii = 0; ii < ip4s; ii++) {
1334 				if (_prison_check_ip4(tpr, &ip4[ii]) == 0) {
1335 					error = EADDRINUSE;
1336 					vfs_opterror(opts,
1337 					    "IPv4 addresses clash");
1338 					goto done_deref_locked;
1339 				}
1340 			}
1341 		}
1342 	}
1343 #endif
1344 #ifdef INET6
1345 	if (ip6s > 0) {
1346 		if (ppr->pr_flags & PR_IP6) {
1347 			/*
1348 			 * Make sure the new set of IP addresses is a
1349 			 * subset of the parent's list.
1350 			 */
1351 			for (ij = 0; ij < ppr->pr_ip6s; ij++)
1352 				if (IN6_ARE_ADDR_EQUAL(&ip6[0],
1353 				    &ppr->pr_ip6[ij]))
1354 					break;
1355 			if (ij == ppr->pr_ip6s) {
1356 				error = EPERM;
1357 				goto done_deref_locked;
1358 			}
1359 			if (ip6s > 1) {
1360 				for (ii = ij = 1; ii < ip6s; ii++) {
1361 					if (IN6_ARE_ADDR_EQUAL(&ip6[ii],
1362 					     &ppr->pr_ip6[0]))
1363 						continue;
1364 					for (; ij < ppr->pr_ip6s; ij++)
1365 						if (IN6_ARE_ADDR_EQUAL(
1366 						    &ip6[ii], &ppr->pr_ip6[ij]))
1367 							break;
1368 					if (ij == ppr->pr_ip6s)
1369 						break;
1370 				}
1371 				if (ij == ppr->pr_ip6s) {
1372 					error = EPERM;
1373 					goto done_deref_locked;
1374 				}
1375 			}
1376 		}
1377 		/* Check for conflicting IP addresses. */
1378 		tppr = ppr;
1379 #ifdef VIMAGE
1380 		for (; tppr != &prison0; tppr = tppr->pr_parent)
1381 			if (tppr->pr_flags & PR_VNET)
1382 				break;
1383 #endif
1384 		FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) {
1385 			if (tpr == pr ||
1386 #ifdef VIMAGE
1387 			    (tpr != tppr && (tpr->pr_flags & PR_VNET)) ||
1388 #endif
1389 			    tpr->pr_uref == 0) {
1390 				descend = 0;
1391 				continue;
1392 			}
1393 			if (!(tpr->pr_flags & PR_IP6_USER))
1394 				continue;
1395 			descend = 0;
1396 			if (tpr->pr_ip6 == NULL ||
1397 			    (ip6s == 1 && tpr->pr_ip6s == 1))
1398 				continue;
1399 			for (ii = 0; ii < ip6s; ii++) {
1400 				if (_prison_check_ip6(tpr, &ip6[ii]) == 0) {
1401 					error = EADDRINUSE;
1402 					vfs_opterror(opts,
1403 					    "IPv6 addresses clash");
1404 					goto done_deref_locked;
1405 				}
1406 			}
1407 		}
1408 	}
1409 #endif
1410 	onamelen = namelen = 0;
1411 	if (name != NULL) {
1412 		/* Give a default name of the jid. */
1413 		if (name[0] == '\0')
1414 			snprintf(name = numbuf, sizeof(numbuf), "%d", jid);
1415 		else if (strtoul(name, &p, 10) != jid && *p == '\0') {
1416 			error = EINVAL;
1417 			vfs_opterror(opts, "name cannot be numeric");
1418 			goto done_deref_locked;
1419 		}
1420 		/*
1421 		 * Make sure the name isn't too long for the prison or its
1422 		 * children.
1423 		 */
1424 		onamelen = strlen(pr->pr_name);
1425 		namelen = strlen(name);
1426 		if (strlen(ppr->pr_name) + namelen + 2 > sizeof(pr->pr_name)) {
1427 			error = ENAMETOOLONG;
1428 			goto done_deref_locked;
1429 		}
1430 		FOREACH_PRISON_DESCENDANT(pr, tpr, descend) {
1431 			if (strlen(tpr->pr_name) + (namelen - onamelen) >=
1432 			    sizeof(pr->pr_name)) {
1433 				error = ENAMETOOLONG;
1434 				goto done_deref_locked;
1435 			}
1436 		}
1437 	}
1438 	if (pr_allow & ~ppr->pr_allow) {
1439 		error = EPERM;
1440 		goto done_deref_locked;
1441 	}
1442 
1443 	/* Set the parameters of the prison. */
1444 #ifdef INET
1445 	redo_ip4 = 0;
1446 	if (pr_flags & PR_IP4_USER) {
1447 		pr->pr_flags |= PR_IP4;
1448 		free(pr->pr_ip4, M_PRISON);
1449 		pr->pr_ip4s = ip4s;
1450 		pr->pr_ip4 = ip4;
1451 		ip4 = NULL;
1452 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1453 #ifdef VIMAGE
1454 			if (tpr->pr_flags & PR_VNET) {
1455 				descend = 0;
1456 				continue;
1457 			}
1458 #endif
1459 			if (prison_restrict_ip4(tpr, NULL)) {
1460 				redo_ip4 = 1;
1461 				descend = 0;
1462 			}
1463 		}
1464 	}
1465 #endif
1466 #ifdef INET6
1467 	redo_ip6 = 0;
1468 	if (pr_flags & PR_IP6_USER) {
1469 		pr->pr_flags |= PR_IP6;
1470 		free(pr->pr_ip6, M_PRISON);
1471 		pr->pr_ip6s = ip6s;
1472 		pr->pr_ip6 = ip6;
1473 		ip6 = NULL;
1474 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1475 #ifdef VIMAGE
1476 			if (tpr->pr_flags & PR_VNET) {
1477 				descend = 0;
1478 				continue;
1479 			}
1480 #endif
1481 			if (prison_restrict_ip6(tpr, NULL)) {
1482 				redo_ip6 = 1;
1483 				descend = 0;
1484 			}
1485 		}
1486 	}
1487 #endif
1488 	if (gotslevel) {
1489 		pr->pr_securelevel = slevel;
1490 		/* Set all child jails to be at least this level. */
1491 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1492 			if (tpr->pr_securelevel < slevel)
1493 				tpr->pr_securelevel = slevel;
1494 	}
1495 	if (gotchildmax) {
1496 		pr->pr_childmax = childmax;
1497 		/* Set all child jails to under this limit. */
1498 		FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level)
1499 			if (tpr->pr_childmax > childmax - level)
1500 				tpr->pr_childmax = childmax > level
1501 				    ? childmax - level : 0;
1502 	}
1503 	if (gotenforce) {
1504 		pr->pr_enforce_statfs = enforce;
1505 		/* Pass this restriction on to the children. */
1506 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1507 			if (tpr->pr_enforce_statfs < enforce)
1508 				tpr->pr_enforce_statfs = enforce;
1509 	}
1510 	if (name != NULL) {
1511 		if (ppr == &prison0)
1512 			strlcpy(pr->pr_name, name, sizeof(pr->pr_name));
1513 		else
1514 			snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s",
1515 			    ppr->pr_name, name);
1516 		/* Change this component of child names. */
1517 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1518 			bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen,
1519 			    strlen(tpr->pr_name + onamelen) + 1);
1520 			bcopy(pr->pr_name, tpr->pr_name, namelen);
1521 		}
1522 	}
1523 	if (path != NULL) {
1524 		/* Try to keep a real-rooted full pathname. */
1525 		if (path[0] == '/' && strcmp(mypr->pr_path, "/"))
1526 			snprintf(pr->pr_path, sizeof(pr->pr_path), "%s%s",
1527 			    mypr->pr_path, path);
1528 		else
1529 			strlcpy(pr->pr_path, path, sizeof(pr->pr_path));
1530 		pr->pr_root = root;
1531 	}
1532 	if (PR_HOST & ch_flags & ~pr_flags) {
1533 		if (pr->pr_flags & PR_HOST) {
1534 			/*
1535 			 * Copy the parent's host info.  As with pr_ip4 above,
1536 			 * the lack of a lock on the parent is not a problem;
1537 			 * it is always set with allprison_lock at least
1538 			 * shared, and is held exclusively here.
1539 			 */
1540 			strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname,
1541 			    sizeof(pr->pr_hostname));
1542 			strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname,
1543 			    sizeof(pr->pr_domainname));
1544 			strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid,
1545 			    sizeof(pr->pr_hostuuid));
1546 			pr->pr_hostid = pr->pr_parent->pr_hostid;
1547 		}
1548 	} else if (host != NULL || domain != NULL || uuid != NULL || gothid) {
1549 		/* Set this prison, and any descendants without PR_HOST. */
1550 		if (host != NULL)
1551 			strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname));
1552 		if (domain != NULL)
1553 			strlcpy(pr->pr_domainname, domain,
1554 			    sizeof(pr->pr_domainname));
1555 		if (uuid != NULL)
1556 			strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid));
1557 		if (gothid)
1558 			pr->pr_hostid = hid;
1559 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1560 			if (tpr->pr_flags & PR_HOST)
1561 				descend = 0;
1562 			else {
1563 				if (host != NULL)
1564 					strlcpy(tpr->pr_hostname,
1565 					    pr->pr_hostname,
1566 					    sizeof(tpr->pr_hostname));
1567 				if (domain != NULL)
1568 					strlcpy(tpr->pr_domainname,
1569 					    pr->pr_domainname,
1570 					    sizeof(tpr->pr_domainname));
1571 				if (uuid != NULL)
1572 					strlcpy(tpr->pr_hostuuid,
1573 					    pr->pr_hostuuid,
1574 					    sizeof(tpr->pr_hostuuid));
1575 				if (gothid)
1576 					tpr->pr_hostid = hid;
1577 			}
1578 		}
1579 	}
1580 	if ((tallow = ch_allow & ~pr_allow)) {
1581 		/* Clear allow bits in all children. */
1582 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend)
1583 			tpr->pr_allow &= ~tallow;
1584 	}
1585 	pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow;
1586 	/*
1587 	 * Persistent prisons get an extra reference, and prisons losing their
1588 	 * persist flag lose that reference.  Only do this for existing prisons
1589 	 * for now, so new ones will remain unseen until after the module
1590 	 * handlers have completed.
1591 	 */
1592 	if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) {
1593 		if (pr_flags & PR_PERSIST) {
1594 			pr->pr_ref++;
1595 			pr->pr_uref++;
1596 		} else {
1597 			pr->pr_ref--;
1598 			pr->pr_uref--;
1599 		}
1600 	}
1601 	pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags;
1602 	mtx_unlock(&pr->pr_mtx);
1603 
1604 	/* Locks may have prevented a complete restriction of child IP
1605 	 * addresses.  If so, allocate some more memory and try again.
1606 	 */
1607 #ifdef INET
1608 	while (redo_ip4) {
1609 		ip4s = pr->pr_ip4s;
1610 		ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK);
1611 		mtx_lock(&pr->pr_mtx);
1612 		redo_ip4 = 0;
1613 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1614 #ifdef VIMAGE
1615 			if (tpr->pr_flags & PR_VNET) {
1616 				descend = 0;
1617 				continue;
1618 			}
1619 #endif
1620 			if (prison_restrict_ip4(tpr, ip4)) {
1621 				if (ip4 != NULL)
1622 					ip4 = NULL;
1623 				else
1624 					redo_ip4 = 1;
1625 			}
1626 		}
1627 		mtx_unlock(&pr->pr_mtx);
1628 	}
1629 #endif
1630 #ifdef INET6
1631 	while (redo_ip6) {
1632 		ip6s = pr->pr_ip6s;
1633 		ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK);
1634 		mtx_lock(&pr->pr_mtx);
1635 		redo_ip6 = 0;
1636 		FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) {
1637 #ifdef VIMAGE
1638 			if (tpr->pr_flags & PR_VNET) {
1639 				descend = 0;
1640 				continue;
1641 			}
1642 #endif
1643 			if (prison_restrict_ip6(tpr, ip6)) {
1644 				if (ip6 != NULL)
1645 					ip6 = NULL;
1646 				else
1647 					redo_ip6 = 1;
1648 			}
1649 		}
1650 		mtx_unlock(&pr->pr_mtx);
1651 	}
1652 #endif
1653 
1654 	/* Let the modules do their work. */
1655 	sx_downgrade(&allprison_lock);
1656 	if (created) {
1657 		error = osd_jail_call(pr, PR_METHOD_CREATE, opts);
1658 		if (error) {
1659 			prison_deref(pr, PD_LIST_SLOCKED);
1660 			goto done_errmsg;
1661 		}
1662 	}
1663 	error = osd_jail_call(pr, PR_METHOD_SET, opts);
1664 	if (error) {
1665 		prison_deref(pr, created
1666 		    ? PD_LIST_SLOCKED
1667 		    : PD_DEREF | PD_LIST_SLOCKED);
1668 		goto done_errmsg;
1669 	}
1670 
1671 	/* Attach this process to the prison if requested. */
1672 	if (flags & JAIL_ATTACH) {
1673 		mtx_lock(&pr->pr_mtx);
1674 		error = do_jail_attach(td, pr);
1675 		if (error) {
1676 			vfs_opterror(opts, "attach failed");
1677 			if (!created)
1678 				prison_deref(pr, PD_DEREF);
1679 			goto done_errmsg;
1680 		}
1681 	}
1682 
1683 	/*
1684 	 * Now that it is all there, drop the temporary reference from existing
1685 	 * prisons.  Or add a reference to newly created persistent prisons
1686 	 * (which was not done earlier so that the prison would not be publicly
1687 	 * visible).
1688 	 */
1689 	if (!created) {
1690 		prison_deref(pr, (flags & JAIL_ATTACH)
1691 		    ? PD_DEREF
1692 		    : PD_DEREF | PD_LIST_SLOCKED);
1693 	} else {
1694 		if (pr_flags & PR_PERSIST) {
1695 			mtx_lock(&pr->pr_mtx);
1696 			pr->pr_ref++;
1697 			pr->pr_uref++;
1698 			mtx_unlock(&pr->pr_mtx);
1699 		}
1700 		if (!(flags & JAIL_ATTACH))
1701 			sx_sunlock(&allprison_lock);
1702 	}
1703 	td->td_retval[0] = pr->pr_id;
1704 	goto done_errmsg;
1705 
1706  done_deref_locked:
1707 	prison_deref(pr, created
1708 	    ? PD_LOCKED | PD_LIST_XLOCKED
1709 	    : PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
1710 	goto done_releroot;
1711  done_unlock_list:
1712 	sx_xunlock(&allprison_lock);
1713  done_releroot:
1714 	if (root != NULL) {
1715 		vfslocked = VFS_LOCK_GIANT(root->v_mount);
1716 		vrele(root);
1717 		VFS_UNLOCK_GIANT(vfslocked);
1718 	}
1719  done_errmsg:
1720 	if (error) {
1721 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
1722 		if (errmsg_len > 0) {
1723 			errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1;
1724 			if (errmsg_pos > 0) {
1725 				if (optuio->uio_segflg == UIO_SYSSPACE)
1726 					bcopy(errmsg,
1727 					   optuio->uio_iov[errmsg_pos].iov_base,
1728 					   errmsg_len);
1729 				else
1730 					copyout(errmsg,
1731 					   optuio->uio_iov[errmsg_pos].iov_base,
1732 					   errmsg_len);
1733 			}
1734 		}
1735 	}
1736  done_free:
1737 #ifdef INET
1738 	free(ip4, M_PRISON);
1739 #endif
1740 #ifdef INET6
1741 	free(ip6, M_PRISON);
1742 #endif
1743 	vfs_freeopts(opts);
1744 	return (error);
1745 }
1746 
1747 
1748 /*
1749  * struct jail_get_args {
1750  *	struct iovec *iovp;
1751  *	unsigned int iovcnt;
1752  *	int flags;
1753  * };
1754  */
1755 int
1756 jail_get(struct thread *td, struct jail_get_args *uap)
1757 {
1758 	struct uio *auio;
1759 	int error;
1760 
1761 	/* Check that we have an even number of iovecs. */
1762 	if (uap->iovcnt & 1)
1763 		return (EINVAL);
1764 
1765 	error = copyinuio(uap->iovp, uap->iovcnt, &auio);
1766 	if (error)
1767 		return (error);
1768 	error = kern_jail_get(td, auio, uap->flags);
1769 	if (error == 0)
1770 		error = copyout(auio->uio_iov, uap->iovp,
1771 		    uap->iovcnt * sizeof (struct iovec));
1772 	free(auio, M_IOV);
1773 	return (error);
1774 }
1775 
1776 int
1777 kern_jail_get(struct thread *td, struct uio *optuio, int flags)
1778 {
1779 	struct prison *pr, *mypr;
1780 	struct vfsopt *opt;
1781 	struct vfsoptlist *opts;
1782 	char *errmsg, *name;
1783 	int error, errmsg_len, errmsg_pos, fi, i, jid, len, locked, pos;
1784 
1785 	if (flags & ~JAIL_GET_MASK)
1786 		return (EINVAL);
1787 
1788 	/* Get the parameter list. */
1789 	error = vfs_buildopts(optuio, &opts);
1790 	if (error)
1791 		return (error);
1792 	errmsg_pos = vfs_getopt_pos(opts, "errmsg");
1793 	mypr = td->td_ucred->cr_prison;
1794 
1795 	/*
1796 	 * Find the prison specified by one of: lastjid, jid, name.
1797 	 */
1798 	sx_slock(&allprison_lock);
1799 	error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid));
1800 	if (error == 0) {
1801 		TAILQ_FOREACH(pr, &allprison, pr_list) {
1802 			if (pr->pr_id > jid && prison_ischild(mypr, pr)) {
1803 				mtx_lock(&pr->pr_mtx);
1804 				if (pr->pr_ref > 0 &&
1805 				    (pr->pr_uref > 0 || (flags & JAIL_DYING)))
1806 					break;
1807 				mtx_unlock(&pr->pr_mtx);
1808 			}
1809 		}
1810 		if (pr != NULL)
1811 			goto found_prison;
1812 		error = ENOENT;
1813 		vfs_opterror(opts, "no jail after %d", jid);
1814 		goto done_unlock_list;
1815 	} else if (error != ENOENT)
1816 		goto done_unlock_list;
1817 
1818 	error = vfs_copyopt(opts, "jid", &jid, sizeof(jid));
1819 	if (error == 0) {
1820 		if (jid != 0) {
1821 			pr = prison_find_child(mypr, jid);
1822 			if (pr != NULL) {
1823 				if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1824 					mtx_unlock(&pr->pr_mtx);
1825 					error = ENOENT;
1826 					vfs_opterror(opts, "jail %d is dying",
1827 					    jid);
1828 					goto done_unlock_list;
1829 				}
1830 				goto found_prison;
1831 			}
1832 			error = ENOENT;
1833 			vfs_opterror(opts, "jail %d not found", jid);
1834 			goto done_unlock_list;
1835 		}
1836 	} else if (error != ENOENT)
1837 		goto done_unlock_list;
1838 
1839 	error = vfs_getopt(opts, "name", (void **)&name, &len);
1840 	if (error == 0) {
1841 		if (len == 0 || name[len - 1] != '\0') {
1842 			error = EINVAL;
1843 			goto done_unlock_list;
1844 		}
1845 		pr = prison_find_name(mypr, name);
1846 		if (pr != NULL) {
1847 			if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) {
1848 				mtx_unlock(&pr->pr_mtx);
1849 				error = ENOENT;
1850 				vfs_opterror(opts, "jail \"%s\" is dying",
1851 				    name);
1852 				goto done_unlock_list;
1853 			}
1854 			goto found_prison;
1855 		}
1856 		error = ENOENT;
1857 		vfs_opterror(opts, "jail \"%s\" not found", name);
1858 		goto done_unlock_list;
1859 	} else if (error != ENOENT)
1860 		goto done_unlock_list;
1861 
1862 	vfs_opterror(opts, "no jail specified");
1863 	error = ENOENT;
1864 	goto done_unlock_list;
1865 
1866  found_prison:
1867 	/* Get the parameters of the prison. */
1868 	pr->pr_ref++;
1869 	locked = PD_LOCKED;
1870 	td->td_retval[0] = pr->pr_id;
1871 	error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id));
1872 	if (error != 0 && error != ENOENT)
1873 		goto done_deref;
1874 	i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id;
1875 	error = vfs_setopt(opts, "parent", &i, sizeof(i));
1876 	if (error != 0 && error != ENOENT)
1877 		goto done_deref;
1878 	error = vfs_setopts(opts, "name", prison_name(mypr, pr));
1879 	if (error != 0 && error != ENOENT)
1880 		goto done_deref;
1881 	error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id,
1882 	    sizeof(pr->pr_cpuset->cs_id));
1883 	if (error != 0 && error != ENOENT)
1884 		goto done_deref;
1885 	error = vfs_setopts(opts, "path", prison_path(mypr, pr));
1886 	if (error != 0 && error != ENOENT)
1887 		goto done_deref;
1888 #ifdef INET
1889 	error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4,
1890 	    pr->pr_ip4s * sizeof(*pr->pr_ip4));
1891 	if (error != 0 && error != ENOENT)
1892 		goto done_deref;
1893 #endif
1894 #ifdef INET6
1895 	error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6,
1896 	    pr->pr_ip6s * sizeof(*pr->pr_ip6));
1897 	if (error != 0 && error != ENOENT)
1898 		goto done_deref;
1899 #endif
1900 	error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel,
1901 	    sizeof(pr->pr_securelevel));
1902 	if (error != 0 && error != ENOENT)
1903 		goto done_deref;
1904 	error = vfs_setopt(opts, "children.cur", &pr->pr_childcount,
1905 	    sizeof(pr->pr_childcount));
1906 	if (error != 0 && error != ENOENT)
1907 		goto done_deref;
1908 	error = vfs_setopt(opts, "children.max", &pr->pr_childmax,
1909 	    sizeof(pr->pr_childmax));
1910 	if (error != 0 && error != ENOENT)
1911 		goto done_deref;
1912 	error = vfs_setopts(opts, "host.hostname", pr->pr_hostname);
1913 	if (error != 0 && error != ENOENT)
1914 		goto done_deref;
1915 	error = vfs_setopts(opts, "host.domainname", pr->pr_domainname);
1916 	if (error != 0 && error != ENOENT)
1917 		goto done_deref;
1918 	error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid);
1919 	if (error != 0 && error != ENOENT)
1920 		goto done_deref;
1921 #ifdef COMPAT_IA32
1922 	if (td->td_proc->p_sysent->sv_flags & SV_IA32) {
1923 		uint32_t hid32 = pr->pr_hostid;
1924 
1925 		error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32));
1926 	} else
1927 #endif
1928 	error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid,
1929 	    sizeof(pr->pr_hostid));
1930 	if (error != 0 && error != ENOENT)
1931 		goto done_deref;
1932 	error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs,
1933 	    sizeof(pr->pr_enforce_statfs));
1934 	if (error != 0 && error != ENOENT)
1935 		goto done_deref;
1936 	for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
1937 	    fi++) {
1938 		if (pr_flag_names[fi] == NULL)
1939 			continue;
1940 		i = (pr->pr_flags & (1 << fi)) ? 1 : 0;
1941 		error = vfs_setopt(opts, pr_flag_names[fi], &i, sizeof(i));
1942 		if (error != 0 && error != ENOENT)
1943 			goto done_deref;
1944 		i = !i;
1945 		error = vfs_setopt(opts, pr_flag_nonames[fi], &i, sizeof(i));
1946 		if (error != 0 && error != ENOENT)
1947 			goto done_deref;
1948 	}
1949 	for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
1950 	    fi++) {
1951 		i = pr->pr_flags &
1952 		    (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
1953 		i = pr_flag_jailsys[fi].disable &&
1954 		      (i == pr_flag_jailsys[fi].disable) ? JAIL_SYS_DISABLE
1955 		    : (i == pr_flag_jailsys[fi].new) ? JAIL_SYS_NEW
1956 		    : JAIL_SYS_INHERIT;
1957 		error =
1958 		    vfs_setopt(opts, pr_flag_jailsys[fi].name, &i, sizeof(i));
1959 		if (error != 0 && error != ENOENT)
1960 			goto done_deref;
1961 	}
1962 	for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
1963 	    fi++) {
1964 		if (pr_allow_names[fi] == NULL)
1965 			continue;
1966 		i = (pr->pr_allow & (1 << fi)) ? 1 : 0;
1967 		error = vfs_setopt(opts, pr_allow_names[fi], &i, sizeof(i));
1968 		if (error != 0 && error != ENOENT)
1969 			goto done_deref;
1970 		i = !i;
1971 		error = vfs_setopt(opts, pr_allow_nonames[fi], &i, sizeof(i));
1972 		if (error != 0 && error != ENOENT)
1973 			goto done_deref;
1974 	}
1975 	i = (pr->pr_uref == 0);
1976 	error = vfs_setopt(opts, "dying", &i, sizeof(i));
1977 	if (error != 0 && error != ENOENT)
1978 		goto done_deref;
1979 	i = !i;
1980 	error = vfs_setopt(opts, "nodying", &i, sizeof(i));
1981 	if (error != 0 && error != ENOENT)
1982 		goto done_deref;
1983 
1984 	/* Get the module parameters. */
1985 	mtx_unlock(&pr->pr_mtx);
1986 	locked = 0;
1987 	error = osd_jail_call(pr, PR_METHOD_GET, opts);
1988 	if (error)
1989 		goto done_deref;
1990 	prison_deref(pr, PD_DEREF | PD_LIST_SLOCKED);
1991 
1992 	/* By now, all parameters should have been noted. */
1993 	TAILQ_FOREACH(opt, opts, link) {
1994 		if (!opt->seen && strcmp(opt->name, "errmsg")) {
1995 			error = EINVAL;
1996 			vfs_opterror(opts, "unknown parameter: %s", opt->name);
1997 			goto done_errmsg;
1998 		}
1999 	}
2000 
2001 	/* Write the fetched parameters back to userspace. */
2002 	error = 0;
2003 	TAILQ_FOREACH(opt, opts, link) {
2004 		if (opt->pos >= 0 && opt->pos != errmsg_pos) {
2005 			pos = 2 * opt->pos + 1;
2006 			optuio->uio_iov[pos].iov_len = opt->len;
2007 			if (opt->value != NULL) {
2008 				if (optuio->uio_segflg == UIO_SYSSPACE) {
2009 					bcopy(opt->value,
2010 					    optuio->uio_iov[pos].iov_base,
2011 					    opt->len);
2012 				} else {
2013 					error = copyout(opt->value,
2014 					    optuio->uio_iov[pos].iov_base,
2015 					    opt->len);
2016 					if (error)
2017 						break;
2018 				}
2019 			}
2020 		}
2021 	}
2022 	goto done_errmsg;
2023 
2024  done_deref:
2025 	prison_deref(pr, locked | PD_DEREF | PD_LIST_SLOCKED);
2026 	goto done_errmsg;
2027 
2028  done_unlock_list:
2029 	sx_sunlock(&allprison_lock);
2030  done_errmsg:
2031 	if (error && errmsg_pos >= 0) {
2032 		vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len);
2033 		errmsg_pos = 2 * errmsg_pos + 1;
2034 		if (errmsg_len > 0) {
2035 			if (optuio->uio_segflg == UIO_SYSSPACE)
2036 				bcopy(errmsg,
2037 				    optuio->uio_iov[errmsg_pos].iov_base,
2038 				    errmsg_len);
2039 			else
2040 				copyout(errmsg,
2041 				    optuio->uio_iov[errmsg_pos].iov_base,
2042 				    errmsg_len);
2043 		}
2044 	}
2045 	vfs_freeopts(opts);
2046 	return (error);
2047 }
2048 
2049 
2050 /*
2051  * struct jail_remove_args {
2052  *	int jid;
2053  * };
2054  */
2055 int
2056 jail_remove(struct thread *td, struct jail_remove_args *uap)
2057 {
2058 	struct prison *pr, *cpr, *lpr, *tpr;
2059 	int descend, error;
2060 
2061 	error = priv_check(td, PRIV_JAIL_REMOVE);
2062 	if (error)
2063 		return (error);
2064 
2065 	sx_xlock(&allprison_lock);
2066 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2067 	if (pr == NULL) {
2068 		sx_xunlock(&allprison_lock);
2069 		return (EINVAL);
2070 	}
2071 
2072 	/* Remove all descendants of this prison, then remove this prison. */
2073 	pr->pr_ref++;
2074 	pr->pr_flags |= PR_REMOVE;
2075 	if (!LIST_EMPTY(&pr->pr_children)) {
2076 		mtx_unlock(&pr->pr_mtx);
2077 		lpr = NULL;
2078 		FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
2079 			mtx_lock(&cpr->pr_mtx);
2080 			if (cpr->pr_ref > 0) {
2081 				tpr = cpr;
2082 				cpr->pr_ref++;
2083 				cpr->pr_flags |= PR_REMOVE;
2084 			} else {
2085 				/* Already removed - do not do it again. */
2086 				tpr = NULL;
2087 			}
2088 			mtx_unlock(&cpr->pr_mtx);
2089 			if (lpr != NULL) {
2090 				mtx_lock(&lpr->pr_mtx);
2091 				prison_remove_one(lpr);
2092 				sx_xlock(&allprison_lock);
2093 			}
2094 			lpr = tpr;
2095 		}
2096 		if (lpr != NULL) {
2097 			mtx_lock(&lpr->pr_mtx);
2098 			prison_remove_one(lpr);
2099 			sx_xlock(&allprison_lock);
2100 		}
2101 		mtx_lock(&pr->pr_mtx);
2102 	}
2103 	prison_remove_one(pr);
2104 	return (0);
2105 }
2106 
2107 static void
2108 prison_remove_one(struct prison *pr)
2109 {
2110 	struct proc *p;
2111 	int deuref;
2112 
2113 	/* If the prison was persistent, it is not anymore. */
2114 	deuref = 0;
2115 	if (pr->pr_flags & PR_PERSIST) {
2116 		pr->pr_ref--;
2117 		deuref = PD_DEUREF;
2118 		pr->pr_flags &= ~PR_PERSIST;
2119 	}
2120 
2121 	/*
2122 	 * jail_remove added a reference.  If that's the only one, remove
2123 	 * the prison now.
2124 	 */
2125 	KASSERT(pr->pr_ref > 0,
2126 	    ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id));
2127 	if (pr->pr_ref == 1) {
2128 		prison_deref(pr,
2129 		    deuref | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED);
2130 		return;
2131 	}
2132 
2133 	mtx_unlock(&pr->pr_mtx);
2134 	sx_xunlock(&allprison_lock);
2135 	/*
2136 	 * Kill all processes unfortunate enough to be attached to this prison.
2137 	 */
2138 	sx_slock(&allproc_lock);
2139 	LIST_FOREACH(p, &allproc, p_list) {
2140 		PROC_LOCK(p);
2141 		if (p->p_state != PRS_NEW && p->p_ucred &&
2142 		    p->p_ucred->cr_prison == pr)
2143 			psignal(p, SIGKILL);
2144 		PROC_UNLOCK(p);
2145 	}
2146 	sx_sunlock(&allproc_lock);
2147 	/* Remove the temporary reference added by jail_remove. */
2148 	prison_deref(pr, deuref | PD_DEREF);
2149 }
2150 
2151 
2152 /*
2153  * struct jail_attach_args {
2154  *	int jid;
2155  * };
2156  */
2157 int
2158 jail_attach(struct thread *td, struct jail_attach_args *uap)
2159 {
2160 	struct prison *pr;
2161 	int error;
2162 
2163 	error = priv_check(td, PRIV_JAIL_ATTACH);
2164 	if (error)
2165 		return (error);
2166 
2167 	sx_slock(&allprison_lock);
2168 	pr = prison_find_child(td->td_ucred->cr_prison, uap->jid);
2169 	if (pr == NULL) {
2170 		sx_sunlock(&allprison_lock);
2171 		return (EINVAL);
2172 	}
2173 
2174 	/*
2175 	 * Do not allow a process to attach to a prison that is not
2176 	 * considered to be "alive".
2177 	 */
2178 	if (pr->pr_uref == 0) {
2179 		mtx_unlock(&pr->pr_mtx);
2180 		sx_sunlock(&allprison_lock);
2181 		return (EINVAL);
2182 	}
2183 
2184 	return (do_jail_attach(td, pr));
2185 }
2186 
2187 static int
2188 do_jail_attach(struct thread *td, struct prison *pr)
2189 {
2190 	struct prison *ppr;
2191 	struct proc *p;
2192 	struct ucred *newcred, *oldcred;
2193 	int vfslocked, error;
2194 
2195 	/*
2196 	 * XXX: Note that there is a slight race here if two threads
2197 	 * in the same privileged process attempt to attach to two
2198 	 * different jails at the same time.  It is important for
2199 	 * user processes not to do this, or they might end up with
2200 	 * a process root from one prison, but attached to the jail
2201 	 * of another.
2202 	 */
2203 	pr->pr_ref++;
2204 	pr->pr_uref++;
2205 	mtx_unlock(&pr->pr_mtx);
2206 
2207 	/* Let modules do whatever they need to prepare for attaching. */
2208 	error = osd_jail_call(pr, PR_METHOD_ATTACH, td);
2209 	if (error) {
2210 		prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED);
2211 		return (error);
2212 	}
2213 	sx_sunlock(&allprison_lock);
2214 
2215 	/*
2216 	 * Reparent the newly attached process to this jail.
2217 	 */
2218 	ppr = td->td_ucred->cr_prison;
2219 	p = td->td_proc;
2220 	error = cpuset_setproc_update_set(p, pr->pr_cpuset);
2221 	if (error)
2222 		goto e_revert_osd;
2223 
2224 	vfslocked = VFS_LOCK_GIANT(pr->pr_root->v_mount);
2225 	vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY);
2226 	if ((error = change_dir(pr->pr_root, td)) != 0)
2227 		goto e_unlock;
2228 #ifdef MAC
2229 	if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root)))
2230 		goto e_unlock;
2231 #endif
2232 	VOP_UNLOCK(pr->pr_root, 0);
2233 	if ((error = change_root(pr->pr_root, td)))
2234 		goto e_unlock_giant;
2235 	VFS_UNLOCK_GIANT(vfslocked);
2236 
2237 	newcred = crget();
2238 	PROC_LOCK(p);
2239 	oldcred = p->p_ucred;
2240 	setsugid(p);
2241 	crcopy(newcred, oldcred);
2242 	newcred->cr_prison = pr;
2243 	p->p_ucred = newcred;
2244 	PROC_UNLOCK(p);
2245 	crfree(oldcred);
2246 	prison_deref(ppr, PD_DEREF | PD_DEUREF);
2247 	return (0);
2248  e_unlock:
2249 	VOP_UNLOCK(pr->pr_root, 0);
2250  e_unlock_giant:
2251 	VFS_UNLOCK_GIANT(vfslocked);
2252  e_revert_osd:
2253 	/* Tell modules this thread is still in its old jail after all. */
2254 	(void)osd_jail_call(ppr, PR_METHOD_ATTACH, td);
2255 	prison_deref(pr, PD_DEREF | PD_DEUREF);
2256 	return (error);
2257 }
2258 
2259 
2260 /*
2261  * Returns a locked prison instance, or NULL on failure.
2262  */
2263 struct prison *
2264 prison_find(int prid)
2265 {
2266 	struct prison *pr;
2267 
2268 	sx_assert(&allprison_lock, SX_LOCKED);
2269 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2270 		if (pr->pr_id == prid) {
2271 			mtx_lock(&pr->pr_mtx);
2272 			if (pr->pr_ref > 0)
2273 				return (pr);
2274 			mtx_unlock(&pr->pr_mtx);
2275 		}
2276 	}
2277 	return (NULL);
2278 }
2279 
2280 /*
2281  * Find a prison that is a descendant of mypr.  Returns a locked prison or NULL.
2282  */
2283 struct prison *
2284 prison_find_child(struct prison *mypr, int prid)
2285 {
2286 	struct prison *pr;
2287 	int descend;
2288 
2289 	sx_assert(&allprison_lock, SX_LOCKED);
2290 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2291 		if (pr->pr_id == prid) {
2292 			mtx_lock(&pr->pr_mtx);
2293 			if (pr->pr_ref > 0)
2294 				return (pr);
2295 			mtx_unlock(&pr->pr_mtx);
2296 		}
2297 	}
2298 	return (NULL);
2299 }
2300 
2301 /*
2302  * Look for the name relative to mypr.  Returns a locked prison or NULL.
2303  */
2304 struct prison *
2305 prison_find_name(struct prison *mypr, const char *name)
2306 {
2307 	struct prison *pr, *deadpr;
2308 	size_t mylen;
2309 	int descend;
2310 
2311 	sx_assert(&allprison_lock, SX_LOCKED);
2312 	mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1;
2313  again:
2314 	deadpr = NULL;
2315 	FOREACH_PRISON_DESCENDANT(mypr, pr, descend) {
2316 		if (!strcmp(pr->pr_name + mylen, name)) {
2317 			mtx_lock(&pr->pr_mtx);
2318 			if (pr->pr_ref > 0) {
2319 				if (pr->pr_uref > 0)
2320 					return (pr);
2321 				deadpr = pr;
2322 			}
2323 			mtx_unlock(&pr->pr_mtx);
2324 		}
2325 	}
2326 	/* There was no valid prison - perhaps there was a dying one. */
2327 	if (deadpr != NULL) {
2328 		mtx_lock(&deadpr->pr_mtx);
2329 		if (deadpr->pr_ref == 0) {
2330 			mtx_unlock(&deadpr->pr_mtx);
2331 			goto again;
2332 		}
2333 	}
2334 	return (deadpr);
2335 }
2336 
2337 /*
2338  * See if a prison has the specific flag set.
2339  */
2340 int
2341 prison_flag(struct ucred *cred, unsigned flag)
2342 {
2343 
2344 	/* This is an atomic read, so no locking is necessary. */
2345 	return (cred->cr_prison->pr_flags & flag);
2346 }
2347 
2348 int
2349 prison_allow(struct ucred *cred, unsigned flag)
2350 {
2351 
2352 	/* This is an atomic read, so no locking is necessary. */
2353 	return (cred->cr_prison->pr_allow & flag);
2354 }
2355 
2356 /*
2357  * Remove a prison reference.  If that was the last reference, remove the
2358  * prison itself - but not in this context in case there are locks held.
2359  */
2360 void
2361 prison_free_locked(struct prison *pr)
2362 {
2363 
2364 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2365 	pr->pr_ref--;
2366 	if (pr->pr_ref == 0) {
2367 		mtx_unlock(&pr->pr_mtx);
2368 		TASK_INIT(&pr->pr_task, 0, prison_complete, pr);
2369 		taskqueue_enqueue(taskqueue_thread, &pr->pr_task);
2370 		return;
2371 	}
2372 	mtx_unlock(&pr->pr_mtx);
2373 }
2374 
2375 void
2376 prison_free(struct prison *pr)
2377 {
2378 
2379 	mtx_lock(&pr->pr_mtx);
2380 	prison_free_locked(pr);
2381 }
2382 
2383 static void
2384 prison_complete(void *context, int pending)
2385 {
2386 
2387 	prison_deref((struct prison *)context, 0);
2388 }
2389 
2390 /*
2391  * Remove a prison reference (usually).  This internal version assumes no
2392  * mutexes are held, except perhaps the prison itself.  If there are no more
2393  * references, release and delist the prison.  On completion, the prison lock
2394  * and the allprison lock are both unlocked.
2395  */
2396 static void
2397 prison_deref(struct prison *pr, int flags)
2398 {
2399 	struct prison *ppr, *tpr;
2400 	int vfslocked;
2401 
2402 	if (!(flags & PD_LOCKED))
2403 		mtx_lock(&pr->pr_mtx);
2404 	/* Decrement the user references in a separate loop. */
2405 	if (flags & PD_DEUREF) {
2406 		for (tpr = pr;; tpr = tpr->pr_parent) {
2407 			if (tpr != pr)
2408 				mtx_lock(&tpr->pr_mtx);
2409 			if (--tpr->pr_uref > 0)
2410 				break;
2411 			KASSERT(tpr != &prison0, ("prison0 pr_uref=0"));
2412 			mtx_unlock(&tpr->pr_mtx);
2413 		}
2414 		/* Done if there were only user references to remove. */
2415 		if (!(flags & PD_DEREF)) {
2416 			mtx_unlock(&tpr->pr_mtx);
2417 			if (flags & PD_LIST_SLOCKED)
2418 				sx_sunlock(&allprison_lock);
2419 			else if (flags & PD_LIST_XLOCKED)
2420 				sx_xunlock(&allprison_lock);
2421 			return;
2422 		}
2423 		if (tpr != pr) {
2424 			mtx_unlock(&tpr->pr_mtx);
2425 			mtx_lock(&pr->pr_mtx);
2426 		}
2427 	}
2428 
2429 	for (;;) {
2430 		if (flags & PD_DEREF)
2431 			pr->pr_ref--;
2432 		/* If the prison still has references, nothing else to do. */
2433 		if (pr->pr_ref > 0) {
2434 			mtx_unlock(&pr->pr_mtx);
2435 			if (flags & PD_LIST_SLOCKED)
2436 				sx_sunlock(&allprison_lock);
2437 			else if (flags & PD_LIST_XLOCKED)
2438 				sx_xunlock(&allprison_lock);
2439 			return;
2440 		}
2441 
2442 		mtx_unlock(&pr->pr_mtx);
2443 		if (flags & PD_LIST_SLOCKED) {
2444 			if (!sx_try_upgrade(&allprison_lock)) {
2445 				sx_sunlock(&allprison_lock);
2446 				sx_xlock(&allprison_lock);
2447 			}
2448 		} else if (!(flags & PD_LIST_XLOCKED))
2449 			sx_xlock(&allprison_lock);
2450 
2451 		TAILQ_REMOVE(&allprison, pr, pr_list);
2452 		LIST_REMOVE(pr, pr_sibling);
2453 		ppr = pr->pr_parent;
2454 		for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent)
2455 			tpr->pr_childcount--;
2456 		sx_downgrade(&allprison_lock);
2457 
2458 #ifdef VIMAGE
2459 		if (pr->pr_flags & PR_VNET)
2460 			vnet_destroy(pr->pr_vnet);
2461 #endif
2462 		if (pr->pr_root != NULL) {
2463 			vfslocked = VFS_LOCK_GIANT(pr->pr_root->v_mount);
2464 			vrele(pr->pr_root);
2465 			VFS_UNLOCK_GIANT(vfslocked);
2466 		}
2467 		mtx_destroy(&pr->pr_mtx);
2468 #ifdef INET
2469 		free(pr->pr_ip4, M_PRISON);
2470 #endif
2471 #ifdef INET6
2472 		free(pr->pr_ip6, M_PRISON);
2473 #endif
2474 		if (pr->pr_cpuset != NULL)
2475 			cpuset_rel(pr->pr_cpuset);
2476 		osd_jail_exit(pr);
2477 		free(pr, M_PRISON);
2478 
2479 		/* Removing a prison frees a reference on its parent. */
2480 		pr = ppr;
2481 		mtx_lock(&pr->pr_mtx);
2482 		flags = PD_DEREF | PD_LIST_SLOCKED;
2483 	}
2484 }
2485 
2486 void
2487 prison_hold_locked(struct prison *pr)
2488 {
2489 
2490 	mtx_assert(&pr->pr_mtx, MA_OWNED);
2491 	KASSERT(pr->pr_ref > 0,
2492 	    ("Trying to hold dead prison (jid=%d).", pr->pr_id));
2493 	pr->pr_ref++;
2494 }
2495 
2496 void
2497 prison_hold(struct prison *pr)
2498 {
2499 
2500 	mtx_lock(&pr->pr_mtx);
2501 	prison_hold_locked(pr);
2502 	mtx_unlock(&pr->pr_mtx);
2503 }
2504 
2505 void
2506 prison_proc_hold(struct prison *pr)
2507 {
2508 
2509 	mtx_lock(&pr->pr_mtx);
2510 	KASSERT(pr->pr_uref > 0,
2511 	    ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id));
2512 	pr->pr_uref++;
2513 	mtx_unlock(&pr->pr_mtx);
2514 }
2515 
2516 void
2517 prison_proc_free(struct prison *pr)
2518 {
2519 
2520 	mtx_lock(&pr->pr_mtx);
2521 	KASSERT(pr->pr_uref > 0,
2522 	    ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id));
2523 	prison_deref(pr, PD_DEUREF | PD_LOCKED);
2524 }
2525 
2526 
2527 #ifdef INET
2528 /*
2529  * Restrict a prison's IP address list with its parent's, possibly replacing
2530  * it.  Return true if the replacement buffer was used (or would have been).
2531  */
2532 static int
2533 prison_restrict_ip4(struct prison *pr, struct in_addr *newip4)
2534 {
2535 	int ii, ij, used;
2536 	struct prison *ppr;
2537 
2538 	ppr = pr->pr_parent;
2539 	if (!(pr->pr_flags & PR_IP4_USER)) {
2540 		/* This has no user settings, so just copy the parent's list. */
2541 		if (pr->pr_ip4s < ppr->pr_ip4s) {
2542 			/*
2543 			 * There's no room for the parent's list.  Use the
2544 			 * new list buffer, which is assumed to be big enough
2545 			 * (if it was passed).  If there's no buffer, try to
2546 			 * allocate one.
2547 			 */
2548 			used = 1;
2549 			if (newip4 == NULL) {
2550 				newip4 = malloc(ppr->pr_ip4s * sizeof(*newip4),
2551 				    M_PRISON, M_NOWAIT);
2552 				if (newip4 != NULL)
2553 					used = 0;
2554 			}
2555 			if (newip4 != NULL) {
2556 				bcopy(ppr->pr_ip4, newip4,
2557 				    ppr->pr_ip4s * sizeof(*newip4));
2558 				free(pr->pr_ip4, M_PRISON);
2559 				pr->pr_ip4 = newip4;
2560 				pr->pr_ip4s = ppr->pr_ip4s;
2561 			}
2562 			return (used);
2563 		}
2564 		pr->pr_ip4s = ppr->pr_ip4s;
2565 		if (pr->pr_ip4s > 0)
2566 			bcopy(ppr->pr_ip4, pr->pr_ip4,
2567 			    pr->pr_ip4s * sizeof(*newip4));
2568 		else if (pr->pr_ip4 != NULL) {
2569 			free(pr->pr_ip4, M_PRISON);
2570 			pr->pr_ip4 = NULL;
2571 		}
2572 	} else if (pr->pr_ip4s > 0) {
2573 		/* Remove addresses that aren't in the parent. */
2574 		for (ij = 0; ij < ppr->pr_ip4s; ij++)
2575 			if (pr->pr_ip4[0].s_addr == ppr->pr_ip4[ij].s_addr)
2576 				break;
2577 		if (ij < ppr->pr_ip4s)
2578 			ii = 1;
2579 		else {
2580 			bcopy(pr->pr_ip4 + 1, pr->pr_ip4,
2581 			    --pr->pr_ip4s * sizeof(*pr->pr_ip4));
2582 			ii = 0;
2583 		}
2584 		for (ij = 1; ii < pr->pr_ip4s; ) {
2585 			if (pr->pr_ip4[ii].s_addr == ppr->pr_ip4[0].s_addr) {
2586 				ii++;
2587 				continue;
2588 			}
2589 			switch (ij >= ppr->pr_ip4s ? -1 :
2590 				qcmp_v4(&pr->pr_ip4[ii], &ppr->pr_ip4[ij])) {
2591 			case -1:
2592 				bcopy(pr->pr_ip4 + ii + 1, pr->pr_ip4 + ii,
2593 				    (--pr->pr_ip4s - ii) * sizeof(*pr->pr_ip4));
2594 				break;
2595 			case 0:
2596 				ii++;
2597 				ij++;
2598 				break;
2599 			case 1:
2600 				ij++;
2601 				break;
2602 			}
2603 		}
2604 		if (pr->pr_ip4s == 0) {
2605 			pr->pr_flags |= PR_IP4_DISABLE;
2606 			free(pr->pr_ip4, M_PRISON);
2607 			pr->pr_ip4 = NULL;
2608 		}
2609 	}
2610 	return (0);
2611 }
2612 
2613 /*
2614  * Pass back primary IPv4 address of this jail.
2615  *
2616  * If not restricted return success but do not alter the address.  Caller has
2617  * to make sure to initialize it correctly (e.g. INADDR_ANY).
2618  *
2619  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv4.
2620  * Address returned in NBO.
2621  */
2622 int
2623 prison_get_ip4(struct ucred *cred, struct in_addr *ia)
2624 {
2625 	struct prison *pr;
2626 
2627 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2628 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2629 
2630 	pr = cred->cr_prison;
2631 	if (!(pr->pr_flags & PR_IP4))
2632 		return (0);
2633 	mtx_lock(&pr->pr_mtx);
2634 	if (!(pr->pr_flags & PR_IP4)) {
2635 		mtx_unlock(&pr->pr_mtx);
2636 		return (0);
2637 	}
2638 	if (pr->pr_ip4 == NULL) {
2639 		mtx_unlock(&pr->pr_mtx);
2640 		return (EAFNOSUPPORT);
2641 	}
2642 
2643 	ia->s_addr = pr->pr_ip4[0].s_addr;
2644 	mtx_unlock(&pr->pr_mtx);
2645 	return (0);
2646 }
2647 
2648 /*
2649  * Return true if pr1 and pr2 have the same IPv4 address restrictions.
2650  */
2651 int
2652 prison_equal_ip4(struct prison *pr1, struct prison *pr2)
2653 {
2654 
2655 	if (pr1 == pr2)
2656 		return (1);
2657 
2658 	/*
2659 	 * No need to lock since the PR_IP4_USER flag can't be altered for
2660 	 * existing prisons.
2661 	 */
2662 	while (pr1 != &prison0 &&
2663 #ifdef VIMAGE
2664 	       !(pr1->pr_flags & PR_VNET) &&
2665 #endif
2666 	       !(pr1->pr_flags & PR_IP4_USER))
2667 		pr1 = pr1->pr_parent;
2668 	while (pr2 != &prison0 &&
2669 #ifdef VIMAGE
2670 	       !(pr2->pr_flags & PR_VNET) &&
2671 #endif
2672 	       !(pr2->pr_flags & PR_IP4_USER))
2673 		pr2 = pr2->pr_parent;
2674 	return (pr1 == pr2);
2675 }
2676 
2677 /*
2678  * Make sure our (source) address is set to something meaningful to this
2679  * jail.
2680  *
2681  * Returns 0 if jail doesn't restrict IPv4 or if address belongs to jail,
2682  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
2683  * doesn't allow IPv4.  Address passed in in NBO and returned in NBO.
2684  */
2685 int
2686 prison_local_ip4(struct ucred *cred, struct in_addr *ia)
2687 {
2688 	struct prison *pr;
2689 	struct in_addr ia0;
2690 	int error;
2691 
2692 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2693 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2694 
2695 	pr = cred->cr_prison;
2696 	if (!(pr->pr_flags & PR_IP4))
2697 		return (0);
2698 	mtx_lock(&pr->pr_mtx);
2699 	if (!(pr->pr_flags & PR_IP4)) {
2700 		mtx_unlock(&pr->pr_mtx);
2701 		return (0);
2702 	}
2703 	if (pr->pr_ip4 == NULL) {
2704 		mtx_unlock(&pr->pr_mtx);
2705 		return (EAFNOSUPPORT);
2706 	}
2707 
2708 	ia0.s_addr = ntohl(ia->s_addr);
2709 	if (ia0.s_addr == INADDR_LOOPBACK) {
2710 		ia->s_addr = pr->pr_ip4[0].s_addr;
2711 		mtx_unlock(&pr->pr_mtx);
2712 		return (0);
2713 	}
2714 
2715 	if (ia0.s_addr == INADDR_ANY) {
2716 		/*
2717 		 * In case there is only 1 IPv4 address, bind directly.
2718 		 */
2719 		if (pr->pr_ip4s == 1)
2720 			ia->s_addr = pr->pr_ip4[0].s_addr;
2721 		mtx_unlock(&pr->pr_mtx);
2722 		return (0);
2723 	}
2724 
2725 	error = _prison_check_ip4(pr, ia);
2726 	mtx_unlock(&pr->pr_mtx);
2727 	return (error);
2728 }
2729 
2730 /*
2731  * Rewrite destination address in case we will connect to loopback address.
2732  *
2733  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv4.
2734  * Address passed in in NBO and returned in NBO.
2735  */
2736 int
2737 prison_remote_ip4(struct ucred *cred, struct in_addr *ia)
2738 {
2739 	struct prison *pr;
2740 
2741 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2742 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2743 
2744 	pr = cred->cr_prison;
2745 	if (!(pr->pr_flags & PR_IP4))
2746 		return (0);
2747 	mtx_lock(&pr->pr_mtx);
2748 	if (!(pr->pr_flags & PR_IP4)) {
2749 		mtx_unlock(&pr->pr_mtx);
2750 		return (0);
2751 	}
2752 	if (pr->pr_ip4 == NULL) {
2753 		mtx_unlock(&pr->pr_mtx);
2754 		return (EAFNOSUPPORT);
2755 	}
2756 
2757 	if (ntohl(ia->s_addr) == INADDR_LOOPBACK) {
2758 		ia->s_addr = pr->pr_ip4[0].s_addr;
2759 		mtx_unlock(&pr->pr_mtx);
2760 		return (0);
2761 	}
2762 
2763 	/*
2764 	 * Return success because nothing had to be changed.
2765 	 */
2766 	mtx_unlock(&pr->pr_mtx);
2767 	return (0);
2768 }
2769 
2770 /*
2771  * Check if given address belongs to the jail referenced by cred/prison.
2772  *
2773  * Returns 0 if jail doesn't restrict IPv4 or if address belongs to jail,
2774  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
2775  * doesn't allow IPv4.  Address passed in in NBO.
2776  */
2777 static int
2778 _prison_check_ip4(struct prison *pr, struct in_addr *ia)
2779 {
2780 	int i, a, z, d;
2781 
2782 	/*
2783 	 * Check the primary IP.
2784 	 */
2785 	if (pr->pr_ip4[0].s_addr == ia->s_addr)
2786 		return (0);
2787 
2788 	/*
2789 	 * All the other IPs are sorted so we can do a binary search.
2790 	 */
2791 	a = 0;
2792 	z = pr->pr_ip4s - 2;
2793 	while (a <= z) {
2794 		i = (a + z) / 2;
2795 		d = qcmp_v4(&pr->pr_ip4[i+1], ia);
2796 		if (d > 0)
2797 			z = i - 1;
2798 		else if (d < 0)
2799 			a = i + 1;
2800 		else
2801 			return (0);
2802 	}
2803 
2804 	return (EADDRNOTAVAIL);
2805 }
2806 
2807 int
2808 prison_check_ip4(struct ucred *cred, struct in_addr *ia)
2809 {
2810 	struct prison *pr;
2811 	int error;
2812 
2813 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2814 	KASSERT(ia != NULL, ("%s: ia is NULL", __func__));
2815 
2816 	pr = cred->cr_prison;
2817 	if (!(pr->pr_flags & PR_IP4))
2818 		return (0);
2819 	mtx_lock(&pr->pr_mtx);
2820 	if (!(pr->pr_flags & PR_IP4)) {
2821 		mtx_unlock(&pr->pr_mtx);
2822 		return (0);
2823 	}
2824 	if (pr->pr_ip4 == NULL) {
2825 		mtx_unlock(&pr->pr_mtx);
2826 		return (EAFNOSUPPORT);
2827 	}
2828 
2829 	error = _prison_check_ip4(pr, ia);
2830 	mtx_unlock(&pr->pr_mtx);
2831 	return (error);
2832 }
2833 #endif
2834 
2835 #ifdef INET6
2836 static int
2837 prison_restrict_ip6(struct prison *pr, struct in6_addr *newip6)
2838 {
2839 	int ii, ij, used;
2840 	struct prison *ppr;
2841 
2842 	ppr = pr->pr_parent;
2843 	if (!(pr->pr_flags & PR_IP6_USER)) {
2844 		/* This has no user settings, so just copy the parent's list. */
2845 		if (pr->pr_ip6s < ppr->pr_ip6s) {
2846 			/*
2847 			 * There's no room for the parent's list.  Use the
2848 			 * new list buffer, which is assumed to be big enough
2849 			 * (if it was passed).  If there's no buffer, try to
2850 			 * allocate one.
2851 			 */
2852 			used = 1;
2853 			if (newip6 == NULL) {
2854 				newip6 = malloc(ppr->pr_ip6s * sizeof(*newip6),
2855 				    M_PRISON, M_NOWAIT);
2856 				if (newip6 != NULL)
2857 					used = 0;
2858 			}
2859 			if (newip6 != NULL) {
2860 				bcopy(ppr->pr_ip6, newip6,
2861 				    ppr->pr_ip6s * sizeof(*newip6));
2862 				free(pr->pr_ip6, M_PRISON);
2863 				pr->pr_ip6 = newip6;
2864 				pr->pr_ip6s = ppr->pr_ip6s;
2865 			}
2866 			return (used);
2867 		}
2868 		pr->pr_ip6s = ppr->pr_ip6s;
2869 		if (pr->pr_ip6s > 0)
2870 			bcopy(ppr->pr_ip6, pr->pr_ip6,
2871 			    pr->pr_ip6s * sizeof(*newip6));
2872 		else if (pr->pr_ip6 != NULL) {
2873 			free(pr->pr_ip6, M_PRISON);
2874 			pr->pr_ip6 = NULL;
2875 		}
2876 	} else if (pr->pr_ip6s > 0) {
2877 		/* Remove addresses that aren't in the parent. */
2878 		for (ij = 0; ij < ppr->pr_ip6s; ij++)
2879 			if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[0],
2880 			    &ppr->pr_ip6[ij]))
2881 				break;
2882 		if (ij < ppr->pr_ip6s)
2883 			ii = 1;
2884 		else {
2885 			bcopy(pr->pr_ip6 + 1, pr->pr_ip6,
2886 			    --pr->pr_ip6s * sizeof(*pr->pr_ip6));
2887 			ii = 0;
2888 		}
2889 		for (ij = 1; ii < pr->pr_ip6s; ) {
2890 			if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[ii],
2891 			    &ppr->pr_ip6[0])) {
2892 				ii++;
2893 				continue;
2894 			}
2895 			switch (ij >= ppr->pr_ip4s ? -1 :
2896 				qcmp_v6(&pr->pr_ip6[ii], &ppr->pr_ip6[ij])) {
2897 			case -1:
2898 				bcopy(pr->pr_ip6 + ii + 1, pr->pr_ip6 + ii,
2899 				    (--pr->pr_ip6s - ii) * sizeof(*pr->pr_ip6));
2900 				break;
2901 			case 0:
2902 				ii++;
2903 				ij++;
2904 				break;
2905 			case 1:
2906 				ij++;
2907 				break;
2908 			}
2909 		}
2910 		if (pr->pr_ip6s == 0) {
2911 			pr->pr_flags |= PR_IP6_DISABLE;
2912 			free(pr->pr_ip6, M_PRISON);
2913 			pr->pr_ip6 = NULL;
2914 		}
2915 	}
2916 	return 0;
2917 }
2918 
2919 /*
2920  * Pass back primary IPv6 address for this jail.
2921  *
2922  * If not restricted return success but do not alter the address.  Caller has
2923  * to make sure to initialize it correctly (e.g. IN6ADDR_ANY_INIT).
2924  *
2925  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv6.
2926  */
2927 int
2928 prison_get_ip6(struct ucred *cred, struct in6_addr *ia6)
2929 {
2930 	struct prison *pr;
2931 
2932 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2933 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
2934 
2935 	pr = cred->cr_prison;
2936 	if (!(pr->pr_flags & PR_IP6))
2937 		return (0);
2938 	mtx_lock(&pr->pr_mtx);
2939 	if (!(pr->pr_flags & PR_IP6)) {
2940 		mtx_unlock(&pr->pr_mtx);
2941 		return (0);
2942 	}
2943 	if (pr->pr_ip6 == NULL) {
2944 		mtx_unlock(&pr->pr_mtx);
2945 		return (EAFNOSUPPORT);
2946 	}
2947 
2948 	bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
2949 	mtx_unlock(&pr->pr_mtx);
2950 	return (0);
2951 }
2952 
2953 /*
2954  * Return true if pr1 and pr2 have the same IPv6 address restrictions.
2955  */
2956 int
2957 prison_equal_ip6(struct prison *pr1, struct prison *pr2)
2958 {
2959 
2960 	if (pr1 == pr2)
2961 		return (1);
2962 
2963 	while (pr1 != &prison0 &&
2964 #ifdef VIMAGE
2965 	       !(pr1->pr_flags & PR_VNET) &&
2966 #endif
2967 	       !(pr1->pr_flags & PR_IP6_USER))
2968 		pr1 = pr1->pr_parent;
2969 	while (pr2 != &prison0 &&
2970 #ifdef VIMAGE
2971 	       !(pr2->pr_flags & PR_VNET) &&
2972 #endif
2973 	       !(pr2->pr_flags & PR_IP6_USER))
2974 		pr2 = pr2->pr_parent;
2975 	return (pr1 == pr2);
2976 }
2977 
2978 /*
2979  * Make sure our (source) address is set to something meaningful to this jail.
2980  *
2981  * v6only should be set based on (inp->inp_flags & IN6P_IPV6_V6ONLY != 0)
2982  * when needed while binding.
2983  *
2984  * Returns 0 if jail doesn't restrict IPv6 or if address belongs to jail,
2985  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
2986  * doesn't allow IPv6.
2987  */
2988 int
2989 prison_local_ip6(struct ucred *cred, struct in6_addr *ia6, int v6only)
2990 {
2991 	struct prison *pr;
2992 	int error;
2993 
2994 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
2995 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
2996 
2997 	pr = cred->cr_prison;
2998 	if (!(pr->pr_flags & PR_IP6))
2999 		return (0);
3000 	mtx_lock(&pr->pr_mtx);
3001 	if (!(pr->pr_flags & PR_IP6)) {
3002 		mtx_unlock(&pr->pr_mtx);
3003 		return (0);
3004 	}
3005 	if (pr->pr_ip6 == NULL) {
3006 		mtx_unlock(&pr->pr_mtx);
3007 		return (EAFNOSUPPORT);
3008 	}
3009 
3010 	if (IN6_IS_ADDR_LOOPBACK(ia6)) {
3011 		bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3012 		mtx_unlock(&pr->pr_mtx);
3013 		return (0);
3014 	}
3015 
3016 	if (IN6_IS_ADDR_UNSPECIFIED(ia6)) {
3017 		/*
3018 		 * In case there is only 1 IPv6 address, and v6only is true,
3019 		 * then bind directly.
3020 		 */
3021 		if (v6only != 0 && pr->pr_ip6s == 1)
3022 			bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3023 		mtx_unlock(&pr->pr_mtx);
3024 		return (0);
3025 	}
3026 
3027 	error = _prison_check_ip6(pr, ia6);
3028 	mtx_unlock(&pr->pr_mtx);
3029 	return (error);
3030 }
3031 
3032 /*
3033  * Rewrite destination address in case we will connect to loopback address.
3034  *
3035  * Returns 0 on success, EAFNOSUPPORT if the jail doesn't allow IPv6.
3036  */
3037 int
3038 prison_remote_ip6(struct ucred *cred, struct in6_addr *ia6)
3039 {
3040 	struct prison *pr;
3041 
3042 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3043 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3044 
3045 	pr = cred->cr_prison;
3046 	if (!(pr->pr_flags & PR_IP6))
3047 		return (0);
3048 	mtx_lock(&pr->pr_mtx);
3049 	if (!(pr->pr_flags & PR_IP6)) {
3050 		mtx_unlock(&pr->pr_mtx);
3051 		return (0);
3052 	}
3053 	if (pr->pr_ip6 == NULL) {
3054 		mtx_unlock(&pr->pr_mtx);
3055 		return (EAFNOSUPPORT);
3056 	}
3057 
3058 	if (IN6_IS_ADDR_LOOPBACK(ia6)) {
3059 		bcopy(&pr->pr_ip6[0], ia6, sizeof(struct in6_addr));
3060 		mtx_unlock(&pr->pr_mtx);
3061 		return (0);
3062 	}
3063 
3064 	/*
3065 	 * Return success because nothing had to be changed.
3066 	 */
3067 	mtx_unlock(&pr->pr_mtx);
3068 	return (0);
3069 }
3070 
3071 /*
3072  * Check if given address belongs to the jail referenced by cred/prison.
3073  *
3074  * Returns 0 if jail doesn't restrict IPv6 or if address belongs to jail,
3075  * EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if the jail
3076  * doesn't allow IPv6.
3077  */
3078 static int
3079 _prison_check_ip6(struct prison *pr, struct in6_addr *ia6)
3080 {
3081 	int i, a, z, d;
3082 
3083 	/*
3084 	 * Check the primary IP.
3085 	 */
3086 	if (IN6_ARE_ADDR_EQUAL(&pr->pr_ip6[0], ia6))
3087 		return (0);
3088 
3089 	/*
3090 	 * All the other IPs are sorted so we can do a binary search.
3091 	 */
3092 	a = 0;
3093 	z = pr->pr_ip6s - 2;
3094 	while (a <= z) {
3095 		i = (a + z) / 2;
3096 		d = qcmp_v6(&pr->pr_ip6[i+1], ia6);
3097 		if (d > 0)
3098 			z = i - 1;
3099 		else if (d < 0)
3100 			a = i + 1;
3101 		else
3102 			return (0);
3103 	}
3104 
3105 	return (EADDRNOTAVAIL);
3106 }
3107 
3108 int
3109 prison_check_ip6(struct ucred *cred, struct in6_addr *ia6)
3110 {
3111 	struct prison *pr;
3112 	int error;
3113 
3114 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3115 	KASSERT(ia6 != NULL, ("%s: ia6 is NULL", __func__));
3116 
3117 	pr = cred->cr_prison;
3118 	if (!(pr->pr_flags & PR_IP6))
3119 		return (0);
3120 	mtx_lock(&pr->pr_mtx);
3121 	if (!(pr->pr_flags & PR_IP6)) {
3122 		mtx_unlock(&pr->pr_mtx);
3123 		return (0);
3124 	}
3125 	if (pr->pr_ip6 == NULL) {
3126 		mtx_unlock(&pr->pr_mtx);
3127 		return (EAFNOSUPPORT);
3128 	}
3129 
3130 	error = _prison_check_ip6(pr, ia6);
3131 	mtx_unlock(&pr->pr_mtx);
3132 	return (error);
3133 }
3134 #endif
3135 
3136 /*
3137  * Check if a jail supports the given address family.
3138  *
3139  * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT
3140  * if not.
3141  */
3142 int
3143 prison_check_af(struct ucred *cred, int af)
3144 {
3145 	struct prison *pr;
3146 	int error;
3147 
3148 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3149 
3150 	pr = cred->cr_prison;
3151 #ifdef VIMAGE
3152 	/* Prisons with their own network stack are not limited. */
3153 	if (pr->pr_flags & PR_VNET)
3154 		return (0);
3155 #endif
3156 
3157 	error = 0;
3158 	switch (af)
3159 	{
3160 #ifdef INET
3161 	case AF_INET:
3162 		if (pr->pr_flags & PR_IP4)
3163 		{
3164 			mtx_lock(&pr->pr_mtx);
3165 			if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL)
3166 				error = EAFNOSUPPORT;
3167 			mtx_unlock(&pr->pr_mtx);
3168 		}
3169 		break;
3170 #endif
3171 #ifdef INET6
3172 	case AF_INET6:
3173 		if (pr->pr_flags & PR_IP6)
3174 		{
3175 			mtx_lock(&pr->pr_mtx);
3176 			if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL)
3177 				error = EAFNOSUPPORT;
3178 			mtx_unlock(&pr->pr_mtx);
3179 		}
3180 		break;
3181 #endif
3182 	case AF_LOCAL:
3183 	case AF_ROUTE:
3184 		break;
3185 	default:
3186 		if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF))
3187 			error = EAFNOSUPPORT;
3188 	}
3189 	return (error);
3190 }
3191 
3192 /*
3193  * Check if given address belongs to the jail referenced by cred (wrapper to
3194  * prison_check_ip[46]).
3195  *
3196  * Returns 0 if jail doesn't restrict the address family or if address belongs
3197  * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if
3198  * the jail doesn't allow the address family.  IPv4 Address passed in in NBO.
3199  */
3200 int
3201 prison_if(struct ucred *cred, struct sockaddr *sa)
3202 {
3203 #ifdef INET
3204 	struct sockaddr_in *sai;
3205 #endif
3206 #ifdef INET6
3207 	struct sockaddr_in6 *sai6;
3208 #endif
3209 	int error;
3210 
3211 	KASSERT(cred != NULL, ("%s: cred is NULL", __func__));
3212 	KASSERT(sa != NULL, ("%s: sa is NULL", __func__));
3213 
3214 	error = 0;
3215 	switch (sa->sa_family)
3216 	{
3217 #ifdef INET
3218 	case AF_INET:
3219 		sai = (struct sockaddr_in *)sa;
3220 		error = prison_check_ip4(cred, &sai->sin_addr);
3221 		break;
3222 #endif
3223 #ifdef INET6
3224 	case AF_INET6:
3225 		sai6 = (struct sockaddr_in6 *)sa;
3226 		error = prison_check_ip6(cred, &sai6->sin6_addr);
3227 		break;
3228 #endif
3229 	default:
3230 		if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF))
3231 			error = EAFNOSUPPORT;
3232 	}
3233 	return (error);
3234 }
3235 
3236 /*
3237  * Return 0 if jails permit p1 to frob p2, otherwise ESRCH.
3238  */
3239 int
3240 prison_check(struct ucred *cred1, struct ucred *cred2)
3241 {
3242 
3243 	return ((cred1->cr_prison == cred2->cr_prison ||
3244 	    prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH);
3245 }
3246 
3247 /*
3248  * Return 1 if p2 is a child of p1, otherwise 0.
3249  */
3250 int
3251 prison_ischild(struct prison *pr1, struct prison *pr2)
3252 {
3253 
3254 	for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent)
3255 		if (pr1 == pr2)
3256 			return (1);
3257 	return (0);
3258 }
3259 
3260 /*
3261  * Return 1 if the passed credential is in a jail, otherwise 0.
3262  */
3263 int
3264 jailed(struct ucred *cred)
3265 {
3266 
3267 	return (cred->cr_prison != &prison0);
3268 }
3269 
3270 /*
3271  * Return the correct hostname (domainname, et al) for the passed credential.
3272  */
3273 void
3274 getcredhostname(struct ucred *cred, char *buf, size_t size)
3275 {
3276 	struct prison *pr;
3277 
3278 	/*
3279 	 * A NULL credential can be used to shortcut to the physical
3280 	 * system's hostname.
3281 	 */
3282 	pr = (cred != NULL) ? cred->cr_prison : &prison0;
3283 	mtx_lock(&pr->pr_mtx);
3284 	strlcpy(buf, pr->pr_hostname, size);
3285 	mtx_unlock(&pr->pr_mtx);
3286 }
3287 
3288 void
3289 getcreddomainname(struct ucred *cred, char *buf, size_t size)
3290 {
3291 
3292 	mtx_lock(&cred->cr_prison->pr_mtx);
3293 	strlcpy(buf, cred->cr_prison->pr_domainname, size);
3294 	mtx_unlock(&cred->cr_prison->pr_mtx);
3295 }
3296 
3297 void
3298 getcredhostuuid(struct ucred *cred, char *buf, size_t size)
3299 {
3300 
3301 	mtx_lock(&cred->cr_prison->pr_mtx);
3302 	strlcpy(buf, cred->cr_prison->pr_hostuuid, size);
3303 	mtx_unlock(&cred->cr_prison->pr_mtx);
3304 }
3305 
3306 void
3307 getcredhostid(struct ucred *cred, unsigned long *hostid)
3308 {
3309 
3310 	mtx_lock(&cred->cr_prison->pr_mtx);
3311 	*hostid = cred->cr_prison->pr_hostid;
3312 	mtx_unlock(&cred->cr_prison->pr_mtx);
3313 }
3314 
3315 #ifdef VIMAGE
3316 /*
3317  * Determine whether the prison represented by cred owns
3318  * its vnet rather than having it inherited.
3319  *
3320  * Returns 1 in case the prison owns the vnet, 0 otherwise.
3321  */
3322 int
3323 prison_owns_vnet(struct ucred *cred)
3324 {
3325 
3326 	/*
3327 	 * vnets cannot be added/removed after jail creation,
3328 	 * so no need to lock here.
3329 	 */
3330 	return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0);
3331 }
3332 #endif
3333 
3334 /*
3335  * Determine whether the subject represented by cred can "see"
3336  * status of a mount point.
3337  * Returns: 0 for permitted, ENOENT otherwise.
3338  * XXX: This function should be called cr_canseemount() and should be
3339  *      placed in kern_prot.c.
3340  */
3341 int
3342 prison_canseemount(struct ucred *cred, struct mount *mp)
3343 {
3344 	struct prison *pr;
3345 	struct statfs *sp;
3346 	size_t len;
3347 
3348 	pr = cred->cr_prison;
3349 	if (pr->pr_enforce_statfs == 0)
3350 		return (0);
3351 	if (pr->pr_root->v_mount == mp)
3352 		return (0);
3353 	if (pr->pr_enforce_statfs == 2)
3354 		return (ENOENT);
3355 	/*
3356 	 * If jail's chroot directory is set to "/" we should be able to see
3357 	 * all mount-points from inside a jail.
3358 	 * This is ugly check, but this is the only situation when jail's
3359 	 * directory ends with '/'.
3360 	 */
3361 	if (strcmp(pr->pr_path, "/") == 0)
3362 		return (0);
3363 	len = strlen(pr->pr_path);
3364 	sp = &mp->mnt_stat;
3365 	if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0)
3366 		return (ENOENT);
3367 	/*
3368 	 * Be sure that we don't have situation where jail's root directory
3369 	 * is "/some/path" and mount point is "/some/pathpath".
3370 	 */
3371 	if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/')
3372 		return (ENOENT);
3373 	return (0);
3374 }
3375 
3376 void
3377 prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp)
3378 {
3379 	char jpath[MAXPATHLEN];
3380 	struct prison *pr;
3381 	size_t len;
3382 
3383 	pr = cred->cr_prison;
3384 	if (pr->pr_enforce_statfs == 0)
3385 		return;
3386 	if (prison_canseemount(cred, mp) != 0) {
3387 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3388 		strlcpy(sp->f_mntonname, "[restricted]",
3389 		    sizeof(sp->f_mntonname));
3390 		return;
3391 	}
3392 	if (pr->pr_root->v_mount == mp) {
3393 		/*
3394 		 * Clear current buffer data, so we are sure nothing from
3395 		 * the valid path left there.
3396 		 */
3397 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3398 		*sp->f_mntonname = '/';
3399 		return;
3400 	}
3401 	/*
3402 	 * If jail's chroot directory is set to "/" we should be able to see
3403 	 * all mount-points from inside a jail.
3404 	 */
3405 	if (strcmp(pr->pr_path, "/") == 0)
3406 		return;
3407 	len = strlen(pr->pr_path);
3408 	strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath));
3409 	/*
3410 	 * Clear current buffer data, so we are sure nothing from
3411 	 * the valid path left there.
3412 	 */
3413 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3414 	if (*jpath == '\0') {
3415 		/* Should never happen. */
3416 		*sp->f_mntonname = '/';
3417 	} else {
3418 		strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname));
3419 	}
3420 }
3421 
3422 /*
3423  * Check with permission for a specific privilege is granted within jail.  We
3424  * have a specific list of accepted privileges; the rest are denied.
3425  */
3426 int
3427 prison_priv_check(struct ucred *cred, int priv)
3428 {
3429 
3430 	if (!jailed(cred))
3431 		return (0);
3432 
3433 #ifdef VIMAGE
3434 	/*
3435 	 * Privileges specific to prisons with a virtual network stack.
3436 	 * There might be a duplicate entry here in case the privilege
3437 	 * is only granted conditionally in the legacy jail case.
3438 	 */
3439 	switch (priv) {
3440 #ifdef notyet
3441 		/*
3442 		 * NFS-specific privileges.
3443 		 */
3444 	case PRIV_NFS_DAEMON:
3445 	case PRIV_NFS_LOCKD:
3446 #endif
3447 		/*
3448 		 * Network stack privileges.
3449 		 */
3450 	case PRIV_NET_BRIDGE:
3451 	case PRIV_NET_GRE:
3452 	case PRIV_NET_BPF:
3453 	case PRIV_NET_RAW:		/* Dup, cond. in legacy jail case. */
3454 	case PRIV_NET_ROUTE:
3455 	case PRIV_NET_TAP:
3456 	case PRIV_NET_SETIFMTU:
3457 	case PRIV_NET_SETIFFLAGS:
3458 	case PRIV_NET_SETIFCAP:
3459 	case PRIV_NET_SETIFNAME	:
3460 	case PRIV_NET_SETIFMETRIC:
3461 	case PRIV_NET_SETIFPHYS:
3462 	case PRIV_NET_SETIFMAC:
3463 	case PRIV_NET_ADDMULTI:
3464 	case PRIV_NET_DELMULTI:
3465 	case PRIV_NET_HWIOCTL:
3466 	case PRIV_NET_SETLLADDR:
3467 	case PRIV_NET_ADDIFGROUP:
3468 	case PRIV_NET_DELIFGROUP:
3469 	case PRIV_NET_IFCREATE:
3470 	case PRIV_NET_IFDESTROY:
3471 	case PRIV_NET_ADDIFADDR:
3472 	case PRIV_NET_DELIFADDR:
3473 	case PRIV_NET_LAGG:
3474 	case PRIV_NET_GIF:
3475 	case PRIV_NET_SETIFVNET:
3476 
3477 		/*
3478 		 * 802.11-related privileges.
3479 		 */
3480 	case PRIV_NET80211_GETKEY:
3481 #ifdef notyet
3482 	case PRIV_NET80211_MANAGE:		/* XXX-BZ discuss with sam@ */
3483 #endif
3484 
3485 #ifdef notyet
3486 		/*
3487 		 * AppleTalk privileges.
3488 		 */
3489 	case PRIV_NETATALK_RESERVEDPORT:
3490 
3491 		/*
3492 		 * ATM privileges.
3493 		 */
3494 	case PRIV_NETATM_CFG:
3495 	case PRIV_NETATM_ADD:
3496 	case PRIV_NETATM_DEL:
3497 	case PRIV_NETATM_SET:
3498 
3499 		/*
3500 		 * Bluetooth privileges.
3501 		 */
3502 	case PRIV_NETBLUETOOTH_RAW:
3503 #endif
3504 
3505 		/*
3506 		 * Netgraph and netgraph module privileges.
3507 		 */
3508 	case PRIV_NETGRAPH_CONTROL:
3509 #ifdef notyet
3510 	case PRIV_NETGRAPH_TTY:
3511 #endif
3512 
3513 		/*
3514 		 * IPv4 and IPv6 privileges.
3515 		 */
3516 	case PRIV_NETINET_IPFW:
3517 	case PRIV_NETINET_DIVERT:
3518 	case PRIV_NETINET_PF:
3519 	case PRIV_NETINET_DUMMYNET:
3520 	case PRIV_NETINET_CARP:
3521 	case PRIV_NETINET_MROUTE:
3522 	case PRIV_NETINET_RAW:
3523 	case PRIV_NETINET_ADDRCTRL6:
3524 	case PRIV_NETINET_ND6:
3525 	case PRIV_NETINET_SCOPE6:
3526 	case PRIV_NETINET_ALIFETIME6:
3527 	case PRIV_NETINET_IPSEC:
3528 	case PRIV_NETINET_BINDANY:
3529 
3530 #ifdef notyet
3531 		/*
3532 		 * IPX/SPX privileges.
3533 		 */
3534 	case PRIV_NETIPX_RESERVEDPORT:
3535 	case PRIV_NETIPX_RAW:
3536 
3537 		/*
3538 		 * NCP privileges.
3539 		 */
3540 	case PRIV_NETNCP:
3541 
3542 		/*
3543 		 * SMB privileges.
3544 		 */
3545 	case PRIV_NETSMB:
3546 #endif
3547 
3548 	/*
3549 	 * No default: or deny here.
3550 	 * In case of no permit fall through to next switch().
3551 	 */
3552 		if (cred->cr_prison->pr_flags & PR_VNET)
3553 			return (0);
3554 	}
3555 #endif /* VIMAGE */
3556 
3557 	switch (priv) {
3558 
3559 		/*
3560 		 * Allow ktrace privileges for root in jail.
3561 		 */
3562 	case PRIV_KTRACE:
3563 
3564 #if 0
3565 		/*
3566 		 * Allow jailed processes to configure audit identity and
3567 		 * submit audit records (login, etc).  In the future we may
3568 		 * want to further refine the relationship between audit and
3569 		 * jail.
3570 		 */
3571 	case PRIV_AUDIT_GETAUDIT:
3572 	case PRIV_AUDIT_SETAUDIT:
3573 	case PRIV_AUDIT_SUBMIT:
3574 #endif
3575 
3576 		/*
3577 		 * Allow jailed processes to manipulate process UNIX
3578 		 * credentials in any way they see fit.
3579 		 */
3580 	case PRIV_CRED_SETUID:
3581 	case PRIV_CRED_SETEUID:
3582 	case PRIV_CRED_SETGID:
3583 	case PRIV_CRED_SETEGID:
3584 	case PRIV_CRED_SETGROUPS:
3585 	case PRIV_CRED_SETREUID:
3586 	case PRIV_CRED_SETREGID:
3587 	case PRIV_CRED_SETRESUID:
3588 	case PRIV_CRED_SETRESGID:
3589 
3590 		/*
3591 		 * Jail implements visibility constraints already, so allow
3592 		 * jailed root to override uid/gid-based constraints.
3593 		 */
3594 	case PRIV_SEEOTHERGIDS:
3595 	case PRIV_SEEOTHERUIDS:
3596 
3597 		/*
3598 		 * Jail implements inter-process debugging limits already, so
3599 		 * allow jailed root various debugging privileges.
3600 		 */
3601 	case PRIV_DEBUG_DIFFCRED:
3602 	case PRIV_DEBUG_SUGID:
3603 	case PRIV_DEBUG_UNPRIV:
3604 
3605 		/*
3606 		 * Allow jail to set various resource limits and login
3607 		 * properties, and for now, exceed process resource limits.
3608 		 */
3609 	case PRIV_PROC_LIMIT:
3610 	case PRIV_PROC_SETLOGIN:
3611 	case PRIV_PROC_SETRLIMIT:
3612 
3613 		/*
3614 		 * System V and POSIX IPC privileges are granted in jail.
3615 		 */
3616 	case PRIV_IPC_READ:
3617 	case PRIV_IPC_WRITE:
3618 	case PRIV_IPC_ADMIN:
3619 	case PRIV_IPC_MSGSIZE:
3620 	case PRIV_MQ_ADMIN:
3621 
3622 		/*
3623 		 * Jail operations within a jail work on child jails.
3624 		 */
3625 	case PRIV_JAIL_ATTACH:
3626 	case PRIV_JAIL_SET:
3627 	case PRIV_JAIL_REMOVE:
3628 
3629 		/*
3630 		 * Jail implements its own inter-process limits, so allow
3631 		 * root processes in jail to change scheduling on other
3632 		 * processes in the same jail.  Likewise for signalling.
3633 		 */
3634 	case PRIV_SCHED_DIFFCRED:
3635 	case PRIV_SCHED_CPUSET:
3636 	case PRIV_SIGNAL_DIFFCRED:
3637 	case PRIV_SIGNAL_SUGID:
3638 
3639 		/*
3640 		 * Allow jailed processes to write to sysctls marked as jail
3641 		 * writable.
3642 		 */
3643 	case PRIV_SYSCTL_WRITEJAIL:
3644 
3645 		/*
3646 		 * Allow root in jail to manage a variety of quota
3647 		 * properties.  These should likely be conditional on a
3648 		 * configuration option.
3649 		 */
3650 	case PRIV_VFS_GETQUOTA:
3651 	case PRIV_VFS_SETQUOTA:
3652 
3653 		/*
3654 		 * Since Jail relies on chroot() to implement file system
3655 		 * protections, grant many VFS privileges to root in jail.
3656 		 * Be careful to exclude mount-related and NFS-related
3657 		 * privileges.
3658 		 */
3659 	case PRIV_VFS_READ:
3660 	case PRIV_VFS_WRITE:
3661 	case PRIV_VFS_ADMIN:
3662 	case PRIV_VFS_EXEC:
3663 	case PRIV_VFS_LOOKUP:
3664 	case PRIV_VFS_BLOCKRESERVE:	/* XXXRW: Slightly surprising. */
3665 	case PRIV_VFS_CHFLAGS_DEV:
3666 	case PRIV_VFS_CHOWN:
3667 	case PRIV_VFS_CHROOT:
3668 	case PRIV_VFS_RETAINSUGID:
3669 	case PRIV_VFS_FCHROOT:
3670 	case PRIV_VFS_LINK:
3671 	case PRIV_VFS_SETGID:
3672 	case PRIV_VFS_STAT:
3673 	case PRIV_VFS_STICKYFILE:
3674 		return (0);
3675 
3676 		/*
3677 		 * Depending on the global setting, allow privilege of
3678 		 * setting system flags.
3679 		 */
3680 	case PRIV_VFS_SYSFLAGS:
3681 		if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS)
3682 			return (0);
3683 		else
3684 			return (EPERM);
3685 
3686 		/*
3687 		 * Depending on the global setting, allow privilege of
3688 		 * mounting/unmounting file systems.
3689 		 */
3690 	case PRIV_VFS_MOUNT:
3691 	case PRIV_VFS_UNMOUNT:
3692 	case PRIV_VFS_MOUNT_NONUSER:
3693 	case PRIV_VFS_MOUNT_OWNER:
3694 		if (cred->cr_prison->pr_allow & PR_ALLOW_MOUNT)
3695 			return (0);
3696 		else
3697 			return (EPERM);
3698 
3699 		/*
3700 		 * Allow jailed root to bind reserved ports and reuse in-use
3701 		 * ports.
3702 		 */
3703 	case PRIV_NETINET_RESERVEDPORT:
3704 	case PRIV_NETINET_REUSEPORT:
3705 		return (0);
3706 
3707 		/*
3708 		 * Allow jailed root to set certian IPv4/6 (option) headers.
3709 		 */
3710 	case PRIV_NETINET_SETHDROPTS:
3711 		return (0);
3712 
3713 		/*
3714 		 * Conditionally allow creating raw sockets in jail.
3715 		 */
3716 	case PRIV_NETINET_RAW:
3717 		if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS)
3718 			return (0);
3719 		else
3720 			return (EPERM);
3721 
3722 		/*
3723 		 * Since jail implements its own visibility limits on netstat
3724 		 * sysctls, allow getcred.  This allows identd to work in
3725 		 * jail.
3726 		 */
3727 	case PRIV_NETINET_GETCRED:
3728 		return (0);
3729 
3730 	default:
3731 		/*
3732 		 * In all remaining cases, deny the privilege request.  This
3733 		 * includes almost all network privileges, many system
3734 		 * configuration privileges.
3735 		 */
3736 		return (EPERM);
3737 	}
3738 }
3739 
3740 /*
3741  * Return the part of pr2's name that is relative to pr1, or the whole name
3742  * if it does not directly follow.
3743  */
3744 
3745 char *
3746 prison_name(struct prison *pr1, struct prison *pr2)
3747 {
3748 	char *name;
3749 
3750 	/* Jails see themselves as "0" (if they see themselves at all). */
3751 	if (pr1 == pr2)
3752 		return "0";
3753 	name = pr2->pr_name;
3754 	if (prison_ischild(pr1, pr2)) {
3755 		/*
3756 		 * pr1 isn't locked (and allprison_lock may not be either)
3757 		 * so its length can't be counted on.  But the number of dots
3758 		 * can be counted on - and counted.
3759 		 */
3760 		for (; pr1 != &prison0; pr1 = pr1->pr_parent)
3761 			name = strchr(name, '.') + 1;
3762 	}
3763 	return (name);
3764 }
3765 
3766 /*
3767  * Return the part of pr2's path that is relative to pr1, or the whole path
3768  * if it does not directly follow.
3769  */
3770 static char *
3771 prison_path(struct prison *pr1, struct prison *pr2)
3772 {
3773 	char *path1, *path2;
3774 	int len1;
3775 
3776 	path1 = pr1->pr_path;
3777 	path2 = pr2->pr_path;
3778 	if (!strcmp(path1, "/"))
3779 		return (path2);
3780 	len1 = strlen(path1);
3781 	if (strncmp(path1, path2, len1))
3782 		return (path2);
3783 	if (path2[len1] == '\0')
3784 		return "/";
3785 	if (path2[len1] == '/')
3786 		return (path2 + len1);
3787 	return (path2);
3788 }
3789 
3790 
3791 /*
3792  * Jail-related sysctls.
3793  */
3794 SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW, 0,
3795     "Jails");
3796 
3797 static int
3798 sysctl_jail_list(SYSCTL_HANDLER_ARGS)
3799 {
3800 	struct xprison *xp;
3801 	struct prison *pr, *cpr;
3802 #ifdef INET
3803 	struct in_addr *ip4 = NULL;
3804 	int ip4s = 0;
3805 #endif
3806 #ifdef INET6
3807 	struct in_addr *ip6 = NULL;
3808 	int ip6s = 0;
3809 #endif
3810 	int descend, error;
3811 
3812 	xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK);
3813 	pr = req->td->td_ucred->cr_prison;
3814 	error = 0;
3815 	sx_slock(&allprison_lock);
3816 	FOREACH_PRISON_DESCENDANT(pr, cpr, descend) {
3817 #if defined(INET) || defined(INET6)
3818  again:
3819 #endif
3820 		mtx_lock(&cpr->pr_mtx);
3821 #ifdef INET
3822 		if (cpr->pr_ip4s > 0) {
3823 			if (ip4s < cpr->pr_ip4s) {
3824 				ip4s = cpr->pr_ip4s;
3825 				mtx_unlock(&cpr->pr_mtx);
3826 				ip4 = realloc(ip4, ip4s *
3827 				    sizeof(struct in_addr), M_TEMP, M_WAITOK);
3828 				goto again;
3829 			}
3830 			bcopy(cpr->pr_ip4, ip4,
3831 			    cpr->pr_ip4s * sizeof(struct in_addr));
3832 		}
3833 #endif
3834 #ifdef INET6
3835 		if (cpr->pr_ip6s > 0) {
3836 			if (ip6s < cpr->pr_ip6s) {
3837 				ip6s = cpr->pr_ip6s;
3838 				mtx_unlock(&cpr->pr_mtx);
3839 				ip6 = realloc(ip6, ip6s *
3840 				    sizeof(struct in6_addr), M_TEMP, M_WAITOK);
3841 				goto again;
3842 			}
3843 			bcopy(cpr->pr_ip6, ip6,
3844 			    cpr->pr_ip6s * sizeof(struct in6_addr));
3845 		}
3846 #endif
3847 		if (cpr->pr_ref == 0) {
3848 			mtx_unlock(&cpr->pr_mtx);
3849 			continue;
3850 		}
3851 		bzero(xp, sizeof(*xp));
3852 		xp->pr_version = XPRISON_VERSION;
3853 		xp->pr_id = cpr->pr_id;
3854 		xp->pr_state = cpr->pr_uref > 0
3855 		    ? PRISON_STATE_ALIVE : PRISON_STATE_DYING;
3856 		strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path));
3857 		strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host));
3858 		strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name));
3859 #ifdef INET
3860 		xp->pr_ip4s = cpr->pr_ip4s;
3861 #endif
3862 #ifdef INET6
3863 		xp->pr_ip6s = cpr->pr_ip6s;
3864 #endif
3865 		mtx_unlock(&cpr->pr_mtx);
3866 		error = SYSCTL_OUT(req, xp, sizeof(*xp));
3867 		if (error)
3868 			break;
3869 #ifdef INET
3870 		if (xp->pr_ip4s > 0) {
3871 			error = SYSCTL_OUT(req, ip4,
3872 			    xp->pr_ip4s * sizeof(struct in_addr));
3873 			if (error)
3874 				break;
3875 		}
3876 #endif
3877 #ifdef INET6
3878 		if (xp->pr_ip6s > 0) {
3879 			error = SYSCTL_OUT(req, ip6,
3880 			    xp->pr_ip6s * sizeof(struct in6_addr));
3881 			if (error)
3882 				break;
3883 		}
3884 #endif
3885 	}
3886 	sx_sunlock(&allprison_lock);
3887 	free(xp, M_TEMP);
3888 #ifdef INET
3889 	free(ip4, M_TEMP);
3890 #endif
3891 #ifdef INET6
3892 	free(ip6, M_TEMP);
3893 #endif
3894 	return (error);
3895 }
3896 
3897 SYSCTL_OID(_security_jail, OID_AUTO, list,
3898     CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3899     sysctl_jail_list, "S", "List of active jails");
3900 
3901 static int
3902 sysctl_jail_jailed(SYSCTL_HANDLER_ARGS)
3903 {
3904 	int error, injail;
3905 
3906 	injail = jailed(req->td->td_ucred);
3907 	error = SYSCTL_OUT(req, &injail, sizeof(injail));
3908 
3909 	return (error);
3910 }
3911 
3912 SYSCTL_PROC(_security_jail, OID_AUTO, jailed,
3913     CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
3914     sysctl_jail_jailed, "I", "Process in jail?");
3915 
3916 #if defined(INET) || defined(INET6)
3917 SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW,
3918     &jail_max_af_ips, 0,
3919     "Number of IP addresses a jail may have at most per address family");
3920 #endif
3921 
3922 /*
3923  * Default parameters for jail(2) compatability.  For historical reasons,
3924  * the sysctl names have varying similarity to the parameter names.  Prisons
3925  * just see their own parameters, and can't change them.
3926  */
3927 static int
3928 sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS)
3929 {
3930 	struct prison *pr;
3931 	int allow, error, i;
3932 
3933 	pr = req->td->td_ucred->cr_prison;
3934 	allow = (pr == &prison0) ? jail_default_allow : pr->pr_allow;
3935 
3936 	/* Get the current flag value, and convert it to a boolean. */
3937 	i = (allow & arg2) ? 1 : 0;
3938 	if (arg1 != NULL)
3939 		i = !i;
3940 	error = sysctl_handle_int(oidp, &i, 0, req);
3941 	if (error || !req->newptr)
3942 		return (error);
3943 	i = i ? arg2 : 0;
3944 	if (arg1 != NULL)
3945 		i ^= arg2;
3946 	/*
3947 	 * The sysctls don't have CTLFLAGS_PRISON, so assume prison0
3948 	 * for writing.
3949 	 */
3950 	mtx_lock(&prison0.pr_mtx);
3951 	jail_default_allow = (jail_default_allow & ~arg2) | i;
3952 	mtx_unlock(&prison0.pr_mtx);
3953 	return (0);
3954 }
3955 
3956 SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed,
3957     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3958     NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I",
3959     "Processes in jail can set their hostnames");
3960 SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only,
3961     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3962     (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I",
3963     "Processes in jail are limited to creating UNIX/IP/route sockets only");
3964 SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed,
3965     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3966     NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I",
3967     "Processes in jail can use System V IPC primitives");
3968 SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets,
3969     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3970     NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I",
3971     "Prison root can create raw sockets");
3972 SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed,
3973     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3974     NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I",
3975     "Processes in jail can alter system file flags");
3976 SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed,
3977     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3978     NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I",
3979     "Processes in jail can mount/unmount jail-friendly file systems");
3980 
3981 static int
3982 sysctl_jail_default_level(SYSCTL_HANDLER_ARGS)
3983 {
3984 	struct prison *pr;
3985 	int level, error;
3986 
3987 	pr = req->td->td_ucred->cr_prison;
3988 	level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2);
3989 	error = sysctl_handle_int(oidp, &level, 0, req);
3990 	if (error || !req->newptr)
3991 		return (error);
3992 	*(int *)arg1 = level;
3993 	return (0);
3994 }
3995 
3996 SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs,
3997     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
3998     &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs),
3999     sysctl_jail_default_level, "I",
4000     "Processes in jail cannot see all mounted file systems");
4001 
4002 /*
4003  * Nodes to describe jail parameters.  Maximum length of string parameters
4004  * is returned in the string itself, and the other parameters exist merely
4005  * to make themselves and their types known.
4006  */
4007 SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW, 0,
4008     "Jail parameters");
4009 
4010 int
4011 sysctl_jail_param(SYSCTL_HANDLER_ARGS)
4012 {
4013 	int i;
4014 	long l;
4015 	size_t s;
4016 	char numbuf[12];
4017 
4018 	switch (oidp->oid_kind & CTLTYPE)
4019 	{
4020 	case CTLTYPE_LONG:
4021 	case CTLTYPE_ULONG:
4022 		l = 0;
4023 #ifdef SCTL_MASK32
4024 		if (!(req->flags & SCTL_MASK32))
4025 #endif
4026 			return (SYSCTL_OUT(req, &l, sizeof(l)));
4027 	case CTLTYPE_INT:
4028 	case CTLTYPE_UINT:
4029 		i = 0;
4030 		return (SYSCTL_OUT(req, &i, sizeof(i)));
4031 	case CTLTYPE_STRING:
4032 		snprintf(numbuf, sizeof(numbuf), "%d", arg2);
4033 		return
4034 		    (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req));
4035 	case CTLTYPE_STRUCT:
4036 		s = (size_t)arg2;
4037 		return (SYSCTL_OUT(req, &s, sizeof(s)));
4038 	}
4039 	return (0);
4040 }
4041 
4042 SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID");
4043 SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID");
4044 SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name");
4045 SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path");
4046 SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW,
4047     "I", "Jail secure level");
4048 SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW,
4049     "I", "Jail cannot see all mounted file systems");
4050 SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW,
4051     "B", "Jail persistence");
4052 #ifdef VIMAGE
4053 SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN,
4054     "E,jailsys", "Virtual network stack");
4055 #endif
4056 SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD,
4057     "B", "Jail is in the process of shutting down");
4058 
4059 SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails");
4060 SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD,
4061     "I", "Current number of child jails");
4062 SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW,
4063     "I", "Maximum number of child jails");
4064 
4065 SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info");
4066 SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN,
4067     "Jail hostname");
4068 SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN,
4069     "Jail NIS domainname");
4070 SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN,
4071     "Jail host UUID");
4072 SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW,
4073     "LU", "Jail host ID");
4074 
4075 SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset");
4076 SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID");
4077 
4078 #ifdef INET
4079 SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN,
4080     "Jail IPv4 address virtualization");
4081 SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr),
4082     "S,in_addr,a", "Jail IPv4 addresses");
4083 #endif
4084 #ifdef INET6
4085 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN,
4086     "Jail IPv6 address virtualization");
4087 SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr),
4088     "S,in6_addr,a", "Jail IPv6 addresses");
4089 #endif
4090 
4091 SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags");
4092 SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW,
4093     "B", "Jail may set hostname");
4094 SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW,
4095     "B", "Jail may use SYSV IPC");
4096 SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW,
4097     "B", "Jail may create raw sockets");
4098 SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW,
4099     "B", "Jail may alter system file flags");
4100 SYSCTL_JAIL_PARAM(_allow, mount, CTLTYPE_INT | CTLFLAG_RW,
4101     "B", "Jail may mount/unmount jail-friendly file systems");
4102 SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW,
4103     "B", "Jail may set file quotas");
4104 SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW,
4105     "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route");
4106 
4107 
4108 #ifdef DDB
4109 
4110 static void
4111 db_show_prison(struct prison *pr)
4112 {
4113 	int fi;
4114 #if defined(INET) || defined(INET6)
4115 	int ii;
4116 #endif
4117 	unsigned jsf;
4118 #ifdef INET6
4119 	char ip6buf[INET6_ADDRSTRLEN];
4120 #endif
4121 
4122 	db_printf("prison %p:\n", pr);
4123 	db_printf(" jid             = %d\n", pr->pr_id);
4124 	db_printf(" name            = %s\n", pr->pr_name);
4125 	db_printf(" parent          = %p\n", pr->pr_parent);
4126 	db_printf(" ref             = %d\n", pr->pr_ref);
4127 	db_printf(" uref            = %d\n", pr->pr_uref);
4128 	db_printf(" path            = %s\n", pr->pr_path);
4129 	db_printf(" cpuset          = %d\n", pr->pr_cpuset
4130 	    ? pr->pr_cpuset->cs_id : -1);
4131 #ifdef VIMAGE
4132 	db_printf(" vnet            = %p\n", pr->pr_vnet);
4133 #endif
4134 	db_printf(" root            = %p\n", pr->pr_root);
4135 	db_printf(" securelevel     = %d\n", pr->pr_securelevel);
4136 	db_printf(" childcount      = %d\n", pr->pr_childcount);
4137 	db_printf(" child           = %p\n", LIST_FIRST(&pr->pr_children));
4138 	db_printf(" sibling         = %p\n", LIST_NEXT(pr, pr_sibling));
4139 	db_printf(" flags           = %x", pr->pr_flags);
4140 	for (fi = 0; fi < sizeof(pr_flag_names) / sizeof(pr_flag_names[0]);
4141 	    fi++)
4142 		if (pr_flag_names[fi] != NULL && (pr->pr_flags & (1 << fi)))
4143 			db_printf(" %s", pr_flag_names[fi]);
4144 	for (fi = 0; fi < sizeof(pr_flag_jailsys) / sizeof(pr_flag_jailsys[0]);
4145 	    fi++) {
4146 		jsf = pr->pr_flags &
4147 		    (pr_flag_jailsys[fi].disable | pr_flag_jailsys[fi].new);
4148 		db_printf(" %-16s= %s\n", pr_flag_jailsys[fi].name,
4149 		    pr_flag_jailsys[fi].disable &&
4150 		      (jsf == pr_flag_jailsys[fi].disable) ? "disable"
4151 		    : (jsf == pr_flag_jailsys[fi].new) ? "new"
4152 		    : "inherit");
4153 	}
4154 	db_printf(" allow           = %x", pr->pr_allow);
4155 	for (fi = 0; fi < sizeof(pr_allow_names) / sizeof(pr_allow_names[0]);
4156 	    fi++)
4157 		if (pr_allow_names[fi] != NULL && (pr->pr_allow & (1 << fi)))
4158 			db_printf(" %s", pr_allow_names[fi]);
4159 	db_printf("\n");
4160 	db_printf(" enforce_statfs  = %d\n", pr->pr_enforce_statfs);
4161 	db_printf(" host.hostname   = %s\n", pr->pr_hostname);
4162 	db_printf(" host.domainname = %s\n", pr->pr_domainname);
4163 	db_printf(" host.hostuuid   = %s\n", pr->pr_hostuuid);
4164 	db_printf(" host.hostid     = %lu\n", pr->pr_hostid);
4165 #ifdef INET
4166 	db_printf(" ip4s            = %d\n", pr->pr_ip4s);
4167 	for (ii = 0; ii < pr->pr_ip4s; ii++)
4168 		db_printf(" %s %s\n",
4169 		    ii == 0 ? "ip4             =" : "                 ",
4170 		    inet_ntoa(pr->pr_ip4[ii]));
4171 #endif
4172 #ifdef INET6
4173 	db_printf(" ip6s            = %d\n", pr->pr_ip6s);
4174 	for (ii = 0; ii < pr->pr_ip6s; ii++)
4175 		db_printf(" %s %s\n",
4176 		    ii == 0 ? "ip6             =" : "                 ",
4177 		    ip6_sprintf(ip6buf, &pr->pr_ip6[ii]));
4178 #endif
4179 }
4180 
4181 DB_SHOW_COMMAND(prison, db_show_prison_command)
4182 {
4183 	struct prison *pr;
4184 
4185 	if (!have_addr) {
4186 		/*
4187 		 * Show all prisons in the list, and prison0 which is not
4188 		 * listed.
4189 		 */
4190 		db_show_prison(&prison0);
4191 		if (!db_pager_quit) {
4192 			TAILQ_FOREACH(pr, &allprison, pr_list) {
4193 				db_show_prison(pr);
4194 				if (db_pager_quit)
4195 					break;
4196 			}
4197 		}
4198 		return;
4199 	}
4200 
4201 	if (addr == 0)
4202 		pr = &prison0;
4203 	else {
4204 		/* Look for a prison with the ID and with references. */
4205 		TAILQ_FOREACH(pr, &allprison, pr_list)
4206 			if (pr->pr_id == addr && pr->pr_ref > 0)
4207 				break;
4208 		if (pr == NULL)
4209 			/* Look again, without requiring a reference. */
4210 			TAILQ_FOREACH(pr, &allprison, pr_list)
4211 				if (pr->pr_id == addr)
4212 					break;
4213 		if (pr == NULL)
4214 			/* Assume address points to a valid prison. */
4215 			pr = (struct prison *)addr;
4216 	}
4217 	db_show_prison(pr);
4218 }
4219 
4220 #endif /* DDB */
4221